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 39e005d193SJoe Perches #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 40e005d193SJoe Perches 411da177e4SLinus Torvalds #include <linux/module.h> 421da177e4SLinus Torvalds #include <linux/types.h> 431da177e4SLinus Torvalds #include <linux/kernel.h> 44fe55f6d5SVegard Nossum #include <linux/kmemcheck.h> 451da177e4SLinus Torvalds #include <linux/mm.h> 461da177e4SLinus Torvalds #include <linux/interrupt.h> 471da177e4SLinus Torvalds #include <linux/in.h> 481da177e4SLinus Torvalds #include <linux/inet.h> 491da177e4SLinus Torvalds #include <linux/slab.h> 50de960aa9SFlorian Westphal #include <linux/tcp.h> 51de960aa9SFlorian Westphal #include <linux/udp.h> 521da177e4SLinus Torvalds #include <linux/netdevice.h> 531da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 541da177e4SLinus Torvalds #include <net/pkt_sched.h> 551da177e4SLinus Torvalds #endif 561da177e4SLinus Torvalds #include <linux/string.h> 571da177e4SLinus Torvalds #include <linux/skbuff.h> 589c55e01cSJens Axboe #include <linux/splice.h> 591da177e4SLinus Torvalds #include <linux/cache.h> 601da177e4SLinus Torvalds #include <linux/rtnetlink.h> 611da177e4SLinus Torvalds #include <linux/init.h> 62716ea3a7SDavid Howells #include <linux/scatterlist.h> 63ac45f602SPatrick Ohly #include <linux/errqueue.h> 64268bb0ceSLinus Torvalds #include <linux/prefetch.h> 650d5501c1SVlad Yasevich #include <linux/if_vlan.h> 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds #include <net/protocol.h> 681da177e4SLinus Torvalds #include <net/dst.h> 691da177e4SLinus Torvalds #include <net/sock.h> 701da177e4SLinus Torvalds #include <net/checksum.h> 71ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 721da177e4SLinus Torvalds #include <net/xfrm.h> 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds #include <asm/uaccess.h> 75ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7651c56b00SEric Dumazet #include <linux/highmem.h> 77b245be1fSWillem de Bruijn #include <linux/capability.h> 78b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 79a1f8e7f7SAl Viro 80d7e8883cSEric Dumazet struct kmem_cache *skbuff_head_cache __read_mostly; 81e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 821da177e4SLinus Torvalds 831da177e4SLinus Torvalds /** 84f05de73bSJean Sacren * skb_panic - private function for out-of-line support 851da177e4SLinus Torvalds * @skb: buffer 861da177e4SLinus Torvalds * @sz: size 87f05de73bSJean Sacren * @addr: address 8899d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 891da177e4SLinus Torvalds * 90f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 91f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 92f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 93f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 941da177e4SLinus Torvalds */ 95f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 9699d5851eSJames Hogan const char msg[]) 971da177e4SLinus Torvalds { 98e005d193SJoe Perches pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 9999d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1004305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 10126095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1021da177e4SLinus Torvalds BUG(); 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds 105f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1061da177e4SLinus Torvalds { 107f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1081da177e4SLinus Torvalds } 1091da177e4SLinus Torvalds 110f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 111f05de73bSJean Sacren { 112f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 113f05de73bSJean Sacren } 114c93bdd0eSMel Gorman 115c93bdd0eSMel Gorman /* 116c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 117c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 118c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 119c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 120c93bdd0eSMel Gorman * memory is free 121c93bdd0eSMel Gorman */ 122c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 123c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 12461c5e88aSstephen hemminger 12561c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 12661c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 127c93bdd0eSMel Gorman { 128c93bdd0eSMel Gorman void *obj; 129c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 130c93bdd0eSMel Gorman 131c93bdd0eSMel Gorman /* 132c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 133c93bdd0eSMel Gorman * to the reserves, fail. 134c93bdd0eSMel Gorman */ 135c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 136c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 137c93bdd0eSMel Gorman node); 138c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 139c93bdd0eSMel Gorman goto out; 140c93bdd0eSMel Gorman 141c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 142c93bdd0eSMel Gorman ret_pfmemalloc = true; 143c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 144c93bdd0eSMel Gorman 145c93bdd0eSMel Gorman out: 146c93bdd0eSMel Gorman if (pfmemalloc) 147c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 148c93bdd0eSMel Gorman 149c93bdd0eSMel Gorman return obj; 150c93bdd0eSMel Gorman } 151c93bdd0eSMel Gorman 1521da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1531da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1541da177e4SLinus Torvalds * [BEEP] leaks. 1551da177e4SLinus Torvalds * 1561da177e4SLinus Torvalds */ 1571da177e4SLinus Torvalds 1580ebd0ac5SPatrick McHardy struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) 1590ebd0ac5SPatrick McHardy { 1600ebd0ac5SPatrick McHardy struct sk_buff *skb; 1610ebd0ac5SPatrick McHardy 1620ebd0ac5SPatrick McHardy /* Get the HEAD */ 1630ebd0ac5SPatrick McHardy skb = kmem_cache_alloc_node(skbuff_head_cache, 1640ebd0ac5SPatrick McHardy gfp_mask & ~__GFP_DMA, node); 1650ebd0ac5SPatrick McHardy if (!skb) 1660ebd0ac5SPatrick McHardy goto out; 1670ebd0ac5SPatrick McHardy 1680ebd0ac5SPatrick McHardy /* 1690ebd0ac5SPatrick McHardy * Only clear those fields we need to clear, not those that we will 1700ebd0ac5SPatrick McHardy * actually initialise below. Hence, don't put any more fields after 1710ebd0ac5SPatrick McHardy * the tail pointer in struct sk_buff! 1720ebd0ac5SPatrick McHardy */ 1730ebd0ac5SPatrick McHardy memset(skb, 0, offsetof(struct sk_buff, tail)); 1745e71d9d7SPablo Neira skb->head = NULL; 1750ebd0ac5SPatrick McHardy skb->truesize = sizeof(struct sk_buff); 1760ebd0ac5SPatrick McHardy atomic_set(&skb->users, 1); 1770ebd0ac5SPatrick McHardy 17835d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 1790ebd0ac5SPatrick McHardy out: 1800ebd0ac5SPatrick McHardy return skb; 1810ebd0ac5SPatrick McHardy } 1820ebd0ac5SPatrick McHardy 1831da177e4SLinus Torvalds /** 184d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1851da177e4SLinus Torvalds * @size: size to allocate 1861da177e4SLinus Torvalds * @gfp_mask: allocation mask 187c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 188c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 189c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 190c93bdd0eSMel Gorman * allocations in case the data is required for writeback 191b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1921da177e4SLinus Torvalds * 1931da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 19494b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 19594b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 1961da177e4SLinus Torvalds * 1971da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1981da177e4SLinus Torvalds * %GFP_ATOMIC. 1991da177e4SLinus Torvalds */ 200dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 201c93bdd0eSMel Gorman int flags, int node) 2021da177e4SLinus Torvalds { 203e18b890bSChristoph Lameter struct kmem_cache *cache; 2044947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 2051da177e4SLinus Torvalds struct sk_buff *skb; 2061da177e4SLinus Torvalds u8 *data; 207c93bdd0eSMel Gorman bool pfmemalloc; 2081da177e4SLinus Torvalds 209c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 210c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 211c93bdd0eSMel Gorman 212c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 213c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 2148798b3fbSHerbert Xu 2151da177e4SLinus Torvalds /* Get the HEAD */ 216b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 2171da177e4SLinus Torvalds if (!skb) 2181da177e4SLinus Torvalds goto out; 219ec7d2f2cSEric Dumazet prefetchw(skb); 2201da177e4SLinus Torvalds 22187fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 22287fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 22387fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 22487fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 22587fb4b7bSEric Dumazet */ 226bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 22787fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 228c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2291da177e4SLinus Torvalds if (!data) 2301da177e4SLinus Torvalds goto nodata; 23187fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 23287fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 23387fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 23487fb4b7bSEric Dumazet */ 23587fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 236ec7d2f2cSEric Dumazet prefetchw(data + size); 2371da177e4SLinus Torvalds 238ca0605a7SArnaldo Carvalho de Melo /* 239c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 240c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 241c8005785SJohannes Berg * the tail pointer in struct sk_buff! 242ca0605a7SArnaldo Carvalho de Melo */ 243ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 24487fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 24587fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 246c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 2471da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2481da177e4SLinus Torvalds skb->head = data; 2491da177e4SLinus Torvalds skb->data = data; 25027a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2514305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 25235d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 25335d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 25419633e12SStephen Hemminger 2554947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2564947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 257ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2584947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 259c2aa3665SEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 2604947d3efSBenjamin LaHaise 261c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 262d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 2631da177e4SLinus Torvalds 264d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 265d0bf4a9eSEric Dumazet 266d0bf4a9eSEric Dumazet kmemcheck_annotate_bitfield(&fclones->skb2, flags1); 267d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 268d0bf4a9eSEric Dumazet atomic_set(&fclones->fclone_ref, 1); 269d179cd12SDavid S. Miller 2706ffe75ebSEric Dumazet fclones->skb2.fclone = SKB_FCLONE_CLONE; 271d0bf4a9eSEric Dumazet fclones->skb2.pfmemalloc = pfmemalloc; 272d179cd12SDavid S. Miller } 2731da177e4SLinus Torvalds out: 2741da177e4SLinus Torvalds return skb; 2751da177e4SLinus Torvalds nodata: 2768798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2771da177e4SLinus Torvalds skb = NULL; 2781da177e4SLinus Torvalds goto out; 2791da177e4SLinus Torvalds } 280b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2811da177e4SLinus Torvalds 2821da177e4SLinus Torvalds /** 2832ea2f62cSEric Dumazet * __build_skb - build a network buffer 284b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2852ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 286b2b5ce9dSEric Dumazet * 287b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 288deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2892ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2902ea2f62cSEric Dumazet * or vmalloc() 291b2b5ce9dSEric Dumazet * The return is the new skb buffer. 292b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 293b2b5ce9dSEric Dumazet * Notes : 294b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 295b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 296b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 297b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 298b2b5ce9dSEric Dumazet * before giving packet to stack. 299b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 300b2b5ce9dSEric Dumazet */ 3012ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 302b2b5ce9dSEric Dumazet { 303b2b5ce9dSEric Dumazet struct skb_shared_info *shinfo; 304b2b5ce9dSEric Dumazet struct sk_buff *skb; 305d3836f21SEric Dumazet unsigned int size = frag_size ? : ksize(data); 306b2b5ce9dSEric Dumazet 307b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 308b2b5ce9dSEric Dumazet if (!skb) 309b2b5ce9dSEric Dumazet return NULL; 310b2b5ce9dSEric Dumazet 311d3836f21SEric Dumazet size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 312b2b5ce9dSEric Dumazet 313b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 314b2b5ce9dSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 315b2b5ce9dSEric Dumazet atomic_set(&skb->users, 1); 316b2b5ce9dSEric Dumazet skb->head = data; 317b2b5ce9dSEric Dumazet skb->data = data; 318b2b5ce9dSEric Dumazet skb_reset_tail_pointer(skb); 319b2b5ce9dSEric Dumazet skb->end = skb->tail + size; 32035d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 32135d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 322b2b5ce9dSEric Dumazet 323b2b5ce9dSEric Dumazet /* make sure we initialize shinfo sequentially */ 324b2b5ce9dSEric Dumazet shinfo = skb_shinfo(skb); 325b2b5ce9dSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 326b2b5ce9dSEric Dumazet atomic_set(&shinfo->dataref, 1); 327b2b5ce9dSEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 328b2b5ce9dSEric Dumazet 329b2b5ce9dSEric Dumazet return skb; 330b2b5ce9dSEric Dumazet } 3312ea2f62cSEric Dumazet 3322ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3332ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3342ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 3352ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 3362ea2f62cSEric Dumazet */ 3372ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3382ea2f62cSEric Dumazet { 3392ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3402ea2f62cSEric Dumazet 3412ea2f62cSEric Dumazet if (skb && frag_size) { 3422ea2f62cSEric Dumazet skb->head_frag = 1; 3432ea2f62cSEric Dumazet if (virt_to_head_page(data)->pfmemalloc) 3442ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3452ea2f62cSEric Dumazet } 3462ea2f62cSEric Dumazet return skb; 3472ea2f62cSEric Dumazet } 348b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 349b2b5ce9dSEric Dumazet 350a1c7fff7SEric Dumazet struct netdev_alloc_cache { 35169b08f62SEric Dumazet struct page_frag frag; 35269b08f62SEric Dumazet /* we maintain a pagecount bias, so that we dont dirty cache line 35369b08f62SEric Dumazet * containing page->_count every time we allocate a fragment. 35469b08f62SEric Dumazet */ 355540eb7bfSAlexander Duyck unsigned int pagecnt_bias; 3569451980aSAlexander Duyck bool pfmemalloc; 357a1c7fff7SEric Dumazet }; 358a1c7fff7SEric Dumazet static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache); 359ffde7328SAlexander Duyck static DEFINE_PER_CPU(struct netdev_alloc_cache, napi_alloc_cache); 360a1c7fff7SEric Dumazet 361ffde7328SAlexander Duyck static struct page *__page_frag_refill(struct netdev_alloc_cache *nc, 362ffde7328SAlexander Duyck gfp_t gfp_mask) 3636f532612SEric Dumazet { 364ffde7328SAlexander Duyck const unsigned int order = NETDEV_FRAG_PAGE_MAX_ORDER; 365ffde7328SAlexander Duyck struct page *page = NULL; 36669b08f62SEric Dumazet gfp_t gfp = gfp_mask; 36769b08f62SEric Dumazet 368ffde7328SAlexander Duyck if (order) { 36979930f58SEric Dumazet gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | 37079930f58SEric Dumazet __GFP_NOMEMALLOC; 371ffde7328SAlexander Duyck page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 372ffde7328SAlexander Duyck nc->frag.size = PAGE_SIZE << (page ? order : 0); 37369b08f62SEric Dumazet } 374ffde7328SAlexander Duyck 375ffde7328SAlexander Duyck if (unlikely(!page)) 376ffde7328SAlexander Duyck page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); 377ffde7328SAlexander Duyck 378ffde7328SAlexander Duyck nc->frag.page = page; 379ffde7328SAlexander Duyck 380ffde7328SAlexander Duyck return page; 381ffde7328SAlexander Duyck } 382ffde7328SAlexander Duyck 3839451980aSAlexander Duyck static void *__alloc_page_frag(struct netdev_alloc_cache *nc, 384ffde7328SAlexander Duyck unsigned int fragsz, gfp_t gfp_mask) 385ffde7328SAlexander Duyck { 386ffde7328SAlexander Duyck struct page *page = nc->frag.page; 387ffde7328SAlexander Duyck unsigned int size; 388ffde7328SAlexander Duyck int offset; 389ffde7328SAlexander Duyck 390ffde7328SAlexander Duyck if (unlikely(!page)) { 391ffde7328SAlexander Duyck refill: 392ffde7328SAlexander Duyck page = __page_frag_refill(nc, gfp_mask); 393ffde7328SAlexander Duyck if (!page) 394ffde7328SAlexander Duyck return NULL; 395ffde7328SAlexander Duyck 396ffde7328SAlexander Duyck /* if size can vary use frag.size else just use PAGE_SIZE */ 397ffde7328SAlexander Duyck size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE; 398ffde7328SAlexander Duyck 3994c450583SEric Dumazet /* Even if we own the page, we do not use atomic_set(). 4004c450583SEric Dumazet * This would break get_page_unless_zero() users. 4014c450583SEric Dumazet */ 402ffde7328SAlexander Duyck atomic_add(size - 1, &page->_count); 403ffde7328SAlexander Duyck 404ffde7328SAlexander Duyck /* reset page count bias and offset to start of new frag */ 4059451980aSAlexander Duyck nc->pfmemalloc = page->pfmemalloc; 406ffde7328SAlexander Duyck nc->pagecnt_bias = size; 407ffde7328SAlexander Duyck nc->frag.offset = size; 4086f532612SEric Dumazet } 409540eb7bfSAlexander Duyck 410ffde7328SAlexander Duyck offset = nc->frag.offset - fragsz; 411ffde7328SAlexander Duyck if (unlikely(offset < 0)) { 412ffde7328SAlexander Duyck if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count)) 4136f532612SEric Dumazet goto refill; 414ffde7328SAlexander Duyck 415ffde7328SAlexander Duyck /* if size can vary use frag.size else just use PAGE_SIZE */ 416ffde7328SAlexander Duyck size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE; 417ffde7328SAlexander Duyck 4184c450583SEric Dumazet /* OK, page count is 0, we can safely set it */ 419ffde7328SAlexander Duyck atomic_set(&page->_count, size); 420ffde7328SAlexander Duyck 421ffde7328SAlexander Duyck /* reset page count bias and offset to start of new frag */ 422ffde7328SAlexander Duyck nc->pagecnt_bias = size; 423ffde7328SAlexander Duyck offset = size - fragsz; 4246f532612SEric Dumazet } 425540eb7bfSAlexander Duyck 426540eb7bfSAlexander Duyck nc->pagecnt_bias--; 427ffde7328SAlexander Duyck nc->frag.offset = offset; 428ffde7328SAlexander Duyck 429ffde7328SAlexander Duyck return page_address(page) + offset; 430ffde7328SAlexander Duyck } 431ffde7328SAlexander Duyck 432ffde7328SAlexander Duyck static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 433ffde7328SAlexander Duyck { 4349451980aSAlexander Duyck struct netdev_alloc_cache *nc; 435ffde7328SAlexander Duyck unsigned long flags; 436ffde7328SAlexander Duyck void *data; 437ffde7328SAlexander Duyck 438ffde7328SAlexander Duyck local_irq_save(flags); 4399451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 4409451980aSAlexander Duyck data = __alloc_page_frag(nc, fragsz, gfp_mask); 4416f532612SEric Dumazet local_irq_restore(flags); 4426f532612SEric Dumazet return data; 4436f532612SEric Dumazet } 444c93bdd0eSMel Gorman 445c93bdd0eSMel Gorman /** 446c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 447c93bdd0eSMel Gorman * @fragsz: fragment size 448c93bdd0eSMel Gorman * 449c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 450c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 451c93bdd0eSMel Gorman */ 452c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 453c93bdd0eSMel Gorman { 454c93bdd0eSMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 455c93bdd0eSMel Gorman } 4566f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 4576f532612SEric Dumazet 458ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 459ffde7328SAlexander Duyck { 4609451980aSAlexander Duyck struct netdev_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 4619451980aSAlexander Duyck 4629451980aSAlexander Duyck return __alloc_page_frag(nc, fragsz, gfp_mask); 463ffde7328SAlexander Duyck } 464ffde7328SAlexander Duyck 465ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 466ffde7328SAlexander Duyck { 467ffde7328SAlexander Duyck return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 468ffde7328SAlexander Duyck } 469ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 470ffde7328SAlexander Duyck 4716f532612SEric Dumazet /** 472fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 473fd11a83dSAlexander Duyck * @dev: network device to receive on 474fd11a83dSAlexander Duyck * @length: length to allocate 475fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 476fd11a83dSAlexander Duyck * 477fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 478fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 479fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 480fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 481fd11a83dSAlexander Duyck * 482fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 483fd11a83dSAlexander Duyck */ 4849451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4859451980aSAlexander Duyck gfp_t gfp_mask) 486fd11a83dSAlexander Duyck { 4879451980aSAlexander Duyck struct netdev_alloc_cache *nc; 4889451980aSAlexander Duyck unsigned long flags; 489fd11a83dSAlexander Duyck struct sk_buff *skb; 4909451980aSAlexander Duyck bool pfmemalloc; 4919451980aSAlexander Duyck void *data; 492fd11a83dSAlexander Duyck 4939451980aSAlexander Duyck len += NET_SKB_PAD; 494fd11a83dSAlexander Duyck 4959451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 4969451980aSAlexander Duyck (gfp_mask & (__GFP_WAIT | GFP_DMA))) 4979451980aSAlexander Duyck return __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 4989451980aSAlexander Duyck 4999451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5009451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5019451980aSAlexander Duyck 5029451980aSAlexander Duyck if (sk_memalloc_socks()) 5039451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5049451980aSAlexander Duyck 5059451980aSAlexander Duyck local_irq_save(flags); 5069451980aSAlexander Duyck 5079451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 5089451980aSAlexander Duyck data = __alloc_page_frag(nc, len, gfp_mask); 5099451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 5109451980aSAlexander Duyck 5119451980aSAlexander Duyck local_irq_restore(flags); 5129451980aSAlexander Duyck 5139451980aSAlexander Duyck if (unlikely(!data)) 5149451980aSAlexander Duyck return NULL; 5159451980aSAlexander Duyck 5169451980aSAlexander Duyck skb = __build_skb(data, len); 5179451980aSAlexander Duyck if (unlikely(!skb)) { 5189451980aSAlexander Duyck put_page(virt_to_head_page(data)); 5199451980aSAlexander Duyck return NULL; 5209451980aSAlexander Duyck } 5219451980aSAlexander Duyck 5229451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 5239451980aSAlexander Duyck if (pfmemalloc) 5249451980aSAlexander Duyck skb->pfmemalloc = 1; 5259451980aSAlexander Duyck skb->head_frag = 1; 5269451980aSAlexander Duyck 5278af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 5287b2e497aSChristoph Hellwig skb->dev = dev; 529fd11a83dSAlexander Duyck 5308af27456SChristoph Hellwig return skb; 5318af27456SChristoph Hellwig } 532b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 5331da177e4SLinus Torvalds 534fd11a83dSAlexander Duyck /** 535fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 536fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 537fd11a83dSAlexander Duyck * @length: length to allocate 538fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 539fd11a83dSAlexander Duyck * 540fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 541fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 542fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 543fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 544fd11a83dSAlexander Duyck * 545fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 546fd11a83dSAlexander Duyck */ 5479451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 5489451980aSAlexander Duyck gfp_t gfp_mask) 549fd11a83dSAlexander Duyck { 5509451980aSAlexander Duyck struct netdev_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 551fd11a83dSAlexander Duyck struct sk_buff *skb; 5529451980aSAlexander Duyck void *data; 553fd11a83dSAlexander Duyck 5549451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 555fd11a83dSAlexander Duyck 5569451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 5579451980aSAlexander Duyck (gfp_mask & (__GFP_WAIT | GFP_DMA))) 5589451980aSAlexander Duyck return __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 5599451980aSAlexander Duyck 5609451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5619451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5629451980aSAlexander Duyck 5639451980aSAlexander Duyck if (sk_memalloc_socks()) 5649451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5659451980aSAlexander Duyck 5669451980aSAlexander Duyck data = __alloc_page_frag(nc, len, gfp_mask); 5679451980aSAlexander Duyck if (unlikely(!data)) 5689451980aSAlexander Duyck return NULL; 5699451980aSAlexander Duyck 5709451980aSAlexander Duyck skb = __build_skb(data, len); 5719451980aSAlexander Duyck if (unlikely(!skb)) { 5729451980aSAlexander Duyck put_page(virt_to_head_page(data)); 5739451980aSAlexander Duyck return NULL; 5749451980aSAlexander Duyck } 5759451980aSAlexander Duyck 5769451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 5779451980aSAlexander Duyck if (nc->pfmemalloc) 5789451980aSAlexander Duyck skb->pfmemalloc = 1; 5799451980aSAlexander Duyck skb->head_frag = 1; 5809451980aSAlexander Duyck 581fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 582fd11a83dSAlexander Duyck skb->dev = napi->dev; 583fd11a83dSAlexander Duyck 584fd11a83dSAlexander Duyck return skb; 585fd11a83dSAlexander Duyck } 586fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 587fd11a83dSAlexander Duyck 588654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 58950269e19SEric Dumazet int size, unsigned int truesize) 590654bed16SPeter Zijlstra { 591654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 592654bed16SPeter Zijlstra skb->len += size; 593654bed16SPeter Zijlstra skb->data_len += size; 59450269e19SEric Dumazet skb->truesize += truesize; 595654bed16SPeter Zijlstra } 596654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 597654bed16SPeter Zijlstra 598f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 599f8e617e1SJason Wang unsigned int truesize) 600f8e617e1SJason Wang { 601f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 602f8e617e1SJason Wang 603f8e617e1SJason Wang skb_frag_size_add(frag, size); 604f8e617e1SJason Wang skb->len += size; 605f8e617e1SJason Wang skb->data_len += size; 606f8e617e1SJason Wang skb->truesize += truesize; 607f8e617e1SJason Wang } 608f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 609f8e617e1SJason Wang 61027b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 6111da177e4SLinus Torvalds { 612bd8a7036SEric Dumazet kfree_skb_list(*listp); 61327b437c8SHerbert Xu *listp = NULL; 6141da177e4SLinus Torvalds } 6151da177e4SLinus Torvalds 61627b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 61727b437c8SHerbert Xu { 61827b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 61927b437c8SHerbert Xu } 62027b437c8SHerbert Xu 6211da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 6221da177e4SLinus Torvalds { 6231da177e4SLinus Torvalds struct sk_buff *list; 6241da177e4SLinus Torvalds 625fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 6261da177e4SLinus Torvalds skb_get(list); 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds 629d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 630d3836f21SEric Dumazet { 631d3836f21SEric Dumazet if (skb->head_frag) 632d3836f21SEric Dumazet put_page(virt_to_head_page(skb->head)); 633d3836f21SEric Dumazet else 634d3836f21SEric Dumazet kfree(skb->head); 635d3836f21SEric Dumazet } 636d3836f21SEric Dumazet 6375bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 6381da177e4SLinus Torvalds { 639ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6401da177e4SLinus Torvalds int i; 641ff04a771SEric Dumazet 642ff04a771SEric Dumazet if (skb->cloned && 643ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 644ff04a771SEric Dumazet &shinfo->dataref)) 645ff04a771SEric Dumazet return; 646ff04a771SEric Dumazet 647ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 648ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 6491da177e4SLinus Torvalds 650a6686f2fSShirley Ma /* 651a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 652a6686f2fSShirley Ma * the lower device DMA has done; 653a6686f2fSShirley Ma */ 654ff04a771SEric Dumazet if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) { 655a6686f2fSShirley Ma struct ubuf_info *uarg; 656a6686f2fSShirley Ma 657ff04a771SEric Dumazet uarg = shinfo->destructor_arg; 658a6686f2fSShirley Ma if (uarg->callback) 659e19d6763SMichael S. Tsirkin uarg->callback(uarg, true); 660a6686f2fSShirley Ma } 661a6686f2fSShirley Ma 662ff04a771SEric Dumazet if (shinfo->frag_list) 663ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6641da177e4SLinus Torvalds 665d3836f21SEric Dumazet skb_free_head(skb); 6661da177e4SLinus Torvalds } 6671da177e4SLinus Torvalds 6681da177e4SLinus Torvalds /* 6691da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6701da177e4SLinus Torvalds */ 6712d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6721da177e4SLinus Torvalds { 673d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 674d179cd12SDavid S. Miller 675d179cd12SDavid S. Miller switch (skb->fclone) { 676d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6771da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6786ffe75ebSEric Dumazet return; 679d179cd12SDavid S. Miller 680d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 681d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6826ffe75ebSEric Dumazet 6836ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6846ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6856ffe75ebSEric Dumazet * while here, branch prediction will be good. 6866ffe75ebSEric Dumazet */ 6876ffe75ebSEric Dumazet if (atomic_read(&fclones->fclone_ref) == 1) 6886ffe75ebSEric Dumazet goto fastpath; 689d179cd12SDavid S. Miller break; 690d179cd12SDavid S. Miller 6916ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 692d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 693d179cd12SDavid S. Miller break; 6943ff50b79SStephen Hemminger } 6956ffe75ebSEric Dumazet if (!atomic_dec_and_test(&fclones->fclone_ref)) 6966ffe75ebSEric Dumazet return; 6976ffe75ebSEric Dumazet fastpath: 6986ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6991da177e4SLinus Torvalds } 7001da177e4SLinus Torvalds 70104a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 7021da177e4SLinus Torvalds { 703adf30907SEric Dumazet skb_dst_drop(skb); 7041da177e4SLinus Torvalds #ifdef CONFIG_XFRM 7051da177e4SLinus Torvalds secpath_put(skb->sp); 7061da177e4SLinus Torvalds #endif 7071da177e4SLinus Torvalds if (skb->destructor) { 7089c2b3328SStephen Hemminger WARN_ON(in_irq()); 7091da177e4SLinus Torvalds skb->destructor(skb); 7101da177e4SLinus Torvalds } 711a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 7125f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 7132fc72c7bSKOVACS Krisztian #endif 7141109a90cSPablo Neira Ayuso #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 7151da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 7161da177e4SLinus Torvalds #endif 71704a4bb55SLennert Buytenhek } 71804a4bb55SLennert Buytenhek 71904a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 72004a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 72104a4bb55SLennert Buytenhek { 72204a4bb55SLennert Buytenhek skb_release_head_state(skb); 7235e71d9d7SPablo Neira if (likely(skb->head)) 7242d4baff8SHerbert Xu skb_release_data(skb); 7252d4baff8SHerbert Xu } 7261da177e4SLinus Torvalds 7272d4baff8SHerbert Xu /** 7282d4baff8SHerbert Xu * __kfree_skb - private function 7292d4baff8SHerbert Xu * @skb: buffer 7302d4baff8SHerbert Xu * 7312d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 7322d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 7332d4baff8SHerbert Xu * always call kfree_skb 7342d4baff8SHerbert Xu */ 7352d4baff8SHerbert Xu 7362d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 7372d4baff8SHerbert Xu { 7382d4baff8SHerbert Xu skb_release_all(skb); 7391da177e4SLinus Torvalds kfree_skbmem(skb); 7401da177e4SLinus Torvalds } 741b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds /** 744231d06aeSJörn Engel * kfree_skb - free an sk_buff 745231d06aeSJörn Engel * @skb: buffer to free 746231d06aeSJörn Engel * 747231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 748231d06aeSJörn Engel * hit zero. 749231d06aeSJörn Engel */ 750231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 751231d06aeSJörn Engel { 752231d06aeSJörn Engel if (unlikely(!skb)) 753231d06aeSJörn Engel return; 754231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 755231d06aeSJörn Engel smp_rmb(); 756231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 757231d06aeSJörn Engel return; 758ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 759231d06aeSJörn Engel __kfree_skb(skb); 760231d06aeSJörn Engel } 761b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 762231d06aeSJörn Engel 763bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 764bd8a7036SEric Dumazet { 765bd8a7036SEric Dumazet while (segs) { 766bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 767bd8a7036SEric Dumazet 768bd8a7036SEric Dumazet kfree_skb(segs); 769bd8a7036SEric Dumazet segs = next; 770bd8a7036SEric Dumazet } 771bd8a7036SEric Dumazet } 772bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 773bd8a7036SEric Dumazet 774d1a203eaSStephen Hemminger /** 77525121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 77625121173SMichael S. Tsirkin * @skb: buffer that triggered an error 77725121173SMichael S. Tsirkin * 77825121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 77925121173SMichael S. Tsirkin * skb must be freed afterwards. 78025121173SMichael S. Tsirkin */ 78125121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 78225121173SMichael S. Tsirkin { 78325121173SMichael S. Tsirkin if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 78425121173SMichael S. Tsirkin struct ubuf_info *uarg; 78525121173SMichael S. Tsirkin 78625121173SMichael S. Tsirkin uarg = skb_shinfo(skb)->destructor_arg; 78725121173SMichael S. Tsirkin if (uarg->callback) 78825121173SMichael S. Tsirkin uarg->callback(uarg, false); 78925121173SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 79025121173SMichael S. Tsirkin } 79125121173SMichael S. Tsirkin } 79225121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 79325121173SMichael S. Tsirkin 79425121173SMichael S. Tsirkin /** 795ead2ceb0SNeil Horman * consume_skb - free an skbuff 796ead2ceb0SNeil Horman * @skb: buffer to free 797ead2ceb0SNeil Horman * 798ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 799ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 800ead2ceb0SNeil Horman * is being dropped after a failure and notes that 801ead2ceb0SNeil Horman */ 802ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 803ead2ceb0SNeil Horman { 804ead2ceb0SNeil Horman if (unlikely(!skb)) 805ead2ceb0SNeil Horman return; 806ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 807ead2ceb0SNeil Horman smp_rmb(); 808ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 809ead2ceb0SNeil Horman return; 81007dc22e7SKoki Sanagi trace_consume_skb(skb); 811ead2ceb0SNeil Horman __kfree_skb(skb); 812ead2ceb0SNeil Horman } 813ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 814ead2ceb0SNeil Horman 815b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 816b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 817b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 818b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 819b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 820b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 821b1937227SEric Dumazet 822dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 823dec18810SHerbert Xu { 824dec18810SHerbert Xu new->tstamp = old->tstamp; 825b1937227SEric Dumazet /* We do not copy old->sk */ 826dec18810SHerbert Xu new->dev = old->dev; 827b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 8287fee226aSEric Dumazet skb_dst_copy(new, old); 829def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 830dec18810SHerbert Xu new->sp = secpath_get(old->sp); 831dec18810SHerbert Xu #endif 832b1937227SEric Dumazet __nf_copy(new, old, false); 8336aa895b0SPatrick McHardy 834b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 835b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 836b1937227SEric Dumazet */ 837b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 83806021292SEliezer Tamir 839b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 840b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 841b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 842b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 843b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 844b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 845b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 846b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 847b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 848b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 849b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 850b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 851b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 852b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 853b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 854b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 855b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 856b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 857b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 858b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 859b1937227SEric Dumazet #endif 860e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 861b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 86206021292SEliezer Tamir #endif 8632bd82484SEric Dumazet #ifdef CONFIG_XPS 8642bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 8652bd82484SEric Dumazet #endif 866b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 867b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 868b1937227SEric Dumazet #ifdef CONFIG_NET_CLS_ACT 869b1937227SEric Dumazet CHECK_SKB_FIELD(tc_verd); 870b1937227SEric Dumazet #endif 871b1937227SEric Dumazet #endif 872b1937227SEric Dumazet 873dec18810SHerbert Xu } 874dec18810SHerbert Xu 87582c49a35SHerbert Xu /* 87682c49a35SHerbert Xu * You should not add any new code to this function. Add it to 87782c49a35SHerbert Xu * __copy_skb_header above instead. 87882c49a35SHerbert Xu */ 879e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 8801da177e4SLinus Torvalds { 8811da177e4SLinus Torvalds #define C(x) n->x = skb->x 8821da177e4SLinus Torvalds 8831da177e4SLinus Torvalds n->next = n->prev = NULL; 8841da177e4SLinus Torvalds n->sk = NULL; 885dec18810SHerbert Xu __copy_skb_header(n, skb); 886dec18810SHerbert Xu 8871da177e4SLinus Torvalds C(len); 8881da177e4SLinus Torvalds C(data_len); 8893e6b3b2eSAlexey Dobriyan C(mac_len); 890334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 89102f1c89dSPaul Moore n->cloned = 1; 8921da177e4SLinus Torvalds n->nohdr = 0; 8931da177e4SLinus Torvalds n->destructor = NULL; 8941da177e4SLinus Torvalds C(tail); 8951da177e4SLinus Torvalds C(end); 89602f1c89dSPaul Moore C(head); 897d3836f21SEric Dumazet C(head_frag); 89802f1c89dSPaul Moore C(data); 89902f1c89dSPaul Moore C(truesize); 90002f1c89dSPaul Moore atomic_set(&n->users, 1); 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 9031da177e4SLinus Torvalds skb->cloned = 1; 9041da177e4SLinus Torvalds 9051da177e4SLinus Torvalds return n; 906e0053ec0SHerbert Xu #undef C 907e0053ec0SHerbert Xu } 908e0053ec0SHerbert Xu 909e0053ec0SHerbert Xu /** 910e0053ec0SHerbert Xu * skb_morph - morph one skb into another 911e0053ec0SHerbert Xu * @dst: the skb to receive the contents 912e0053ec0SHerbert Xu * @src: the skb to supply the contents 913e0053ec0SHerbert Xu * 914e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 915e0053ec0SHerbert Xu * supplied by the user. 916e0053ec0SHerbert Xu * 917e0053ec0SHerbert Xu * The target skb is returned upon exit. 918e0053ec0SHerbert Xu */ 919e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 920e0053ec0SHerbert Xu { 9212d4baff8SHerbert Xu skb_release_all(dst); 922e0053ec0SHerbert Xu return __skb_clone(dst, src); 923e0053ec0SHerbert Xu } 924e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 925e0053ec0SHerbert Xu 9262c53040fSBen Hutchings /** 9272c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 92848c83012SMichael S. Tsirkin * @skb: the skb to modify 92948c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 93048c83012SMichael S. Tsirkin * 93148c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 93248c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 93348c83012SMichael S. Tsirkin * to userspace pages. 93448c83012SMichael S. Tsirkin * 93548c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 93648c83012SMichael S. Tsirkin * %GFP_ATOMIC. 93748c83012SMichael S. Tsirkin * 93848c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 93948c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 94048c83012SMichael S. Tsirkin */ 94148c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 942a6686f2fSShirley Ma { 943a6686f2fSShirley Ma int i; 944a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 945a6686f2fSShirley Ma struct page *page, *head = NULL; 946a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 947a6686f2fSShirley Ma 948a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 949a6686f2fSShirley Ma u8 *vaddr; 950a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 951a6686f2fSShirley Ma 95202756ed4SKrishna Kumar page = alloc_page(gfp_mask); 953a6686f2fSShirley Ma if (!page) { 954a6686f2fSShirley Ma while (head) { 95540dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 956a6686f2fSShirley Ma put_page(head); 957a6686f2fSShirley Ma head = next; 958a6686f2fSShirley Ma } 959a6686f2fSShirley Ma return -ENOMEM; 960a6686f2fSShirley Ma } 96151c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 962a6686f2fSShirley Ma memcpy(page_address(page), 9639e903e08SEric Dumazet vaddr + f->page_offset, skb_frag_size(f)); 96451c56b00SEric Dumazet kunmap_atomic(vaddr); 96540dadff2SSunghan Suh set_page_private(page, (unsigned long)head); 966a6686f2fSShirley Ma head = page; 967a6686f2fSShirley Ma } 968a6686f2fSShirley Ma 969a6686f2fSShirley Ma /* skb frags release userspace buffers */ 97002756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 971a8605c60SIan Campbell skb_frag_unref(skb, i); 972a6686f2fSShirley Ma 973e19d6763SMichael S. Tsirkin uarg->callback(uarg, false); 974a6686f2fSShirley Ma 975a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 97602756ed4SKrishna Kumar for (i = num_frags - 1; i >= 0; i--) { 97702756ed4SKrishna Kumar __skb_fill_page_desc(skb, i, head, 0, 97802756ed4SKrishna Kumar skb_shinfo(skb)->frags[i].size); 97940dadff2SSunghan Suh head = (struct page *)page_private(head); 980a6686f2fSShirley Ma } 98148c83012SMichael S. Tsirkin 98248c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 983a6686f2fSShirley Ma return 0; 984a6686f2fSShirley Ma } 985dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 986a6686f2fSShirley Ma 987e0053ec0SHerbert Xu /** 988e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 989e0053ec0SHerbert Xu * @skb: buffer to clone 990e0053ec0SHerbert Xu * @gfp_mask: allocation priority 991e0053ec0SHerbert Xu * 992e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 993e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 994e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 995e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 996e0053ec0SHerbert Xu * 997e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 998e0053ec0SHerbert Xu * %GFP_ATOMIC. 999e0053ec0SHerbert Xu */ 1000e0053ec0SHerbert Xu 1001e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1002e0053ec0SHerbert Xu { 1003d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1004d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1005d0bf4a9eSEric Dumazet skb1); 10066ffe75ebSEric Dumazet struct sk_buff *n; 1007e0053ec0SHerbert Xu 100870008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1009a6686f2fSShirley Ma return NULL; 1010a6686f2fSShirley Ma 1011e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 10126ffe75ebSEric Dumazet atomic_read(&fclones->fclone_ref) == 1) { 10136ffe75ebSEric Dumazet n = &fclones->skb2; 10146ffe75ebSEric Dumazet atomic_set(&fclones->fclone_ref, 2); 1015e0053ec0SHerbert Xu } else { 1016c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1017c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1018c93bdd0eSMel Gorman 1019e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1020e0053ec0SHerbert Xu if (!n) 1021e0053ec0SHerbert Xu return NULL; 1022fe55f6d5SVegard Nossum 1023fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 1024e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1025e0053ec0SHerbert Xu } 1026e0053ec0SHerbert Xu 1027e0053ec0SHerbert Xu return __skb_clone(n, skb); 10281da177e4SLinus Torvalds } 1029b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 10301da177e4SLinus Torvalds 1031f5b17294SPravin B Shelar static void skb_headers_offset_update(struct sk_buff *skb, int off) 1032f5b17294SPravin B Shelar { 1033030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1034030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1035030737bcSEric Dumazet skb->csum_start += off; 1036f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1037f5b17294SPravin B Shelar skb->transport_header += off; 1038f5b17294SPravin B Shelar skb->network_header += off; 1039f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1040f5b17294SPravin B Shelar skb->mac_header += off; 1041f5b17294SPravin B Shelar skb->inner_transport_header += off; 1042f5b17294SPravin B Shelar skb->inner_network_header += off; 1043aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1044f5b17294SPravin B Shelar } 1045f5b17294SPravin B Shelar 10461da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 10471da177e4SLinus Torvalds { 1048dec18810SHerbert Xu __copy_skb_header(new, old); 1049dec18810SHerbert Xu 10507967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 10517967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 10527967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds 1055c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1056c93bdd0eSMel Gorman { 1057c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1058c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1059c93bdd0eSMel Gorman return 0; 1060c93bdd0eSMel Gorman } 1061c93bdd0eSMel Gorman 10621da177e4SLinus Torvalds /** 10631da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 10641da177e4SLinus Torvalds * @skb: buffer to copy 10651da177e4SLinus Torvalds * @gfp_mask: allocation priority 10661da177e4SLinus Torvalds * 10671da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 10681da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 10691da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 10701da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 10711da177e4SLinus Torvalds * 10721da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 10731da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 10741da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 10751da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 10761da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 10771da177e4SLinus Torvalds */ 10781da177e4SLinus Torvalds 1079dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 10801da177e4SLinus Torvalds { 10816602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1082ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1083c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1084c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 10856602cebbSEric Dumazet 10861da177e4SLinus Torvalds if (!n) 10871da177e4SLinus Torvalds return NULL; 10881da177e4SLinus Torvalds 10891da177e4SLinus Torvalds /* Set the data pointer */ 10901da177e4SLinus Torvalds skb_reserve(n, headerlen); 10911da177e4SLinus Torvalds /* Set the tail pointer and length */ 10921da177e4SLinus Torvalds skb_put(n, skb->len); 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 10951da177e4SLinus Torvalds BUG(); 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds copy_skb_header(n, skb); 10981da177e4SLinus Torvalds return n; 10991da177e4SLinus Torvalds } 1100b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds /** 1103bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 11041da177e4SLinus Torvalds * @skb: buffer to copy 1105117632e6SEric Dumazet * @headroom: headroom of new skb 11061da177e4SLinus Torvalds * @gfp_mask: allocation priority 1107bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1108bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1109bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 11101da177e4SLinus Torvalds * 11111da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 11121da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 11131da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 11141da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 11151da177e4SLinus Torvalds * or the pointer to the buffer on success. 11161da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 11171da177e4SLinus Torvalds */ 11181da177e4SLinus Torvalds 1119bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1120bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 11211da177e4SLinus Torvalds { 1122117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1123bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1124bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 11256602cebbSEric Dumazet 11261da177e4SLinus Torvalds if (!n) 11271da177e4SLinus Torvalds goto out; 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds /* Set the data pointer */ 1130117632e6SEric Dumazet skb_reserve(n, headroom); 11311da177e4SLinus Torvalds /* Set the tail pointer and length */ 11321da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 11331da177e4SLinus Torvalds /* Copy the bytes */ 1134d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 11351da177e4SLinus Torvalds 113625f484a6SHerbert Xu n->truesize += skb->data_len; 11371da177e4SLinus Torvalds n->data_len = skb->data_len; 11381da177e4SLinus Torvalds n->len = skb->len; 11391da177e4SLinus Torvalds 11401da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 11411da177e4SLinus Torvalds int i; 11421da177e4SLinus Torvalds 114370008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) { 11441511022cSDan Carpenter kfree_skb(n); 11451511022cSDan Carpenter n = NULL; 1146a6686f2fSShirley Ma goto out; 1147a6686f2fSShirley Ma } 11481da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 11491da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1150ea2ab693SIan Campbell skb_frag_ref(skb, i); 11511da177e4SLinus Torvalds } 11521da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 11531da177e4SLinus Torvalds } 11541da177e4SLinus Torvalds 115521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 11561da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 11571da177e4SLinus Torvalds skb_clone_fraglist(n); 11581da177e4SLinus Torvalds } 11591da177e4SLinus Torvalds 11601da177e4SLinus Torvalds copy_skb_header(n, skb); 11611da177e4SLinus Torvalds out: 11621da177e4SLinus Torvalds return n; 11631da177e4SLinus Torvalds } 1164bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds /** 11671da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 11681da177e4SLinus Torvalds * @skb: buffer to reallocate 11691da177e4SLinus Torvalds * @nhead: room to add at head 11701da177e4SLinus Torvalds * @ntail: room to add at tail 11711da177e4SLinus Torvalds * @gfp_mask: allocation priority 11721da177e4SLinus Torvalds * 1173bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1174bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 11751da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 11761da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 11771da177e4SLinus Torvalds * 11781da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 11791da177e4SLinus Torvalds * reloaded after call to this function. 11801da177e4SLinus Torvalds */ 11811da177e4SLinus Torvalds 118286a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1183dd0fc66fSAl Viro gfp_t gfp_mask) 11841da177e4SLinus Torvalds { 11851da177e4SLinus Torvalds int i; 11861da177e4SLinus Torvalds u8 *data; 1187ec47ea82SAlexander Duyck int size = nhead + skb_end_offset(skb) + ntail; 11881da177e4SLinus Torvalds long off; 11891da177e4SLinus Torvalds 11904edd87adSHerbert Xu BUG_ON(nhead < 0); 11914edd87adSHerbert Xu 11921da177e4SLinus Torvalds if (skb_shared(skb)) 11931da177e4SLinus Torvalds BUG(); 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 11961da177e4SLinus Torvalds 1197c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1198c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1199c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1200c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 12011da177e4SLinus Torvalds if (!data) 12021da177e4SLinus Torvalds goto nodata; 120387151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 12041da177e4SLinus Torvalds 12051da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 12066602cebbSEric Dumazet * optimized for the cases when header is void. 12076602cebbSEric Dumazet */ 12086602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 12096602cebbSEric Dumazet 12106602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 12116602cebbSEric Dumazet skb_shinfo(skb), 1212fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 12131da177e4SLinus Torvalds 12143e24591aSAlexander Duyck /* 12153e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 12163e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 12173e24591aSAlexander Duyck * be since all we did is relocate the values 12183e24591aSAlexander Duyck */ 12193e24591aSAlexander Duyck if (skb_cloned(skb)) { 1220a6686f2fSShirley Ma /* copy this zero copy skb frags */ 122170008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1222a6686f2fSShirley Ma goto nofrags; 12231da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1224ea2ab693SIan Campbell skb_frag_ref(skb, i); 12251da177e4SLinus Torvalds 122621dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 12271da177e4SLinus Torvalds skb_clone_fraglist(skb); 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds skb_release_data(skb); 12303e24591aSAlexander Duyck } else { 12313e24591aSAlexander Duyck skb_free_head(skb); 12321fd63041SEric Dumazet } 12331da177e4SLinus Torvalds off = (data + nhead) - skb->head; 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds skb->head = data; 1236d3836f21SEric Dumazet skb->head_frag = 0; 12371da177e4SLinus Torvalds skb->data += off; 12384305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 12394305b541SArnaldo Carvalho de Melo skb->end = size; 124056eb8882SPatrick McHardy off = nhead; 12414305b541SArnaldo Carvalho de Melo #else 12424305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 124356eb8882SPatrick McHardy #endif 124427a884dcSArnaldo Carvalho de Melo skb->tail += off; 1245b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 12461da177e4SLinus Torvalds skb->cloned = 0; 1247334a8132SPatrick McHardy skb->hdr_len = 0; 12481da177e4SLinus Torvalds skb->nohdr = 0; 12491da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 12501da177e4SLinus Torvalds return 0; 12511da177e4SLinus Torvalds 1252a6686f2fSShirley Ma nofrags: 1253a6686f2fSShirley Ma kfree(data); 12541da177e4SLinus Torvalds nodata: 12551da177e4SLinus Torvalds return -ENOMEM; 12561da177e4SLinus Torvalds } 1257b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 12581da177e4SLinus Torvalds 12591da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 12621da177e4SLinus Torvalds { 12631da177e4SLinus Torvalds struct sk_buff *skb2; 12641da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds if (delta <= 0) 12671da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 12681da177e4SLinus Torvalds else { 12691da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 12701da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 12711da177e4SLinus Torvalds GFP_ATOMIC)) { 12721da177e4SLinus Torvalds kfree_skb(skb2); 12731da177e4SLinus Torvalds skb2 = NULL; 12741da177e4SLinus Torvalds } 12751da177e4SLinus Torvalds } 12761da177e4SLinus Torvalds return skb2; 12771da177e4SLinus Torvalds } 1278b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 12791da177e4SLinus Torvalds 12801da177e4SLinus Torvalds /** 12811da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 12821da177e4SLinus Torvalds * @skb: buffer to copy 12831da177e4SLinus Torvalds * @newheadroom: new free bytes at head 12841da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 12851da177e4SLinus Torvalds * @gfp_mask: allocation priority 12861da177e4SLinus Torvalds * 12871da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 12881da177e4SLinus Torvalds * allocate additional space. 12891da177e4SLinus Torvalds * 12901da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 12911da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 12921da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 12931da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 12941da177e4SLinus Torvalds * 12951da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 12961da177e4SLinus Torvalds * is called from an interrupt. 12971da177e4SLinus Torvalds */ 12981da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 129986a76cafSVictor Fusco int newheadroom, int newtailroom, 1300dd0fc66fSAl Viro gfp_t gfp_mask) 13011da177e4SLinus Torvalds { 13021da177e4SLinus Torvalds /* 13031da177e4SLinus Torvalds * Allocate the copy buffer 13041da177e4SLinus Torvalds */ 1305c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1306c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1307c93bdd0eSMel Gorman NUMA_NO_NODE); 1308efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 13091da177e4SLinus Torvalds int head_copy_len, head_copy_off; 13101da177e4SLinus Torvalds 13111da177e4SLinus Torvalds if (!n) 13121da177e4SLinus Torvalds return NULL; 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds skb_reserve(n, newheadroom); 13151da177e4SLinus Torvalds 13161da177e4SLinus Torvalds /* Set the tail pointer and length */ 13171da177e4SLinus Torvalds skb_put(n, skb->len); 13181da177e4SLinus Torvalds 1319efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 13201da177e4SLinus Torvalds head_copy_off = 0; 13211da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 13221da177e4SLinus Torvalds head_copy_len = newheadroom; 13231da177e4SLinus Torvalds else 13241da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds /* Copy the linear header and data. */ 13271da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 13281da177e4SLinus Torvalds skb->len + head_copy_len)) 13291da177e4SLinus Torvalds BUG(); 13301da177e4SLinus Torvalds 13311da177e4SLinus Torvalds copy_skb_header(n, skb); 13321da177e4SLinus Torvalds 1333030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1334efd1e8d5SPatrick McHardy 13351da177e4SLinus Torvalds return n; 13361da177e4SLinus Torvalds } 1337b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds /** 13401da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 13411da177e4SLinus Torvalds * @skb: buffer to pad 13421da177e4SLinus Torvalds * @pad: space to pad 13431da177e4SLinus Torvalds * 13441da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 13451da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 13461da177e4SLinus Torvalds * beyond the buffer end onto the wire. 13471da177e4SLinus Torvalds * 13485b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 13491da177e4SLinus Torvalds */ 13501da177e4SLinus Torvalds 13515b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 13521da177e4SLinus Torvalds { 13535b057c6bSHerbert Xu int err; 13545b057c6bSHerbert Xu int ntail; 13551da177e4SLinus Torvalds 13561da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 13575b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 13581da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 13595b057c6bSHerbert Xu return 0; 13601da177e4SLinus Torvalds } 13611da177e4SLinus Torvalds 13624305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 13635b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 13645b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 13655b057c6bSHerbert Xu if (unlikely(err)) 13665b057c6bSHerbert Xu goto free_skb; 13675b057c6bSHerbert Xu } 13685b057c6bSHerbert Xu 13695b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 13705b057c6bSHerbert Xu * to be audited. 13715b057c6bSHerbert Xu */ 13725b057c6bSHerbert Xu err = skb_linearize(skb); 13735b057c6bSHerbert Xu if (unlikely(err)) 13745b057c6bSHerbert Xu goto free_skb; 13755b057c6bSHerbert Xu 13765b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 13775b057c6bSHerbert Xu return 0; 13785b057c6bSHerbert Xu 13795b057c6bSHerbert Xu free_skb: 13801da177e4SLinus Torvalds kfree_skb(skb); 13815b057c6bSHerbert Xu return err; 13821da177e4SLinus Torvalds } 1383b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 13841da177e4SLinus Torvalds 13850dde3e16SIlpo Järvinen /** 13860c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 13870c7ddf36SMathias Krause * @skb: start of the buffer to use 13880c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 13890c7ddf36SMathias Krause * @len: amount of data to add 13900c7ddf36SMathias Krause * 13910c7ddf36SMathias Krause * This function extends the used data area of the potentially 13920c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 13930c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 13940c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 13950c7ddf36SMathias Krause * returned. 13960c7ddf36SMathias Krause */ 13970c7ddf36SMathias Krause 13980c7ddf36SMathias Krause unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 13990c7ddf36SMathias Krause { 14000c7ddf36SMathias Krause if (tail != skb) { 14010c7ddf36SMathias Krause skb->data_len += len; 14020c7ddf36SMathias Krause skb->len += len; 14030c7ddf36SMathias Krause } 14040c7ddf36SMathias Krause return skb_put(tail, len); 14050c7ddf36SMathias Krause } 14060c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 14070c7ddf36SMathias Krause 14080c7ddf36SMathias Krause /** 14090dde3e16SIlpo Järvinen * skb_put - add data to a buffer 14100dde3e16SIlpo Järvinen * @skb: buffer to use 14110dde3e16SIlpo Järvinen * @len: amount of data to add 14120dde3e16SIlpo Järvinen * 14130dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 14140dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 14150dde3e16SIlpo Järvinen * first byte of the extra data is returned. 14160dde3e16SIlpo Järvinen */ 14170dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 14180dde3e16SIlpo Järvinen { 14190dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 14200dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 14210dde3e16SIlpo Järvinen skb->tail += len; 14220dde3e16SIlpo Järvinen skb->len += len; 14230dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 14240dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 14250dde3e16SIlpo Järvinen return tmp; 14260dde3e16SIlpo Järvinen } 14270dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 14280dde3e16SIlpo Järvinen 14296be8ac2fSIlpo Järvinen /** 1430c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1431c2aa270aSIlpo Järvinen * @skb: buffer to use 1432c2aa270aSIlpo Järvinen * @len: amount of data to add 1433c2aa270aSIlpo Järvinen * 1434c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1435c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1436c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1437c2aa270aSIlpo Järvinen */ 1438c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1439c2aa270aSIlpo Järvinen { 1440c2aa270aSIlpo Järvinen skb->data -= len; 1441c2aa270aSIlpo Järvinen skb->len += len; 1442c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1443c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1444c2aa270aSIlpo Järvinen return skb->data; 1445c2aa270aSIlpo Järvinen } 1446c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1447c2aa270aSIlpo Järvinen 1448c2aa270aSIlpo Järvinen /** 14496be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 14506be8ac2fSIlpo Järvinen * @skb: buffer to use 14516be8ac2fSIlpo Järvinen * @len: amount of data to remove 14526be8ac2fSIlpo Järvinen * 14536be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 14546be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 14556be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 14566be8ac2fSIlpo Järvinen * the old data. 14576be8ac2fSIlpo Järvinen */ 14586be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 14596be8ac2fSIlpo Järvinen { 146047d29646SDavid S. Miller return skb_pull_inline(skb, len); 14616be8ac2fSIlpo Järvinen } 14626be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 14636be8ac2fSIlpo Järvinen 1464419ae74eSIlpo Järvinen /** 1465419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1466419ae74eSIlpo Järvinen * @skb: buffer to alter 1467419ae74eSIlpo Järvinen * @len: new length 1468419ae74eSIlpo Järvinen * 1469419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1470419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1471419ae74eSIlpo Järvinen * The skb must be linear. 1472419ae74eSIlpo Järvinen */ 1473419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1474419ae74eSIlpo Järvinen { 1475419ae74eSIlpo Järvinen if (skb->len > len) 1476419ae74eSIlpo Järvinen __skb_trim(skb, len); 1477419ae74eSIlpo Järvinen } 1478419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1479419ae74eSIlpo Järvinen 14803cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 14811da177e4SLinus Torvalds */ 14821da177e4SLinus Torvalds 14833cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 14841da177e4SLinus Torvalds { 148527b437c8SHerbert Xu struct sk_buff **fragp; 148627b437c8SHerbert Xu struct sk_buff *frag; 14871da177e4SLinus Torvalds int offset = skb_headlen(skb); 14881da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 14891da177e4SLinus Torvalds int i; 149027b437c8SHerbert Xu int err; 149127b437c8SHerbert Xu 149227b437c8SHerbert Xu if (skb_cloned(skb) && 149327b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 149427b437c8SHerbert Xu return err; 14951da177e4SLinus Torvalds 1496f4d26fb3SHerbert Xu i = 0; 1497f4d26fb3SHerbert Xu if (offset >= len) 1498f4d26fb3SHerbert Xu goto drop_pages; 1499f4d26fb3SHerbert Xu 1500f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 15019e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 150227b437c8SHerbert Xu 150327b437c8SHerbert Xu if (end < len) { 15041da177e4SLinus Torvalds offset = end; 150527b437c8SHerbert Xu continue; 15061da177e4SLinus Torvalds } 15071da177e4SLinus Torvalds 15089e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 150927b437c8SHerbert Xu 1510f4d26fb3SHerbert Xu drop_pages: 151127b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 151227b437c8SHerbert Xu 151327b437c8SHerbert Xu for (; i < nfrags; i++) 1514ea2ab693SIan Campbell skb_frag_unref(skb, i); 151527b437c8SHerbert Xu 151621dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 151727b437c8SHerbert Xu skb_drop_fraglist(skb); 1518f4d26fb3SHerbert Xu goto done; 151927b437c8SHerbert Xu } 152027b437c8SHerbert Xu 152127b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 152227b437c8SHerbert Xu fragp = &frag->next) { 152327b437c8SHerbert Xu int end = offset + frag->len; 152427b437c8SHerbert Xu 152527b437c8SHerbert Xu if (skb_shared(frag)) { 152627b437c8SHerbert Xu struct sk_buff *nfrag; 152727b437c8SHerbert Xu 152827b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 152927b437c8SHerbert Xu if (unlikely(!nfrag)) 153027b437c8SHerbert Xu return -ENOMEM; 153127b437c8SHerbert Xu 153227b437c8SHerbert Xu nfrag->next = frag->next; 153385bb2a60SEric Dumazet consume_skb(frag); 153427b437c8SHerbert Xu frag = nfrag; 153527b437c8SHerbert Xu *fragp = frag; 153627b437c8SHerbert Xu } 153727b437c8SHerbert Xu 153827b437c8SHerbert Xu if (end < len) { 153927b437c8SHerbert Xu offset = end; 154027b437c8SHerbert Xu continue; 154127b437c8SHerbert Xu } 154227b437c8SHerbert Xu 154327b437c8SHerbert Xu if (end > len && 154427b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 154527b437c8SHerbert Xu return err; 154627b437c8SHerbert Xu 154727b437c8SHerbert Xu if (frag->next) 154827b437c8SHerbert Xu skb_drop_list(&frag->next); 154927b437c8SHerbert Xu break; 155027b437c8SHerbert Xu } 155127b437c8SHerbert Xu 1552f4d26fb3SHerbert Xu done: 155327b437c8SHerbert Xu if (len > skb_headlen(skb)) { 15541da177e4SLinus Torvalds skb->data_len -= skb->len - len; 15551da177e4SLinus Torvalds skb->len = len; 15561da177e4SLinus Torvalds } else { 15571da177e4SLinus Torvalds skb->len = len; 15581da177e4SLinus Torvalds skb->data_len = 0; 155927a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 15601da177e4SLinus Torvalds } 15611da177e4SLinus Torvalds 15621da177e4SLinus Torvalds return 0; 15631da177e4SLinus Torvalds } 1564b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 15651da177e4SLinus Torvalds 15661da177e4SLinus Torvalds /** 15671da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 15681da177e4SLinus Torvalds * @skb: buffer to reallocate 15691da177e4SLinus Torvalds * @delta: number of bytes to advance tail 15701da177e4SLinus Torvalds * 15711da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 15721da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 15731da177e4SLinus Torvalds * data from fragmented part. 15741da177e4SLinus Torvalds * 15751da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 15761da177e4SLinus Torvalds * 15771da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 15781da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 15791da177e4SLinus Torvalds * 15801da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 15811da177e4SLinus Torvalds * reloaded after call to this function. 15821da177e4SLinus Torvalds */ 15831da177e4SLinus Torvalds 15841da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 15851da177e4SLinus Torvalds * when it is necessary. 15861da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 15871da177e4SLinus Torvalds * 2. It may change skb pointers. 15881da177e4SLinus Torvalds * 15891da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 15901da177e4SLinus Torvalds */ 15911da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 15921da177e4SLinus Torvalds { 15931da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 15941da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 15951da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 15961da177e4SLinus Torvalds */ 15974305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 15981da177e4SLinus Torvalds 15991da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 16001da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 16011da177e4SLinus Torvalds GFP_ATOMIC)) 16021da177e4SLinus Torvalds return NULL; 16031da177e4SLinus Torvalds } 16041da177e4SLinus Torvalds 160527a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 16061da177e4SLinus Torvalds BUG(); 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 16091da177e4SLinus Torvalds * size of pulled pages. Superb. 16101da177e4SLinus Torvalds */ 161121dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 16121da177e4SLinus Torvalds goto pull_pages; 16131da177e4SLinus Torvalds 16141da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 16151da177e4SLinus Torvalds eat = delta; 16161da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16179e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 16189e903e08SEric Dumazet 16199e903e08SEric Dumazet if (size >= eat) 16201da177e4SLinus Torvalds goto pull_pages; 16219e903e08SEric Dumazet eat -= size; 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 16251da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 16261da177e4SLinus Torvalds * but taking into account that pulling is expected to 16271da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 16281da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 16291da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 16301da177e4SLinus Torvalds */ 16311da177e4SLinus Torvalds if (eat) { 16321da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 16331da177e4SLinus Torvalds struct sk_buff *clone = NULL; 16341da177e4SLinus Torvalds struct sk_buff *insp = NULL; 16351da177e4SLinus Torvalds 16361da177e4SLinus Torvalds do { 163709a62660SKris Katterjohn BUG_ON(!list); 16381da177e4SLinus Torvalds 16391da177e4SLinus Torvalds if (list->len <= eat) { 16401da177e4SLinus Torvalds /* Eaten as whole. */ 16411da177e4SLinus Torvalds eat -= list->len; 16421da177e4SLinus Torvalds list = list->next; 16431da177e4SLinus Torvalds insp = list; 16441da177e4SLinus Torvalds } else { 16451da177e4SLinus Torvalds /* Eaten partially. */ 16461da177e4SLinus Torvalds 16471da177e4SLinus Torvalds if (skb_shared(list)) { 16481da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 16491da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 16501da177e4SLinus Torvalds if (!clone) 16511da177e4SLinus Torvalds return NULL; 16521da177e4SLinus Torvalds insp = list->next; 16531da177e4SLinus Torvalds list = clone; 16541da177e4SLinus Torvalds } else { 16551da177e4SLinus Torvalds /* This may be pulled without 16561da177e4SLinus Torvalds * problems. */ 16571da177e4SLinus Torvalds insp = list; 16581da177e4SLinus Torvalds } 16591da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 16601da177e4SLinus Torvalds kfree_skb(clone); 16611da177e4SLinus Torvalds return NULL; 16621da177e4SLinus Torvalds } 16631da177e4SLinus Torvalds break; 16641da177e4SLinus Torvalds } 16651da177e4SLinus Torvalds } while (eat); 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds /* Free pulled out fragments. */ 16681da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 16691da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 16701da177e4SLinus Torvalds kfree_skb(list); 16711da177e4SLinus Torvalds } 16721da177e4SLinus Torvalds /* And insert new clone at head. */ 16731da177e4SLinus Torvalds if (clone) { 16741da177e4SLinus Torvalds clone->next = list; 16751da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 16761da177e4SLinus Torvalds } 16771da177e4SLinus Torvalds } 16781da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 16791da177e4SLinus Torvalds 16801da177e4SLinus Torvalds pull_pages: 16811da177e4SLinus Torvalds eat = delta; 16821da177e4SLinus Torvalds k = 0; 16831da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16849e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 16859e903e08SEric Dumazet 16869e903e08SEric Dumazet if (size <= eat) { 1687ea2ab693SIan Campbell skb_frag_unref(skb, i); 16889e903e08SEric Dumazet eat -= size; 16891da177e4SLinus Torvalds } else { 16901da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 16911da177e4SLinus Torvalds if (eat) { 16921da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 16939e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 16941da177e4SLinus Torvalds eat = 0; 16951da177e4SLinus Torvalds } 16961da177e4SLinus Torvalds k++; 16971da177e4SLinus Torvalds } 16981da177e4SLinus Torvalds } 16991da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 17001da177e4SLinus Torvalds 17011da177e4SLinus Torvalds skb->tail += delta; 17021da177e4SLinus Torvalds skb->data_len -= delta; 17031da177e4SLinus Torvalds 170427a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 17051da177e4SLinus Torvalds } 1706b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 17071da177e4SLinus Torvalds 170822019b17SEric Dumazet /** 170922019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 171022019b17SEric Dumazet * @skb: source skb 171122019b17SEric Dumazet * @offset: offset in source 171222019b17SEric Dumazet * @to: destination buffer 171322019b17SEric Dumazet * @len: number of bytes to copy 171422019b17SEric Dumazet * 171522019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 171622019b17SEric Dumazet * destination buffer. 171722019b17SEric Dumazet * 171822019b17SEric Dumazet * CAUTION ! : 171922019b17SEric Dumazet * If its prototype is ever changed, 172022019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 172122019b17SEric Dumazet * since it is called from BPF assembly code. 172222019b17SEric Dumazet */ 17231da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 17241da177e4SLinus Torvalds { 17251a028e50SDavid S. Miller int start = skb_headlen(skb); 1726fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1727fbb398a8SDavid S. Miller int i, copy; 17281da177e4SLinus Torvalds 17291da177e4SLinus Torvalds if (offset > (int)skb->len - len) 17301da177e4SLinus Torvalds goto fault; 17311da177e4SLinus Torvalds 17321da177e4SLinus Torvalds /* Copy header. */ 17331a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 17341da177e4SLinus Torvalds if (copy > len) 17351da177e4SLinus Torvalds copy = len; 1736d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 17371da177e4SLinus Torvalds if ((len -= copy) == 0) 17381da177e4SLinus Torvalds return 0; 17391da177e4SLinus Torvalds offset += copy; 17401da177e4SLinus Torvalds to += copy; 17411da177e4SLinus Torvalds } 17421da177e4SLinus Torvalds 17431da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 17441a028e50SDavid S. Miller int end; 174551c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 17461da177e4SLinus Torvalds 1747547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17481a028e50SDavid S. Miller 174951c56b00SEric Dumazet end = start + skb_frag_size(f); 17501da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17511da177e4SLinus Torvalds u8 *vaddr; 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds if (copy > len) 17541da177e4SLinus Torvalds copy = len; 17551da177e4SLinus Torvalds 175651c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 17571da177e4SLinus Torvalds memcpy(to, 175851c56b00SEric Dumazet vaddr + f->page_offset + offset - start, 175951c56b00SEric Dumazet copy); 176051c56b00SEric Dumazet kunmap_atomic(vaddr); 17611da177e4SLinus Torvalds 17621da177e4SLinus Torvalds if ((len -= copy) == 0) 17631da177e4SLinus Torvalds return 0; 17641da177e4SLinus Torvalds offset += copy; 17651da177e4SLinus Torvalds to += copy; 17661da177e4SLinus Torvalds } 17671a028e50SDavid S. Miller start = end; 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds 1770fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 17711a028e50SDavid S. Miller int end; 17721da177e4SLinus Torvalds 1773547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17741a028e50SDavid S. Miller 1775fbb398a8SDavid S. Miller end = start + frag_iter->len; 17761da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17771da177e4SLinus Torvalds if (copy > len) 17781da177e4SLinus Torvalds copy = len; 1779fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 17801da177e4SLinus Torvalds goto fault; 17811da177e4SLinus Torvalds if ((len -= copy) == 0) 17821da177e4SLinus Torvalds return 0; 17831da177e4SLinus Torvalds offset += copy; 17841da177e4SLinus Torvalds to += copy; 17851da177e4SLinus Torvalds } 17861a028e50SDavid S. Miller start = end; 17871da177e4SLinus Torvalds } 1788a6686f2fSShirley Ma 17891da177e4SLinus Torvalds if (!len) 17901da177e4SLinus Torvalds return 0; 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds fault: 17931da177e4SLinus Torvalds return -EFAULT; 17941da177e4SLinus Torvalds } 1795b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 17961da177e4SLinus Torvalds 17979c55e01cSJens Axboe /* 17989c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 17999c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 18009c55e01cSJens Axboe */ 18019c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 18029c55e01cSJens Axboe { 18038b9d3728SJarek Poplawski put_page(spd->pages[i]); 18048b9d3728SJarek Poplawski } 18059c55e01cSJens Axboe 1806a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 18074fb66994SJarek Poplawski unsigned int *offset, 180818aafc62SEric Dumazet struct sock *sk) 18098b9d3728SJarek Poplawski { 18105640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 18118b9d3728SJarek Poplawski 18125640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 18138b9d3728SJarek Poplawski return NULL; 18144fb66994SJarek Poplawski 18155640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 18164fb66994SJarek Poplawski 18175640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 18185640f768SEric Dumazet page_address(page) + *offset, *len); 18195640f768SEric Dumazet *offset = pfrag->offset; 18205640f768SEric Dumazet pfrag->offset += *len; 18214fb66994SJarek Poplawski 18225640f768SEric Dumazet return pfrag->page; 18239c55e01cSJens Axboe } 18249c55e01cSJens Axboe 182541c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 182641c73a0dSEric Dumazet struct page *page, 182741c73a0dSEric Dumazet unsigned int offset) 182841c73a0dSEric Dumazet { 182941c73a0dSEric Dumazet return spd->nr_pages && 183041c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 183141c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 183241c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 183341c73a0dSEric Dumazet } 183441c73a0dSEric Dumazet 18359c55e01cSJens Axboe /* 18369c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 18379c55e01cSJens Axboe */ 1838a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 183935f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 18404fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 184118aafc62SEric Dumazet bool linear, 18427a67e56fSJarek Poplawski struct sock *sk) 18439c55e01cSJens Axboe { 184441c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1845a108d5f3SDavid S. Miller return true; 18469c55e01cSJens Axboe 18478b9d3728SJarek Poplawski if (linear) { 184818aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 18498b9d3728SJarek Poplawski if (!page) 1850a108d5f3SDavid S. Miller return true; 185141c73a0dSEric Dumazet } 185241c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 185341c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 1854a108d5f3SDavid S. Miller return false; 185541c73a0dSEric Dumazet } 18568b9d3728SJarek Poplawski get_page(page); 18579c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 18584fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 18599c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 18609c55e01cSJens Axboe spd->nr_pages++; 18618b9d3728SJarek Poplawski 1862a108d5f3SDavid S. Miller return false; 18639c55e01cSJens Axboe } 18649c55e01cSJens Axboe 1865a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 18662870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 186718aafc62SEric Dumazet unsigned int *len, 1868d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 186935f3d14dSJens Axboe struct sock *sk, 187035f3d14dSJens Axboe struct pipe_inode_info *pipe) 18719c55e01cSJens Axboe { 18722870c43dSOctavian Purdila if (!*len) 1873a108d5f3SDavid S. Miller return true; 18749c55e01cSJens Axboe 18752870c43dSOctavian Purdila /* skip this segment if already processed */ 18762870c43dSOctavian Purdila if (*off >= plen) { 18772870c43dSOctavian Purdila *off -= plen; 1878a108d5f3SDavid S. Miller return false; 18792870c43dSOctavian Purdila } 18802870c43dSOctavian Purdila 18812870c43dSOctavian Purdila /* ignore any bits we already processed */ 18829ca1b22dSEric Dumazet poff += *off; 18839ca1b22dSEric Dumazet plen -= *off; 18842870c43dSOctavian Purdila *off = 0; 18852870c43dSOctavian Purdila 188618aafc62SEric Dumazet do { 188718aafc62SEric Dumazet unsigned int flen = min(*len, plen); 18882870c43dSOctavian Purdila 188918aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 189018aafc62SEric Dumazet linear, sk)) 1891a108d5f3SDavid S. Miller return true; 189218aafc62SEric Dumazet poff += flen; 189318aafc62SEric Dumazet plen -= flen; 18942870c43dSOctavian Purdila *len -= flen; 189518aafc62SEric Dumazet } while (*len && plen); 18962870c43dSOctavian Purdila 1897a108d5f3SDavid S. Miller return false; 1898db43a282SOctavian Purdila } 18999c55e01cSJens Axboe 19009c55e01cSJens Axboe /* 1901a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 19022870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 19039c55e01cSJens Axboe */ 1904a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 190535f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 190635f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 19072870c43dSOctavian Purdila { 19082870c43dSOctavian Purdila int seg; 19099c55e01cSJens Axboe 19101d0c0b32SEric Dumazet /* map the linear part : 19112996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 19122996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 19132996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 19149c55e01cSJens Axboe */ 19152870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 19162870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 19172870c43dSOctavian Purdila skb_headlen(skb), 191818aafc62SEric Dumazet offset, len, spd, 19193a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 19201d0c0b32SEric Dumazet sk, pipe)) 1921a108d5f3SDavid S. Miller return true; 19229c55e01cSJens Axboe 19239c55e01cSJens Axboe /* 19249c55e01cSJens Axboe * then map the fragments 19259c55e01cSJens Axboe */ 19269c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 19279c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 19289c55e01cSJens Axboe 1929ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 19309e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 193118aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 1932a108d5f3SDavid S. Miller return true; 19339c55e01cSJens Axboe } 19349c55e01cSJens Axboe 1935a108d5f3SDavid S. Miller return false; 19369c55e01cSJens Axboe } 19379c55e01cSJens Axboe 19389c55e01cSJens Axboe /* 19399c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 19409c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 19419c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 19429c55e01cSJens Axboe * handle that cleanly. 19439c55e01cSJens Axboe */ 19448b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 19459c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 19469c55e01cSJens Axboe unsigned int flags) 19479c55e01cSJens Axboe { 194841c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 194941c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 19509c55e01cSJens Axboe struct splice_pipe_desc spd = { 19519c55e01cSJens Axboe .pages = pages, 19529c55e01cSJens Axboe .partial = partial, 1953047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 19549c55e01cSJens Axboe .flags = flags, 195528a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 19569c55e01cSJens Axboe .spd_release = sock_spd_release, 19579c55e01cSJens Axboe }; 1958fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 19597a67e56fSJarek Poplawski struct sock *sk = skb->sk; 196035f3d14dSJens Axboe int ret = 0; 196135f3d14dSJens Axboe 19629c55e01cSJens Axboe /* 19639c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 19649c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 19659c55e01cSJens Axboe */ 196635f3d14dSJens Axboe if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 19679c55e01cSJens Axboe goto done; 19689c55e01cSJens Axboe else if (!tlen) 19699c55e01cSJens Axboe goto done; 19709c55e01cSJens Axboe 19719c55e01cSJens Axboe /* 19729c55e01cSJens Axboe * now see if we have a frag_list to map 19739c55e01cSJens Axboe */ 1974fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 1975fbb398a8SDavid S. Miller if (!tlen) 19769c55e01cSJens Axboe break; 197735f3d14dSJens Axboe if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1978fbb398a8SDavid S. Miller break; 19799c55e01cSJens Axboe } 19809c55e01cSJens Axboe 19819c55e01cSJens Axboe done: 19829c55e01cSJens Axboe if (spd.nr_pages) { 19839c55e01cSJens Axboe /* 19849c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 19859c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 19869c55e01cSJens Axboe * here as compared to sendfile(). We enter here 19879c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 19889c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 19899c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 19909c55e01cSJens Axboe * and networking will grab the socket lock. 19919c55e01cSJens Axboe */ 1992293ad604SOctavian Purdila release_sock(sk); 19939c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1994293ad604SOctavian Purdila lock_sock(sk); 19959c55e01cSJens Axboe } 19969c55e01cSJens Axboe 199735f3d14dSJens Axboe return ret; 19989c55e01cSJens Axboe } 19999c55e01cSJens Axboe 2000357b40a1SHerbert Xu /** 2001357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2002357b40a1SHerbert Xu * @skb: destination buffer 2003357b40a1SHerbert Xu * @offset: offset in destination 2004357b40a1SHerbert Xu * @from: source buffer 2005357b40a1SHerbert Xu * @len: number of bytes to copy 2006357b40a1SHerbert Xu * 2007357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2008357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2009357b40a1SHerbert Xu * traversing fragment lists and such. 2010357b40a1SHerbert Xu */ 2011357b40a1SHerbert Xu 20120c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2013357b40a1SHerbert Xu { 20141a028e50SDavid S. Miller int start = skb_headlen(skb); 2015fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2016fbb398a8SDavid S. Miller int i, copy; 2017357b40a1SHerbert Xu 2018357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2019357b40a1SHerbert Xu goto fault; 2020357b40a1SHerbert Xu 20211a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2022357b40a1SHerbert Xu if (copy > len) 2023357b40a1SHerbert Xu copy = len; 202427d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2025357b40a1SHerbert Xu if ((len -= copy) == 0) 2026357b40a1SHerbert Xu return 0; 2027357b40a1SHerbert Xu offset += copy; 2028357b40a1SHerbert Xu from += copy; 2029357b40a1SHerbert Xu } 2030357b40a1SHerbert Xu 2031357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2032357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 20331a028e50SDavid S. Miller int end; 2034357b40a1SHerbert Xu 2035547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20361a028e50SDavid S. Miller 20379e903e08SEric Dumazet end = start + skb_frag_size(frag); 2038357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2039357b40a1SHerbert Xu u8 *vaddr; 2040357b40a1SHerbert Xu 2041357b40a1SHerbert Xu if (copy > len) 2042357b40a1SHerbert Xu copy = len; 2043357b40a1SHerbert Xu 204451c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 20451a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 20461a028e50SDavid S. Miller from, copy); 204751c56b00SEric Dumazet kunmap_atomic(vaddr); 2048357b40a1SHerbert Xu 2049357b40a1SHerbert Xu if ((len -= copy) == 0) 2050357b40a1SHerbert Xu return 0; 2051357b40a1SHerbert Xu offset += copy; 2052357b40a1SHerbert Xu from += copy; 2053357b40a1SHerbert Xu } 20541a028e50SDavid S. Miller start = end; 2055357b40a1SHerbert Xu } 2056357b40a1SHerbert Xu 2057fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 20581a028e50SDavid S. Miller int end; 2059357b40a1SHerbert Xu 2060547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20611a028e50SDavid S. Miller 2062fbb398a8SDavid S. Miller end = start + frag_iter->len; 2063357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2064357b40a1SHerbert Xu if (copy > len) 2065357b40a1SHerbert Xu copy = len; 2066fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 20671a028e50SDavid S. Miller from, copy)) 2068357b40a1SHerbert Xu goto fault; 2069357b40a1SHerbert Xu if ((len -= copy) == 0) 2070357b40a1SHerbert Xu return 0; 2071357b40a1SHerbert Xu offset += copy; 2072357b40a1SHerbert Xu from += copy; 2073357b40a1SHerbert Xu } 20741a028e50SDavid S. Miller start = end; 2075357b40a1SHerbert Xu } 2076357b40a1SHerbert Xu if (!len) 2077357b40a1SHerbert Xu return 0; 2078357b40a1SHerbert Xu 2079357b40a1SHerbert Xu fault: 2080357b40a1SHerbert Xu return -EFAULT; 2081357b40a1SHerbert Xu } 2082357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2083357b40a1SHerbert Xu 20841da177e4SLinus Torvalds /* Checksum skb data. */ 20852817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 20862817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 20871da177e4SLinus Torvalds { 20881a028e50SDavid S. Miller int start = skb_headlen(skb); 20891a028e50SDavid S. Miller int i, copy = start - offset; 2090fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 20911da177e4SLinus Torvalds int pos = 0; 20921da177e4SLinus Torvalds 20931da177e4SLinus Torvalds /* Checksum header. */ 20941da177e4SLinus Torvalds if (copy > 0) { 20951da177e4SLinus Torvalds if (copy > len) 20961da177e4SLinus Torvalds copy = len; 20972817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 20981da177e4SLinus Torvalds if ((len -= copy) == 0) 20991da177e4SLinus Torvalds return csum; 21001da177e4SLinus Torvalds offset += copy; 21011da177e4SLinus Torvalds pos = copy; 21021da177e4SLinus Torvalds } 21031da177e4SLinus Torvalds 21041da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21051a028e50SDavid S. Miller int end; 210651c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 21071da177e4SLinus Torvalds 2108547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21091a028e50SDavid S. Miller 211051c56b00SEric Dumazet end = start + skb_frag_size(frag); 21111da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 211244bb9363SAl Viro __wsum csum2; 21131da177e4SLinus Torvalds u8 *vaddr; 21141da177e4SLinus Torvalds 21151da177e4SLinus Torvalds if (copy > len) 21161da177e4SLinus Torvalds copy = len; 211751c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 21182817a336SDaniel Borkmann csum2 = ops->update(vaddr + frag->page_offset + 21191a028e50SDavid S. Miller offset - start, copy, 0); 212051c56b00SEric Dumazet kunmap_atomic(vaddr); 21212817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 21221da177e4SLinus Torvalds if (!(len -= copy)) 21231da177e4SLinus Torvalds return csum; 21241da177e4SLinus Torvalds offset += copy; 21251da177e4SLinus Torvalds pos += copy; 21261da177e4SLinus Torvalds } 21271a028e50SDavid S. Miller start = end; 21281da177e4SLinus Torvalds } 21291da177e4SLinus Torvalds 2130fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 21311a028e50SDavid S. Miller int end; 21321da177e4SLinus Torvalds 2133547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21341a028e50SDavid S. Miller 2135fbb398a8SDavid S. Miller end = start + frag_iter->len; 21361da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21375f92a738SAl Viro __wsum csum2; 21381da177e4SLinus Torvalds if (copy > len) 21391da177e4SLinus Torvalds copy = len; 21402817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 21412817a336SDaniel Borkmann copy, 0, ops); 21422817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 21431da177e4SLinus Torvalds if ((len -= copy) == 0) 21441da177e4SLinus Torvalds return csum; 21451da177e4SLinus Torvalds offset += copy; 21461da177e4SLinus Torvalds pos += copy; 21471da177e4SLinus Torvalds } 21481a028e50SDavid S. Miller start = end; 21491da177e4SLinus Torvalds } 215009a62660SKris Katterjohn BUG_ON(len); 21511da177e4SLinus Torvalds 21521da177e4SLinus Torvalds return csum; 21531da177e4SLinus Torvalds } 21542817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 21552817a336SDaniel Borkmann 21562817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 21572817a336SDaniel Borkmann int len, __wsum csum) 21582817a336SDaniel Borkmann { 21592817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2160cea80ea8SDaniel Borkmann .update = csum_partial_ext, 21612817a336SDaniel Borkmann .combine = csum_block_add_ext, 21622817a336SDaniel Borkmann }; 21632817a336SDaniel Borkmann 21642817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 21652817a336SDaniel Borkmann } 2166b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 21671da177e4SLinus Torvalds 21681da177e4SLinus Torvalds /* Both of above in one bottle. */ 21691da177e4SLinus Torvalds 217081d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 217181d77662SAl Viro u8 *to, int len, __wsum csum) 21721da177e4SLinus Torvalds { 21731a028e50SDavid S. Miller int start = skb_headlen(skb); 21741a028e50SDavid S. Miller int i, copy = start - offset; 2175fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 21761da177e4SLinus Torvalds int pos = 0; 21771da177e4SLinus Torvalds 21781da177e4SLinus Torvalds /* Copy header. */ 21791da177e4SLinus Torvalds if (copy > 0) { 21801da177e4SLinus Torvalds if (copy > len) 21811da177e4SLinus Torvalds copy = len; 21821da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 21831da177e4SLinus Torvalds copy, csum); 21841da177e4SLinus Torvalds if ((len -= copy) == 0) 21851da177e4SLinus Torvalds return csum; 21861da177e4SLinus Torvalds offset += copy; 21871da177e4SLinus Torvalds to += copy; 21881da177e4SLinus Torvalds pos = copy; 21891da177e4SLinus Torvalds } 21901da177e4SLinus Torvalds 21911da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21921a028e50SDavid S. Miller int end; 21931da177e4SLinus Torvalds 2194547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21951a028e50SDavid S. Miller 21969e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 21971da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21985084205fSAl Viro __wsum csum2; 21991da177e4SLinus Torvalds u8 *vaddr; 22001da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 22011da177e4SLinus Torvalds 22021da177e4SLinus Torvalds if (copy > len) 22031da177e4SLinus Torvalds copy = len; 220451c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 22051da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 22061a028e50SDavid S. Miller frag->page_offset + 22071a028e50SDavid S. Miller offset - start, to, 22081a028e50SDavid S. Miller copy, 0); 220951c56b00SEric Dumazet kunmap_atomic(vaddr); 22101da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 22111da177e4SLinus Torvalds if (!(len -= copy)) 22121da177e4SLinus Torvalds return csum; 22131da177e4SLinus Torvalds offset += copy; 22141da177e4SLinus Torvalds to += copy; 22151da177e4SLinus Torvalds pos += copy; 22161da177e4SLinus Torvalds } 22171a028e50SDavid S. Miller start = end; 22181da177e4SLinus Torvalds } 22191da177e4SLinus Torvalds 2220fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 222181d77662SAl Viro __wsum csum2; 22221a028e50SDavid S. Miller int end; 22231da177e4SLinus Torvalds 2224547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22251a028e50SDavid S. Miller 2226fbb398a8SDavid S. Miller end = start + frag_iter->len; 22271da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 22281da177e4SLinus Torvalds if (copy > len) 22291da177e4SLinus Torvalds copy = len; 2230fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 22311a028e50SDavid S. Miller offset - start, 22321da177e4SLinus Torvalds to, copy, 0); 22331da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 22341da177e4SLinus Torvalds if ((len -= copy) == 0) 22351da177e4SLinus Torvalds return csum; 22361da177e4SLinus Torvalds offset += copy; 22371da177e4SLinus Torvalds to += copy; 22381da177e4SLinus Torvalds pos += copy; 22391da177e4SLinus Torvalds } 22401a028e50SDavid S. Miller start = end; 22411da177e4SLinus Torvalds } 224209a62660SKris Katterjohn BUG_ON(len); 22431da177e4SLinus Torvalds return csum; 22441da177e4SLinus Torvalds } 2245b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 22461da177e4SLinus Torvalds 2247af2806f8SThomas Graf /** 2248af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2249af2806f8SThomas Graf * @from: source buffer 2250af2806f8SThomas Graf * 2251af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2252af2806f8SThomas Graf * into skb_zerocopy(). 2253af2806f8SThomas Graf */ 2254af2806f8SThomas Graf unsigned int 2255af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2256af2806f8SThomas Graf { 2257af2806f8SThomas Graf unsigned int hlen = 0; 2258af2806f8SThomas Graf 2259af2806f8SThomas Graf if (!from->head_frag || 2260af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2261af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2262af2806f8SThomas Graf hlen = skb_headlen(from); 2263af2806f8SThomas Graf 2264af2806f8SThomas Graf if (skb_has_frag_list(from)) 2265af2806f8SThomas Graf hlen = from->len; 2266af2806f8SThomas Graf 2267af2806f8SThomas Graf return hlen; 2268af2806f8SThomas Graf } 2269af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2270af2806f8SThomas Graf 2271af2806f8SThomas Graf /** 2272af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2273af2806f8SThomas Graf * @to: destination buffer 22747fceb4deSMasanari Iida * @from: source buffer 2275af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2276af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2277af2806f8SThomas Graf * 2278af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2279af2806f8SThomas Graf * to the frags in the source buffer. 2280af2806f8SThomas Graf * 2281af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2282af2806f8SThomas Graf * headroom in the `to` buffer. 228336d5fe6aSZoltan Kiss * 228436d5fe6aSZoltan Kiss * Return value: 228536d5fe6aSZoltan Kiss * 0: everything is OK 228636d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 228736d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2288af2806f8SThomas Graf */ 228936d5fe6aSZoltan Kiss int 229036d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2291af2806f8SThomas Graf { 2292af2806f8SThomas Graf int i, j = 0; 2293af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 229436d5fe6aSZoltan Kiss int ret; 2295af2806f8SThomas Graf struct page *page; 2296af2806f8SThomas Graf unsigned int offset; 2297af2806f8SThomas Graf 2298af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2299af2806f8SThomas Graf 2300af2806f8SThomas Graf /* dont bother with small payloads */ 230136d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 230236d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2303af2806f8SThomas Graf 2304af2806f8SThomas Graf if (hlen) { 230536d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 230636d5fe6aSZoltan Kiss if (unlikely(ret)) 230736d5fe6aSZoltan Kiss return ret; 2308af2806f8SThomas Graf len -= hlen; 2309af2806f8SThomas Graf } else { 2310af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2311af2806f8SThomas Graf if (plen) { 2312af2806f8SThomas Graf page = virt_to_head_page(from->head); 2313af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2314af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2315af2806f8SThomas Graf get_page(page); 2316af2806f8SThomas Graf j = 1; 2317af2806f8SThomas Graf len -= plen; 2318af2806f8SThomas Graf } 2319af2806f8SThomas Graf } 2320af2806f8SThomas Graf 2321af2806f8SThomas Graf to->truesize += len + plen; 2322af2806f8SThomas Graf to->len += len + plen; 2323af2806f8SThomas Graf to->data_len += len + plen; 2324af2806f8SThomas Graf 232536d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 232636d5fe6aSZoltan Kiss skb_tx_error(from); 232736d5fe6aSZoltan Kiss return -ENOMEM; 232836d5fe6aSZoltan Kiss } 232936d5fe6aSZoltan Kiss 2330af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2331af2806f8SThomas Graf if (!len) 2332af2806f8SThomas Graf break; 2333af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2334af2806f8SThomas Graf skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2335af2806f8SThomas Graf len -= skb_shinfo(to)->frags[j].size; 2336af2806f8SThomas Graf skb_frag_ref(to, j); 2337af2806f8SThomas Graf j++; 2338af2806f8SThomas Graf } 2339af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 234036d5fe6aSZoltan Kiss 234136d5fe6aSZoltan Kiss return 0; 2342af2806f8SThomas Graf } 2343af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2344af2806f8SThomas Graf 23451da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 23461da177e4SLinus Torvalds { 2347d3bc23e7SAl Viro __wsum csum; 23481da177e4SLinus Torvalds long csstart; 23491da177e4SLinus Torvalds 235084fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 235155508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 23521da177e4SLinus Torvalds else 23531da177e4SLinus Torvalds csstart = skb_headlen(skb); 23541da177e4SLinus Torvalds 235509a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 23561da177e4SLinus Torvalds 2357d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 23581da177e4SLinus Torvalds 23591da177e4SLinus Torvalds csum = 0; 23601da177e4SLinus Torvalds if (csstart != skb->len) 23611da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 23621da177e4SLinus Torvalds skb->len - csstart, 0); 23631da177e4SLinus Torvalds 236484fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2365ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 23661da177e4SLinus Torvalds 2367d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 23681da177e4SLinus Torvalds } 23691da177e4SLinus Torvalds } 2370b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 23711da177e4SLinus Torvalds 23721da177e4SLinus Torvalds /** 23731da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 23741da177e4SLinus Torvalds * @list: list to dequeue from 23751da177e4SLinus Torvalds * 23761da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 23771da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 23781da177e4SLinus Torvalds * returned or %NULL if the list is empty. 23791da177e4SLinus Torvalds */ 23801da177e4SLinus Torvalds 23811da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 23821da177e4SLinus Torvalds { 23831da177e4SLinus Torvalds unsigned long flags; 23841da177e4SLinus Torvalds struct sk_buff *result; 23851da177e4SLinus Torvalds 23861da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23871da177e4SLinus Torvalds result = __skb_dequeue(list); 23881da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23891da177e4SLinus Torvalds return result; 23901da177e4SLinus Torvalds } 2391b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 23921da177e4SLinus Torvalds 23931da177e4SLinus Torvalds /** 23941da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 23951da177e4SLinus Torvalds * @list: list to dequeue from 23961da177e4SLinus Torvalds * 23971da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 23981da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 23991da177e4SLinus Torvalds * returned or %NULL if the list is empty. 24001da177e4SLinus Torvalds */ 24011da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 24021da177e4SLinus Torvalds { 24031da177e4SLinus Torvalds unsigned long flags; 24041da177e4SLinus Torvalds struct sk_buff *result; 24051da177e4SLinus Torvalds 24061da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24071da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 24081da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24091da177e4SLinus Torvalds return result; 24101da177e4SLinus Torvalds } 2411b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 24121da177e4SLinus Torvalds 24131da177e4SLinus Torvalds /** 24141da177e4SLinus Torvalds * skb_queue_purge - empty a list 24151da177e4SLinus Torvalds * @list: list to empty 24161da177e4SLinus Torvalds * 24171da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 24181da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 24191da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 24201da177e4SLinus Torvalds */ 24211da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 24221da177e4SLinus Torvalds { 24231da177e4SLinus Torvalds struct sk_buff *skb; 24241da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 24251da177e4SLinus Torvalds kfree_skb(skb); 24261da177e4SLinus Torvalds } 2427b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 24281da177e4SLinus Torvalds 24291da177e4SLinus Torvalds /** 24301da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 24311da177e4SLinus Torvalds * @list: list to use 24321da177e4SLinus Torvalds * @newsk: buffer to queue 24331da177e4SLinus Torvalds * 24341da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 24351da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 24361da177e4SLinus Torvalds * safely. 24371da177e4SLinus Torvalds * 24381da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24391da177e4SLinus Torvalds */ 24401da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 24411da177e4SLinus Torvalds { 24421da177e4SLinus Torvalds unsigned long flags; 24431da177e4SLinus Torvalds 24441da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24451da177e4SLinus Torvalds __skb_queue_head(list, newsk); 24461da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24471da177e4SLinus Torvalds } 2448b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 24491da177e4SLinus Torvalds 24501da177e4SLinus Torvalds /** 24511da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 24521da177e4SLinus Torvalds * @list: list to use 24531da177e4SLinus Torvalds * @newsk: buffer to queue 24541da177e4SLinus Torvalds * 24551da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 24561da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 24571da177e4SLinus Torvalds * safely. 24581da177e4SLinus Torvalds * 24591da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24601da177e4SLinus Torvalds */ 24611da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 24621da177e4SLinus Torvalds { 24631da177e4SLinus Torvalds unsigned long flags; 24641da177e4SLinus Torvalds 24651da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24661da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 24671da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24681da177e4SLinus Torvalds } 2469b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 24708728b834SDavid S. Miller 24711da177e4SLinus Torvalds /** 24721da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 24731da177e4SLinus Torvalds * @skb: buffer to remove 24748728b834SDavid S. Miller * @list: list to use 24751da177e4SLinus Torvalds * 24768728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 24778728b834SDavid S. Miller * function is atomic with respect to other list locked calls 24781da177e4SLinus Torvalds * 24798728b834SDavid S. Miller * You must know what list the SKB is on. 24801da177e4SLinus Torvalds */ 24818728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 24821da177e4SLinus Torvalds { 24831da177e4SLinus Torvalds unsigned long flags; 24841da177e4SLinus Torvalds 24851da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24868728b834SDavid S. Miller __skb_unlink(skb, list); 24871da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24881da177e4SLinus Torvalds } 2489b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 24901da177e4SLinus Torvalds 24911da177e4SLinus Torvalds /** 24921da177e4SLinus Torvalds * skb_append - append a buffer 24931da177e4SLinus Torvalds * @old: buffer to insert after 24941da177e4SLinus Torvalds * @newsk: buffer to insert 24958728b834SDavid S. Miller * @list: list to use 24961da177e4SLinus Torvalds * 24971da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 24981da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 24991da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 25001da177e4SLinus Torvalds */ 25018728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 25021da177e4SLinus Torvalds { 25031da177e4SLinus Torvalds unsigned long flags; 25041da177e4SLinus Torvalds 25058728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 25067de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 25078728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 25081da177e4SLinus Torvalds } 2509b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 25101da177e4SLinus Torvalds 25111da177e4SLinus Torvalds /** 25121da177e4SLinus Torvalds * skb_insert - insert a buffer 25131da177e4SLinus Torvalds * @old: buffer to insert before 25141da177e4SLinus Torvalds * @newsk: buffer to insert 25158728b834SDavid S. Miller * @list: list to use 25161da177e4SLinus Torvalds * 25178728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 25188728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 25198728b834SDavid S. Miller * calls. 25208728b834SDavid S. Miller * 25211da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 25221da177e4SLinus Torvalds */ 25238728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 25241da177e4SLinus Torvalds { 25251da177e4SLinus Torvalds unsigned long flags; 25261da177e4SLinus Torvalds 25278728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 25288728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 25298728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 25301da177e4SLinus Torvalds } 2531b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 25321da177e4SLinus Torvalds 25331da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 25341da177e4SLinus Torvalds struct sk_buff* skb1, 25351da177e4SLinus Torvalds const u32 len, const int pos) 25361da177e4SLinus Torvalds { 25371da177e4SLinus Torvalds int i; 25381da177e4SLinus Torvalds 2539d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2540d626f62bSArnaldo Carvalho de Melo pos - len); 25411da177e4SLinus Torvalds /* And move data appendix as is. */ 25421da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 25431da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 25441da177e4SLinus Torvalds 25451da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 25461da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 25471da177e4SLinus Torvalds skb1->data_len = skb->data_len; 25481da177e4SLinus Torvalds skb1->len += skb1->data_len; 25491da177e4SLinus Torvalds skb->data_len = 0; 25501da177e4SLinus Torvalds skb->len = len; 255127a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 25521da177e4SLinus Torvalds } 25531da177e4SLinus Torvalds 25541da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 25551da177e4SLinus Torvalds struct sk_buff* skb1, 25561da177e4SLinus Torvalds const u32 len, int pos) 25571da177e4SLinus Torvalds { 25581da177e4SLinus Torvalds int i, k = 0; 25591da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 25601da177e4SLinus Torvalds 25611da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 25621da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 25631da177e4SLinus Torvalds skb->len = len; 25641da177e4SLinus Torvalds skb->data_len = len - pos; 25651da177e4SLinus Torvalds 25661da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 25679e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 25681da177e4SLinus Torvalds 25691da177e4SLinus Torvalds if (pos + size > len) { 25701da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 25711da177e4SLinus Torvalds 25721da177e4SLinus Torvalds if (pos < len) { 25731da177e4SLinus Torvalds /* Split frag. 25741da177e4SLinus Torvalds * We have two variants in this case: 25751da177e4SLinus Torvalds * 1. Move all the frag to the second 25761da177e4SLinus Torvalds * part, if it is possible. F.e. 25771da177e4SLinus Torvalds * this approach is mandatory for TUX, 25781da177e4SLinus Torvalds * where splitting is expensive. 25791da177e4SLinus Torvalds * 2. Split is accurately. We make this. 25801da177e4SLinus Torvalds */ 2581ea2ab693SIan Campbell skb_frag_ref(skb, i); 25821da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 25839e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 25849e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 25851da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 25861da177e4SLinus Torvalds } 25871da177e4SLinus Torvalds k++; 25881da177e4SLinus Torvalds } else 25891da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 25901da177e4SLinus Torvalds pos += size; 25911da177e4SLinus Torvalds } 25921da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 25931da177e4SLinus Torvalds } 25941da177e4SLinus Torvalds 25951da177e4SLinus Torvalds /** 25961da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 25971da177e4SLinus Torvalds * @skb: the buffer to split 25981da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 25991da177e4SLinus Torvalds * @len: new length for skb 26001da177e4SLinus Torvalds */ 26011da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 26021da177e4SLinus Torvalds { 26031da177e4SLinus Torvalds int pos = skb_headlen(skb); 26041da177e4SLinus Torvalds 260568534c68SAmerigo Wang skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 26061da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 26071da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 26081da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 26091da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 26101da177e4SLinus Torvalds } 2611b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 26121da177e4SLinus Torvalds 26139f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 26149f782db3SIlpo Järvinen * 26159f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 26169f782db3SIlpo Järvinen */ 2617832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2618832d11c5SIlpo Järvinen { 26190ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2620832d11c5SIlpo Järvinen } 2621832d11c5SIlpo Järvinen 2622832d11c5SIlpo Järvinen /** 2623832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2624832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2625832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2626832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2627832d11c5SIlpo Järvinen * 2628832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 262920e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 2630832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2631832d11c5SIlpo Järvinen * 2632832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2633832d11c5SIlpo Järvinen * 2634832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2635832d11c5SIlpo Järvinen * to have non-paged data as well. 2636832d11c5SIlpo Järvinen * 2637832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2638832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2639832d11c5SIlpo Järvinen */ 2640832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2641832d11c5SIlpo Järvinen { 2642832d11c5SIlpo Järvinen int from, to, merge, todo; 2643832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2644832d11c5SIlpo Järvinen 2645832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2646832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2647832d11c5SIlpo Järvinen 2648832d11c5SIlpo Järvinen todo = shiftlen; 2649832d11c5SIlpo Järvinen from = 0; 2650832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2651832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2652832d11c5SIlpo Järvinen 2653832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2654832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2655832d11c5SIlpo Järvinen */ 2656832d11c5SIlpo Järvinen if (!to || 2657ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2658ea2ab693SIan Campbell fragfrom->page_offset)) { 2659832d11c5SIlpo Järvinen merge = -1; 2660832d11c5SIlpo Järvinen } else { 2661832d11c5SIlpo Järvinen merge = to - 1; 2662832d11c5SIlpo Järvinen 26639e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2664832d11c5SIlpo Järvinen if (todo < 0) { 2665832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2666832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2667832d11c5SIlpo Järvinen return 0; 2668832d11c5SIlpo Järvinen 26699f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 26709f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2671832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2672832d11c5SIlpo Järvinen 26739e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 26749e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 2675832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2676832d11c5SIlpo Järvinen 2677832d11c5SIlpo Järvinen goto onlymerged; 2678832d11c5SIlpo Järvinen } 2679832d11c5SIlpo Järvinen 2680832d11c5SIlpo Järvinen from++; 2681832d11c5SIlpo Järvinen } 2682832d11c5SIlpo Järvinen 2683832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2684832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2685832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2686832d11c5SIlpo Järvinen return 0; 2687832d11c5SIlpo Järvinen 2688832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2689832d11c5SIlpo Järvinen return 0; 2690832d11c5SIlpo Järvinen 2691832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2692832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2693832d11c5SIlpo Järvinen return 0; 2694832d11c5SIlpo Järvinen 2695832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2696832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2697832d11c5SIlpo Järvinen 26989e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 2699832d11c5SIlpo Järvinen *fragto = *fragfrom; 27009e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2701832d11c5SIlpo Järvinen from++; 2702832d11c5SIlpo Järvinen to++; 2703832d11c5SIlpo Järvinen 2704832d11c5SIlpo Järvinen } else { 2705ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2706832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2707832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 27089e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 2709832d11c5SIlpo Järvinen 2710832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 27119e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 2712832d11c5SIlpo Järvinen todo = 0; 2713832d11c5SIlpo Järvinen 2714832d11c5SIlpo Järvinen to++; 2715832d11c5SIlpo Järvinen break; 2716832d11c5SIlpo Järvinen } 2717832d11c5SIlpo Järvinen } 2718832d11c5SIlpo Järvinen 2719832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2720832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2721832d11c5SIlpo Järvinen 2722832d11c5SIlpo Järvinen if (merge >= 0) { 2723832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2724832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2725832d11c5SIlpo Järvinen 27269e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2727ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2728832d11c5SIlpo Järvinen } 2729832d11c5SIlpo Järvinen 2730832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2731832d11c5SIlpo Järvinen to = 0; 2732832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2733832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2734832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2735832d11c5SIlpo Järvinen 2736832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2737832d11c5SIlpo Järvinen 2738832d11c5SIlpo Järvinen onlymerged: 2739832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2740832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2741832d11c5SIlpo Järvinen */ 2742832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2743832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2744832d11c5SIlpo Järvinen 2745832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2746832d11c5SIlpo Järvinen skb->len -= shiftlen; 2747832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2748832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2749832d11c5SIlpo Järvinen tgt->len += shiftlen; 2750832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2751832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2752832d11c5SIlpo Järvinen 2753832d11c5SIlpo Järvinen return shiftlen; 2754832d11c5SIlpo Järvinen } 2755832d11c5SIlpo Järvinen 2756677e90edSThomas Graf /** 2757677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2758677e90edSThomas Graf * @skb: the buffer to read 2759677e90edSThomas Graf * @from: lower offset of data to be read 2760677e90edSThomas Graf * @to: upper offset of data to be read 2761677e90edSThomas Graf * @st: state variable 2762677e90edSThomas Graf * 2763677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2764677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2765677e90edSThomas Graf */ 2766677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2767677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2768677e90edSThomas Graf { 2769677e90edSThomas Graf st->lower_offset = from; 2770677e90edSThomas Graf st->upper_offset = to; 2771677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2772677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2773677e90edSThomas Graf st->frag_data = NULL; 2774677e90edSThomas Graf } 2775b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2776677e90edSThomas Graf 2777677e90edSThomas Graf /** 2778677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2779677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2780677e90edSThomas Graf * @data: destination pointer for data to be returned 2781677e90edSThomas Graf * @st: state variable 2782677e90edSThomas Graf * 2783bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 2784677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2785bc32383cSMathias Krause * the head of the data block to @data and returns the length 2786677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2787677e90edSThomas Graf * offset has been reached. 2788677e90edSThomas Graf * 2789677e90edSThomas Graf * The caller is not required to consume all of the data 2790bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 2791677e90edSThomas Graf * of bytes already consumed and the next call to 2792677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2793677e90edSThomas Graf * 279425985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2795e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 2796677e90edSThomas Graf * reads of potentially non linear data. 2797677e90edSThomas Graf * 2798bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2799677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2800677e90edSThomas Graf * a stack for this purpose. 2801677e90edSThomas Graf */ 2802677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2803677e90edSThomas Graf struct skb_seq_state *st) 2804677e90edSThomas Graf { 2805677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2806677e90edSThomas Graf skb_frag_t *frag; 2807677e90edSThomas Graf 2808aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 2809aeb193eaSWedson Almeida Filho if (st->frag_data) { 2810aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 2811aeb193eaSWedson Almeida Filho st->frag_data = NULL; 2812aeb193eaSWedson Almeida Filho } 2813677e90edSThomas Graf return 0; 2814aeb193eaSWedson Almeida Filho } 2815677e90edSThomas Graf 2816677e90edSThomas Graf next_skb: 281795e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2818677e90edSThomas Graf 2819995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 282095e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2821677e90edSThomas Graf return block_limit - abs_offset; 2822677e90edSThomas Graf } 2823677e90edSThomas Graf 2824677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2825677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2826677e90edSThomas Graf 2827677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2828677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 28299e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 2830677e90edSThomas Graf 2831677e90edSThomas Graf if (abs_offset < block_limit) { 2832677e90edSThomas Graf if (!st->frag_data) 283351c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 2834677e90edSThomas Graf 2835677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2836677e90edSThomas Graf (abs_offset - st->stepped_offset); 2837677e90edSThomas Graf 2838677e90edSThomas Graf return block_limit - abs_offset; 2839677e90edSThomas Graf } 2840677e90edSThomas Graf 2841677e90edSThomas Graf if (st->frag_data) { 284251c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2843677e90edSThomas Graf st->frag_data = NULL; 2844677e90edSThomas Graf } 2845677e90edSThomas Graf 2846677e90edSThomas Graf st->frag_idx++; 28479e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 2848677e90edSThomas Graf } 2849677e90edSThomas Graf 28505b5a60daSOlaf Kirch if (st->frag_data) { 285151c56b00SEric Dumazet kunmap_atomic(st->frag_data); 28525b5a60daSOlaf Kirch st->frag_data = NULL; 28535b5a60daSOlaf Kirch } 28545b5a60daSOlaf Kirch 285521dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2856677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 285795e3b24cSHerbert Xu st->frag_idx = 0; 2858677e90edSThomas Graf goto next_skb; 285971b3346dSShyam Iyer } else if (st->cur_skb->next) { 286071b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 286171b3346dSShyam Iyer st->frag_idx = 0; 2862677e90edSThomas Graf goto next_skb; 2863677e90edSThomas Graf } 2864677e90edSThomas Graf 2865677e90edSThomas Graf return 0; 2866677e90edSThomas Graf } 2867b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2868677e90edSThomas Graf 2869677e90edSThomas Graf /** 2870677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2871677e90edSThomas Graf * @st: state variable 2872677e90edSThomas Graf * 2873677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2874677e90edSThomas Graf * returned 0. 2875677e90edSThomas Graf */ 2876677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2877677e90edSThomas Graf { 2878677e90edSThomas Graf if (st->frag_data) 287951c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2880677e90edSThomas Graf } 2881b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2882677e90edSThomas Graf 28833fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 28843fc7e8a6SThomas Graf 28853fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 28863fc7e8a6SThomas Graf struct ts_config *conf, 28873fc7e8a6SThomas Graf struct ts_state *state) 28883fc7e8a6SThomas Graf { 28893fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 28903fc7e8a6SThomas Graf } 28913fc7e8a6SThomas Graf 28923fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 28933fc7e8a6SThomas Graf { 28943fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 28953fc7e8a6SThomas Graf } 28963fc7e8a6SThomas Graf 28973fc7e8a6SThomas Graf /** 28983fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 28993fc7e8a6SThomas Graf * @skb: the buffer to look in 29003fc7e8a6SThomas Graf * @from: search offset 29013fc7e8a6SThomas Graf * @to: search limit 29023fc7e8a6SThomas Graf * @config: textsearch configuration 29033fc7e8a6SThomas Graf * 29043fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 29053fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 29063fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 29073fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 29083fc7e8a6SThomas Graf */ 29093fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 2910059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 29113fc7e8a6SThomas Graf { 2912059a2440SBojan Prtvar struct ts_state state; 2913f72b948dSPhil Oester unsigned int ret; 2914f72b948dSPhil Oester 29153fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 29163fc7e8a6SThomas Graf config->finish = skb_ts_finish; 29173fc7e8a6SThomas Graf 2918059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 29193fc7e8a6SThomas Graf 2920059a2440SBojan Prtvar ret = textsearch_find(config, &state); 2921f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 29223fc7e8a6SThomas Graf } 2923b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 29243fc7e8a6SThomas Graf 2925e89e9cf5SAnanda Raju /** 29262c53040fSBen Hutchings * skb_append_datato_frags - append the user data to a skb 2927e89e9cf5SAnanda Raju * @sk: sock structure 2928e793c0f7SMasanari Iida * @skb: skb structure to be appended with user data. 2929e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2930e89e9cf5SAnanda Raju * @from: pointer to user message iov 2931e89e9cf5SAnanda Raju * @length: length of the iov message 2932e89e9cf5SAnanda Raju * 2933e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2934e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2935e89e9cf5SAnanda Raju */ 2936e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2937dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2938e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2939e89e9cf5SAnanda Raju void *from, int length) 2940e89e9cf5SAnanda Raju { 2941b2111724SEric Dumazet int frg_cnt = skb_shinfo(skb)->nr_frags; 2942b2111724SEric Dumazet int copy; 2943e89e9cf5SAnanda Raju int offset = 0; 2944e89e9cf5SAnanda Raju int ret; 2945b2111724SEric Dumazet struct page_frag *pfrag = ¤t->task_frag; 2946e89e9cf5SAnanda Raju 2947e89e9cf5SAnanda Raju do { 2948e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2949e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2950b2111724SEric Dumazet return -EMSGSIZE; 2951e89e9cf5SAnanda Raju 2952b2111724SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 2953e89e9cf5SAnanda Raju return -ENOMEM; 2954e89e9cf5SAnanda Raju 2955e89e9cf5SAnanda Raju /* copy the user data to page */ 2956b2111724SEric Dumazet copy = min_t(int, length, pfrag->size - pfrag->offset); 2957e89e9cf5SAnanda Raju 2958b2111724SEric Dumazet ret = getfrag(from, page_address(pfrag->page) + pfrag->offset, 2959e89e9cf5SAnanda Raju offset, copy, 0, skb); 2960e89e9cf5SAnanda Raju if (ret < 0) 2961e89e9cf5SAnanda Raju return -EFAULT; 2962e89e9cf5SAnanda Raju 2963e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2964b2111724SEric Dumazet skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset, 2965b2111724SEric Dumazet copy); 2966b2111724SEric Dumazet frg_cnt++; 2967b2111724SEric Dumazet pfrag->offset += copy; 2968b2111724SEric Dumazet get_page(pfrag->page); 2969b2111724SEric Dumazet 2970b2111724SEric Dumazet skb->truesize += copy; 2971b2111724SEric Dumazet atomic_add(copy, &sk->sk_wmem_alloc); 2972e89e9cf5SAnanda Raju skb->len += copy; 2973e89e9cf5SAnanda Raju skb->data_len += copy; 2974e89e9cf5SAnanda Raju offset += copy; 2975e89e9cf5SAnanda Raju length -= copy; 2976e89e9cf5SAnanda Raju 2977e89e9cf5SAnanda Raju } while (length > 0); 2978e89e9cf5SAnanda Raju 2979e89e9cf5SAnanda Raju return 0; 2980e89e9cf5SAnanda Raju } 2981b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2982e89e9cf5SAnanda Raju 2983cbb042f9SHerbert Xu /** 2984cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2985cbb042f9SHerbert Xu * @skb: buffer to update 2986cbb042f9SHerbert Xu * @len: length of data pulled 2987cbb042f9SHerbert Xu * 2988cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2989fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 299084fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 299184fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 299284fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2993cbb042f9SHerbert Xu */ 2994cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2995cbb042f9SHerbert Xu { 2996cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2997cbb042f9SHerbert Xu skb->len -= len; 2998cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2999cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 3000cbb042f9SHerbert Xu return skb->data += len; 3001cbb042f9SHerbert Xu } 3002f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3003f94691acSArnaldo Carvalho de Melo 3004f4c50d99SHerbert Xu /** 3005f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3006df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3007576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3008f4c50d99SHerbert Xu * 3009f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 30104c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 30114c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3012f4c50d99SHerbert Xu */ 3013df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3014df5771ffSMichael S. Tsirkin netdev_features_t features) 3015f4c50d99SHerbert Xu { 3016f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3017f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 30181a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3019df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3020df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3021df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 30221fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3023f4c50d99SHerbert Xu unsigned int offset = doffset; 3024df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3025f4c50d99SHerbert Xu unsigned int headroom; 3026f4c50d99SHerbert Xu unsigned int len; 3027ec5f0615SPravin B Shelar __be16 proto; 3028ec5f0615SPravin B Shelar bool csum; 302904ed3e74SMichał Mirosław int sg = !!(features & NETIF_F_SG); 3030df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3031f4c50d99SHerbert Xu int err = -ENOMEM; 3032f4c50d99SHerbert Xu int i = 0; 3033f4c50d99SHerbert Xu int pos; 303453d6471cSVlad Yasevich int dummy; 3035f4c50d99SHerbert Xu 30365882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 303753d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 3038ec5f0615SPravin B Shelar if (unlikely(!proto)) 3039ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3040ec5f0615SPravin B Shelar 30417e2b10c1STom Herbert csum = !head_skb->encap_hdr_csum && 30427e2b10c1STom Herbert !!can_checksum_protocol(features, proto); 30437e2b10c1STom Herbert 3044df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3045df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3046f4c50d99SHerbert Xu 3047f4c50d99SHerbert Xu do { 3048f4c50d99SHerbert Xu struct sk_buff *nskb; 30498cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3050c8884eddSHerbert Xu int hsize; 3051f4c50d99SHerbert Xu int size; 3052f4c50d99SHerbert Xu 3053df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3054f4c50d99SHerbert Xu if (len > mss) 3055f4c50d99SHerbert Xu len = mss; 3056f4c50d99SHerbert Xu 3057df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3058f4c50d99SHerbert Xu if (hsize < 0) 3059f4c50d99SHerbert Xu hsize = 0; 3060c8884eddSHerbert Xu if (hsize > len || !sg) 3061c8884eddSHerbert Xu hsize = len; 3062f4c50d99SHerbert Xu 30631a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 30641a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 30651a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 306689319d38SHerbert Xu 30679d8506ccSHerbert Xu i = 0; 30681a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 30691a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 30701fd819ecSMichael S. Tsirkin frag_skb = list_skb; 30711a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 30729d8506ccSHerbert Xu 30739d8506ccSHerbert Xu while (pos < offset + len) { 30749d8506ccSHerbert Xu BUG_ON(i >= nfrags); 30759d8506ccSHerbert Xu 30764e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 30779d8506ccSHerbert Xu if (pos + size > offset + len) 30789d8506ccSHerbert Xu break; 30799d8506ccSHerbert Xu 30809d8506ccSHerbert Xu i++; 30819d8506ccSHerbert Xu pos += size; 30824e1beba1SMichael S. Tsirkin frag++; 30839d8506ccSHerbert Xu } 30849d8506ccSHerbert Xu 30851a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 30861a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 308789319d38SHerbert Xu 3088f4c50d99SHerbert Xu if (unlikely(!nskb)) 3089f4c50d99SHerbert Xu goto err; 3090f4c50d99SHerbert Xu 30919d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 30929d8506ccSHerbert Xu kfree_skb(nskb); 30939d8506ccSHerbert Xu goto err; 30949d8506ccSHerbert Xu } 30959d8506ccSHerbert Xu 3096ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 309789319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 309889319d38SHerbert Xu kfree_skb(nskb); 309989319d38SHerbert Xu goto err; 310089319d38SHerbert Xu } 310189319d38SHerbert Xu 3102ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 310389319d38SHerbert Xu skb_release_head_state(nskb); 310489319d38SHerbert Xu __skb_push(nskb, doffset); 310589319d38SHerbert Xu } else { 3106c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3107df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3108c93bdd0eSMel Gorman NUMA_NO_NODE); 310989319d38SHerbert Xu 311089319d38SHerbert Xu if (unlikely(!nskb)) 311189319d38SHerbert Xu goto err; 311289319d38SHerbert Xu 311389319d38SHerbert Xu skb_reserve(nskb, headroom); 311489319d38SHerbert Xu __skb_put(nskb, doffset); 311589319d38SHerbert Xu } 311689319d38SHerbert Xu 3117f4c50d99SHerbert Xu if (segs) 3118f4c50d99SHerbert Xu tail->next = nskb; 3119f4c50d99SHerbert Xu else 3120f4c50d99SHerbert Xu segs = nskb; 3121f4c50d99SHerbert Xu tail = nskb; 3122f4c50d99SHerbert Xu 3123df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3124f4c50d99SHerbert Xu 3125030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3126fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 312768c33163SPravin B Shelar 3128df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 312968c33163SPravin B Shelar nskb->data - tnl_hlen, 313068c33163SPravin B Shelar doffset + tnl_hlen); 313189319d38SHerbert Xu 31329d8506ccSHerbert Xu if (nskb->len == len + doffset) 31331cdbcb79SSimon Horman goto perform_csum_check; 313489319d38SHerbert Xu 3135e585f236STom Herbert if (!sg && !nskb->remcsum_offload) { 31366f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 3137df5771ffSMichael S. Tsirkin nskb->csum = skb_copy_and_csum_bits(head_skb, offset, 3138f4c50d99SHerbert Xu skb_put(nskb, len), 3139f4c50d99SHerbert Xu len, 0); 31407e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3141de843723STom Herbert skb_headroom(nskb) + doffset; 3142f4c50d99SHerbert Xu continue; 3143f4c50d99SHerbert Xu } 3144f4c50d99SHerbert Xu 31458cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3146f4c50d99SHerbert Xu 3147df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3148d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3149f4c50d99SHerbert Xu 3150df5771ffSMichael S. Tsirkin skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags & 3151df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3152cef401deSEric Dumazet 31539d8506ccSHerbert Xu while (pos < offset + len) { 31549d8506ccSHerbert Xu if (i >= nfrags) { 31551a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb)); 31569d8506ccSHerbert Xu 31579d8506ccSHerbert Xu i = 0; 31581a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 31591a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 31601fd819ecSMichael S. Tsirkin frag_skb = list_skb; 31619d8506ccSHerbert Xu 31629d8506ccSHerbert Xu BUG_ON(!nfrags); 31639d8506ccSHerbert Xu 31641a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 31659d8506ccSHerbert Xu } 31669d8506ccSHerbert Xu 31679d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 31689d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 31699d8506ccSHerbert Xu net_warn_ratelimited( 31709d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 31719d8506ccSHerbert Xu pos, mss); 31729d8506ccSHerbert Xu goto err; 31739d8506ccSHerbert Xu } 31749d8506ccSHerbert Xu 31751fd819ecSMichael S. Tsirkin if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC))) 31761fd819ecSMichael S. Tsirkin goto err; 31771fd819ecSMichael S. Tsirkin 31784e1beba1SMichael S. Tsirkin *nskb_frag = *frag; 31798cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 31808cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3181f4c50d99SHerbert Xu 3182f4c50d99SHerbert Xu if (pos < offset) { 31838cb19905SMichael S. Tsirkin nskb_frag->page_offset += offset - pos; 31848cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3185f4c50d99SHerbert Xu } 3186f4c50d99SHerbert Xu 318789319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3188f4c50d99SHerbert Xu 3189f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3190f4c50d99SHerbert Xu i++; 31914e1beba1SMichael S. Tsirkin frag++; 3192f4c50d99SHerbert Xu pos += size; 3193f4c50d99SHerbert Xu } else { 31948cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 319589319d38SHerbert Xu goto skip_fraglist; 3196f4c50d99SHerbert Xu } 3197f4c50d99SHerbert Xu 31988cb19905SMichael S. Tsirkin nskb_frag++; 3199f4c50d99SHerbert Xu } 3200f4c50d99SHerbert Xu 320189319d38SHerbert Xu skip_fraglist: 3202f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3203f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3204f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3205ec5f0615SPravin B Shelar 32061cdbcb79SSimon Horman perform_csum_check: 3207e585f236STom Herbert if (!csum && !nskb->remcsum_offload) { 3208ec5f0615SPravin B Shelar nskb->csum = skb_checksum(nskb, doffset, 3209ec5f0615SPravin B Shelar nskb->len - doffset, 0); 3210ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 32117e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 32127e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3213ec5f0615SPravin B Shelar } 3214df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3215f4c50d99SHerbert Xu 3216bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3217bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3218bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3219bec3cfdcSEric Dumazet */ 3220bec3cfdcSEric Dumazet segs->prev = tail; 3221432c856fSToshiaki Makita 3222432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 3223432c856fSToshiaki Makita * using skb_set_owner_w(). 3224432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 3225432c856fSToshiaki Makita */ 3226432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 3227432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 3228432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 3229432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 3230432c856fSToshiaki Makita } 3231f4c50d99SHerbert Xu return segs; 3232f4c50d99SHerbert Xu 3233f4c50d99SHerbert Xu err: 3234289dccbeSEric Dumazet kfree_skb_list(segs); 3235f4c50d99SHerbert Xu return ERR_PTR(err); 3236f4c50d99SHerbert Xu } 3237f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3238f4c50d99SHerbert Xu 323971d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 324071d93b39SHerbert Xu { 32418a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 324267147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 324367147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 32448a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 324558025e46SEric Dumazet struct sk_buff *lp, *p = *head; 3246715dc1f3SEric Dumazet unsigned int delta_truesize; 324771d93b39SHerbert Xu 32488a29111cSEric Dumazet if (unlikely(p->len + len >= 65536)) 324971d93b39SHerbert Xu return -E2BIG; 325071d93b39SHerbert Xu 325129e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 32528a29111cSEric Dumazet pinfo = skb_shinfo(lp); 32538a29111cSEric Dumazet 32548a29111cSEric Dumazet if (headlen <= offset) { 325542da6994SHerbert Xu skb_frag_t *frag; 325666e92fcfSHerbert Xu skb_frag_t *frag2; 32579aaa156cSHerbert Xu int i = skbinfo->nr_frags; 32589aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 325942da6994SHerbert Xu 326066e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 32618a29111cSEric Dumazet goto merge; 326281705ad1SHerbert Xu 32638a29111cSEric Dumazet offset -= headlen; 32649aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 32659aaa156cSHerbert Xu skbinfo->nr_frags = 0; 3266f5572068SHerbert Xu 32679aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 32689aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 326966e92fcfSHerbert Xu do { 327066e92fcfSHerbert Xu *--frag = *--frag2; 327166e92fcfSHerbert Xu } while (--i); 327266e92fcfSHerbert Xu 327366e92fcfSHerbert Xu frag->page_offset += offset; 32749e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 327566e92fcfSHerbert Xu 3276715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3277ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3278ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3279715dc1f3SEric Dumazet 3280f5572068SHerbert Xu skb->truesize -= skb->data_len; 3281f5572068SHerbert Xu skb->len -= skb->data_len; 3282f5572068SHerbert Xu skb->data_len = 0; 3283f5572068SHerbert Xu 3284715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 32855d38a079SHerbert Xu goto done; 3286d7e8883cSEric Dumazet } else if (skb->head_frag) { 3287d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3288d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3289d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3290d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3291d7e8883cSEric Dumazet unsigned int first_offset; 3292d7e8883cSEric Dumazet 3293d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 32948a29111cSEric Dumazet goto merge; 3295d7e8883cSEric Dumazet 3296d7e8883cSEric Dumazet first_offset = skb->data - 3297d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3298d7e8883cSEric Dumazet offset; 3299d7e8883cSEric Dumazet 3300d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3301d7e8883cSEric Dumazet 3302d7e8883cSEric Dumazet frag->page.p = page; 3303d7e8883cSEric Dumazet frag->page_offset = first_offset; 3304d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3305d7e8883cSEric Dumazet 3306d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3307d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3308d7e8883cSEric Dumazet 3309715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3310d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3311d7e8883cSEric Dumazet goto done; 33128a29111cSEric Dumazet } 331371d93b39SHerbert Xu 331471d93b39SHerbert Xu merge: 3315715dc1f3SEric Dumazet delta_truesize = skb->truesize; 331667147ba9SHerbert Xu if (offset > headlen) { 3317d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3318d1dc7abfSMichal Schmidt 3319d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 33209e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3321d1dc7abfSMichal Schmidt skb->data_len -= eat; 3322d1dc7abfSMichal Schmidt skb->len -= eat; 332367147ba9SHerbert Xu offset = headlen; 332456035022SHerbert Xu } 332556035022SHerbert Xu 332667147ba9SHerbert Xu __skb_pull(skb, offset); 332756035022SHerbert Xu 332829e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 33298a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 33308a29111cSEric Dumazet else 3331c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3332c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 3333f4a775d1SEric Dumazet __skb_header_release(skb); 33348a29111cSEric Dumazet lp = p; 333571d93b39SHerbert Xu 33365d38a079SHerbert Xu done: 33375d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 333837fe4732SHerbert Xu p->data_len += len; 3339715dc1f3SEric Dumazet p->truesize += delta_truesize; 334037fe4732SHerbert Xu p->len += len; 33418a29111cSEric Dumazet if (lp != p) { 33428a29111cSEric Dumazet lp->data_len += len; 33438a29111cSEric Dumazet lp->truesize += delta_truesize; 33448a29111cSEric Dumazet lp->len += len; 33458a29111cSEric Dumazet } 334671d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 334771d93b39SHerbert Xu return 0; 334871d93b39SHerbert Xu } 334971d93b39SHerbert Xu 33501da177e4SLinus Torvalds void __init skb_init(void) 33511da177e4SLinus Torvalds { 33521da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 33531da177e4SLinus Torvalds sizeof(struct sk_buff), 33541da177e4SLinus Torvalds 0, 3355e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 335620c2df83SPaul Mundt NULL); 3357d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3358d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 3359d179cd12SDavid S. Miller 0, 3360e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 336120c2df83SPaul Mundt NULL); 33621da177e4SLinus Torvalds } 33631da177e4SLinus Torvalds 3364716ea3a7SDavid Howells /** 3365716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3366716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 3367716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 3368716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 3369716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 3370716ea3a7SDavid Howells * 3371716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 3372716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 3373716ea3a7SDavid Howells */ 337451c739d1SDavid S. Miller static int 337551c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3376716ea3a7SDavid Howells { 33771a028e50SDavid S. Miller int start = skb_headlen(skb); 33781a028e50SDavid S. Miller int i, copy = start - offset; 3379fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3380716ea3a7SDavid Howells int elt = 0; 3381716ea3a7SDavid Howells 3382716ea3a7SDavid Howells if (copy > 0) { 3383716ea3a7SDavid Howells if (copy > len) 3384716ea3a7SDavid Howells copy = len; 3385642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 3386716ea3a7SDavid Howells elt++; 3387716ea3a7SDavid Howells if ((len -= copy) == 0) 3388716ea3a7SDavid Howells return elt; 3389716ea3a7SDavid Howells offset += copy; 3390716ea3a7SDavid Howells } 3391716ea3a7SDavid Howells 3392716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 33931a028e50SDavid S. Miller int end; 3394716ea3a7SDavid Howells 3395547b792cSIlpo Järvinen WARN_ON(start > offset + len); 33961a028e50SDavid S. Miller 33979e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3398716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3399716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3400716ea3a7SDavid Howells 3401716ea3a7SDavid Howells if (copy > len) 3402716ea3a7SDavid Howells copy = len; 3403ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3404642f1490SJens Axboe frag->page_offset+offset-start); 3405716ea3a7SDavid Howells elt++; 3406716ea3a7SDavid Howells if (!(len -= copy)) 3407716ea3a7SDavid Howells return elt; 3408716ea3a7SDavid Howells offset += copy; 3409716ea3a7SDavid Howells } 34101a028e50SDavid S. Miller start = end; 3411716ea3a7SDavid Howells } 3412716ea3a7SDavid Howells 3413fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 34141a028e50SDavid S. Miller int end; 3415716ea3a7SDavid Howells 3416547b792cSIlpo Järvinen WARN_ON(start > offset + len); 34171a028e50SDavid S. Miller 3418fbb398a8SDavid S. Miller end = start + frag_iter->len; 3419716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3420716ea3a7SDavid Howells if (copy > len) 3421716ea3a7SDavid Howells copy = len; 3422fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 342351c739d1SDavid S. Miller copy); 3424716ea3a7SDavid Howells if ((len -= copy) == 0) 3425716ea3a7SDavid Howells return elt; 3426716ea3a7SDavid Howells offset += copy; 3427716ea3a7SDavid Howells } 34281a028e50SDavid S. Miller start = end; 3429716ea3a7SDavid Howells } 3430716ea3a7SDavid Howells BUG_ON(len); 3431716ea3a7SDavid Howells return elt; 3432716ea3a7SDavid Howells } 3433716ea3a7SDavid Howells 343425a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 343525a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 343625a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 343725a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 343825a91d8dSFan Du * 343925a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 344025a91d8dSFan Du * 1. sg_init_table 344125a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 344225a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 344325a91d8dSFan Du * 344425a91d8dSFan Du * This is equivalent to: 344525a91d8dSFan Du * 1. sg_init_table 344625a91d8dSFan Du * 2. skb_to_sgvec(payload1) 344725a91d8dSFan Du * 3. sg_unmark_end 344825a91d8dSFan Du * 4. skb_to_sgvec(payload2) 344925a91d8dSFan Du * 345025a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 345125a91d8dSFan Du * is more preferable. 345225a91d8dSFan Du */ 345325a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 345425a91d8dSFan Du int offset, int len) 345525a91d8dSFan Du { 345625a91d8dSFan Du return __skb_to_sgvec(skb, sg, offset, len); 345725a91d8dSFan Du } 345825a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 345925a91d8dSFan Du 346051c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 346151c739d1SDavid S. Miller { 346251c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 346351c739d1SDavid S. Miller 3464c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 346551c739d1SDavid S. Miller 346651c739d1SDavid S. Miller return nsg; 346751c739d1SDavid S. Miller } 3468b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 346951c739d1SDavid S. Miller 3470716ea3a7SDavid Howells /** 3471716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 3472716ea3a7SDavid Howells * @skb: The socket buffer to check. 3473716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 3474716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 3475716ea3a7SDavid Howells * 3476716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 3477716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 3478716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 3479716ea3a7SDavid Howells * 3480716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 3481716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 3482716ea3a7SDavid Howells * set to point to the skb in which this space begins. 3483716ea3a7SDavid Howells * 3484716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3485716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3486716ea3a7SDavid Howells */ 3487716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3488716ea3a7SDavid Howells { 3489716ea3a7SDavid Howells int copyflag; 3490716ea3a7SDavid Howells int elt; 3491716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3492716ea3a7SDavid Howells 3493716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3494716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3495716ea3a7SDavid Howells * at the moment even if they are anonymous). 3496716ea3a7SDavid Howells */ 3497716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3498716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3499716ea3a7SDavid Howells return -ENOMEM; 3500716ea3a7SDavid Howells 3501716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 350221dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3503716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3504716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3505716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3506716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3507716ea3a7SDavid Howells 3508716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3509716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3510716ea3a7SDavid Howells return -ENOMEM; 3511716ea3a7SDavid Howells 3512716ea3a7SDavid Howells /* Voila! */ 3513716ea3a7SDavid Howells *trailer = skb; 3514716ea3a7SDavid Howells return 1; 3515716ea3a7SDavid Howells } 3516716ea3a7SDavid Howells 3517716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3518716ea3a7SDavid Howells 3519716ea3a7SDavid Howells elt = 1; 3520716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3521716ea3a7SDavid Howells copyflag = 0; 3522716ea3a7SDavid Howells 3523716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3524716ea3a7SDavid Howells int ntail = 0; 3525716ea3a7SDavid Howells 3526716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3527716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3528716ea3a7SDavid Howells * after it. */ 3529716ea3a7SDavid Howells 3530716ea3a7SDavid Howells if (skb_shared(skb1)) 3531716ea3a7SDavid Howells copyflag = 1; 3532716ea3a7SDavid Howells 3533716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3534716ea3a7SDavid Howells 3535716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3536716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 353721dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3538716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3539716ea3a7SDavid Howells ntail = tailbits + 128; 3540716ea3a7SDavid Howells } 3541716ea3a7SDavid Howells 3542716ea3a7SDavid Howells if (copyflag || 3543716ea3a7SDavid Howells skb_cloned(skb1) || 3544716ea3a7SDavid Howells ntail || 3545716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 354621dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3547716ea3a7SDavid Howells struct sk_buff *skb2; 3548716ea3a7SDavid Howells 3549716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3550716ea3a7SDavid Howells if (ntail == 0) 3551716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3552716ea3a7SDavid Howells else 3553716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3554716ea3a7SDavid Howells skb_headroom(skb1), 3555716ea3a7SDavid Howells ntail, 3556716ea3a7SDavid Howells GFP_ATOMIC); 3557716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3558716ea3a7SDavid Howells return -ENOMEM; 3559716ea3a7SDavid Howells 3560716ea3a7SDavid Howells if (skb1->sk) 3561716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3562716ea3a7SDavid Howells 3563716ea3a7SDavid Howells /* Looking around. Are we still alive? 3564716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3565716ea3a7SDavid Howells 3566716ea3a7SDavid Howells skb2->next = skb1->next; 3567716ea3a7SDavid Howells *skb_p = skb2; 3568716ea3a7SDavid Howells kfree_skb(skb1); 3569716ea3a7SDavid Howells skb1 = skb2; 3570716ea3a7SDavid Howells } 3571716ea3a7SDavid Howells elt++; 3572716ea3a7SDavid Howells *trailer = skb1; 3573716ea3a7SDavid Howells skb_p = &skb1->next; 3574716ea3a7SDavid Howells } 3575716ea3a7SDavid Howells 3576716ea3a7SDavid Howells return elt; 3577716ea3a7SDavid Howells } 3578b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3579716ea3a7SDavid Howells 3580b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3581b1faf566SEric Dumazet { 3582b1faf566SEric Dumazet struct sock *sk = skb->sk; 3583b1faf566SEric Dumazet 3584b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3585b1faf566SEric Dumazet } 3586b1faf566SEric Dumazet 3587b1faf566SEric Dumazet /* 3588b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3589b1faf566SEric Dumazet */ 3590b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3591b1faf566SEric Dumazet { 3592b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 359395c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 3594b1faf566SEric Dumazet return -ENOMEM; 3595b1faf566SEric Dumazet 3596b1faf566SEric Dumazet skb_orphan(skb); 3597b1faf566SEric Dumazet skb->sk = sk; 3598b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3599b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3600b1faf566SEric Dumazet 3601abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3602abb57ea4SEric Dumazet skb_dst_force(skb); 3603abb57ea4SEric Dumazet 3604b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3605b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3606676d2369SDavid S. Miller sk->sk_data_ready(sk); 3607b1faf566SEric Dumazet return 0; 3608b1faf566SEric Dumazet } 3609b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3610b1faf566SEric Dumazet 3611364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 3612364a9e93SWillem de Bruijn { 3613364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 3614364a9e93SWillem de Bruijn struct sk_buff *skb, *skb_next; 3615997d5c3fSEric Dumazet unsigned long flags; 3616364a9e93SWillem de Bruijn int err = 0; 3617364a9e93SWillem de Bruijn 3618997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 3619364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 3620364a9e93SWillem de Bruijn if (skb && (skb_next = skb_peek(q))) 3621364a9e93SWillem de Bruijn err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 3622997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 3623364a9e93SWillem de Bruijn 3624364a9e93SWillem de Bruijn sk->sk_err = err; 3625364a9e93SWillem de Bruijn if (err) 3626364a9e93SWillem de Bruijn sk->sk_error_report(sk); 3627364a9e93SWillem de Bruijn 3628364a9e93SWillem de Bruijn return skb; 3629364a9e93SWillem de Bruijn } 3630364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 3631364a9e93SWillem de Bruijn 3632cab41c47SAlexander Duyck /** 3633cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 3634cab41c47SAlexander Duyck * @skb: the skb to clone 3635cab41c47SAlexander Duyck * 3636cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 3637cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 3638cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 3639cab41c47SAlexander Duyck * 3640cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 3641cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 3642cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 3643cab41c47SAlexander Duyck * the sk_error_queue. 3644cab41c47SAlexander Duyck */ 364562bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 364662bccb8cSAlexander Duyck { 364762bccb8cSAlexander Duyck struct sock *sk = skb->sk; 364862bccb8cSAlexander Duyck struct sk_buff *clone; 364962bccb8cSAlexander Duyck 365062bccb8cSAlexander Duyck if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt)) 365162bccb8cSAlexander Duyck return NULL; 365262bccb8cSAlexander Duyck 365362bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 365462bccb8cSAlexander Duyck if (!clone) { 365562bccb8cSAlexander Duyck sock_put(sk); 365662bccb8cSAlexander Duyck return NULL; 365762bccb8cSAlexander Duyck } 365862bccb8cSAlexander Duyck 365962bccb8cSAlexander Duyck clone->sk = sk; 366062bccb8cSAlexander Duyck clone->destructor = sock_efree; 366162bccb8cSAlexander Duyck 366262bccb8cSAlexander Duyck return clone; 366362bccb8cSAlexander Duyck } 366462bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 366562bccb8cSAlexander Duyck 366637846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 366737846ef0SAlexander Duyck struct sock *sk, 366837846ef0SAlexander Duyck int tstype) 3669ac45f602SPatrick Ohly { 3670ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3671ac45f602SPatrick Ohly int err; 3672ac45f602SPatrick Ohly 3673ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3674ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3675ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3676ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 3677e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 36784ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 367909c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 36804ed2d765SWillem de Bruijn if (sk->sk_protocol == IPPROTO_TCP) 36814ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 36824ed2d765SWillem de Bruijn } 368329030374SEric Dumazet 3684ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 368529030374SEric Dumazet 3686ac45f602SPatrick Ohly if (err) 3687ac45f602SPatrick Ohly kfree_skb(skb); 3688ac45f602SPatrick Ohly } 368937846ef0SAlexander Duyck 3690b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 3691b245be1fSWillem de Bruijn { 3692b245be1fSWillem de Bruijn bool ret; 3693b245be1fSWillem de Bruijn 3694b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 3695b245be1fSWillem de Bruijn return true; 3696b245be1fSWillem de Bruijn 3697b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 3698b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 3699b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 3700b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 3701b245be1fSWillem de Bruijn return ret; 3702b245be1fSWillem de Bruijn } 3703b245be1fSWillem de Bruijn 370437846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 370537846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 370637846ef0SAlexander Duyck { 370737846ef0SAlexander Duyck struct sock *sk = skb->sk; 370837846ef0SAlexander Duyck 3709b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 3710b245be1fSWillem de Bruijn return; 3711b245be1fSWillem de Bruijn 371262bccb8cSAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 371362bccb8cSAlexander Duyck sock_hold(sk); 371437846ef0SAlexander Duyck 371537846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 371637846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND); 371737846ef0SAlexander Duyck 371837846ef0SAlexander Duyck sock_put(sk); 371937846ef0SAlexander Duyck } 372037846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 372137846ef0SAlexander Duyck 372237846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 372337846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 372437846ef0SAlexander Duyck struct sock *sk, int tstype) 372537846ef0SAlexander Duyck { 372637846ef0SAlexander Duyck struct sk_buff *skb; 37273a8dd971SWillem de Bruijn bool tsonly; 372837846ef0SAlexander Duyck 37293a8dd971SWillem de Bruijn if (!sk) 37303a8dd971SWillem de Bruijn return; 37313a8dd971SWillem de Bruijn 37323a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 37333a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 373437846ef0SAlexander Duyck return; 373537846ef0SAlexander Duyck 373649ca0d8bSWillem de Bruijn if (tsonly) 373749ca0d8bSWillem de Bruijn skb = alloc_skb(0, GFP_ATOMIC); 373837846ef0SAlexander Duyck else 373937846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 374037846ef0SAlexander Duyck if (!skb) 374137846ef0SAlexander Duyck return; 374237846ef0SAlexander Duyck 374349ca0d8bSWillem de Bruijn if (tsonly) { 374449ca0d8bSWillem de Bruijn skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags; 374549ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 374649ca0d8bSWillem de Bruijn } 374749ca0d8bSWillem de Bruijn 374849ca0d8bSWillem de Bruijn if (hwtstamps) 374949ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 375049ca0d8bSWillem de Bruijn else 375149ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 375249ca0d8bSWillem de Bruijn 375337846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, tstype); 375437846ef0SAlexander Duyck } 3755e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 3756e7fd2885SWillem de Bruijn 3757e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 3758e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 3759e7fd2885SWillem de Bruijn { 3760e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 3761e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 3762e7fd2885SWillem de Bruijn } 3763ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3764ac45f602SPatrick Ohly 37656e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 37666e3e939fSJohannes Berg { 37676e3e939fSJohannes Berg struct sock *sk = skb->sk; 37686e3e939fSJohannes Berg struct sock_exterr_skb *serr; 37696e3e939fSJohannes Berg int err; 37706e3e939fSJohannes Berg 37716e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 37726e3e939fSJohannes Berg skb->wifi_acked = acked; 37736e3e939fSJohannes Berg 37746e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 37756e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 37766e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 37776e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 37786e3e939fSJohannes Berg 3779bf7fa551SAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 3780bf7fa551SAlexander Duyck sock_hold(sk); 3781bf7fa551SAlexander Duyck 37826e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 37836e3e939fSJohannes Berg if (err) 37846e3e939fSJohannes Berg kfree_skb(skb); 3785bf7fa551SAlexander Duyck 3786bf7fa551SAlexander Duyck sock_put(sk); 37876e3e939fSJohannes Berg } 37886e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 37896e3e939fSJohannes Berg 3790f35d9d8aSRusty Russell /** 3791f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3792f35d9d8aSRusty Russell * @skb: the skb to set 3793f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3794f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3795f35d9d8aSRusty Russell * 3796f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3797f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3798f35d9d8aSRusty Russell * 3799f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3800f35d9d8aSRusty Russell * returns false you should drop the packet. 3801f35d9d8aSRusty Russell */ 3802f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3803f35d9d8aSRusty Russell { 38045ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 38055ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3806e87cc472SJoe Perches net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 38075ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3808f35d9d8aSRusty Russell return false; 3809f35d9d8aSRusty Russell } 3810f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3811f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3812f35d9d8aSRusty Russell skb->csum_offset = off; 3813e5d5decaSJason Wang skb_set_transport_header(skb, start); 3814f35d9d8aSRusty Russell return true; 3815f35d9d8aSRusty Russell } 3816b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3817f35d9d8aSRusty Russell 3818ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 3819ed1f50c3SPaul Durrant unsigned int max) 3820ed1f50c3SPaul Durrant { 3821ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 3822ed1f50c3SPaul Durrant return 0; 3823ed1f50c3SPaul Durrant 3824ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 3825ed1f50c3SPaul Durrant * won't need to do it again. 3826ed1f50c3SPaul Durrant */ 3827ed1f50c3SPaul Durrant if (max > skb->len) 3828ed1f50c3SPaul Durrant max = skb->len; 3829ed1f50c3SPaul Durrant 3830ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 3831ed1f50c3SPaul Durrant return -ENOMEM; 3832ed1f50c3SPaul Durrant 3833ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 3834ed1f50c3SPaul Durrant return -EPROTO; 3835ed1f50c3SPaul Durrant 3836ed1f50c3SPaul Durrant return 0; 3837ed1f50c3SPaul Durrant } 3838ed1f50c3SPaul Durrant 3839f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 3840f9708b43SJan Beulich 3841f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 3842f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 3843f9708b43SJan Beulich unsigned int off) 3844f9708b43SJan Beulich { 3845f9708b43SJan Beulich switch (proto) { 3846f9708b43SJan Beulich int err; 3847f9708b43SJan Beulich 3848f9708b43SJan Beulich case IPPROTO_TCP: 3849f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 3850f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 3851f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3852f9708b43SJan Beulich offsetof(struct tcphdr, 3853f9708b43SJan Beulich check))) 3854f9708b43SJan Beulich err = -EPROTO; 3855f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 3856f9708b43SJan Beulich 3857f9708b43SJan Beulich case IPPROTO_UDP: 3858f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 3859f9708b43SJan Beulich off + sizeof(struct udphdr)); 3860f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3861f9708b43SJan Beulich offsetof(struct udphdr, 3862f9708b43SJan Beulich check))) 3863f9708b43SJan Beulich err = -EPROTO; 3864f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 3865f9708b43SJan Beulich } 3866f9708b43SJan Beulich 3867f9708b43SJan Beulich return ERR_PTR(-EPROTO); 3868f9708b43SJan Beulich } 3869f9708b43SJan Beulich 3870ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 3871ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 3872ed1f50c3SPaul Durrant */ 3873ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 3874ed1f50c3SPaul Durrant 3875f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 3876ed1f50c3SPaul Durrant { 3877ed1f50c3SPaul Durrant unsigned int off; 3878ed1f50c3SPaul Durrant bool fragment; 3879f9708b43SJan Beulich __sum16 *csum; 3880ed1f50c3SPaul Durrant int err; 3881ed1f50c3SPaul Durrant 3882ed1f50c3SPaul Durrant fragment = false; 3883ed1f50c3SPaul Durrant 3884ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3885ed1f50c3SPaul Durrant sizeof(struct iphdr), 3886ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 3887ed1f50c3SPaul Durrant if (err < 0) 3888ed1f50c3SPaul Durrant goto out; 3889ed1f50c3SPaul Durrant 3890ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 3891ed1f50c3SPaul Durrant fragment = true; 3892ed1f50c3SPaul Durrant 3893ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 3894ed1f50c3SPaul Durrant 3895ed1f50c3SPaul Durrant err = -EPROTO; 3896ed1f50c3SPaul Durrant 3897ed1f50c3SPaul Durrant if (fragment) 3898ed1f50c3SPaul Durrant goto out; 3899ed1f50c3SPaul Durrant 3900f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 3901f9708b43SJan Beulich if (IS_ERR(csum)) 3902f9708b43SJan Beulich return PTR_ERR(csum); 3903ed1f50c3SPaul Durrant 3904ed1f50c3SPaul Durrant if (recalculate) 3905f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 3906ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 3907ed1f50c3SPaul Durrant skb->len - off, 3908f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 3909ed1f50c3SPaul Durrant err = 0; 3910ed1f50c3SPaul Durrant 3911ed1f50c3SPaul Durrant out: 3912ed1f50c3SPaul Durrant return err; 3913ed1f50c3SPaul Durrant } 3914ed1f50c3SPaul Durrant 3915ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 3916ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 3917ed1f50c3SPaul Durrant */ 3918ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 3919ed1f50c3SPaul Durrant 3920ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 3921ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 3922ed1f50c3SPaul Durrant 3923ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 3924ed1f50c3SPaul Durrant { 3925ed1f50c3SPaul Durrant int err; 3926ed1f50c3SPaul Durrant u8 nexthdr; 3927ed1f50c3SPaul Durrant unsigned int off; 3928ed1f50c3SPaul Durrant unsigned int len; 3929ed1f50c3SPaul Durrant bool fragment; 3930ed1f50c3SPaul Durrant bool done; 3931f9708b43SJan Beulich __sum16 *csum; 3932ed1f50c3SPaul Durrant 3933ed1f50c3SPaul Durrant fragment = false; 3934ed1f50c3SPaul Durrant done = false; 3935ed1f50c3SPaul Durrant 3936ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 3937ed1f50c3SPaul Durrant 3938ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 3939ed1f50c3SPaul Durrant if (err < 0) 3940ed1f50c3SPaul Durrant goto out; 3941ed1f50c3SPaul Durrant 3942ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 3943ed1f50c3SPaul Durrant 3944ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 3945ed1f50c3SPaul Durrant while (off <= len && !done) { 3946ed1f50c3SPaul Durrant switch (nexthdr) { 3947ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 3948ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 3949ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 3950ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 3951ed1f50c3SPaul Durrant 3952ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3953ed1f50c3SPaul Durrant off + 3954ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 3955ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3956ed1f50c3SPaul Durrant if (err < 0) 3957ed1f50c3SPaul Durrant goto out; 3958ed1f50c3SPaul Durrant 3959ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 3960ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3961ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 3962ed1f50c3SPaul Durrant break; 3963ed1f50c3SPaul Durrant } 3964ed1f50c3SPaul Durrant case IPPROTO_AH: { 3965ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 3966ed1f50c3SPaul Durrant 3967ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3968ed1f50c3SPaul Durrant off + 3969ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 3970ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3971ed1f50c3SPaul Durrant if (err < 0) 3972ed1f50c3SPaul Durrant goto out; 3973ed1f50c3SPaul Durrant 3974ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 3975ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3976ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 3977ed1f50c3SPaul Durrant break; 3978ed1f50c3SPaul Durrant } 3979ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 3980ed1f50c3SPaul Durrant struct frag_hdr *hp; 3981ed1f50c3SPaul Durrant 3982ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3983ed1f50c3SPaul Durrant off + 3984ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 3985ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3986ed1f50c3SPaul Durrant if (err < 0) 3987ed1f50c3SPaul Durrant goto out; 3988ed1f50c3SPaul Durrant 3989ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 3990ed1f50c3SPaul Durrant 3991ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 3992ed1f50c3SPaul Durrant fragment = true; 3993ed1f50c3SPaul Durrant 3994ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3995ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 3996ed1f50c3SPaul Durrant break; 3997ed1f50c3SPaul Durrant } 3998ed1f50c3SPaul Durrant default: 3999ed1f50c3SPaul Durrant done = true; 4000ed1f50c3SPaul Durrant break; 4001ed1f50c3SPaul Durrant } 4002ed1f50c3SPaul Durrant } 4003ed1f50c3SPaul Durrant 4004ed1f50c3SPaul Durrant err = -EPROTO; 4005ed1f50c3SPaul Durrant 4006ed1f50c3SPaul Durrant if (!done || fragment) 4007ed1f50c3SPaul Durrant goto out; 4008ed1f50c3SPaul Durrant 4009f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 4010f9708b43SJan Beulich if (IS_ERR(csum)) 4011f9708b43SJan Beulich return PTR_ERR(csum); 4012ed1f50c3SPaul Durrant 4013ed1f50c3SPaul Durrant if (recalculate) 4014f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 4015ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 4016f9708b43SJan Beulich skb->len - off, nexthdr, 0); 4017ed1f50c3SPaul Durrant err = 0; 4018ed1f50c3SPaul Durrant 4019ed1f50c3SPaul Durrant out: 4020ed1f50c3SPaul Durrant return err; 4021ed1f50c3SPaul Durrant } 4022ed1f50c3SPaul Durrant 4023ed1f50c3SPaul Durrant /** 4024ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 4025ed1f50c3SPaul Durrant * @skb: the skb to set up 4026ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 4027ed1f50c3SPaul Durrant */ 4028ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 4029ed1f50c3SPaul Durrant { 4030ed1f50c3SPaul Durrant int err; 4031ed1f50c3SPaul Durrant 4032ed1f50c3SPaul Durrant switch (skb->protocol) { 4033ed1f50c3SPaul Durrant case htons(ETH_P_IP): 4034f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 4035ed1f50c3SPaul Durrant break; 4036ed1f50c3SPaul Durrant 4037ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 4038ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 4039ed1f50c3SPaul Durrant break; 4040ed1f50c3SPaul Durrant 4041ed1f50c3SPaul Durrant default: 4042ed1f50c3SPaul Durrant err = -EPROTO; 4043ed1f50c3SPaul Durrant break; 4044ed1f50c3SPaul Durrant } 4045ed1f50c3SPaul Durrant 4046ed1f50c3SPaul Durrant return err; 4047ed1f50c3SPaul Durrant } 4048ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 4049ed1f50c3SPaul Durrant 40509afd85c9SLinus Lüssing /** 40519afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 40529afd85c9SLinus Lüssing * @skb: the skb to check 40539afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 40549afd85c9SLinus Lüssing * 40559afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 40569afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 40579afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 40589afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 40599afd85c9SLinus Lüssing * 40609afd85c9SLinus Lüssing * Caller needs to set the skb transport header and release the returned skb. 40619afd85c9SLinus Lüssing * Provided skb is consumed. 40629afd85c9SLinus Lüssing */ 40639afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 40649afd85c9SLinus Lüssing unsigned int transport_len) 40659afd85c9SLinus Lüssing { 40669afd85c9SLinus Lüssing struct sk_buff *skb_chk; 40679afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 40689afd85c9SLinus Lüssing int ret; 40699afd85c9SLinus Lüssing 40709afd85c9SLinus Lüssing if (skb->len < len) { 40719afd85c9SLinus Lüssing kfree_skb(skb); 40729afd85c9SLinus Lüssing return NULL; 40739afd85c9SLinus Lüssing } else if (skb->len == len) { 40749afd85c9SLinus Lüssing return skb; 40759afd85c9SLinus Lüssing } 40769afd85c9SLinus Lüssing 40779afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 40789afd85c9SLinus Lüssing kfree_skb(skb); 40799afd85c9SLinus Lüssing 40809afd85c9SLinus Lüssing if (!skb_chk) 40819afd85c9SLinus Lüssing return NULL; 40829afd85c9SLinus Lüssing 40839afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 40849afd85c9SLinus Lüssing if (ret) { 40859afd85c9SLinus Lüssing kfree_skb(skb_chk); 40869afd85c9SLinus Lüssing return NULL; 40879afd85c9SLinus Lüssing } 40889afd85c9SLinus Lüssing 40899afd85c9SLinus Lüssing return skb_chk; 40909afd85c9SLinus Lüssing } 40919afd85c9SLinus Lüssing 40929afd85c9SLinus Lüssing /** 40939afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 40949afd85c9SLinus Lüssing * @skb: the skb to check 40959afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 40969afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 40979afd85c9SLinus Lüssing * 40989afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 40999afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 41009afd85c9SLinus Lüssing * 41019afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 41029afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 41039afd85c9SLinus Lüssing * 41049afd85c9SLinus Lüssing * Caller needs to set the skb transport header and release the returned skb. 41059afd85c9SLinus Lüssing * Provided skb is consumed. 41069afd85c9SLinus Lüssing */ 41079afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 41089afd85c9SLinus Lüssing unsigned int transport_len, 41099afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 41109afd85c9SLinus Lüssing { 41119afd85c9SLinus Lüssing struct sk_buff *skb_chk; 41129afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 4113fcba67c9SLinus Lüssing __sum16 ret; 41149afd85c9SLinus Lüssing 41159afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 41169afd85c9SLinus Lüssing if (!skb_chk) 41179afd85c9SLinus Lüssing return NULL; 41189afd85c9SLinus Lüssing 41199afd85c9SLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) { 41209afd85c9SLinus Lüssing kfree_skb(skb_chk); 41219afd85c9SLinus Lüssing return NULL; 41229afd85c9SLinus Lüssing } 41239afd85c9SLinus Lüssing 41249afd85c9SLinus Lüssing __skb_pull(skb_chk, offset); 41259afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 41269afd85c9SLinus Lüssing __skb_push(skb_chk, offset); 41279afd85c9SLinus Lüssing 41289afd85c9SLinus Lüssing if (ret) { 41299afd85c9SLinus Lüssing kfree_skb(skb_chk); 41309afd85c9SLinus Lüssing return NULL; 41319afd85c9SLinus Lüssing } 41329afd85c9SLinus Lüssing 41339afd85c9SLinus Lüssing return skb_chk; 41349afd85c9SLinus Lüssing } 41359afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 41369afd85c9SLinus Lüssing 41374497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 41384497b076SBen Hutchings { 4139e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 4140e87cc472SJoe Perches skb->dev->name); 41414497b076SBen Hutchings } 41424497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 4143bad43ca8SEric Dumazet 4144bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 4145bad43ca8SEric Dumazet { 41463d861f66SEric Dumazet if (head_stolen) { 41473d861f66SEric Dumazet skb_release_head_state(skb); 4148bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 41493d861f66SEric Dumazet } else { 4150bad43ca8SEric Dumazet __kfree_skb(skb); 4151bad43ca8SEric Dumazet } 41523d861f66SEric Dumazet } 4153bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 4154bad43ca8SEric Dumazet 4155bad43ca8SEric Dumazet /** 4156bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 4157bad43ca8SEric Dumazet * @to: prior buffer 4158bad43ca8SEric Dumazet * @from: buffer to add 4159bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 4160c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 4161bad43ca8SEric Dumazet */ 4162bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 4163bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 4164bad43ca8SEric Dumazet { 4165bad43ca8SEric Dumazet int i, delta, len = from->len; 4166bad43ca8SEric Dumazet 4167bad43ca8SEric Dumazet *fragstolen = false; 4168bad43ca8SEric Dumazet 4169bad43ca8SEric Dumazet if (skb_cloned(to)) 4170bad43ca8SEric Dumazet return false; 4171bad43ca8SEric Dumazet 4172bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 4173e93a0435SEric Dumazet if (len) 4174bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 4175bad43ca8SEric Dumazet *delta_truesize = 0; 4176bad43ca8SEric Dumazet return true; 4177bad43ca8SEric Dumazet } 4178bad43ca8SEric Dumazet 4179bad43ca8SEric Dumazet if (skb_has_frag_list(to) || skb_has_frag_list(from)) 4180bad43ca8SEric Dumazet return false; 4181bad43ca8SEric Dumazet 4182bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 4183bad43ca8SEric Dumazet struct page *page; 4184bad43ca8SEric Dumazet unsigned int offset; 4185bad43ca8SEric Dumazet 4186bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4187bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 4188bad43ca8SEric Dumazet return false; 4189bad43ca8SEric Dumazet 4190bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 4191bad43ca8SEric Dumazet return false; 4192bad43ca8SEric Dumazet 4193bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4194bad43ca8SEric Dumazet 4195bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 4196bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 4197bad43ca8SEric Dumazet 4198bad43ca8SEric Dumazet skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 4199bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 4200bad43ca8SEric Dumazet *fragstolen = true; 4201bad43ca8SEric Dumazet } else { 4202bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4203bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 4204bad43ca8SEric Dumazet return false; 4205bad43ca8SEric Dumazet 4206f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 4207bad43ca8SEric Dumazet } 4208bad43ca8SEric Dumazet 4209bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 4210bad43ca8SEric Dumazet 4211bad43ca8SEric Dumazet memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 4212bad43ca8SEric Dumazet skb_shinfo(from)->frags, 4213bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 4214bad43ca8SEric Dumazet skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 4215bad43ca8SEric Dumazet 4216bad43ca8SEric Dumazet if (!skb_cloned(from)) 4217bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags = 0; 4218bad43ca8SEric Dumazet 42198ea853fdSLi RongQing /* if the skb is not cloned this does nothing 42208ea853fdSLi RongQing * since we set nr_frags to 0. 42218ea853fdSLi RongQing */ 4222bad43ca8SEric Dumazet for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 4223bad43ca8SEric Dumazet skb_frag_ref(from, i); 4224bad43ca8SEric Dumazet 4225bad43ca8SEric Dumazet to->truesize += delta; 4226bad43ca8SEric Dumazet to->len += len; 4227bad43ca8SEric Dumazet to->data_len += len; 4228bad43ca8SEric Dumazet 4229bad43ca8SEric Dumazet *delta_truesize = delta; 4230bad43ca8SEric Dumazet return true; 4231bad43ca8SEric Dumazet } 4232bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 4233621e84d6SNicolas Dichtel 4234621e84d6SNicolas Dichtel /** 42358b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 4236621e84d6SNicolas Dichtel * 4237621e84d6SNicolas Dichtel * @skb: buffer to clean 42388b27f277SNicolas Dichtel * @xnet: packet is crossing netns 4239621e84d6SNicolas Dichtel * 42408b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 42418b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 42428b27f277SNicolas Dichtel * operations. 42438b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 42448b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 42458b27f277SNicolas Dichtel * skb that could impact namespace isolation. 4246621e84d6SNicolas Dichtel */ 42478b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4248621e84d6SNicolas Dichtel { 4249621e84d6SNicolas Dichtel skb->tstamp.tv64 = 0; 4250621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 4251621e84d6SNicolas Dichtel skb->skb_iif = 0; 425260ff7467SWANG Cong skb->ignore_df = 0; 4253621e84d6SNicolas Dichtel skb_dst_drop(skb); 4254c29390c6SEric Dumazet skb_sender_cpu_clear(skb); 4255621e84d6SNicolas Dichtel secpath_reset(skb); 4256621e84d6SNicolas Dichtel nf_reset(skb); 4257621e84d6SNicolas Dichtel nf_reset_trace(skb); 4258213dd74aSHerbert Xu 4259213dd74aSHerbert Xu if (!xnet) 4260213dd74aSHerbert Xu return; 4261213dd74aSHerbert Xu 4262213dd74aSHerbert Xu skb_orphan(skb); 4263213dd74aSHerbert Xu skb->mark = 0; 4264621e84d6SNicolas Dichtel } 4265621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 4266de960aa9SFlorian Westphal 4267de960aa9SFlorian Westphal /** 4268de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4269de960aa9SFlorian Westphal * 4270de960aa9SFlorian Westphal * @skb: GSO skb 4271de960aa9SFlorian Westphal * 4272de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 4273de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 4274de960aa9SFlorian Westphal * 4275de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4276de960aa9SFlorian Westphal */ 4277de960aa9SFlorian Westphal unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4278de960aa9SFlorian Westphal { 4279de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 4280f993bc25SFlorian Westphal unsigned int thlen = 0; 4281f993bc25SFlorian Westphal 4282f993bc25SFlorian Westphal if (skb->encapsulation) { 4283f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 4284f993bc25SFlorian Westphal skb_transport_header(skb); 4285de960aa9SFlorian Westphal 4286de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 4287f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 4288f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 4289f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 4290f993bc25SFlorian Westphal } 42916d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 42926d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 42936d39d589SFlorian Westphal * accounted for. 42946d39d589SFlorian Westphal */ 4295f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 4296de960aa9SFlorian Westphal } 4297de960aa9SFlorian Westphal EXPORT_SYMBOL_GPL(skb_gso_transport_seglen); 42980d5501c1SVlad Yasevich 42990d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 43000d5501c1SVlad Yasevich { 43010d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 43020d5501c1SVlad Yasevich kfree_skb(skb); 43030d5501c1SVlad Yasevich return NULL; 43040d5501c1SVlad Yasevich } 43050d5501c1SVlad Yasevich 43060d5501c1SVlad Yasevich memmove(skb->data - ETH_HLEN, skb->data - VLAN_ETH_HLEN, 2 * ETH_ALEN); 43070d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 43080d5501c1SVlad Yasevich return skb; 43090d5501c1SVlad Yasevich } 43100d5501c1SVlad Yasevich 43110d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 43120d5501c1SVlad Yasevich { 43130d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 43140d5501c1SVlad Yasevich u16 vlan_tci; 43150d5501c1SVlad Yasevich 4316df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 43170d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 43180d5501c1SVlad Yasevich return skb; 43190d5501c1SVlad Yasevich } 43200d5501c1SVlad Yasevich 43210d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 43220d5501c1SVlad Yasevich if (unlikely(!skb)) 43230d5501c1SVlad Yasevich goto err_free; 43240d5501c1SVlad Yasevich 43250d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 43260d5501c1SVlad Yasevich goto err_free; 43270d5501c1SVlad Yasevich 43280d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 43290d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 43300d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 43310d5501c1SVlad Yasevich 43320d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 43330d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 43340d5501c1SVlad Yasevich 43350d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 43360d5501c1SVlad Yasevich if (unlikely(!skb)) 43370d5501c1SVlad Yasevich goto err_free; 43380d5501c1SVlad Yasevich 43390d5501c1SVlad Yasevich skb_reset_network_header(skb); 43400d5501c1SVlad Yasevich skb_reset_transport_header(skb); 43410d5501c1SVlad Yasevich skb_reset_mac_len(skb); 43420d5501c1SVlad Yasevich 43430d5501c1SVlad Yasevich return skb; 43440d5501c1SVlad Yasevich 43450d5501c1SVlad Yasevich err_free: 43460d5501c1SVlad Yasevich kfree_skb(skb); 43470d5501c1SVlad Yasevich return NULL; 43480d5501c1SVlad Yasevich } 43490d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 43502e4e4410SEric Dumazet 4351e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 4352e2195121SJiri Pirko { 4353e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 4354e2195121SJiri Pirko return -ENOMEM; 4355e2195121SJiri Pirko 4356e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 4357e2195121SJiri Pirko return 0; 4358e2195121SJiri Pirko 4359e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 4360e2195121SJiri Pirko } 4361e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 4362e2195121SJiri Pirko 436393515d53SJiri Pirko /* remove VLAN header from packet and update csum accordingly. */ 436493515d53SJiri Pirko static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 436593515d53SJiri Pirko { 436693515d53SJiri Pirko struct vlan_hdr *vhdr; 436793515d53SJiri Pirko unsigned int offset = skb->data - skb_mac_header(skb); 436893515d53SJiri Pirko int err; 436993515d53SJiri Pirko 437093515d53SJiri Pirko __skb_push(skb, offset); 437193515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 437293515d53SJiri Pirko if (unlikely(err)) 437393515d53SJiri Pirko goto pull; 437493515d53SJiri Pirko 437593515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 437693515d53SJiri Pirko 437793515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 437893515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 437993515d53SJiri Pirko 438093515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 438193515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 438293515d53SJiri Pirko 438393515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 438493515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 438593515d53SJiri Pirko 438693515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 438793515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 438893515d53SJiri Pirko 438993515d53SJiri Pirko skb_reset_mac_len(skb); 439093515d53SJiri Pirko pull: 439193515d53SJiri Pirko __skb_pull(skb, offset); 439293515d53SJiri Pirko 439393515d53SJiri Pirko return err; 439493515d53SJiri Pirko } 439593515d53SJiri Pirko 439693515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 439793515d53SJiri Pirko { 439893515d53SJiri Pirko u16 vlan_tci; 439993515d53SJiri Pirko __be16 vlan_proto; 440093515d53SJiri Pirko int err; 440193515d53SJiri Pirko 4402df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 440393515d53SJiri Pirko skb->vlan_tci = 0; 440493515d53SJiri Pirko } else { 440593515d53SJiri Pirko if (unlikely((skb->protocol != htons(ETH_P_8021Q) && 440693515d53SJiri Pirko skb->protocol != htons(ETH_P_8021AD)) || 440793515d53SJiri Pirko skb->len < VLAN_ETH_HLEN)) 440893515d53SJiri Pirko return 0; 440993515d53SJiri Pirko 441093515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 441193515d53SJiri Pirko if (err) 441293515d53SJiri Pirko return err; 441393515d53SJiri Pirko } 441493515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 441593515d53SJiri Pirko if (likely((skb->protocol != htons(ETH_P_8021Q) && 441693515d53SJiri Pirko skb->protocol != htons(ETH_P_8021AD)) || 441793515d53SJiri Pirko skb->len < VLAN_ETH_HLEN)) 441893515d53SJiri Pirko return 0; 441993515d53SJiri Pirko 442093515d53SJiri Pirko vlan_proto = skb->protocol; 442193515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 442293515d53SJiri Pirko if (unlikely(err)) 442393515d53SJiri Pirko return err; 442493515d53SJiri Pirko 442593515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 442693515d53SJiri Pirko return 0; 442793515d53SJiri Pirko } 442893515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 442993515d53SJiri Pirko 443093515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 443193515d53SJiri Pirko { 4432df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 443393515d53SJiri Pirko unsigned int offset = skb->data - skb_mac_header(skb); 443493515d53SJiri Pirko int err; 443593515d53SJiri Pirko 443693515d53SJiri Pirko /* __vlan_insert_tag expect skb->data pointing to mac header. 443793515d53SJiri Pirko * So change skb->data before calling it and change back to 443893515d53SJiri Pirko * original position later 443993515d53SJiri Pirko */ 444093515d53SJiri Pirko __skb_push(skb, offset); 444193515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 4442df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 444393515d53SJiri Pirko if (err) 444493515d53SJiri Pirko return err; 444593515d53SJiri Pirko skb->protocol = skb->vlan_proto; 444693515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 444793515d53SJiri Pirko __skb_pull(skb, offset); 444893515d53SJiri Pirko 444993515d53SJiri Pirko if (skb->ip_summed == CHECKSUM_COMPLETE) 445093515d53SJiri Pirko skb->csum = csum_add(skb->csum, csum_partial(skb->data 445193515d53SJiri Pirko + (2 * ETH_ALEN), VLAN_HLEN, 0)); 445293515d53SJiri Pirko } 445393515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 445493515d53SJiri Pirko return 0; 445593515d53SJiri Pirko } 445693515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 445793515d53SJiri Pirko 44582e4e4410SEric Dumazet /** 44592e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 44602e4e4410SEric Dumazet * 4461de3f0d0eSMasanari Iida * @header_len: size of linear part 4462de3f0d0eSMasanari Iida * @data_len: needed length in frags 4463de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 4464de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 4465de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 44662e4e4410SEric Dumazet * 44672e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 44682e4e4410SEric Dumazet */ 44692e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 44702e4e4410SEric Dumazet unsigned long data_len, 44712e4e4410SEric Dumazet int max_page_order, 44722e4e4410SEric Dumazet int *errcode, 44732e4e4410SEric Dumazet gfp_t gfp_mask) 44742e4e4410SEric Dumazet { 44752e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 44762e4e4410SEric Dumazet unsigned long chunk; 44772e4e4410SEric Dumazet struct sk_buff *skb; 44782e4e4410SEric Dumazet struct page *page; 44792e4e4410SEric Dumazet gfp_t gfp_head; 44802e4e4410SEric Dumazet int i; 44812e4e4410SEric Dumazet 44822e4e4410SEric Dumazet *errcode = -EMSGSIZE; 44832e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 44842e4e4410SEric Dumazet * high order pages... 44852e4e4410SEric Dumazet */ 44862e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 44872e4e4410SEric Dumazet return NULL; 44882e4e4410SEric Dumazet 44892e4e4410SEric Dumazet gfp_head = gfp_mask; 44902e4e4410SEric Dumazet if (gfp_head & __GFP_WAIT) 44912e4e4410SEric Dumazet gfp_head |= __GFP_REPEAT; 44922e4e4410SEric Dumazet 44932e4e4410SEric Dumazet *errcode = -ENOBUFS; 44942e4e4410SEric Dumazet skb = alloc_skb(header_len, gfp_head); 44952e4e4410SEric Dumazet if (!skb) 44962e4e4410SEric Dumazet return NULL; 44972e4e4410SEric Dumazet 44982e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 44992e4e4410SEric Dumazet 45002e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 45012e4e4410SEric Dumazet int order = max_page_order; 45022e4e4410SEric Dumazet 45032e4e4410SEric Dumazet while (order) { 45042e4e4410SEric Dumazet if (npages >= 1 << order) { 45052e4e4410SEric Dumazet page = alloc_pages(gfp_mask | 45062e4e4410SEric Dumazet __GFP_COMP | 45072e4e4410SEric Dumazet __GFP_NOWARN | 45082e4e4410SEric Dumazet __GFP_NORETRY, 45092e4e4410SEric Dumazet order); 45102e4e4410SEric Dumazet if (page) 45112e4e4410SEric Dumazet goto fill_page; 45122e4e4410SEric Dumazet /* Do not retry other high order allocations */ 45132e4e4410SEric Dumazet order = 1; 45142e4e4410SEric Dumazet max_page_order = 0; 45152e4e4410SEric Dumazet } 45162e4e4410SEric Dumazet order--; 45172e4e4410SEric Dumazet } 45182e4e4410SEric Dumazet page = alloc_page(gfp_mask); 45192e4e4410SEric Dumazet if (!page) 45202e4e4410SEric Dumazet goto failure; 45212e4e4410SEric Dumazet fill_page: 45222e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 45232e4e4410SEric Dumazet PAGE_SIZE << order); 45242e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 45252e4e4410SEric Dumazet data_len -= chunk; 45262e4e4410SEric Dumazet npages -= 1 << order; 45272e4e4410SEric Dumazet } 45282e4e4410SEric Dumazet return skb; 45292e4e4410SEric Dumazet 45302e4e4410SEric Dumazet failure: 45312e4e4410SEric Dumazet kfree_skb(skb); 45322e4e4410SEric Dumazet return NULL; 45332e4e4410SEric Dumazet } 45342e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 4535