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> 60268bb0ceSLinus Torvalds #include <linux/prefetch.h> 611da177e4SLinus Torvalds 621da177e4SLinus Torvalds #include <net/protocol.h> 631da177e4SLinus Torvalds #include <net/dst.h> 641da177e4SLinus Torvalds #include <net/sock.h> 651da177e4SLinus Torvalds #include <net/checksum.h> 661da177e4SLinus Torvalds #include <net/xfrm.h> 671da177e4SLinus Torvalds 681da177e4SLinus Torvalds #include <asm/uaccess.h> 691da177e4SLinus Torvalds #include <asm/system.h> 70ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 711da177e4SLinus Torvalds 72a1f8e7f7SAl Viro #include "kmap_skb.h" 73a1f8e7f7SAl Viro 74e18b890bSChristoph Lameter static struct kmem_cache *skbuff_head_cache __read_mostly; 75e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 761da177e4SLinus Torvalds 779c55e01cSJens Axboe static void sock_pipe_buf_release(struct pipe_inode_info *pipe, 789c55e01cSJens Axboe struct pipe_buffer *buf) 799c55e01cSJens Axboe { 808b9d3728SJarek Poplawski put_page(buf->page); 819c55e01cSJens Axboe } 829c55e01cSJens Axboe 839c55e01cSJens Axboe static void sock_pipe_buf_get(struct pipe_inode_info *pipe, 849c55e01cSJens Axboe struct pipe_buffer *buf) 859c55e01cSJens Axboe { 868b9d3728SJarek Poplawski get_page(buf->page); 879c55e01cSJens Axboe } 889c55e01cSJens Axboe 899c55e01cSJens Axboe static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, 909c55e01cSJens Axboe struct pipe_buffer *buf) 919c55e01cSJens Axboe { 929c55e01cSJens Axboe return 1; 939c55e01cSJens Axboe } 949c55e01cSJens Axboe 959c55e01cSJens Axboe 969c55e01cSJens Axboe /* Pipe buffer operations for a socket. */ 9728dfef8fSAlexey Dobriyan static const struct pipe_buf_operations sock_pipe_buf_ops = { 989c55e01cSJens Axboe .can_merge = 0, 999c55e01cSJens Axboe .map = generic_pipe_buf_map, 1009c55e01cSJens Axboe .unmap = generic_pipe_buf_unmap, 1019c55e01cSJens Axboe .confirm = generic_pipe_buf_confirm, 1029c55e01cSJens Axboe .release = sock_pipe_buf_release, 1039c55e01cSJens Axboe .steal = sock_pipe_buf_steal, 1049c55e01cSJens Axboe .get = sock_pipe_buf_get, 1059c55e01cSJens Axboe }; 1069c55e01cSJens Axboe 1071da177e4SLinus Torvalds /* 1081da177e4SLinus Torvalds * Keep out-of-line to prevent kernel bloat. 1091da177e4SLinus Torvalds * __builtin_return_address is not used because it is not always 1101da177e4SLinus Torvalds * reliable. 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds /** 1141da177e4SLinus Torvalds * skb_over_panic - private function 1151da177e4SLinus Torvalds * @skb: buffer 1161da177e4SLinus Torvalds * @sz: size 1171da177e4SLinus Torvalds * @here: address 1181da177e4SLinus Torvalds * 1191da177e4SLinus Torvalds * Out of line support code for skb_put(). Not user callable. 1201da177e4SLinus Torvalds */ 121ccb7c773SRami Rosen static void skb_over_panic(struct sk_buff *skb, int sz, void *here) 1221da177e4SLinus Torvalds { 12326095455SPatrick McHardy printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " 1244305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 12527a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1264305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 12726095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1281da177e4SLinus Torvalds BUG(); 1291da177e4SLinus Torvalds } 1301da177e4SLinus Torvalds 1311da177e4SLinus Torvalds /** 1321da177e4SLinus Torvalds * skb_under_panic - private function 1331da177e4SLinus Torvalds * @skb: buffer 1341da177e4SLinus Torvalds * @sz: size 1351da177e4SLinus Torvalds * @here: address 1361da177e4SLinus Torvalds * 1371da177e4SLinus Torvalds * Out of line support code for skb_push(). Not user callable. 1381da177e4SLinus Torvalds */ 1391da177e4SLinus Torvalds 140ccb7c773SRami Rosen static void skb_under_panic(struct sk_buff *skb, int sz, void *here) 1411da177e4SLinus Torvalds { 14226095455SPatrick McHardy printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " 1434305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 14427a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1454305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 14626095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1471da177e4SLinus Torvalds BUG(); 1481da177e4SLinus Torvalds } 1491da177e4SLinus Torvalds 1501da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1511da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1521da177e4SLinus Torvalds * [BEEP] leaks. 1531da177e4SLinus Torvalds * 1541da177e4SLinus Torvalds */ 1551da177e4SLinus Torvalds 1561da177e4SLinus Torvalds /** 157d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1581da177e4SLinus Torvalds * @size: size to allocate 1591da177e4SLinus Torvalds * @gfp_mask: allocation mask 160c83c2486SRandy Dunlap * @fclone: allocate from fclone cache instead of head cache 161c83c2486SRandy Dunlap * and allocate a cloned (child) skb 162b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1631da177e4SLinus Torvalds * 1641da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 1651da177e4SLinus Torvalds * tail room of size bytes. The object has a reference count of one. 1661da177e4SLinus Torvalds * The return is the buffer. On a failure the return is %NULL. 1671da177e4SLinus Torvalds * 1681da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1691da177e4SLinus Torvalds * %GFP_ATOMIC. 1701da177e4SLinus Torvalds */ 171dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 172b30973f8SChristoph Hellwig int fclone, int node) 1731da177e4SLinus Torvalds { 174e18b890bSChristoph Lameter struct kmem_cache *cache; 1754947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1761da177e4SLinus Torvalds struct sk_buff *skb; 1771da177e4SLinus Torvalds u8 *data; 1781da177e4SLinus Torvalds 1798798b3fbSHerbert Xu cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; 1808798b3fbSHerbert Xu 1811da177e4SLinus Torvalds /* Get the HEAD */ 182b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1831da177e4SLinus Torvalds if (!skb) 1841da177e4SLinus Torvalds goto out; 185ec7d2f2cSEric Dumazet prefetchw(skb); 1861da177e4SLinus Torvalds 18787fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 18887fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 18987fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 19087fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 19187fb4b7bSEric Dumazet */ 19287fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 19387fb4b7bSEric Dumazet data = kmalloc_node_track_caller(size, gfp_mask, node); 1941da177e4SLinus Torvalds if (!data) 1951da177e4SLinus Torvalds goto nodata; 19687fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 19787fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 19887fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 19987fb4b7bSEric Dumazet */ 20087fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 201ec7d2f2cSEric Dumazet prefetchw(data + size); 2021da177e4SLinus Torvalds 203ca0605a7SArnaldo Carvalho de Melo /* 204c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 205c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 206c8005785SJohannes Berg * the tail pointer in struct sk_buff! 207ca0605a7SArnaldo Carvalho de Melo */ 208ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 20987fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 21087fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 2111da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2121da177e4SLinus Torvalds skb->head = data; 2131da177e4SLinus Torvalds skb->data = data; 21427a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2154305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 21619633e12SStephen Hemminger #ifdef NET_SKBUFF_DATA_USES_OFFSET 21719633e12SStephen Hemminger skb->mac_header = ~0U; 21819633e12SStephen Hemminger #endif 21919633e12SStephen Hemminger 2204947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2214947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 222ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2234947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 224c2aa3665SEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 2254947d3efSBenjamin LaHaise 226d179cd12SDavid S. Miller if (fclone) { 227d179cd12SDavid S. Miller struct sk_buff *child = skb + 1; 228d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (child + 1); 2291da177e4SLinus Torvalds 230fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags1); 231fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags2); 232d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 233d179cd12SDavid S. Miller atomic_set(fclone_ref, 1); 234d179cd12SDavid S. Miller 235d179cd12SDavid S. Miller child->fclone = SKB_FCLONE_UNAVAILABLE; 236d179cd12SDavid S. Miller } 2371da177e4SLinus Torvalds out: 2381da177e4SLinus Torvalds return skb; 2391da177e4SLinus Torvalds nodata: 2408798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2411da177e4SLinus Torvalds skb = NULL; 2421da177e4SLinus Torvalds goto out; 2431da177e4SLinus Torvalds } 244b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2451da177e4SLinus Torvalds 2461da177e4SLinus Torvalds /** 2478af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 2488af27456SChristoph Hellwig * @dev: network device to receive on 2498af27456SChristoph Hellwig * @length: length to allocate 2508af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 2518af27456SChristoph Hellwig * 2528af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 2538af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 2548af27456SChristoph Hellwig * the headroom they think they need without accounting for the 2558af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 2568af27456SChristoph Hellwig * 2578af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 2588af27456SChristoph Hellwig */ 2598af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 2608af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 2618af27456SChristoph Hellwig { 2628af27456SChristoph Hellwig struct sk_buff *skb; 2638af27456SChristoph Hellwig 264564824b0SEric Dumazet skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE); 2657b2e497aSChristoph Hellwig if (likely(skb)) { 2668af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 2677b2e497aSChristoph Hellwig skb->dev = dev; 2687b2e497aSChristoph Hellwig } 2698af27456SChristoph Hellwig return skb; 2708af27456SChristoph Hellwig } 271b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 2721da177e4SLinus Torvalds 273654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 274654bed16SPeter Zijlstra int size) 275654bed16SPeter Zijlstra { 276654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 277654bed16SPeter Zijlstra skb->len += size; 278654bed16SPeter Zijlstra skb->data_len += size; 279654bed16SPeter Zijlstra skb->truesize += size; 280654bed16SPeter Zijlstra } 281654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 282654bed16SPeter Zijlstra 283f58518e6SIlpo Järvinen /** 284f58518e6SIlpo Järvinen * dev_alloc_skb - allocate an skbuff for receiving 285f58518e6SIlpo Järvinen * @length: length to allocate 286f58518e6SIlpo Järvinen * 287f58518e6SIlpo Järvinen * Allocate a new &sk_buff and assign it a usage count of one. The 288f58518e6SIlpo Järvinen * buffer has unspecified headroom built in. Users should allocate 289f58518e6SIlpo Järvinen * the headroom they think they need without accounting for the 290f58518e6SIlpo Järvinen * built in space. The built in space is used for optimisations. 291f58518e6SIlpo Järvinen * 292f58518e6SIlpo Järvinen * %NULL is returned if there is no free memory. Although this function 293f58518e6SIlpo Järvinen * allocates memory it can be called from an interrupt. 294f58518e6SIlpo Järvinen */ 295f58518e6SIlpo Järvinen struct sk_buff *dev_alloc_skb(unsigned int length) 296f58518e6SIlpo Järvinen { 2971483b874SDenys Vlasenko /* 2981483b874SDenys Vlasenko * There is more code here than it seems: 299a0f55e0eSDavid S. Miller * __dev_alloc_skb is an inline 3001483b874SDenys Vlasenko */ 301f58518e6SIlpo Järvinen return __dev_alloc_skb(length, GFP_ATOMIC); 302f58518e6SIlpo Järvinen } 303f58518e6SIlpo Järvinen EXPORT_SYMBOL(dev_alloc_skb); 304f58518e6SIlpo Järvinen 30527b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 3061da177e4SLinus Torvalds { 30727b437c8SHerbert Xu struct sk_buff *list = *listp; 3081da177e4SLinus Torvalds 30927b437c8SHerbert Xu *listp = NULL; 3101da177e4SLinus Torvalds 3111da177e4SLinus Torvalds do { 3121da177e4SLinus Torvalds struct sk_buff *this = list; 3131da177e4SLinus Torvalds list = list->next; 3141da177e4SLinus Torvalds kfree_skb(this); 3151da177e4SLinus Torvalds } while (list); 3161da177e4SLinus Torvalds } 3171da177e4SLinus Torvalds 31827b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 31927b437c8SHerbert Xu { 32027b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 32127b437c8SHerbert Xu } 32227b437c8SHerbert Xu 3231da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 3241da177e4SLinus Torvalds { 3251da177e4SLinus Torvalds struct sk_buff *list; 3261da177e4SLinus Torvalds 327fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 3281da177e4SLinus Torvalds skb_get(list); 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 3315bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 3321da177e4SLinus Torvalds { 3331da177e4SLinus Torvalds if (!skb->cloned || 3341da177e4SLinus Torvalds !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 3351da177e4SLinus Torvalds &skb_shinfo(skb)->dataref)) { 3361da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 3371da177e4SLinus Torvalds int i; 3381da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 339ea2ab693SIan Campbell skb_frag_unref(skb, i); 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 342a6686f2fSShirley Ma /* 343a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 344a6686f2fSShirley Ma * the lower device DMA has done; 345a6686f2fSShirley Ma */ 346a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 347a6686f2fSShirley Ma struct ubuf_info *uarg; 348a6686f2fSShirley Ma 349a6686f2fSShirley Ma uarg = skb_shinfo(skb)->destructor_arg; 350a6686f2fSShirley Ma if (uarg->callback) 351a6686f2fSShirley Ma uarg->callback(uarg); 352a6686f2fSShirley Ma } 353a6686f2fSShirley Ma 35421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 3551da177e4SLinus Torvalds skb_drop_fraglist(skb); 3561da177e4SLinus Torvalds 3571da177e4SLinus Torvalds kfree(skb->head); 3581da177e4SLinus Torvalds } 3591da177e4SLinus Torvalds } 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds /* 3621da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 3631da177e4SLinus Torvalds */ 3642d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 3651da177e4SLinus Torvalds { 366d179cd12SDavid S. Miller struct sk_buff *other; 367d179cd12SDavid S. Miller atomic_t *fclone_ref; 368d179cd12SDavid S. Miller 369d179cd12SDavid S. Miller switch (skb->fclone) { 370d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 3711da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 372d179cd12SDavid S. Miller break; 373d179cd12SDavid S. Miller 374d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 375d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 2); 376d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 377d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, skb); 378d179cd12SDavid S. Miller break; 379d179cd12SDavid S. Miller 380d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 381d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 1); 382d179cd12SDavid S. Miller other = skb - 1; 383d179cd12SDavid S. Miller 384d179cd12SDavid S. Miller /* The clone portion is available for 385d179cd12SDavid S. Miller * fast-cloning again. 386d179cd12SDavid S. Miller */ 387d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_UNAVAILABLE; 388d179cd12SDavid S. Miller 389d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 390d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, other); 391d179cd12SDavid S. Miller break; 3923ff50b79SStephen Hemminger } 3931da177e4SLinus Torvalds } 3941da177e4SLinus Torvalds 39504a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 3961da177e4SLinus Torvalds { 397adf30907SEric Dumazet skb_dst_drop(skb); 3981da177e4SLinus Torvalds #ifdef CONFIG_XFRM 3991da177e4SLinus Torvalds secpath_put(skb->sp); 4001da177e4SLinus Torvalds #endif 4011da177e4SLinus Torvalds if (skb->destructor) { 4029c2b3328SStephen Hemminger WARN_ON(in_irq()); 4031da177e4SLinus Torvalds skb->destructor(skb); 4041da177e4SLinus Torvalds } 4059fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4065f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 4072fc72c7bSKOVACS Krisztian #endif 4082fc72c7bSKOVACS Krisztian #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 4099fb9cbb1SYasuyuki Kozakai nf_conntrack_put_reasm(skb->nfct_reasm); 4109fb9cbb1SYasuyuki Kozakai #endif 4111da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 4121da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 4131da177e4SLinus Torvalds #endif 4141da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 4151da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 4161da177e4SLinus Torvalds skb->tc_index = 0; 4171da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 4181da177e4SLinus Torvalds skb->tc_verd = 0; 4191da177e4SLinus Torvalds #endif 4201da177e4SLinus Torvalds #endif 42104a4bb55SLennert Buytenhek } 42204a4bb55SLennert Buytenhek 42304a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 42404a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 42504a4bb55SLennert Buytenhek { 42604a4bb55SLennert Buytenhek skb_release_head_state(skb); 4272d4baff8SHerbert Xu skb_release_data(skb); 4282d4baff8SHerbert Xu } 4291da177e4SLinus Torvalds 4302d4baff8SHerbert Xu /** 4312d4baff8SHerbert Xu * __kfree_skb - private function 4322d4baff8SHerbert Xu * @skb: buffer 4332d4baff8SHerbert Xu * 4342d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 4352d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 4362d4baff8SHerbert Xu * always call kfree_skb 4372d4baff8SHerbert Xu */ 4382d4baff8SHerbert Xu 4392d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 4402d4baff8SHerbert Xu { 4412d4baff8SHerbert Xu skb_release_all(skb); 4421da177e4SLinus Torvalds kfree_skbmem(skb); 4431da177e4SLinus Torvalds } 444b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 4451da177e4SLinus Torvalds 4461da177e4SLinus Torvalds /** 447231d06aeSJörn Engel * kfree_skb - free an sk_buff 448231d06aeSJörn Engel * @skb: buffer to free 449231d06aeSJörn Engel * 450231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 451231d06aeSJörn Engel * hit zero. 452231d06aeSJörn Engel */ 453231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 454231d06aeSJörn Engel { 455231d06aeSJörn Engel if (unlikely(!skb)) 456231d06aeSJörn Engel return; 457231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 458231d06aeSJörn Engel smp_rmb(); 459231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 460231d06aeSJörn Engel return; 461ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 462231d06aeSJörn Engel __kfree_skb(skb); 463231d06aeSJörn Engel } 464b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 465231d06aeSJörn Engel 466d1a203eaSStephen Hemminger /** 467ead2ceb0SNeil Horman * consume_skb - free an skbuff 468ead2ceb0SNeil Horman * @skb: buffer to free 469ead2ceb0SNeil Horman * 470ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 471ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 472ead2ceb0SNeil Horman * is being dropped after a failure and notes that 473ead2ceb0SNeil Horman */ 474ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 475ead2ceb0SNeil Horman { 476ead2ceb0SNeil Horman if (unlikely(!skb)) 477ead2ceb0SNeil Horman return; 478ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 479ead2ceb0SNeil Horman smp_rmb(); 480ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 481ead2ceb0SNeil Horman return; 48207dc22e7SKoki Sanagi trace_consume_skb(skb); 483ead2ceb0SNeil Horman __kfree_skb(skb); 484ead2ceb0SNeil Horman } 485ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 486ead2ceb0SNeil Horman 487ead2ceb0SNeil Horman /** 488d1a203eaSStephen Hemminger * skb_recycle_check - check if skb can be reused for receive 489d1a203eaSStephen Hemminger * @skb: buffer 490d1a203eaSStephen Hemminger * @skb_size: minimum receive buffer size 491d1a203eaSStephen Hemminger * 492d1a203eaSStephen Hemminger * Checks that the skb passed in is not shared or cloned, and 493d1a203eaSStephen Hemminger * that it is linear and its head portion at least as large as 494d1a203eaSStephen Hemminger * skb_size so that it can be recycled as a receive buffer. 495d1a203eaSStephen Hemminger * If these conditions are met, this function does any necessary 496d1a203eaSStephen Hemminger * reference count dropping and cleans up the skbuff as if it 497d1a203eaSStephen Hemminger * just came from __alloc_skb(). 498d1a203eaSStephen Hemminger */ 4995b0daa34SChangli Gao bool skb_recycle_check(struct sk_buff *skb, int skb_size) 50004a4bb55SLennert Buytenhek { 50104a4bb55SLennert Buytenhek struct skb_shared_info *shinfo; 50204a4bb55SLennert Buytenhek 503e84af6ddSAnton Vorontsov if (irqs_disabled()) 5045b0daa34SChangli Gao return false; 505e84af6ddSAnton Vorontsov 506a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) 507a6686f2fSShirley Ma return false; 508a6686f2fSShirley Ma 50904a4bb55SLennert Buytenhek if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE) 5105b0daa34SChangli Gao return false; 51104a4bb55SLennert Buytenhek 51204a4bb55SLennert Buytenhek skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD); 51304a4bb55SLennert Buytenhek if (skb_end_pointer(skb) - skb->head < skb_size) 5145b0daa34SChangli Gao return false; 51504a4bb55SLennert Buytenhek 51604a4bb55SLennert Buytenhek if (skb_shared(skb) || skb_cloned(skb)) 5175b0daa34SChangli Gao return false; 51804a4bb55SLennert Buytenhek 51904a4bb55SLennert Buytenhek skb_release_head_state(skb); 520ec7d2f2cSEric Dumazet 52104a4bb55SLennert Buytenhek shinfo = skb_shinfo(skb); 522ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 52304a4bb55SLennert Buytenhek atomic_set(&shinfo->dataref, 1); 52404a4bb55SLennert Buytenhek 52504a4bb55SLennert Buytenhek memset(skb, 0, offsetof(struct sk_buff, tail)); 52604a4bb55SLennert Buytenhek skb->data = skb->head + NET_SKB_PAD; 5275cd33db2SLennert Buytenhek skb_reset_tail_pointer(skb); 52804a4bb55SLennert Buytenhek 5295b0daa34SChangli Gao return true; 53004a4bb55SLennert Buytenhek } 53104a4bb55SLennert Buytenhek EXPORT_SYMBOL(skb_recycle_check); 53204a4bb55SLennert Buytenhek 533dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 534dec18810SHerbert Xu { 535dec18810SHerbert Xu new->tstamp = old->tstamp; 536dec18810SHerbert Xu new->dev = old->dev; 537dec18810SHerbert Xu new->transport_header = old->transport_header; 538dec18810SHerbert Xu new->network_header = old->network_header; 539dec18810SHerbert Xu new->mac_header = old->mac_header; 5407fee226aSEric Dumazet skb_dst_copy(new, old); 5410a9627f2STom Herbert new->rxhash = old->rxhash; 5426461be3aSChangli Gao new->ooo_okay = old->ooo_okay; 543bdeab991STom Herbert new->l4_rxhash = old->l4_rxhash; 544def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 545dec18810SHerbert Xu new->sp = secpath_get(old->sp); 546dec18810SHerbert Xu #endif 547dec18810SHerbert Xu memcpy(new->cb, old->cb, sizeof(old->cb)); 5489bcb97caSHerbert Xu new->csum = old->csum; 549dec18810SHerbert Xu new->local_df = old->local_df; 550dec18810SHerbert Xu new->pkt_type = old->pkt_type; 551dec18810SHerbert Xu new->ip_summed = old->ip_summed; 552dec18810SHerbert Xu skb_copy_queue_mapping(new, old); 553dec18810SHerbert Xu new->priority = old->priority; 554dec18810SHerbert Xu #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 555dec18810SHerbert Xu new->ipvs_property = old->ipvs_property; 556dec18810SHerbert Xu #endif 557dec18810SHerbert Xu new->protocol = old->protocol; 558dec18810SHerbert Xu new->mark = old->mark; 5598964be4aSEric Dumazet new->skb_iif = old->skb_iif; 560dec18810SHerbert Xu __nf_copy(new, old); 561dec18810SHerbert Xu #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ 562dec18810SHerbert Xu defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) 563dec18810SHerbert Xu new->nf_trace = old->nf_trace; 564dec18810SHerbert Xu #endif 565dec18810SHerbert Xu #ifdef CONFIG_NET_SCHED 566dec18810SHerbert Xu new->tc_index = old->tc_index; 567dec18810SHerbert Xu #ifdef CONFIG_NET_CLS_ACT 568dec18810SHerbert Xu new->tc_verd = old->tc_verd; 569dec18810SHerbert Xu #endif 570dec18810SHerbert Xu #endif 5716aa895b0SPatrick McHardy new->vlan_tci = old->vlan_tci; 5726aa895b0SPatrick McHardy 573dec18810SHerbert Xu skb_copy_secmark(new, old); 574dec18810SHerbert Xu } 575dec18810SHerbert Xu 57682c49a35SHerbert Xu /* 57782c49a35SHerbert Xu * You should not add any new code to this function. Add it to 57882c49a35SHerbert Xu * __copy_skb_header above instead. 57982c49a35SHerbert Xu */ 580e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 5811da177e4SLinus Torvalds { 5821da177e4SLinus Torvalds #define C(x) n->x = skb->x 5831da177e4SLinus Torvalds 5841da177e4SLinus Torvalds n->next = n->prev = NULL; 5851da177e4SLinus Torvalds n->sk = NULL; 586dec18810SHerbert Xu __copy_skb_header(n, skb); 587dec18810SHerbert Xu 5881da177e4SLinus Torvalds C(len); 5891da177e4SLinus Torvalds C(data_len); 5903e6b3b2eSAlexey Dobriyan C(mac_len); 591334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 59202f1c89dSPaul Moore n->cloned = 1; 5931da177e4SLinus Torvalds n->nohdr = 0; 5941da177e4SLinus Torvalds n->destructor = NULL; 5951da177e4SLinus Torvalds C(tail); 5961da177e4SLinus Torvalds C(end); 59702f1c89dSPaul Moore C(head); 59802f1c89dSPaul Moore C(data); 59902f1c89dSPaul Moore C(truesize); 60002f1c89dSPaul Moore atomic_set(&n->users, 1); 6011da177e4SLinus Torvalds 6021da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 6031da177e4SLinus Torvalds skb->cloned = 1; 6041da177e4SLinus Torvalds 6051da177e4SLinus Torvalds return n; 606e0053ec0SHerbert Xu #undef C 607e0053ec0SHerbert Xu } 608e0053ec0SHerbert Xu 609e0053ec0SHerbert Xu /** 610e0053ec0SHerbert Xu * skb_morph - morph one skb into another 611e0053ec0SHerbert Xu * @dst: the skb to receive the contents 612e0053ec0SHerbert Xu * @src: the skb to supply the contents 613e0053ec0SHerbert Xu * 614e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 615e0053ec0SHerbert Xu * supplied by the user. 616e0053ec0SHerbert Xu * 617e0053ec0SHerbert Xu * The target skb is returned upon exit. 618e0053ec0SHerbert Xu */ 619e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 620e0053ec0SHerbert Xu { 6212d4baff8SHerbert Xu skb_release_all(dst); 622e0053ec0SHerbert Xu return __skb_clone(dst, src); 623e0053ec0SHerbert Xu } 624e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 625e0053ec0SHerbert Xu 62648c83012SMichael S. Tsirkin /* skb_copy_ubufs - copy userspace skb frags buffers to kernel 62748c83012SMichael S. Tsirkin * @skb: the skb to modify 62848c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 62948c83012SMichael S. Tsirkin * 63048c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 63148c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 63248c83012SMichael S. Tsirkin * to userspace pages. 63348c83012SMichael S. Tsirkin * 63448c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 63548c83012SMichael S. Tsirkin * %GFP_ATOMIC. 63648c83012SMichael S. Tsirkin * 63748c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 63848c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 63948c83012SMichael S. Tsirkin */ 64048c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 641a6686f2fSShirley Ma { 642a6686f2fSShirley Ma int i; 643a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 644a6686f2fSShirley Ma struct page *page, *head = NULL; 645a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 646a6686f2fSShirley Ma 647a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 648a6686f2fSShirley Ma u8 *vaddr; 649a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 650a6686f2fSShirley Ma 651a6686f2fSShirley Ma page = alloc_page(GFP_ATOMIC); 652a6686f2fSShirley Ma if (!page) { 653a6686f2fSShirley Ma while (head) { 654a6686f2fSShirley Ma struct page *next = (struct page *)head->private; 655a6686f2fSShirley Ma put_page(head); 656a6686f2fSShirley Ma head = next; 657a6686f2fSShirley Ma } 658a6686f2fSShirley Ma return -ENOMEM; 659a6686f2fSShirley Ma } 660a6686f2fSShirley Ma vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]); 661a6686f2fSShirley Ma memcpy(page_address(page), 662a6686f2fSShirley Ma vaddr + f->page_offset, f->size); 663a6686f2fSShirley Ma kunmap_skb_frag(vaddr); 664a6686f2fSShirley Ma page->private = (unsigned long)head; 665a6686f2fSShirley Ma head = page; 666a6686f2fSShirley Ma } 667a6686f2fSShirley Ma 668a6686f2fSShirley Ma /* skb frags release userspace buffers */ 669a6686f2fSShirley Ma for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 670a6686f2fSShirley Ma put_page(skb_shinfo(skb)->frags[i].page); 671a6686f2fSShirley Ma 672a6686f2fSShirley Ma uarg->callback(uarg); 673a6686f2fSShirley Ma 674a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 675a6686f2fSShirley Ma for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) { 676a6686f2fSShirley Ma skb_shinfo(skb)->frags[i - 1].page_offset = 0; 677a6686f2fSShirley Ma skb_shinfo(skb)->frags[i - 1].page = head; 678a6686f2fSShirley Ma head = (struct page *)head->private; 679a6686f2fSShirley Ma } 68048c83012SMichael S. Tsirkin 68148c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 682a6686f2fSShirley Ma return 0; 683a6686f2fSShirley Ma } 684a6686f2fSShirley Ma 685a6686f2fSShirley Ma 686e0053ec0SHerbert Xu /** 687e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 688e0053ec0SHerbert Xu * @skb: buffer to clone 689e0053ec0SHerbert Xu * @gfp_mask: allocation priority 690e0053ec0SHerbert Xu * 691e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 692e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 693e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 694e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 695e0053ec0SHerbert Xu * 696e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 697e0053ec0SHerbert Xu * %GFP_ATOMIC. 698e0053ec0SHerbert Xu */ 699e0053ec0SHerbert Xu 700e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 701e0053ec0SHerbert Xu { 702e0053ec0SHerbert Xu struct sk_buff *n; 703e0053ec0SHerbert Xu 704a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 705a6686f2fSShirley Ma if (skb_copy_ubufs(skb, gfp_mask)) 706a6686f2fSShirley Ma return NULL; 707a6686f2fSShirley Ma } 708a6686f2fSShirley Ma 709e0053ec0SHerbert Xu n = skb + 1; 710e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 711e0053ec0SHerbert Xu n->fclone == SKB_FCLONE_UNAVAILABLE) { 712e0053ec0SHerbert Xu atomic_t *fclone_ref = (atomic_t *) (n + 1); 713e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_CLONE; 714e0053ec0SHerbert Xu atomic_inc(fclone_ref); 715e0053ec0SHerbert Xu } else { 716e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 717e0053ec0SHerbert Xu if (!n) 718e0053ec0SHerbert Xu return NULL; 719fe55f6d5SVegard Nossum 720fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 721fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags2); 722e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 723e0053ec0SHerbert Xu } 724e0053ec0SHerbert Xu 725e0053ec0SHerbert Xu return __skb_clone(n, skb); 7261da177e4SLinus Torvalds } 727b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 7281da177e4SLinus Torvalds 7291da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 7301da177e4SLinus Torvalds { 7312e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 7321da177e4SLinus Torvalds /* 7331da177e4SLinus Torvalds * Shift between the two data areas in bytes 7341da177e4SLinus Torvalds */ 7351da177e4SLinus Torvalds unsigned long offset = new->data - old->data; 7362e07fa9cSArnaldo Carvalho de Melo #endif 737dec18810SHerbert Xu 738dec18810SHerbert Xu __copy_skb_header(new, old); 739dec18810SHerbert Xu 7402e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 7412e07fa9cSArnaldo Carvalho de Melo /* {transport,network,mac}_header are relative to skb->head */ 7422e07fa9cSArnaldo Carvalho de Melo new->transport_header += offset; 7432e07fa9cSArnaldo Carvalho de Melo new->network_header += offset; 744603a8bbeSStephen Hemminger if (skb_mac_header_was_set(new)) 7452e07fa9cSArnaldo Carvalho de Melo new->mac_header += offset; 7462e07fa9cSArnaldo Carvalho de Melo #endif 7477967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 7487967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 7497967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 7501da177e4SLinus Torvalds } 7511da177e4SLinus Torvalds 7521da177e4SLinus Torvalds /** 7531da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 7541da177e4SLinus Torvalds * @skb: buffer to copy 7551da177e4SLinus Torvalds * @gfp_mask: allocation priority 7561da177e4SLinus Torvalds * 7571da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 7581da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 7591da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 7601da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 7611da177e4SLinus Torvalds * 7621da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 7631da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 7641da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 7651da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 7661da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 7671da177e4SLinus Torvalds */ 7681da177e4SLinus Torvalds 769dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 7701da177e4SLinus Torvalds { 7716602cebbSEric Dumazet int headerlen = skb_headroom(skb); 7726602cebbSEric Dumazet unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len; 7736602cebbSEric Dumazet struct sk_buff *n = alloc_skb(size, gfp_mask); 7746602cebbSEric Dumazet 7751da177e4SLinus Torvalds if (!n) 7761da177e4SLinus Torvalds return NULL; 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds /* Set the data pointer */ 7791da177e4SLinus Torvalds skb_reserve(n, headerlen); 7801da177e4SLinus Torvalds /* Set the tail pointer and length */ 7811da177e4SLinus Torvalds skb_put(n, skb->len); 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 7841da177e4SLinus Torvalds BUG(); 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds copy_skb_header(n, skb); 7871da177e4SLinus Torvalds return n; 7881da177e4SLinus Torvalds } 789b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 7901da177e4SLinus Torvalds 7911da177e4SLinus Torvalds /** 7921da177e4SLinus Torvalds * pskb_copy - create copy of an sk_buff with private head. 7931da177e4SLinus Torvalds * @skb: buffer to copy 7941da177e4SLinus Torvalds * @gfp_mask: allocation priority 7951da177e4SLinus Torvalds * 7961da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 7971da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 7981da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 7991da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 8001da177e4SLinus Torvalds * or the pointer to the buffer on success. 8011da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 8021da177e4SLinus Torvalds */ 8031da177e4SLinus Torvalds 804dd0fc66fSAl Viro struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask) 8051da177e4SLinus Torvalds { 8066602cebbSEric Dumazet unsigned int size = skb_end_pointer(skb) - skb->head; 8076602cebbSEric Dumazet struct sk_buff *n = alloc_skb(size, gfp_mask); 8086602cebbSEric Dumazet 8091da177e4SLinus Torvalds if (!n) 8101da177e4SLinus Torvalds goto out; 8111da177e4SLinus Torvalds 8121da177e4SLinus Torvalds /* Set the data pointer */ 8136602cebbSEric Dumazet skb_reserve(n, skb_headroom(skb)); 8141da177e4SLinus Torvalds /* Set the tail pointer and length */ 8151da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 8161da177e4SLinus Torvalds /* Copy the bytes */ 817d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 8181da177e4SLinus Torvalds 81925f484a6SHerbert Xu n->truesize += skb->data_len; 8201da177e4SLinus Torvalds n->data_len = skb->data_len; 8211da177e4SLinus Torvalds n->len = skb->len; 8221da177e4SLinus Torvalds 8231da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 8241da177e4SLinus Torvalds int i; 8251da177e4SLinus Torvalds 826a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 827a6686f2fSShirley Ma if (skb_copy_ubufs(skb, gfp_mask)) { 8281511022cSDan Carpenter kfree_skb(n); 8291511022cSDan Carpenter n = NULL; 830a6686f2fSShirley Ma goto out; 831a6686f2fSShirley Ma } 832a6686f2fSShirley Ma } 8331da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 8341da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 835ea2ab693SIan Campbell skb_frag_ref(skb, i); 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 8381da177e4SLinus Torvalds } 8391da177e4SLinus Torvalds 84021dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 8411da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 8421da177e4SLinus Torvalds skb_clone_fraglist(n); 8431da177e4SLinus Torvalds } 8441da177e4SLinus Torvalds 8451da177e4SLinus Torvalds copy_skb_header(n, skb); 8461da177e4SLinus Torvalds out: 8471da177e4SLinus Torvalds return n; 8481da177e4SLinus Torvalds } 849b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_copy); 8501da177e4SLinus Torvalds 8511da177e4SLinus Torvalds /** 8521da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 8531da177e4SLinus Torvalds * @skb: buffer to reallocate 8541da177e4SLinus Torvalds * @nhead: room to add at head 8551da177e4SLinus Torvalds * @ntail: room to add at tail 8561da177e4SLinus Torvalds * @gfp_mask: allocation priority 8571da177e4SLinus Torvalds * 8581da177e4SLinus Torvalds * Expands (or creates identical copy, if &nhead and &ntail are zero) 8591da177e4SLinus Torvalds * header of skb. &sk_buff itself is not changed. &sk_buff MUST have 8601da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 8611da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 8621da177e4SLinus Torvalds * 8631da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 8641da177e4SLinus Torvalds * reloaded after call to this function. 8651da177e4SLinus Torvalds */ 8661da177e4SLinus Torvalds 86786a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 868dd0fc66fSAl Viro gfp_t gfp_mask) 8691da177e4SLinus Torvalds { 8701da177e4SLinus Torvalds int i; 8711da177e4SLinus Torvalds u8 *data; 8726602cebbSEric Dumazet int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail; 8731da177e4SLinus Torvalds long off; 8741fd63041SEric Dumazet bool fastpath; 8751da177e4SLinus Torvalds 8764edd87adSHerbert Xu BUG_ON(nhead < 0); 8774edd87adSHerbert Xu 8781da177e4SLinus Torvalds if (skb_shared(skb)) 8791da177e4SLinus Torvalds BUG(); 8801da177e4SLinus Torvalds 8811da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 8821da177e4SLinus Torvalds 883ca44ac38SChangli Gao /* Check if we can avoid taking references on fragments if we own 884ca44ac38SChangli Gao * the last reference on skb->head. (see skb_release_data()) 885ca44ac38SChangli Gao */ 886ca44ac38SChangli Gao if (!skb->cloned) 887ca44ac38SChangli Gao fastpath = true; 888ca44ac38SChangli Gao else { 889ca44ac38SChangli Gao int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 890ca44ac38SChangli Gao fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta; 891ca44ac38SChangli Gao } 892ca44ac38SChangli Gao 893ca44ac38SChangli Gao if (fastpath && 894ca44ac38SChangli Gao size + sizeof(struct skb_shared_info) <= ksize(skb->head)) { 895ca44ac38SChangli Gao memmove(skb->head + size, skb_shinfo(skb), 896ca44ac38SChangli Gao offsetof(struct skb_shared_info, 897ca44ac38SChangli Gao frags[skb_shinfo(skb)->nr_frags])); 898ca44ac38SChangli Gao memmove(skb->head + nhead, skb->head, 899ca44ac38SChangli Gao skb_tail_pointer(skb) - skb->head); 900ca44ac38SChangli Gao off = nhead; 901ca44ac38SChangli Gao goto adjust_others; 902ca44ac38SChangli Gao } 903ca44ac38SChangli Gao 9041da177e4SLinus Torvalds data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); 9051da177e4SLinus Torvalds if (!data) 9061da177e4SLinus Torvalds goto nodata; 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 9096602cebbSEric Dumazet * optimized for the cases when header is void. 9106602cebbSEric Dumazet */ 9116602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 9126602cebbSEric Dumazet 9136602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 9146602cebbSEric Dumazet skb_shinfo(skb), 915fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 9161da177e4SLinus Torvalds 9171fd63041SEric Dumazet if (fastpath) { 9181fd63041SEric Dumazet kfree(skb->head); 9191fd63041SEric Dumazet } else { 920a6686f2fSShirley Ma /* copy this zero copy skb frags */ 921a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 922a6686f2fSShirley Ma if (skb_copy_ubufs(skb, gfp_mask)) 923a6686f2fSShirley Ma goto nofrags; 924a6686f2fSShirley Ma } 9251da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 926ea2ab693SIan Campbell skb_frag_ref(skb, i); 9271da177e4SLinus Torvalds 92821dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 9291da177e4SLinus Torvalds skb_clone_fraglist(skb); 9301da177e4SLinus Torvalds 9311da177e4SLinus Torvalds skb_release_data(skb); 9321fd63041SEric Dumazet } 9331da177e4SLinus Torvalds off = (data + nhead) - skb->head; 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds skb->head = data; 936ca44ac38SChangli Gao adjust_others: 9371da177e4SLinus Torvalds skb->data += off; 9384305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 9394305b541SArnaldo Carvalho de Melo skb->end = size; 94056eb8882SPatrick McHardy off = nhead; 9414305b541SArnaldo Carvalho de Melo #else 9424305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 94356eb8882SPatrick McHardy #endif 94427a884dcSArnaldo Carvalho de Melo /* {transport,network,mac}_header and tail are relative to skb->head */ 94527a884dcSArnaldo Carvalho de Melo skb->tail += off; 946b0e380b1SArnaldo Carvalho de Melo skb->transport_header += off; 947b0e380b1SArnaldo Carvalho de Melo skb->network_header += off; 948603a8bbeSStephen Hemminger if (skb_mac_header_was_set(skb)) 949b0e380b1SArnaldo Carvalho de Melo skb->mac_header += off; 95000c5a983SAndrea Shepard /* Only adjust this if it actually is csum_start rather than csum */ 95100c5a983SAndrea Shepard if (skb->ip_summed == CHECKSUM_PARTIAL) 952172a863fSHerbert Xu skb->csum_start += nhead; 9531da177e4SLinus Torvalds skb->cloned = 0; 954334a8132SPatrick McHardy skb->hdr_len = 0; 9551da177e4SLinus Torvalds skb->nohdr = 0; 9561da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 9571da177e4SLinus Torvalds return 0; 9581da177e4SLinus Torvalds 959a6686f2fSShirley Ma nofrags: 960a6686f2fSShirley Ma kfree(data); 9611da177e4SLinus Torvalds nodata: 9621da177e4SLinus Torvalds return -ENOMEM; 9631da177e4SLinus Torvalds } 964b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 9671da177e4SLinus Torvalds 9681da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 9691da177e4SLinus Torvalds { 9701da177e4SLinus Torvalds struct sk_buff *skb2; 9711da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 9721da177e4SLinus Torvalds 9731da177e4SLinus Torvalds if (delta <= 0) 9741da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 9751da177e4SLinus Torvalds else { 9761da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 9771da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 9781da177e4SLinus Torvalds GFP_ATOMIC)) { 9791da177e4SLinus Torvalds kfree_skb(skb2); 9801da177e4SLinus Torvalds skb2 = NULL; 9811da177e4SLinus Torvalds } 9821da177e4SLinus Torvalds } 9831da177e4SLinus Torvalds return skb2; 9841da177e4SLinus Torvalds } 985b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds /** 9881da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 9891da177e4SLinus Torvalds * @skb: buffer to copy 9901da177e4SLinus Torvalds * @newheadroom: new free bytes at head 9911da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 9921da177e4SLinus Torvalds * @gfp_mask: allocation priority 9931da177e4SLinus Torvalds * 9941da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 9951da177e4SLinus Torvalds * allocate additional space. 9961da177e4SLinus Torvalds * 9971da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 9981da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 9991da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 10001da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 10011da177e4SLinus Torvalds * 10021da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 10031da177e4SLinus Torvalds * is called from an interrupt. 10041da177e4SLinus Torvalds */ 10051da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 100686a76cafSVictor Fusco int newheadroom, int newtailroom, 1007dd0fc66fSAl Viro gfp_t gfp_mask) 10081da177e4SLinus Torvalds { 10091da177e4SLinus Torvalds /* 10101da177e4SLinus Torvalds * Allocate the copy buffer 10111da177e4SLinus Torvalds */ 10121da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom, 10131da177e4SLinus Torvalds gfp_mask); 1014efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 10151da177e4SLinus Torvalds int head_copy_len, head_copy_off; 101652886051SHerbert Xu int off; 10171da177e4SLinus Torvalds 10181da177e4SLinus Torvalds if (!n) 10191da177e4SLinus Torvalds return NULL; 10201da177e4SLinus Torvalds 10211da177e4SLinus Torvalds skb_reserve(n, newheadroom); 10221da177e4SLinus Torvalds 10231da177e4SLinus Torvalds /* Set the tail pointer and length */ 10241da177e4SLinus Torvalds skb_put(n, skb->len); 10251da177e4SLinus Torvalds 1026efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 10271da177e4SLinus Torvalds head_copy_off = 0; 10281da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 10291da177e4SLinus Torvalds head_copy_len = newheadroom; 10301da177e4SLinus Torvalds else 10311da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 10321da177e4SLinus Torvalds 10331da177e4SLinus Torvalds /* Copy the linear header and data. */ 10341da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 10351da177e4SLinus Torvalds skb->len + head_copy_len)) 10361da177e4SLinus Torvalds BUG(); 10371da177e4SLinus Torvalds 10381da177e4SLinus Torvalds copy_skb_header(n, skb); 10391da177e4SLinus Torvalds 1040efd1e8d5SPatrick McHardy off = newheadroom - oldheadroom; 1041be2b6e62SDavid S. Miller if (n->ip_summed == CHECKSUM_PARTIAL) 104252886051SHerbert Xu n->csum_start += off; 104352886051SHerbert Xu #ifdef NET_SKBUFF_DATA_USES_OFFSET 1044efd1e8d5SPatrick McHardy n->transport_header += off; 1045efd1e8d5SPatrick McHardy n->network_header += off; 1046603a8bbeSStephen Hemminger if (skb_mac_header_was_set(skb)) 1047efd1e8d5SPatrick McHardy n->mac_header += off; 104852886051SHerbert Xu #endif 1049efd1e8d5SPatrick McHardy 10501da177e4SLinus Torvalds return n; 10511da177e4SLinus Torvalds } 1052b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds /** 10551da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 10561da177e4SLinus Torvalds * @skb: buffer to pad 10571da177e4SLinus Torvalds * @pad: space to pad 10581da177e4SLinus Torvalds * 10591da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 10601da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 10611da177e4SLinus Torvalds * beyond the buffer end onto the wire. 10621da177e4SLinus Torvalds * 10635b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 10641da177e4SLinus Torvalds */ 10651da177e4SLinus Torvalds 10665b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 10671da177e4SLinus Torvalds { 10685b057c6bSHerbert Xu int err; 10695b057c6bSHerbert Xu int ntail; 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 10725b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 10731da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 10745b057c6bSHerbert Xu return 0; 10751da177e4SLinus Torvalds } 10761da177e4SLinus Torvalds 10774305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 10785b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 10795b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 10805b057c6bSHerbert Xu if (unlikely(err)) 10815b057c6bSHerbert Xu goto free_skb; 10825b057c6bSHerbert Xu } 10835b057c6bSHerbert Xu 10845b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 10855b057c6bSHerbert Xu * to be audited. 10865b057c6bSHerbert Xu */ 10875b057c6bSHerbert Xu err = skb_linearize(skb); 10885b057c6bSHerbert Xu if (unlikely(err)) 10895b057c6bSHerbert Xu goto free_skb; 10905b057c6bSHerbert Xu 10915b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 10925b057c6bSHerbert Xu return 0; 10935b057c6bSHerbert Xu 10945b057c6bSHerbert Xu free_skb: 10951da177e4SLinus Torvalds kfree_skb(skb); 10965b057c6bSHerbert Xu return err; 10971da177e4SLinus Torvalds } 1098b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 10991da177e4SLinus Torvalds 11000dde3e16SIlpo Järvinen /** 11010dde3e16SIlpo Järvinen * skb_put - add data to a buffer 11020dde3e16SIlpo Järvinen * @skb: buffer to use 11030dde3e16SIlpo Järvinen * @len: amount of data to add 11040dde3e16SIlpo Järvinen * 11050dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 11060dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 11070dde3e16SIlpo Järvinen * first byte of the extra data is returned. 11080dde3e16SIlpo Järvinen */ 11090dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 11100dde3e16SIlpo Järvinen { 11110dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 11120dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 11130dde3e16SIlpo Järvinen skb->tail += len; 11140dde3e16SIlpo Järvinen skb->len += len; 11150dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 11160dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 11170dde3e16SIlpo Järvinen return tmp; 11180dde3e16SIlpo Järvinen } 11190dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 11200dde3e16SIlpo Järvinen 11216be8ac2fSIlpo Järvinen /** 1122c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1123c2aa270aSIlpo Järvinen * @skb: buffer to use 1124c2aa270aSIlpo Järvinen * @len: amount of data to add 1125c2aa270aSIlpo Järvinen * 1126c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1127c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1128c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1129c2aa270aSIlpo Järvinen */ 1130c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1131c2aa270aSIlpo Järvinen { 1132c2aa270aSIlpo Järvinen skb->data -= len; 1133c2aa270aSIlpo Järvinen skb->len += len; 1134c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1135c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1136c2aa270aSIlpo Järvinen return skb->data; 1137c2aa270aSIlpo Järvinen } 1138c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1139c2aa270aSIlpo Järvinen 1140c2aa270aSIlpo Järvinen /** 11416be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 11426be8ac2fSIlpo Järvinen * @skb: buffer to use 11436be8ac2fSIlpo Järvinen * @len: amount of data to remove 11446be8ac2fSIlpo Järvinen * 11456be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 11466be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 11476be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 11486be8ac2fSIlpo Järvinen * the old data. 11496be8ac2fSIlpo Järvinen */ 11506be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 11516be8ac2fSIlpo Järvinen { 115247d29646SDavid S. Miller return skb_pull_inline(skb, len); 11536be8ac2fSIlpo Järvinen } 11546be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 11556be8ac2fSIlpo Järvinen 1156419ae74eSIlpo Järvinen /** 1157419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1158419ae74eSIlpo Järvinen * @skb: buffer to alter 1159419ae74eSIlpo Järvinen * @len: new length 1160419ae74eSIlpo Järvinen * 1161419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1162419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1163419ae74eSIlpo Järvinen * The skb must be linear. 1164419ae74eSIlpo Järvinen */ 1165419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1166419ae74eSIlpo Järvinen { 1167419ae74eSIlpo Järvinen if (skb->len > len) 1168419ae74eSIlpo Järvinen __skb_trim(skb, len); 1169419ae74eSIlpo Järvinen } 1170419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1171419ae74eSIlpo Järvinen 11723cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 11731da177e4SLinus Torvalds */ 11741da177e4SLinus Torvalds 11753cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 11761da177e4SLinus Torvalds { 117727b437c8SHerbert Xu struct sk_buff **fragp; 117827b437c8SHerbert Xu struct sk_buff *frag; 11791da177e4SLinus Torvalds int offset = skb_headlen(skb); 11801da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 11811da177e4SLinus Torvalds int i; 118227b437c8SHerbert Xu int err; 118327b437c8SHerbert Xu 118427b437c8SHerbert Xu if (skb_cloned(skb) && 118527b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 118627b437c8SHerbert Xu return err; 11871da177e4SLinus Torvalds 1188f4d26fb3SHerbert Xu i = 0; 1189f4d26fb3SHerbert Xu if (offset >= len) 1190f4d26fb3SHerbert Xu goto drop_pages; 1191f4d26fb3SHerbert Xu 1192f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 11931da177e4SLinus Torvalds int end = offset + skb_shinfo(skb)->frags[i].size; 119427b437c8SHerbert Xu 119527b437c8SHerbert Xu if (end < len) { 11961da177e4SLinus Torvalds offset = end; 119727b437c8SHerbert Xu continue; 11981da177e4SLinus Torvalds } 11991da177e4SLinus Torvalds 120027b437c8SHerbert Xu skb_shinfo(skb)->frags[i++].size = len - offset; 120127b437c8SHerbert Xu 1202f4d26fb3SHerbert Xu drop_pages: 120327b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 120427b437c8SHerbert Xu 120527b437c8SHerbert Xu for (; i < nfrags; i++) 1206ea2ab693SIan Campbell skb_frag_unref(skb, i); 120727b437c8SHerbert Xu 120821dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 120927b437c8SHerbert Xu skb_drop_fraglist(skb); 1210f4d26fb3SHerbert Xu goto done; 121127b437c8SHerbert Xu } 121227b437c8SHerbert Xu 121327b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 121427b437c8SHerbert Xu fragp = &frag->next) { 121527b437c8SHerbert Xu int end = offset + frag->len; 121627b437c8SHerbert Xu 121727b437c8SHerbert Xu if (skb_shared(frag)) { 121827b437c8SHerbert Xu struct sk_buff *nfrag; 121927b437c8SHerbert Xu 122027b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 122127b437c8SHerbert Xu if (unlikely(!nfrag)) 122227b437c8SHerbert Xu return -ENOMEM; 122327b437c8SHerbert Xu 122427b437c8SHerbert Xu nfrag->next = frag->next; 1225f4d26fb3SHerbert Xu kfree_skb(frag); 122627b437c8SHerbert Xu frag = nfrag; 122727b437c8SHerbert Xu *fragp = frag; 122827b437c8SHerbert Xu } 122927b437c8SHerbert Xu 123027b437c8SHerbert Xu if (end < len) { 123127b437c8SHerbert Xu offset = end; 123227b437c8SHerbert Xu continue; 123327b437c8SHerbert Xu } 123427b437c8SHerbert Xu 123527b437c8SHerbert Xu if (end > len && 123627b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 123727b437c8SHerbert Xu return err; 123827b437c8SHerbert Xu 123927b437c8SHerbert Xu if (frag->next) 124027b437c8SHerbert Xu skb_drop_list(&frag->next); 124127b437c8SHerbert Xu break; 124227b437c8SHerbert Xu } 124327b437c8SHerbert Xu 1244f4d26fb3SHerbert Xu done: 124527b437c8SHerbert Xu if (len > skb_headlen(skb)) { 12461da177e4SLinus Torvalds skb->data_len -= skb->len - len; 12471da177e4SLinus Torvalds skb->len = len; 12481da177e4SLinus Torvalds } else { 12491da177e4SLinus Torvalds skb->len = len; 12501da177e4SLinus Torvalds skb->data_len = 0; 125127a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 12521da177e4SLinus Torvalds } 12531da177e4SLinus Torvalds 12541da177e4SLinus Torvalds return 0; 12551da177e4SLinus Torvalds } 1256b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds /** 12591da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 12601da177e4SLinus Torvalds * @skb: buffer to reallocate 12611da177e4SLinus Torvalds * @delta: number of bytes to advance tail 12621da177e4SLinus Torvalds * 12631da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 12641da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 12651da177e4SLinus Torvalds * data from fragmented part. 12661da177e4SLinus Torvalds * 12671da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 12681da177e4SLinus Torvalds * 12691da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 12701da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 12711da177e4SLinus Torvalds * 12721da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 12731da177e4SLinus Torvalds * reloaded after call to this function. 12741da177e4SLinus Torvalds */ 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 12771da177e4SLinus Torvalds * when it is necessary. 12781da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 12791da177e4SLinus Torvalds * 2. It may change skb pointers. 12801da177e4SLinus Torvalds * 12811da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 12821da177e4SLinus Torvalds */ 12831da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 12841da177e4SLinus Torvalds { 12851da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 12861da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 12871da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 12881da177e4SLinus Torvalds */ 12894305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 12921da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 12931da177e4SLinus Torvalds GFP_ATOMIC)) 12941da177e4SLinus Torvalds return NULL; 12951da177e4SLinus Torvalds } 12961da177e4SLinus Torvalds 129727a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 12981da177e4SLinus Torvalds BUG(); 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 13011da177e4SLinus Torvalds * size of pulled pages. Superb. 13021da177e4SLinus Torvalds */ 130321dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 13041da177e4SLinus Torvalds goto pull_pages; 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 13071da177e4SLinus Torvalds eat = delta; 13081da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 13091da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size >= eat) 13101da177e4SLinus Torvalds goto pull_pages; 13111da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 13151da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 13161da177e4SLinus Torvalds * but taking into account that pulling is expected to 13171da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 13181da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 13191da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 13201da177e4SLinus Torvalds */ 13211da177e4SLinus Torvalds if (eat) { 13221da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 13231da177e4SLinus Torvalds struct sk_buff *clone = NULL; 13241da177e4SLinus Torvalds struct sk_buff *insp = NULL; 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds do { 132709a62660SKris Katterjohn BUG_ON(!list); 13281da177e4SLinus Torvalds 13291da177e4SLinus Torvalds if (list->len <= eat) { 13301da177e4SLinus Torvalds /* Eaten as whole. */ 13311da177e4SLinus Torvalds eat -= list->len; 13321da177e4SLinus Torvalds list = list->next; 13331da177e4SLinus Torvalds insp = list; 13341da177e4SLinus Torvalds } else { 13351da177e4SLinus Torvalds /* Eaten partially. */ 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds if (skb_shared(list)) { 13381da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 13391da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 13401da177e4SLinus Torvalds if (!clone) 13411da177e4SLinus Torvalds return NULL; 13421da177e4SLinus Torvalds insp = list->next; 13431da177e4SLinus Torvalds list = clone; 13441da177e4SLinus Torvalds } else { 13451da177e4SLinus Torvalds /* This may be pulled without 13461da177e4SLinus Torvalds * problems. */ 13471da177e4SLinus Torvalds insp = list; 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 13501da177e4SLinus Torvalds kfree_skb(clone); 13511da177e4SLinus Torvalds return NULL; 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds break; 13541da177e4SLinus Torvalds } 13551da177e4SLinus Torvalds } while (eat); 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds /* Free pulled out fragments. */ 13581da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 13591da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 13601da177e4SLinus Torvalds kfree_skb(list); 13611da177e4SLinus Torvalds } 13621da177e4SLinus Torvalds /* And insert new clone at head. */ 13631da177e4SLinus Torvalds if (clone) { 13641da177e4SLinus Torvalds clone->next = list; 13651da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 13661da177e4SLinus Torvalds } 13671da177e4SLinus Torvalds } 13681da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds pull_pages: 13711da177e4SLinus Torvalds eat = delta; 13721da177e4SLinus Torvalds k = 0; 13731da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 13741da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size <= eat) { 1375ea2ab693SIan Campbell skb_frag_unref(skb, i); 13761da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 13771da177e4SLinus Torvalds } else { 13781da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 13791da177e4SLinus Torvalds if (eat) { 13801da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 13811da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].size -= eat; 13821da177e4SLinus Torvalds eat = 0; 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds k++; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds } 13871da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds skb->tail += delta; 13901da177e4SLinus Torvalds skb->data_len -= delta; 13911da177e4SLinus Torvalds 139227a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 13931da177e4SLinus Torvalds } 1394b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 13951da177e4SLinus Torvalds 139622019b17SEric Dumazet /** 139722019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 139822019b17SEric Dumazet * @skb: source skb 139922019b17SEric Dumazet * @offset: offset in source 140022019b17SEric Dumazet * @to: destination buffer 140122019b17SEric Dumazet * @len: number of bytes to copy 140222019b17SEric Dumazet * 140322019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 140422019b17SEric Dumazet * destination buffer. 140522019b17SEric Dumazet * 140622019b17SEric Dumazet * CAUTION ! : 140722019b17SEric Dumazet * If its prototype is ever changed, 140822019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 140922019b17SEric Dumazet * since it is called from BPF assembly code. 141022019b17SEric Dumazet */ 14111da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 14121da177e4SLinus Torvalds { 14131a028e50SDavid S. Miller int start = skb_headlen(skb); 1414fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1415fbb398a8SDavid S. Miller int i, copy; 14161da177e4SLinus Torvalds 14171da177e4SLinus Torvalds if (offset > (int)skb->len - len) 14181da177e4SLinus Torvalds goto fault; 14191da177e4SLinus Torvalds 14201da177e4SLinus Torvalds /* Copy header. */ 14211a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 14221da177e4SLinus Torvalds if (copy > len) 14231da177e4SLinus Torvalds copy = len; 1424d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 14251da177e4SLinus Torvalds if ((len -= copy) == 0) 14261da177e4SLinus Torvalds return 0; 14271da177e4SLinus Torvalds offset += copy; 14281da177e4SLinus Torvalds to += copy; 14291da177e4SLinus Torvalds } 14301da177e4SLinus Torvalds 14311da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 14321a028e50SDavid S. Miller int end; 14331da177e4SLinus Torvalds 1434547b792cSIlpo Järvinen WARN_ON(start > offset + len); 14351a028e50SDavid S. Miller 14361a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 14371da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 14381da177e4SLinus Torvalds u8 *vaddr; 14391da177e4SLinus Torvalds 14401da177e4SLinus Torvalds if (copy > len) 14411da177e4SLinus Torvalds copy = len; 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]); 14441da177e4SLinus Torvalds memcpy(to, 14451a028e50SDavid S. Miller vaddr + skb_shinfo(skb)->frags[i].page_offset+ 14461a028e50SDavid S. Miller offset - start, copy); 14471da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 14481da177e4SLinus Torvalds 14491da177e4SLinus Torvalds if ((len -= copy) == 0) 14501da177e4SLinus Torvalds return 0; 14511da177e4SLinus Torvalds offset += copy; 14521da177e4SLinus Torvalds to += copy; 14531da177e4SLinus Torvalds } 14541a028e50SDavid S. Miller start = end; 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds 1457fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 14581a028e50SDavid S. Miller int end; 14591da177e4SLinus Torvalds 1460547b792cSIlpo Järvinen WARN_ON(start > offset + len); 14611a028e50SDavid S. Miller 1462fbb398a8SDavid S. Miller end = start + frag_iter->len; 14631da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 14641da177e4SLinus Torvalds if (copy > len) 14651da177e4SLinus Torvalds copy = len; 1466fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 14671da177e4SLinus Torvalds goto fault; 14681da177e4SLinus Torvalds if ((len -= copy) == 0) 14691da177e4SLinus Torvalds return 0; 14701da177e4SLinus Torvalds offset += copy; 14711da177e4SLinus Torvalds to += copy; 14721da177e4SLinus Torvalds } 14731a028e50SDavid S. Miller start = end; 14741da177e4SLinus Torvalds } 1475a6686f2fSShirley Ma 14761da177e4SLinus Torvalds if (!len) 14771da177e4SLinus Torvalds return 0; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds fault: 14801da177e4SLinus Torvalds return -EFAULT; 14811da177e4SLinus Torvalds } 1482b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 14831da177e4SLinus Torvalds 14849c55e01cSJens Axboe /* 14859c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 14869c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 14879c55e01cSJens Axboe */ 14889c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 14899c55e01cSJens Axboe { 14908b9d3728SJarek Poplawski put_page(spd->pages[i]); 14918b9d3728SJarek Poplawski } 14929c55e01cSJens Axboe 14934fb66994SJarek Poplawski static inline struct page *linear_to_page(struct page *page, unsigned int *len, 14944fb66994SJarek Poplawski unsigned int *offset, 14957a67e56fSJarek Poplawski struct sk_buff *skb, struct sock *sk) 14968b9d3728SJarek Poplawski { 14974fb66994SJarek Poplawski struct page *p = sk->sk_sndmsg_page; 14984fb66994SJarek Poplawski unsigned int off; 14998b9d3728SJarek Poplawski 15004fb66994SJarek Poplawski if (!p) { 15014fb66994SJarek Poplawski new_page: 15024fb66994SJarek Poplawski p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0); 15038b9d3728SJarek Poplawski if (!p) 15048b9d3728SJarek Poplawski return NULL; 15054fb66994SJarek Poplawski 15064fb66994SJarek Poplawski off = sk->sk_sndmsg_off = 0; 15074fb66994SJarek Poplawski /* hold one ref to this page until it's full */ 15084fb66994SJarek Poplawski } else { 15094fb66994SJarek Poplawski unsigned int mlen; 15104fb66994SJarek Poplawski 15114fb66994SJarek Poplawski off = sk->sk_sndmsg_off; 15124fb66994SJarek Poplawski mlen = PAGE_SIZE - off; 15134fb66994SJarek Poplawski if (mlen < 64 && mlen < *len) { 15144fb66994SJarek Poplawski put_page(p); 15154fb66994SJarek Poplawski goto new_page; 15164fb66994SJarek Poplawski } 15174fb66994SJarek Poplawski 15184fb66994SJarek Poplawski *len = min_t(unsigned int, *len, mlen); 15194fb66994SJarek Poplawski } 15204fb66994SJarek Poplawski 15214fb66994SJarek Poplawski memcpy(page_address(p) + off, page_address(page) + *offset, *len); 15224fb66994SJarek Poplawski sk->sk_sndmsg_off += *len; 15234fb66994SJarek Poplawski *offset = off; 15244fb66994SJarek Poplawski get_page(p); 15258b9d3728SJarek Poplawski 15268b9d3728SJarek Poplawski return p; 15279c55e01cSJens Axboe } 15289c55e01cSJens Axboe 15299c55e01cSJens Axboe /* 15309c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 15319c55e01cSJens Axboe */ 153235f3d14dSJens Axboe static inline int spd_fill_page(struct splice_pipe_desc *spd, 153335f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 15344fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 15357a67e56fSJarek Poplawski struct sk_buff *skb, int linear, 15367a67e56fSJarek Poplawski struct sock *sk) 15379c55e01cSJens Axboe { 153835f3d14dSJens Axboe if (unlikely(spd->nr_pages == pipe->buffers)) 15399c55e01cSJens Axboe return 1; 15409c55e01cSJens Axboe 15418b9d3728SJarek Poplawski if (linear) { 15427a67e56fSJarek Poplawski page = linear_to_page(page, len, &offset, skb, sk); 15438b9d3728SJarek Poplawski if (!page) 15448b9d3728SJarek Poplawski return 1; 15458b9d3728SJarek Poplawski } else 15468b9d3728SJarek Poplawski get_page(page); 15478b9d3728SJarek Poplawski 15489c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 15494fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 15509c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 15519c55e01cSJens Axboe spd->nr_pages++; 15528b9d3728SJarek Poplawski 15539c55e01cSJens Axboe return 0; 15549c55e01cSJens Axboe } 15559c55e01cSJens Axboe 15562870c43dSOctavian Purdila static inline void __segment_seek(struct page **page, unsigned int *poff, 15572870c43dSOctavian Purdila unsigned int *plen, unsigned int off) 15582870c43dSOctavian Purdila { 1559ce3dd395SJarek Poplawski unsigned long n; 1560ce3dd395SJarek Poplawski 15612870c43dSOctavian Purdila *poff += off; 1562ce3dd395SJarek Poplawski n = *poff / PAGE_SIZE; 1563ce3dd395SJarek Poplawski if (n) 1564ce3dd395SJarek Poplawski *page = nth_page(*page, n); 1565ce3dd395SJarek Poplawski 15662870c43dSOctavian Purdila *poff = *poff % PAGE_SIZE; 15672870c43dSOctavian Purdila *plen -= off; 15682870c43dSOctavian Purdila } 15692870c43dSOctavian Purdila 15702870c43dSOctavian Purdila static inline int __splice_segment(struct page *page, unsigned int poff, 15712870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 15722870c43dSOctavian Purdila unsigned int *len, struct sk_buff *skb, 15737a67e56fSJarek Poplawski struct splice_pipe_desc *spd, int linear, 157435f3d14dSJens Axboe struct sock *sk, 157535f3d14dSJens Axboe struct pipe_inode_info *pipe) 15769c55e01cSJens Axboe { 15772870c43dSOctavian Purdila if (!*len) 15782870c43dSOctavian Purdila return 1; 15799c55e01cSJens Axboe 15802870c43dSOctavian Purdila /* skip this segment if already processed */ 15812870c43dSOctavian Purdila if (*off >= plen) { 15822870c43dSOctavian Purdila *off -= plen; 15832870c43dSOctavian Purdila return 0; 15842870c43dSOctavian Purdila } 15852870c43dSOctavian Purdila 15862870c43dSOctavian Purdila /* ignore any bits we already processed */ 15872870c43dSOctavian Purdila if (*off) { 15882870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, *off); 15892870c43dSOctavian Purdila *off = 0; 15902870c43dSOctavian Purdila } 15912870c43dSOctavian Purdila 15922870c43dSOctavian Purdila do { 15932870c43dSOctavian Purdila unsigned int flen = min(*len, plen); 15942870c43dSOctavian Purdila 15952870c43dSOctavian Purdila /* the linear region may spread across several pages */ 15962870c43dSOctavian Purdila flen = min_t(unsigned int, flen, PAGE_SIZE - poff); 15972870c43dSOctavian Purdila 159835f3d14dSJens Axboe if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk)) 15992870c43dSOctavian Purdila return 1; 16002870c43dSOctavian Purdila 16012870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, flen); 16022870c43dSOctavian Purdila *len -= flen; 16032870c43dSOctavian Purdila 16042870c43dSOctavian Purdila } while (*len && plen); 16052870c43dSOctavian Purdila 16062870c43dSOctavian Purdila return 0; 1607db43a282SOctavian Purdila } 16089c55e01cSJens Axboe 16099c55e01cSJens Axboe /* 16102870c43dSOctavian Purdila * Map linear and fragment data from the skb to spd. It reports failure if the 16112870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 16129c55e01cSJens Axboe */ 161335f3d14dSJens Axboe static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 161435f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 161535f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 16162870c43dSOctavian Purdila { 16172870c43dSOctavian Purdila int seg; 16189c55e01cSJens Axboe 16199c55e01cSJens Axboe /* 16202870c43dSOctavian Purdila * map the linear part 16219c55e01cSJens Axboe */ 16222870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 16232870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 16242870c43dSOctavian Purdila skb_headlen(skb), 162535f3d14dSJens Axboe offset, len, skb, spd, 1, sk, pipe)) 16262870c43dSOctavian Purdila return 1; 16279c55e01cSJens Axboe 16289c55e01cSJens Axboe /* 16299c55e01cSJens Axboe * then map the fragments 16309c55e01cSJens Axboe */ 16319c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 16329c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 16339c55e01cSJens Axboe 1634ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 1635ea2ab693SIan Campbell f->page_offset, f->size, 163635f3d14dSJens Axboe offset, len, skb, spd, 0, sk, pipe)) 16372870c43dSOctavian Purdila return 1; 16389c55e01cSJens Axboe } 16399c55e01cSJens Axboe 16409c55e01cSJens Axboe return 0; 16419c55e01cSJens Axboe } 16429c55e01cSJens Axboe 16439c55e01cSJens Axboe /* 16449c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 16459c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 16469c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 16479c55e01cSJens Axboe * handle that cleanly. 16489c55e01cSJens Axboe */ 16498b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 16509c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 16519c55e01cSJens Axboe unsigned int flags) 16529c55e01cSJens Axboe { 165335f3d14dSJens Axboe struct partial_page partial[PIPE_DEF_BUFFERS]; 165435f3d14dSJens Axboe struct page *pages[PIPE_DEF_BUFFERS]; 16559c55e01cSJens Axboe struct splice_pipe_desc spd = { 16569c55e01cSJens Axboe .pages = pages, 16579c55e01cSJens Axboe .partial = partial, 16589c55e01cSJens Axboe .flags = flags, 16599c55e01cSJens Axboe .ops = &sock_pipe_buf_ops, 16609c55e01cSJens Axboe .spd_release = sock_spd_release, 16619c55e01cSJens Axboe }; 1662fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 16637a67e56fSJarek Poplawski struct sock *sk = skb->sk; 166435f3d14dSJens Axboe int ret = 0; 166535f3d14dSJens Axboe 166635f3d14dSJens Axboe if (splice_grow_spd(pipe, &spd)) 166735f3d14dSJens Axboe return -ENOMEM; 16689c55e01cSJens Axboe 16699c55e01cSJens Axboe /* 16709c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 16719c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 16729c55e01cSJens Axboe */ 167335f3d14dSJens Axboe if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 16749c55e01cSJens Axboe goto done; 16759c55e01cSJens Axboe else if (!tlen) 16769c55e01cSJens Axboe goto done; 16779c55e01cSJens Axboe 16789c55e01cSJens Axboe /* 16799c55e01cSJens Axboe * now see if we have a frag_list to map 16809c55e01cSJens Axboe */ 1681fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 1682fbb398a8SDavid S. Miller if (!tlen) 16839c55e01cSJens Axboe break; 168435f3d14dSJens Axboe if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1685fbb398a8SDavid S. Miller break; 16869c55e01cSJens Axboe } 16879c55e01cSJens Axboe 16889c55e01cSJens Axboe done: 16899c55e01cSJens Axboe if (spd.nr_pages) { 16909c55e01cSJens Axboe /* 16919c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 16929c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 16939c55e01cSJens Axboe * here as compared to sendfile(). We enter here 16949c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 16959c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 16969c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 16979c55e01cSJens Axboe * and networking will grab the socket lock. 16989c55e01cSJens Axboe */ 1699293ad604SOctavian Purdila release_sock(sk); 17009c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1701293ad604SOctavian Purdila lock_sock(sk); 17029c55e01cSJens Axboe } 17039c55e01cSJens Axboe 170435f3d14dSJens Axboe splice_shrink_spd(pipe, &spd); 170535f3d14dSJens Axboe return ret; 17069c55e01cSJens Axboe } 17079c55e01cSJens Axboe 1708357b40a1SHerbert Xu /** 1709357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1710357b40a1SHerbert Xu * @skb: destination buffer 1711357b40a1SHerbert Xu * @offset: offset in destination 1712357b40a1SHerbert Xu * @from: source buffer 1713357b40a1SHerbert Xu * @len: number of bytes to copy 1714357b40a1SHerbert Xu * 1715357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1716357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1717357b40a1SHerbert Xu * traversing fragment lists and such. 1718357b40a1SHerbert Xu */ 1719357b40a1SHerbert Xu 17200c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1721357b40a1SHerbert Xu { 17221a028e50SDavid S. Miller int start = skb_headlen(skb); 1723fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1724fbb398a8SDavid S. Miller int i, copy; 1725357b40a1SHerbert Xu 1726357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1727357b40a1SHerbert Xu goto fault; 1728357b40a1SHerbert Xu 17291a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1730357b40a1SHerbert Xu if (copy > len) 1731357b40a1SHerbert Xu copy = len; 173227d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1733357b40a1SHerbert Xu if ((len -= copy) == 0) 1734357b40a1SHerbert Xu return 0; 1735357b40a1SHerbert Xu offset += copy; 1736357b40a1SHerbert Xu from += copy; 1737357b40a1SHerbert Xu } 1738357b40a1SHerbert Xu 1739357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1740357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 17411a028e50SDavid S. Miller int end; 1742357b40a1SHerbert Xu 1743547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17441a028e50SDavid S. Miller 17451a028e50SDavid S. Miller end = start + frag->size; 1746357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1747357b40a1SHerbert Xu u8 *vaddr; 1748357b40a1SHerbert Xu 1749357b40a1SHerbert Xu if (copy > len) 1750357b40a1SHerbert Xu copy = len; 1751357b40a1SHerbert Xu 1752357b40a1SHerbert Xu vaddr = kmap_skb_frag(frag); 17531a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 17541a028e50SDavid S. Miller from, copy); 1755357b40a1SHerbert Xu kunmap_skb_frag(vaddr); 1756357b40a1SHerbert Xu 1757357b40a1SHerbert Xu if ((len -= copy) == 0) 1758357b40a1SHerbert Xu return 0; 1759357b40a1SHerbert Xu offset += copy; 1760357b40a1SHerbert Xu from += copy; 1761357b40a1SHerbert Xu } 17621a028e50SDavid S. Miller start = end; 1763357b40a1SHerbert Xu } 1764357b40a1SHerbert Xu 1765fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 17661a028e50SDavid S. Miller int end; 1767357b40a1SHerbert Xu 1768547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17691a028e50SDavid S. Miller 1770fbb398a8SDavid S. Miller end = start + frag_iter->len; 1771357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1772357b40a1SHerbert Xu if (copy > len) 1773357b40a1SHerbert Xu copy = len; 1774fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 17751a028e50SDavid S. Miller from, copy)) 1776357b40a1SHerbert Xu goto fault; 1777357b40a1SHerbert Xu if ((len -= copy) == 0) 1778357b40a1SHerbert Xu return 0; 1779357b40a1SHerbert Xu offset += copy; 1780357b40a1SHerbert Xu from += copy; 1781357b40a1SHerbert Xu } 17821a028e50SDavid S. Miller start = end; 1783357b40a1SHerbert Xu } 1784357b40a1SHerbert Xu if (!len) 1785357b40a1SHerbert Xu return 0; 1786357b40a1SHerbert Xu 1787357b40a1SHerbert Xu fault: 1788357b40a1SHerbert Xu return -EFAULT; 1789357b40a1SHerbert Xu } 1790357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1791357b40a1SHerbert Xu 17921da177e4SLinus Torvalds /* Checksum skb data. */ 17931da177e4SLinus Torvalds 17942bbbc868SAl Viro __wsum skb_checksum(const struct sk_buff *skb, int offset, 17952bbbc868SAl Viro int len, __wsum csum) 17961da177e4SLinus Torvalds { 17971a028e50SDavid S. Miller int start = skb_headlen(skb); 17981a028e50SDavid S. Miller int i, copy = start - offset; 1799fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 18001da177e4SLinus Torvalds int pos = 0; 18011da177e4SLinus Torvalds 18021da177e4SLinus Torvalds /* Checksum header. */ 18031da177e4SLinus Torvalds if (copy > 0) { 18041da177e4SLinus Torvalds if (copy > len) 18051da177e4SLinus Torvalds copy = len; 18061da177e4SLinus Torvalds csum = csum_partial(skb->data + offset, copy, csum); 18071da177e4SLinus Torvalds if ((len -= copy) == 0) 18081da177e4SLinus Torvalds return csum; 18091da177e4SLinus Torvalds offset += copy; 18101da177e4SLinus Torvalds pos = copy; 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds 18131da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 18141a028e50SDavid S. Miller int end; 18151da177e4SLinus Torvalds 1816547b792cSIlpo Järvinen WARN_ON(start > offset + len); 18171a028e50SDavid S. Miller 18181a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 18191da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 182044bb9363SAl Viro __wsum csum2; 18211da177e4SLinus Torvalds u8 *vaddr; 18221da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 18231da177e4SLinus Torvalds 18241da177e4SLinus Torvalds if (copy > len) 18251da177e4SLinus Torvalds copy = len; 18261da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 18271a028e50SDavid S. Miller csum2 = csum_partial(vaddr + frag->page_offset + 18281a028e50SDavid S. Miller offset - start, copy, 0); 18291da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 18301da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 18311da177e4SLinus Torvalds if (!(len -= copy)) 18321da177e4SLinus Torvalds return csum; 18331da177e4SLinus Torvalds offset += copy; 18341da177e4SLinus Torvalds pos += copy; 18351da177e4SLinus Torvalds } 18361a028e50SDavid S. Miller start = end; 18371da177e4SLinus Torvalds } 18381da177e4SLinus Torvalds 1839fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 18401a028e50SDavid S. Miller int end; 18411da177e4SLinus Torvalds 1842547b792cSIlpo Järvinen WARN_ON(start > offset + len); 18431a028e50SDavid S. Miller 1844fbb398a8SDavid S. Miller end = start + frag_iter->len; 18451da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 18465f92a738SAl Viro __wsum csum2; 18471da177e4SLinus Torvalds if (copy > len) 18481da177e4SLinus Torvalds copy = len; 1849fbb398a8SDavid S. Miller csum2 = skb_checksum(frag_iter, offset - start, 18501a028e50SDavid S. Miller copy, 0); 18511da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 18521da177e4SLinus Torvalds if ((len -= copy) == 0) 18531da177e4SLinus Torvalds return csum; 18541da177e4SLinus Torvalds offset += copy; 18551da177e4SLinus Torvalds pos += copy; 18561da177e4SLinus Torvalds } 18571a028e50SDavid S. Miller start = end; 18581da177e4SLinus Torvalds } 185909a62660SKris Katterjohn BUG_ON(len); 18601da177e4SLinus Torvalds 18611da177e4SLinus Torvalds return csum; 18621da177e4SLinus Torvalds } 1863b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 18641da177e4SLinus Torvalds 18651da177e4SLinus Torvalds /* Both of above in one bottle. */ 18661da177e4SLinus Torvalds 186781d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 186881d77662SAl Viro u8 *to, int len, __wsum csum) 18691da177e4SLinus Torvalds { 18701a028e50SDavid S. Miller int start = skb_headlen(skb); 18711a028e50SDavid S. Miller int i, copy = start - offset; 1872fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 18731da177e4SLinus Torvalds int pos = 0; 18741da177e4SLinus Torvalds 18751da177e4SLinus Torvalds /* Copy header. */ 18761da177e4SLinus Torvalds if (copy > 0) { 18771da177e4SLinus Torvalds if (copy > len) 18781da177e4SLinus Torvalds copy = len; 18791da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 18801da177e4SLinus Torvalds copy, csum); 18811da177e4SLinus Torvalds if ((len -= copy) == 0) 18821da177e4SLinus Torvalds return csum; 18831da177e4SLinus Torvalds offset += copy; 18841da177e4SLinus Torvalds to += copy; 18851da177e4SLinus Torvalds pos = copy; 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds 18881da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 18891a028e50SDavid S. Miller int end; 18901da177e4SLinus Torvalds 1891547b792cSIlpo Järvinen WARN_ON(start > offset + len); 18921a028e50SDavid S. Miller 18931a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 18941da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 18955084205fSAl Viro __wsum csum2; 18961da177e4SLinus Torvalds u8 *vaddr; 18971da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 18981da177e4SLinus Torvalds 18991da177e4SLinus Torvalds if (copy > len) 19001da177e4SLinus Torvalds copy = len; 19011da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 19021da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 19031a028e50SDavid S. Miller frag->page_offset + 19041a028e50SDavid S. Miller offset - start, to, 19051a028e50SDavid S. Miller copy, 0); 19061da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 19071da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 19081da177e4SLinus Torvalds if (!(len -= copy)) 19091da177e4SLinus Torvalds return csum; 19101da177e4SLinus Torvalds offset += copy; 19111da177e4SLinus Torvalds to += copy; 19121da177e4SLinus Torvalds pos += copy; 19131da177e4SLinus Torvalds } 19141a028e50SDavid S. Miller start = end; 19151da177e4SLinus Torvalds } 19161da177e4SLinus Torvalds 1917fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 191881d77662SAl Viro __wsum csum2; 19191a028e50SDavid S. Miller int end; 19201da177e4SLinus Torvalds 1921547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19221a028e50SDavid S. Miller 1923fbb398a8SDavid S. Miller end = start + frag_iter->len; 19241da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 19251da177e4SLinus Torvalds if (copy > len) 19261da177e4SLinus Torvalds copy = len; 1927fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 19281a028e50SDavid S. Miller offset - start, 19291da177e4SLinus Torvalds to, copy, 0); 19301da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 19311da177e4SLinus Torvalds if ((len -= copy) == 0) 19321da177e4SLinus Torvalds return csum; 19331da177e4SLinus Torvalds offset += copy; 19341da177e4SLinus Torvalds to += copy; 19351da177e4SLinus Torvalds pos += copy; 19361da177e4SLinus Torvalds } 19371a028e50SDavid S. Miller start = end; 19381da177e4SLinus Torvalds } 193909a62660SKris Katterjohn BUG_ON(len); 19401da177e4SLinus Torvalds return csum; 19411da177e4SLinus Torvalds } 1942b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 19431da177e4SLinus Torvalds 19441da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 19451da177e4SLinus Torvalds { 1946d3bc23e7SAl Viro __wsum csum; 19471da177e4SLinus Torvalds long csstart; 19481da177e4SLinus Torvalds 194984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 195055508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 19511da177e4SLinus Torvalds else 19521da177e4SLinus Torvalds csstart = skb_headlen(skb); 19531da177e4SLinus Torvalds 195409a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 19551da177e4SLinus Torvalds 1956d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 19571da177e4SLinus Torvalds 19581da177e4SLinus Torvalds csum = 0; 19591da177e4SLinus Torvalds if (csstart != skb->len) 19601da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 19611da177e4SLinus Torvalds skb->len - csstart, 0); 19621da177e4SLinus Torvalds 196384fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 1964ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 19651da177e4SLinus Torvalds 1966d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 19671da177e4SLinus Torvalds } 19681da177e4SLinus Torvalds } 1969b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 19701da177e4SLinus Torvalds 19711da177e4SLinus Torvalds /** 19721da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 19731da177e4SLinus Torvalds * @list: list to dequeue from 19741da177e4SLinus Torvalds * 19751da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 19761da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 19771da177e4SLinus Torvalds * returned or %NULL if the list is empty. 19781da177e4SLinus Torvalds */ 19791da177e4SLinus Torvalds 19801da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 19811da177e4SLinus Torvalds { 19821da177e4SLinus Torvalds unsigned long flags; 19831da177e4SLinus Torvalds struct sk_buff *result; 19841da177e4SLinus Torvalds 19851da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 19861da177e4SLinus Torvalds result = __skb_dequeue(list); 19871da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 19881da177e4SLinus Torvalds return result; 19891da177e4SLinus Torvalds } 1990b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 19911da177e4SLinus Torvalds 19921da177e4SLinus Torvalds /** 19931da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 19941da177e4SLinus Torvalds * @list: list to dequeue from 19951da177e4SLinus Torvalds * 19961da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 19971da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 19981da177e4SLinus Torvalds * returned or %NULL if the list is empty. 19991da177e4SLinus Torvalds */ 20001da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 20011da177e4SLinus Torvalds { 20021da177e4SLinus Torvalds unsigned long flags; 20031da177e4SLinus Torvalds struct sk_buff *result; 20041da177e4SLinus Torvalds 20051da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 20061da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 20071da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 20081da177e4SLinus Torvalds return result; 20091da177e4SLinus Torvalds } 2010b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 20111da177e4SLinus Torvalds 20121da177e4SLinus Torvalds /** 20131da177e4SLinus Torvalds * skb_queue_purge - empty a list 20141da177e4SLinus Torvalds * @list: list to empty 20151da177e4SLinus Torvalds * 20161da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 20171da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 20181da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 20191da177e4SLinus Torvalds */ 20201da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 20211da177e4SLinus Torvalds { 20221da177e4SLinus Torvalds struct sk_buff *skb; 20231da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 20241da177e4SLinus Torvalds kfree_skb(skb); 20251da177e4SLinus Torvalds } 2026b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 20271da177e4SLinus Torvalds 20281da177e4SLinus Torvalds /** 20291da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 20301da177e4SLinus Torvalds * @list: list to use 20311da177e4SLinus Torvalds * @newsk: buffer to queue 20321da177e4SLinus Torvalds * 20331da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 20341da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 20351da177e4SLinus Torvalds * safely. 20361da177e4SLinus Torvalds * 20371da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 20381da177e4SLinus Torvalds */ 20391da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 20401da177e4SLinus Torvalds { 20411da177e4SLinus Torvalds unsigned long flags; 20421da177e4SLinus Torvalds 20431da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 20441da177e4SLinus Torvalds __skb_queue_head(list, newsk); 20451da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 20461da177e4SLinus Torvalds } 2047b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 20481da177e4SLinus Torvalds 20491da177e4SLinus Torvalds /** 20501da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 20511da177e4SLinus Torvalds * @list: list to use 20521da177e4SLinus Torvalds * @newsk: buffer to queue 20531da177e4SLinus Torvalds * 20541da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 20551da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 20561da177e4SLinus Torvalds * safely. 20571da177e4SLinus Torvalds * 20581da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 20591da177e4SLinus Torvalds */ 20601da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 20611da177e4SLinus Torvalds { 20621da177e4SLinus Torvalds unsigned long flags; 20631da177e4SLinus Torvalds 20641da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 20651da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 20661da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 20671da177e4SLinus Torvalds } 2068b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 20698728b834SDavid S. Miller 20701da177e4SLinus Torvalds /** 20711da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 20721da177e4SLinus Torvalds * @skb: buffer to remove 20738728b834SDavid S. Miller * @list: list to use 20741da177e4SLinus Torvalds * 20758728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 20768728b834SDavid S. Miller * function is atomic with respect to other list locked calls 20771da177e4SLinus Torvalds * 20788728b834SDavid S. Miller * You must know what list the SKB is on. 20791da177e4SLinus Torvalds */ 20808728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 20811da177e4SLinus Torvalds { 20821da177e4SLinus Torvalds unsigned long flags; 20831da177e4SLinus Torvalds 20841da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 20858728b834SDavid S. Miller __skb_unlink(skb, list); 20861da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 20871da177e4SLinus Torvalds } 2088b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 20891da177e4SLinus Torvalds 20901da177e4SLinus Torvalds /** 20911da177e4SLinus Torvalds * skb_append - append a buffer 20921da177e4SLinus Torvalds * @old: buffer to insert after 20931da177e4SLinus Torvalds * @newsk: buffer to insert 20948728b834SDavid S. Miller * @list: list to use 20951da177e4SLinus Torvalds * 20961da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 20971da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 20981da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 20991da177e4SLinus Torvalds */ 21008728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 21011da177e4SLinus Torvalds { 21021da177e4SLinus Torvalds unsigned long flags; 21031da177e4SLinus Torvalds 21048728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 21057de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 21068728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 21071da177e4SLinus Torvalds } 2108b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 21091da177e4SLinus Torvalds 21101da177e4SLinus Torvalds /** 21111da177e4SLinus Torvalds * skb_insert - insert a buffer 21121da177e4SLinus Torvalds * @old: buffer to insert before 21131da177e4SLinus Torvalds * @newsk: buffer to insert 21148728b834SDavid S. Miller * @list: list to use 21151da177e4SLinus Torvalds * 21168728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 21178728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 21188728b834SDavid S. Miller * calls. 21198728b834SDavid S. Miller * 21201da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 21211da177e4SLinus Torvalds */ 21228728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 21231da177e4SLinus Torvalds { 21241da177e4SLinus Torvalds unsigned long flags; 21251da177e4SLinus Torvalds 21268728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 21278728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 21288728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 21291da177e4SLinus Torvalds } 2130b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 21311da177e4SLinus Torvalds 21321da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 21331da177e4SLinus Torvalds struct sk_buff* skb1, 21341da177e4SLinus Torvalds const u32 len, const int pos) 21351da177e4SLinus Torvalds { 21361da177e4SLinus Torvalds int i; 21371da177e4SLinus Torvalds 2138d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2139d626f62bSArnaldo Carvalho de Melo pos - len); 21401da177e4SLinus Torvalds /* And move data appendix as is. */ 21411da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 21421da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 21431da177e4SLinus Torvalds 21441da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 21451da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 21461da177e4SLinus Torvalds skb1->data_len = skb->data_len; 21471da177e4SLinus Torvalds skb1->len += skb1->data_len; 21481da177e4SLinus Torvalds skb->data_len = 0; 21491da177e4SLinus Torvalds skb->len = len; 215027a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 21511da177e4SLinus Torvalds } 21521da177e4SLinus Torvalds 21531da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 21541da177e4SLinus Torvalds struct sk_buff* skb1, 21551da177e4SLinus Torvalds const u32 len, int pos) 21561da177e4SLinus Torvalds { 21571da177e4SLinus Torvalds int i, k = 0; 21581da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 21591da177e4SLinus Torvalds 21601da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 21611da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 21621da177e4SLinus Torvalds skb->len = len; 21631da177e4SLinus Torvalds skb->data_len = len - pos; 21641da177e4SLinus Torvalds 21651da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 21661da177e4SLinus Torvalds int size = skb_shinfo(skb)->frags[i].size; 21671da177e4SLinus Torvalds 21681da177e4SLinus Torvalds if (pos + size > len) { 21691da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 21701da177e4SLinus Torvalds 21711da177e4SLinus Torvalds if (pos < len) { 21721da177e4SLinus Torvalds /* Split frag. 21731da177e4SLinus Torvalds * We have two variants in this case: 21741da177e4SLinus Torvalds * 1. Move all the frag to the second 21751da177e4SLinus Torvalds * part, if it is possible. F.e. 21761da177e4SLinus Torvalds * this approach is mandatory for TUX, 21771da177e4SLinus Torvalds * where splitting is expensive. 21781da177e4SLinus Torvalds * 2. Split is accurately. We make this. 21791da177e4SLinus Torvalds */ 2180ea2ab693SIan Campbell skb_frag_ref(skb, i); 21811da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 21821da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].size -= len - pos; 21831da177e4SLinus Torvalds skb_shinfo(skb)->frags[i].size = len - pos; 21841da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 21851da177e4SLinus Torvalds } 21861da177e4SLinus Torvalds k++; 21871da177e4SLinus Torvalds } else 21881da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 21891da177e4SLinus Torvalds pos += size; 21901da177e4SLinus Torvalds } 21911da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 21921da177e4SLinus Torvalds } 21931da177e4SLinus Torvalds 21941da177e4SLinus Torvalds /** 21951da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 21961da177e4SLinus Torvalds * @skb: the buffer to split 21971da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 21981da177e4SLinus Torvalds * @len: new length for skb 21991da177e4SLinus Torvalds */ 22001da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 22011da177e4SLinus Torvalds { 22021da177e4SLinus Torvalds int pos = skb_headlen(skb); 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 22051da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 22061da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 22071da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 22081da177e4SLinus Torvalds } 2209b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 22101da177e4SLinus Torvalds 22119f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 22129f782db3SIlpo Järvinen * 22139f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 22149f782db3SIlpo Järvinen */ 2215832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2216832d11c5SIlpo Järvinen { 22170ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2218832d11c5SIlpo Järvinen } 2219832d11c5SIlpo Järvinen 2220832d11c5SIlpo Järvinen /** 2221832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2222832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2223832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2224832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2225832d11c5SIlpo Järvinen * 2226832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 2227832d11c5SIlpo Järvinen * the length of the skb, from tgt to skb. Returns number bytes shifted. 2228832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2229832d11c5SIlpo Järvinen * 2230832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2231832d11c5SIlpo Järvinen * 2232832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2233832d11c5SIlpo Järvinen * to have non-paged data as well. 2234832d11c5SIlpo Järvinen * 2235832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2236832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2237832d11c5SIlpo Järvinen */ 2238832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2239832d11c5SIlpo Järvinen { 2240832d11c5SIlpo Järvinen int from, to, merge, todo; 2241832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2242832d11c5SIlpo Järvinen 2243832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2244832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2245832d11c5SIlpo Järvinen 2246832d11c5SIlpo Järvinen todo = shiftlen; 2247832d11c5SIlpo Järvinen from = 0; 2248832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2249832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2250832d11c5SIlpo Järvinen 2251832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2252832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2253832d11c5SIlpo Järvinen */ 2254832d11c5SIlpo Järvinen if (!to || 2255ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2256ea2ab693SIan Campbell fragfrom->page_offset)) { 2257832d11c5SIlpo Järvinen merge = -1; 2258832d11c5SIlpo Järvinen } else { 2259832d11c5SIlpo Järvinen merge = to - 1; 2260832d11c5SIlpo Järvinen 2261832d11c5SIlpo Järvinen todo -= fragfrom->size; 2262832d11c5SIlpo Järvinen if (todo < 0) { 2263832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2264832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2265832d11c5SIlpo Järvinen return 0; 2266832d11c5SIlpo Järvinen 22679f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 22689f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2269832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2270832d11c5SIlpo Järvinen 2271832d11c5SIlpo Järvinen fragto->size += shiftlen; 2272832d11c5SIlpo Järvinen fragfrom->size -= shiftlen; 2273832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2274832d11c5SIlpo Järvinen 2275832d11c5SIlpo Järvinen goto onlymerged; 2276832d11c5SIlpo Järvinen } 2277832d11c5SIlpo Järvinen 2278832d11c5SIlpo Järvinen from++; 2279832d11c5SIlpo Järvinen } 2280832d11c5SIlpo Järvinen 2281832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2282832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2283832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2284832d11c5SIlpo Järvinen return 0; 2285832d11c5SIlpo Järvinen 2286832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2287832d11c5SIlpo Järvinen return 0; 2288832d11c5SIlpo Järvinen 2289832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2290832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2291832d11c5SIlpo Järvinen return 0; 2292832d11c5SIlpo Järvinen 2293832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2294832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2295832d11c5SIlpo Järvinen 2296832d11c5SIlpo Järvinen if (todo >= fragfrom->size) { 2297832d11c5SIlpo Järvinen *fragto = *fragfrom; 2298832d11c5SIlpo Järvinen todo -= fragfrom->size; 2299832d11c5SIlpo Järvinen from++; 2300832d11c5SIlpo Järvinen to++; 2301832d11c5SIlpo Järvinen 2302832d11c5SIlpo Järvinen } else { 2303ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2304832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2305832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 2306832d11c5SIlpo Järvinen fragto->size = todo; 2307832d11c5SIlpo Järvinen 2308832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 2309832d11c5SIlpo Järvinen fragfrom->size -= todo; 2310832d11c5SIlpo Järvinen todo = 0; 2311832d11c5SIlpo Järvinen 2312832d11c5SIlpo Järvinen to++; 2313832d11c5SIlpo Järvinen break; 2314832d11c5SIlpo Järvinen } 2315832d11c5SIlpo Järvinen } 2316832d11c5SIlpo Järvinen 2317832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2318832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2319832d11c5SIlpo Järvinen 2320832d11c5SIlpo Järvinen if (merge >= 0) { 2321832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2322832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2323832d11c5SIlpo Järvinen 2324832d11c5SIlpo Järvinen fragto->size += fragfrom->size; 2325ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2326832d11c5SIlpo Järvinen } 2327832d11c5SIlpo Järvinen 2328832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2329832d11c5SIlpo Järvinen to = 0; 2330832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2331832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2332832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2333832d11c5SIlpo Järvinen 2334832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2335832d11c5SIlpo Järvinen 2336832d11c5SIlpo Järvinen onlymerged: 2337832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2338832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2339832d11c5SIlpo Järvinen */ 2340832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2341832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2342832d11c5SIlpo Järvinen 2343832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2344832d11c5SIlpo Järvinen skb->len -= shiftlen; 2345832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2346832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2347832d11c5SIlpo Järvinen tgt->len += shiftlen; 2348832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2349832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2350832d11c5SIlpo Järvinen 2351832d11c5SIlpo Järvinen return shiftlen; 2352832d11c5SIlpo Järvinen } 2353832d11c5SIlpo Järvinen 2354677e90edSThomas Graf /** 2355677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2356677e90edSThomas Graf * @skb: the buffer to read 2357677e90edSThomas Graf * @from: lower offset of data to be read 2358677e90edSThomas Graf * @to: upper offset of data to be read 2359677e90edSThomas Graf * @st: state variable 2360677e90edSThomas Graf * 2361677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2362677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2363677e90edSThomas Graf */ 2364677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2365677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2366677e90edSThomas Graf { 2367677e90edSThomas Graf st->lower_offset = from; 2368677e90edSThomas Graf st->upper_offset = to; 2369677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2370677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2371677e90edSThomas Graf st->frag_data = NULL; 2372677e90edSThomas Graf } 2373b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2374677e90edSThomas Graf 2375677e90edSThomas Graf /** 2376677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2377677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2378677e90edSThomas Graf * @data: destination pointer for data to be returned 2379677e90edSThomas Graf * @st: state variable 2380677e90edSThomas Graf * 2381677e90edSThomas Graf * Reads a block of skb data at &consumed relative to the 2382677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2383677e90edSThomas Graf * the head of the data block to &data and returns the length 2384677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2385677e90edSThomas Graf * offset has been reached. 2386677e90edSThomas Graf * 2387677e90edSThomas Graf * The caller is not required to consume all of the data 2388677e90edSThomas Graf * returned, i.e. &consumed is typically set to the number 2389677e90edSThomas Graf * of bytes already consumed and the next call to 2390677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2391677e90edSThomas Graf * 239225985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2393677e90edSThomas Graf * this limitation is the cost for zerocopy seqeuental 2394677e90edSThomas Graf * reads of potentially non linear data. 2395677e90edSThomas Graf * 2396bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2397677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2398677e90edSThomas Graf * a stack for this purpose. 2399677e90edSThomas Graf */ 2400677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2401677e90edSThomas Graf struct skb_seq_state *st) 2402677e90edSThomas Graf { 2403677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2404677e90edSThomas Graf skb_frag_t *frag; 2405677e90edSThomas Graf 2406677e90edSThomas Graf if (unlikely(abs_offset >= st->upper_offset)) 2407677e90edSThomas Graf return 0; 2408677e90edSThomas Graf 2409677e90edSThomas Graf next_skb: 241095e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2411677e90edSThomas Graf 2412995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 241395e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2414677e90edSThomas Graf return block_limit - abs_offset; 2415677e90edSThomas Graf } 2416677e90edSThomas Graf 2417677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2418677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2419677e90edSThomas Graf 2420677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2421677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 2422677e90edSThomas Graf block_limit = frag->size + st->stepped_offset; 2423677e90edSThomas Graf 2424677e90edSThomas Graf if (abs_offset < block_limit) { 2425677e90edSThomas Graf if (!st->frag_data) 2426677e90edSThomas Graf st->frag_data = kmap_skb_frag(frag); 2427677e90edSThomas Graf 2428677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2429677e90edSThomas Graf (abs_offset - st->stepped_offset); 2430677e90edSThomas Graf 2431677e90edSThomas Graf return block_limit - abs_offset; 2432677e90edSThomas Graf } 2433677e90edSThomas Graf 2434677e90edSThomas Graf if (st->frag_data) { 2435677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2436677e90edSThomas Graf st->frag_data = NULL; 2437677e90edSThomas Graf } 2438677e90edSThomas Graf 2439677e90edSThomas Graf st->frag_idx++; 2440677e90edSThomas Graf st->stepped_offset += frag->size; 2441677e90edSThomas Graf } 2442677e90edSThomas Graf 24435b5a60daSOlaf Kirch if (st->frag_data) { 24445b5a60daSOlaf Kirch kunmap_skb_frag(st->frag_data); 24455b5a60daSOlaf Kirch st->frag_data = NULL; 24465b5a60daSOlaf Kirch } 24475b5a60daSOlaf Kirch 244821dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2449677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 245095e3b24cSHerbert Xu st->frag_idx = 0; 2451677e90edSThomas Graf goto next_skb; 245271b3346dSShyam Iyer } else if (st->cur_skb->next) { 245371b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 245471b3346dSShyam Iyer st->frag_idx = 0; 2455677e90edSThomas Graf goto next_skb; 2456677e90edSThomas Graf } 2457677e90edSThomas Graf 2458677e90edSThomas Graf return 0; 2459677e90edSThomas Graf } 2460b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2461677e90edSThomas Graf 2462677e90edSThomas Graf /** 2463677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2464677e90edSThomas Graf * @st: state variable 2465677e90edSThomas Graf * 2466677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2467677e90edSThomas Graf * returned 0. 2468677e90edSThomas Graf */ 2469677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2470677e90edSThomas Graf { 2471677e90edSThomas Graf if (st->frag_data) 2472677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2473677e90edSThomas Graf } 2474b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2475677e90edSThomas Graf 24763fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 24773fc7e8a6SThomas Graf 24783fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 24793fc7e8a6SThomas Graf struct ts_config *conf, 24803fc7e8a6SThomas Graf struct ts_state *state) 24813fc7e8a6SThomas Graf { 24823fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 24833fc7e8a6SThomas Graf } 24843fc7e8a6SThomas Graf 24853fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 24863fc7e8a6SThomas Graf { 24873fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 24883fc7e8a6SThomas Graf } 24893fc7e8a6SThomas Graf 24903fc7e8a6SThomas Graf /** 24913fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 24923fc7e8a6SThomas Graf * @skb: the buffer to look in 24933fc7e8a6SThomas Graf * @from: search offset 24943fc7e8a6SThomas Graf * @to: search limit 24953fc7e8a6SThomas Graf * @config: textsearch configuration 24963fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 24973fc7e8a6SThomas Graf * 24983fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 24993fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 25003fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 25013fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 25023fc7e8a6SThomas Graf */ 25033fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 25043fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 25053fc7e8a6SThomas Graf struct ts_state *state) 25063fc7e8a6SThomas Graf { 2507f72b948dSPhil Oester unsigned int ret; 2508f72b948dSPhil Oester 25093fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 25103fc7e8a6SThomas Graf config->finish = skb_ts_finish; 25113fc7e8a6SThomas Graf 25123fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 25133fc7e8a6SThomas Graf 2514f72b948dSPhil Oester ret = textsearch_find(config, state); 2515f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 25163fc7e8a6SThomas Graf } 2517b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 25183fc7e8a6SThomas Graf 2519e89e9cf5SAnanda Raju /** 2520e89e9cf5SAnanda Raju * skb_append_datato_frags: - append the user data to a skb 2521e89e9cf5SAnanda Raju * @sk: sock structure 2522e89e9cf5SAnanda Raju * @skb: skb structure to be appened with user data. 2523e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2524e89e9cf5SAnanda Raju * @from: pointer to user message iov 2525e89e9cf5SAnanda Raju * @length: length of the iov message 2526e89e9cf5SAnanda Raju * 2527e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2528e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2529e89e9cf5SAnanda Raju */ 2530e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2531dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2532e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2533e89e9cf5SAnanda Raju void *from, int length) 2534e89e9cf5SAnanda Raju { 2535e89e9cf5SAnanda Raju int frg_cnt = 0; 2536e89e9cf5SAnanda Raju skb_frag_t *frag = NULL; 2537e89e9cf5SAnanda Raju struct page *page = NULL; 2538e89e9cf5SAnanda Raju int copy, left; 2539e89e9cf5SAnanda Raju int offset = 0; 2540e89e9cf5SAnanda Raju int ret; 2541e89e9cf5SAnanda Raju 2542e89e9cf5SAnanda Raju do { 2543e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2544e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2545e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2546e89e9cf5SAnanda Raju return -EFAULT; 2547e89e9cf5SAnanda Raju 2548e89e9cf5SAnanda Raju /* allocate a new page for next frag */ 2549e89e9cf5SAnanda Raju page = alloc_pages(sk->sk_allocation, 0); 2550e89e9cf5SAnanda Raju 2551e89e9cf5SAnanda Raju /* If alloc_page fails just return failure and caller will 2552e89e9cf5SAnanda Raju * free previous allocated pages by doing kfree_skb() 2553e89e9cf5SAnanda Raju */ 2554e89e9cf5SAnanda Raju if (page == NULL) 2555e89e9cf5SAnanda Raju return -ENOMEM; 2556e89e9cf5SAnanda Raju 2557e89e9cf5SAnanda Raju /* initialize the next frag */ 2558e89e9cf5SAnanda Raju skb_fill_page_desc(skb, frg_cnt, page, 0, 0); 2559e89e9cf5SAnanda Raju skb->truesize += PAGE_SIZE; 2560e89e9cf5SAnanda Raju atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 2561e89e9cf5SAnanda Raju 2562e89e9cf5SAnanda Raju /* get the new initialized frag */ 2563e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2564e89e9cf5SAnanda Raju frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; 2565e89e9cf5SAnanda Raju 2566e89e9cf5SAnanda Raju /* copy the user data to page */ 2567e89e9cf5SAnanda Raju left = PAGE_SIZE - frag->page_offset; 2568e89e9cf5SAnanda Raju copy = (length > left)? left : length; 2569e89e9cf5SAnanda Raju 2570ea2ab693SIan Campbell ret = getfrag(from, skb_frag_address(frag) + frag->size, 2571e89e9cf5SAnanda Raju offset, copy, 0, skb); 2572e89e9cf5SAnanda Raju if (ret < 0) 2573e89e9cf5SAnanda Raju return -EFAULT; 2574e89e9cf5SAnanda Raju 2575e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2576e89e9cf5SAnanda Raju frag->size += copy; 2577e89e9cf5SAnanda Raju skb->len += copy; 2578e89e9cf5SAnanda Raju skb->data_len += copy; 2579e89e9cf5SAnanda Raju offset += copy; 2580e89e9cf5SAnanda Raju length -= copy; 2581e89e9cf5SAnanda Raju 2582e89e9cf5SAnanda Raju } while (length > 0); 2583e89e9cf5SAnanda Raju 2584e89e9cf5SAnanda Raju return 0; 2585e89e9cf5SAnanda Raju } 2586b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2587e89e9cf5SAnanda Raju 2588cbb042f9SHerbert Xu /** 2589cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2590cbb042f9SHerbert Xu * @skb: buffer to update 2591cbb042f9SHerbert Xu * @len: length of data pulled 2592cbb042f9SHerbert Xu * 2593cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2594fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 259584fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 259684fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 259784fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2598cbb042f9SHerbert Xu */ 2599cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2600cbb042f9SHerbert Xu { 2601cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2602cbb042f9SHerbert Xu skb->len -= len; 2603cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2604cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2605cbb042f9SHerbert Xu return skb->data += len; 2606cbb042f9SHerbert Xu } 2607f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2608f94691acSArnaldo Carvalho de Melo 2609f4c50d99SHerbert Xu /** 2610f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2611f4c50d99SHerbert Xu * @skb: buffer to segment 2612576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2613f4c50d99SHerbert Xu * 2614f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 26154c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 26164c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2617f4c50d99SHerbert Xu */ 261804ed3e74SMichał Mirosław struct sk_buff *skb_segment(struct sk_buff *skb, u32 features) 2619f4c50d99SHerbert Xu { 2620f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2621f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 262289319d38SHerbert Xu struct sk_buff *fskb = skb_shinfo(skb)->frag_list; 2623f4c50d99SHerbert Xu unsigned int mss = skb_shinfo(skb)->gso_size; 262498e399f8SArnaldo Carvalho de Melo unsigned int doffset = skb->data - skb_mac_header(skb); 2625f4c50d99SHerbert Xu unsigned int offset = doffset; 2626f4c50d99SHerbert Xu unsigned int headroom; 2627f4c50d99SHerbert Xu unsigned int len; 262804ed3e74SMichał Mirosław int sg = !!(features & NETIF_F_SG); 2629f4c50d99SHerbert Xu int nfrags = skb_shinfo(skb)->nr_frags; 2630f4c50d99SHerbert Xu int err = -ENOMEM; 2631f4c50d99SHerbert Xu int i = 0; 2632f4c50d99SHerbert Xu int pos; 2633f4c50d99SHerbert Xu 2634f4c50d99SHerbert Xu __skb_push(skb, doffset); 2635f4c50d99SHerbert Xu headroom = skb_headroom(skb); 2636f4c50d99SHerbert Xu pos = skb_headlen(skb); 2637f4c50d99SHerbert Xu 2638f4c50d99SHerbert Xu do { 2639f4c50d99SHerbert Xu struct sk_buff *nskb; 2640f4c50d99SHerbert Xu skb_frag_t *frag; 2641c8884eddSHerbert Xu int hsize; 2642f4c50d99SHerbert Xu int size; 2643f4c50d99SHerbert Xu 2644f4c50d99SHerbert Xu len = skb->len - offset; 2645f4c50d99SHerbert Xu if (len > mss) 2646f4c50d99SHerbert Xu len = mss; 2647f4c50d99SHerbert Xu 2648f4c50d99SHerbert Xu hsize = skb_headlen(skb) - offset; 2649f4c50d99SHerbert Xu if (hsize < 0) 2650f4c50d99SHerbert Xu hsize = 0; 2651c8884eddSHerbert Xu if (hsize > len || !sg) 2652c8884eddSHerbert Xu hsize = len; 2653f4c50d99SHerbert Xu 265489319d38SHerbert Xu if (!hsize && i >= nfrags) { 265589319d38SHerbert Xu BUG_ON(fskb->len != len); 265689319d38SHerbert Xu 265789319d38SHerbert Xu pos += len; 265889319d38SHerbert Xu nskb = skb_clone(fskb, GFP_ATOMIC); 265989319d38SHerbert Xu fskb = fskb->next; 266089319d38SHerbert Xu 2661f4c50d99SHerbert Xu if (unlikely(!nskb)) 2662f4c50d99SHerbert Xu goto err; 2663f4c50d99SHerbert Xu 266489319d38SHerbert Xu hsize = skb_end_pointer(nskb) - nskb->head; 266589319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 266689319d38SHerbert Xu kfree_skb(nskb); 266789319d38SHerbert Xu goto err; 266889319d38SHerbert Xu } 266989319d38SHerbert Xu 267089319d38SHerbert Xu nskb->truesize += skb_end_pointer(nskb) - nskb->head - 267189319d38SHerbert Xu hsize; 267289319d38SHerbert Xu skb_release_head_state(nskb); 267389319d38SHerbert Xu __skb_push(nskb, doffset); 267489319d38SHerbert Xu } else { 267589319d38SHerbert Xu nskb = alloc_skb(hsize + doffset + headroom, 267689319d38SHerbert Xu GFP_ATOMIC); 267789319d38SHerbert Xu 267889319d38SHerbert Xu if (unlikely(!nskb)) 267989319d38SHerbert Xu goto err; 268089319d38SHerbert Xu 268189319d38SHerbert Xu skb_reserve(nskb, headroom); 268289319d38SHerbert Xu __skb_put(nskb, doffset); 268389319d38SHerbert Xu } 268489319d38SHerbert Xu 2685f4c50d99SHerbert Xu if (segs) 2686f4c50d99SHerbert Xu tail->next = nskb; 2687f4c50d99SHerbert Xu else 2688f4c50d99SHerbert Xu segs = nskb; 2689f4c50d99SHerbert Xu tail = nskb; 2690f4c50d99SHerbert Xu 26916f85a124SHerbert Xu __copy_skb_header(nskb, skb); 2692f4c50d99SHerbert Xu nskb->mac_len = skb->mac_len; 2693f4c50d99SHerbert Xu 26943d3be433SEric Dumazet /* nskb and skb might have different headroom */ 26953d3be433SEric Dumazet if (nskb->ip_summed == CHECKSUM_PARTIAL) 26963d3be433SEric Dumazet nskb->csum_start += skb_headroom(nskb) - headroom; 26973d3be433SEric Dumazet 2698459a98edSArnaldo Carvalho de Melo skb_reset_mac_header(nskb); 2699ddc7b8e3SArnaldo Carvalho de Melo skb_set_network_header(nskb, skb->mac_len); 2700b0e380b1SArnaldo Carvalho de Melo nskb->transport_header = (nskb->network_header + 2701b0e380b1SArnaldo Carvalho de Melo skb_network_header_len(skb)); 270289319d38SHerbert Xu skb_copy_from_linear_data(skb, nskb->data, doffset); 270389319d38SHerbert Xu 27042f181855SHerbert Xu if (fskb != skb_shinfo(skb)->frag_list) 270589319d38SHerbert Xu continue; 270689319d38SHerbert Xu 2707f4c50d99SHerbert Xu if (!sg) { 27086f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 2709f4c50d99SHerbert Xu nskb->csum = skb_copy_and_csum_bits(skb, offset, 2710f4c50d99SHerbert Xu skb_put(nskb, len), 2711f4c50d99SHerbert Xu len, 0); 2712f4c50d99SHerbert Xu continue; 2713f4c50d99SHerbert Xu } 2714f4c50d99SHerbert Xu 2715f4c50d99SHerbert Xu frag = skb_shinfo(nskb)->frags; 2716f4c50d99SHerbert Xu 2717d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, 2718d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 2719f4c50d99SHerbert Xu 272089319d38SHerbert Xu while (pos < offset + len && i < nfrags) { 2721f4c50d99SHerbert Xu *frag = skb_shinfo(skb)->frags[i]; 2722ea2ab693SIan Campbell __skb_frag_ref(frag); 2723f4c50d99SHerbert Xu size = frag->size; 2724f4c50d99SHerbert Xu 2725f4c50d99SHerbert Xu if (pos < offset) { 2726f4c50d99SHerbert Xu frag->page_offset += offset - pos; 2727f4c50d99SHerbert Xu frag->size -= offset - pos; 2728f4c50d99SHerbert Xu } 2729f4c50d99SHerbert Xu 273089319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 2731f4c50d99SHerbert Xu 2732f4c50d99SHerbert Xu if (pos + size <= offset + len) { 2733f4c50d99SHerbert Xu i++; 2734f4c50d99SHerbert Xu pos += size; 2735f4c50d99SHerbert Xu } else { 2736f4c50d99SHerbert Xu frag->size -= pos + size - (offset + len); 273789319d38SHerbert Xu goto skip_fraglist; 2738f4c50d99SHerbert Xu } 2739f4c50d99SHerbert Xu 2740f4c50d99SHerbert Xu frag++; 2741f4c50d99SHerbert Xu } 2742f4c50d99SHerbert Xu 274389319d38SHerbert Xu if (pos < offset + len) { 274489319d38SHerbert Xu struct sk_buff *fskb2 = fskb; 274589319d38SHerbert Xu 274689319d38SHerbert Xu BUG_ON(pos + fskb->len != offset + len); 274789319d38SHerbert Xu 274889319d38SHerbert Xu pos += fskb->len; 274989319d38SHerbert Xu fskb = fskb->next; 275089319d38SHerbert Xu 275189319d38SHerbert Xu if (fskb2->next) { 275289319d38SHerbert Xu fskb2 = skb_clone(fskb2, GFP_ATOMIC); 275389319d38SHerbert Xu if (!fskb2) 275489319d38SHerbert Xu goto err; 275589319d38SHerbert Xu } else 275689319d38SHerbert Xu skb_get(fskb2); 275789319d38SHerbert Xu 2758fbb398a8SDavid S. Miller SKB_FRAG_ASSERT(nskb); 275989319d38SHerbert Xu skb_shinfo(nskb)->frag_list = fskb2; 276089319d38SHerbert Xu } 276189319d38SHerbert Xu 276289319d38SHerbert Xu skip_fraglist: 2763f4c50d99SHerbert Xu nskb->data_len = len - hsize; 2764f4c50d99SHerbert Xu nskb->len += nskb->data_len; 2765f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 2766f4c50d99SHerbert Xu } while ((offset += len) < skb->len); 2767f4c50d99SHerbert Xu 2768f4c50d99SHerbert Xu return segs; 2769f4c50d99SHerbert Xu 2770f4c50d99SHerbert Xu err: 2771f4c50d99SHerbert Xu while ((skb = segs)) { 2772f4c50d99SHerbert Xu segs = skb->next; 2773b08d5840SPatrick McHardy kfree_skb(skb); 2774f4c50d99SHerbert Xu } 2775f4c50d99SHerbert Xu return ERR_PTR(err); 2776f4c50d99SHerbert Xu } 2777f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 2778f4c50d99SHerbert Xu 277971d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 278071d93b39SHerbert Xu { 278171d93b39SHerbert Xu struct sk_buff *p = *head; 278271d93b39SHerbert Xu struct sk_buff *nskb; 27839aaa156cSHerbert Xu struct skb_shared_info *skbinfo = skb_shinfo(skb); 27849aaa156cSHerbert Xu struct skb_shared_info *pinfo = skb_shinfo(p); 278571d93b39SHerbert Xu unsigned int headroom; 278686911732SHerbert Xu unsigned int len = skb_gro_len(skb); 278767147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 278867147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 278971d93b39SHerbert Xu 279086911732SHerbert Xu if (p->len + len >= 65536) 279171d93b39SHerbert Xu return -E2BIG; 279271d93b39SHerbert Xu 27939aaa156cSHerbert Xu if (pinfo->frag_list) 279471d93b39SHerbert Xu goto merge; 279567147ba9SHerbert Xu else if (headlen <= offset) { 279642da6994SHerbert Xu skb_frag_t *frag; 279766e92fcfSHerbert Xu skb_frag_t *frag2; 27989aaa156cSHerbert Xu int i = skbinfo->nr_frags; 27999aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 280042da6994SHerbert Xu 280166e92fcfSHerbert Xu offset -= headlen; 280266e92fcfSHerbert Xu 280366e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 280481705ad1SHerbert Xu return -E2BIG; 280581705ad1SHerbert Xu 28069aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 28079aaa156cSHerbert Xu skbinfo->nr_frags = 0; 2808f5572068SHerbert Xu 28099aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 28109aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 281166e92fcfSHerbert Xu do { 281266e92fcfSHerbert Xu *--frag = *--frag2; 281366e92fcfSHerbert Xu } while (--i); 281466e92fcfSHerbert Xu 281566e92fcfSHerbert Xu frag->page_offset += offset; 281666e92fcfSHerbert Xu frag->size -= offset; 281766e92fcfSHerbert Xu 2818f5572068SHerbert Xu skb->truesize -= skb->data_len; 2819f5572068SHerbert Xu skb->len -= skb->data_len; 2820f5572068SHerbert Xu skb->data_len = 0; 2821f5572068SHerbert Xu 28225d38a079SHerbert Xu NAPI_GRO_CB(skb)->free = 1; 28235d38a079SHerbert Xu goto done; 282469c0cab1SHerbert Xu } else if (skb_gro_len(p) != pinfo->gso_size) 282569c0cab1SHerbert Xu return -E2BIG; 282671d93b39SHerbert Xu 282771d93b39SHerbert Xu headroom = skb_headroom(p); 28283d3be433SEric Dumazet nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); 282971d93b39SHerbert Xu if (unlikely(!nskb)) 283071d93b39SHerbert Xu return -ENOMEM; 283171d93b39SHerbert Xu 283271d93b39SHerbert Xu __copy_skb_header(nskb, p); 283371d93b39SHerbert Xu nskb->mac_len = p->mac_len; 283471d93b39SHerbert Xu 283571d93b39SHerbert Xu skb_reserve(nskb, headroom); 283686911732SHerbert Xu __skb_put(nskb, skb_gro_offset(p)); 283771d93b39SHerbert Xu 283886911732SHerbert Xu skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 283971d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 284071d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 284171d93b39SHerbert Xu 284286911732SHerbert Xu __skb_pull(p, skb_gro_offset(p)); 284386911732SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), 284486911732SHerbert Xu p->data - skb_mac_header(p)); 284571d93b39SHerbert Xu 284671d93b39SHerbert Xu *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p); 284771d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 28489aaa156cSHerbert Xu skb_shinfo(nskb)->gso_size = pinfo->gso_size; 2849622e0ca1SHerbert Xu pinfo->gso_size = 0; 285071d93b39SHerbert Xu skb_header_release(p); 285171d93b39SHerbert Xu nskb->prev = p; 285271d93b39SHerbert Xu 285371d93b39SHerbert Xu nskb->data_len += p->len; 285471d93b39SHerbert Xu nskb->truesize += p->len; 285571d93b39SHerbert Xu nskb->len += p->len; 285671d93b39SHerbert Xu 285771d93b39SHerbert Xu *head = nskb; 285871d93b39SHerbert Xu nskb->next = p->next; 285971d93b39SHerbert Xu p->next = NULL; 286071d93b39SHerbert Xu 286171d93b39SHerbert Xu p = nskb; 286271d93b39SHerbert Xu 286371d93b39SHerbert Xu merge: 286467147ba9SHerbert Xu if (offset > headlen) { 2865d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 2866d1dc7abfSMichal Schmidt 2867d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 2868d1dc7abfSMichal Schmidt skbinfo->frags[0].size -= eat; 2869d1dc7abfSMichal Schmidt skb->data_len -= eat; 2870d1dc7abfSMichal Schmidt skb->len -= eat; 287167147ba9SHerbert Xu offset = headlen; 287256035022SHerbert Xu } 287356035022SHerbert Xu 287467147ba9SHerbert Xu __skb_pull(skb, offset); 287556035022SHerbert Xu 287671d93b39SHerbert Xu p->prev->next = skb; 287771d93b39SHerbert Xu p->prev = skb; 287871d93b39SHerbert Xu skb_header_release(skb); 287971d93b39SHerbert Xu 28805d38a079SHerbert Xu done: 28815d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 288237fe4732SHerbert Xu p->data_len += len; 288337fe4732SHerbert Xu p->truesize += len; 288437fe4732SHerbert Xu p->len += len; 288571d93b39SHerbert Xu 288671d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 288771d93b39SHerbert Xu return 0; 288871d93b39SHerbert Xu } 288971d93b39SHerbert Xu EXPORT_SYMBOL_GPL(skb_gro_receive); 289071d93b39SHerbert Xu 28911da177e4SLinus Torvalds void __init skb_init(void) 28921da177e4SLinus Torvalds { 28931da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 28941da177e4SLinus Torvalds sizeof(struct sk_buff), 28951da177e4SLinus Torvalds 0, 2896e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 289720c2df83SPaul Mundt NULL); 2898d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 2899d179cd12SDavid S. Miller (2*sizeof(struct sk_buff)) + 2900d179cd12SDavid S. Miller sizeof(atomic_t), 2901d179cd12SDavid S. Miller 0, 2902e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 290320c2df83SPaul Mundt NULL); 29041da177e4SLinus Torvalds } 29051da177e4SLinus Torvalds 2906716ea3a7SDavid Howells /** 2907716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 2908716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 2909716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 2910716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 2911716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 2912716ea3a7SDavid Howells * 2913716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 2914716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 2915716ea3a7SDavid Howells */ 291651c739d1SDavid S. Miller static int 291751c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 2918716ea3a7SDavid Howells { 29191a028e50SDavid S. Miller int start = skb_headlen(skb); 29201a028e50SDavid S. Miller int i, copy = start - offset; 2921fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2922716ea3a7SDavid Howells int elt = 0; 2923716ea3a7SDavid Howells 2924716ea3a7SDavid Howells if (copy > 0) { 2925716ea3a7SDavid Howells if (copy > len) 2926716ea3a7SDavid Howells copy = len; 2927642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 2928716ea3a7SDavid Howells elt++; 2929716ea3a7SDavid Howells if ((len -= copy) == 0) 2930716ea3a7SDavid Howells return elt; 2931716ea3a7SDavid Howells offset += copy; 2932716ea3a7SDavid Howells } 2933716ea3a7SDavid Howells 2934716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 29351a028e50SDavid S. Miller int end; 2936716ea3a7SDavid Howells 2937547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29381a028e50SDavid S. Miller 29391a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 2940716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2941716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2942716ea3a7SDavid Howells 2943716ea3a7SDavid Howells if (copy > len) 2944716ea3a7SDavid Howells copy = len; 2945ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 2946642f1490SJens Axboe frag->page_offset+offset-start); 2947716ea3a7SDavid Howells elt++; 2948716ea3a7SDavid Howells if (!(len -= copy)) 2949716ea3a7SDavid Howells return elt; 2950716ea3a7SDavid Howells offset += copy; 2951716ea3a7SDavid Howells } 29521a028e50SDavid S. Miller start = end; 2953716ea3a7SDavid Howells } 2954716ea3a7SDavid Howells 2955fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 29561a028e50SDavid S. Miller int end; 2957716ea3a7SDavid Howells 2958547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29591a028e50SDavid S. Miller 2960fbb398a8SDavid S. Miller end = start + frag_iter->len; 2961716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2962716ea3a7SDavid Howells if (copy > len) 2963716ea3a7SDavid Howells copy = len; 2964fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 296551c739d1SDavid S. Miller copy); 2966716ea3a7SDavid Howells if ((len -= copy) == 0) 2967716ea3a7SDavid Howells return elt; 2968716ea3a7SDavid Howells offset += copy; 2969716ea3a7SDavid Howells } 29701a028e50SDavid S. Miller start = end; 2971716ea3a7SDavid Howells } 2972716ea3a7SDavid Howells BUG_ON(len); 2973716ea3a7SDavid Howells return elt; 2974716ea3a7SDavid Howells } 2975716ea3a7SDavid Howells 297651c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 297751c739d1SDavid S. Miller { 297851c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 297951c739d1SDavid S. Miller 2980c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 298151c739d1SDavid S. Miller 298251c739d1SDavid S. Miller return nsg; 298351c739d1SDavid S. Miller } 2984b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 298551c739d1SDavid S. Miller 2986716ea3a7SDavid Howells /** 2987716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 2988716ea3a7SDavid Howells * @skb: The socket buffer to check. 2989716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 2990716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 2991716ea3a7SDavid Howells * 2992716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 2993716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 2994716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 2995716ea3a7SDavid Howells * 2996716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 2997716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 2998716ea3a7SDavid Howells * set to point to the skb in which this space begins. 2999716ea3a7SDavid Howells * 3000716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3001716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3002716ea3a7SDavid Howells */ 3003716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3004716ea3a7SDavid Howells { 3005716ea3a7SDavid Howells int copyflag; 3006716ea3a7SDavid Howells int elt; 3007716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3008716ea3a7SDavid Howells 3009716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3010716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3011716ea3a7SDavid Howells * at the moment even if they are anonymous). 3012716ea3a7SDavid Howells */ 3013716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3014716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3015716ea3a7SDavid Howells return -ENOMEM; 3016716ea3a7SDavid Howells 3017716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 301821dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3019716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3020716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3021716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3022716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3023716ea3a7SDavid Howells 3024716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3025716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3026716ea3a7SDavid Howells return -ENOMEM; 3027716ea3a7SDavid Howells 3028716ea3a7SDavid Howells /* Voila! */ 3029716ea3a7SDavid Howells *trailer = skb; 3030716ea3a7SDavid Howells return 1; 3031716ea3a7SDavid Howells } 3032716ea3a7SDavid Howells 3033716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3034716ea3a7SDavid Howells 3035716ea3a7SDavid Howells elt = 1; 3036716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3037716ea3a7SDavid Howells copyflag = 0; 3038716ea3a7SDavid Howells 3039716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3040716ea3a7SDavid Howells int ntail = 0; 3041716ea3a7SDavid Howells 3042716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3043716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3044716ea3a7SDavid Howells * after it. */ 3045716ea3a7SDavid Howells 3046716ea3a7SDavid Howells if (skb_shared(skb1)) 3047716ea3a7SDavid Howells copyflag = 1; 3048716ea3a7SDavid Howells 3049716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3050716ea3a7SDavid Howells 3051716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3052716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 305321dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3054716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3055716ea3a7SDavid Howells ntail = tailbits + 128; 3056716ea3a7SDavid Howells } 3057716ea3a7SDavid Howells 3058716ea3a7SDavid Howells if (copyflag || 3059716ea3a7SDavid Howells skb_cloned(skb1) || 3060716ea3a7SDavid Howells ntail || 3061716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 306221dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3063716ea3a7SDavid Howells struct sk_buff *skb2; 3064716ea3a7SDavid Howells 3065716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3066716ea3a7SDavid Howells if (ntail == 0) 3067716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3068716ea3a7SDavid Howells else 3069716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3070716ea3a7SDavid Howells skb_headroom(skb1), 3071716ea3a7SDavid Howells ntail, 3072716ea3a7SDavid Howells GFP_ATOMIC); 3073716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3074716ea3a7SDavid Howells return -ENOMEM; 3075716ea3a7SDavid Howells 3076716ea3a7SDavid Howells if (skb1->sk) 3077716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3078716ea3a7SDavid Howells 3079716ea3a7SDavid Howells /* Looking around. Are we still alive? 3080716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3081716ea3a7SDavid Howells 3082716ea3a7SDavid Howells skb2->next = skb1->next; 3083716ea3a7SDavid Howells *skb_p = skb2; 3084716ea3a7SDavid Howells kfree_skb(skb1); 3085716ea3a7SDavid Howells skb1 = skb2; 3086716ea3a7SDavid Howells } 3087716ea3a7SDavid Howells elt++; 3088716ea3a7SDavid Howells *trailer = skb1; 3089716ea3a7SDavid Howells skb_p = &skb1->next; 3090716ea3a7SDavid Howells } 3091716ea3a7SDavid Howells 3092716ea3a7SDavid Howells return elt; 3093716ea3a7SDavid Howells } 3094b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3095716ea3a7SDavid Howells 3096b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3097b1faf566SEric Dumazet { 3098b1faf566SEric Dumazet struct sock *sk = skb->sk; 3099b1faf566SEric Dumazet 3100b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3101b1faf566SEric Dumazet } 3102b1faf566SEric Dumazet 3103b1faf566SEric Dumazet /* 3104b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3105b1faf566SEric Dumazet */ 3106b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3107b1faf566SEric Dumazet { 3108b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 3109b1faf566SEric Dumazet (unsigned)sk->sk_rcvbuf) 3110b1faf566SEric Dumazet return -ENOMEM; 3111b1faf566SEric Dumazet 3112b1faf566SEric Dumazet skb_orphan(skb); 3113b1faf566SEric Dumazet skb->sk = sk; 3114b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3115b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3116b1faf566SEric Dumazet 3117abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3118abb57ea4SEric Dumazet skb_dst_force(skb); 3119abb57ea4SEric Dumazet 3120b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3121b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3122b1faf566SEric Dumazet sk->sk_data_ready(sk, skb->len); 3123b1faf566SEric Dumazet return 0; 3124b1faf566SEric Dumazet } 3125b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3126b1faf566SEric Dumazet 3127ac45f602SPatrick Ohly void skb_tstamp_tx(struct sk_buff *orig_skb, 3128ac45f602SPatrick Ohly struct skb_shared_hwtstamps *hwtstamps) 3129ac45f602SPatrick Ohly { 3130ac45f602SPatrick Ohly struct sock *sk = orig_skb->sk; 3131ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3132ac45f602SPatrick Ohly struct sk_buff *skb; 3133ac45f602SPatrick Ohly int err; 3134ac45f602SPatrick Ohly 3135ac45f602SPatrick Ohly if (!sk) 3136ac45f602SPatrick Ohly return; 3137ac45f602SPatrick Ohly 3138ac45f602SPatrick Ohly skb = skb_clone(orig_skb, GFP_ATOMIC); 3139ac45f602SPatrick Ohly if (!skb) 3140ac45f602SPatrick Ohly return; 3141ac45f602SPatrick Ohly 3142ac45f602SPatrick Ohly if (hwtstamps) { 3143ac45f602SPatrick Ohly *skb_hwtstamps(skb) = 3144ac45f602SPatrick Ohly *hwtstamps; 3145ac45f602SPatrick Ohly } else { 3146ac45f602SPatrick Ohly /* 3147ac45f602SPatrick Ohly * no hardware time stamps available, 31482244d07bSOliver Hartkopp * so keep the shared tx_flags and only 3149ac45f602SPatrick Ohly * store software time stamp 3150ac45f602SPatrick Ohly */ 3151ac45f602SPatrick Ohly skb->tstamp = ktime_get_real(); 3152ac45f602SPatrick Ohly } 3153ac45f602SPatrick Ohly 3154ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3155ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3156ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3157ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 315829030374SEric Dumazet 3159ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 316029030374SEric Dumazet 3161ac45f602SPatrick Ohly if (err) 3162ac45f602SPatrick Ohly kfree_skb(skb); 3163ac45f602SPatrick Ohly } 3164ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3165ac45f602SPatrick Ohly 3166ac45f602SPatrick Ohly 3167f35d9d8aSRusty Russell /** 3168f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3169f35d9d8aSRusty Russell * @skb: the skb to set 3170f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3171f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3172f35d9d8aSRusty Russell * 3173f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3174f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3175f35d9d8aSRusty Russell * 3176f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3177f35d9d8aSRusty Russell * returns false you should drop the packet. 3178f35d9d8aSRusty Russell */ 3179f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3180f35d9d8aSRusty Russell { 31815ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 31825ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3183f35d9d8aSRusty Russell if (net_ratelimit()) 3184f35d9d8aSRusty Russell printk(KERN_WARNING 3185f35d9d8aSRusty Russell "bad partial csum: csum=%u/%u len=%u\n", 31865ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3187f35d9d8aSRusty Russell return false; 3188f35d9d8aSRusty Russell } 3189f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3190f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3191f35d9d8aSRusty Russell skb->csum_offset = off; 3192f35d9d8aSRusty Russell return true; 3193f35d9d8aSRusty Russell } 3194b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3195f35d9d8aSRusty Russell 31964497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 31974497b076SBen Hutchings { 31984497b076SBen Hutchings if (net_ratelimit()) 31994497b076SBen Hutchings pr_warning("%s: received packets cannot be forwarded" 32004497b076SBen Hutchings " while LRO is enabled\n", skb->dev->name); 32014497b076SBen Hutchings } 32024497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 3203