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> 42fe55f6d5SVegard Nossum #include <linux/kmemcheck.h> 431da177e4SLinus Torvalds #include <linux/mm.h> 441da177e4SLinus Torvalds #include <linux/interrupt.h> 451da177e4SLinus Torvalds #include <linux/in.h> 461da177e4SLinus Torvalds #include <linux/inet.h> 471da177e4SLinus Torvalds #include <linux/slab.h> 481da177e4SLinus Torvalds #include <linux/netdevice.h> 491da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 501da177e4SLinus Torvalds #include <net/pkt_sched.h> 511da177e4SLinus Torvalds #endif 521da177e4SLinus Torvalds #include <linux/string.h> 531da177e4SLinus Torvalds #include <linux/skbuff.h> 549c55e01cSJens Axboe #include <linux/splice.h> 551da177e4SLinus Torvalds #include <linux/cache.h> 561da177e4SLinus Torvalds #include <linux/rtnetlink.h> 571da177e4SLinus Torvalds #include <linux/init.h> 58716ea3a7SDavid Howells #include <linux/scatterlist.h> 59ac45f602SPatrick Ohly #include <linux/errqueue.h> 601da177e4SLinus Torvalds 611da177e4SLinus Torvalds #include <net/protocol.h> 621da177e4SLinus Torvalds #include <net/dst.h> 631da177e4SLinus Torvalds #include <net/sock.h> 641da177e4SLinus Torvalds #include <net/checksum.h> 651da177e4SLinus Torvalds #include <net/xfrm.h> 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds #include <asm/uaccess.h> 681da177e4SLinus Torvalds #include <asm/system.h> 69ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 701da177e4SLinus Torvalds 71a1f8e7f7SAl Viro #include "kmap_skb.h" 72a1f8e7f7SAl Viro 73e18b890bSChristoph Lameter static struct kmem_cache *skbuff_head_cache __read_mostly; 74e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 751da177e4SLinus Torvalds 769c55e01cSJens Axboe static void sock_pipe_buf_release(struct pipe_inode_info *pipe, 779c55e01cSJens Axboe struct pipe_buffer *buf) 789c55e01cSJens Axboe { 798b9d3728SJarek Poplawski put_page(buf->page); 809c55e01cSJens Axboe } 819c55e01cSJens Axboe 829c55e01cSJens Axboe static void sock_pipe_buf_get(struct pipe_inode_info *pipe, 839c55e01cSJens Axboe struct pipe_buffer *buf) 849c55e01cSJens Axboe { 858b9d3728SJarek Poplawski get_page(buf->page); 869c55e01cSJens Axboe } 879c55e01cSJens Axboe 889c55e01cSJens Axboe static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, 899c55e01cSJens Axboe struct pipe_buffer *buf) 909c55e01cSJens Axboe { 919c55e01cSJens Axboe return 1; 929c55e01cSJens Axboe } 939c55e01cSJens Axboe 949c55e01cSJens Axboe 959c55e01cSJens Axboe /* Pipe buffer operations for a socket. */ 9628dfef8fSAlexey Dobriyan static const struct pipe_buf_operations sock_pipe_buf_ops = { 979c55e01cSJens Axboe .can_merge = 0, 989c55e01cSJens Axboe .map = generic_pipe_buf_map, 999c55e01cSJens Axboe .unmap = generic_pipe_buf_unmap, 1009c55e01cSJens Axboe .confirm = generic_pipe_buf_confirm, 1019c55e01cSJens Axboe .release = sock_pipe_buf_release, 1029c55e01cSJens Axboe .steal = sock_pipe_buf_steal, 1039c55e01cSJens Axboe .get = sock_pipe_buf_get, 1049c55e01cSJens Axboe }; 1059c55e01cSJens Axboe 1061da177e4SLinus Torvalds /* 1071da177e4SLinus Torvalds * Keep out-of-line to prevent kernel bloat. 1081da177e4SLinus Torvalds * __builtin_return_address is not used because it is not always 1091da177e4SLinus Torvalds * reliable. 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds 1121da177e4SLinus Torvalds /** 1131da177e4SLinus Torvalds * skb_over_panic - private function 1141da177e4SLinus Torvalds * @skb: buffer 1151da177e4SLinus Torvalds * @sz: size 1161da177e4SLinus Torvalds * @here: address 1171da177e4SLinus Torvalds * 1181da177e4SLinus Torvalds * Out of line support code for skb_put(). Not user callable. 1191da177e4SLinus Torvalds */ 120ccb7c773SRami Rosen static void skb_over_panic(struct sk_buff *skb, int sz, void *here) 1211da177e4SLinus Torvalds { 12226095455SPatrick McHardy printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " 1234305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 12427a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1254305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 12626095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1271da177e4SLinus Torvalds BUG(); 1281da177e4SLinus Torvalds } 1291da177e4SLinus Torvalds 1301da177e4SLinus Torvalds /** 1311da177e4SLinus Torvalds * skb_under_panic - private function 1321da177e4SLinus Torvalds * @skb: buffer 1331da177e4SLinus Torvalds * @sz: size 1341da177e4SLinus Torvalds * @here: address 1351da177e4SLinus Torvalds * 1361da177e4SLinus Torvalds * Out of line support code for skb_push(). Not user callable. 1371da177e4SLinus Torvalds */ 1381da177e4SLinus Torvalds 139ccb7c773SRami Rosen static void skb_under_panic(struct sk_buff *skb, int sz, void *here) 1401da177e4SLinus Torvalds { 14126095455SPatrick McHardy printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " 1424305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 14327a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1444305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 14526095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1461da177e4SLinus Torvalds BUG(); 1471da177e4SLinus Torvalds } 1481da177e4SLinus Torvalds 1491da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1501da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1511da177e4SLinus Torvalds * [BEEP] leaks. 1521da177e4SLinus Torvalds * 1531da177e4SLinus Torvalds */ 1541da177e4SLinus Torvalds 1551da177e4SLinus Torvalds /** 156d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1571da177e4SLinus Torvalds * @size: size to allocate 1581da177e4SLinus Torvalds * @gfp_mask: allocation mask 159c83c2486SRandy Dunlap * @fclone: allocate from fclone cache instead of head cache 160c83c2486SRandy Dunlap * and allocate a cloned (child) skb 161b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1621da177e4SLinus Torvalds * 1631da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 1641da177e4SLinus Torvalds * tail room of size bytes. The object has a reference count of one. 1651da177e4SLinus Torvalds * The return is the buffer. On a failure the return is %NULL. 1661da177e4SLinus Torvalds * 1671da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1681da177e4SLinus Torvalds * %GFP_ATOMIC. 1691da177e4SLinus Torvalds */ 170dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 171b30973f8SChristoph Hellwig int fclone, int node) 1721da177e4SLinus Torvalds { 173e18b890bSChristoph Lameter struct kmem_cache *cache; 1744947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1751da177e4SLinus Torvalds struct sk_buff *skb; 1761da177e4SLinus Torvalds u8 *data; 1771da177e4SLinus Torvalds 1788798b3fbSHerbert Xu cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; 1798798b3fbSHerbert Xu 1801da177e4SLinus Torvalds /* Get the HEAD */ 181b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1821da177e4SLinus Torvalds if (!skb) 1831da177e4SLinus Torvalds goto out; 184ec7d2f2cSEric Dumazet prefetchw(skb); 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 187b30973f8SChristoph Hellwig data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), 188b30973f8SChristoph Hellwig gfp_mask, node); 1891da177e4SLinus Torvalds if (!data) 1901da177e4SLinus Torvalds goto nodata; 191ec7d2f2cSEric Dumazet prefetchw(data + size); 1921da177e4SLinus Torvalds 193ca0605a7SArnaldo Carvalho de Melo /* 194c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 195c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 196c8005785SJohannes Berg * the tail pointer in struct sk_buff! 197ca0605a7SArnaldo Carvalho de Melo */ 198ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 1991da177e4SLinus Torvalds skb->truesize = size + sizeof(struct sk_buff); 2001da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2011da177e4SLinus Torvalds skb->head = data; 2021da177e4SLinus Torvalds skb->data = data; 20327a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2044305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 20519633e12SStephen Hemminger #ifdef NET_SKBUFF_DATA_USES_OFFSET 20619633e12SStephen Hemminger skb->mac_header = ~0U; 20719633e12SStephen Hemminger #endif 20819633e12SStephen Hemminger 2094947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2104947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 211ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2124947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 213c2aa3665SEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 2144947d3efSBenjamin LaHaise 215d179cd12SDavid S. Miller if (fclone) { 216d179cd12SDavid S. Miller struct sk_buff *child = skb + 1; 217d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (child + 1); 2181da177e4SLinus Torvalds 219fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags1); 220fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags2); 221d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 222d179cd12SDavid S. Miller atomic_set(fclone_ref, 1); 223d179cd12SDavid S. Miller 224d179cd12SDavid S. Miller child->fclone = SKB_FCLONE_UNAVAILABLE; 225d179cd12SDavid S. Miller } 2261da177e4SLinus Torvalds out: 2271da177e4SLinus Torvalds return skb; 2281da177e4SLinus Torvalds nodata: 2298798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2301da177e4SLinus Torvalds skb = NULL; 2311da177e4SLinus Torvalds goto out; 2321da177e4SLinus Torvalds } 233b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2341da177e4SLinus Torvalds 2351da177e4SLinus Torvalds /** 2368af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 2378af27456SChristoph Hellwig * @dev: network device to receive on 2388af27456SChristoph Hellwig * @length: length to allocate 2398af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 2408af27456SChristoph Hellwig * 2418af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 2428af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 2438af27456SChristoph Hellwig * the headroom they think they need without accounting for the 2448af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 2458af27456SChristoph Hellwig * 2468af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 2478af27456SChristoph Hellwig */ 2488af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 2498af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 2508af27456SChristoph Hellwig { 2518af27456SChristoph Hellwig struct sk_buff *skb; 2528af27456SChristoph Hellwig 253564824b0SEric Dumazet skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE); 2547b2e497aSChristoph Hellwig if (likely(skb)) { 2558af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 2567b2e497aSChristoph Hellwig skb->dev = dev; 2577b2e497aSChristoph Hellwig } 2588af27456SChristoph Hellwig return skb; 2598af27456SChristoph Hellwig } 260b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 2611da177e4SLinus Torvalds 262654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 263654bed16SPeter Zijlstra int size) 264654bed16SPeter Zijlstra { 265654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 266654bed16SPeter Zijlstra skb->len += size; 267654bed16SPeter Zijlstra skb->data_len += size; 268654bed16SPeter Zijlstra skb->truesize += size; 269654bed16SPeter Zijlstra } 270654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 271654bed16SPeter Zijlstra 272f58518e6SIlpo Järvinen /** 273f58518e6SIlpo Järvinen * dev_alloc_skb - allocate an skbuff for receiving 274f58518e6SIlpo Järvinen * @length: length to allocate 275f58518e6SIlpo Järvinen * 276f58518e6SIlpo Järvinen * Allocate a new &sk_buff and assign it a usage count of one. The 277f58518e6SIlpo Järvinen * buffer has unspecified headroom built in. Users should allocate 278f58518e6SIlpo Järvinen * the headroom they think they need without accounting for the 279f58518e6SIlpo Järvinen * built in space. The built in space is used for optimisations. 280f58518e6SIlpo Järvinen * 281f58518e6SIlpo Järvinen * %NULL is returned if there is no free memory. Although this function 282f58518e6SIlpo Järvinen * allocates memory it can be called from an interrupt. 283f58518e6SIlpo Järvinen */ 284f58518e6SIlpo Järvinen struct sk_buff *dev_alloc_skb(unsigned int length) 285f58518e6SIlpo Järvinen { 2861483b874SDenys Vlasenko /* 2871483b874SDenys Vlasenko * There is more code here than it seems: 288a0f55e0eSDavid S. Miller * __dev_alloc_skb is an inline 2891483b874SDenys Vlasenko */ 290f58518e6SIlpo Järvinen return __dev_alloc_skb(length, GFP_ATOMIC); 291f58518e6SIlpo Järvinen } 292f58518e6SIlpo Järvinen EXPORT_SYMBOL(dev_alloc_skb); 293f58518e6SIlpo Järvinen 29427b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 2951da177e4SLinus Torvalds { 29627b437c8SHerbert Xu struct sk_buff *list = *listp; 2971da177e4SLinus Torvalds 29827b437c8SHerbert Xu *listp = NULL; 2991da177e4SLinus Torvalds 3001da177e4SLinus Torvalds do { 3011da177e4SLinus Torvalds struct sk_buff *this = list; 3021da177e4SLinus Torvalds list = list->next; 3031da177e4SLinus Torvalds kfree_skb(this); 3041da177e4SLinus Torvalds } while (list); 3051da177e4SLinus Torvalds } 3061da177e4SLinus Torvalds 30727b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 30827b437c8SHerbert Xu { 30927b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 31027b437c8SHerbert Xu } 31127b437c8SHerbert Xu 3121da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 3131da177e4SLinus Torvalds { 3141da177e4SLinus Torvalds struct sk_buff *list; 3151da177e4SLinus Torvalds 316fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 3171da177e4SLinus Torvalds skb_get(list); 3181da177e4SLinus Torvalds } 3191da177e4SLinus Torvalds 3205bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 3211da177e4SLinus Torvalds { 3221da177e4SLinus Torvalds if (!skb->cloned || 3231da177e4SLinus Torvalds !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 3241da177e4SLinus Torvalds &skb_shinfo(skb)->dataref)) { 3251da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 3261da177e4SLinus Torvalds int i; 3271da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 3281da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 33121dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 3321da177e4SLinus Torvalds skb_drop_fraglist(skb); 3331da177e4SLinus Torvalds 3341da177e4SLinus Torvalds kfree(skb->head); 3351da177e4SLinus Torvalds } 3361da177e4SLinus Torvalds } 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds /* 3391da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 3401da177e4SLinus Torvalds */ 3412d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 3421da177e4SLinus Torvalds { 343d179cd12SDavid S. Miller struct sk_buff *other; 344d179cd12SDavid S. Miller atomic_t *fclone_ref; 345d179cd12SDavid S. Miller 346d179cd12SDavid S. Miller switch (skb->fclone) { 347d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 3481da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 349d179cd12SDavid S. Miller break; 350d179cd12SDavid S. Miller 351d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 352d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 2); 353d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 354d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, skb); 355d179cd12SDavid S. Miller break; 356d179cd12SDavid S. Miller 357d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 358d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 1); 359d179cd12SDavid S. Miller other = skb - 1; 360d179cd12SDavid S. Miller 361d179cd12SDavid S. Miller /* The clone portion is available for 362d179cd12SDavid S. Miller * fast-cloning again. 363d179cd12SDavid S. Miller */ 364d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_UNAVAILABLE; 365d179cd12SDavid S. Miller 366d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 367d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, other); 368d179cd12SDavid S. Miller break; 3693ff50b79SStephen Hemminger } 3701da177e4SLinus Torvalds } 3711da177e4SLinus Torvalds 37204a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 3731da177e4SLinus Torvalds { 374adf30907SEric Dumazet skb_dst_drop(skb); 3751da177e4SLinus Torvalds #ifdef CONFIG_XFRM 3761da177e4SLinus Torvalds secpath_put(skb->sp); 3771da177e4SLinus Torvalds #endif 3781da177e4SLinus Torvalds if (skb->destructor) { 3799c2b3328SStephen Hemminger WARN_ON(in_irq()); 3801da177e4SLinus Torvalds skb->destructor(skb); 3811da177e4SLinus Torvalds } 3829fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3835f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 3842fc72c7bSKOVACS Krisztian #endif 3852fc72c7bSKOVACS Krisztian #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 3869fb9cbb1SYasuyuki Kozakai nf_conntrack_put_reasm(skb->nfct_reasm); 3879fb9cbb1SYasuyuki Kozakai #endif 3881da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 3891da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 3901da177e4SLinus Torvalds #endif 3911da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 3921da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 3931da177e4SLinus Torvalds skb->tc_index = 0; 3941da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 3951da177e4SLinus Torvalds skb->tc_verd = 0; 3961da177e4SLinus Torvalds #endif 3971da177e4SLinus Torvalds #endif 39804a4bb55SLennert Buytenhek } 39904a4bb55SLennert Buytenhek 40004a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 40104a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 40204a4bb55SLennert Buytenhek { 40304a4bb55SLennert Buytenhek skb_release_head_state(skb); 4042d4baff8SHerbert Xu skb_release_data(skb); 4052d4baff8SHerbert Xu } 4061da177e4SLinus Torvalds 4072d4baff8SHerbert Xu /** 4082d4baff8SHerbert Xu * __kfree_skb - private function 4092d4baff8SHerbert Xu * @skb: buffer 4102d4baff8SHerbert Xu * 4112d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 4122d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 4132d4baff8SHerbert Xu * always call kfree_skb 4142d4baff8SHerbert Xu */ 4152d4baff8SHerbert Xu 4162d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 4172d4baff8SHerbert Xu { 4182d4baff8SHerbert Xu skb_release_all(skb); 4191da177e4SLinus Torvalds kfree_skbmem(skb); 4201da177e4SLinus Torvalds } 421b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 4221da177e4SLinus Torvalds 4231da177e4SLinus Torvalds /** 424231d06aeSJörn Engel * kfree_skb - free an sk_buff 425231d06aeSJörn Engel * @skb: buffer to free 426231d06aeSJörn Engel * 427231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 428231d06aeSJörn Engel * hit zero. 429231d06aeSJörn Engel */ 430231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 431231d06aeSJörn Engel { 432231d06aeSJörn Engel if (unlikely(!skb)) 433231d06aeSJörn Engel return; 434231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 435231d06aeSJörn Engel smp_rmb(); 436231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 437231d06aeSJörn Engel return; 438ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 439231d06aeSJörn Engel __kfree_skb(skb); 440231d06aeSJörn Engel } 441b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 442231d06aeSJörn Engel 443d1a203eaSStephen Hemminger /** 444ead2ceb0SNeil Horman * consume_skb - free an skbuff 445ead2ceb0SNeil Horman * @skb: buffer to free 446ead2ceb0SNeil Horman * 447ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 448ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 449ead2ceb0SNeil Horman * is being dropped after a failure and notes that 450ead2ceb0SNeil Horman */ 451ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 452ead2ceb0SNeil Horman { 453ead2ceb0SNeil Horman if (unlikely(!skb)) 454ead2ceb0SNeil Horman return; 455ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 456ead2ceb0SNeil Horman smp_rmb(); 457ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 458ead2ceb0SNeil Horman return; 45907dc22e7SKoki Sanagi trace_consume_skb(skb); 460ead2ceb0SNeil Horman __kfree_skb(skb); 461ead2ceb0SNeil Horman } 462ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 463ead2ceb0SNeil Horman 464ead2ceb0SNeil Horman /** 465d1a203eaSStephen Hemminger * skb_recycle_check - check if skb can be reused for receive 466d1a203eaSStephen Hemminger * @skb: buffer 467d1a203eaSStephen Hemminger * @skb_size: minimum receive buffer size 468d1a203eaSStephen Hemminger * 469d1a203eaSStephen Hemminger * Checks that the skb passed in is not shared or cloned, and 470d1a203eaSStephen Hemminger * that it is linear and its head portion at least as large as 471d1a203eaSStephen Hemminger * skb_size so that it can be recycled as a receive buffer. 472d1a203eaSStephen Hemminger * If these conditions are met, this function does any necessary 473d1a203eaSStephen Hemminger * reference count dropping and cleans up the skbuff as if it 474d1a203eaSStephen Hemminger * just came from __alloc_skb(). 475d1a203eaSStephen Hemminger */ 4765b0daa34SChangli Gao bool skb_recycle_check(struct sk_buff *skb, int skb_size) 47704a4bb55SLennert Buytenhek { 47804a4bb55SLennert Buytenhek struct skb_shared_info *shinfo; 47904a4bb55SLennert Buytenhek 480e84af6ddSAnton Vorontsov if (irqs_disabled()) 4815b0daa34SChangli Gao return false; 482e84af6ddSAnton Vorontsov 48304a4bb55SLennert Buytenhek if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE) 4845b0daa34SChangli Gao return false; 48504a4bb55SLennert Buytenhek 48604a4bb55SLennert Buytenhek skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD); 48704a4bb55SLennert Buytenhek if (skb_end_pointer(skb) - skb->head < skb_size) 4885b0daa34SChangli Gao return false; 48904a4bb55SLennert Buytenhek 49004a4bb55SLennert Buytenhek if (skb_shared(skb) || skb_cloned(skb)) 4915b0daa34SChangli Gao return false; 49204a4bb55SLennert Buytenhek 49304a4bb55SLennert Buytenhek skb_release_head_state(skb); 494ec7d2f2cSEric Dumazet 49504a4bb55SLennert Buytenhek shinfo = skb_shinfo(skb); 496ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 49704a4bb55SLennert Buytenhek atomic_set(&shinfo->dataref, 1); 49804a4bb55SLennert Buytenhek 49904a4bb55SLennert Buytenhek memset(skb, 0, offsetof(struct sk_buff, tail)); 50004a4bb55SLennert Buytenhek skb->data = skb->head + NET_SKB_PAD; 5015cd33db2SLennert Buytenhek skb_reset_tail_pointer(skb); 50204a4bb55SLennert Buytenhek 5035b0daa34SChangli Gao return true; 50404a4bb55SLennert Buytenhek } 50504a4bb55SLennert Buytenhek EXPORT_SYMBOL(skb_recycle_check); 50604a4bb55SLennert Buytenhek 507dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 508dec18810SHerbert Xu { 509dec18810SHerbert Xu new->tstamp = old->tstamp; 510dec18810SHerbert Xu new->dev = old->dev; 511dec18810SHerbert Xu new->transport_header = old->transport_header; 512dec18810SHerbert Xu new->network_header = old->network_header; 513dec18810SHerbert Xu new->mac_header = old->mac_header; 5147fee226aSEric Dumazet skb_dst_copy(new, old); 5150a9627f2STom Herbert new->rxhash = old->rxhash; 516def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 517dec18810SHerbert Xu new->sp = secpath_get(old->sp); 518dec18810SHerbert Xu #endif 519dec18810SHerbert Xu memcpy(new->cb, old->cb, sizeof(old->cb)); 5209bcb97caSHerbert Xu new->csum = old->csum; 521dec18810SHerbert Xu new->local_df = old->local_df; 522dec18810SHerbert Xu new->pkt_type = old->pkt_type; 523dec18810SHerbert Xu new->ip_summed = old->ip_summed; 524dec18810SHerbert Xu skb_copy_queue_mapping(new, old); 525dec18810SHerbert Xu new->priority = old->priority; 526dec18810SHerbert Xu #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 527dec18810SHerbert Xu new->ipvs_property = old->ipvs_property; 528dec18810SHerbert Xu #endif 529dec18810SHerbert Xu new->protocol = old->protocol; 530dec18810SHerbert Xu new->mark = old->mark; 5318964be4aSEric Dumazet new->skb_iif = old->skb_iif; 532dec18810SHerbert Xu __nf_copy(new, old); 533dec18810SHerbert Xu #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ 534dec18810SHerbert Xu defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) 535dec18810SHerbert Xu new->nf_trace = old->nf_trace; 536dec18810SHerbert Xu #endif 537dec18810SHerbert Xu #ifdef CONFIG_NET_SCHED 538dec18810SHerbert Xu new->tc_index = old->tc_index; 539dec18810SHerbert Xu #ifdef CONFIG_NET_CLS_ACT 540dec18810SHerbert Xu new->tc_verd = old->tc_verd; 541dec18810SHerbert Xu #endif 542dec18810SHerbert Xu #endif 5436aa895b0SPatrick McHardy new->vlan_tci = old->vlan_tci; 5446aa895b0SPatrick McHardy 545dec18810SHerbert Xu skb_copy_secmark(new, old); 546dec18810SHerbert Xu } 547dec18810SHerbert Xu 54882c49a35SHerbert Xu /* 54982c49a35SHerbert Xu * You should not add any new code to this function. Add it to 55082c49a35SHerbert Xu * __copy_skb_header above instead. 55182c49a35SHerbert Xu */ 552e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 5531da177e4SLinus Torvalds { 5541da177e4SLinus Torvalds #define C(x) n->x = skb->x 5551da177e4SLinus Torvalds 5561da177e4SLinus Torvalds n->next = n->prev = NULL; 5571da177e4SLinus Torvalds n->sk = NULL; 558dec18810SHerbert Xu __copy_skb_header(n, skb); 559dec18810SHerbert Xu 5601da177e4SLinus Torvalds C(len); 5611da177e4SLinus Torvalds C(data_len); 5623e6b3b2eSAlexey Dobriyan C(mac_len); 563334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 56402f1c89dSPaul Moore n->cloned = 1; 5651da177e4SLinus Torvalds n->nohdr = 0; 5661da177e4SLinus Torvalds n->destructor = NULL; 5671da177e4SLinus Torvalds C(tail); 5681da177e4SLinus Torvalds C(end); 56902f1c89dSPaul Moore C(head); 57002f1c89dSPaul Moore C(data); 57102f1c89dSPaul Moore C(truesize); 57202f1c89dSPaul Moore atomic_set(&n->users, 1); 5731da177e4SLinus Torvalds 5741da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 5751da177e4SLinus Torvalds skb->cloned = 1; 5761da177e4SLinus Torvalds 5771da177e4SLinus Torvalds return n; 578e0053ec0SHerbert Xu #undef C 579e0053ec0SHerbert Xu } 580e0053ec0SHerbert Xu 581e0053ec0SHerbert Xu /** 582e0053ec0SHerbert Xu * skb_morph - morph one skb into another 583e0053ec0SHerbert Xu * @dst: the skb to receive the contents 584e0053ec0SHerbert Xu * @src: the skb to supply the contents 585e0053ec0SHerbert Xu * 586e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 587e0053ec0SHerbert Xu * supplied by the user. 588e0053ec0SHerbert Xu * 589e0053ec0SHerbert Xu * The target skb is returned upon exit. 590e0053ec0SHerbert Xu */ 591e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 592e0053ec0SHerbert Xu { 5932d4baff8SHerbert Xu skb_release_all(dst); 594e0053ec0SHerbert Xu return __skb_clone(dst, src); 595e0053ec0SHerbert Xu } 596e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 597e0053ec0SHerbert Xu 598e0053ec0SHerbert Xu /** 599e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 600e0053ec0SHerbert Xu * @skb: buffer to clone 601e0053ec0SHerbert Xu * @gfp_mask: allocation priority 602e0053ec0SHerbert Xu * 603e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 604e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 605e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 606e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 607e0053ec0SHerbert Xu * 608e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 609e0053ec0SHerbert Xu * %GFP_ATOMIC. 610e0053ec0SHerbert Xu */ 611e0053ec0SHerbert Xu 612e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 613e0053ec0SHerbert Xu { 614e0053ec0SHerbert Xu struct sk_buff *n; 615e0053ec0SHerbert Xu 616e0053ec0SHerbert Xu n = skb + 1; 617e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 618e0053ec0SHerbert Xu n->fclone == SKB_FCLONE_UNAVAILABLE) { 619e0053ec0SHerbert Xu atomic_t *fclone_ref = (atomic_t *) (n + 1); 620e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_CLONE; 621e0053ec0SHerbert Xu atomic_inc(fclone_ref); 622e0053ec0SHerbert Xu } else { 623e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 624e0053ec0SHerbert Xu if (!n) 625e0053ec0SHerbert Xu return NULL; 626fe55f6d5SVegard Nossum 627fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 628fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags2); 629e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 630e0053ec0SHerbert Xu } 631e0053ec0SHerbert Xu 632e0053ec0SHerbert Xu return __skb_clone(n, skb); 6331da177e4SLinus Torvalds } 634b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 6351da177e4SLinus Torvalds 6361da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 6371da177e4SLinus Torvalds { 6382e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 6391da177e4SLinus Torvalds /* 6401da177e4SLinus Torvalds * Shift between the two data areas in bytes 6411da177e4SLinus Torvalds */ 6421da177e4SLinus Torvalds unsigned long offset = new->data - old->data; 6432e07fa9cSArnaldo Carvalho de Melo #endif 644dec18810SHerbert Xu 645dec18810SHerbert Xu __copy_skb_header(new, old); 646dec18810SHerbert Xu 6472e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 6482e07fa9cSArnaldo Carvalho de Melo /* {transport,network,mac}_header are relative to skb->head */ 6492e07fa9cSArnaldo Carvalho de Melo new->transport_header += offset; 6502e07fa9cSArnaldo Carvalho de Melo new->network_header += offset; 651603a8bbeSStephen Hemminger if (skb_mac_header_was_set(new)) 6522e07fa9cSArnaldo Carvalho de Melo new->mac_header += offset; 6532e07fa9cSArnaldo Carvalho de Melo #endif 6547967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 6557967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 6567967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 6571da177e4SLinus Torvalds } 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds /** 6601da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 6611da177e4SLinus Torvalds * @skb: buffer to copy 6621da177e4SLinus Torvalds * @gfp_mask: allocation priority 6631da177e4SLinus Torvalds * 6641da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 6651da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 6661da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 6671da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 6681da177e4SLinus Torvalds * 6691da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 6701da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 6711da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 6721da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 6731da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 6741da177e4SLinus Torvalds */ 6751da177e4SLinus Torvalds 676dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 6771da177e4SLinus Torvalds { 6786602cebbSEric Dumazet int headerlen = skb_headroom(skb); 6796602cebbSEric Dumazet unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len; 6806602cebbSEric Dumazet struct sk_buff *n = alloc_skb(size, gfp_mask); 6816602cebbSEric Dumazet 6821da177e4SLinus Torvalds if (!n) 6831da177e4SLinus Torvalds return NULL; 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds /* Set the data pointer */ 6861da177e4SLinus Torvalds skb_reserve(n, headerlen); 6871da177e4SLinus Torvalds /* Set the tail pointer and length */ 6881da177e4SLinus Torvalds skb_put(n, skb->len); 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 6911da177e4SLinus Torvalds BUG(); 6921da177e4SLinus Torvalds 6931da177e4SLinus Torvalds copy_skb_header(n, skb); 6941da177e4SLinus Torvalds return n; 6951da177e4SLinus Torvalds } 696b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds /** 6991da177e4SLinus Torvalds * pskb_copy - create copy of an sk_buff with private head. 7001da177e4SLinus Torvalds * @skb: buffer to copy 7011da177e4SLinus Torvalds * @gfp_mask: allocation priority 7021da177e4SLinus Torvalds * 7031da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 7041da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 7051da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 7061da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 7071da177e4SLinus Torvalds * or the pointer to the buffer on success. 7081da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 7091da177e4SLinus Torvalds */ 7101da177e4SLinus Torvalds 711dd0fc66fSAl Viro struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask) 7121da177e4SLinus Torvalds { 7136602cebbSEric Dumazet unsigned int size = skb_end_pointer(skb) - skb->head; 7146602cebbSEric Dumazet struct sk_buff *n = alloc_skb(size, gfp_mask); 7156602cebbSEric Dumazet 7161da177e4SLinus Torvalds if (!n) 7171da177e4SLinus Torvalds goto out; 7181da177e4SLinus Torvalds 7191da177e4SLinus Torvalds /* Set the data pointer */ 7206602cebbSEric Dumazet skb_reserve(n, skb_headroom(skb)); 7211da177e4SLinus Torvalds /* Set the tail pointer and length */ 7221da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 7231da177e4SLinus Torvalds /* Copy the bytes */ 724d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 7251da177e4SLinus Torvalds 72625f484a6SHerbert Xu n->truesize += skb->data_len; 7271da177e4SLinus Torvalds n->data_len = skb->data_len; 7281da177e4SLinus Torvalds n->len = skb->len; 7291da177e4SLinus Torvalds 7301da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 7311da177e4SLinus Torvalds int i; 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 7341da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 7351da177e4SLinus Torvalds get_page(skb_shinfo(n)->frags[i].page); 7361da177e4SLinus Torvalds } 7371da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 7381da177e4SLinus Torvalds } 7391da177e4SLinus Torvalds 74021dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 7411da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 7421da177e4SLinus Torvalds skb_clone_fraglist(n); 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds 7451da177e4SLinus Torvalds copy_skb_header(n, skb); 7461da177e4SLinus Torvalds out: 7471da177e4SLinus Torvalds return n; 7481da177e4SLinus Torvalds } 749b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_copy); 7501da177e4SLinus Torvalds 7511da177e4SLinus Torvalds /** 7521da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 7531da177e4SLinus Torvalds * @skb: buffer to reallocate 7541da177e4SLinus Torvalds * @nhead: room to add at head 7551da177e4SLinus Torvalds * @ntail: room to add at tail 7561da177e4SLinus Torvalds * @gfp_mask: allocation priority 7571da177e4SLinus Torvalds * 7581da177e4SLinus Torvalds * Expands (or creates identical copy, if &nhead and &ntail are zero) 7591da177e4SLinus Torvalds * header of skb. &sk_buff itself is not changed. &sk_buff MUST have 7601da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 7611da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 7621da177e4SLinus Torvalds * 7631da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 7641da177e4SLinus Torvalds * reloaded after call to this function. 7651da177e4SLinus Torvalds */ 7661da177e4SLinus Torvalds 76786a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 768dd0fc66fSAl Viro gfp_t gfp_mask) 7691da177e4SLinus Torvalds { 7701da177e4SLinus Torvalds int i; 7711da177e4SLinus Torvalds u8 *data; 7726602cebbSEric Dumazet int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail; 7731da177e4SLinus Torvalds long off; 7741fd63041SEric Dumazet bool fastpath; 7751da177e4SLinus Torvalds 7764edd87adSHerbert Xu BUG_ON(nhead < 0); 7774edd87adSHerbert Xu 7781da177e4SLinus Torvalds if (skb_shared(skb)) 7791da177e4SLinus Torvalds BUG(); 7801da177e4SLinus Torvalds 7811da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 7821da177e4SLinus Torvalds 783ca44ac38SChangli Gao /* Check if we can avoid taking references on fragments if we own 784ca44ac38SChangli Gao * the last reference on skb->head. (see skb_release_data()) 785ca44ac38SChangli Gao */ 786ca44ac38SChangli Gao if (!skb->cloned) 787ca44ac38SChangli Gao fastpath = true; 788ca44ac38SChangli Gao else { 789ca44ac38SChangli Gao int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 790ca44ac38SChangli Gao 791ca44ac38SChangli Gao fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta; 792ca44ac38SChangli Gao } 793ca44ac38SChangli Gao 794ca44ac38SChangli Gao if (fastpath && 795ca44ac38SChangli Gao size + sizeof(struct skb_shared_info) <= ksize(skb->head)) { 796ca44ac38SChangli Gao memmove(skb->head + size, skb_shinfo(skb), 797ca44ac38SChangli Gao offsetof(struct skb_shared_info, 798ca44ac38SChangli Gao frags[skb_shinfo(skb)->nr_frags])); 799ca44ac38SChangli Gao memmove(skb->head + nhead, skb->head, 800ca44ac38SChangli Gao skb_tail_pointer(skb) - skb->head); 801ca44ac38SChangli Gao off = nhead; 802ca44ac38SChangli Gao goto adjust_others; 803ca44ac38SChangli Gao } 804ca44ac38SChangli Gao 8051da177e4SLinus Torvalds data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); 8061da177e4SLinus Torvalds if (!data) 8071da177e4SLinus Torvalds goto nodata; 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 8106602cebbSEric Dumazet * optimized for the cases when header is void. 8116602cebbSEric Dumazet */ 8126602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 8136602cebbSEric Dumazet 8146602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 8156602cebbSEric Dumazet skb_shinfo(skb), 816fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 8171da177e4SLinus Torvalds 8181fd63041SEric Dumazet if (fastpath) { 8191fd63041SEric Dumazet kfree(skb->head); 8201fd63041SEric Dumazet } else { 8211da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 8221da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 8231da177e4SLinus Torvalds 82421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 8251da177e4SLinus Torvalds skb_clone_fraglist(skb); 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds skb_release_data(skb); 8281fd63041SEric Dumazet } 8291da177e4SLinus Torvalds off = (data + nhead) - skb->head; 8301da177e4SLinus Torvalds 8311da177e4SLinus Torvalds skb->head = data; 832ca44ac38SChangli Gao adjust_others: 8331da177e4SLinus Torvalds skb->data += off; 8344305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 8354305b541SArnaldo Carvalho de Melo skb->end = size; 83656eb8882SPatrick McHardy off = nhead; 8374305b541SArnaldo Carvalho de Melo #else 8384305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 83956eb8882SPatrick McHardy #endif 84027a884dcSArnaldo Carvalho de Melo /* {transport,network,mac}_header and tail are relative to skb->head */ 84127a884dcSArnaldo Carvalho de Melo skb->tail += off; 842b0e380b1SArnaldo Carvalho de Melo skb->transport_header += off; 843b0e380b1SArnaldo Carvalho de Melo skb->network_header += off; 844603a8bbeSStephen Hemminger if (skb_mac_header_was_set(skb)) 845b0e380b1SArnaldo Carvalho de Melo skb->mac_header += off; 84600c5a983SAndrea Shepard /* Only adjust this if it actually is csum_start rather than csum */ 84700c5a983SAndrea Shepard if (skb->ip_summed == CHECKSUM_PARTIAL) 848172a863fSHerbert Xu skb->csum_start += nhead; 8491da177e4SLinus Torvalds skb->cloned = 0; 850334a8132SPatrick McHardy skb->hdr_len = 0; 8511da177e4SLinus Torvalds skb->nohdr = 0; 8521da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 8531da177e4SLinus Torvalds return 0; 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds nodata: 8561da177e4SLinus Torvalds return -ENOMEM; 8571da177e4SLinus Torvalds } 858b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 8611da177e4SLinus Torvalds 8621da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 8631da177e4SLinus Torvalds { 8641da177e4SLinus Torvalds struct sk_buff *skb2; 8651da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 8661da177e4SLinus Torvalds 8671da177e4SLinus Torvalds if (delta <= 0) 8681da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 8691da177e4SLinus Torvalds else { 8701da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 8711da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 8721da177e4SLinus Torvalds GFP_ATOMIC)) { 8731da177e4SLinus Torvalds kfree_skb(skb2); 8741da177e4SLinus Torvalds skb2 = NULL; 8751da177e4SLinus Torvalds } 8761da177e4SLinus Torvalds } 8771da177e4SLinus Torvalds return skb2; 8781da177e4SLinus Torvalds } 879b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 8801da177e4SLinus Torvalds 8811da177e4SLinus Torvalds /** 8821da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 8831da177e4SLinus Torvalds * @skb: buffer to copy 8841da177e4SLinus Torvalds * @newheadroom: new free bytes at head 8851da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 8861da177e4SLinus Torvalds * @gfp_mask: allocation priority 8871da177e4SLinus Torvalds * 8881da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 8891da177e4SLinus Torvalds * allocate additional space. 8901da177e4SLinus Torvalds * 8911da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 8921da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 8931da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 8941da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 8951da177e4SLinus Torvalds * 8961da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 8971da177e4SLinus Torvalds * is called from an interrupt. 8981da177e4SLinus Torvalds */ 8991da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 90086a76cafSVictor Fusco int newheadroom, int newtailroom, 901dd0fc66fSAl Viro gfp_t gfp_mask) 9021da177e4SLinus Torvalds { 9031da177e4SLinus Torvalds /* 9041da177e4SLinus Torvalds * Allocate the copy buffer 9051da177e4SLinus Torvalds */ 9061da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom, 9071da177e4SLinus Torvalds gfp_mask); 908efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 9091da177e4SLinus Torvalds int head_copy_len, head_copy_off; 91052886051SHerbert Xu int off; 9111da177e4SLinus Torvalds 9121da177e4SLinus Torvalds if (!n) 9131da177e4SLinus Torvalds return NULL; 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds skb_reserve(n, newheadroom); 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds /* Set the tail pointer and length */ 9181da177e4SLinus Torvalds skb_put(n, skb->len); 9191da177e4SLinus Torvalds 920efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 9211da177e4SLinus Torvalds head_copy_off = 0; 9221da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 9231da177e4SLinus Torvalds head_copy_len = newheadroom; 9241da177e4SLinus Torvalds else 9251da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds /* Copy the linear header and data. */ 9281da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 9291da177e4SLinus Torvalds skb->len + head_copy_len)) 9301da177e4SLinus Torvalds BUG(); 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds copy_skb_header(n, skb); 9331da177e4SLinus Torvalds 934efd1e8d5SPatrick McHardy off = newheadroom - oldheadroom; 935be2b6e62SDavid S. Miller if (n->ip_summed == CHECKSUM_PARTIAL) 93652886051SHerbert Xu n->csum_start += off; 93752886051SHerbert Xu #ifdef NET_SKBUFF_DATA_USES_OFFSET 938efd1e8d5SPatrick McHardy n->transport_header += off; 939efd1e8d5SPatrick McHardy n->network_header += off; 940603a8bbeSStephen Hemminger if (skb_mac_header_was_set(skb)) 941efd1e8d5SPatrick McHardy n->mac_header += off; 94252886051SHerbert Xu #endif 943efd1e8d5SPatrick McHardy 9441da177e4SLinus Torvalds return n; 9451da177e4SLinus Torvalds } 946b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 9471da177e4SLinus Torvalds 9481da177e4SLinus Torvalds /** 9491da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 9501da177e4SLinus Torvalds * @skb: buffer to pad 9511da177e4SLinus Torvalds * @pad: space to pad 9521da177e4SLinus Torvalds * 9531da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 9541da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 9551da177e4SLinus Torvalds * beyond the buffer end onto the wire. 9561da177e4SLinus Torvalds * 9575b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 9581da177e4SLinus Torvalds */ 9591da177e4SLinus Torvalds 9605b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 9611da177e4SLinus Torvalds { 9625b057c6bSHerbert Xu int err; 9635b057c6bSHerbert Xu int ntail; 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 9665b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 9671da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 9685b057c6bSHerbert Xu return 0; 9691da177e4SLinus Torvalds } 9701da177e4SLinus Torvalds 9714305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 9725b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 9735b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 9745b057c6bSHerbert Xu if (unlikely(err)) 9755b057c6bSHerbert Xu goto free_skb; 9765b057c6bSHerbert Xu } 9775b057c6bSHerbert Xu 9785b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 9795b057c6bSHerbert Xu * to be audited. 9805b057c6bSHerbert Xu */ 9815b057c6bSHerbert Xu err = skb_linearize(skb); 9825b057c6bSHerbert Xu if (unlikely(err)) 9835b057c6bSHerbert Xu goto free_skb; 9845b057c6bSHerbert Xu 9855b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 9865b057c6bSHerbert Xu return 0; 9875b057c6bSHerbert Xu 9885b057c6bSHerbert Xu free_skb: 9891da177e4SLinus Torvalds kfree_skb(skb); 9905b057c6bSHerbert Xu return err; 9911da177e4SLinus Torvalds } 992b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 9931da177e4SLinus Torvalds 9940dde3e16SIlpo Järvinen /** 9950dde3e16SIlpo Järvinen * skb_put - add data to a buffer 9960dde3e16SIlpo Järvinen * @skb: buffer to use 9970dde3e16SIlpo Järvinen * @len: amount of data to add 9980dde3e16SIlpo Järvinen * 9990dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 10000dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 10010dde3e16SIlpo Järvinen * first byte of the extra data is returned. 10020dde3e16SIlpo Järvinen */ 10030dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 10040dde3e16SIlpo Järvinen { 10050dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 10060dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 10070dde3e16SIlpo Järvinen skb->tail += len; 10080dde3e16SIlpo Järvinen skb->len += len; 10090dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 10100dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 10110dde3e16SIlpo Järvinen return tmp; 10120dde3e16SIlpo Järvinen } 10130dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 10140dde3e16SIlpo Järvinen 10156be8ac2fSIlpo Järvinen /** 1016c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1017c2aa270aSIlpo Järvinen * @skb: buffer to use 1018c2aa270aSIlpo Järvinen * @len: amount of data to add 1019c2aa270aSIlpo Järvinen * 1020c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1021c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1022c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1023c2aa270aSIlpo Järvinen */ 1024c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1025c2aa270aSIlpo Järvinen { 1026c2aa270aSIlpo Järvinen skb->data -= len; 1027c2aa270aSIlpo Järvinen skb->len += len; 1028c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1029c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1030c2aa270aSIlpo Järvinen return skb->data; 1031c2aa270aSIlpo Järvinen } 1032c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1033c2aa270aSIlpo Järvinen 1034c2aa270aSIlpo Järvinen /** 10356be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 10366be8ac2fSIlpo Järvinen * @skb: buffer to use 10376be8ac2fSIlpo Järvinen * @len: amount of data to remove 10386be8ac2fSIlpo Järvinen * 10396be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 10406be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 10416be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 10426be8ac2fSIlpo Järvinen * the old data. 10436be8ac2fSIlpo Järvinen */ 10446be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 10456be8ac2fSIlpo Järvinen { 104647d29646SDavid S. Miller return skb_pull_inline(skb, len); 10476be8ac2fSIlpo Järvinen } 10486be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 10496be8ac2fSIlpo Järvinen 1050419ae74eSIlpo Järvinen /** 1051419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1052419ae74eSIlpo Järvinen * @skb: buffer to alter 1053419ae74eSIlpo Järvinen * @len: new length 1054419ae74eSIlpo Järvinen * 1055419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1056419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1057419ae74eSIlpo Järvinen * The skb must be linear. 1058419ae74eSIlpo Järvinen */ 1059419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1060419ae74eSIlpo Järvinen { 1061419ae74eSIlpo Järvinen if (skb->len > len) 1062419ae74eSIlpo Järvinen __skb_trim(skb, len); 1063419ae74eSIlpo Järvinen } 1064419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1065419ae74eSIlpo Järvinen 10663cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 10671da177e4SLinus Torvalds */ 10681da177e4SLinus Torvalds 10693cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 10701da177e4SLinus Torvalds { 107127b437c8SHerbert Xu struct sk_buff **fragp; 107227b437c8SHerbert Xu struct sk_buff *frag; 10731da177e4SLinus Torvalds int offset = skb_headlen(skb); 10741da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 10751da177e4SLinus Torvalds int i; 107627b437c8SHerbert Xu int err; 107727b437c8SHerbert Xu 107827b437c8SHerbert Xu if (skb_cloned(skb) && 107927b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 108027b437c8SHerbert Xu return err; 10811da177e4SLinus Torvalds 1082f4d26fb3SHerbert Xu i = 0; 1083f4d26fb3SHerbert Xu if (offset >= len) 1084f4d26fb3SHerbert Xu goto drop_pages; 1085f4d26fb3SHerbert Xu 1086f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 10871da177e4SLinus Torvalds int end = offset + skb_shinfo(skb)->frags[i].size; 108827b437c8SHerbert Xu 108927b437c8SHerbert Xu if (end < len) { 10901da177e4SLinus Torvalds offset = end; 109127b437c8SHerbert Xu continue; 10921da177e4SLinus Torvalds } 10931da177e4SLinus Torvalds 109427b437c8SHerbert Xu skb_shinfo(skb)->frags[i++].size = len - offset; 109527b437c8SHerbert Xu 1096f4d26fb3SHerbert Xu drop_pages: 109727b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 109827b437c8SHerbert Xu 109927b437c8SHerbert Xu for (; i < nfrags; i++) 110027b437c8SHerbert Xu put_page(skb_shinfo(skb)->frags[i].page); 110127b437c8SHerbert Xu 110221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 110327b437c8SHerbert Xu skb_drop_fraglist(skb); 1104f4d26fb3SHerbert Xu goto done; 110527b437c8SHerbert Xu } 110627b437c8SHerbert Xu 110727b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 110827b437c8SHerbert Xu fragp = &frag->next) { 110927b437c8SHerbert Xu int end = offset + frag->len; 111027b437c8SHerbert Xu 111127b437c8SHerbert Xu if (skb_shared(frag)) { 111227b437c8SHerbert Xu struct sk_buff *nfrag; 111327b437c8SHerbert Xu 111427b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 111527b437c8SHerbert Xu if (unlikely(!nfrag)) 111627b437c8SHerbert Xu return -ENOMEM; 111727b437c8SHerbert Xu 111827b437c8SHerbert Xu nfrag->next = frag->next; 1119f4d26fb3SHerbert Xu kfree_skb(frag); 112027b437c8SHerbert Xu frag = nfrag; 112127b437c8SHerbert Xu *fragp = frag; 112227b437c8SHerbert Xu } 112327b437c8SHerbert Xu 112427b437c8SHerbert Xu if (end < len) { 112527b437c8SHerbert Xu offset = end; 112627b437c8SHerbert Xu continue; 112727b437c8SHerbert Xu } 112827b437c8SHerbert Xu 112927b437c8SHerbert Xu if (end > len && 113027b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 113127b437c8SHerbert Xu return err; 113227b437c8SHerbert Xu 113327b437c8SHerbert Xu if (frag->next) 113427b437c8SHerbert Xu skb_drop_list(&frag->next); 113527b437c8SHerbert Xu break; 113627b437c8SHerbert Xu } 113727b437c8SHerbert Xu 1138f4d26fb3SHerbert Xu done: 113927b437c8SHerbert Xu if (len > skb_headlen(skb)) { 11401da177e4SLinus Torvalds skb->data_len -= skb->len - len; 11411da177e4SLinus Torvalds skb->len = len; 11421da177e4SLinus Torvalds } else { 11431da177e4SLinus Torvalds skb->len = len; 11441da177e4SLinus Torvalds skb->data_len = 0; 114527a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 11461da177e4SLinus Torvalds } 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds return 0; 11491da177e4SLinus Torvalds } 1150b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds /** 11531da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 11541da177e4SLinus Torvalds * @skb: buffer to reallocate 11551da177e4SLinus Torvalds * @delta: number of bytes to advance tail 11561da177e4SLinus Torvalds * 11571da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 11581da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 11591da177e4SLinus Torvalds * data from fragmented part. 11601da177e4SLinus Torvalds * 11611da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 11621da177e4SLinus Torvalds * 11631da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 11641da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 11651da177e4SLinus Torvalds * 11661da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 11671da177e4SLinus Torvalds * reloaded after call to this function. 11681da177e4SLinus Torvalds */ 11691da177e4SLinus Torvalds 11701da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 11711da177e4SLinus Torvalds * when it is necessary. 11721da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 11731da177e4SLinus Torvalds * 2. It may change skb pointers. 11741da177e4SLinus Torvalds * 11751da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 11761da177e4SLinus Torvalds */ 11771da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 11781da177e4SLinus Torvalds { 11791da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 11801da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 11811da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 11821da177e4SLinus Torvalds */ 11834305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 11861da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 11871da177e4SLinus Torvalds GFP_ATOMIC)) 11881da177e4SLinus Torvalds return NULL; 11891da177e4SLinus Torvalds } 11901da177e4SLinus Torvalds 119127a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 11921da177e4SLinus Torvalds BUG(); 11931da177e4SLinus Torvalds 11941da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 11951da177e4SLinus Torvalds * size of pulled pages. Superb. 11961da177e4SLinus Torvalds */ 119721dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 11981da177e4SLinus Torvalds goto pull_pages; 11991da177e4SLinus Torvalds 12001da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 12011da177e4SLinus Torvalds eat = delta; 12021da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 12031da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size >= eat) 12041da177e4SLinus Torvalds goto pull_pages; 12051da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 12091da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 12101da177e4SLinus Torvalds * but taking into account that pulling is expected to 12111da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 12121da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 12131da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 12141da177e4SLinus Torvalds */ 12151da177e4SLinus Torvalds if (eat) { 12161da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 12171da177e4SLinus Torvalds struct sk_buff *clone = NULL; 12181da177e4SLinus Torvalds struct sk_buff *insp = NULL; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds do { 122109a62660SKris Katterjohn BUG_ON(!list); 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds if (list->len <= eat) { 12241da177e4SLinus Torvalds /* Eaten as whole. */ 12251da177e4SLinus Torvalds eat -= list->len; 12261da177e4SLinus Torvalds list = list->next; 12271da177e4SLinus Torvalds insp = list; 12281da177e4SLinus Torvalds } else { 12291da177e4SLinus Torvalds /* Eaten partially. */ 12301da177e4SLinus Torvalds 12311da177e4SLinus Torvalds if (skb_shared(list)) { 12321da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 12331da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 12341da177e4SLinus Torvalds if (!clone) 12351da177e4SLinus Torvalds return NULL; 12361da177e4SLinus Torvalds insp = list->next; 12371da177e4SLinus Torvalds list = clone; 12381da177e4SLinus Torvalds } else { 12391da177e4SLinus Torvalds /* This may be pulled without 12401da177e4SLinus Torvalds * problems. */ 12411da177e4SLinus Torvalds insp = list; 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 12441da177e4SLinus Torvalds kfree_skb(clone); 12451da177e4SLinus Torvalds return NULL; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds break; 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds } while (eat); 12501da177e4SLinus Torvalds 12511da177e4SLinus Torvalds /* Free pulled out fragments. */ 12521da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 12531da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 12541da177e4SLinus Torvalds kfree_skb(list); 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds /* And insert new clone at head. */ 12571da177e4SLinus Torvalds if (clone) { 12581da177e4SLinus Torvalds clone->next = list; 12591da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds } 12621da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds pull_pages: 12651da177e4SLinus Torvalds eat = delta; 12661da177e4SLinus Torvalds k = 0; 12671da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 12681da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size <= eat) { 12691da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 12701da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 12711da177e4SLinus Torvalds } else { 12721da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 12731da177e4SLinus Torvalds if (eat) { 12741da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 12751da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].size -= eat; 12761da177e4SLinus Torvalds eat = 0; 12771da177e4SLinus Torvalds } 12781da177e4SLinus Torvalds k++; 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds skb->tail += delta; 12841da177e4SLinus Torvalds skb->data_len -= delta; 12851da177e4SLinus Torvalds 128627a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 12871da177e4SLinus Torvalds } 1288b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds /* Copy some data bits from skb to kernel buffer. */ 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 12931da177e4SLinus Torvalds { 12941a028e50SDavid S. Miller int start = skb_headlen(skb); 1295fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1296fbb398a8SDavid S. Miller int i, copy; 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds if (offset > (int)skb->len - len) 12991da177e4SLinus Torvalds goto fault; 13001da177e4SLinus Torvalds 13011da177e4SLinus Torvalds /* Copy header. */ 13021a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 13031da177e4SLinus Torvalds if (copy > len) 13041da177e4SLinus Torvalds copy = len; 1305d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 13061da177e4SLinus Torvalds if ((len -= copy) == 0) 13071da177e4SLinus Torvalds return 0; 13081da177e4SLinus Torvalds offset += copy; 13091da177e4SLinus Torvalds to += copy; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds 13121da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 13131a028e50SDavid S. Miller int end; 13141da177e4SLinus Torvalds 1315547b792cSIlpo Järvinen WARN_ON(start > offset + len); 13161a028e50SDavid S. Miller 13171a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 13181da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 13191da177e4SLinus Torvalds u8 *vaddr; 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds if (copy > len) 13221da177e4SLinus Torvalds copy = len; 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]); 13251da177e4SLinus Torvalds memcpy(to, 13261a028e50SDavid S. Miller vaddr + skb_shinfo(skb)->frags[i].page_offset+ 13271a028e50SDavid S. Miller offset - start, copy); 13281da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 13291da177e4SLinus Torvalds 13301da177e4SLinus Torvalds if ((len -= copy) == 0) 13311da177e4SLinus Torvalds return 0; 13321da177e4SLinus Torvalds offset += copy; 13331da177e4SLinus Torvalds to += copy; 13341da177e4SLinus Torvalds } 13351a028e50SDavid S. Miller start = end; 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 1338fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 13391a028e50SDavid S. Miller int end; 13401da177e4SLinus Torvalds 1341547b792cSIlpo Järvinen WARN_ON(start > offset + len); 13421a028e50SDavid S. Miller 1343fbb398a8SDavid S. Miller end = start + frag_iter->len; 13441da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 13451da177e4SLinus Torvalds if (copy > len) 13461da177e4SLinus Torvalds copy = len; 1347fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 13481da177e4SLinus Torvalds goto fault; 13491da177e4SLinus Torvalds if ((len -= copy) == 0) 13501da177e4SLinus Torvalds return 0; 13511da177e4SLinus Torvalds offset += copy; 13521da177e4SLinus Torvalds to += copy; 13531da177e4SLinus Torvalds } 13541a028e50SDavid S. Miller start = end; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds if (!len) 13571da177e4SLinus Torvalds return 0; 13581da177e4SLinus Torvalds 13591da177e4SLinus Torvalds fault: 13601da177e4SLinus Torvalds return -EFAULT; 13611da177e4SLinus Torvalds } 1362b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 13631da177e4SLinus Torvalds 13649c55e01cSJens Axboe /* 13659c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 13669c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 13679c55e01cSJens Axboe */ 13689c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 13699c55e01cSJens Axboe { 13708b9d3728SJarek Poplawski put_page(spd->pages[i]); 13718b9d3728SJarek Poplawski } 13729c55e01cSJens Axboe 13734fb66994SJarek Poplawski static inline struct page *linear_to_page(struct page *page, unsigned int *len, 13744fb66994SJarek Poplawski unsigned int *offset, 13757a67e56fSJarek Poplawski struct sk_buff *skb, struct sock *sk) 13768b9d3728SJarek Poplawski { 13774fb66994SJarek Poplawski struct page *p = sk->sk_sndmsg_page; 13784fb66994SJarek Poplawski unsigned int off; 13798b9d3728SJarek Poplawski 13804fb66994SJarek Poplawski if (!p) { 13814fb66994SJarek Poplawski new_page: 13824fb66994SJarek Poplawski p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0); 13838b9d3728SJarek Poplawski if (!p) 13848b9d3728SJarek Poplawski return NULL; 13854fb66994SJarek Poplawski 13864fb66994SJarek Poplawski off = sk->sk_sndmsg_off = 0; 13874fb66994SJarek Poplawski /* hold one ref to this page until it's full */ 13884fb66994SJarek Poplawski } else { 13894fb66994SJarek Poplawski unsigned int mlen; 13904fb66994SJarek Poplawski 13914fb66994SJarek Poplawski off = sk->sk_sndmsg_off; 13924fb66994SJarek Poplawski mlen = PAGE_SIZE - off; 13934fb66994SJarek Poplawski if (mlen < 64 && mlen < *len) { 13944fb66994SJarek Poplawski put_page(p); 13954fb66994SJarek Poplawski goto new_page; 13964fb66994SJarek Poplawski } 13974fb66994SJarek Poplawski 13984fb66994SJarek Poplawski *len = min_t(unsigned int, *len, mlen); 13994fb66994SJarek Poplawski } 14004fb66994SJarek Poplawski 14014fb66994SJarek Poplawski memcpy(page_address(p) + off, page_address(page) + *offset, *len); 14024fb66994SJarek Poplawski sk->sk_sndmsg_off += *len; 14034fb66994SJarek Poplawski *offset = off; 14044fb66994SJarek Poplawski get_page(p); 14058b9d3728SJarek Poplawski 14068b9d3728SJarek Poplawski return p; 14079c55e01cSJens Axboe } 14089c55e01cSJens Axboe 14099c55e01cSJens Axboe /* 14109c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 14119c55e01cSJens Axboe */ 141235f3d14dSJens Axboe static inline int spd_fill_page(struct splice_pipe_desc *spd, 141335f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 14144fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 14157a67e56fSJarek Poplawski struct sk_buff *skb, int linear, 14167a67e56fSJarek Poplawski struct sock *sk) 14179c55e01cSJens Axboe { 141835f3d14dSJens Axboe if (unlikely(spd->nr_pages == pipe->buffers)) 14199c55e01cSJens Axboe return 1; 14209c55e01cSJens Axboe 14218b9d3728SJarek Poplawski if (linear) { 14227a67e56fSJarek Poplawski page = linear_to_page(page, len, &offset, skb, sk); 14238b9d3728SJarek Poplawski if (!page) 14248b9d3728SJarek Poplawski return 1; 14258b9d3728SJarek Poplawski } else 14268b9d3728SJarek Poplawski get_page(page); 14278b9d3728SJarek Poplawski 14289c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 14294fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 14309c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 14319c55e01cSJens Axboe spd->nr_pages++; 14328b9d3728SJarek Poplawski 14339c55e01cSJens Axboe return 0; 14349c55e01cSJens Axboe } 14359c55e01cSJens Axboe 14362870c43dSOctavian Purdila static inline void __segment_seek(struct page **page, unsigned int *poff, 14372870c43dSOctavian Purdila unsigned int *plen, unsigned int off) 14382870c43dSOctavian Purdila { 1439ce3dd395SJarek Poplawski unsigned long n; 1440ce3dd395SJarek Poplawski 14412870c43dSOctavian Purdila *poff += off; 1442ce3dd395SJarek Poplawski n = *poff / PAGE_SIZE; 1443ce3dd395SJarek Poplawski if (n) 1444ce3dd395SJarek Poplawski *page = nth_page(*page, n); 1445ce3dd395SJarek Poplawski 14462870c43dSOctavian Purdila *poff = *poff % PAGE_SIZE; 14472870c43dSOctavian Purdila *plen -= off; 14482870c43dSOctavian Purdila } 14492870c43dSOctavian Purdila 14502870c43dSOctavian Purdila static inline int __splice_segment(struct page *page, unsigned int poff, 14512870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 14522870c43dSOctavian Purdila unsigned int *len, struct sk_buff *skb, 14537a67e56fSJarek Poplawski struct splice_pipe_desc *spd, int linear, 145435f3d14dSJens Axboe struct sock *sk, 145535f3d14dSJens Axboe struct pipe_inode_info *pipe) 14569c55e01cSJens Axboe { 14572870c43dSOctavian Purdila if (!*len) 14582870c43dSOctavian Purdila return 1; 14599c55e01cSJens Axboe 14602870c43dSOctavian Purdila /* skip this segment if already processed */ 14612870c43dSOctavian Purdila if (*off >= plen) { 14622870c43dSOctavian Purdila *off -= plen; 14632870c43dSOctavian Purdila return 0; 14642870c43dSOctavian Purdila } 14652870c43dSOctavian Purdila 14662870c43dSOctavian Purdila /* ignore any bits we already processed */ 14672870c43dSOctavian Purdila if (*off) { 14682870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, *off); 14692870c43dSOctavian Purdila *off = 0; 14702870c43dSOctavian Purdila } 14712870c43dSOctavian Purdila 14722870c43dSOctavian Purdila do { 14732870c43dSOctavian Purdila unsigned int flen = min(*len, plen); 14742870c43dSOctavian Purdila 14752870c43dSOctavian Purdila /* the linear region may spread across several pages */ 14762870c43dSOctavian Purdila flen = min_t(unsigned int, flen, PAGE_SIZE - poff); 14772870c43dSOctavian Purdila 147835f3d14dSJens Axboe if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk)) 14792870c43dSOctavian Purdila return 1; 14802870c43dSOctavian Purdila 14812870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, flen); 14822870c43dSOctavian Purdila *len -= flen; 14832870c43dSOctavian Purdila 14842870c43dSOctavian Purdila } while (*len && plen); 14852870c43dSOctavian Purdila 14862870c43dSOctavian Purdila return 0; 1487db43a282SOctavian Purdila } 14889c55e01cSJens Axboe 14899c55e01cSJens Axboe /* 14902870c43dSOctavian Purdila * Map linear and fragment data from the skb to spd. It reports failure if the 14912870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 14929c55e01cSJens Axboe */ 149335f3d14dSJens Axboe static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 149435f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 149535f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 14962870c43dSOctavian Purdila { 14972870c43dSOctavian Purdila int seg; 14989c55e01cSJens Axboe 14999c55e01cSJens Axboe /* 15002870c43dSOctavian Purdila * map the linear part 15019c55e01cSJens Axboe */ 15022870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 15032870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 15042870c43dSOctavian Purdila skb_headlen(skb), 150535f3d14dSJens Axboe offset, len, skb, spd, 1, sk, pipe)) 15062870c43dSOctavian Purdila return 1; 15079c55e01cSJens Axboe 15089c55e01cSJens Axboe /* 15099c55e01cSJens Axboe * then map the fragments 15109c55e01cSJens Axboe */ 15119c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 15129c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 15139c55e01cSJens Axboe 15142870c43dSOctavian Purdila if (__splice_segment(f->page, f->page_offset, f->size, 151535f3d14dSJens Axboe offset, len, skb, spd, 0, sk, pipe)) 15162870c43dSOctavian Purdila return 1; 15179c55e01cSJens Axboe } 15189c55e01cSJens Axboe 15199c55e01cSJens Axboe return 0; 15209c55e01cSJens Axboe } 15219c55e01cSJens Axboe 15229c55e01cSJens Axboe /* 15239c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 15249c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 15259c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 15269c55e01cSJens Axboe * handle that cleanly. 15279c55e01cSJens Axboe */ 15288b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 15299c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 15309c55e01cSJens Axboe unsigned int flags) 15319c55e01cSJens Axboe { 153235f3d14dSJens Axboe struct partial_page partial[PIPE_DEF_BUFFERS]; 153335f3d14dSJens Axboe struct page *pages[PIPE_DEF_BUFFERS]; 15349c55e01cSJens Axboe struct splice_pipe_desc spd = { 15359c55e01cSJens Axboe .pages = pages, 15369c55e01cSJens Axboe .partial = partial, 15379c55e01cSJens Axboe .flags = flags, 15389c55e01cSJens Axboe .ops = &sock_pipe_buf_ops, 15399c55e01cSJens Axboe .spd_release = sock_spd_release, 15409c55e01cSJens Axboe }; 1541fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 15427a67e56fSJarek Poplawski struct sock *sk = skb->sk; 154335f3d14dSJens Axboe int ret = 0; 154435f3d14dSJens Axboe 154535f3d14dSJens Axboe if (splice_grow_spd(pipe, &spd)) 154635f3d14dSJens Axboe return -ENOMEM; 15479c55e01cSJens Axboe 15489c55e01cSJens Axboe /* 15499c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 15509c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 15519c55e01cSJens Axboe */ 155235f3d14dSJens Axboe if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 15539c55e01cSJens Axboe goto done; 15549c55e01cSJens Axboe else if (!tlen) 15559c55e01cSJens Axboe goto done; 15569c55e01cSJens Axboe 15579c55e01cSJens Axboe /* 15589c55e01cSJens Axboe * now see if we have a frag_list to map 15599c55e01cSJens Axboe */ 1560fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 1561fbb398a8SDavid S. Miller if (!tlen) 15629c55e01cSJens Axboe break; 156335f3d14dSJens Axboe if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1564fbb398a8SDavid S. Miller break; 15659c55e01cSJens Axboe } 15669c55e01cSJens Axboe 15679c55e01cSJens Axboe done: 15689c55e01cSJens Axboe if (spd.nr_pages) { 15699c55e01cSJens Axboe /* 15709c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 15719c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 15729c55e01cSJens Axboe * here as compared to sendfile(). We enter here 15739c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 15749c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 15759c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 15769c55e01cSJens Axboe * and networking will grab the socket lock. 15779c55e01cSJens Axboe */ 1578293ad604SOctavian Purdila release_sock(sk); 15799c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1580293ad604SOctavian Purdila lock_sock(sk); 15819c55e01cSJens Axboe } 15829c55e01cSJens Axboe 158335f3d14dSJens Axboe splice_shrink_spd(pipe, &spd); 158435f3d14dSJens Axboe return ret; 15859c55e01cSJens Axboe } 15869c55e01cSJens Axboe 1587357b40a1SHerbert Xu /** 1588357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1589357b40a1SHerbert Xu * @skb: destination buffer 1590357b40a1SHerbert Xu * @offset: offset in destination 1591357b40a1SHerbert Xu * @from: source buffer 1592357b40a1SHerbert Xu * @len: number of bytes to copy 1593357b40a1SHerbert Xu * 1594357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1595357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1596357b40a1SHerbert Xu * traversing fragment lists and such. 1597357b40a1SHerbert Xu */ 1598357b40a1SHerbert Xu 15990c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1600357b40a1SHerbert Xu { 16011a028e50SDavid S. Miller int start = skb_headlen(skb); 1602fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1603fbb398a8SDavid S. Miller int i, copy; 1604357b40a1SHerbert Xu 1605357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1606357b40a1SHerbert Xu goto fault; 1607357b40a1SHerbert Xu 16081a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1609357b40a1SHerbert Xu if (copy > len) 1610357b40a1SHerbert Xu copy = len; 161127d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1612357b40a1SHerbert Xu if ((len -= copy) == 0) 1613357b40a1SHerbert Xu return 0; 1614357b40a1SHerbert Xu offset += copy; 1615357b40a1SHerbert Xu from += copy; 1616357b40a1SHerbert Xu } 1617357b40a1SHerbert Xu 1618357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1619357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 16201a028e50SDavid S. Miller int end; 1621357b40a1SHerbert Xu 1622547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16231a028e50SDavid S. Miller 16241a028e50SDavid S. Miller end = start + frag->size; 1625357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1626357b40a1SHerbert Xu u8 *vaddr; 1627357b40a1SHerbert Xu 1628357b40a1SHerbert Xu if (copy > len) 1629357b40a1SHerbert Xu copy = len; 1630357b40a1SHerbert Xu 1631357b40a1SHerbert Xu vaddr = kmap_skb_frag(frag); 16321a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 16331a028e50SDavid S. Miller from, copy); 1634357b40a1SHerbert Xu kunmap_skb_frag(vaddr); 1635357b40a1SHerbert Xu 1636357b40a1SHerbert Xu if ((len -= copy) == 0) 1637357b40a1SHerbert Xu return 0; 1638357b40a1SHerbert Xu offset += copy; 1639357b40a1SHerbert Xu from += copy; 1640357b40a1SHerbert Xu } 16411a028e50SDavid S. Miller start = end; 1642357b40a1SHerbert Xu } 1643357b40a1SHerbert Xu 1644fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 16451a028e50SDavid S. Miller int end; 1646357b40a1SHerbert Xu 1647547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16481a028e50SDavid S. Miller 1649fbb398a8SDavid S. Miller end = start + frag_iter->len; 1650357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1651357b40a1SHerbert Xu if (copy > len) 1652357b40a1SHerbert Xu copy = len; 1653fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 16541a028e50SDavid S. Miller from, copy)) 1655357b40a1SHerbert Xu goto fault; 1656357b40a1SHerbert Xu if ((len -= copy) == 0) 1657357b40a1SHerbert Xu return 0; 1658357b40a1SHerbert Xu offset += copy; 1659357b40a1SHerbert Xu from += copy; 1660357b40a1SHerbert Xu } 16611a028e50SDavid S. Miller start = end; 1662357b40a1SHerbert Xu } 1663357b40a1SHerbert Xu if (!len) 1664357b40a1SHerbert Xu return 0; 1665357b40a1SHerbert Xu 1666357b40a1SHerbert Xu fault: 1667357b40a1SHerbert Xu return -EFAULT; 1668357b40a1SHerbert Xu } 1669357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1670357b40a1SHerbert Xu 16711da177e4SLinus Torvalds /* Checksum skb data. */ 16721da177e4SLinus Torvalds 16732bbbc868SAl Viro __wsum skb_checksum(const struct sk_buff *skb, int offset, 16742bbbc868SAl Viro int len, __wsum csum) 16751da177e4SLinus Torvalds { 16761a028e50SDavid S. Miller int start = skb_headlen(skb); 16771a028e50SDavid S. Miller int i, copy = start - offset; 1678fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 16791da177e4SLinus Torvalds int pos = 0; 16801da177e4SLinus Torvalds 16811da177e4SLinus Torvalds /* Checksum header. */ 16821da177e4SLinus Torvalds if (copy > 0) { 16831da177e4SLinus Torvalds if (copy > len) 16841da177e4SLinus Torvalds copy = len; 16851da177e4SLinus Torvalds csum = csum_partial(skb->data + offset, copy, csum); 16861da177e4SLinus Torvalds if ((len -= copy) == 0) 16871da177e4SLinus Torvalds return csum; 16881da177e4SLinus Torvalds offset += copy; 16891da177e4SLinus Torvalds pos = copy; 16901da177e4SLinus Torvalds } 16911da177e4SLinus Torvalds 16921da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16931a028e50SDavid S. Miller int end; 16941da177e4SLinus Torvalds 1695547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16961a028e50SDavid S. Miller 16971a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 16981da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 169944bb9363SAl Viro __wsum csum2; 17001da177e4SLinus Torvalds u8 *vaddr; 17011da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 17021da177e4SLinus Torvalds 17031da177e4SLinus Torvalds if (copy > len) 17041da177e4SLinus Torvalds copy = len; 17051da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 17061a028e50SDavid S. Miller csum2 = csum_partial(vaddr + frag->page_offset + 17071a028e50SDavid S. Miller offset - start, copy, 0); 17081da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 17091da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17101da177e4SLinus Torvalds if (!(len -= copy)) 17111da177e4SLinus Torvalds return csum; 17121da177e4SLinus Torvalds offset += copy; 17131da177e4SLinus Torvalds pos += copy; 17141da177e4SLinus Torvalds } 17151a028e50SDavid S. Miller start = end; 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds 1718fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 17191a028e50SDavid S. Miller int end; 17201da177e4SLinus Torvalds 1721547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17221a028e50SDavid S. Miller 1723fbb398a8SDavid S. Miller end = start + frag_iter->len; 17241da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17255f92a738SAl Viro __wsum csum2; 17261da177e4SLinus Torvalds if (copy > len) 17271da177e4SLinus Torvalds copy = len; 1728fbb398a8SDavid S. Miller csum2 = skb_checksum(frag_iter, offset - start, 17291a028e50SDavid S. Miller copy, 0); 17301da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17311da177e4SLinus Torvalds if ((len -= copy) == 0) 17321da177e4SLinus Torvalds return csum; 17331da177e4SLinus Torvalds offset += copy; 17341da177e4SLinus Torvalds pos += copy; 17351da177e4SLinus Torvalds } 17361a028e50SDavid S. Miller start = end; 17371da177e4SLinus Torvalds } 173809a62660SKris Katterjohn BUG_ON(len); 17391da177e4SLinus Torvalds 17401da177e4SLinus Torvalds return csum; 17411da177e4SLinus Torvalds } 1742b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 17431da177e4SLinus Torvalds 17441da177e4SLinus Torvalds /* Both of above in one bottle. */ 17451da177e4SLinus Torvalds 174681d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 174781d77662SAl Viro u8 *to, int len, __wsum csum) 17481da177e4SLinus Torvalds { 17491a028e50SDavid S. Miller int start = skb_headlen(skb); 17501a028e50SDavid S. Miller int i, copy = start - offset; 1751fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 17521da177e4SLinus Torvalds int pos = 0; 17531da177e4SLinus Torvalds 17541da177e4SLinus Torvalds /* Copy header. */ 17551da177e4SLinus Torvalds if (copy > 0) { 17561da177e4SLinus Torvalds if (copy > len) 17571da177e4SLinus Torvalds copy = len; 17581da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 17591da177e4SLinus Torvalds copy, csum); 17601da177e4SLinus Torvalds if ((len -= copy) == 0) 17611da177e4SLinus Torvalds return csum; 17621da177e4SLinus Torvalds offset += copy; 17631da177e4SLinus Torvalds to += copy; 17641da177e4SLinus Torvalds pos = copy; 17651da177e4SLinus Torvalds } 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 17681a028e50SDavid S. Miller int end; 17691da177e4SLinus Torvalds 1770547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17711a028e50SDavid S. Miller 17721a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 17731da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17745084205fSAl Viro __wsum csum2; 17751da177e4SLinus Torvalds u8 *vaddr; 17761da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 17771da177e4SLinus Torvalds 17781da177e4SLinus Torvalds if (copy > len) 17791da177e4SLinus Torvalds copy = len; 17801da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 17811da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 17821a028e50SDavid S. Miller frag->page_offset + 17831a028e50SDavid S. Miller offset - start, to, 17841a028e50SDavid S. Miller copy, 0); 17851da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 17861da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17871da177e4SLinus Torvalds if (!(len -= copy)) 17881da177e4SLinus Torvalds return csum; 17891da177e4SLinus Torvalds offset += copy; 17901da177e4SLinus Torvalds to += copy; 17911da177e4SLinus Torvalds pos += copy; 17921da177e4SLinus Torvalds } 17931a028e50SDavid S. Miller start = end; 17941da177e4SLinus Torvalds } 17951da177e4SLinus Torvalds 1796fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 179781d77662SAl Viro __wsum csum2; 17981a028e50SDavid S. Miller int end; 17991da177e4SLinus Torvalds 1800547b792cSIlpo Järvinen WARN_ON(start > offset + len); 18011a028e50SDavid S. Miller 1802fbb398a8SDavid S. Miller end = start + frag_iter->len; 18031da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 18041da177e4SLinus Torvalds if (copy > len) 18051da177e4SLinus Torvalds copy = len; 1806fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 18071a028e50SDavid S. Miller offset - start, 18081da177e4SLinus Torvalds to, copy, 0); 18091da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 18101da177e4SLinus Torvalds if ((len -= copy) == 0) 18111da177e4SLinus Torvalds return csum; 18121da177e4SLinus Torvalds offset += copy; 18131da177e4SLinus Torvalds to += copy; 18141da177e4SLinus Torvalds pos += copy; 18151da177e4SLinus Torvalds } 18161a028e50SDavid S. Miller start = end; 18171da177e4SLinus Torvalds } 181809a62660SKris Katterjohn BUG_ON(len); 18191da177e4SLinus Torvalds return csum; 18201da177e4SLinus Torvalds } 1821b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 18221da177e4SLinus Torvalds 18231da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 18241da177e4SLinus Torvalds { 1825d3bc23e7SAl Viro __wsum csum; 18261da177e4SLinus Torvalds long csstart; 18271da177e4SLinus Torvalds 182884fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 182955508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 18301da177e4SLinus Torvalds else 18311da177e4SLinus Torvalds csstart = skb_headlen(skb); 18321da177e4SLinus Torvalds 183309a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 18341da177e4SLinus Torvalds 1835d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 18361da177e4SLinus Torvalds 18371da177e4SLinus Torvalds csum = 0; 18381da177e4SLinus Torvalds if (csstart != skb->len) 18391da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 18401da177e4SLinus Torvalds skb->len - csstart, 0); 18411da177e4SLinus Torvalds 184284fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 1843ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 18441da177e4SLinus Torvalds 1845d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 18461da177e4SLinus Torvalds } 18471da177e4SLinus Torvalds } 1848b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 18491da177e4SLinus Torvalds 18501da177e4SLinus Torvalds /** 18511da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 18521da177e4SLinus Torvalds * @list: list to dequeue from 18531da177e4SLinus Torvalds * 18541da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 18551da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 18561da177e4SLinus Torvalds * returned or %NULL if the list is empty. 18571da177e4SLinus Torvalds */ 18581da177e4SLinus Torvalds 18591da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 18601da177e4SLinus Torvalds { 18611da177e4SLinus Torvalds unsigned long flags; 18621da177e4SLinus Torvalds struct sk_buff *result; 18631da177e4SLinus Torvalds 18641da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18651da177e4SLinus Torvalds result = __skb_dequeue(list); 18661da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18671da177e4SLinus Torvalds return result; 18681da177e4SLinus Torvalds } 1869b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds /** 18721da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 18731da177e4SLinus Torvalds * @list: list to dequeue from 18741da177e4SLinus Torvalds * 18751da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 18761da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 18771da177e4SLinus Torvalds * returned or %NULL if the list is empty. 18781da177e4SLinus Torvalds */ 18791da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 18801da177e4SLinus Torvalds { 18811da177e4SLinus Torvalds unsigned long flags; 18821da177e4SLinus Torvalds struct sk_buff *result; 18831da177e4SLinus Torvalds 18841da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18851da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 18861da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18871da177e4SLinus Torvalds return result; 18881da177e4SLinus Torvalds } 1889b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 18901da177e4SLinus Torvalds 18911da177e4SLinus Torvalds /** 18921da177e4SLinus Torvalds * skb_queue_purge - empty a list 18931da177e4SLinus Torvalds * @list: list to empty 18941da177e4SLinus Torvalds * 18951da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 18961da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 18971da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 18981da177e4SLinus Torvalds */ 18991da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 19001da177e4SLinus Torvalds { 19011da177e4SLinus Torvalds struct sk_buff *skb; 19021da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 19031da177e4SLinus Torvalds kfree_skb(skb); 19041da177e4SLinus Torvalds } 1905b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds /** 19081da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 19091da177e4SLinus Torvalds * @list: list to use 19101da177e4SLinus Torvalds * @newsk: buffer to queue 19111da177e4SLinus Torvalds * 19121da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 19131da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 19141da177e4SLinus Torvalds * safely. 19151da177e4SLinus Torvalds * 19161da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19171da177e4SLinus Torvalds */ 19181da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 19191da177e4SLinus Torvalds { 19201da177e4SLinus Torvalds unsigned long flags; 19211da177e4SLinus Torvalds 19221da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 19231da177e4SLinus Torvalds __skb_queue_head(list, newsk); 19241da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 19251da177e4SLinus Torvalds } 1926b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 19271da177e4SLinus Torvalds 19281da177e4SLinus Torvalds /** 19291da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 19301da177e4SLinus Torvalds * @list: list to use 19311da177e4SLinus Torvalds * @newsk: buffer to queue 19321da177e4SLinus Torvalds * 19331da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 19341da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 19351da177e4SLinus Torvalds * safely. 19361da177e4SLinus Torvalds * 19371da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19381da177e4SLinus Torvalds */ 19391da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 19401da177e4SLinus Torvalds { 19411da177e4SLinus Torvalds unsigned long flags; 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 19441da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 19451da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 19461da177e4SLinus Torvalds } 1947b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 19488728b834SDavid S. Miller 19491da177e4SLinus Torvalds /** 19501da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 19511da177e4SLinus Torvalds * @skb: buffer to remove 19528728b834SDavid S. Miller * @list: list to use 19531da177e4SLinus Torvalds * 19548728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 19558728b834SDavid S. Miller * function is atomic with respect to other list locked calls 19561da177e4SLinus Torvalds * 19578728b834SDavid S. Miller * You must know what list the SKB is on. 19581da177e4SLinus Torvalds */ 19598728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 19601da177e4SLinus Torvalds { 19611da177e4SLinus Torvalds unsigned long flags; 19621da177e4SLinus Torvalds 19631da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 19648728b834SDavid S. Miller __skb_unlink(skb, list); 19651da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 19661da177e4SLinus Torvalds } 1967b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 19681da177e4SLinus Torvalds 19691da177e4SLinus Torvalds /** 19701da177e4SLinus Torvalds * skb_append - append a buffer 19711da177e4SLinus Torvalds * @old: buffer to insert after 19721da177e4SLinus Torvalds * @newsk: buffer to insert 19738728b834SDavid S. Miller * @list: list to use 19741da177e4SLinus Torvalds * 19751da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 19761da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 19771da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19781da177e4SLinus Torvalds */ 19798728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 19801da177e4SLinus Torvalds { 19811da177e4SLinus Torvalds unsigned long flags; 19821da177e4SLinus Torvalds 19838728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 19847de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 19858728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 19861da177e4SLinus Torvalds } 1987b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 19881da177e4SLinus Torvalds 19891da177e4SLinus Torvalds /** 19901da177e4SLinus Torvalds * skb_insert - insert a buffer 19911da177e4SLinus Torvalds * @old: buffer to insert before 19921da177e4SLinus Torvalds * @newsk: buffer to insert 19938728b834SDavid S. Miller * @list: list to use 19941da177e4SLinus Torvalds * 19958728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 19968728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 19978728b834SDavid S. Miller * calls. 19988728b834SDavid S. Miller * 19991da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 20001da177e4SLinus Torvalds */ 20018728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 20021da177e4SLinus Torvalds { 20031da177e4SLinus Torvalds unsigned long flags; 20041da177e4SLinus Torvalds 20058728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 20068728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 20078728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 20081da177e4SLinus Torvalds } 2009b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 20101da177e4SLinus Torvalds 20111da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 20121da177e4SLinus Torvalds struct sk_buff* skb1, 20131da177e4SLinus Torvalds const u32 len, const int pos) 20141da177e4SLinus Torvalds { 20151da177e4SLinus Torvalds int i; 20161da177e4SLinus Torvalds 2017d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2018d626f62bSArnaldo Carvalho de Melo pos - len); 20191da177e4SLinus Torvalds /* And move data appendix as is. */ 20201da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 20211da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 20221da177e4SLinus Torvalds 20231da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 20241da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 20251da177e4SLinus Torvalds skb1->data_len = skb->data_len; 20261da177e4SLinus Torvalds skb1->len += skb1->data_len; 20271da177e4SLinus Torvalds skb->data_len = 0; 20281da177e4SLinus Torvalds skb->len = len; 202927a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 20301da177e4SLinus Torvalds } 20311da177e4SLinus Torvalds 20321da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 20331da177e4SLinus Torvalds struct sk_buff* skb1, 20341da177e4SLinus Torvalds const u32 len, int pos) 20351da177e4SLinus Torvalds { 20361da177e4SLinus Torvalds int i, k = 0; 20371da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 20381da177e4SLinus Torvalds 20391da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 20401da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 20411da177e4SLinus Torvalds skb->len = len; 20421da177e4SLinus Torvalds skb->data_len = len - pos; 20431da177e4SLinus Torvalds 20441da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 20451da177e4SLinus Torvalds int size = skb_shinfo(skb)->frags[i].size; 20461da177e4SLinus Torvalds 20471da177e4SLinus Torvalds if (pos + size > len) { 20481da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 20491da177e4SLinus Torvalds 20501da177e4SLinus Torvalds if (pos < len) { 20511da177e4SLinus Torvalds /* Split frag. 20521da177e4SLinus Torvalds * We have two variants in this case: 20531da177e4SLinus Torvalds * 1. Move all the frag to the second 20541da177e4SLinus Torvalds * part, if it is possible. F.e. 20551da177e4SLinus Torvalds * this approach is mandatory for TUX, 20561da177e4SLinus Torvalds * where splitting is expensive. 20571da177e4SLinus Torvalds * 2. Split is accurately. We make this. 20581da177e4SLinus Torvalds */ 20591da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 20601da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 20611da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].size -= len - pos; 20621da177e4SLinus Torvalds skb_shinfo(skb)->frags[i].size = len - pos; 20631da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 20641da177e4SLinus Torvalds } 20651da177e4SLinus Torvalds k++; 20661da177e4SLinus Torvalds } else 20671da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 20681da177e4SLinus Torvalds pos += size; 20691da177e4SLinus Torvalds } 20701da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 20711da177e4SLinus Torvalds } 20721da177e4SLinus Torvalds 20731da177e4SLinus Torvalds /** 20741da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 20751da177e4SLinus Torvalds * @skb: the buffer to split 20761da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 20771da177e4SLinus Torvalds * @len: new length for skb 20781da177e4SLinus Torvalds */ 20791da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 20801da177e4SLinus Torvalds { 20811da177e4SLinus Torvalds int pos = skb_headlen(skb); 20821da177e4SLinus Torvalds 20831da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 20841da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 20851da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 20861da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 20871da177e4SLinus Torvalds } 2088b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 20891da177e4SLinus Torvalds 20909f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 20919f782db3SIlpo Järvinen * 20929f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 20939f782db3SIlpo Järvinen */ 2094832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2095832d11c5SIlpo Järvinen { 20960ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2097832d11c5SIlpo Järvinen } 2098832d11c5SIlpo Järvinen 2099832d11c5SIlpo Järvinen /** 2100832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2101832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2102832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2103832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2104832d11c5SIlpo Järvinen * 2105832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 2106832d11c5SIlpo Järvinen * the length of the skb, from tgt to skb. Returns number bytes shifted. 2107832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2108832d11c5SIlpo Järvinen * 2109832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2110832d11c5SIlpo Järvinen * 2111832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2112832d11c5SIlpo Järvinen * to have non-paged data as well. 2113832d11c5SIlpo Järvinen * 2114832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2115832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2116832d11c5SIlpo Järvinen */ 2117832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2118832d11c5SIlpo Järvinen { 2119832d11c5SIlpo Järvinen int from, to, merge, todo; 2120832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2121832d11c5SIlpo Järvinen 2122832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2123832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2124832d11c5SIlpo Järvinen 2125832d11c5SIlpo Järvinen todo = shiftlen; 2126832d11c5SIlpo Järvinen from = 0; 2127832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2128832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2129832d11c5SIlpo Järvinen 2130832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2131832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2132832d11c5SIlpo Järvinen */ 2133832d11c5SIlpo Järvinen if (!to || 2134832d11c5SIlpo Järvinen !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) { 2135832d11c5SIlpo Järvinen merge = -1; 2136832d11c5SIlpo Järvinen } else { 2137832d11c5SIlpo Järvinen merge = to - 1; 2138832d11c5SIlpo Järvinen 2139832d11c5SIlpo Järvinen todo -= fragfrom->size; 2140832d11c5SIlpo Järvinen if (todo < 0) { 2141832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2142832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2143832d11c5SIlpo Järvinen return 0; 2144832d11c5SIlpo Järvinen 21459f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 21469f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2147832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2148832d11c5SIlpo Järvinen 2149832d11c5SIlpo Järvinen fragto->size += shiftlen; 2150832d11c5SIlpo Järvinen fragfrom->size -= shiftlen; 2151832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2152832d11c5SIlpo Järvinen 2153832d11c5SIlpo Järvinen goto onlymerged; 2154832d11c5SIlpo Järvinen } 2155832d11c5SIlpo Järvinen 2156832d11c5SIlpo Järvinen from++; 2157832d11c5SIlpo Järvinen } 2158832d11c5SIlpo Järvinen 2159832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2160832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2161832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2162832d11c5SIlpo Järvinen return 0; 2163832d11c5SIlpo Järvinen 2164832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2165832d11c5SIlpo Järvinen return 0; 2166832d11c5SIlpo Järvinen 2167832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2168832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2169832d11c5SIlpo Järvinen return 0; 2170832d11c5SIlpo Järvinen 2171832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2172832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2173832d11c5SIlpo Järvinen 2174832d11c5SIlpo Järvinen if (todo >= fragfrom->size) { 2175832d11c5SIlpo Järvinen *fragto = *fragfrom; 2176832d11c5SIlpo Järvinen todo -= fragfrom->size; 2177832d11c5SIlpo Järvinen from++; 2178832d11c5SIlpo Järvinen to++; 2179832d11c5SIlpo Järvinen 2180832d11c5SIlpo Järvinen } else { 2181832d11c5SIlpo Järvinen get_page(fragfrom->page); 2182832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2183832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 2184832d11c5SIlpo Järvinen fragto->size = todo; 2185832d11c5SIlpo Järvinen 2186832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 2187832d11c5SIlpo Järvinen fragfrom->size -= todo; 2188832d11c5SIlpo Järvinen todo = 0; 2189832d11c5SIlpo Järvinen 2190832d11c5SIlpo Järvinen to++; 2191832d11c5SIlpo Järvinen break; 2192832d11c5SIlpo Järvinen } 2193832d11c5SIlpo Järvinen } 2194832d11c5SIlpo Järvinen 2195832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2196832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2197832d11c5SIlpo Järvinen 2198832d11c5SIlpo Järvinen if (merge >= 0) { 2199832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2200832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2201832d11c5SIlpo Järvinen 2202832d11c5SIlpo Järvinen fragto->size += fragfrom->size; 2203832d11c5SIlpo Järvinen put_page(fragfrom->page); 2204832d11c5SIlpo Järvinen } 2205832d11c5SIlpo Järvinen 2206832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2207832d11c5SIlpo Järvinen to = 0; 2208832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2209832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2210832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2211832d11c5SIlpo Järvinen 2212832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2213832d11c5SIlpo Järvinen 2214832d11c5SIlpo Järvinen onlymerged: 2215832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2216832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2217832d11c5SIlpo Järvinen */ 2218832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2219832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2220832d11c5SIlpo Järvinen 2221832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2222832d11c5SIlpo Järvinen skb->len -= shiftlen; 2223832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2224832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2225832d11c5SIlpo Järvinen tgt->len += shiftlen; 2226832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2227832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2228832d11c5SIlpo Järvinen 2229832d11c5SIlpo Järvinen return shiftlen; 2230832d11c5SIlpo Järvinen } 2231832d11c5SIlpo Järvinen 2232677e90edSThomas Graf /** 2233677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2234677e90edSThomas Graf * @skb: the buffer to read 2235677e90edSThomas Graf * @from: lower offset of data to be read 2236677e90edSThomas Graf * @to: upper offset of data to be read 2237677e90edSThomas Graf * @st: state variable 2238677e90edSThomas Graf * 2239677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2240677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2241677e90edSThomas Graf */ 2242677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2243677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2244677e90edSThomas Graf { 2245677e90edSThomas Graf st->lower_offset = from; 2246677e90edSThomas Graf st->upper_offset = to; 2247677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2248677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2249677e90edSThomas Graf st->frag_data = NULL; 2250677e90edSThomas Graf } 2251b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2252677e90edSThomas Graf 2253677e90edSThomas Graf /** 2254677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2255677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2256677e90edSThomas Graf * @data: destination pointer for data to be returned 2257677e90edSThomas Graf * @st: state variable 2258677e90edSThomas Graf * 2259677e90edSThomas Graf * Reads a block of skb data at &consumed relative to the 2260677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2261677e90edSThomas Graf * the head of the data block to &data and returns the length 2262677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2263677e90edSThomas Graf * offset has been reached. 2264677e90edSThomas Graf * 2265677e90edSThomas Graf * The caller is not required to consume all of the data 2266677e90edSThomas Graf * returned, i.e. &consumed is typically set to the number 2267677e90edSThomas Graf * of bytes already consumed and the next call to 2268677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2269677e90edSThomas Graf * 227025985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2271677e90edSThomas Graf * this limitation is the cost for zerocopy seqeuental 2272677e90edSThomas Graf * reads of potentially non linear data. 2273677e90edSThomas Graf * 2274bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2275677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2276677e90edSThomas Graf * a stack for this purpose. 2277677e90edSThomas Graf */ 2278677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2279677e90edSThomas Graf struct skb_seq_state *st) 2280677e90edSThomas Graf { 2281677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2282677e90edSThomas Graf skb_frag_t *frag; 2283677e90edSThomas Graf 2284677e90edSThomas Graf if (unlikely(abs_offset >= st->upper_offset)) 2285677e90edSThomas Graf return 0; 2286677e90edSThomas Graf 2287677e90edSThomas Graf next_skb: 228895e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2289677e90edSThomas Graf 2290995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 229195e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2292677e90edSThomas Graf return block_limit - abs_offset; 2293677e90edSThomas Graf } 2294677e90edSThomas Graf 2295677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2296677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2297677e90edSThomas Graf 2298677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2299677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 2300677e90edSThomas Graf block_limit = frag->size + st->stepped_offset; 2301677e90edSThomas Graf 2302677e90edSThomas Graf if (abs_offset < block_limit) { 2303677e90edSThomas Graf if (!st->frag_data) 2304677e90edSThomas Graf st->frag_data = kmap_skb_frag(frag); 2305677e90edSThomas Graf 2306677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2307677e90edSThomas Graf (abs_offset - st->stepped_offset); 2308677e90edSThomas Graf 2309677e90edSThomas Graf return block_limit - abs_offset; 2310677e90edSThomas Graf } 2311677e90edSThomas Graf 2312677e90edSThomas Graf if (st->frag_data) { 2313677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2314677e90edSThomas Graf st->frag_data = NULL; 2315677e90edSThomas Graf } 2316677e90edSThomas Graf 2317677e90edSThomas Graf st->frag_idx++; 2318677e90edSThomas Graf st->stepped_offset += frag->size; 2319677e90edSThomas Graf } 2320677e90edSThomas Graf 23215b5a60daSOlaf Kirch if (st->frag_data) { 23225b5a60daSOlaf Kirch kunmap_skb_frag(st->frag_data); 23235b5a60daSOlaf Kirch st->frag_data = NULL; 23245b5a60daSOlaf Kirch } 23255b5a60daSOlaf Kirch 232621dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2327677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 232895e3b24cSHerbert Xu st->frag_idx = 0; 2329677e90edSThomas Graf goto next_skb; 233071b3346dSShyam Iyer } else if (st->cur_skb->next) { 233171b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 233271b3346dSShyam Iyer st->frag_idx = 0; 2333677e90edSThomas Graf goto next_skb; 2334677e90edSThomas Graf } 2335677e90edSThomas Graf 2336677e90edSThomas Graf return 0; 2337677e90edSThomas Graf } 2338b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2339677e90edSThomas Graf 2340677e90edSThomas Graf /** 2341677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2342677e90edSThomas Graf * @st: state variable 2343677e90edSThomas Graf * 2344677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2345677e90edSThomas Graf * returned 0. 2346677e90edSThomas Graf */ 2347677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2348677e90edSThomas Graf { 2349677e90edSThomas Graf if (st->frag_data) 2350677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2351677e90edSThomas Graf } 2352b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2353677e90edSThomas Graf 23543fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 23553fc7e8a6SThomas Graf 23563fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 23573fc7e8a6SThomas Graf struct ts_config *conf, 23583fc7e8a6SThomas Graf struct ts_state *state) 23593fc7e8a6SThomas Graf { 23603fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 23613fc7e8a6SThomas Graf } 23623fc7e8a6SThomas Graf 23633fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 23643fc7e8a6SThomas Graf { 23653fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 23663fc7e8a6SThomas Graf } 23673fc7e8a6SThomas Graf 23683fc7e8a6SThomas Graf /** 23693fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 23703fc7e8a6SThomas Graf * @skb: the buffer to look in 23713fc7e8a6SThomas Graf * @from: search offset 23723fc7e8a6SThomas Graf * @to: search limit 23733fc7e8a6SThomas Graf * @config: textsearch configuration 23743fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 23753fc7e8a6SThomas Graf * 23763fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 23773fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 23783fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 23793fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 23803fc7e8a6SThomas Graf */ 23813fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 23823fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 23833fc7e8a6SThomas Graf struct ts_state *state) 23843fc7e8a6SThomas Graf { 2385f72b948dSPhil Oester unsigned int ret; 2386f72b948dSPhil Oester 23873fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 23883fc7e8a6SThomas Graf config->finish = skb_ts_finish; 23893fc7e8a6SThomas Graf 23903fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 23913fc7e8a6SThomas Graf 2392f72b948dSPhil Oester ret = textsearch_find(config, state); 2393f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 23943fc7e8a6SThomas Graf } 2395b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 23963fc7e8a6SThomas Graf 2397e89e9cf5SAnanda Raju /** 2398e89e9cf5SAnanda Raju * skb_append_datato_frags: - append the user data to a skb 2399e89e9cf5SAnanda Raju * @sk: sock structure 2400e89e9cf5SAnanda Raju * @skb: skb structure to be appened with user data. 2401e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2402e89e9cf5SAnanda Raju * @from: pointer to user message iov 2403e89e9cf5SAnanda Raju * @length: length of the iov message 2404e89e9cf5SAnanda Raju * 2405e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2406e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2407e89e9cf5SAnanda Raju */ 2408e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2409dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2410e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2411e89e9cf5SAnanda Raju void *from, int length) 2412e89e9cf5SAnanda Raju { 2413e89e9cf5SAnanda Raju int frg_cnt = 0; 2414e89e9cf5SAnanda Raju skb_frag_t *frag = NULL; 2415e89e9cf5SAnanda Raju struct page *page = NULL; 2416e89e9cf5SAnanda Raju int copy, left; 2417e89e9cf5SAnanda Raju int offset = 0; 2418e89e9cf5SAnanda Raju int ret; 2419e89e9cf5SAnanda Raju 2420e89e9cf5SAnanda Raju do { 2421e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2422e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2423e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2424e89e9cf5SAnanda Raju return -EFAULT; 2425e89e9cf5SAnanda Raju 2426e89e9cf5SAnanda Raju /* allocate a new page for next frag */ 2427e89e9cf5SAnanda Raju page = alloc_pages(sk->sk_allocation, 0); 2428e89e9cf5SAnanda Raju 2429e89e9cf5SAnanda Raju /* If alloc_page fails just return failure and caller will 2430e89e9cf5SAnanda Raju * free previous allocated pages by doing kfree_skb() 2431e89e9cf5SAnanda Raju */ 2432e89e9cf5SAnanda Raju if (page == NULL) 2433e89e9cf5SAnanda Raju return -ENOMEM; 2434e89e9cf5SAnanda Raju 2435e89e9cf5SAnanda Raju /* initialize the next frag */ 2436e89e9cf5SAnanda Raju skb_fill_page_desc(skb, frg_cnt, page, 0, 0); 2437e89e9cf5SAnanda Raju skb->truesize += PAGE_SIZE; 2438e89e9cf5SAnanda Raju atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 2439e89e9cf5SAnanda Raju 2440e89e9cf5SAnanda Raju /* get the new initialized frag */ 2441e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2442e89e9cf5SAnanda Raju frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; 2443e89e9cf5SAnanda Raju 2444e89e9cf5SAnanda Raju /* copy the user data to page */ 2445e89e9cf5SAnanda Raju left = PAGE_SIZE - frag->page_offset; 2446e89e9cf5SAnanda Raju copy = (length > left)? left : length; 2447e89e9cf5SAnanda Raju 2448e89e9cf5SAnanda Raju ret = getfrag(from, (page_address(frag->page) + 2449e89e9cf5SAnanda Raju frag->page_offset + frag->size), 2450e89e9cf5SAnanda Raju offset, copy, 0, skb); 2451e89e9cf5SAnanda Raju if (ret < 0) 2452e89e9cf5SAnanda Raju return -EFAULT; 2453e89e9cf5SAnanda Raju 2454e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2455e89e9cf5SAnanda Raju frag->size += copy; 2456e89e9cf5SAnanda Raju skb->len += copy; 2457e89e9cf5SAnanda Raju skb->data_len += copy; 2458e89e9cf5SAnanda Raju offset += copy; 2459e89e9cf5SAnanda Raju length -= copy; 2460e89e9cf5SAnanda Raju 2461e89e9cf5SAnanda Raju } while (length > 0); 2462e89e9cf5SAnanda Raju 2463e89e9cf5SAnanda Raju return 0; 2464e89e9cf5SAnanda Raju } 2465b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2466e89e9cf5SAnanda Raju 2467cbb042f9SHerbert Xu /** 2468cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2469cbb042f9SHerbert Xu * @skb: buffer to update 2470cbb042f9SHerbert Xu * @len: length of data pulled 2471cbb042f9SHerbert Xu * 2472cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2473fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 247484fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 247584fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 247684fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2477cbb042f9SHerbert Xu */ 2478cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2479cbb042f9SHerbert Xu { 2480cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2481cbb042f9SHerbert Xu skb->len -= len; 2482cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2483cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2484cbb042f9SHerbert Xu return skb->data += len; 2485cbb042f9SHerbert Xu } 2486f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2487f94691acSArnaldo Carvalho de Melo 2488f4c50d99SHerbert Xu /** 2489f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2490f4c50d99SHerbert Xu * @skb: buffer to segment 2491576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2492f4c50d99SHerbert Xu * 2493f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 24944c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 24954c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2496f4c50d99SHerbert Xu */ 249704ed3e74SMichał Mirosław struct sk_buff *skb_segment(struct sk_buff *skb, u32 features) 2498f4c50d99SHerbert Xu { 2499f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2500f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 250189319d38SHerbert Xu struct sk_buff *fskb = skb_shinfo(skb)->frag_list; 2502f4c50d99SHerbert Xu unsigned int mss = skb_shinfo(skb)->gso_size; 250398e399f8SArnaldo Carvalho de Melo unsigned int doffset = skb->data - skb_mac_header(skb); 2504f4c50d99SHerbert Xu unsigned int offset = doffset; 2505f4c50d99SHerbert Xu unsigned int headroom; 2506f4c50d99SHerbert Xu unsigned int len; 250704ed3e74SMichał Mirosław int sg = !!(features & NETIF_F_SG); 2508f4c50d99SHerbert Xu int nfrags = skb_shinfo(skb)->nr_frags; 2509f4c50d99SHerbert Xu int err = -ENOMEM; 2510f4c50d99SHerbert Xu int i = 0; 2511f4c50d99SHerbert Xu int pos; 2512f4c50d99SHerbert Xu 2513f4c50d99SHerbert Xu __skb_push(skb, doffset); 2514f4c50d99SHerbert Xu headroom = skb_headroom(skb); 2515f4c50d99SHerbert Xu pos = skb_headlen(skb); 2516f4c50d99SHerbert Xu 2517f4c50d99SHerbert Xu do { 2518f4c50d99SHerbert Xu struct sk_buff *nskb; 2519f4c50d99SHerbert Xu skb_frag_t *frag; 2520c8884eddSHerbert Xu int hsize; 2521f4c50d99SHerbert Xu int size; 2522f4c50d99SHerbert Xu 2523f4c50d99SHerbert Xu len = skb->len - offset; 2524f4c50d99SHerbert Xu if (len > mss) 2525f4c50d99SHerbert Xu len = mss; 2526f4c50d99SHerbert Xu 2527f4c50d99SHerbert Xu hsize = skb_headlen(skb) - offset; 2528f4c50d99SHerbert Xu if (hsize < 0) 2529f4c50d99SHerbert Xu hsize = 0; 2530c8884eddSHerbert Xu if (hsize > len || !sg) 2531c8884eddSHerbert Xu hsize = len; 2532f4c50d99SHerbert Xu 253389319d38SHerbert Xu if (!hsize && i >= nfrags) { 253489319d38SHerbert Xu BUG_ON(fskb->len != len); 253589319d38SHerbert Xu 253689319d38SHerbert Xu pos += len; 253789319d38SHerbert Xu nskb = skb_clone(fskb, GFP_ATOMIC); 253889319d38SHerbert Xu fskb = fskb->next; 253989319d38SHerbert Xu 2540f4c50d99SHerbert Xu if (unlikely(!nskb)) 2541f4c50d99SHerbert Xu goto err; 2542f4c50d99SHerbert Xu 254389319d38SHerbert Xu hsize = skb_end_pointer(nskb) - nskb->head; 254489319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 254589319d38SHerbert Xu kfree_skb(nskb); 254689319d38SHerbert Xu goto err; 254789319d38SHerbert Xu } 254889319d38SHerbert Xu 254989319d38SHerbert Xu nskb->truesize += skb_end_pointer(nskb) - nskb->head - 255089319d38SHerbert Xu hsize; 255189319d38SHerbert Xu skb_release_head_state(nskb); 255289319d38SHerbert Xu __skb_push(nskb, doffset); 255389319d38SHerbert Xu } else { 255489319d38SHerbert Xu nskb = alloc_skb(hsize + doffset + headroom, 255589319d38SHerbert Xu GFP_ATOMIC); 255689319d38SHerbert Xu 255789319d38SHerbert Xu if (unlikely(!nskb)) 255889319d38SHerbert Xu goto err; 255989319d38SHerbert Xu 256089319d38SHerbert Xu skb_reserve(nskb, headroom); 256189319d38SHerbert Xu __skb_put(nskb, doffset); 256289319d38SHerbert Xu } 256389319d38SHerbert Xu 2564f4c50d99SHerbert Xu if (segs) 2565f4c50d99SHerbert Xu tail->next = nskb; 2566f4c50d99SHerbert Xu else 2567f4c50d99SHerbert Xu segs = nskb; 2568f4c50d99SHerbert Xu tail = nskb; 2569f4c50d99SHerbert Xu 25706f85a124SHerbert Xu __copy_skb_header(nskb, skb); 2571f4c50d99SHerbert Xu nskb->mac_len = skb->mac_len; 2572f4c50d99SHerbert Xu 25733d3be433SEric Dumazet /* nskb and skb might have different headroom */ 25743d3be433SEric Dumazet if (nskb->ip_summed == CHECKSUM_PARTIAL) 25753d3be433SEric Dumazet nskb->csum_start += skb_headroom(nskb) - headroom; 25763d3be433SEric Dumazet 2577459a98edSArnaldo Carvalho de Melo skb_reset_mac_header(nskb); 2578ddc7b8e3SArnaldo Carvalho de Melo skb_set_network_header(nskb, skb->mac_len); 2579b0e380b1SArnaldo Carvalho de Melo nskb->transport_header = (nskb->network_header + 2580b0e380b1SArnaldo Carvalho de Melo skb_network_header_len(skb)); 258189319d38SHerbert Xu skb_copy_from_linear_data(skb, nskb->data, doffset); 258289319d38SHerbert Xu 25832f181855SHerbert Xu if (fskb != skb_shinfo(skb)->frag_list) 258489319d38SHerbert Xu continue; 258589319d38SHerbert Xu 2586f4c50d99SHerbert Xu if (!sg) { 25876f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 2588f4c50d99SHerbert Xu nskb->csum = skb_copy_and_csum_bits(skb, offset, 2589f4c50d99SHerbert Xu skb_put(nskb, len), 2590f4c50d99SHerbert Xu len, 0); 2591f4c50d99SHerbert Xu continue; 2592f4c50d99SHerbert Xu } 2593f4c50d99SHerbert Xu 2594f4c50d99SHerbert Xu frag = skb_shinfo(nskb)->frags; 2595f4c50d99SHerbert Xu 2596d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, 2597d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 2598f4c50d99SHerbert Xu 259989319d38SHerbert Xu while (pos < offset + len && i < nfrags) { 2600f4c50d99SHerbert Xu *frag = skb_shinfo(skb)->frags[i]; 2601f4c50d99SHerbert Xu get_page(frag->page); 2602f4c50d99SHerbert Xu size = frag->size; 2603f4c50d99SHerbert Xu 2604f4c50d99SHerbert Xu if (pos < offset) { 2605f4c50d99SHerbert Xu frag->page_offset += offset - pos; 2606f4c50d99SHerbert Xu frag->size -= offset - pos; 2607f4c50d99SHerbert Xu } 2608f4c50d99SHerbert Xu 260989319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 2610f4c50d99SHerbert Xu 2611f4c50d99SHerbert Xu if (pos + size <= offset + len) { 2612f4c50d99SHerbert Xu i++; 2613f4c50d99SHerbert Xu pos += size; 2614f4c50d99SHerbert Xu } else { 2615f4c50d99SHerbert Xu frag->size -= pos + size - (offset + len); 261689319d38SHerbert Xu goto skip_fraglist; 2617f4c50d99SHerbert Xu } 2618f4c50d99SHerbert Xu 2619f4c50d99SHerbert Xu frag++; 2620f4c50d99SHerbert Xu } 2621f4c50d99SHerbert Xu 262289319d38SHerbert Xu if (pos < offset + len) { 262389319d38SHerbert Xu struct sk_buff *fskb2 = fskb; 262489319d38SHerbert Xu 262589319d38SHerbert Xu BUG_ON(pos + fskb->len != offset + len); 262689319d38SHerbert Xu 262789319d38SHerbert Xu pos += fskb->len; 262889319d38SHerbert Xu fskb = fskb->next; 262989319d38SHerbert Xu 263089319d38SHerbert Xu if (fskb2->next) { 263189319d38SHerbert Xu fskb2 = skb_clone(fskb2, GFP_ATOMIC); 263289319d38SHerbert Xu if (!fskb2) 263389319d38SHerbert Xu goto err; 263489319d38SHerbert Xu } else 263589319d38SHerbert Xu skb_get(fskb2); 263689319d38SHerbert Xu 2637fbb398a8SDavid S. Miller SKB_FRAG_ASSERT(nskb); 263889319d38SHerbert Xu skb_shinfo(nskb)->frag_list = fskb2; 263989319d38SHerbert Xu } 264089319d38SHerbert Xu 264189319d38SHerbert Xu skip_fraglist: 2642f4c50d99SHerbert Xu nskb->data_len = len - hsize; 2643f4c50d99SHerbert Xu nskb->len += nskb->data_len; 2644f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 2645f4c50d99SHerbert Xu } while ((offset += len) < skb->len); 2646f4c50d99SHerbert Xu 2647f4c50d99SHerbert Xu return segs; 2648f4c50d99SHerbert Xu 2649f4c50d99SHerbert Xu err: 2650f4c50d99SHerbert Xu while ((skb = segs)) { 2651f4c50d99SHerbert Xu segs = skb->next; 2652b08d5840SPatrick McHardy kfree_skb(skb); 2653f4c50d99SHerbert Xu } 2654f4c50d99SHerbert Xu return ERR_PTR(err); 2655f4c50d99SHerbert Xu } 2656f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 2657f4c50d99SHerbert Xu 265871d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 265971d93b39SHerbert Xu { 266071d93b39SHerbert Xu struct sk_buff *p = *head; 266171d93b39SHerbert Xu struct sk_buff *nskb; 26629aaa156cSHerbert Xu struct skb_shared_info *skbinfo = skb_shinfo(skb); 26639aaa156cSHerbert Xu struct skb_shared_info *pinfo = skb_shinfo(p); 266471d93b39SHerbert Xu unsigned int headroom; 266586911732SHerbert Xu unsigned int len = skb_gro_len(skb); 266667147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 266767147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 266871d93b39SHerbert Xu 266986911732SHerbert Xu if (p->len + len >= 65536) 267071d93b39SHerbert Xu return -E2BIG; 267171d93b39SHerbert Xu 26729aaa156cSHerbert Xu if (pinfo->frag_list) 267371d93b39SHerbert Xu goto merge; 267467147ba9SHerbert Xu else if (headlen <= offset) { 267542da6994SHerbert Xu skb_frag_t *frag; 267666e92fcfSHerbert Xu skb_frag_t *frag2; 26779aaa156cSHerbert Xu int i = skbinfo->nr_frags; 26789aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 267942da6994SHerbert Xu 268066e92fcfSHerbert Xu offset -= headlen; 268166e92fcfSHerbert Xu 268266e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 268381705ad1SHerbert Xu return -E2BIG; 268481705ad1SHerbert Xu 26859aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 26869aaa156cSHerbert Xu skbinfo->nr_frags = 0; 2687f5572068SHerbert Xu 26889aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 26899aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 269066e92fcfSHerbert Xu do { 269166e92fcfSHerbert Xu *--frag = *--frag2; 269266e92fcfSHerbert Xu } while (--i); 269366e92fcfSHerbert Xu 269466e92fcfSHerbert Xu frag->page_offset += offset; 269566e92fcfSHerbert Xu frag->size -= offset; 269666e92fcfSHerbert Xu 2697f5572068SHerbert Xu skb->truesize -= skb->data_len; 2698f5572068SHerbert Xu skb->len -= skb->data_len; 2699f5572068SHerbert Xu skb->data_len = 0; 2700f5572068SHerbert Xu 27015d38a079SHerbert Xu NAPI_GRO_CB(skb)->free = 1; 27025d38a079SHerbert Xu goto done; 270369c0cab1SHerbert Xu } else if (skb_gro_len(p) != pinfo->gso_size) 270469c0cab1SHerbert Xu return -E2BIG; 270571d93b39SHerbert Xu 270671d93b39SHerbert Xu headroom = skb_headroom(p); 27073d3be433SEric Dumazet nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); 270871d93b39SHerbert Xu if (unlikely(!nskb)) 270971d93b39SHerbert Xu return -ENOMEM; 271071d93b39SHerbert Xu 271171d93b39SHerbert Xu __copy_skb_header(nskb, p); 271271d93b39SHerbert Xu nskb->mac_len = p->mac_len; 271371d93b39SHerbert Xu 271471d93b39SHerbert Xu skb_reserve(nskb, headroom); 271586911732SHerbert Xu __skb_put(nskb, skb_gro_offset(p)); 271671d93b39SHerbert Xu 271786911732SHerbert Xu skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 271871d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 271971d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 272071d93b39SHerbert Xu 272186911732SHerbert Xu __skb_pull(p, skb_gro_offset(p)); 272286911732SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), 272386911732SHerbert Xu p->data - skb_mac_header(p)); 272471d93b39SHerbert Xu 272571d93b39SHerbert Xu *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p); 272671d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 27279aaa156cSHerbert Xu skb_shinfo(nskb)->gso_size = pinfo->gso_size; 2728622e0ca1SHerbert Xu pinfo->gso_size = 0; 272971d93b39SHerbert Xu skb_header_release(p); 273071d93b39SHerbert Xu nskb->prev = p; 273171d93b39SHerbert Xu 273271d93b39SHerbert Xu nskb->data_len += p->len; 273371d93b39SHerbert Xu nskb->truesize += p->len; 273471d93b39SHerbert Xu nskb->len += p->len; 273571d93b39SHerbert Xu 273671d93b39SHerbert Xu *head = nskb; 273771d93b39SHerbert Xu nskb->next = p->next; 273871d93b39SHerbert Xu p->next = NULL; 273971d93b39SHerbert Xu 274071d93b39SHerbert Xu p = nskb; 274171d93b39SHerbert Xu 274271d93b39SHerbert Xu merge: 274367147ba9SHerbert Xu if (offset > headlen) { 2744d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 2745d1dc7abfSMichal Schmidt 2746d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 2747d1dc7abfSMichal Schmidt skbinfo->frags[0].size -= eat; 2748d1dc7abfSMichal Schmidt skb->data_len -= eat; 2749d1dc7abfSMichal Schmidt skb->len -= eat; 275067147ba9SHerbert Xu offset = headlen; 275156035022SHerbert Xu } 275256035022SHerbert Xu 275367147ba9SHerbert Xu __skb_pull(skb, offset); 275456035022SHerbert Xu 275571d93b39SHerbert Xu p->prev->next = skb; 275671d93b39SHerbert Xu p->prev = skb; 275771d93b39SHerbert Xu skb_header_release(skb); 275871d93b39SHerbert Xu 27595d38a079SHerbert Xu done: 27605d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 276137fe4732SHerbert Xu p->data_len += len; 276237fe4732SHerbert Xu p->truesize += len; 276337fe4732SHerbert Xu p->len += len; 276471d93b39SHerbert Xu 276571d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 276671d93b39SHerbert Xu return 0; 276771d93b39SHerbert Xu } 276871d93b39SHerbert Xu EXPORT_SYMBOL_GPL(skb_gro_receive); 276971d93b39SHerbert Xu 27701da177e4SLinus Torvalds void __init skb_init(void) 27711da177e4SLinus Torvalds { 27721da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 27731da177e4SLinus Torvalds sizeof(struct sk_buff), 27741da177e4SLinus Torvalds 0, 2775e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 277620c2df83SPaul Mundt NULL); 2777d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 2778d179cd12SDavid S. Miller (2*sizeof(struct sk_buff)) + 2779d179cd12SDavid S. Miller sizeof(atomic_t), 2780d179cd12SDavid S. Miller 0, 2781e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 278220c2df83SPaul Mundt NULL); 27831da177e4SLinus Torvalds } 27841da177e4SLinus Torvalds 2785716ea3a7SDavid Howells /** 2786716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 2787716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 2788716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 2789716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 2790716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 2791716ea3a7SDavid Howells * 2792716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 2793716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 2794716ea3a7SDavid Howells */ 279551c739d1SDavid S. Miller static int 279651c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 2797716ea3a7SDavid Howells { 27981a028e50SDavid S. Miller int start = skb_headlen(skb); 27991a028e50SDavid S. Miller int i, copy = start - offset; 2800fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2801716ea3a7SDavid Howells int elt = 0; 2802716ea3a7SDavid Howells 2803716ea3a7SDavid Howells if (copy > 0) { 2804716ea3a7SDavid Howells if (copy > len) 2805716ea3a7SDavid Howells copy = len; 2806642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 2807716ea3a7SDavid Howells elt++; 2808716ea3a7SDavid Howells if ((len -= copy) == 0) 2809716ea3a7SDavid Howells return elt; 2810716ea3a7SDavid Howells offset += copy; 2811716ea3a7SDavid Howells } 2812716ea3a7SDavid Howells 2813716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 28141a028e50SDavid S. Miller int end; 2815716ea3a7SDavid Howells 2816547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28171a028e50SDavid S. Miller 28181a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 2819716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2820716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2821716ea3a7SDavid Howells 2822716ea3a7SDavid Howells if (copy > len) 2823716ea3a7SDavid Howells copy = len; 2824642f1490SJens Axboe sg_set_page(&sg[elt], frag->page, copy, 2825642f1490SJens Axboe frag->page_offset+offset-start); 2826716ea3a7SDavid Howells elt++; 2827716ea3a7SDavid Howells if (!(len -= copy)) 2828716ea3a7SDavid Howells return elt; 2829716ea3a7SDavid Howells offset += copy; 2830716ea3a7SDavid Howells } 28311a028e50SDavid S. Miller start = end; 2832716ea3a7SDavid Howells } 2833716ea3a7SDavid Howells 2834fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 28351a028e50SDavid S. Miller int end; 2836716ea3a7SDavid Howells 2837547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28381a028e50SDavid S. Miller 2839fbb398a8SDavid S. Miller end = start + frag_iter->len; 2840716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2841716ea3a7SDavid Howells if (copy > len) 2842716ea3a7SDavid Howells copy = len; 2843fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 284451c739d1SDavid S. Miller copy); 2845716ea3a7SDavid Howells if ((len -= copy) == 0) 2846716ea3a7SDavid Howells return elt; 2847716ea3a7SDavid Howells offset += copy; 2848716ea3a7SDavid Howells } 28491a028e50SDavid S. Miller start = end; 2850716ea3a7SDavid Howells } 2851716ea3a7SDavid Howells BUG_ON(len); 2852716ea3a7SDavid Howells return elt; 2853716ea3a7SDavid Howells } 2854716ea3a7SDavid Howells 285551c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 285651c739d1SDavid S. Miller { 285751c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 285851c739d1SDavid S. Miller 2859c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 286051c739d1SDavid S. Miller 286151c739d1SDavid S. Miller return nsg; 286251c739d1SDavid S. Miller } 2863b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 286451c739d1SDavid S. Miller 2865716ea3a7SDavid Howells /** 2866716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 2867716ea3a7SDavid Howells * @skb: The socket buffer to check. 2868716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 2869716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 2870716ea3a7SDavid Howells * 2871716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 2872716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 2873716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 2874716ea3a7SDavid Howells * 2875716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 2876716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 2877716ea3a7SDavid Howells * set to point to the skb in which this space begins. 2878716ea3a7SDavid Howells * 2879716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 2880716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 2881716ea3a7SDavid Howells */ 2882716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 2883716ea3a7SDavid Howells { 2884716ea3a7SDavid Howells int copyflag; 2885716ea3a7SDavid Howells int elt; 2886716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 2887716ea3a7SDavid Howells 2888716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 2889716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 2890716ea3a7SDavid Howells * at the moment even if they are anonymous). 2891716ea3a7SDavid Howells */ 2892716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 2893716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 2894716ea3a7SDavid Howells return -ENOMEM; 2895716ea3a7SDavid Howells 2896716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 289721dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 2898716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 2899716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 2900716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 2901716ea3a7SDavid Howells * space, 128 bytes is fair. */ 2902716ea3a7SDavid Howells 2903716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 2904716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 2905716ea3a7SDavid Howells return -ENOMEM; 2906716ea3a7SDavid Howells 2907716ea3a7SDavid Howells /* Voila! */ 2908716ea3a7SDavid Howells *trailer = skb; 2909716ea3a7SDavid Howells return 1; 2910716ea3a7SDavid Howells } 2911716ea3a7SDavid Howells 2912716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 2913716ea3a7SDavid Howells 2914716ea3a7SDavid Howells elt = 1; 2915716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 2916716ea3a7SDavid Howells copyflag = 0; 2917716ea3a7SDavid Howells 2918716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 2919716ea3a7SDavid Howells int ntail = 0; 2920716ea3a7SDavid Howells 2921716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 2922716ea3a7SDavid Howells * this can happen on input. Copy it and everything 2923716ea3a7SDavid Howells * after it. */ 2924716ea3a7SDavid Howells 2925716ea3a7SDavid Howells if (skb_shared(skb1)) 2926716ea3a7SDavid Howells copyflag = 1; 2927716ea3a7SDavid Howells 2928716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 2929716ea3a7SDavid Howells 2930716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 2931716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 293221dc3301SDavid S. Miller skb_has_frag_list(skb1) || 2933716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 2934716ea3a7SDavid Howells ntail = tailbits + 128; 2935716ea3a7SDavid Howells } 2936716ea3a7SDavid Howells 2937716ea3a7SDavid Howells if (copyflag || 2938716ea3a7SDavid Howells skb_cloned(skb1) || 2939716ea3a7SDavid Howells ntail || 2940716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 294121dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 2942716ea3a7SDavid Howells struct sk_buff *skb2; 2943716ea3a7SDavid Howells 2944716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 2945716ea3a7SDavid Howells if (ntail == 0) 2946716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 2947716ea3a7SDavid Howells else 2948716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 2949716ea3a7SDavid Howells skb_headroom(skb1), 2950716ea3a7SDavid Howells ntail, 2951716ea3a7SDavid Howells GFP_ATOMIC); 2952716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 2953716ea3a7SDavid Howells return -ENOMEM; 2954716ea3a7SDavid Howells 2955716ea3a7SDavid Howells if (skb1->sk) 2956716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 2957716ea3a7SDavid Howells 2958716ea3a7SDavid Howells /* Looking around. Are we still alive? 2959716ea3a7SDavid Howells * OK, link new skb, drop old one */ 2960716ea3a7SDavid Howells 2961716ea3a7SDavid Howells skb2->next = skb1->next; 2962716ea3a7SDavid Howells *skb_p = skb2; 2963716ea3a7SDavid Howells kfree_skb(skb1); 2964716ea3a7SDavid Howells skb1 = skb2; 2965716ea3a7SDavid Howells } 2966716ea3a7SDavid Howells elt++; 2967716ea3a7SDavid Howells *trailer = skb1; 2968716ea3a7SDavid Howells skb_p = &skb1->next; 2969716ea3a7SDavid Howells } 2970716ea3a7SDavid Howells 2971716ea3a7SDavid Howells return elt; 2972716ea3a7SDavid Howells } 2973b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 2974716ea3a7SDavid Howells 2975b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 2976b1faf566SEric Dumazet { 2977b1faf566SEric Dumazet struct sock *sk = skb->sk; 2978b1faf566SEric Dumazet 2979b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 2980b1faf566SEric Dumazet } 2981b1faf566SEric Dumazet 2982b1faf566SEric Dumazet /* 2983b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 2984b1faf566SEric Dumazet */ 2985b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 2986b1faf566SEric Dumazet { 2987b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 2988b1faf566SEric Dumazet (unsigned)sk->sk_rcvbuf) 2989b1faf566SEric Dumazet return -ENOMEM; 2990b1faf566SEric Dumazet 2991b1faf566SEric Dumazet skb_orphan(skb); 2992b1faf566SEric Dumazet skb->sk = sk; 2993b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 2994b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2995b1faf566SEric Dumazet 2996b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 2997b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 2998b1faf566SEric Dumazet sk->sk_data_ready(sk, skb->len); 2999b1faf566SEric Dumazet return 0; 3000b1faf566SEric Dumazet } 3001b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3002b1faf566SEric Dumazet 3003ac45f602SPatrick Ohly void skb_tstamp_tx(struct sk_buff *orig_skb, 3004ac45f602SPatrick Ohly struct skb_shared_hwtstamps *hwtstamps) 3005ac45f602SPatrick Ohly { 3006ac45f602SPatrick Ohly struct sock *sk = orig_skb->sk; 3007ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3008ac45f602SPatrick Ohly struct sk_buff *skb; 3009ac45f602SPatrick Ohly int err; 3010ac45f602SPatrick Ohly 3011ac45f602SPatrick Ohly if (!sk) 3012ac45f602SPatrick Ohly return; 3013ac45f602SPatrick Ohly 3014ac45f602SPatrick Ohly skb = skb_clone(orig_skb, GFP_ATOMIC); 3015ac45f602SPatrick Ohly if (!skb) 3016ac45f602SPatrick Ohly return; 3017ac45f602SPatrick Ohly 3018ac45f602SPatrick Ohly if (hwtstamps) { 3019ac45f602SPatrick Ohly *skb_hwtstamps(skb) = 3020ac45f602SPatrick Ohly *hwtstamps; 3021ac45f602SPatrick Ohly } else { 3022ac45f602SPatrick Ohly /* 3023ac45f602SPatrick Ohly * no hardware time stamps available, 30242244d07bSOliver Hartkopp * so keep the shared tx_flags and only 3025ac45f602SPatrick Ohly * store software time stamp 3026ac45f602SPatrick Ohly */ 3027ac45f602SPatrick Ohly skb->tstamp = ktime_get_real(); 3028ac45f602SPatrick Ohly } 3029ac45f602SPatrick Ohly 3030ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3031ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3032ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3033ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 303429030374SEric Dumazet 3035ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 303629030374SEric Dumazet 3037ac45f602SPatrick Ohly if (err) 3038ac45f602SPatrick Ohly kfree_skb(skb); 3039ac45f602SPatrick Ohly } 3040ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3041ac45f602SPatrick Ohly 3042ac45f602SPatrick Ohly 3043f35d9d8aSRusty Russell /** 3044f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3045f35d9d8aSRusty Russell * @skb: the skb to set 3046f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3047f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3048f35d9d8aSRusty Russell * 3049f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3050f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3051f35d9d8aSRusty Russell * 3052f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3053f35d9d8aSRusty Russell * returns false you should drop the packet. 3054f35d9d8aSRusty Russell */ 3055f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3056f35d9d8aSRusty Russell { 30575ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 30585ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3059f35d9d8aSRusty Russell if (net_ratelimit()) 3060f35d9d8aSRusty Russell printk(KERN_WARNING 3061f35d9d8aSRusty Russell "bad partial csum: csum=%u/%u len=%u\n", 30625ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3063f35d9d8aSRusty Russell return false; 3064f35d9d8aSRusty Russell } 3065f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3066f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3067f35d9d8aSRusty Russell skb->csum_offset = off; 3068f35d9d8aSRusty Russell return true; 3069f35d9d8aSRusty Russell } 3070b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3071f35d9d8aSRusty Russell 30724497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 30734497b076SBen Hutchings { 30744497b076SBen Hutchings if (net_ratelimit()) 30754497b076SBen Hutchings pr_warning("%s: received packets cannot be forwarded" 30764497b076SBen Hutchings " while LRO is enabled\n", skb->dev->name); 30774497b076SBen Hutchings } 30784497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 3079