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> 77a1f8e7f7SAl Viro 78d7e8883cSEric Dumazet struct kmem_cache *skbuff_head_cache __read_mostly; 79e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 801da177e4SLinus Torvalds 811da177e4SLinus Torvalds /** 82f05de73bSJean Sacren * skb_panic - private function for out-of-line support 831da177e4SLinus Torvalds * @skb: buffer 841da177e4SLinus Torvalds * @sz: size 85f05de73bSJean Sacren * @addr: address 8699d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 871da177e4SLinus Torvalds * 88f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 89f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 90f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 91f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 921da177e4SLinus Torvalds */ 93f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 9499d5851eSJames Hogan const char msg[]) 951da177e4SLinus Torvalds { 96e005d193SJoe Perches pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 9799d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 984305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 9926095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1001da177e4SLinus Torvalds BUG(); 1011da177e4SLinus Torvalds } 1021da177e4SLinus Torvalds 103f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1041da177e4SLinus Torvalds { 105f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1061da177e4SLinus Torvalds } 1071da177e4SLinus Torvalds 108f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 109f05de73bSJean Sacren { 110f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 111f05de73bSJean Sacren } 112c93bdd0eSMel Gorman 113c93bdd0eSMel Gorman /* 114c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 115c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 116c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 117c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 118c93bdd0eSMel Gorman * memory is free 119c93bdd0eSMel Gorman */ 120c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 121c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 12261c5e88aSstephen hemminger 12361c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 12461c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 125c93bdd0eSMel Gorman { 126c93bdd0eSMel Gorman void *obj; 127c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 128c93bdd0eSMel Gorman 129c93bdd0eSMel Gorman /* 130c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 131c93bdd0eSMel Gorman * to the reserves, fail. 132c93bdd0eSMel Gorman */ 133c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 134c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 135c93bdd0eSMel Gorman node); 136c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 137c93bdd0eSMel Gorman goto out; 138c93bdd0eSMel Gorman 139c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 140c93bdd0eSMel Gorman ret_pfmemalloc = true; 141c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 142c93bdd0eSMel Gorman 143c93bdd0eSMel Gorman out: 144c93bdd0eSMel Gorman if (pfmemalloc) 145c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 146c93bdd0eSMel Gorman 147c93bdd0eSMel Gorman return obj; 148c93bdd0eSMel Gorman } 149c93bdd0eSMel Gorman 1501da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1511da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1521da177e4SLinus Torvalds * [BEEP] leaks. 1531da177e4SLinus Torvalds * 1541da177e4SLinus Torvalds */ 1551da177e4SLinus Torvalds 1560ebd0ac5SPatrick McHardy struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) 1570ebd0ac5SPatrick McHardy { 1580ebd0ac5SPatrick McHardy struct sk_buff *skb; 1590ebd0ac5SPatrick McHardy 1600ebd0ac5SPatrick McHardy /* Get the HEAD */ 1610ebd0ac5SPatrick McHardy skb = kmem_cache_alloc_node(skbuff_head_cache, 1620ebd0ac5SPatrick McHardy gfp_mask & ~__GFP_DMA, node); 1630ebd0ac5SPatrick McHardy if (!skb) 1640ebd0ac5SPatrick McHardy goto out; 1650ebd0ac5SPatrick McHardy 1660ebd0ac5SPatrick McHardy /* 1670ebd0ac5SPatrick McHardy * Only clear those fields we need to clear, not those that we will 1680ebd0ac5SPatrick McHardy * actually initialise below. Hence, don't put any more fields after 1690ebd0ac5SPatrick McHardy * the tail pointer in struct sk_buff! 1700ebd0ac5SPatrick McHardy */ 1710ebd0ac5SPatrick McHardy memset(skb, 0, offsetof(struct sk_buff, tail)); 1725e71d9d7SPablo Neira skb->head = NULL; 1730ebd0ac5SPatrick McHardy skb->truesize = sizeof(struct sk_buff); 1740ebd0ac5SPatrick McHardy atomic_set(&skb->users, 1); 1750ebd0ac5SPatrick McHardy 17635d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 1770ebd0ac5SPatrick McHardy out: 1780ebd0ac5SPatrick McHardy return skb; 1790ebd0ac5SPatrick McHardy } 1800ebd0ac5SPatrick McHardy 1811da177e4SLinus Torvalds /** 182d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1831da177e4SLinus Torvalds * @size: size to allocate 1841da177e4SLinus Torvalds * @gfp_mask: allocation mask 185c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 186c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 187c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 188c93bdd0eSMel Gorman * allocations in case the data is required for writeback 189b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1901da177e4SLinus Torvalds * 1911da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 19294b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 19394b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 1941da177e4SLinus Torvalds * 1951da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1961da177e4SLinus Torvalds * %GFP_ATOMIC. 1971da177e4SLinus Torvalds */ 198dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 199c93bdd0eSMel Gorman int flags, int node) 2001da177e4SLinus Torvalds { 201e18b890bSChristoph Lameter struct kmem_cache *cache; 2024947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 2031da177e4SLinus Torvalds struct sk_buff *skb; 2041da177e4SLinus Torvalds u8 *data; 205c93bdd0eSMel Gorman bool pfmemalloc; 2061da177e4SLinus Torvalds 207c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 208c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 209c93bdd0eSMel Gorman 210c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 211c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 2128798b3fbSHerbert Xu 2131da177e4SLinus Torvalds /* Get the HEAD */ 214b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 2151da177e4SLinus Torvalds if (!skb) 2161da177e4SLinus Torvalds goto out; 217ec7d2f2cSEric Dumazet prefetchw(skb); 2181da177e4SLinus Torvalds 21987fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 22087fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 22187fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 22287fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 22387fb4b7bSEric Dumazet */ 224bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 22587fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 226c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2271da177e4SLinus Torvalds if (!data) 2281da177e4SLinus Torvalds goto nodata; 22987fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 23087fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 23187fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 23287fb4b7bSEric Dumazet */ 23387fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 234ec7d2f2cSEric Dumazet prefetchw(data + size); 2351da177e4SLinus Torvalds 236ca0605a7SArnaldo Carvalho de Melo /* 237c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 238c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 239c8005785SJohannes Berg * the tail pointer in struct sk_buff! 240ca0605a7SArnaldo Carvalho de Melo */ 241ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 24287fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 24387fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 244c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 2451da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2461da177e4SLinus Torvalds skb->head = data; 2471da177e4SLinus Torvalds skb->data = data; 24827a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2494305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 25035d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 25135d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 25219633e12SStephen Hemminger 2534947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2544947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 255ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2564947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 257c2aa3665SEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 2584947d3efSBenjamin LaHaise 259c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 260d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 2611da177e4SLinus Torvalds 262d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 263d0bf4a9eSEric Dumazet 264d0bf4a9eSEric Dumazet kmemcheck_annotate_bitfield(&fclones->skb2, flags1); 265d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 266d0bf4a9eSEric Dumazet atomic_set(&fclones->fclone_ref, 1); 267d179cd12SDavid S. Miller 268c8753d55SVijay Subramanian fclones->skb2.fclone = SKB_FCLONE_FREE; 269d0bf4a9eSEric Dumazet fclones->skb2.pfmemalloc = pfmemalloc; 270d179cd12SDavid S. Miller } 2711da177e4SLinus Torvalds out: 2721da177e4SLinus Torvalds return skb; 2731da177e4SLinus Torvalds nodata: 2748798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2751da177e4SLinus Torvalds skb = NULL; 2761da177e4SLinus Torvalds goto out; 2771da177e4SLinus Torvalds } 278b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2791da177e4SLinus Torvalds 2801da177e4SLinus Torvalds /** 281b2b5ce9dSEric Dumazet * build_skb - build a network buffer 282b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 283d3836f21SEric Dumazet * @frag_size: size of fragment, or 0 if head was kmalloced 284b2b5ce9dSEric Dumazet * 285b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 286deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 287deceb4c0SFlorian Fainelli * @frag_size is 0, otherwise data should come from the page allocator. 288b2b5ce9dSEric Dumazet * The return is the new skb buffer. 289b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 290b2b5ce9dSEric Dumazet * Notes : 291b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 292b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 293b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 294b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 295b2b5ce9dSEric Dumazet * before giving packet to stack. 296b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 297b2b5ce9dSEric Dumazet */ 298d3836f21SEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 299b2b5ce9dSEric Dumazet { 300b2b5ce9dSEric Dumazet struct skb_shared_info *shinfo; 301b2b5ce9dSEric Dumazet struct sk_buff *skb; 302d3836f21SEric Dumazet unsigned int size = frag_size ? : ksize(data); 303b2b5ce9dSEric Dumazet 304b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 305b2b5ce9dSEric Dumazet if (!skb) 306b2b5ce9dSEric Dumazet return NULL; 307b2b5ce9dSEric Dumazet 308d3836f21SEric Dumazet size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 309b2b5ce9dSEric Dumazet 310b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 311b2b5ce9dSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 312d3836f21SEric Dumazet skb->head_frag = frag_size != 0; 313b2b5ce9dSEric Dumazet atomic_set(&skb->users, 1); 314b2b5ce9dSEric Dumazet skb->head = data; 315b2b5ce9dSEric Dumazet skb->data = data; 316b2b5ce9dSEric Dumazet skb_reset_tail_pointer(skb); 317b2b5ce9dSEric Dumazet skb->end = skb->tail + size; 31835d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 31935d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 320b2b5ce9dSEric Dumazet 321b2b5ce9dSEric Dumazet /* make sure we initialize shinfo sequentially */ 322b2b5ce9dSEric Dumazet shinfo = skb_shinfo(skb); 323b2b5ce9dSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 324b2b5ce9dSEric Dumazet atomic_set(&shinfo->dataref, 1); 325b2b5ce9dSEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 326b2b5ce9dSEric Dumazet 327b2b5ce9dSEric Dumazet return skb; 328b2b5ce9dSEric Dumazet } 329b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 330b2b5ce9dSEric Dumazet 331a1c7fff7SEric Dumazet struct netdev_alloc_cache { 33269b08f62SEric Dumazet struct page_frag frag; 33369b08f62SEric Dumazet /* we maintain a pagecount bias, so that we dont dirty cache line 33469b08f62SEric Dumazet * containing page->_count every time we allocate a fragment. 33569b08f62SEric Dumazet */ 336540eb7bfSAlexander Duyck unsigned int pagecnt_bias; 337a1c7fff7SEric Dumazet }; 338a1c7fff7SEric Dumazet static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache); 339a1c7fff7SEric Dumazet 340c93bdd0eSMel Gorman static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 3416f532612SEric Dumazet { 3426f532612SEric Dumazet struct netdev_alloc_cache *nc; 3436f532612SEric Dumazet void *data = NULL; 34469b08f62SEric Dumazet int order; 3456f532612SEric Dumazet unsigned long flags; 3466f532612SEric Dumazet 3476f532612SEric Dumazet local_irq_save(flags); 3486f532612SEric Dumazet nc = &__get_cpu_var(netdev_alloc_cache); 34969b08f62SEric Dumazet if (unlikely(!nc->frag.page)) { 3506f532612SEric Dumazet refill: 35169b08f62SEric Dumazet for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) { 35269b08f62SEric Dumazet gfp_t gfp = gfp_mask; 35369b08f62SEric Dumazet 35469b08f62SEric Dumazet if (order) 35569b08f62SEric Dumazet gfp |= __GFP_COMP | __GFP_NOWARN; 35669b08f62SEric Dumazet nc->frag.page = alloc_pages(gfp, order); 35769b08f62SEric Dumazet if (likely(nc->frag.page)) 35869b08f62SEric Dumazet break; 35969b08f62SEric Dumazet if (--order < 0) 360540eb7bfSAlexander Duyck goto end; 36169b08f62SEric Dumazet } 36269b08f62SEric Dumazet nc->frag.size = PAGE_SIZE << order; 363540eb7bfSAlexander Duyck recycle: 36469b08f62SEric Dumazet atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS); 36569b08f62SEric Dumazet nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS; 36669b08f62SEric Dumazet nc->frag.offset = 0; 3676f532612SEric Dumazet } 368540eb7bfSAlexander Duyck 36969b08f62SEric Dumazet if (nc->frag.offset + fragsz > nc->frag.size) { 370540eb7bfSAlexander Duyck /* avoid unnecessary locked operations if possible */ 37169b08f62SEric Dumazet if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) || 37269b08f62SEric Dumazet atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count)) 373540eb7bfSAlexander Duyck goto recycle; 3746f532612SEric Dumazet goto refill; 3756f532612SEric Dumazet } 376540eb7bfSAlexander Duyck 37769b08f62SEric Dumazet data = page_address(nc->frag.page) + nc->frag.offset; 37869b08f62SEric Dumazet nc->frag.offset += fragsz; 379540eb7bfSAlexander Duyck nc->pagecnt_bias--; 380540eb7bfSAlexander Duyck end: 3816f532612SEric Dumazet local_irq_restore(flags); 3826f532612SEric Dumazet return data; 3836f532612SEric Dumazet } 384c93bdd0eSMel Gorman 385c93bdd0eSMel Gorman /** 386c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 387c93bdd0eSMel Gorman * @fragsz: fragment size 388c93bdd0eSMel Gorman * 389c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 390c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 391c93bdd0eSMel Gorman */ 392c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 393c93bdd0eSMel Gorman { 394c93bdd0eSMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 395c93bdd0eSMel Gorman } 3966f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 3976f532612SEric Dumazet 3986f532612SEric Dumazet /** 3998af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 4008af27456SChristoph Hellwig * @dev: network device to receive on 4018af27456SChristoph Hellwig * @length: length to allocate 4028af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 4038af27456SChristoph Hellwig * 4048af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 4058af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 4068af27456SChristoph Hellwig * the headroom they think they need without accounting for the 4078af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 4088af27456SChristoph Hellwig * 4098af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 4108af27456SChristoph Hellwig */ 4118af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 4128af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 4138af27456SChristoph Hellwig { 4146f532612SEric Dumazet struct sk_buff *skb = NULL; 415a1c7fff7SEric Dumazet unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) + 416a1c7fff7SEric Dumazet SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4178af27456SChristoph Hellwig 418310e158cSEric Dumazet if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) { 419c93bdd0eSMel Gorman void *data; 420c93bdd0eSMel Gorman 421c93bdd0eSMel Gorman if (sk_memalloc_socks()) 422c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 423c93bdd0eSMel Gorman 424c93bdd0eSMel Gorman data = __netdev_alloc_frag(fragsz, gfp_mask); 425a1c7fff7SEric Dumazet 4266f532612SEric Dumazet if (likely(data)) { 4276f532612SEric Dumazet skb = build_skb(data, fragsz); 4286f532612SEric Dumazet if (unlikely(!skb)) 4296f532612SEric Dumazet put_page(virt_to_head_page(data)); 430a1c7fff7SEric Dumazet } 431a1c7fff7SEric Dumazet } else { 432c93bdd0eSMel Gorman skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 433c93bdd0eSMel Gorman SKB_ALLOC_RX, NUMA_NO_NODE); 434a1c7fff7SEric Dumazet } 4357b2e497aSChristoph Hellwig if (likely(skb)) { 4368af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4377b2e497aSChristoph Hellwig skb->dev = dev; 4387b2e497aSChristoph Hellwig } 4398af27456SChristoph Hellwig return skb; 4408af27456SChristoph Hellwig } 441b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4421da177e4SLinus Torvalds 443654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 44450269e19SEric Dumazet int size, unsigned int truesize) 445654bed16SPeter Zijlstra { 446654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 447654bed16SPeter Zijlstra skb->len += size; 448654bed16SPeter Zijlstra skb->data_len += size; 44950269e19SEric Dumazet skb->truesize += truesize; 450654bed16SPeter Zijlstra } 451654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 452654bed16SPeter Zijlstra 453f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 454f8e617e1SJason Wang unsigned int truesize) 455f8e617e1SJason Wang { 456f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 457f8e617e1SJason Wang 458f8e617e1SJason Wang skb_frag_size_add(frag, size); 459f8e617e1SJason Wang skb->len += size; 460f8e617e1SJason Wang skb->data_len += size; 461f8e617e1SJason Wang skb->truesize += truesize; 462f8e617e1SJason Wang } 463f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 464f8e617e1SJason Wang 46527b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 4661da177e4SLinus Torvalds { 467bd8a7036SEric Dumazet kfree_skb_list(*listp); 46827b437c8SHerbert Xu *listp = NULL; 4691da177e4SLinus Torvalds } 4701da177e4SLinus Torvalds 47127b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 47227b437c8SHerbert Xu { 47327b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 47427b437c8SHerbert Xu } 47527b437c8SHerbert Xu 4761da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 4771da177e4SLinus Torvalds { 4781da177e4SLinus Torvalds struct sk_buff *list; 4791da177e4SLinus Torvalds 480fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 4811da177e4SLinus Torvalds skb_get(list); 4821da177e4SLinus Torvalds } 4831da177e4SLinus Torvalds 484d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 485d3836f21SEric Dumazet { 486d3836f21SEric Dumazet if (skb->head_frag) 487d3836f21SEric Dumazet put_page(virt_to_head_page(skb->head)); 488d3836f21SEric Dumazet else 489d3836f21SEric Dumazet kfree(skb->head); 490d3836f21SEric Dumazet } 491d3836f21SEric Dumazet 4925bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 4931da177e4SLinus Torvalds { 494ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 4951da177e4SLinus Torvalds int i; 496ff04a771SEric Dumazet 497ff04a771SEric Dumazet if (skb->cloned && 498ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 499ff04a771SEric Dumazet &shinfo->dataref)) 500ff04a771SEric Dumazet return; 501ff04a771SEric Dumazet 502ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 503ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 5041da177e4SLinus Torvalds 505a6686f2fSShirley Ma /* 506a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 507a6686f2fSShirley Ma * the lower device DMA has done; 508a6686f2fSShirley Ma */ 509ff04a771SEric Dumazet if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) { 510a6686f2fSShirley Ma struct ubuf_info *uarg; 511a6686f2fSShirley Ma 512ff04a771SEric Dumazet uarg = shinfo->destructor_arg; 513a6686f2fSShirley Ma if (uarg->callback) 514e19d6763SMichael S. Tsirkin uarg->callback(uarg, true); 515a6686f2fSShirley Ma } 516a6686f2fSShirley Ma 517ff04a771SEric Dumazet if (shinfo->frag_list) 518ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 5191da177e4SLinus Torvalds 520d3836f21SEric Dumazet skb_free_head(skb); 5211da177e4SLinus Torvalds } 5221da177e4SLinus Torvalds 5231da177e4SLinus Torvalds /* 5241da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 5251da177e4SLinus Torvalds */ 5262d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 5271da177e4SLinus Torvalds { 528d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 529d179cd12SDavid S. Miller 530d179cd12SDavid S. Miller switch (skb->fclone) { 531d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 5321da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 533d179cd12SDavid S. Miller break; 534d179cd12SDavid S. Miller 535d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 536d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 537d0bf4a9eSEric Dumazet if (atomic_dec_and_test(&fclones->fclone_ref)) 538d0bf4a9eSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 539d179cd12SDavid S. Miller break; 540d179cd12SDavid S. Miller 541d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 542d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 543d179cd12SDavid S. Miller 544ce1a4ea3SEric Dumazet /* Warning : We must perform the atomic_dec_and_test() before 545c8753d55SVijay Subramanian * setting skb->fclone back to SKB_FCLONE_FREE, otherwise 546ce1a4ea3SEric Dumazet * skb_clone() could set clone_ref to 2 before our decrement. 547ce1a4ea3SEric Dumazet * Anyway, if we are going to free the structure, no need to 548ce1a4ea3SEric Dumazet * rewrite skb->fclone. 549ce1a4ea3SEric Dumazet */ 550ce1a4ea3SEric Dumazet if (atomic_dec_and_test(&fclones->fclone_ref)) { 551ce1a4ea3SEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 552ce1a4ea3SEric Dumazet } else { 553d179cd12SDavid S. Miller /* The clone portion is available for 554d179cd12SDavid S. Miller * fast-cloning again. 555d179cd12SDavid S. Miller */ 556c8753d55SVijay Subramanian skb->fclone = SKB_FCLONE_FREE; 557ce1a4ea3SEric Dumazet } 558d179cd12SDavid S. Miller break; 5593ff50b79SStephen Hemminger } 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds 56204a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 5631da177e4SLinus Torvalds { 564adf30907SEric Dumazet skb_dst_drop(skb); 5651da177e4SLinus Torvalds #ifdef CONFIG_XFRM 5661da177e4SLinus Torvalds secpath_put(skb->sp); 5671da177e4SLinus Torvalds #endif 5681da177e4SLinus Torvalds if (skb->destructor) { 5699c2b3328SStephen Hemminger WARN_ON(in_irq()); 5701da177e4SLinus Torvalds skb->destructor(skb); 5711da177e4SLinus Torvalds } 572a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5735f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 5742fc72c7bSKOVACS Krisztian #endif 5751109a90cSPablo Neira Ayuso #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 5761da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 5771da177e4SLinus Torvalds #endif 5781da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 5791da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 5801da177e4SLinus Torvalds skb->tc_index = 0; 5811da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 5821da177e4SLinus Torvalds skb->tc_verd = 0; 5831da177e4SLinus Torvalds #endif 5841da177e4SLinus Torvalds #endif 58504a4bb55SLennert Buytenhek } 58604a4bb55SLennert Buytenhek 58704a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 58804a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 58904a4bb55SLennert Buytenhek { 59004a4bb55SLennert Buytenhek skb_release_head_state(skb); 5915e71d9d7SPablo Neira if (likely(skb->head)) 5922d4baff8SHerbert Xu skb_release_data(skb); 5932d4baff8SHerbert Xu } 5941da177e4SLinus Torvalds 5952d4baff8SHerbert Xu /** 5962d4baff8SHerbert Xu * __kfree_skb - private function 5972d4baff8SHerbert Xu * @skb: buffer 5982d4baff8SHerbert Xu * 5992d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6002d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6012d4baff8SHerbert Xu * always call kfree_skb 6022d4baff8SHerbert Xu */ 6032d4baff8SHerbert Xu 6042d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6052d4baff8SHerbert Xu { 6062d4baff8SHerbert Xu skb_release_all(skb); 6071da177e4SLinus Torvalds kfree_skbmem(skb); 6081da177e4SLinus Torvalds } 609b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6101da177e4SLinus Torvalds 6111da177e4SLinus Torvalds /** 612231d06aeSJörn Engel * kfree_skb - free an sk_buff 613231d06aeSJörn Engel * @skb: buffer to free 614231d06aeSJörn Engel * 615231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 616231d06aeSJörn Engel * hit zero. 617231d06aeSJörn Engel */ 618231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 619231d06aeSJörn Engel { 620231d06aeSJörn Engel if (unlikely(!skb)) 621231d06aeSJörn Engel return; 622231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 623231d06aeSJörn Engel smp_rmb(); 624231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 625231d06aeSJörn Engel return; 626ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 627231d06aeSJörn Engel __kfree_skb(skb); 628231d06aeSJörn Engel } 629b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 630231d06aeSJörn Engel 631bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 632bd8a7036SEric Dumazet { 633bd8a7036SEric Dumazet while (segs) { 634bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 635bd8a7036SEric Dumazet 636bd8a7036SEric Dumazet kfree_skb(segs); 637bd8a7036SEric Dumazet segs = next; 638bd8a7036SEric Dumazet } 639bd8a7036SEric Dumazet } 640bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 641bd8a7036SEric Dumazet 642d1a203eaSStephen Hemminger /** 64325121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 64425121173SMichael S. Tsirkin * @skb: buffer that triggered an error 64525121173SMichael S. Tsirkin * 64625121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 64725121173SMichael S. Tsirkin * skb must be freed afterwards. 64825121173SMichael S. Tsirkin */ 64925121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 65025121173SMichael S. Tsirkin { 65125121173SMichael S. Tsirkin if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 65225121173SMichael S. Tsirkin struct ubuf_info *uarg; 65325121173SMichael S. Tsirkin 65425121173SMichael S. Tsirkin uarg = skb_shinfo(skb)->destructor_arg; 65525121173SMichael S. Tsirkin if (uarg->callback) 65625121173SMichael S. Tsirkin uarg->callback(uarg, false); 65725121173SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 65825121173SMichael S. Tsirkin } 65925121173SMichael S. Tsirkin } 66025121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 66125121173SMichael S. Tsirkin 66225121173SMichael S. Tsirkin /** 663ead2ceb0SNeil Horman * consume_skb - free an skbuff 664ead2ceb0SNeil Horman * @skb: buffer to free 665ead2ceb0SNeil Horman * 666ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 667ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 668ead2ceb0SNeil Horman * is being dropped after a failure and notes that 669ead2ceb0SNeil Horman */ 670ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 671ead2ceb0SNeil Horman { 672ead2ceb0SNeil Horman if (unlikely(!skb)) 673ead2ceb0SNeil Horman return; 674ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 675ead2ceb0SNeil Horman smp_rmb(); 676ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 677ead2ceb0SNeil Horman return; 67807dc22e7SKoki Sanagi trace_consume_skb(skb); 679ead2ceb0SNeil Horman __kfree_skb(skb); 680ead2ceb0SNeil Horman } 681ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 682ead2ceb0SNeil Horman 683b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 684b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 685b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 686b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 687b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 688b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 689b1937227SEric Dumazet 690dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 691dec18810SHerbert Xu { 692dec18810SHerbert Xu new->tstamp = old->tstamp; 693b1937227SEric Dumazet /* We do not copy old->sk */ 694dec18810SHerbert Xu new->dev = old->dev; 695b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 6967fee226aSEric Dumazet skb_dst_copy(new, old); 697def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 698dec18810SHerbert Xu new->sp = secpath_get(old->sp); 699dec18810SHerbert Xu #endif 700b1937227SEric Dumazet __nf_copy(new, old, false); 7016aa895b0SPatrick McHardy 702b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 703b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 704b1937227SEric Dumazet */ 705b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 70606021292SEliezer Tamir 707b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 708b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 709b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 710b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 711b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 712b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 713b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 714b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 715b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 716b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 717b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 718b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 719b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 720b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 721b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 722b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 723b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 724b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 725b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 726b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 727b1937227SEric Dumazet #endif 728e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 729b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 73006021292SEliezer Tamir #endif 731b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 732b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 733b1937227SEric Dumazet #ifdef CONFIG_NET_CLS_ACT 734b1937227SEric Dumazet CHECK_SKB_FIELD(tc_verd); 735b1937227SEric Dumazet #endif 736b1937227SEric Dumazet #endif 737b1937227SEric Dumazet 738dec18810SHerbert Xu } 739dec18810SHerbert Xu 74082c49a35SHerbert Xu /* 74182c49a35SHerbert Xu * You should not add any new code to this function. Add it to 74282c49a35SHerbert Xu * __copy_skb_header above instead. 74382c49a35SHerbert Xu */ 744e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 7451da177e4SLinus Torvalds { 7461da177e4SLinus Torvalds #define C(x) n->x = skb->x 7471da177e4SLinus Torvalds 7481da177e4SLinus Torvalds n->next = n->prev = NULL; 7491da177e4SLinus Torvalds n->sk = NULL; 750dec18810SHerbert Xu __copy_skb_header(n, skb); 751dec18810SHerbert Xu 7521da177e4SLinus Torvalds C(len); 7531da177e4SLinus Torvalds C(data_len); 7543e6b3b2eSAlexey Dobriyan C(mac_len); 755334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 75602f1c89dSPaul Moore n->cloned = 1; 7571da177e4SLinus Torvalds n->nohdr = 0; 7581da177e4SLinus Torvalds n->destructor = NULL; 7591da177e4SLinus Torvalds C(tail); 7601da177e4SLinus Torvalds C(end); 76102f1c89dSPaul Moore C(head); 762d3836f21SEric Dumazet C(head_frag); 76302f1c89dSPaul Moore C(data); 76402f1c89dSPaul Moore C(truesize); 76502f1c89dSPaul Moore atomic_set(&n->users, 1); 7661da177e4SLinus Torvalds 7671da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 7681da177e4SLinus Torvalds skb->cloned = 1; 7691da177e4SLinus Torvalds 7701da177e4SLinus Torvalds return n; 771e0053ec0SHerbert Xu #undef C 772e0053ec0SHerbert Xu } 773e0053ec0SHerbert Xu 774e0053ec0SHerbert Xu /** 775e0053ec0SHerbert Xu * skb_morph - morph one skb into another 776e0053ec0SHerbert Xu * @dst: the skb to receive the contents 777e0053ec0SHerbert Xu * @src: the skb to supply the contents 778e0053ec0SHerbert Xu * 779e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 780e0053ec0SHerbert Xu * supplied by the user. 781e0053ec0SHerbert Xu * 782e0053ec0SHerbert Xu * The target skb is returned upon exit. 783e0053ec0SHerbert Xu */ 784e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 785e0053ec0SHerbert Xu { 7862d4baff8SHerbert Xu skb_release_all(dst); 787e0053ec0SHerbert Xu return __skb_clone(dst, src); 788e0053ec0SHerbert Xu } 789e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 790e0053ec0SHerbert Xu 7912c53040fSBen Hutchings /** 7922c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 79348c83012SMichael S. Tsirkin * @skb: the skb to modify 79448c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 79548c83012SMichael S. Tsirkin * 79648c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 79748c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 79848c83012SMichael S. Tsirkin * to userspace pages. 79948c83012SMichael S. Tsirkin * 80048c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 80148c83012SMichael S. Tsirkin * %GFP_ATOMIC. 80248c83012SMichael S. Tsirkin * 80348c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 80448c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 80548c83012SMichael S. Tsirkin */ 80648c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 807a6686f2fSShirley Ma { 808a6686f2fSShirley Ma int i; 809a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 810a6686f2fSShirley Ma struct page *page, *head = NULL; 811a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 812a6686f2fSShirley Ma 813a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 814a6686f2fSShirley Ma u8 *vaddr; 815a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 816a6686f2fSShirley Ma 81702756ed4SKrishna Kumar page = alloc_page(gfp_mask); 818a6686f2fSShirley Ma if (!page) { 819a6686f2fSShirley Ma while (head) { 82040dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 821a6686f2fSShirley Ma put_page(head); 822a6686f2fSShirley Ma head = next; 823a6686f2fSShirley Ma } 824a6686f2fSShirley Ma return -ENOMEM; 825a6686f2fSShirley Ma } 82651c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 827a6686f2fSShirley Ma memcpy(page_address(page), 8289e903e08SEric Dumazet vaddr + f->page_offset, skb_frag_size(f)); 82951c56b00SEric Dumazet kunmap_atomic(vaddr); 83040dadff2SSunghan Suh set_page_private(page, (unsigned long)head); 831a6686f2fSShirley Ma head = page; 832a6686f2fSShirley Ma } 833a6686f2fSShirley Ma 834a6686f2fSShirley Ma /* skb frags release userspace buffers */ 83502756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 836a8605c60SIan Campbell skb_frag_unref(skb, i); 837a6686f2fSShirley Ma 838e19d6763SMichael S. Tsirkin uarg->callback(uarg, false); 839a6686f2fSShirley Ma 840a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 84102756ed4SKrishna Kumar for (i = num_frags - 1; i >= 0; i--) { 84202756ed4SKrishna Kumar __skb_fill_page_desc(skb, i, head, 0, 84302756ed4SKrishna Kumar skb_shinfo(skb)->frags[i].size); 84440dadff2SSunghan Suh head = (struct page *)page_private(head); 845a6686f2fSShirley Ma } 84648c83012SMichael S. Tsirkin 84748c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 848a6686f2fSShirley Ma return 0; 849a6686f2fSShirley Ma } 850dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 851a6686f2fSShirley Ma 852e0053ec0SHerbert Xu /** 853e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 854e0053ec0SHerbert Xu * @skb: buffer to clone 855e0053ec0SHerbert Xu * @gfp_mask: allocation priority 856e0053ec0SHerbert Xu * 857e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 858e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 859e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 860e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 861e0053ec0SHerbert Xu * 862e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 863e0053ec0SHerbert Xu * %GFP_ATOMIC. 864e0053ec0SHerbert Xu */ 865e0053ec0SHerbert Xu 866e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 867e0053ec0SHerbert Xu { 868d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 869d0bf4a9eSEric Dumazet struct sk_buff_fclones, 870d0bf4a9eSEric Dumazet skb1); 871d0bf4a9eSEric Dumazet struct sk_buff *n = &fclones->skb2; 872e0053ec0SHerbert Xu 87370008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 874a6686f2fSShirley Ma return NULL; 875a6686f2fSShirley Ma 876e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 877c8753d55SVijay Subramanian n->fclone == SKB_FCLONE_FREE) { 878e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_CLONE; 879ce1a4ea3SEric Dumazet /* As our fastclone was free, clone_ref must be 1 at this point. 880ce1a4ea3SEric Dumazet * We could use atomic_inc() here, but it is faster 881ce1a4ea3SEric Dumazet * to set the final value. 882ce1a4ea3SEric Dumazet */ 883ce1a4ea3SEric Dumazet atomic_set(&fclones->fclone_ref, 2); 884e0053ec0SHerbert Xu } else { 885c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 886c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 887c93bdd0eSMel Gorman 888e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 889e0053ec0SHerbert Xu if (!n) 890e0053ec0SHerbert Xu return NULL; 891fe55f6d5SVegard Nossum 892fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 893e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 894e0053ec0SHerbert Xu } 895e0053ec0SHerbert Xu 896e0053ec0SHerbert Xu return __skb_clone(n, skb); 8971da177e4SLinus Torvalds } 898b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 8991da177e4SLinus Torvalds 900f5b17294SPravin B Shelar static void skb_headers_offset_update(struct sk_buff *skb, int off) 901f5b17294SPravin B Shelar { 902030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 903030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 904030737bcSEric Dumazet skb->csum_start += off; 905f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 906f5b17294SPravin B Shelar skb->transport_header += off; 907f5b17294SPravin B Shelar skb->network_header += off; 908f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 909f5b17294SPravin B Shelar skb->mac_header += off; 910f5b17294SPravin B Shelar skb->inner_transport_header += off; 911f5b17294SPravin B Shelar skb->inner_network_header += off; 912aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 913f5b17294SPravin B Shelar } 914f5b17294SPravin B Shelar 9151da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 9161da177e4SLinus Torvalds { 917dec18810SHerbert Xu __copy_skb_header(new, old); 918dec18810SHerbert Xu 9197967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 9207967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 9217967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 9221da177e4SLinus Torvalds } 9231da177e4SLinus Torvalds 924c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 925c93bdd0eSMel Gorman { 926c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 927c93bdd0eSMel Gorman return SKB_ALLOC_RX; 928c93bdd0eSMel Gorman return 0; 929c93bdd0eSMel Gorman } 930c93bdd0eSMel Gorman 9311da177e4SLinus Torvalds /** 9321da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 9331da177e4SLinus Torvalds * @skb: buffer to copy 9341da177e4SLinus Torvalds * @gfp_mask: allocation priority 9351da177e4SLinus Torvalds * 9361da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 9371da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 9381da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 9391da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 9401da177e4SLinus Torvalds * 9411da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 9421da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 9431da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 9441da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 9451da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 9461da177e4SLinus Torvalds */ 9471da177e4SLinus Torvalds 948dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 9491da177e4SLinus Torvalds { 9506602cebbSEric Dumazet int headerlen = skb_headroom(skb); 951ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 952c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 953c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 9546602cebbSEric Dumazet 9551da177e4SLinus Torvalds if (!n) 9561da177e4SLinus Torvalds return NULL; 9571da177e4SLinus Torvalds 9581da177e4SLinus Torvalds /* Set the data pointer */ 9591da177e4SLinus Torvalds skb_reserve(n, headerlen); 9601da177e4SLinus Torvalds /* Set the tail pointer and length */ 9611da177e4SLinus Torvalds skb_put(n, skb->len); 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 9641da177e4SLinus Torvalds BUG(); 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds copy_skb_header(n, skb); 9671da177e4SLinus Torvalds return n; 9681da177e4SLinus Torvalds } 969b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 9701da177e4SLinus Torvalds 9711da177e4SLinus Torvalds /** 972bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 9731da177e4SLinus Torvalds * @skb: buffer to copy 974117632e6SEric Dumazet * @headroom: headroom of new skb 9751da177e4SLinus Torvalds * @gfp_mask: allocation priority 976bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 977bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 978bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 9791da177e4SLinus Torvalds * 9801da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 9811da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 9821da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 9831da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 9841da177e4SLinus Torvalds * or the pointer to the buffer on success. 9851da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 9861da177e4SLinus Torvalds */ 9871da177e4SLinus Torvalds 988bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 989bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 9901da177e4SLinus Torvalds { 991117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 992bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 993bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 9946602cebbSEric Dumazet 9951da177e4SLinus Torvalds if (!n) 9961da177e4SLinus Torvalds goto out; 9971da177e4SLinus Torvalds 9981da177e4SLinus Torvalds /* Set the data pointer */ 999117632e6SEric Dumazet skb_reserve(n, headroom); 10001da177e4SLinus Torvalds /* Set the tail pointer and length */ 10011da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 10021da177e4SLinus Torvalds /* Copy the bytes */ 1003d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 10041da177e4SLinus Torvalds 100525f484a6SHerbert Xu n->truesize += skb->data_len; 10061da177e4SLinus Torvalds n->data_len = skb->data_len; 10071da177e4SLinus Torvalds n->len = skb->len; 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 10101da177e4SLinus Torvalds int i; 10111da177e4SLinus Torvalds 101270008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) { 10131511022cSDan Carpenter kfree_skb(n); 10141511022cSDan Carpenter n = NULL; 1015a6686f2fSShirley Ma goto out; 1016a6686f2fSShirley Ma } 10171da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 10181da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1019ea2ab693SIan Campbell skb_frag_ref(skb, i); 10201da177e4SLinus Torvalds } 10211da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 10221da177e4SLinus Torvalds } 10231da177e4SLinus Torvalds 102421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 10251da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 10261da177e4SLinus Torvalds skb_clone_fraglist(n); 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds copy_skb_header(n, skb); 10301da177e4SLinus Torvalds out: 10311da177e4SLinus Torvalds return n; 10321da177e4SLinus Torvalds } 1033bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds /** 10361da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 10371da177e4SLinus Torvalds * @skb: buffer to reallocate 10381da177e4SLinus Torvalds * @nhead: room to add at head 10391da177e4SLinus Torvalds * @ntail: room to add at tail 10401da177e4SLinus Torvalds * @gfp_mask: allocation priority 10411da177e4SLinus Torvalds * 1042bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1043bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 10441da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 10451da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 10461da177e4SLinus Torvalds * 10471da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 10481da177e4SLinus Torvalds * reloaded after call to this function. 10491da177e4SLinus Torvalds */ 10501da177e4SLinus Torvalds 105186a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1052dd0fc66fSAl Viro gfp_t gfp_mask) 10531da177e4SLinus Torvalds { 10541da177e4SLinus Torvalds int i; 10551da177e4SLinus Torvalds u8 *data; 1056ec47ea82SAlexander Duyck int size = nhead + skb_end_offset(skb) + ntail; 10571da177e4SLinus Torvalds long off; 10581da177e4SLinus Torvalds 10594edd87adSHerbert Xu BUG_ON(nhead < 0); 10604edd87adSHerbert Xu 10611da177e4SLinus Torvalds if (skb_shared(skb)) 10621da177e4SLinus Torvalds BUG(); 10631da177e4SLinus Torvalds 10641da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 10651da177e4SLinus Torvalds 1066c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1067c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1068c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1069c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 10701da177e4SLinus Torvalds if (!data) 10711da177e4SLinus Torvalds goto nodata; 107287151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 10731da177e4SLinus Torvalds 10741da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 10756602cebbSEric Dumazet * optimized for the cases when header is void. 10766602cebbSEric Dumazet */ 10776602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 10786602cebbSEric Dumazet 10796602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 10806602cebbSEric Dumazet skb_shinfo(skb), 1081fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 10821da177e4SLinus Torvalds 10833e24591aSAlexander Duyck /* 10843e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 10853e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 10863e24591aSAlexander Duyck * be since all we did is relocate the values 10873e24591aSAlexander Duyck */ 10883e24591aSAlexander Duyck if (skb_cloned(skb)) { 1089a6686f2fSShirley Ma /* copy this zero copy skb frags */ 109070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1091a6686f2fSShirley Ma goto nofrags; 10921da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1093ea2ab693SIan Campbell skb_frag_ref(skb, i); 10941da177e4SLinus Torvalds 109521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 10961da177e4SLinus Torvalds skb_clone_fraglist(skb); 10971da177e4SLinus Torvalds 10981da177e4SLinus Torvalds skb_release_data(skb); 10993e24591aSAlexander Duyck } else { 11003e24591aSAlexander Duyck skb_free_head(skb); 11011fd63041SEric Dumazet } 11021da177e4SLinus Torvalds off = (data + nhead) - skb->head; 11031da177e4SLinus Torvalds 11041da177e4SLinus Torvalds skb->head = data; 1105d3836f21SEric Dumazet skb->head_frag = 0; 11061da177e4SLinus Torvalds skb->data += off; 11074305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 11084305b541SArnaldo Carvalho de Melo skb->end = size; 110956eb8882SPatrick McHardy off = nhead; 11104305b541SArnaldo Carvalho de Melo #else 11114305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 111256eb8882SPatrick McHardy #endif 111327a884dcSArnaldo Carvalho de Melo skb->tail += off; 1114b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 11151da177e4SLinus Torvalds skb->cloned = 0; 1116334a8132SPatrick McHardy skb->hdr_len = 0; 11171da177e4SLinus Torvalds skb->nohdr = 0; 11181da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 11191da177e4SLinus Torvalds return 0; 11201da177e4SLinus Torvalds 1121a6686f2fSShirley Ma nofrags: 1122a6686f2fSShirley Ma kfree(data); 11231da177e4SLinus Torvalds nodata: 11241da177e4SLinus Torvalds return -ENOMEM; 11251da177e4SLinus Torvalds } 1126b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 11291da177e4SLinus Torvalds 11301da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 11311da177e4SLinus Torvalds { 11321da177e4SLinus Torvalds struct sk_buff *skb2; 11331da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 11341da177e4SLinus Torvalds 11351da177e4SLinus Torvalds if (delta <= 0) 11361da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 11371da177e4SLinus Torvalds else { 11381da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 11391da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 11401da177e4SLinus Torvalds GFP_ATOMIC)) { 11411da177e4SLinus Torvalds kfree_skb(skb2); 11421da177e4SLinus Torvalds skb2 = NULL; 11431da177e4SLinus Torvalds } 11441da177e4SLinus Torvalds } 11451da177e4SLinus Torvalds return skb2; 11461da177e4SLinus Torvalds } 1147b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds /** 11501da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 11511da177e4SLinus Torvalds * @skb: buffer to copy 11521da177e4SLinus Torvalds * @newheadroom: new free bytes at head 11531da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 11541da177e4SLinus Torvalds * @gfp_mask: allocation priority 11551da177e4SLinus Torvalds * 11561da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 11571da177e4SLinus Torvalds * allocate additional space. 11581da177e4SLinus Torvalds * 11591da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 11601da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 11611da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 11621da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 11631da177e4SLinus Torvalds * 11641da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 11651da177e4SLinus Torvalds * is called from an interrupt. 11661da177e4SLinus Torvalds */ 11671da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 116886a76cafSVictor Fusco int newheadroom, int newtailroom, 1169dd0fc66fSAl Viro gfp_t gfp_mask) 11701da177e4SLinus Torvalds { 11711da177e4SLinus Torvalds /* 11721da177e4SLinus Torvalds * Allocate the copy buffer 11731da177e4SLinus Torvalds */ 1174c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1175c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1176c93bdd0eSMel Gorman NUMA_NO_NODE); 1177efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 11781da177e4SLinus Torvalds int head_copy_len, head_copy_off; 11791da177e4SLinus Torvalds 11801da177e4SLinus Torvalds if (!n) 11811da177e4SLinus Torvalds return NULL; 11821da177e4SLinus Torvalds 11831da177e4SLinus Torvalds skb_reserve(n, newheadroom); 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds /* Set the tail pointer and length */ 11861da177e4SLinus Torvalds skb_put(n, skb->len); 11871da177e4SLinus Torvalds 1188efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 11891da177e4SLinus Torvalds head_copy_off = 0; 11901da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 11911da177e4SLinus Torvalds head_copy_len = newheadroom; 11921da177e4SLinus Torvalds else 11931da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds /* Copy the linear header and data. */ 11961da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 11971da177e4SLinus Torvalds skb->len + head_copy_len)) 11981da177e4SLinus Torvalds BUG(); 11991da177e4SLinus Torvalds 12001da177e4SLinus Torvalds copy_skb_header(n, skb); 12011da177e4SLinus Torvalds 1202030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1203efd1e8d5SPatrick McHardy 12041da177e4SLinus Torvalds return n; 12051da177e4SLinus Torvalds } 1206b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds /** 12091da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 12101da177e4SLinus Torvalds * @skb: buffer to pad 12111da177e4SLinus Torvalds * @pad: space to pad 12121da177e4SLinus Torvalds * 12131da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 12141da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 12151da177e4SLinus Torvalds * beyond the buffer end onto the wire. 12161da177e4SLinus Torvalds * 12175b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 12181da177e4SLinus Torvalds */ 12191da177e4SLinus Torvalds 12205b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 12211da177e4SLinus Torvalds { 12225b057c6bSHerbert Xu int err; 12235b057c6bSHerbert Xu int ntail; 12241da177e4SLinus Torvalds 12251da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 12265b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 12271da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 12285b057c6bSHerbert Xu return 0; 12291da177e4SLinus Torvalds } 12301da177e4SLinus Torvalds 12314305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 12325b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 12335b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 12345b057c6bSHerbert Xu if (unlikely(err)) 12355b057c6bSHerbert Xu goto free_skb; 12365b057c6bSHerbert Xu } 12375b057c6bSHerbert Xu 12385b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 12395b057c6bSHerbert Xu * to be audited. 12405b057c6bSHerbert Xu */ 12415b057c6bSHerbert Xu err = skb_linearize(skb); 12425b057c6bSHerbert Xu if (unlikely(err)) 12435b057c6bSHerbert Xu goto free_skb; 12445b057c6bSHerbert Xu 12455b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 12465b057c6bSHerbert Xu return 0; 12475b057c6bSHerbert Xu 12485b057c6bSHerbert Xu free_skb: 12491da177e4SLinus Torvalds kfree_skb(skb); 12505b057c6bSHerbert Xu return err; 12511da177e4SLinus Torvalds } 1252b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 12531da177e4SLinus Torvalds 12540dde3e16SIlpo Järvinen /** 12550c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 12560c7ddf36SMathias Krause * @skb: start of the buffer to use 12570c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 12580c7ddf36SMathias Krause * @len: amount of data to add 12590c7ddf36SMathias Krause * 12600c7ddf36SMathias Krause * This function extends the used data area of the potentially 12610c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 12620c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 12630c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 12640c7ddf36SMathias Krause * returned. 12650c7ddf36SMathias Krause */ 12660c7ddf36SMathias Krause 12670c7ddf36SMathias Krause unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 12680c7ddf36SMathias Krause { 12690c7ddf36SMathias Krause if (tail != skb) { 12700c7ddf36SMathias Krause skb->data_len += len; 12710c7ddf36SMathias Krause skb->len += len; 12720c7ddf36SMathias Krause } 12730c7ddf36SMathias Krause return skb_put(tail, len); 12740c7ddf36SMathias Krause } 12750c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 12760c7ddf36SMathias Krause 12770c7ddf36SMathias Krause /** 12780dde3e16SIlpo Järvinen * skb_put - add data to a buffer 12790dde3e16SIlpo Järvinen * @skb: buffer to use 12800dde3e16SIlpo Järvinen * @len: amount of data to add 12810dde3e16SIlpo Järvinen * 12820dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 12830dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 12840dde3e16SIlpo Järvinen * first byte of the extra data is returned. 12850dde3e16SIlpo Järvinen */ 12860dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 12870dde3e16SIlpo Järvinen { 12880dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 12890dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 12900dde3e16SIlpo Järvinen skb->tail += len; 12910dde3e16SIlpo Järvinen skb->len += len; 12920dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 12930dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 12940dde3e16SIlpo Järvinen return tmp; 12950dde3e16SIlpo Järvinen } 12960dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 12970dde3e16SIlpo Järvinen 12986be8ac2fSIlpo Järvinen /** 1299c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1300c2aa270aSIlpo Järvinen * @skb: buffer to use 1301c2aa270aSIlpo Järvinen * @len: amount of data to add 1302c2aa270aSIlpo Järvinen * 1303c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1304c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1305c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1306c2aa270aSIlpo Järvinen */ 1307c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1308c2aa270aSIlpo Järvinen { 1309c2aa270aSIlpo Järvinen skb->data -= len; 1310c2aa270aSIlpo Järvinen skb->len += len; 1311c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1312c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1313c2aa270aSIlpo Järvinen return skb->data; 1314c2aa270aSIlpo Järvinen } 1315c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1316c2aa270aSIlpo Järvinen 1317c2aa270aSIlpo Järvinen /** 13186be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 13196be8ac2fSIlpo Järvinen * @skb: buffer to use 13206be8ac2fSIlpo Järvinen * @len: amount of data to remove 13216be8ac2fSIlpo Järvinen * 13226be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 13236be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 13246be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 13256be8ac2fSIlpo Järvinen * the old data. 13266be8ac2fSIlpo Järvinen */ 13276be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 13286be8ac2fSIlpo Järvinen { 132947d29646SDavid S. Miller return skb_pull_inline(skb, len); 13306be8ac2fSIlpo Järvinen } 13316be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 13326be8ac2fSIlpo Järvinen 1333419ae74eSIlpo Järvinen /** 1334419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1335419ae74eSIlpo Järvinen * @skb: buffer to alter 1336419ae74eSIlpo Järvinen * @len: new length 1337419ae74eSIlpo Järvinen * 1338419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1339419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1340419ae74eSIlpo Järvinen * The skb must be linear. 1341419ae74eSIlpo Järvinen */ 1342419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1343419ae74eSIlpo Järvinen { 1344419ae74eSIlpo Järvinen if (skb->len > len) 1345419ae74eSIlpo Järvinen __skb_trim(skb, len); 1346419ae74eSIlpo Järvinen } 1347419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1348419ae74eSIlpo Järvinen 13493cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 13501da177e4SLinus Torvalds */ 13511da177e4SLinus Torvalds 13523cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 13531da177e4SLinus Torvalds { 135427b437c8SHerbert Xu struct sk_buff **fragp; 135527b437c8SHerbert Xu struct sk_buff *frag; 13561da177e4SLinus Torvalds int offset = skb_headlen(skb); 13571da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 13581da177e4SLinus Torvalds int i; 135927b437c8SHerbert Xu int err; 136027b437c8SHerbert Xu 136127b437c8SHerbert Xu if (skb_cloned(skb) && 136227b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 136327b437c8SHerbert Xu return err; 13641da177e4SLinus Torvalds 1365f4d26fb3SHerbert Xu i = 0; 1366f4d26fb3SHerbert Xu if (offset >= len) 1367f4d26fb3SHerbert Xu goto drop_pages; 1368f4d26fb3SHerbert Xu 1369f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 13709e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 137127b437c8SHerbert Xu 137227b437c8SHerbert Xu if (end < len) { 13731da177e4SLinus Torvalds offset = end; 137427b437c8SHerbert Xu continue; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 13779e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 137827b437c8SHerbert Xu 1379f4d26fb3SHerbert Xu drop_pages: 138027b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 138127b437c8SHerbert Xu 138227b437c8SHerbert Xu for (; i < nfrags; i++) 1383ea2ab693SIan Campbell skb_frag_unref(skb, i); 138427b437c8SHerbert Xu 138521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 138627b437c8SHerbert Xu skb_drop_fraglist(skb); 1387f4d26fb3SHerbert Xu goto done; 138827b437c8SHerbert Xu } 138927b437c8SHerbert Xu 139027b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 139127b437c8SHerbert Xu fragp = &frag->next) { 139227b437c8SHerbert Xu int end = offset + frag->len; 139327b437c8SHerbert Xu 139427b437c8SHerbert Xu if (skb_shared(frag)) { 139527b437c8SHerbert Xu struct sk_buff *nfrag; 139627b437c8SHerbert Xu 139727b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 139827b437c8SHerbert Xu if (unlikely(!nfrag)) 139927b437c8SHerbert Xu return -ENOMEM; 140027b437c8SHerbert Xu 140127b437c8SHerbert Xu nfrag->next = frag->next; 140285bb2a60SEric Dumazet consume_skb(frag); 140327b437c8SHerbert Xu frag = nfrag; 140427b437c8SHerbert Xu *fragp = frag; 140527b437c8SHerbert Xu } 140627b437c8SHerbert Xu 140727b437c8SHerbert Xu if (end < len) { 140827b437c8SHerbert Xu offset = end; 140927b437c8SHerbert Xu continue; 141027b437c8SHerbert Xu } 141127b437c8SHerbert Xu 141227b437c8SHerbert Xu if (end > len && 141327b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 141427b437c8SHerbert Xu return err; 141527b437c8SHerbert Xu 141627b437c8SHerbert Xu if (frag->next) 141727b437c8SHerbert Xu skb_drop_list(&frag->next); 141827b437c8SHerbert Xu break; 141927b437c8SHerbert Xu } 142027b437c8SHerbert Xu 1421f4d26fb3SHerbert Xu done: 142227b437c8SHerbert Xu if (len > skb_headlen(skb)) { 14231da177e4SLinus Torvalds skb->data_len -= skb->len - len; 14241da177e4SLinus Torvalds skb->len = len; 14251da177e4SLinus Torvalds } else { 14261da177e4SLinus Torvalds skb->len = len; 14271da177e4SLinus Torvalds skb->data_len = 0; 142827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 14291da177e4SLinus Torvalds } 14301da177e4SLinus Torvalds 14311da177e4SLinus Torvalds return 0; 14321da177e4SLinus Torvalds } 1433b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 14341da177e4SLinus Torvalds 14351da177e4SLinus Torvalds /** 14361da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 14371da177e4SLinus Torvalds * @skb: buffer to reallocate 14381da177e4SLinus Torvalds * @delta: number of bytes to advance tail 14391da177e4SLinus Torvalds * 14401da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 14411da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 14421da177e4SLinus Torvalds * data from fragmented part. 14431da177e4SLinus Torvalds * 14441da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 14451da177e4SLinus Torvalds * 14461da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 14471da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 14481da177e4SLinus Torvalds * 14491da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 14501da177e4SLinus Torvalds * reloaded after call to this function. 14511da177e4SLinus Torvalds */ 14521da177e4SLinus Torvalds 14531da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 14541da177e4SLinus Torvalds * when it is necessary. 14551da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 14561da177e4SLinus Torvalds * 2. It may change skb pointers. 14571da177e4SLinus Torvalds * 14581da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 14591da177e4SLinus Torvalds */ 14601da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 14611da177e4SLinus Torvalds { 14621da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 14631da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 14641da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 14651da177e4SLinus Torvalds */ 14664305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 14671da177e4SLinus Torvalds 14681da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 14691da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 14701da177e4SLinus Torvalds GFP_ATOMIC)) 14711da177e4SLinus Torvalds return NULL; 14721da177e4SLinus Torvalds } 14731da177e4SLinus Torvalds 147427a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 14751da177e4SLinus Torvalds BUG(); 14761da177e4SLinus Torvalds 14771da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 14781da177e4SLinus Torvalds * size of pulled pages. Superb. 14791da177e4SLinus Torvalds */ 148021dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 14811da177e4SLinus Torvalds goto pull_pages; 14821da177e4SLinus Torvalds 14831da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 14841da177e4SLinus Torvalds eat = delta; 14851da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 14869e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 14879e903e08SEric Dumazet 14889e903e08SEric Dumazet if (size >= eat) 14891da177e4SLinus Torvalds goto pull_pages; 14909e903e08SEric Dumazet eat -= size; 14911da177e4SLinus Torvalds } 14921da177e4SLinus Torvalds 14931da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 14941da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 14951da177e4SLinus Torvalds * but taking into account that pulling is expected to 14961da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 14971da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 14981da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 14991da177e4SLinus Torvalds */ 15001da177e4SLinus Torvalds if (eat) { 15011da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 15021da177e4SLinus Torvalds struct sk_buff *clone = NULL; 15031da177e4SLinus Torvalds struct sk_buff *insp = NULL; 15041da177e4SLinus Torvalds 15051da177e4SLinus Torvalds do { 150609a62660SKris Katterjohn BUG_ON(!list); 15071da177e4SLinus Torvalds 15081da177e4SLinus Torvalds if (list->len <= eat) { 15091da177e4SLinus Torvalds /* Eaten as whole. */ 15101da177e4SLinus Torvalds eat -= list->len; 15111da177e4SLinus Torvalds list = list->next; 15121da177e4SLinus Torvalds insp = list; 15131da177e4SLinus Torvalds } else { 15141da177e4SLinus Torvalds /* Eaten partially. */ 15151da177e4SLinus Torvalds 15161da177e4SLinus Torvalds if (skb_shared(list)) { 15171da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 15181da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 15191da177e4SLinus Torvalds if (!clone) 15201da177e4SLinus Torvalds return NULL; 15211da177e4SLinus Torvalds insp = list->next; 15221da177e4SLinus Torvalds list = clone; 15231da177e4SLinus Torvalds } else { 15241da177e4SLinus Torvalds /* This may be pulled without 15251da177e4SLinus Torvalds * problems. */ 15261da177e4SLinus Torvalds insp = list; 15271da177e4SLinus Torvalds } 15281da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 15291da177e4SLinus Torvalds kfree_skb(clone); 15301da177e4SLinus Torvalds return NULL; 15311da177e4SLinus Torvalds } 15321da177e4SLinus Torvalds break; 15331da177e4SLinus Torvalds } 15341da177e4SLinus Torvalds } while (eat); 15351da177e4SLinus Torvalds 15361da177e4SLinus Torvalds /* Free pulled out fragments. */ 15371da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 15381da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 15391da177e4SLinus Torvalds kfree_skb(list); 15401da177e4SLinus Torvalds } 15411da177e4SLinus Torvalds /* And insert new clone at head. */ 15421da177e4SLinus Torvalds if (clone) { 15431da177e4SLinus Torvalds clone->next = list; 15441da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 15451da177e4SLinus Torvalds } 15461da177e4SLinus Torvalds } 15471da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 15481da177e4SLinus Torvalds 15491da177e4SLinus Torvalds pull_pages: 15501da177e4SLinus Torvalds eat = delta; 15511da177e4SLinus Torvalds k = 0; 15521da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 15539e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 15549e903e08SEric Dumazet 15559e903e08SEric Dumazet if (size <= eat) { 1556ea2ab693SIan Campbell skb_frag_unref(skb, i); 15579e903e08SEric Dumazet eat -= size; 15581da177e4SLinus Torvalds } else { 15591da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 15601da177e4SLinus Torvalds if (eat) { 15611da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 15629e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 15631da177e4SLinus Torvalds eat = 0; 15641da177e4SLinus Torvalds } 15651da177e4SLinus Torvalds k++; 15661da177e4SLinus Torvalds } 15671da177e4SLinus Torvalds } 15681da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds skb->tail += delta; 15711da177e4SLinus Torvalds skb->data_len -= delta; 15721da177e4SLinus Torvalds 157327a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 15741da177e4SLinus Torvalds } 1575b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 15761da177e4SLinus Torvalds 157722019b17SEric Dumazet /** 157822019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 157922019b17SEric Dumazet * @skb: source skb 158022019b17SEric Dumazet * @offset: offset in source 158122019b17SEric Dumazet * @to: destination buffer 158222019b17SEric Dumazet * @len: number of bytes to copy 158322019b17SEric Dumazet * 158422019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 158522019b17SEric Dumazet * destination buffer. 158622019b17SEric Dumazet * 158722019b17SEric Dumazet * CAUTION ! : 158822019b17SEric Dumazet * If its prototype is ever changed, 158922019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 159022019b17SEric Dumazet * since it is called from BPF assembly code. 159122019b17SEric Dumazet */ 15921da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 15931da177e4SLinus Torvalds { 15941a028e50SDavid S. Miller int start = skb_headlen(skb); 1595fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1596fbb398a8SDavid S. Miller int i, copy; 15971da177e4SLinus Torvalds 15981da177e4SLinus Torvalds if (offset > (int)skb->len - len) 15991da177e4SLinus Torvalds goto fault; 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds /* Copy header. */ 16021a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 16031da177e4SLinus Torvalds if (copy > len) 16041da177e4SLinus Torvalds copy = len; 1605d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 16061da177e4SLinus Torvalds if ((len -= copy) == 0) 16071da177e4SLinus Torvalds return 0; 16081da177e4SLinus Torvalds offset += copy; 16091da177e4SLinus Torvalds to += copy; 16101da177e4SLinus Torvalds } 16111da177e4SLinus Torvalds 16121da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16131a028e50SDavid S. Miller int end; 161451c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 16151da177e4SLinus Torvalds 1616547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16171a028e50SDavid S. Miller 161851c56b00SEric Dumazet end = start + skb_frag_size(f); 16191da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 16201da177e4SLinus Torvalds u8 *vaddr; 16211da177e4SLinus Torvalds 16221da177e4SLinus Torvalds if (copy > len) 16231da177e4SLinus Torvalds copy = len; 16241da177e4SLinus Torvalds 162551c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 16261da177e4SLinus Torvalds memcpy(to, 162751c56b00SEric Dumazet vaddr + f->page_offset + offset - start, 162851c56b00SEric Dumazet copy); 162951c56b00SEric Dumazet kunmap_atomic(vaddr); 16301da177e4SLinus Torvalds 16311da177e4SLinus Torvalds if ((len -= copy) == 0) 16321da177e4SLinus Torvalds return 0; 16331da177e4SLinus Torvalds offset += copy; 16341da177e4SLinus Torvalds to += copy; 16351da177e4SLinus Torvalds } 16361a028e50SDavid S. Miller start = end; 16371da177e4SLinus Torvalds } 16381da177e4SLinus Torvalds 1639fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 16401a028e50SDavid S. Miller int end; 16411da177e4SLinus Torvalds 1642547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16431a028e50SDavid S. Miller 1644fbb398a8SDavid S. Miller end = start + frag_iter->len; 16451da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 16461da177e4SLinus Torvalds if (copy > len) 16471da177e4SLinus Torvalds copy = len; 1648fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 16491da177e4SLinus Torvalds goto fault; 16501da177e4SLinus Torvalds if ((len -= copy) == 0) 16511da177e4SLinus Torvalds return 0; 16521da177e4SLinus Torvalds offset += copy; 16531da177e4SLinus Torvalds to += copy; 16541da177e4SLinus Torvalds } 16551a028e50SDavid S. Miller start = end; 16561da177e4SLinus Torvalds } 1657a6686f2fSShirley Ma 16581da177e4SLinus Torvalds if (!len) 16591da177e4SLinus Torvalds return 0; 16601da177e4SLinus Torvalds 16611da177e4SLinus Torvalds fault: 16621da177e4SLinus Torvalds return -EFAULT; 16631da177e4SLinus Torvalds } 1664b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 16651da177e4SLinus Torvalds 16669c55e01cSJens Axboe /* 16679c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 16689c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 16699c55e01cSJens Axboe */ 16709c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 16719c55e01cSJens Axboe { 16728b9d3728SJarek Poplawski put_page(spd->pages[i]); 16738b9d3728SJarek Poplawski } 16749c55e01cSJens Axboe 1675a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 16764fb66994SJarek Poplawski unsigned int *offset, 167718aafc62SEric Dumazet struct sock *sk) 16788b9d3728SJarek Poplawski { 16795640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 16808b9d3728SJarek Poplawski 16815640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 16828b9d3728SJarek Poplawski return NULL; 16834fb66994SJarek Poplawski 16845640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 16854fb66994SJarek Poplawski 16865640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 16875640f768SEric Dumazet page_address(page) + *offset, *len); 16885640f768SEric Dumazet *offset = pfrag->offset; 16895640f768SEric Dumazet pfrag->offset += *len; 16904fb66994SJarek Poplawski 16915640f768SEric Dumazet return pfrag->page; 16929c55e01cSJens Axboe } 16939c55e01cSJens Axboe 169441c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 169541c73a0dSEric Dumazet struct page *page, 169641c73a0dSEric Dumazet unsigned int offset) 169741c73a0dSEric Dumazet { 169841c73a0dSEric Dumazet return spd->nr_pages && 169941c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 170041c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 170141c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 170241c73a0dSEric Dumazet } 170341c73a0dSEric Dumazet 17049c55e01cSJens Axboe /* 17059c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 17069c55e01cSJens Axboe */ 1707a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 170835f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 17094fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 171018aafc62SEric Dumazet bool linear, 17117a67e56fSJarek Poplawski struct sock *sk) 17129c55e01cSJens Axboe { 171341c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1714a108d5f3SDavid S. Miller return true; 17159c55e01cSJens Axboe 17168b9d3728SJarek Poplawski if (linear) { 171718aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 17188b9d3728SJarek Poplawski if (!page) 1719a108d5f3SDavid S. Miller return true; 172041c73a0dSEric Dumazet } 172141c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 172241c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 1723a108d5f3SDavid S. Miller return false; 172441c73a0dSEric Dumazet } 17258b9d3728SJarek Poplawski get_page(page); 17269c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 17274fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 17289c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 17299c55e01cSJens Axboe spd->nr_pages++; 17308b9d3728SJarek Poplawski 1731a108d5f3SDavid S. Miller return false; 17329c55e01cSJens Axboe } 17339c55e01cSJens Axboe 1734a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 17352870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 173618aafc62SEric Dumazet unsigned int *len, 1737d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 173835f3d14dSJens Axboe struct sock *sk, 173935f3d14dSJens Axboe struct pipe_inode_info *pipe) 17409c55e01cSJens Axboe { 17412870c43dSOctavian Purdila if (!*len) 1742a108d5f3SDavid S. Miller return true; 17439c55e01cSJens Axboe 17442870c43dSOctavian Purdila /* skip this segment if already processed */ 17452870c43dSOctavian Purdila if (*off >= plen) { 17462870c43dSOctavian Purdila *off -= plen; 1747a108d5f3SDavid S. Miller return false; 17482870c43dSOctavian Purdila } 17492870c43dSOctavian Purdila 17502870c43dSOctavian Purdila /* ignore any bits we already processed */ 17519ca1b22dSEric Dumazet poff += *off; 17529ca1b22dSEric Dumazet plen -= *off; 17532870c43dSOctavian Purdila *off = 0; 17542870c43dSOctavian Purdila 175518aafc62SEric Dumazet do { 175618aafc62SEric Dumazet unsigned int flen = min(*len, plen); 17572870c43dSOctavian Purdila 175818aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 175918aafc62SEric Dumazet linear, sk)) 1760a108d5f3SDavid S. Miller return true; 176118aafc62SEric Dumazet poff += flen; 176218aafc62SEric Dumazet plen -= flen; 17632870c43dSOctavian Purdila *len -= flen; 176418aafc62SEric Dumazet } while (*len && plen); 17652870c43dSOctavian Purdila 1766a108d5f3SDavid S. Miller return false; 1767db43a282SOctavian Purdila } 17689c55e01cSJens Axboe 17699c55e01cSJens Axboe /* 1770a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 17712870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 17729c55e01cSJens Axboe */ 1773a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 177435f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 177535f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 17762870c43dSOctavian Purdila { 17772870c43dSOctavian Purdila int seg; 17789c55e01cSJens Axboe 17791d0c0b32SEric Dumazet /* map the linear part : 17802996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 17812996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 17822996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 17839c55e01cSJens Axboe */ 17842870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 17852870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 17862870c43dSOctavian Purdila skb_headlen(skb), 178718aafc62SEric Dumazet offset, len, spd, 17883a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 17891d0c0b32SEric Dumazet sk, pipe)) 1790a108d5f3SDavid S. Miller return true; 17919c55e01cSJens Axboe 17929c55e01cSJens Axboe /* 17939c55e01cSJens Axboe * then map the fragments 17949c55e01cSJens Axboe */ 17959c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 17969c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 17979c55e01cSJens Axboe 1798ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 17999e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 180018aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 1801a108d5f3SDavid S. Miller return true; 18029c55e01cSJens Axboe } 18039c55e01cSJens Axboe 1804a108d5f3SDavid S. Miller return false; 18059c55e01cSJens Axboe } 18069c55e01cSJens Axboe 18079c55e01cSJens Axboe /* 18089c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 18099c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 18109c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 18119c55e01cSJens Axboe * handle that cleanly. 18129c55e01cSJens Axboe */ 18138b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 18149c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 18159c55e01cSJens Axboe unsigned int flags) 18169c55e01cSJens Axboe { 181741c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 181841c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 18199c55e01cSJens Axboe struct splice_pipe_desc spd = { 18209c55e01cSJens Axboe .pages = pages, 18219c55e01cSJens Axboe .partial = partial, 1822047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 18239c55e01cSJens Axboe .flags = flags, 182428a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 18259c55e01cSJens Axboe .spd_release = sock_spd_release, 18269c55e01cSJens Axboe }; 1827fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 18287a67e56fSJarek Poplawski struct sock *sk = skb->sk; 182935f3d14dSJens Axboe int ret = 0; 183035f3d14dSJens Axboe 18319c55e01cSJens Axboe /* 18329c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 18339c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 18349c55e01cSJens Axboe */ 183535f3d14dSJens Axboe if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 18369c55e01cSJens Axboe goto done; 18379c55e01cSJens Axboe else if (!tlen) 18389c55e01cSJens Axboe goto done; 18399c55e01cSJens Axboe 18409c55e01cSJens Axboe /* 18419c55e01cSJens Axboe * now see if we have a frag_list to map 18429c55e01cSJens Axboe */ 1843fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 1844fbb398a8SDavid S. Miller if (!tlen) 18459c55e01cSJens Axboe break; 184635f3d14dSJens Axboe if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1847fbb398a8SDavid S. Miller break; 18489c55e01cSJens Axboe } 18499c55e01cSJens Axboe 18509c55e01cSJens Axboe done: 18519c55e01cSJens Axboe if (spd.nr_pages) { 18529c55e01cSJens Axboe /* 18539c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 18549c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 18559c55e01cSJens Axboe * here as compared to sendfile(). We enter here 18569c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 18579c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 18589c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 18599c55e01cSJens Axboe * and networking will grab the socket lock. 18609c55e01cSJens Axboe */ 1861293ad604SOctavian Purdila release_sock(sk); 18629c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1863293ad604SOctavian Purdila lock_sock(sk); 18649c55e01cSJens Axboe } 18659c55e01cSJens Axboe 186635f3d14dSJens Axboe return ret; 18679c55e01cSJens Axboe } 18689c55e01cSJens Axboe 1869357b40a1SHerbert Xu /** 1870357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1871357b40a1SHerbert Xu * @skb: destination buffer 1872357b40a1SHerbert Xu * @offset: offset in destination 1873357b40a1SHerbert Xu * @from: source buffer 1874357b40a1SHerbert Xu * @len: number of bytes to copy 1875357b40a1SHerbert Xu * 1876357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1877357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1878357b40a1SHerbert Xu * traversing fragment lists and such. 1879357b40a1SHerbert Xu */ 1880357b40a1SHerbert Xu 18810c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1882357b40a1SHerbert Xu { 18831a028e50SDavid S. Miller int start = skb_headlen(skb); 1884fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1885fbb398a8SDavid S. Miller int i, copy; 1886357b40a1SHerbert Xu 1887357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1888357b40a1SHerbert Xu goto fault; 1889357b40a1SHerbert Xu 18901a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1891357b40a1SHerbert Xu if (copy > len) 1892357b40a1SHerbert Xu copy = len; 189327d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1894357b40a1SHerbert Xu if ((len -= copy) == 0) 1895357b40a1SHerbert Xu return 0; 1896357b40a1SHerbert Xu offset += copy; 1897357b40a1SHerbert Xu from += copy; 1898357b40a1SHerbert Xu } 1899357b40a1SHerbert Xu 1900357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1901357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 19021a028e50SDavid S. Miller int end; 1903357b40a1SHerbert Xu 1904547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19051a028e50SDavid S. Miller 19069e903e08SEric Dumazet end = start + skb_frag_size(frag); 1907357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1908357b40a1SHerbert Xu u8 *vaddr; 1909357b40a1SHerbert Xu 1910357b40a1SHerbert Xu if (copy > len) 1911357b40a1SHerbert Xu copy = len; 1912357b40a1SHerbert Xu 191351c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 19141a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 19151a028e50SDavid S. Miller from, copy); 191651c56b00SEric Dumazet kunmap_atomic(vaddr); 1917357b40a1SHerbert Xu 1918357b40a1SHerbert Xu if ((len -= copy) == 0) 1919357b40a1SHerbert Xu return 0; 1920357b40a1SHerbert Xu offset += copy; 1921357b40a1SHerbert Xu from += copy; 1922357b40a1SHerbert Xu } 19231a028e50SDavid S. Miller start = end; 1924357b40a1SHerbert Xu } 1925357b40a1SHerbert Xu 1926fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 19271a028e50SDavid S. Miller int end; 1928357b40a1SHerbert Xu 1929547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19301a028e50SDavid S. Miller 1931fbb398a8SDavid S. Miller end = start + frag_iter->len; 1932357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1933357b40a1SHerbert Xu if (copy > len) 1934357b40a1SHerbert Xu copy = len; 1935fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 19361a028e50SDavid S. Miller from, copy)) 1937357b40a1SHerbert Xu goto fault; 1938357b40a1SHerbert Xu if ((len -= copy) == 0) 1939357b40a1SHerbert Xu return 0; 1940357b40a1SHerbert Xu offset += copy; 1941357b40a1SHerbert Xu from += copy; 1942357b40a1SHerbert Xu } 19431a028e50SDavid S. Miller start = end; 1944357b40a1SHerbert Xu } 1945357b40a1SHerbert Xu if (!len) 1946357b40a1SHerbert Xu return 0; 1947357b40a1SHerbert Xu 1948357b40a1SHerbert Xu fault: 1949357b40a1SHerbert Xu return -EFAULT; 1950357b40a1SHerbert Xu } 1951357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1952357b40a1SHerbert Xu 19531da177e4SLinus Torvalds /* Checksum skb data. */ 19542817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 19552817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 19561da177e4SLinus Torvalds { 19571a028e50SDavid S. Miller int start = skb_headlen(skb); 19581a028e50SDavid S. Miller int i, copy = start - offset; 1959fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 19601da177e4SLinus Torvalds int pos = 0; 19611da177e4SLinus Torvalds 19621da177e4SLinus Torvalds /* Checksum header. */ 19631da177e4SLinus Torvalds if (copy > 0) { 19641da177e4SLinus Torvalds if (copy > len) 19651da177e4SLinus Torvalds copy = len; 19662817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 19671da177e4SLinus Torvalds if ((len -= copy) == 0) 19681da177e4SLinus Torvalds return csum; 19691da177e4SLinus Torvalds offset += copy; 19701da177e4SLinus Torvalds pos = copy; 19711da177e4SLinus Torvalds } 19721da177e4SLinus Torvalds 19731da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 19741a028e50SDavid S. Miller int end; 197551c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 19761da177e4SLinus Torvalds 1977547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19781a028e50SDavid S. Miller 197951c56b00SEric Dumazet end = start + skb_frag_size(frag); 19801da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 198144bb9363SAl Viro __wsum csum2; 19821da177e4SLinus Torvalds u8 *vaddr; 19831da177e4SLinus Torvalds 19841da177e4SLinus Torvalds if (copy > len) 19851da177e4SLinus Torvalds copy = len; 198651c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 19872817a336SDaniel Borkmann csum2 = ops->update(vaddr + frag->page_offset + 19881a028e50SDavid S. Miller offset - start, copy, 0); 198951c56b00SEric Dumazet kunmap_atomic(vaddr); 19902817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 19911da177e4SLinus Torvalds if (!(len -= copy)) 19921da177e4SLinus Torvalds return csum; 19931da177e4SLinus Torvalds offset += copy; 19941da177e4SLinus Torvalds pos += copy; 19951da177e4SLinus Torvalds } 19961a028e50SDavid S. Miller start = end; 19971da177e4SLinus Torvalds } 19981da177e4SLinus Torvalds 1999fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 20001a028e50SDavid S. Miller int end; 20011da177e4SLinus Torvalds 2002547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20031a028e50SDavid S. Miller 2004fbb398a8SDavid S. Miller end = start + frag_iter->len; 20051da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 20065f92a738SAl Viro __wsum csum2; 20071da177e4SLinus Torvalds if (copy > len) 20081da177e4SLinus Torvalds copy = len; 20092817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 20102817a336SDaniel Borkmann copy, 0, ops); 20112817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 20121da177e4SLinus Torvalds if ((len -= copy) == 0) 20131da177e4SLinus Torvalds return csum; 20141da177e4SLinus Torvalds offset += copy; 20151da177e4SLinus Torvalds pos += copy; 20161da177e4SLinus Torvalds } 20171a028e50SDavid S. Miller start = end; 20181da177e4SLinus Torvalds } 201909a62660SKris Katterjohn BUG_ON(len); 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds return csum; 20221da177e4SLinus Torvalds } 20232817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 20242817a336SDaniel Borkmann 20252817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 20262817a336SDaniel Borkmann int len, __wsum csum) 20272817a336SDaniel Borkmann { 20282817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2029cea80ea8SDaniel Borkmann .update = csum_partial_ext, 20302817a336SDaniel Borkmann .combine = csum_block_add_ext, 20312817a336SDaniel Borkmann }; 20322817a336SDaniel Borkmann 20332817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 20342817a336SDaniel Borkmann } 2035b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 20361da177e4SLinus Torvalds 20371da177e4SLinus Torvalds /* Both of above in one bottle. */ 20381da177e4SLinus Torvalds 203981d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 204081d77662SAl Viro u8 *to, int len, __wsum csum) 20411da177e4SLinus Torvalds { 20421a028e50SDavid S. Miller int start = skb_headlen(skb); 20431a028e50SDavid S. Miller int i, copy = start - offset; 2044fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 20451da177e4SLinus Torvalds int pos = 0; 20461da177e4SLinus Torvalds 20471da177e4SLinus Torvalds /* Copy header. */ 20481da177e4SLinus Torvalds if (copy > 0) { 20491da177e4SLinus Torvalds if (copy > len) 20501da177e4SLinus Torvalds copy = len; 20511da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 20521da177e4SLinus Torvalds copy, csum); 20531da177e4SLinus Torvalds if ((len -= copy) == 0) 20541da177e4SLinus Torvalds return csum; 20551da177e4SLinus Torvalds offset += copy; 20561da177e4SLinus Torvalds to += copy; 20571da177e4SLinus Torvalds pos = copy; 20581da177e4SLinus Torvalds } 20591da177e4SLinus Torvalds 20601da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20611a028e50SDavid S. Miller int end; 20621da177e4SLinus Torvalds 2063547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20641a028e50SDavid S. Miller 20659e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 20661da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 20675084205fSAl Viro __wsum csum2; 20681da177e4SLinus Torvalds u8 *vaddr; 20691da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 20701da177e4SLinus Torvalds 20711da177e4SLinus Torvalds if (copy > len) 20721da177e4SLinus Torvalds copy = len; 207351c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 20741da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 20751a028e50SDavid S. Miller frag->page_offset + 20761a028e50SDavid S. Miller offset - start, to, 20771a028e50SDavid S. Miller copy, 0); 207851c56b00SEric Dumazet kunmap_atomic(vaddr); 20791da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 20801da177e4SLinus Torvalds if (!(len -= copy)) 20811da177e4SLinus Torvalds return csum; 20821da177e4SLinus Torvalds offset += copy; 20831da177e4SLinus Torvalds to += copy; 20841da177e4SLinus Torvalds pos += copy; 20851da177e4SLinus Torvalds } 20861a028e50SDavid S. Miller start = end; 20871da177e4SLinus Torvalds } 20881da177e4SLinus Torvalds 2089fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 209081d77662SAl Viro __wsum csum2; 20911a028e50SDavid S. Miller int end; 20921da177e4SLinus Torvalds 2093547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20941a028e50SDavid S. Miller 2095fbb398a8SDavid S. Miller end = start + frag_iter->len; 20961da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 20971da177e4SLinus Torvalds if (copy > len) 20981da177e4SLinus Torvalds copy = len; 2099fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 21001a028e50SDavid S. Miller offset - start, 21011da177e4SLinus Torvalds to, copy, 0); 21021da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 21031da177e4SLinus Torvalds if ((len -= copy) == 0) 21041da177e4SLinus Torvalds return csum; 21051da177e4SLinus Torvalds offset += copy; 21061da177e4SLinus Torvalds to += copy; 21071da177e4SLinus Torvalds pos += copy; 21081da177e4SLinus Torvalds } 21091a028e50SDavid S. Miller start = end; 21101da177e4SLinus Torvalds } 211109a62660SKris Katterjohn BUG_ON(len); 21121da177e4SLinus Torvalds return csum; 21131da177e4SLinus Torvalds } 2114b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 21151da177e4SLinus Torvalds 2116af2806f8SThomas Graf /** 2117af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2118af2806f8SThomas Graf * @from: source buffer 2119af2806f8SThomas Graf * 2120af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2121af2806f8SThomas Graf * into skb_zerocopy(). 2122af2806f8SThomas Graf */ 2123af2806f8SThomas Graf unsigned int 2124af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2125af2806f8SThomas Graf { 2126af2806f8SThomas Graf unsigned int hlen = 0; 2127af2806f8SThomas Graf 2128af2806f8SThomas Graf if (!from->head_frag || 2129af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2130af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2131af2806f8SThomas Graf hlen = skb_headlen(from); 2132af2806f8SThomas Graf 2133af2806f8SThomas Graf if (skb_has_frag_list(from)) 2134af2806f8SThomas Graf hlen = from->len; 2135af2806f8SThomas Graf 2136af2806f8SThomas Graf return hlen; 2137af2806f8SThomas Graf } 2138af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2139af2806f8SThomas Graf 2140af2806f8SThomas Graf /** 2141af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2142af2806f8SThomas Graf * @to: destination buffer 21437fceb4deSMasanari Iida * @from: source buffer 2144af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2145af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2146af2806f8SThomas Graf * 2147af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2148af2806f8SThomas Graf * to the frags in the source buffer. 2149af2806f8SThomas Graf * 2150af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2151af2806f8SThomas Graf * headroom in the `to` buffer. 215236d5fe6aSZoltan Kiss * 215336d5fe6aSZoltan Kiss * Return value: 215436d5fe6aSZoltan Kiss * 0: everything is OK 215536d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 215636d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2157af2806f8SThomas Graf */ 215836d5fe6aSZoltan Kiss int 215936d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2160af2806f8SThomas Graf { 2161af2806f8SThomas Graf int i, j = 0; 2162af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 216336d5fe6aSZoltan Kiss int ret; 2164af2806f8SThomas Graf struct page *page; 2165af2806f8SThomas Graf unsigned int offset; 2166af2806f8SThomas Graf 2167af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2168af2806f8SThomas Graf 2169af2806f8SThomas Graf /* dont bother with small payloads */ 217036d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 217136d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2172af2806f8SThomas Graf 2173af2806f8SThomas Graf if (hlen) { 217436d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 217536d5fe6aSZoltan Kiss if (unlikely(ret)) 217636d5fe6aSZoltan Kiss return ret; 2177af2806f8SThomas Graf len -= hlen; 2178af2806f8SThomas Graf } else { 2179af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2180af2806f8SThomas Graf if (plen) { 2181af2806f8SThomas Graf page = virt_to_head_page(from->head); 2182af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2183af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2184af2806f8SThomas Graf get_page(page); 2185af2806f8SThomas Graf j = 1; 2186af2806f8SThomas Graf len -= plen; 2187af2806f8SThomas Graf } 2188af2806f8SThomas Graf } 2189af2806f8SThomas Graf 2190af2806f8SThomas Graf to->truesize += len + plen; 2191af2806f8SThomas Graf to->len += len + plen; 2192af2806f8SThomas Graf to->data_len += len + plen; 2193af2806f8SThomas Graf 219436d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 219536d5fe6aSZoltan Kiss skb_tx_error(from); 219636d5fe6aSZoltan Kiss return -ENOMEM; 219736d5fe6aSZoltan Kiss } 219836d5fe6aSZoltan Kiss 2199af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2200af2806f8SThomas Graf if (!len) 2201af2806f8SThomas Graf break; 2202af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2203af2806f8SThomas Graf skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2204af2806f8SThomas Graf len -= skb_shinfo(to)->frags[j].size; 2205af2806f8SThomas Graf skb_frag_ref(to, j); 2206af2806f8SThomas Graf j++; 2207af2806f8SThomas Graf } 2208af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 220936d5fe6aSZoltan Kiss 221036d5fe6aSZoltan Kiss return 0; 2211af2806f8SThomas Graf } 2212af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2213af2806f8SThomas Graf 22141da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 22151da177e4SLinus Torvalds { 2216d3bc23e7SAl Viro __wsum csum; 22171da177e4SLinus Torvalds long csstart; 22181da177e4SLinus Torvalds 221984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 222055508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 22211da177e4SLinus Torvalds else 22221da177e4SLinus Torvalds csstart = skb_headlen(skb); 22231da177e4SLinus Torvalds 222409a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 22251da177e4SLinus Torvalds 2226d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 22271da177e4SLinus Torvalds 22281da177e4SLinus Torvalds csum = 0; 22291da177e4SLinus Torvalds if (csstart != skb->len) 22301da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 22311da177e4SLinus Torvalds skb->len - csstart, 0); 22321da177e4SLinus Torvalds 223384fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2234ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 22351da177e4SLinus Torvalds 2236d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 22371da177e4SLinus Torvalds } 22381da177e4SLinus Torvalds } 2239b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 22401da177e4SLinus Torvalds 22411da177e4SLinus Torvalds /** 22421da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 22431da177e4SLinus Torvalds * @list: list to dequeue from 22441da177e4SLinus Torvalds * 22451da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 22461da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 22471da177e4SLinus Torvalds * returned or %NULL if the list is empty. 22481da177e4SLinus Torvalds */ 22491da177e4SLinus Torvalds 22501da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 22511da177e4SLinus Torvalds { 22521da177e4SLinus Torvalds unsigned long flags; 22531da177e4SLinus Torvalds struct sk_buff *result; 22541da177e4SLinus Torvalds 22551da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 22561da177e4SLinus Torvalds result = __skb_dequeue(list); 22571da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 22581da177e4SLinus Torvalds return result; 22591da177e4SLinus Torvalds } 2260b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 22611da177e4SLinus Torvalds 22621da177e4SLinus Torvalds /** 22631da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 22641da177e4SLinus Torvalds * @list: list to dequeue from 22651da177e4SLinus Torvalds * 22661da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 22671da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 22681da177e4SLinus Torvalds * returned or %NULL if the list is empty. 22691da177e4SLinus Torvalds */ 22701da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 22711da177e4SLinus Torvalds { 22721da177e4SLinus Torvalds unsigned long flags; 22731da177e4SLinus Torvalds struct sk_buff *result; 22741da177e4SLinus Torvalds 22751da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 22761da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 22771da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 22781da177e4SLinus Torvalds return result; 22791da177e4SLinus Torvalds } 2280b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 22811da177e4SLinus Torvalds 22821da177e4SLinus Torvalds /** 22831da177e4SLinus Torvalds * skb_queue_purge - empty a list 22841da177e4SLinus Torvalds * @list: list to empty 22851da177e4SLinus Torvalds * 22861da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 22871da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 22881da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 22891da177e4SLinus Torvalds */ 22901da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 22911da177e4SLinus Torvalds { 22921da177e4SLinus Torvalds struct sk_buff *skb; 22931da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 22941da177e4SLinus Torvalds kfree_skb(skb); 22951da177e4SLinus Torvalds } 2296b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 22971da177e4SLinus Torvalds 22981da177e4SLinus Torvalds /** 22991da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 23001da177e4SLinus Torvalds * @list: list to use 23011da177e4SLinus Torvalds * @newsk: buffer to queue 23021da177e4SLinus Torvalds * 23031da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 23041da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 23051da177e4SLinus Torvalds * safely. 23061da177e4SLinus Torvalds * 23071da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 23081da177e4SLinus Torvalds */ 23091da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 23101da177e4SLinus Torvalds { 23111da177e4SLinus Torvalds unsigned long flags; 23121da177e4SLinus Torvalds 23131da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23141da177e4SLinus Torvalds __skb_queue_head(list, newsk); 23151da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23161da177e4SLinus Torvalds } 2317b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 23181da177e4SLinus Torvalds 23191da177e4SLinus Torvalds /** 23201da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 23211da177e4SLinus Torvalds * @list: list to use 23221da177e4SLinus Torvalds * @newsk: buffer to queue 23231da177e4SLinus Torvalds * 23241da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 23251da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 23261da177e4SLinus Torvalds * safely. 23271da177e4SLinus Torvalds * 23281da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 23291da177e4SLinus Torvalds */ 23301da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 23311da177e4SLinus Torvalds { 23321da177e4SLinus Torvalds unsigned long flags; 23331da177e4SLinus Torvalds 23341da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23351da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 23361da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23371da177e4SLinus Torvalds } 2338b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 23398728b834SDavid S. Miller 23401da177e4SLinus Torvalds /** 23411da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 23421da177e4SLinus Torvalds * @skb: buffer to remove 23438728b834SDavid S. Miller * @list: list to use 23441da177e4SLinus Torvalds * 23458728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 23468728b834SDavid S. Miller * function is atomic with respect to other list locked calls 23471da177e4SLinus Torvalds * 23488728b834SDavid S. Miller * You must know what list the SKB is on. 23491da177e4SLinus Torvalds */ 23508728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 23511da177e4SLinus Torvalds { 23521da177e4SLinus Torvalds unsigned long flags; 23531da177e4SLinus Torvalds 23541da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23558728b834SDavid S. Miller __skb_unlink(skb, list); 23561da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23571da177e4SLinus Torvalds } 2358b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 23591da177e4SLinus Torvalds 23601da177e4SLinus Torvalds /** 23611da177e4SLinus Torvalds * skb_append - append a buffer 23621da177e4SLinus Torvalds * @old: buffer to insert after 23631da177e4SLinus Torvalds * @newsk: buffer to insert 23648728b834SDavid S. Miller * @list: list to use 23651da177e4SLinus Torvalds * 23661da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 23671da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 23681da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 23691da177e4SLinus Torvalds */ 23708728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 23711da177e4SLinus Torvalds { 23721da177e4SLinus Torvalds unsigned long flags; 23731da177e4SLinus Torvalds 23748728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 23757de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 23768728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 23771da177e4SLinus Torvalds } 2378b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 23791da177e4SLinus Torvalds 23801da177e4SLinus Torvalds /** 23811da177e4SLinus Torvalds * skb_insert - insert a buffer 23821da177e4SLinus Torvalds * @old: buffer to insert before 23831da177e4SLinus Torvalds * @newsk: buffer to insert 23848728b834SDavid S. Miller * @list: list to use 23851da177e4SLinus Torvalds * 23868728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 23878728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 23888728b834SDavid S. Miller * calls. 23898728b834SDavid S. Miller * 23901da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 23911da177e4SLinus Torvalds */ 23928728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 23931da177e4SLinus Torvalds { 23941da177e4SLinus Torvalds unsigned long flags; 23951da177e4SLinus Torvalds 23968728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 23978728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 23988728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 23991da177e4SLinus Torvalds } 2400b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 24011da177e4SLinus Torvalds 24021da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 24031da177e4SLinus Torvalds struct sk_buff* skb1, 24041da177e4SLinus Torvalds const u32 len, const int pos) 24051da177e4SLinus Torvalds { 24061da177e4SLinus Torvalds int i; 24071da177e4SLinus Torvalds 2408d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2409d626f62bSArnaldo Carvalho de Melo pos - len); 24101da177e4SLinus Torvalds /* And move data appendix as is. */ 24111da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 24121da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 24131da177e4SLinus Torvalds 24141da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 24151da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 24161da177e4SLinus Torvalds skb1->data_len = skb->data_len; 24171da177e4SLinus Torvalds skb1->len += skb1->data_len; 24181da177e4SLinus Torvalds skb->data_len = 0; 24191da177e4SLinus Torvalds skb->len = len; 242027a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 24211da177e4SLinus Torvalds } 24221da177e4SLinus Torvalds 24231da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 24241da177e4SLinus Torvalds struct sk_buff* skb1, 24251da177e4SLinus Torvalds const u32 len, int pos) 24261da177e4SLinus Torvalds { 24271da177e4SLinus Torvalds int i, k = 0; 24281da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 24291da177e4SLinus Torvalds 24301da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 24311da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 24321da177e4SLinus Torvalds skb->len = len; 24331da177e4SLinus Torvalds skb->data_len = len - pos; 24341da177e4SLinus Torvalds 24351da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 24369e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 24371da177e4SLinus Torvalds 24381da177e4SLinus Torvalds if (pos + size > len) { 24391da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 24401da177e4SLinus Torvalds 24411da177e4SLinus Torvalds if (pos < len) { 24421da177e4SLinus Torvalds /* Split frag. 24431da177e4SLinus Torvalds * We have two variants in this case: 24441da177e4SLinus Torvalds * 1. Move all the frag to the second 24451da177e4SLinus Torvalds * part, if it is possible. F.e. 24461da177e4SLinus Torvalds * this approach is mandatory for TUX, 24471da177e4SLinus Torvalds * where splitting is expensive. 24481da177e4SLinus Torvalds * 2. Split is accurately. We make this. 24491da177e4SLinus Torvalds */ 2450ea2ab693SIan Campbell skb_frag_ref(skb, i); 24511da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 24529e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 24539e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 24541da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 24551da177e4SLinus Torvalds } 24561da177e4SLinus Torvalds k++; 24571da177e4SLinus Torvalds } else 24581da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 24591da177e4SLinus Torvalds pos += size; 24601da177e4SLinus Torvalds } 24611da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 24621da177e4SLinus Torvalds } 24631da177e4SLinus Torvalds 24641da177e4SLinus Torvalds /** 24651da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 24661da177e4SLinus Torvalds * @skb: the buffer to split 24671da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 24681da177e4SLinus Torvalds * @len: new length for skb 24691da177e4SLinus Torvalds */ 24701da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 24711da177e4SLinus Torvalds { 24721da177e4SLinus Torvalds int pos = skb_headlen(skb); 24731da177e4SLinus Torvalds 247468534c68SAmerigo Wang skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 24751da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 24761da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 24771da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 24781da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 24791da177e4SLinus Torvalds } 2480b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 24811da177e4SLinus Torvalds 24829f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 24839f782db3SIlpo Järvinen * 24849f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 24859f782db3SIlpo Järvinen */ 2486832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2487832d11c5SIlpo Järvinen { 24880ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2489832d11c5SIlpo Järvinen } 2490832d11c5SIlpo Järvinen 2491832d11c5SIlpo Järvinen /** 2492832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2493832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2494832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2495832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2496832d11c5SIlpo Järvinen * 2497832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 249820e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 2499832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2500832d11c5SIlpo Järvinen * 2501832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2502832d11c5SIlpo Järvinen * 2503832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2504832d11c5SIlpo Järvinen * to have non-paged data as well. 2505832d11c5SIlpo Järvinen * 2506832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2507832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2508832d11c5SIlpo Järvinen */ 2509832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2510832d11c5SIlpo Järvinen { 2511832d11c5SIlpo Järvinen int from, to, merge, todo; 2512832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2513832d11c5SIlpo Järvinen 2514832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2515832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2516832d11c5SIlpo Järvinen 2517832d11c5SIlpo Järvinen todo = shiftlen; 2518832d11c5SIlpo Järvinen from = 0; 2519832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2520832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2521832d11c5SIlpo Järvinen 2522832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2523832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2524832d11c5SIlpo Järvinen */ 2525832d11c5SIlpo Järvinen if (!to || 2526ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2527ea2ab693SIan Campbell fragfrom->page_offset)) { 2528832d11c5SIlpo Järvinen merge = -1; 2529832d11c5SIlpo Järvinen } else { 2530832d11c5SIlpo Järvinen merge = to - 1; 2531832d11c5SIlpo Järvinen 25329e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2533832d11c5SIlpo Järvinen if (todo < 0) { 2534832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2535832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2536832d11c5SIlpo Järvinen return 0; 2537832d11c5SIlpo Järvinen 25389f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 25399f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2540832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2541832d11c5SIlpo Järvinen 25429e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 25439e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 2544832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2545832d11c5SIlpo Järvinen 2546832d11c5SIlpo Järvinen goto onlymerged; 2547832d11c5SIlpo Järvinen } 2548832d11c5SIlpo Järvinen 2549832d11c5SIlpo Järvinen from++; 2550832d11c5SIlpo Järvinen } 2551832d11c5SIlpo Järvinen 2552832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2553832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2554832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2555832d11c5SIlpo Järvinen return 0; 2556832d11c5SIlpo Järvinen 2557832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2558832d11c5SIlpo Järvinen return 0; 2559832d11c5SIlpo Järvinen 2560832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2561832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2562832d11c5SIlpo Järvinen return 0; 2563832d11c5SIlpo Järvinen 2564832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2565832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2566832d11c5SIlpo Järvinen 25679e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 2568832d11c5SIlpo Järvinen *fragto = *fragfrom; 25699e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2570832d11c5SIlpo Järvinen from++; 2571832d11c5SIlpo Järvinen to++; 2572832d11c5SIlpo Järvinen 2573832d11c5SIlpo Järvinen } else { 2574ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2575832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2576832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 25779e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 2578832d11c5SIlpo Järvinen 2579832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 25809e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 2581832d11c5SIlpo Järvinen todo = 0; 2582832d11c5SIlpo Järvinen 2583832d11c5SIlpo Järvinen to++; 2584832d11c5SIlpo Järvinen break; 2585832d11c5SIlpo Järvinen } 2586832d11c5SIlpo Järvinen } 2587832d11c5SIlpo Järvinen 2588832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2589832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2590832d11c5SIlpo Järvinen 2591832d11c5SIlpo Järvinen if (merge >= 0) { 2592832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2593832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2594832d11c5SIlpo Järvinen 25959e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2596ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2597832d11c5SIlpo Järvinen } 2598832d11c5SIlpo Järvinen 2599832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2600832d11c5SIlpo Järvinen to = 0; 2601832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2602832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2603832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2604832d11c5SIlpo Järvinen 2605832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2606832d11c5SIlpo Järvinen 2607832d11c5SIlpo Järvinen onlymerged: 2608832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2609832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2610832d11c5SIlpo Järvinen */ 2611832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2612832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2613832d11c5SIlpo Järvinen 2614832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2615832d11c5SIlpo Järvinen skb->len -= shiftlen; 2616832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2617832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2618832d11c5SIlpo Järvinen tgt->len += shiftlen; 2619832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2620832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2621832d11c5SIlpo Järvinen 2622832d11c5SIlpo Järvinen return shiftlen; 2623832d11c5SIlpo Järvinen } 2624832d11c5SIlpo Järvinen 2625677e90edSThomas Graf /** 2626677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2627677e90edSThomas Graf * @skb: the buffer to read 2628677e90edSThomas Graf * @from: lower offset of data to be read 2629677e90edSThomas Graf * @to: upper offset of data to be read 2630677e90edSThomas Graf * @st: state variable 2631677e90edSThomas Graf * 2632677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2633677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2634677e90edSThomas Graf */ 2635677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2636677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2637677e90edSThomas Graf { 2638677e90edSThomas Graf st->lower_offset = from; 2639677e90edSThomas Graf st->upper_offset = to; 2640677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2641677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2642677e90edSThomas Graf st->frag_data = NULL; 2643677e90edSThomas Graf } 2644b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2645677e90edSThomas Graf 2646677e90edSThomas Graf /** 2647677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2648677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2649677e90edSThomas Graf * @data: destination pointer for data to be returned 2650677e90edSThomas Graf * @st: state variable 2651677e90edSThomas Graf * 2652bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 2653677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2654bc32383cSMathias Krause * the head of the data block to @data and returns the length 2655677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2656677e90edSThomas Graf * offset has been reached. 2657677e90edSThomas Graf * 2658677e90edSThomas Graf * The caller is not required to consume all of the data 2659bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 2660677e90edSThomas Graf * of bytes already consumed and the next call to 2661677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2662677e90edSThomas Graf * 266325985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2664e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 2665677e90edSThomas Graf * reads of potentially non linear data. 2666677e90edSThomas Graf * 2667bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2668677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2669677e90edSThomas Graf * a stack for this purpose. 2670677e90edSThomas Graf */ 2671677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2672677e90edSThomas Graf struct skb_seq_state *st) 2673677e90edSThomas Graf { 2674677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2675677e90edSThomas Graf skb_frag_t *frag; 2676677e90edSThomas Graf 2677aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 2678aeb193eaSWedson Almeida Filho if (st->frag_data) { 2679aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 2680aeb193eaSWedson Almeida Filho st->frag_data = NULL; 2681aeb193eaSWedson Almeida Filho } 2682677e90edSThomas Graf return 0; 2683aeb193eaSWedson Almeida Filho } 2684677e90edSThomas Graf 2685677e90edSThomas Graf next_skb: 268695e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2687677e90edSThomas Graf 2688995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 268995e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2690677e90edSThomas Graf return block_limit - abs_offset; 2691677e90edSThomas Graf } 2692677e90edSThomas Graf 2693677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2694677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2695677e90edSThomas Graf 2696677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2697677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 26989e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 2699677e90edSThomas Graf 2700677e90edSThomas Graf if (abs_offset < block_limit) { 2701677e90edSThomas Graf if (!st->frag_data) 270251c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 2703677e90edSThomas Graf 2704677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2705677e90edSThomas Graf (abs_offset - st->stepped_offset); 2706677e90edSThomas Graf 2707677e90edSThomas Graf return block_limit - abs_offset; 2708677e90edSThomas Graf } 2709677e90edSThomas Graf 2710677e90edSThomas Graf if (st->frag_data) { 271151c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2712677e90edSThomas Graf st->frag_data = NULL; 2713677e90edSThomas Graf } 2714677e90edSThomas Graf 2715677e90edSThomas Graf st->frag_idx++; 27169e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 2717677e90edSThomas Graf } 2718677e90edSThomas Graf 27195b5a60daSOlaf Kirch if (st->frag_data) { 272051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 27215b5a60daSOlaf Kirch st->frag_data = NULL; 27225b5a60daSOlaf Kirch } 27235b5a60daSOlaf Kirch 272421dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2725677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 272695e3b24cSHerbert Xu st->frag_idx = 0; 2727677e90edSThomas Graf goto next_skb; 272871b3346dSShyam Iyer } else if (st->cur_skb->next) { 272971b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 273071b3346dSShyam Iyer st->frag_idx = 0; 2731677e90edSThomas Graf goto next_skb; 2732677e90edSThomas Graf } 2733677e90edSThomas Graf 2734677e90edSThomas Graf return 0; 2735677e90edSThomas Graf } 2736b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2737677e90edSThomas Graf 2738677e90edSThomas Graf /** 2739677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2740677e90edSThomas Graf * @st: state variable 2741677e90edSThomas Graf * 2742677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2743677e90edSThomas Graf * returned 0. 2744677e90edSThomas Graf */ 2745677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2746677e90edSThomas Graf { 2747677e90edSThomas Graf if (st->frag_data) 274851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2749677e90edSThomas Graf } 2750b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2751677e90edSThomas Graf 27523fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 27533fc7e8a6SThomas Graf 27543fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 27553fc7e8a6SThomas Graf struct ts_config *conf, 27563fc7e8a6SThomas Graf struct ts_state *state) 27573fc7e8a6SThomas Graf { 27583fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 27593fc7e8a6SThomas Graf } 27603fc7e8a6SThomas Graf 27613fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 27623fc7e8a6SThomas Graf { 27633fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 27643fc7e8a6SThomas Graf } 27653fc7e8a6SThomas Graf 27663fc7e8a6SThomas Graf /** 27673fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 27683fc7e8a6SThomas Graf * @skb: the buffer to look in 27693fc7e8a6SThomas Graf * @from: search offset 27703fc7e8a6SThomas Graf * @to: search limit 27713fc7e8a6SThomas Graf * @config: textsearch configuration 27723fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 27733fc7e8a6SThomas Graf * 27743fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 27753fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 27763fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 27773fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 27783fc7e8a6SThomas Graf */ 27793fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 27803fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 27813fc7e8a6SThomas Graf struct ts_state *state) 27823fc7e8a6SThomas Graf { 2783f72b948dSPhil Oester unsigned int ret; 2784f72b948dSPhil Oester 27853fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 27863fc7e8a6SThomas Graf config->finish = skb_ts_finish; 27873fc7e8a6SThomas Graf 27883fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 27893fc7e8a6SThomas Graf 2790f72b948dSPhil Oester ret = textsearch_find(config, state); 2791f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 27923fc7e8a6SThomas Graf } 2793b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 27943fc7e8a6SThomas Graf 2795e89e9cf5SAnanda Raju /** 27962c53040fSBen Hutchings * skb_append_datato_frags - append the user data to a skb 2797e89e9cf5SAnanda Raju * @sk: sock structure 2798e793c0f7SMasanari Iida * @skb: skb structure to be appended with user data. 2799e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2800e89e9cf5SAnanda Raju * @from: pointer to user message iov 2801e89e9cf5SAnanda Raju * @length: length of the iov message 2802e89e9cf5SAnanda Raju * 2803e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2804e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2805e89e9cf5SAnanda Raju */ 2806e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2807dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2808e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2809e89e9cf5SAnanda Raju void *from, int length) 2810e89e9cf5SAnanda Raju { 2811b2111724SEric Dumazet int frg_cnt = skb_shinfo(skb)->nr_frags; 2812b2111724SEric Dumazet int copy; 2813e89e9cf5SAnanda Raju int offset = 0; 2814e89e9cf5SAnanda Raju int ret; 2815b2111724SEric Dumazet struct page_frag *pfrag = ¤t->task_frag; 2816e89e9cf5SAnanda Raju 2817e89e9cf5SAnanda Raju do { 2818e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2819e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2820b2111724SEric Dumazet return -EMSGSIZE; 2821e89e9cf5SAnanda Raju 2822b2111724SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 2823e89e9cf5SAnanda Raju return -ENOMEM; 2824e89e9cf5SAnanda Raju 2825e89e9cf5SAnanda Raju /* copy the user data to page */ 2826b2111724SEric Dumazet copy = min_t(int, length, pfrag->size - pfrag->offset); 2827e89e9cf5SAnanda Raju 2828b2111724SEric Dumazet ret = getfrag(from, page_address(pfrag->page) + pfrag->offset, 2829e89e9cf5SAnanda Raju offset, copy, 0, skb); 2830e89e9cf5SAnanda Raju if (ret < 0) 2831e89e9cf5SAnanda Raju return -EFAULT; 2832e89e9cf5SAnanda Raju 2833e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2834b2111724SEric Dumazet skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset, 2835b2111724SEric Dumazet copy); 2836b2111724SEric Dumazet frg_cnt++; 2837b2111724SEric Dumazet pfrag->offset += copy; 2838b2111724SEric Dumazet get_page(pfrag->page); 2839b2111724SEric Dumazet 2840b2111724SEric Dumazet skb->truesize += copy; 2841b2111724SEric Dumazet atomic_add(copy, &sk->sk_wmem_alloc); 2842e89e9cf5SAnanda Raju skb->len += copy; 2843e89e9cf5SAnanda Raju skb->data_len += copy; 2844e89e9cf5SAnanda Raju offset += copy; 2845e89e9cf5SAnanda Raju length -= copy; 2846e89e9cf5SAnanda Raju 2847e89e9cf5SAnanda Raju } while (length > 0); 2848e89e9cf5SAnanda Raju 2849e89e9cf5SAnanda Raju return 0; 2850e89e9cf5SAnanda Raju } 2851b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2852e89e9cf5SAnanda Raju 2853cbb042f9SHerbert Xu /** 2854cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2855cbb042f9SHerbert Xu * @skb: buffer to update 2856cbb042f9SHerbert Xu * @len: length of data pulled 2857cbb042f9SHerbert Xu * 2858cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2859fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 286084fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 286184fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 286284fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2863cbb042f9SHerbert Xu */ 2864cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2865cbb042f9SHerbert Xu { 2866cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2867cbb042f9SHerbert Xu skb->len -= len; 2868cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2869cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2870cbb042f9SHerbert Xu return skb->data += len; 2871cbb042f9SHerbert Xu } 2872f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2873f94691acSArnaldo Carvalho de Melo 2874f4c50d99SHerbert Xu /** 2875f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2876df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 2877576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2878f4c50d99SHerbert Xu * 2879f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 28804c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 28814c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2882f4c50d99SHerbert Xu */ 2883df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 2884df5771ffSMichael S. Tsirkin netdev_features_t features) 2885f4c50d99SHerbert Xu { 2886f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2887f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 28881a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 2889df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 2890df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 2891df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 28921fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 2893f4c50d99SHerbert Xu unsigned int offset = doffset; 2894df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 2895f4c50d99SHerbert Xu unsigned int headroom; 2896f4c50d99SHerbert Xu unsigned int len; 2897ec5f0615SPravin B Shelar __be16 proto; 2898ec5f0615SPravin B Shelar bool csum; 289904ed3e74SMichał Mirosław int sg = !!(features & NETIF_F_SG); 2900df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 2901f4c50d99SHerbert Xu int err = -ENOMEM; 2902f4c50d99SHerbert Xu int i = 0; 2903f4c50d99SHerbert Xu int pos; 290453d6471cSVlad Yasevich int dummy; 2905f4c50d99SHerbert Xu 29065882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 290753d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 2908ec5f0615SPravin B Shelar if (unlikely(!proto)) 2909ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 2910ec5f0615SPravin B Shelar 29117e2b10c1STom Herbert csum = !head_skb->encap_hdr_csum && 29127e2b10c1STom Herbert !!can_checksum_protocol(features, proto); 29137e2b10c1STom Herbert 2914df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 2915df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 2916f4c50d99SHerbert Xu 2917f4c50d99SHerbert Xu do { 2918f4c50d99SHerbert Xu struct sk_buff *nskb; 29198cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 2920c8884eddSHerbert Xu int hsize; 2921f4c50d99SHerbert Xu int size; 2922f4c50d99SHerbert Xu 2923df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 2924f4c50d99SHerbert Xu if (len > mss) 2925f4c50d99SHerbert Xu len = mss; 2926f4c50d99SHerbert Xu 2927df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 2928f4c50d99SHerbert Xu if (hsize < 0) 2929f4c50d99SHerbert Xu hsize = 0; 2930c8884eddSHerbert Xu if (hsize > len || !sg) 2931c8884eddSHerbert Xu hsize = len; 2932f4c50d99SHerbert Xu 29331a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 29341a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 29351a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 293689319d38SHerbert Xu 29379d8506ccSHerbert Xu i = 0; 29381a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 29391a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 29401fd819ecSMichael S. Tsirkin frag_skb = list_skb; 29411a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 29429d8506ccSHerbert Xu 29439d8506ccSHerbert Xu while (pos < offset + len) { 29449d8506ccSHerbert Xu BUG_ON(i >= nfrags); 29459d8506ccSHerbert Xu 29464e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 29479d8506ccSHerbert Xu if (pos + size > offset + len) 29489d8506ccSHerbert Xu break; 29499d8506ccSHerbert Xu 29509d8506ccSHerbert Xu i++; 29519d8506ccSHerbert Xu pos += size; 29524e1beba1SMichael S. Tsirkin frag++; 29539d8506ccSHerbert Xu } 29549d8506ccSHerbert Xu 29551a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 29561a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 295789319d38SHerbert Xu 2958f4c50d99SHerbert Xu if (unlikely(!nskb)) 2959f4c50d99SHerbert Xu goto err; 2960f4c50d99SHerbert Xu 29619d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 29629d8506ccSHerbert Xu kfree_skb(nskb); 29639d8506ccSHerbert Xu goto err; 29649d8506ccSHerbert Xu } 29659d8506ccSHerbert Xu 2966ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 296789319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 296889319d38SHerbert Xu kfree_skb(nskb); 296989319d38SHerbert Xu goto err; 297089319d38SHerbert Xu } 297189319d38SHerbert Xu 2972ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 297389319d38SHerbert Xu skb_release_head_state(nskb); 297489319d38SHerbert Xu __skb_push(nskb, doffset); 297589319d38SHerbert Xu } else { 2976c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 2977df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 2978c93bdd0eSMel Gorman NUMA_NO_NODE); 297989319d38SHerbert Xu 298089319d38SHerbert Xu if (unlikely(!nskb)) 298189319d38SHerbert Xu goto err; 298289319d38SHerbert Xu 298389319d38SHerbert Xu skb_reserve(nskb, headroom); 298489319d38SHerbert Xu __skb_put(nskb, doffset); 298589319d38SHerbert Xu } 298689319d38SHerbert Xu 2987f4c50d99SHerbert Xu if (segs) 2988f4c50d99SHerbert Xu tail->next = nskb; 2989f4c50d99SHerbert Xu else 2990f4c50d99SHerbert Xu segs = nskb; 2991f4c50d99SHerbert Xu tail = nskb; 2992f4c50d99SHerbert Xu 2993df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 2994f4c50d99SHerbert Xu 2995030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 2996fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 299768c33163SPravin B Shelar 2998df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 299968c33163SPravin B Shelar nskb->data - tnl_hlen, 300068c33163SPravin B Shelar doffset + tnl_hlen); 300189319d38SHerbert Xu 30029d8506ccSHerbert Xu if (nskb->len == len + doffset) 30031cdbcb79SSimon Horman goto perform_csum_check; 300489319d38SHerbert Xu 3005f4c50d99SHerbert Xu if (!sg) { 30066f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 3007df5771ffSMichael S. Tsirkin nskb->csum = skb_copy_and_csum_bits(head_skb, offset, 3008f4c50d99SHerbert Xu skb_put(nskb, len), 3009f4c50d99SHerbert Xu len, 0); 30107e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3011de843723STom Herbert skb_headroom(nskb) + doffset; 3012f4c50d99SHerbert Xu continue; 3013f4c50d99SHerbert Xu } 3014f4c50d99SHerbert Xu 30158cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3016f4c50d99SHerbert Xu 3017df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3018d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3019f4c50d99SHerbert Xu 3020df5771ffSMichael S. Tsirkin skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags & 3021df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3022cef401deSEric Dumazet 30239d8506ccSHerbert Xu while (pos < offset + len) { 30249d8506ccSHerbert Xu if (i >= nfrags) { 30251a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb)); 30269d8506ccSHerbert Xu 30279d8506ccSHerbert Xu i = 0; 30281a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 30291a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 30301fd819ecSMichael S. Tsirkin frag_skb = list_skb; 30319d8506ccSHerbert Xu 30329d8506ccSHerbert Xu BUG_ON(!nfrags); 30339d8506ccSHerbert Xu 30341a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 30359d8506ccSHerbert Xu } 30369d8506ccSHerbert Xu 30379d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 30389d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 30399d8506ccSHerbert Xu net_warn_ratelimited( 30409d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 30419d8506ccSHerbert Xu pos, mss); 30429d8506ccSHerbert Xu goto err; 30439d8506ccSHerbert Xu } 30449d8506ccSHerbert Xu 30451fd819ecSMichael S. Tsirkin if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC))) 30461fd819ecSMichael S. Tsirkin goto err; 30471fd819ecSMichael S. Tsirkin 30484e1beba1SMichael S. Tsirkin *nskb_frag = *frag; 30498cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 30508cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3051f4c50d99SHerbert Xu 3052f4c50d99SHerbert Xu if (pos < offset) { 30538cb19905SMichael S. Tsirkin nskb_frag->page_offset += offset - pos; 30548cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3055f4c50d99SHerbert Xu } 3056f4c50d99SHerbert Xu 305789319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3058f4c50d99SHerbert Xu 3059f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3060f4c50d99SHerbert Xu i++; 30614e1beba1SMichael S. Tsirkin frag++; 3062f4c50d99SHerbert Xu pos += size; 3063f4c50d99SHerbert Xu } else { 30648cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 306589319d38SHerbert Xu goto skip_fraglist; 3066f4c50d99SHerbert Xu } 3067f4c50d99SHerbert Xu 30688cb19905SMichael S. Tsirkin nskb_frag++; 3069f4c50d99SHerbert Xu } 3070f4c50d99SHerbert Xu 307189319d38SHerbert Xu skip_fraglist: 3072f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3073f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3074f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3075ec5f0615SPravin B Shelar 30761cdbcb79SSimon Horman perform_csum_check: 3077ec5f0615SPravin B Shelar if (!csum) { 3078ec5f0615SPravin B Shelar nskb->csum = skb_checksum(nskb, doffset, 3079ec5f0615SPravin B Shelar nskb->len - doffset, 0); 3080ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 30817e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 30827e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3083ec5f0615SPravin B Shelar } 3084df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3085f4c50d99SHerbert Xu 3086bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3087bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3088bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3089bec3cfdcSEric Dumazet */ 3090bec3cfdcSEric Dumazet segs->prev = tail; 3091f4c50d99SHerbert Xu return segs; 3092f4c50d99SHerbert Xu 3093f4c50d99SHerbert Xu err: 3094289dccbeSEric Dumazet kfree_skb_list(segs); 3095f4c50d99SHerbert Xu return ERR_PTR(err); 3096f4c50d99SHerbert Xu } 3097f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3098f4c50d99SHerbert Xu 309971d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 310071d93b39SHerbert Xu { 31018a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 310267147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 310367147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 31048a29111cSEric Dumazet struct sk_buff *nskb, *lp, *p = *head; 31058a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 3106715dc1f3SEric Dumazet unsigned int delta_truesize; 31078a29111cSEric Dumazet unsigned int headroom; 310871d93b39SHerbert Xu 31098a29111cSEric Dumazet if (unlikely(p->len + len >= 65536)) 311071d93b39SHerbert Xu return -E2BIG; 311171d93b39SHerbert Xu 311229e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 31138a29111cSEric Dumazet pinfo = skb_shinfo(lp); 31148a29111cSEric Dumazet 31158a29111cSEric Dumazet if (headlen <= offset) { 311642da6994SHerbert Xu skb_frag_t *frag; 311766e92fcfSHerbert Xu skb_frag_t *frag2; 31189aaa156cSHerbert Xu int i = skbinfo->nr_frags; 31199aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 312042da6994SHerbert Xu 312166e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 31228a29111cSEric Dumazet goto merge; 312381705ad1SHerbert Xu 31248a29111cSEric Dumazet offset -= headlen; 31259aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 31269aaa156cSHerbert Xu skbinfo->nr_frags = 0; 3127f5572068SHerbert Xu 31289aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 31299aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 313066e92fcfSHerbert Xu do { 313166e92fcfSHerbert Xu *--frag = *--frag2; 313266e92fcfSHerbert Xu } while (--i); 313366e92fcfSHerbert Xu 313466e92fcfSHerbert Xu frag->page_offset += offset; 31359e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 313666e92fcfSHerbert Xu 3137715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3138ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3139ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3140715dc1f3SEric Dumazet 3141f5572068SHerbert Xu skb->truesize -= skb->data_len; 3142f5572068SHerbert Xu skb->len -= skb->data_len; 3143f5572068SHerbert Xu skb->data_len = 0; 3144f5572068SHerbert Xu 3145715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 31465d38a079SHerbert Xu goto done; 3147d7e8883cSEric Dumazet } else if (skb->head_frag) { 3148d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3149d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3150d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3151d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3152d7e8883cSEric Dumazet unsigned int first_offset; 3153d7e8883cSEric Dumazet 3154d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 31558a29111cSEric Dumazet goto merge; 3156d7e8883cSEric Dumazet 3157d7e8883cSEric Dumazet first_offset = skb->data - 3158d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3159d7e8883cSEric Dumazet offset; 3160d7e8883cSEric Dumazet 3161d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3162d7e8883cSEric Dumazet 3163d7e8883cSEric Dumazet frag->page.p = page; 3164d7e8883cSEric Dumazet frag->page_offset = first_offset; 3165d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3166d7e8883cSEric Dumazet 3167d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3168d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3169d7e8883cSEric Dumazet 3170715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3171d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3172d7e8883cSEric Dumazet goto done; 31738a29111cSEric Dumazet } 317473d3fe6dSEric Dumazet /* switch back to head shinfo */ 317573d3fe6dSEric Dumazet pinfo = skb_shinfo(p); 317673d3fe6dSEric Dumazet 31778a29111cSEric Dumazet if (pinfo->frag_list) 31788a29111cSEric Dumazet goto merge; 31798a29111cSEric Dumazet if (skb_gro_len(p) != pinfo->gso_size) 318069c0cab1SHerbert Xu return -E2BIG; 318171d93b39SHerbert Xu 318271d93b39SHerbert Xu headroom = skb_headroom(p); 31833d3be433SEric Dumazet nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); 318471d93b39SHerbert Xu if (unlikely(!nskb)) 318571d93b39SHerbert Xu return -ENOMEM; 318671d93b39SHerbert Xu 318771d93b39SHerbert Xu __copy_skb_header(nskb, p); 318871d93b39SHerbert Xu nskb->mac_len = p->mac_len; 318971d93b39SHerbert Xu 319071d93b39SHerbert Xu skb_reserve(nskb, headroom); 319186911732SHerbert Xu __skb_put(nskb, skb_gro_offset(p)); 319271d93b39SHerbert Xu 319386911732SHerbert Xu skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 319471d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 319571d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 319671d93b39SHerbert Xu 319786911732SHerbert Xu __skb_pull(p, skb_gro_offset(p)); 319886911732SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), 319986911732SHerbert Xu p->data - skb_mac_header(p)); 320071d93b39SHerbert Xu 320171d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 32029aaa156cSHerbert Xu skb_shinfo(nskb)->gso_size = pinfo->gso_size; 3203622e0ca1SHerbert Xu pinfo->gso_size = 0; 3204f4a775d1SEric Dumazet __skb_header_release(p); 3205c3c7c254SEric Dumazet NAPI_GRO_CB(nskb)->last = p; 320671d93b39SHerbert Xu 320771d93b39SHerbert Xu nskb->data_len += p->len; 3208de8261c2SEric Dumazet nskb->truesize += p->truesize; 320971d93b39SHerbert Xu nskb->len += p->len; 321071d93b39SHerbert Xu 321171d93b39SHerbert Xu *head = nskb; 321271d93b39SHerbert Xu nskb->next = p->next; 321371d93b39SHerbert Xu p->next = NULL; 321471d93b39SHerbert Xu 321571d93b39SHerbert Xu p = nskb; 321671d93b39SHerbert Xu 321771d93b39SHerbert Xu merge: 3218715dc1f3SEric Dumazet delta_truesize = skb->truesize; 321967147ba9SHerbert Xu if (offset > headlen) { 3220d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3221d1dc7abfSMichal Schmidt 3222d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 32239e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3224d1dc7abfSMichal Schmidt skb->data_len -= eat; 3225d1dc7abfSMichal Schmidt skb->len -= eat; 322667147ba9SHerbert Xu offset = headlen; 322756035022SHerbert Xu } 322856035022SHerbert Xu 322967147ba9SHerbert Xu __skb_pull(skb, offset); 323056035022SHerbert Xu 323129e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 32328a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 32338a29111cSEric Dumazet else 3234c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3235c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 3236f4a775d1SEric Dumazet __skb_header_release(skb); 32378a29111cSEric Dumazet lp = p; 323871d93b39SHerbert Xu 32395d38a079SHerbert Xu done: 32405d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 324137fe4732SHerbert Xu p->data_len += len; 3242715dc1f3SEric Dumazet p->truesize += delta_truesize; 324337fe4732SHerbert Xu p->len += len; 32448a29111cSEric Dumazet if (lp != p) { 32458a29111cSEric Dumazet lp->data_len += len; 32468a29111cSEric Dumazet lp->truesize += delta_truesize; 32478a29111cSEric Dumazet lp->len += len; 32488a29111cSEric Dumazet } 324971d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 325071d93b39SHerbert Xu return 0; 325171d93b39SHerbert Xu } 325271d93b39SHerbert Xu 32531da177e4SLinus Torvalds void __init skb_init(void) 32541da177e4SLinus Torvalds { 32551da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 32561da177e4SLinus Torvalds sizeof(struct sk_buff), 32571da177e4SLinus Torvalds 0, 3258e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 325920c2df83SPaul Mundt NULL); 3260d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3261d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 3262d179cd12SDavid S. Miller 0, 3263e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 326420c2df83SPaul Mundt NULL); 32651da177e4SLinus Torvalds } 32661da177e4SLinus Torvalds 3267716ea3a7SDavid Howells /** 3268716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3269716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 3270716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 3271716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 3272716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 3273716ea3a7SDavid Howells * 3274716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 3275716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 3276716ea3a7SDavid Howells */ 327751c739d1SDavid S. Miller static int 327851c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3279716ea3a7SDavid Howells { 32801a028e50SDavid S. Miller int start = skb_headlen(skb); 32811a028e50SDavid S. Miller int i, copy = start - offset; 3282fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3283716ea3a7SDavid Howells int elt = 0; 3284716ea3a7SDavid Howells 3285716ea3a7SDavid Howells if (copy > 0) { 3286716ea3a7SDavid Howells if (copy > len) 3287716ea3a7SDavid Howells copy = len; 3288642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 3289716ea3a7SDavid Howells elt++; 3290716ea3a7SDavid Howells if ((len -= copy) == 0) 3291716ea3a7SDavid Howells return elt; 3292716ea3a7SDavid Howells offset += copy; 3293716ea3a7SDavid Howells } 3294716ea3a7SDavid Howells 3295716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 32961a028e50SDavid S. Miller int end; 3297716ea3a7SDavid Howells 3298547b792cSIlpo Järvinen WARN_ON(start > offset + len); 32991a028e50SDavid S. Miller 33009e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3301716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3302716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3303716ea3a7SDavid Howells 3304716ea3a7SDavid Howells if (copy > len) 3305716ea3a7SDavid Howells copy = len; 3306ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3307642f1490SJens Axboe frag->page_offset+offset-start); 3308716ea3a7SDavid Howells elt++; 3309716ea3a7SDavid Howells if (!(len -= copy)) 3310716ea3a7SDavid Howells return elt; 3311716ea3a7SDavid Howells offset += copy; 3312716ea3a7SDavid Howells } 33131a028e50SDavid S. Miller start = end; 3314716ea3a7SDavid Howells } 3315716ea3a7SDavid Howells 3316fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 33171a028e50SDavid S. Miller int end; 3318716ea3a7SDavid Howells 3319547b792cSIlpo Järvinen WARN_ON(start > offset + len); 33201a028e50SDavid S. Miller 3321fbb398a8SDavid S. Miller end = start + frag_iter->len; 3322716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3323716ea3a7SDavid Howells if (copy > len) 3324716ea3a7SDavid Howells copy = len; 3325fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 332651c739d1SDavid S. Miller copy); 3327716ea3a7SDavid Howells if ((len -= copy) == 0) 3328716ea3a7SDavid Howells return elt; 3329716ea3a7SDavid Howells offset += copy; 3330716ea3a7SDavid Howells } 33311a028e50SDavid S. Miller start = end; 3332716ea3a7SDavid Howells } 3333716ea3a7SDavid Howells BUG_ON(len); 3334716ea3a7SDavid Howells return elt; 3335716ea3a7SDavid Howells } 3336716ea3a7SDavid Howells 333725a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 333825a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 333925a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 334025a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 334125a91d8dSFan Du * 334225a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 334325a91d8dSFan Du * 1. sg_init_table 334425a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 334525a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 334625a91d8dSFan Du * 334725a91d8dSFan Du * This is equivalent to: 334825a91d8dSFan Du * 1. sg_init_table 334925a91d8dSFan Du * 2. skb_to_sgvec(payload1) 335025a91d8dSFan Du * 3. sg_unmark_end 335125a91d8dSFan Du * 4. skb_to_sgvec(payload2) 335225a91d8dSFan Du * 335325a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 335425a91d8dSFan Du * is more preferable. 335525a91d8dSFan Du */ 335625a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 335725a91d8dSFan Du int offset, int len) 335825a91d8dSFan Du { 335925a91d8dSFan Du return __skb_to_sgvec(skb, sg, offset, len); 336025a91d8dSFan Du } 336125a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 336225a91d8dSFan Du 336351c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 336451c739d1SDavid S. Miller { 336551c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 336651c739d1SDavid S. Miller 3367c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 336851c739d1SDavid S. Miller 336951c739d1SDavid S. Miller return nsg; 337051c739d1SDavid S. Miller } 3371b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 337251c739d1SDavid S. Miller 3373716ea3a7SDavid Howells /** 3374716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 3375716ea3a7SDavid Howells * @skb: The socket buffer to check. 3376716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 3377716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 3378716ea3a7SDavid Howells * 3379716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 3380716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 3381716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 3382716ea3a7SDavid Howells * 3383716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 3384716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 3385716ea3a7SDavid Howells * set to point to the skb in which this space begins. 3386716ea3a7SDavid Howells * 3387716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3388716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3389716ea3a7SDavid Howells */ 3390716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3391716ea3a7SDavid Howells { 3392716ea3a7SDavid Howells int copyflag; 3393716ea3a7SDavid Howells int elt; 3394716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3395716ea3a7SDavid Howells 3396716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3397716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3398716ea3a7SDavid Howells * at the moment even if they are anonymous). 3399716ea3a7SDavid Howells */ 3400716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3401716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3402716ea3a7SDavid Howells return -ENOMEM; 3403716ea3a7SDavid Howells 3404716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 340521dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3406716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3407716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3408716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3409716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3410716ea3a7SDavid Howells 3411716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3412716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3413716ea3a7SDavid Howells return -ENOMEM; 3414716ea3a7SDavid Howells 3415716ea3a7SDavid Howells /* Voila! */ 3416716ea3a7SDavid Howells *trailer = skb; 3417716ea3a7SDavid Howells return 1; 3418716ea3a7SDavid Howells } 3419716ea3a7SDavid Howells 3420716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3421716ea3a7SDavid Howells 3422716ea3a7SDavid Howells elt = 1; 3423716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3424716ea3a7SDavid Howells copyflag = 0; 3425716ea3a7SDavid Howells 3426716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3427716ea3a7SDavid Howells int ntail = 0; 3428716ea3a7SDavid Howells 3429716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3430716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3431716ea3a7SDavid Howells * after it. */ 3432716ea3a7SDavid Howells 3433716ea3a7SDavid Howells if (skb_shared(skb1)) 3434716ea3a7SDavid Howells copyflag = 1; 3435716ea3a7SDavid Howells 3436716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3437716ea3a7SDavid Howells 3438716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3439716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 344021dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3441716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3442716ea3a7SDavid Howells ntail = tailbits + 128; 3443716ea3a7SDavid Howells } 3444716ea3a7SDavid Howells 3445716ea3a7SDavid Howells if (copyflag || 3446716ea3a7SDavid Howells skb_cloned(skb1) || 3447716ea3a7SDavid Howells ntail || 3448716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 344921dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3450716ea3a7SDavid Howells struct sk_buff *skb2; 3451716ea3a7SDavid Howells 3452716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3453716ea3a7SDavid Howells if (ntail == 0) 3454716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3455716ea3a7SDavid Howells else 3456716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3457716ea3a7SDavid Howells skb_headroom(skb1), 3458716ea3a7SDavid Howells ntail, 3459716ea3a7SDavid Howells GFP_ATOMIC); 3460716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3461716ea3a7SDavid Howells return -ENOMEM; 3462716ea3a7SDavid Howells 3463716ea3a7SDavid Howells if (skb1->sk) 3464716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3465716ea3a7SDavid Howells 3466716ea3a7SDavid Howells /* Looking around. Are we still alive? 3467716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3468716ea3a7SDavid Howells 3469716ea3a7SDavid Howells skb2->next = skb1->next; 3470716ea3a7SDavid Howells *skb_p = skb2; 3471716ea3a7SDavid Howells kfree_skb(skb1); 3472716ea3a7SDavid Howells skb1 = skb2; 3473716ea3a7SDavid Howells } 3474716ea3a7SDavid Howells elt++; 3475716ea3a7SDavid Howells *trailer = skb1; 3476716ea3a7SDavid Howells skb_p = &skb1->next; 3477716ea3a7SDavid Howells } 3478716ea3a7SDavid Howells 3479716ea3a7SDavid Howells return elt; 3480716ea3a7SDavid Howells } 3481b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3482716ea3a7SDavid Howells 3483b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3484b1faf566SEric Dumazet { 3485b1faf566SEric Dumazet struct sock *sk = skb->sk; 3486b1faf566SEric Dumazet 3487b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3488b1faf566SEric Dumazet } 3489b1faf566SEric Dumazet 3490b1faf566SEric Dumazet /* 3491b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3492b1faf566SEric Dumazet */ 3493b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3494b1faf566SEric Dumazet { 3495b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 349695c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 3497b1faf566SEric Dumazet return -ENOMEM; 3498b1faf566SEric Dumazet 3499b1faf566SEric Dumazet skb_orphan(skb); 3500b1faf566SEric Dumazet skb->sk = sk; 3501b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3502b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3503b1faf566SEric Dumazet 3504abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3505abb57ea4SEric Dumazet skb_dst_force(skb); 3506abb57ea4SEric Dumazet 3507b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3508b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3509676d2369SDavid S. Miller sk->sk_data_ready(sk); 3510b1faf566SEric Dumazet return 0; 3511b1faf566SEric Dumazet } 3512b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3513b1faf566SEric Dumazet 3514364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 3515364a9e93SWillem de Bruijn { 3516364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 3517364a9e93SWillem de Bruijn struct sk_buff *skb, *skb_next; 3518364a9e93SWillem de Bruijn int err = 0; 3519364a9e93SWillem de Bruijn 3520364a9e93SWillem de Bruijn spin_lock_bh(&q->lock); 3521364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 3522364a9e93SWillem de Bruijn if (skb && (skb_next = skb_peek(q))) 3523364a9e93SWillem de Bruijn err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 3524364a9e93SWillem de Bruijn spin_unlock_bh(&q->lock); 3525364a9e93SWillem de Bruijn 3526364a9e93SWillem de Bruijn sk->sk_err = err; 3527364a9e93SWillem de Bruijn if (err) 3528364a9e93SWillem de Bruijn sk->sk_error_report(sk); 3529364a9e93SWillem de Bruijn 3530364a9e93SWillem de Bruijn return skb; 3531364a9e93SWillem de Bruijn } 3532364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 3533364a9e93SWillem de Bruijn 3534cab41c47SAlexander Duyck /** 3535cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 3536cab41c47SAlexander Duyck * @skb: the skb to clone 3537cab41c47SAlexander Duyck * 3538cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 3539cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 3540cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 3541cab41c47SAlexander Duyck * 3542cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 3543cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 3544cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 3545cab41c47SAlexander Duyck * the sk_error_queue. 3546cab41c47SAlexander Duyck */ 354762bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 354862bccb8cSAlexander Duyck { 354962bccb8cSAlexander Duyck struct sock *sk = skb->sk; 355062bccb8cSAlexander Duyck struct sk_buff *clone; 355162bccb8cSAlexander Duyck 355262bccb8cSAlexander Duyck if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt)) 355362bccb8cSAlexander Duyck return NULL; 355462bccb8cSAlexander Duyck 355562bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 355662bccb8cSAlexander Duyck if (!clone) { 355762bccb8cSAlexander Duyck sock_put(sk); 355862bccb8cSAlexander Duyck return NULL; 355962bccb8cSAlexander Duyck } 356062bccb8cSAlexander Duyck 356162bccb8cSAlexander Duyck clone->sk = sk; 356262bccb8cSAlexander Duyck clone->destructor = sock_efree; 356362bccb8cSAlexander Duyck 356462bccb8cSAlexander Duyck return clone; 356562bccb8cSAlexander Duyck } 356662bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 356762bccb8cSAlexander Duyck 356837846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 356937846ef0SAlexander Duyck struct sock *sk, 357037846ef0SAlexander Duyck int tstype) 3571ac45f602SPatrick Ohly { 3572ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3573ac45f602SPatrick Ohly int err; 3574ac45f602SPatrick Ohly 3575ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3576ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3577ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3578ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 3579e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 35804ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 358109c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 35824ed2d765SWillem de Bruijn if (sk->sk_protocol == IPPROTO_TCP) 35834ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 35844ed2d765SWillem de Bruijn } 358529030374SEric Dumazet 3586ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 358729030374SEric Dumazet 3588ac45f602SPatrick Ohly if (err) 3589ac45f602SPatrick Ohly kfree_skb(skb); 3590ac45f602SPatrick Ohly } 359137846ef0SAlexander Duyck 359237846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 359337846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 359437846ef0SAlexander Duyck { 359537846ef0SAlexander Duyck struct sock *sk = skb->sk; 359637846ef0SAlexander Duyck 359762bccb8cSAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 359862bccb8cSAlexander Duyck sock_hold(sk); 359937846ef0SAlexander Duyck 360037846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 360137846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND); 360237846ef0SAlexander Duyck 360337846ef0SAlexander Duyck sock_put(sk); 360437846ef0SAlexander Duyck } 360537846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 360637846ef0SAlexander Duyck 360737846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 360837846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 360937846ef0SAlexander Duyck struct sock *sk, int tstype) 361037846ef0SAlexander Duyck { 361137846ef0SAlexander Duyck struct sk_buff *skb; 361237846ef0SAlexander Duyck 361337846ef0SAlexander Duyck if (!sk) 361437846ef0SAlexander Duyck return; 361537846ef0SAlexander Duyck 361637846ef0SAlexander Duyck if (hwtstamps) 361737846ef0SAlexander Duyck *skb_hwtstamps(orig_skb) = *hwtstamps; 361837846ef0SAlexander Duyck else 361937846ef0SAlexander Duyck orig_skb->tstamp = ktime_get_real(); 362037846ef0SAlexander Duyck 362137846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 362237846ef0SAlexander Duyck if (!skb) 362337846ef0SAlexander Duyck return; 362437846ef0SAlexander Duyck 362537846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, tstype); 362637846ef0SAlexander Duyck } 3627e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 3628e7fd2885SWillem de Bruijn 3629e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 3630e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 3631e7fd2885SWillem de Bruijn { 3632e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 3633e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 3634e7fd2885SWillem de Bruijn } 3635ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3636ac45f602SPatrick Ohly 36376e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 36386e3e939fSJohannes Berg { 36396e3e939fSJohannes Berg struct sock *sk = skb->sk; 36406e3e939fSJohannes Berg struct sock_exterr_skb *serr; 36416e3e939fSJohannes Berg int err; 36426e3e939fSJohannes Berg 36436e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 36446e3e939fSJohannes Berg skb->wifi_acked = acked; 36456e3e939fSJohannes Berg 36466e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 36476e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 36486e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 36496e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 36506e3e939fSJohannes Berg 3651bf7fa551SAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 3652bf7fa551SAlexander Duyck sock_hold(sk); 3653bf7fa551SAlexander Duyck 36546e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 36556e3e939fSJohannes Berg if (err) 36566e3e939fSJohannes Berg kfree_skb(skb); 3657bf7fa551SAlexander Duyck 3658bf7fa551SAlexander Duyck sock_put(sk); 36596e3e939fSJohannes Berg } 36606e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 36616e3e939fSJohannes Berg 3662ac45f602SPatrick Ohly 3663f35d9d8aSRusty Russell /** 3664f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3665f35d9d8aSRusty Russell * @skb: the skb to set 3666f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3667f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3668f35d9d8aSRusty Russell * 3669f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3670f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3671f35d9d8aSRusty Russell * 3672f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3673f35d9d8aSRusty Russell * returns false you should drop the packet. 3674f35d9d8aSRusty Russell */ 3675f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3676f35d9d8aSRusty Russell { 36775ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 36785ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3679e87cc472SJoe Perches net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 36805ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3681f35d9d8aSRusty Russell return false; 3682f35d9d8aSRusty Russell } 3683f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3684f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3685f35d9d8aSRusty Russell skb->csum_offset = off; 3686e5d5decaSJason Wang skb_set_transport_header(skb, start); 3687f35d9d8aSRusty Russell return true; 3688f35d9d8aSRusty Russell } 3689b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3690f35d9d8aSRusty Russell 3691ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 3692ed1f50c3SPaul Durrant unsigned int max) 3693ed1f50c3SPaul Durrant { 3694ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 3695ed1f50c3SPaul Durrant return 0; 3696ed1f50c3SPaul Durrant 3697ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 3698ed1f50c3SPaul Durrant * won't need to do it again. 3699ed1f50c3SPaul Durrant */ 3700ed1f50c3SPaul Durrant if (max > skb->len) 3701ed1f50c3SPaul Durrant max = skb->len; 3702ed1f50c3SPaul Durrant 3703ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 3704ed1f50c3SPaul Durrant return -ENOMEM; 3705ed1f50c3SPaul Durrant 3706ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 3707ed1f50c3SPaul Durrant return -EPROTO; 3708ed1f50c3SPaul Durrant 3709ed1f50c3SPaul Durrant return 0; 3710ed1f50c3SPaul Durrant } 3711ed1f50c3SPaul Durrant 3712f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 3713f9708b43SJan Beulich 3714f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 3715f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 3716f9708b43SJan Beulich unsigned int off) 3717f9708b43SJan Beulich { 3718f9708b43SJan Beulich switch (proto) { 3719f9708b43SJan Beulich int err; 3720f9708b43SJan Beulich 3721f9708b43SJan Beulich case IPPROTO_TCP: 3722f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 3723f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 3724f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3725f9708b43SJan Beulich offsetof(struct tcphdr, 3726f9708b43SJan Beulich check))) 3727f9708b43SJan Beulich err = -EPROTO; 3728f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 3729f9708b43SJan Beulich 3730f9708b43SJan Beulich case IPPROTO_UDP: 3731f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 3732f9708b43SJan Beulich off + sizeof(struct udphdr)); 3733f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3734f9708b43SJan Beulich offsetof(struct udphdr, 3735f9708b43SJan Beulich check))) 3736f9708b43SJan Beulich err = -EPROTO; 3737f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 3738f9708b43SJan Beulich } 3739f9708b43SJan Beulich 3740f9708b43SJan Beulich return ERR_PTR(-EPROTO); 3741f9708b43SJan Beulich } 3742f9708b43SJan Beulich 3743ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 3744ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 3745ed1f50c3SPaul Durrant */ 3746ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 3747ed1f50c3SPaul Durrant 3748f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 3749ed1f50c3SPaul Durrant { 3750ed1f50c3SPaul Durrant unsigned int off; 3751ed1f50c3SPaul Durrant bool fragment; 3752f9708b43SJan Beulich __sum16 *csum; 3753ed1f50c3SPaul Durrant int err; 3754ed1f50c3SPaul Durrant 3755ed1f50c3SPaul Durrant fragment = false; 3756ed1f50c3SPaul Durrant 3757ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3758ed1f50c3SPaul Durrant sizeof(struct iphdr), 3759ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 3760ed1f50c3SPaul Durrant if (err < 0) 3761ed1f50c3SPaul Durrant goto out; 3762ed1f50c3SPaul Durrant 3763ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 3764ed1f50c3SPaul Durrant fragment = true; 3765ed1f50c3SPaul Durrant 3766ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 3767ed1f50c3SPaul Durrant 3768ed1f50c3SPaul Durrant err = -EPROTO; 3769ed1f50c3SPaul Durrant 3770ed1f50c3SPaul Durrant if (fragment) 3771ed1f50c3SPaul Durrant goto out; 3772ed1f50c3SPaul Durrant 3773f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 3774f9708b43SJan Beulich if (IS_ERR(csum)) 3775f9708b43SJan Beulich return PTR_ERR(csum); 3776ed1f50c3SPaul Durrant 3777ed1f50c3SPaul Durrant if (recalculate) 3778f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 3779ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 3780ed1f50c3SPaul Durrant skb->len - off, 3781f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 3782ed1f50c3SPaul Durrant err = 0; 3783ed1f50c3SPaul Durrant 3784ed1f50c3SPaul Durrant out: 3785ed1f50c3SPaul Durrant return err; 3786ed1f50c3SPaul Durrant } 3787ed1f50c3SPaul Durrant 3788ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 3789ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 3790ed1f50c3SPaul Durrant */ 3791ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 3792ed1f50c3SPaul Durrant 3793ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 3794ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 3795ed1f50c3SPaul Durrant 3796ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 3797ed1f50c3SPaul Durrant { 3798ed1f50c3SPaul Durrant int err; 3799ed1f50c3SPaul Durrant u8 nexthdr; 3800ed1f50c3SPaul Durrant unsigned int off; 3801ed1f50c3SPaul Durrant unsigned int len; 3802ed1f50c3SPaul Durrant bool fragment; 3803ed1f50c3SPaul Durrant bool done; 3804f9708b43SJan Beulich __sum16 *csum; 3805ed1f50c3SPaul Durrant 3806ed1f50c3SPaul Durrant fragment = false; 3807ed1f50c3SPaul Durrant done = false; 3808ed1f50c3SPaul Durrant 3809ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 3810ed1f50c3SPaul Durrant 3811ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 3812ed1f50c3SPaul Durrant if (err < 0) 3813ed1f50c3SPaul Durrant goto out; 3814ed1f50c3SPaul Durrant 3815ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 3816ed1f50c3SPaul Durrant 3817ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 3818ed1f50c3SPaul Durrant while (off <= len && !done) { 3819ed1f50c3SPaul Durrant switch (nexthdr) { 3820ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 3821ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 3822ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 3823ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 3824ed1f50c3SPaul Durrant 3825ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3826ed1f50c3SPaul Durrant off + 3827ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 3828ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3829ed1f50c3SPaul Durrant if (err < 0) 3830ed1f50c3SPaul Durrant goto out; 3831ed1f50c3SPaul Durrant 3832ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 3833ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3834ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 3835ed1f50c3SPaul Durrant break; 3836ed1f50c3SPaul Durrant } 3837ed1f50c3SPaul Durrant case IPPROTO_AH: { 3838ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 3839ed1f50c3SPaul Durrant 3840ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3841ed1f50c3SPaul Durrant off + 3842ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 3843ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3844ed1f50c3SPaul Durrant if (err < 0) 3845ed1f50c3SPaul Durrant goto out; 3846ed1f50c3SPaul Durrant 3847ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 3848ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3849ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 3850ed1f50c3SPaul Durrant break; 3851ed1f50c3SPaul Durrant } 3852ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 3853ed1f50c3SPaul Durrant struct frag_hdr *hp; 3854ed1f50c3SPaul Durrant 3855ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3856ed1f50c3SPaul Durrant off + 3857ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 3858ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3859ed1f50c3SPaul Durrant if (err < 0) 3860ed1f50c3SPaul Durrant goto out; 3861ed1f50c3SPaul Durrant 3862ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 3863ed1f50c3SPaul Durrant 3864ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 3865ed1f50c3SPaul Durrant fragment = true; 3866ed1f50c3SPaul Durrant 3867ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3868ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 3869ed1f50c3SPaul Durrant break; 3870ed1f50c3SPaul Durrant } 3871ed1f50c3SPaul Durrant default: 3872ed1f50c3SPaul Durrant done = true; 3873ed1f50c3SPaul Durrant break; 3874ed1f50c3SPaul Durrant } 3875ed1f50c3SPaul Durrant } 3876ed1f50c3SPaul Durrant 3877ed1f50c3SPaul Durrant err = -EPROTO; 3878ed1f50c3SPaul Durrant 3879ed1f50c3SPaul Durrant if (!done || fragment) 3880ed1f50c3SPaul Durrant goto out; 3881ed1f50c3SPaul Durrant 3882f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 3883f9708b43SJan Beulich if (IS_ERR(csum)) 3884f9708b43SJan Beulich return PTR_ERR(csum); 3885ed1f50c3SPaul Durrant 3886ed1f50c3SPaul Durrant if (recalculate) 3887f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 3888ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 3889f9708b43SJan Beulich skb->len - off, nexthdr, 0); 3890ed1f50c3SPaul Durrant err = 0; 3891ed1f50c3SPaul Durrant 3892ed1f50c3SPaul Durrant out: 3893ed1f50c3SPaul Durrant return err; 3894ed1f50c3SPaul Durrant } 3895ed1f50c3SPaul Durrant 3896ed1f50c3SPaul Durrant /** 3897ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 3898ed1f50c3SPaul Durrant * @skb: the skb to set up 3899ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 3900ed1f50c3SPaul Durrant */ 3901ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 3902ed1f50c3SPaul Durrant { 3903ed1f50c3SPaul Durrant int err; 3904ed1f50c3SPaul Durrant 3905ed1f50c3SPaul Durrant switch (skb->protocol) { 3906ed1f50c3SPaul Durrant case htons(ETH_P_IP): 3907f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 3908ed1f50c3SPaul Durrant break; 3909ed1f50c3SPaul Durrant 3910ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 3911ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 3912ed1f50c3SPaul Durrant break; 3913ed1f50c3SPaul Durrant 3914ed1f50c3SPaul Durrant default: 3915ed1f50c3SPaul Durrant err = -EPROTO; 3916ed1f50c3SPaul Durrant break; 3917ed1f50c3SPaul Durrant } 3918ed1f50c3SPaul Durrant 3919ed1f50c3SPaul Durrant return err; 3920ed1f50c3SPaul Durrant } 3921ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 3922ed1f50c3SPaul Durrant 39234497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 39244497b076SBen Hutchings { 3925e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 3926e87cc472SJoe Perches skb->dev->name); 39274497b076SBen Hutchings } 39284497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 3929bad43ca8SEric Dumazet 3930bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 3931bad43ca8SEric Dumazet { 39323d861f66SEric Dumazet if (head_stolen) { 39333d861f66SEric Dumazet skb_release_head_state(skb); 3934bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 39353d861f66SEric Dumazet } else { 3936bad43ca8SEric Dumazet __kfree_skb(skb); 3937bad43ca8SEric Dumazet } 39383d861f66SEric Dumazet } 3939bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 3940bad43ca8SEric Dumazet 3941bad43ca8SEric Dumazet /** 3942bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 3943bad43ca8SEric Dumazet * @to: prior buffer 3944bad43ca8SEric Dumazet * @from: buffer to add 3945bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 3946c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 3947bad43ca8SEric Dumazet */ 3948bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 3949bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 3950bad43ca8SEric Dumazet { 3951bad43ca8SEric Dumazet int i, delta, len = from->len; 3952bad43ca8SEric Dumazet 3953bad43ca8SEric Dumazet *fragstolen = false; 3954bad43ca8SEric Dumazet 3955bad43ca8SEric Dumazet if (skb_cloned(to)) 3956bad43ca8SEric Dumazet return false; 3957bad43ca8SEric Dumazet 3958bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 3959e93a0435SEric Dumazet if (len) 3960bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 3961bad43ca8SEric Dumazet *delta_truesize = 0; 3962bad43ca8SEric Dumazet return true; 3963bad43ca8SEric Dumazet } 3964bad43ca8SEric Dumazet 3965bad43ca8SEric Dumazet if (skb_has_frag_list(to) || skb_has_frag_list(from)) 3966bad43ca8SEric Dumazet return false; 3967bad43ca8SEric Dumazet 3968bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 3969bad43ca8SEric Dumazet struct page *page; 3970bad43ca8SEric Dumazet unsigned int offset; 3971bad43ca8SEric Dumazet 3972bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 3973bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 3974bad43ca8SEric Dumazet return false; 3975bad43ca8SEric Dumazet 3976bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 3977bad43ca8SEric Dumazet return false; 3978bad43ca8SEric Dumazet 3979bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3980bad43ca8SEric Dumazet 3981bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 3982bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 3983bad43ca8SEric Dumazet 3984bad43ca8SEric Dumazet skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 3985bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 3986bad43ca8SEric Dumazet *fragstolen = true; 3987bad43ca8SEric Dumazet } else { 3988bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 3989bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 3990bad43ca8SEric Dumazet return false; 3991bad43ca8SEric Dumazet 3992f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 3993bad43ca8SEric Dumazet } 3994bad43ca8SEric Dumazet 3995bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 3996bad43ca8SEric Dumazet 3997bad43ca8SEric Dumazet memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 3998bad43ca8SEric Dumazet skb_shinfo(from)->frags, 3999bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 4000bad43ca8SEric Dumazet skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 4001bad43ca8SEric Dumazet 4002bad43ca8SEric Dumazet if (!skb_cloned(from)) 4003bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags = 0; 4004bad43ca8SEric Dumazet 40058ea853fdSLi RongQing /* if the skb is not cloned this does nothing 40068ea853fdSLi RongQing * since we set nr_frags to 0. 40078ea853fdSLi RongQing */ 4008bad43ca8SEric Dumazet for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 4009bad43ca8SEric Dumazet skb_frag_ref(from, i); 4010bad43ca8SEric Dumazet 4011bad43ca8SEric Dumazet to->truesize += delta; 4012bad43ca8SEric Dumazet to->len += len; 4013bad43ca8SEric Dumazet to->data_len += len; 4014bad43ca8SEric Dumazet 4015bad43ca8SEric Dumazet *delta_truesize = delta; 4016bad43ca8SEric Dumazet return true; 4017bad43ca8SEric Dumazet } 4018bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 4019621e84d6SNicolas Dichtel 4020621e84d6SNicolas Dichtel /** 40218b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 4022621e84d6SNicolas Dichtel * 4023621e84d6SNicolas Dichtel * @skb: buffer to clean 40248b27f277SNicolas Dichtel * @xnet: packet is crossing netns 4025621e84d6SNicolas Dichtel * 40268b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 40278b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 40288b27f277SNicolas Dichtel * operations. 40298b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 40308b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 40318b27f277SNicolas Dichtel * skb that could impact namespace isolation. 4032621e84d6SNicolas Dichtel */ 40338b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4034621e84d6SNicolas Dichtel { 40358b27f277SNicolas Dichtel if (xnet) 4036621e84d6SNicolas Dichtel skb_orphan(skb); 4037621e84d6SNicolas Dichtel skb->tstamp.tv64 = 0; 4038621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 4039621e84d6SNicolas Dichtel skb->skb_iif = 0; 404060ff7467SWANG Cong skb->ignore_df = 0; 4041621e84d6SNicolas Dichtel skb_dst_drop(skb); 4042621e84d6SNicolas Dichtel skb->mark = 0; 4043621e84d6SNicolas Dichtel secpath_reset(skb); 4044621e84d6SNicolas Dichtel nf_reset(skb); 4045621e84d6SNicolas Dichtel nf_reset_trace(skb); 4046621e84d6SNicolas Dichtel } 4047621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 4048de960aa9SFlorian Westphal 4049de960aa9SFlorian Westphal /** 4050de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4051de960aa9SFlorian Westphal * 4052de960aa9SFlorian Westphal * @skb: GSO skb 4053de960aa9SFlorian Westphal * 4054de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 4055de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 4056de960aa9SFlorian Westphal * 4057de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4058de960aa9SFlorian Westphal */ 4059de960aa9SFlorian Westphal unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4060de960aa9SFlorian Westphal { 4061de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 4062de960aa9SFlorian Westphal 4063de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 40646d39d589SFlorian Westphal return tcp_hdrlen(skb) + shinfo->gso_size; 40656d39d589SFlorian Westphal 40666d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 40676d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 40686d39d589SFlorian Westphal * accounted for. 40696d39d589SFlorian Westphal */ 40706d39d589SFlorian Westphal return shinfo->gso_size; 4071de960aa9SFlorian Westphal } 4072de960aa9SFlorian Westphal EXPORT_SYMBOL_GPL(skb_gso_transport_seglen); 40730d5501c1SVlad Yasevich 40740d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 40750d5501c1SVlad Yasevich { 40760d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 40770d5501c1SVlad Yasevich kfree_skb(skb); 40780d5501c1SVlad Yasevich return NULL; 40790d5501c1SVlad Yasevich } 40800d5501c1SVlad Yasevich 40810d5501c1SVlad Yasevich memmove(skb->data - ETH_HLEN, skb->data - VLAN_ETH_HLEN, 2 * ETH_ALEN); 40820d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 40830d5501c1SVlad Yasevich return skb; 40840d5501c1SVlad Yasevich } 40850d5501c1SVlad Yasevich 40860d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 40870d5501c1SVlad Yasevich { 40880d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 40890d5501c1SVlad Yasevich u16 vlan_tci; 40900d5501c1SVlad Yasevich 40910d5501c1SVlad Yasevich if (unlikely(vlan_tx_tag_present(skb))) { 40920d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 40930d5501c1SVlad Yasevich return skb; 40940d5501c1SVlad Yasevich } 40950d5501c1SVlad Yasevich 40960d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 40970d5501c1SVlad Yasevich if (unlikely(!skb)) 40980d5501c1SVlad Yasevich goto err_free; 40990d5501c1SVlad Yasevich 41000d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 41010d5501c1SVlad Yasevich goto err_free; 41020d5501c1SVlad Yasevich 41030d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 41040d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 41050d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 41060d5501c1SVlad Yasevich 41070d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 41080d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 41090d5501c1SVlad Yasevich 41100d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 41110d5501c1SVlad Yasevich if (unlikely(!skb)) 41120d5501c1SVlad Yasevich goto err_free; 41130d5501c1SVlad Yasevich 41140d5501c1SVlad Yasevich skb_reset_network_header(skb); 41150d5501c1SVlad Yasevich skb_reset_transport_header(skb); 41160d5501c1SVlad Yasevich skb_reset_mac_len(skb); 41170d5501c1SVlad Yasevich 41180d5501c1SVlad Yasevich return skb; 41190d5501c1SVlad Yasevich 41200d5501c1SVlad Yasevich err_free: 41210d5501c1SVlad Yasevich kfree_skb(skb); 41220d5501c1SVlad Yasevich return NULL; 41230d5501c1SVlad Yasevich } 41240d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 41252e4e4410SEric Dumazet 41262e4e4410SEric Dumazet /** 41272e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 41282e4e4410SEric Dumazet * 41292e4e4410SEric Dumazet * header_len: size of linear part 41302e4e4410SEric Dumazet * data_len: needed length in frags 41312e4e4410SEric Dumazet * max_page_order: max page order desired. 41322e4e4410SEric Dumazet * errcode: pointer to error code if any 41332e4e4410SEric Dumazet * gfp_mask: allocation mask 41342e4e4410SEric Dumazet * 41352e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 41362e4e4410SEric Dumazet */ 41372e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 41382e4e4410SEric Dumazet unsigned long data_len, 41392e4e4410SEric Dumazet int max_page_order, 41402e4e4410SEric Dumazet int *errcode, 41412e4e4410SEric Dumazet gfp_t gfp_mask) 41422e4e4410SEric Dumazet { 41432e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 41442e4e4410SEric Dumazet unsigned long chunk; 41452e4e4410SEric Dumazet struct sk_buff *skb; 41462e4e4410SEric Dumazet struct page *page; 41472e4e4410SEric Dumazet gfp_t gfp_head; 41482e4e4410SEric Dumazet int i; 41492e4e4410SEric Dumazet 41502e4e4410SEric Dumazet *errcode = -EMSGSIZE; 41512e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 41522e4e4410SEric Dumazet * high order pages... 41532e4e4410SEric Dumazet */ 41542e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 41552e4e4410SEric Dumazet return NULL; 41562e4e4410SEric Dumazet 41572e4e4410SEric Dumazet gfp_head = gfp_mask; 41582e4e4410SEric Dumazet if (gfp_head & __GFP_WAIT) 41592e4e4410SEric Dumazet gfp_head |= __GFP_REPEAT; 41602e4e4410SEric Dumazet 41612e4e4410SEric Dumazet *errcode = -ENOBUFS; 41622e4e4410SEric Dumazet skb = alloc_skb(header_len, gfp_head); 41632e4e4410SEric Dumazet if (!skb) 41642e4e4410SEric Dumazet return NULL; 41652e4e4410SEric Dumazet 41662e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 41672e4e4410SEric Dumazet 41682e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 41692e4e4410SEric Dumazet int order = max_page_order; 41702e4e4410SEric Dumazet 41712e4e4410SEric Dumazet while (order) { 41722e4e4410SEric Dumazet if (npages >= 1 << order) { 41732e4e4410SEric Dumazet page = alloc_pages(gfp_mask | 41742e4e4410SEric Dumazet __GFP_COMP | 41752e4e4410SEric Dumazet __GFP_NOWARN | 41762e4e4410SEric Dumazet __GFP_NORETRY, 41772e4e4410SEric Dumazet order); 41782e4e4410SEric Dumazet if (page) 41792e4e4410SEric Dumazet goto fill_page; 41802e4e4410SEric Dumazet /* Do not retry other high order allocations */ 41812e4e4410SEric Dumazet order = 1; 41822e4e4410SEric Dumazet max_page_order = 0; 41832e4e4410SEric Dumazet } 41842e4e4410SEric Dumazet order--; 41852e4e4410SEric Dumazet } 41862e4e4410SEric Dumazet page = alloc_page(gfp_mask); 41872e4e4410SEric Dumazet if (!page) 41882e4e4410SEric Dumazet goto failure; 41892e4e4410SEric Dumazet fill_page: 41902e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 41912e4e4410SEric Dumazet PAGE_SIZE << order); 41922e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 41932e4e4410SEric Dumazet data_len -= chunk; 41942e4e4410SEric Dumazet npages -= 1 << order; 41952e4e4410SEric Dumazet } 41962e4e4410SEric Dumazet return skb; 41972e4e4410SEric Dumazet 41982e4e4410SEric Dumazet failure: 41992e4e4410SEric Dumazet kfree_skb(skb); 42002e4e4410SEric Dumazet return NULL; 42012e4e4410SEric Dumazet } 42022e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 4203