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 2686ffe75ebSEric Dumazet fclones->skb2.fclone = SKB_FCLONE_CLONE; 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); 339ffde7328SAlexander Duyck static DEFINE_PER_CPU(struct netdev_alloc_cache, napi_alloc_cache); 340a1c7fff7SEric Dumazet 341ffde7328SAlexander Duyck static struct page *__page_frag_refill(struct netdev_alloc_cache *nc, 342ffde7328SAlexander Duyck gfp_t gfp_mask) 3436f532612SEric Dumazet { 344ffde7328SAlexander Duyck const unsigned int order = NETDEV_FRAG_PAGE_MAX_ORDER; 345ffde7328SAlexander Duyck struct page *page = NULL; 34669b08f62SEric Dumazet gfp_t gfp = gfp_mask; 34769b08f62SEric Dumazet 348ffde7328SAlexander Duyck if (order) { 349ffde7328SAlexander Duyck gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY; 350ffde7328SAlexander Duyck page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 351ffde7328SAlexander Duyck nc->frag.size = PAGE_SIZE << (page ? order : 0); 35269b08f62SEric Dumazet } 353ffde7328SAlexander Duyck 354ffde7328SAlexander Duyck if (unlikely(!page)) 355ffde7328SAlexander Duyck page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); 356ffde7328SAlexander Duyck 357ffde7328SAlexander Duyck nc->frag.page = page; 358ffde7328SAlexander Duyck 359ffde7328SAlexander Duyck return page; 360ffde7328SAlexander Duyck } 361ffde7328SAlexander Duyck 362ffde7328SAlexander Duyck static void *__alloc_page_frag(struct netdev_alloc_cache __percpu *cache, 363ffde7328SAlexander Duyck unsigned int fragsz, gfp_t gfp_mask) 364ffde7328SAlexander Duyck { 365ffde7328SAlexander Duyck struct netdev_alloc_cache *nc = this_cpu_ptr(cache); 366ffde7328SAlexander Duyck struct page *page = nc->frag.page; 367ffde7328SAlexander Duyck unsigned int size; 368ffde7328SAlexander Duyck int offset; 369ffde7328SAlexander Duyck 370ffde7328SAlexander Duyck if (unlikely(!page)) { 371ffde7328SAlexander Duyck refill: 372ffde7328SAlexander Duyck page = __page_frag_refill(nc, gfp_mask); 373ffde7328SAlexander Duyck if (!page) 374ffde7328SAlexander Duyck return NULL; 375ffde7328SAlexander Duyck 376ffde7328SAlexander Duyck /* if size can vary use frag.size else just use PAGE_SIZE */ 377ffde7328SAlexander Duyck size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE; 378ffde7328SAlexander Duyck 3794c450583SEric Dumazet /* Even if we own the page, we do not use atomic_set(). 3804c450583SEric Dumazet * This would break get_page_unless_zero() users. 3814c450583SEric Dumazet */ 382ffde7328SAlexander Duyck atomic_add(size - 1, &page->_count); 383ffde7328SAlexander Duyck 384ffde7328SAlexander Duyck /* reset page count bias and offset to start of new frag */ 385ffde7328SAlexander Duyck nc->pagecnt_bias = size; 386ffde7328SAlexander Duyck nc->frag.offset = size; 3876f532612SEric Dumazet } 388540eb7bfSAlexander Duyck 389ffde7328SAlexander Duyck offset = nc->frag.offset - fragsz; 390ffde7328SAlexander Duyck if (unlikely(offset < 0)) { 391ffde7328SAlexander Duyck if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count)) 3926f532612SEric Dumazet goto refill; 393ffde7328SAlexander Duyck 394ffde7328SAlexander Duyck /* if size can vary use frag.size else just use PAGE_SIZE */ 395ffde7328SAlexander Duyck size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE; 396ffde7328SAlexander Duyck 3974c450583SEric Dumazet /* OK, page count is 0, we can safely set it */ 398ffde7328SAlexander Duyck atomic_set(&page->_count, size); 399ffde7328SAlexander Duyck 400ffde7328SAlexander Duyck /* reset page count bias and offset to start of new frag */ 401ffde7328SAlexander Duyck nc->pagecnt_bias = size; 402ffde7328SAlexander Duyck offset = size - fragsz; 4036f532612SEric Dumazet } 404540eb7bfSAlexander Duyck 405540eb7bfSAlexander Duyck nc->pagecnt_bias--; 406ffde7328SAlexander Duyck nc->frag.offset = offset; 407ffde7328SAlexander Duyck 408ffde7328SAlexander Duyck return page_address(page) + offset; 409ffde7328SAlexander Duyck } 410ffde7328SAlexander Duyck 411ffde7328SAlexander Duyck static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 412ffde7328SAlexander Duyck { 413ffde7328SAlexander Duyck unsigned long flags; 414ffde7328SAlexander Duyck void *data; 415ffde7328SAlexander Duyck 416ffde7328SAlexander Duyck local_irq_save(flags); 417ffde7328SAlexander Duyck data = __alloc_page_frag(&netdev_alloc_cache, fragsz, gfp_mask); 4186f532612SEric Dumazet local_irq_restore(flags); 4196f532612SEric Dumazet return data; 4206f532612SEric Dumazet } 421c93bdd0eSMel Gorman 422c93bdd0eSMel Gorman /** 423c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 424c93bdd0eSMel Gorman * @fragsz: fragment size 425c93bdd0eSMel Gorman * 426c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 427c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 428c93bdd0eSMel Gorman */ 429c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 430c93bdd0eSMel Gorman { 431c93bdd0eSMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 432c93bdd0eSMel Gorman } 4336f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 4346f532612SEric Dumazet 435ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 436ffde7328SAlexander Duyck { 437ffde7328SAlexander Duyck return __alloc_page_frag(&napi_alloc_cache, fragsz, gfp_mask); 438ffde7328SAlexander Duyck } 439ffde7328SAlexander Duyck 440ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 441ffde7328SAlexander Duyck { 442ffde7328SAlexander Duyck return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 443ffde7328SAlexander Duyck } 444ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 445ffde7328SAlexander Duyck 4466f532612SEric Dumazet /** 4478af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 4488af27456SChristoph Hellwig * @dev: network device to receive on 4498af27456SChristoph Hellwig * @length: length to allocate 4508af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 4518af27456SChristoph Hellwig * 4528af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 4538af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 4548af27456SChristoph Hellwig * the headroom they think they need without accounting for the 4558af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 4568af27456SChristoph Hellwig * 4578af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 4588af27456SChristoph Hellwig */ 4598af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 4608af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 4618af27456SChristoph Hellwig { 4626f532612SEric Dumazet struct sk_buff *skb = NULL; 463a1c7fff7SEric Dumazet unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) + 464a1c7fff7SEric Dumazet SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4658af27456SChristoph Hellwig 466310e158cSEric Dumazet if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) { 467c93bdd0eSMel Gorman void *data; 468c93bdd0eSMel Gorman 469c93bdd0eSMel Gorman if (sk_memalloc_socks()) 470c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 471c93bdd0eSMel Gorman 472c93bdd0eSMel Gorman data = __netdev_alloc_frag(fragsz, gfp_mask); 473a1c7fff7SEric Dumazet 4746f532612SEric Dumazet if (likely(data)) { 4756f532612SEric Dumazet skb = build_skb(data, fragsz); 4766f532612SEric Dumazet if (unlikely(!skb)) 4776f532612SEric Dumazet put_page(virt_to_head_page(data)); 478a1c7fff7SEric Dumazet } 479a1c7fff7SEric Dumazet } else { 480c93bdd0eSMel Gorman skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 481c93bdd0eSMel Gorman SKB_ALLOC_RX, NUMA_NO_NODE); 482a1c7fff7SEric Dumazet } 4837b2e497aSChristoph Hellwig if (likely(skb)) { 4848af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4857b2e497aSChristoph Hellwig skb->dev = dev; 4867b2e497aSChristoph Hellwig } 4878af27456SChristoph Hellwig return skb; 4888af27456SChristoph Hellwig } 489b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4901da177e4SLinus Torvalds 491654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 49250269e19SEric Dumazet int size, unsigned int truesize) 493654bed16SPeter Zijlstra { 494654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 495654bed16SPeter Zijlstra skb->len += size; 496654bed16SPeter Zijlstra skb->data_len += size; 49750269e19SEric Dumazet skb->truesize += truesize; 498654bed16SPeter Zijlstra } 499654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 500654bed16SPeter Zijlstra 501f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 502f8e617e1SJason Wang unsigned int truesize) 503f8e617e1SJason Wang { 504f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 505f8e617e1SJason Wang 506f8e617e1SJason Wang skb_frag_size_add(frag, size); 507f8e617e1SJason Wang skb->len += size; 508f8e617e1SJason Wang skb->data_len += size; 509f8e617e1SJason Wang skb->truesize += truesize; 510f8e617e1SJason Wang } 511f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 512f8e617e1SJason Wang 51327b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5141da177e4SLinus Torvalds { 515bd8a7036SEric Dumazet kfree_skb_list(*listp); 51627b437c8SHerbert Xu *listp = NULL; 5171da177e4SLinus Torvalds } 5181da177e4SLinus Torvalds 51927b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 52027b437c8SHerbert Xu { 52127b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 52227b437c8SHerbert Xu } 52327b437c8SHerbert Xu 5241da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5251da177e4SLinus Torvalds { 5261da177e4SLinus Torvalds struct sk_buff *list; 5271da177e4SLinus Torvalds 528fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5291da177e4SLinus Torvalds skb_get(list); 5301da177e4SLinus Torvalds } 5311da177e4SLinus Torvalds 532d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 533d3836f21SEric Dumazet { 534d3836f21SEric Dumazet if (skb->head_frag) 535d3836f21SEric Dumazet put_page(virt_to_head_page(skb->head)); 536d3836f21SEric Dumazet else 537d3836f21SEric Dumazet kfree(skb->head); 538d3836f21SEric Dumazet } 539d3836f21SEric Dumazet 5405bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5411da177e4SLinus Torvalds { 542ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 5431da177e4SLinus Torvalds int i; 544ff04a771SEric Dumazet 545ff04a771SEric Dumazet if (skb->cloned && 546ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 547ff04a771SEric Dumazet &shinfo->dataref)) 548ff04a771SEric Dumazet return; 549ff04a771SEric Dumazet 550ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 551ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 5521da177e4SLinus Torvalds 553a6686f2fSShirley Ma /* 554a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 555a6686f2fSShirley Ma * the lower device DMA has done; 556a6686f2fSShirley Ma */ 557ff04a771SEric Dumazet if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) { 558a6686f2fSShirley Ma struct ubuf_info *uarg; 559a6686f2fSShirley Ma 560ff04a771SEric Dumazet uarg = shinfo->destructor_arg; 561a6686f2fSShirley Ma if (uarg->callback) 562e19d6763SMichael S. Tsirkin uarg->callback(uarg, true); 563a6686f2fSShirley Ma } 564a6686f2fSShirley Ma 565ff04a771SEric Dumazet if (shinfo->frag_list) 566ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 5671da177e4SLinus Torvalds 568d3836f21SEric Dumazet skb_free_head(skb); 5691da177e4SLinus Torvalds } 5701da177e4SLinus Torvalds 5711da177e4SLinus Torvalds /* 5721da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 5731da177e4SLinus Torvalds */ 5742d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 5751da177e4SLinus Torvalds { 576d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 577d179cd12SDavid S. Miller 578d179cd12SDavid S. Miller switch (skb->fclone) { 579d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 5801da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 5816ffe75ebSEric Dumazet return; 582d179cd12SDavid S. Miller 583d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 584d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 5856ffe75ebSEric Dumazet 5866ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 5876ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 5886ffe75ebSEric Dumazet * while here, branch prediction will be good. 5896ffe75ebSEric Dumazet */ 5906ffe75ebSEric Dumazet if (atomic_read(&fclones->fclone_ref) == 1) 5916ffe75ebSEric Dumazet goto fastpath; 592d179cd12SDavid S. Miller break; 593d179cd12SDavid S. Miller 5946ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 595d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 596d179cd12SDavid S. Miller break; 5973ff50b79SStephen Hemminger } 5986ffe75ebSEric Dumazet if (!atomic_dec_and_test(&fclones->fclone_ref)) 5996ffe75ebSEric Dumazet return; 6006ffe75ebSEric Dumazet fastpath: 6016ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6021da177e4SLinus Torvalds } 6031da177e4SLinus Torvalds 60404a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 6051da177e4SLinus Torvalds { 606adf30907SEric Dumazet skb_dst_drop(skb); 6071da177e4SLinus Torvalds #ifdef CONFIG_XFRM 6081da177e4SLinus Torvalds secpath_put(skb->sp); 6091da177e4SLinus Torvalds #endif 6101da177e4SLinus Torvalds if (skb->destructor) { 6119c2b3328SStephen Hemminger WARN_ON(in_irq()); 6121da177e4SLinus Torvalds skb->destructor(skb); 6131da177e4SLinus Torvalds } 614a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 6155f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 6162fc72c7bSKOVACS Krisztian #endif 6171109a90cSPablo Neira Ayuso #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 6181da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 6191da177e4SLinus Torvalds #endif 6201da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 6211da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 6221da177e4SLinus Torvalds skb->tc_index = 0; 6231da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 6241da177e4SLinus Torvalds skb->tc_verd = 0; 6251da177e4SLinus Torvalds #endif 6261da177e4SLinus Torvalds #endif 62704a4bb55SLennert Buytenhek } 62804a4bb55SLennert Buytenhek 62904a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 63004a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 63104a4bb55SLennert Buytenhek { 63204a4bb55SLennert Buytenhek skb_release_head_state(skb); 6335e71d9d7SPablo Neira if (likely(skb->head)) 6342d4baff8SHerbert Xu skb_release_data(skb); 6352d4baff8SHerbert Xu } 6361da177e4SLinus Torvalds 6372d4baff8SHerbert Xu /** 6382d4baff8SHerbert Xu * __kfree_skb - private function 6392d4baff8SHerbert Xu * @skb: buffer 6402d4baff8SHerbert Xu * 6412d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6422d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6432d4baff8SHerbert Xu * always call kfree_skb 6442d4baff8SHerbert Xu */ 6452d4baff8SHerbert Xu 6462d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6472d4baff8SHerbert Xu { 6482d4baff8SHerbert Xu skb_release_all(skb); 6491da177e4SLinus Torvalds kfree_skbmem(skb); 6501da177e4SLinus Torvalds } 651b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6521da177e4SLinus Torvalds 6531da177e4SLinus Torvalds /** 654231d06aeSJörn Engel * kfree_skb - free an sk_buff 655231d06aeSJörn Engel * @skb: buffer to free 656231d06aeSJörn Engel * 657231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 658231d06aeSJörn Engel * hit zero. 659231d06aeSJörn Engel */ 660231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 661231d06aeSJörn Engel { 662231d06aeSJörn Engel if (unlikely(!skb)) 663231d06aeSJörn Engel return; 664231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 665231d06aeSJörn Engel smp_rmb(); 666231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 667231d06aeSJörn Engel return; 668ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 669231d06aeSJörn Engel __kfree_skb(skb); 670231d06aeSJörn Engel } 671b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 672231d06aeSJörn Engel 673bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 674bd8a7036SEric Dumazet { 675bd8a7036SEric Dumazet while (segs) { 676bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 677bd8a7036SEric Dumazet 678bd8a7036SEric Dumazet kfree_skb(segs); 679bd8a7036SEric Dumazet segs = next; 680bd8a7036SEric Dumazet } 681bd8a7036SEric Dumazet } 682bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 683bd8a7036SEric Dumazet 684d1a203eaSStephen Hemminger /** 68525121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 68625121173SMichael S. Tsirkin * @skb: buffer that triggered an error 68725121173SMichael S. Tsirkin * 68825121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 68925121173SMichael S. Tsirkin * skb must be freed afterwards. 69025121173SMichael S. Tsirkin */ 69125121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 69225121173SMichael S. Tsirkin { 69325121173SMichael S. Tsirkin if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 69425121173SMichael S. Tsirkin struct ubuf_info *uarg; 69525121173SMichael S. Tsirkin 69625121173SMichael S. Tsirkin uarg = skb_shinfo(skb)->destructor_arg; 69725121173SMichael S. Tsirkin if (uarg->callback) 69825121173SMichael S. Tsirkin uarg->callback(uarg, false); 69925121173SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 70025121173SMichael S. Tsirkin } 70125121173SMichael S. Tsirkin } 70225121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 70325121173SMichael S. Tsirkin 70425121173SMichael S. Tsirkin /** 705ead2ceb0SNeil Horman * consume_skb - free an skbuff 706ead2ceb0SNeil Horman * @skb: buffer to free 707ead2ceb0SNeil Horman * 708ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 709ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 710ead2ceb0SNeil Horman * is being dropped after a failure and notes that 711ead2ceb0SNeil Horman */ 712ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 713ead2ceb0SNeil Horman { 714ead2ceb0SNeil Horman if (unlikely(!skb)) 715ead2ceb0SNeil Horman return; 716ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 717ead2ceb0SNeil Horman smp_rmb(); 718ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 719ead2ceb0SNeil Horman return; 72007dc22e7SKoki Sanagi trace_consume_skb(skb); 721ead2ceb0SNeil Horman __kfree_skb(skb); 722ead2ceb0SNeil Horman } 723ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 724ead2ceb0SNeil Horman 725b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 726b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 727b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 728b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 729b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 730b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 731b1937227SEric Dumazet 732dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 733dec18810SHerbert Xu { 734dec18810SHerbert Xu new->tstamp = old->tstamp; 735b1937227SEric Dumazet /* We do not copy old->sk */ 736dec18810SHerbert Xu new->dev = old->dev; 737b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 7387fee226aSEric Dumazet skb_dst_copy(new, old); 739def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 740dec18810SHerbert Xu new->sp = secpath_get(old->sp); 741dec18810SHerbert Xu #endif 742b1937227SEric Dumazet __nf_copy(new, old, false); 7436aa895b0SPatrick McHardy 744b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 745b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 746b1937227SEric Dumazet */ 747b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 74806021292SEliezer Tamir 749b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 750b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 751b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 752b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 753b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 754b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 755b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 756b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 757b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 758b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 759b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 760b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 761b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 762b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 763b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 764b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 765b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 766b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 767b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 768b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 769b1937227SEric Dumazet #endif 770e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 771b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 77206021292SEliezer Tamir #endif 773b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 774b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 775b1937227SEric Dumazet #ifdef CONFIG_NET_CLS_ACT 776b1937227SEric Dumazet CHECK_SKB_FIELD(tc_verd); 777b1937227SEric Dumazet #endif 778b1937227SEric Dumazet #endif 779b1937227SEric Dumazet 780dec18810SHerbert Xu } 781dec18810SHerbert Xu 78282c49a35SHerbert Xu /* 78382c49a35SHerbert Xu * You should not add any new code to this function. Add it to 78482c49a35SHerbert Xu * __copy_skb_header above instead. 78582c49a35SHerbert Xu */ 786e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 7871da177e4SLinus Torvalds { 7881da177e4SLinus Torvalds #define C(x) n->x = skb->x 7891da177e4SLinus Torvalds 7901da177e4SLinus Torvalds n->next = n->prev = NULL; 7911da177e4SLinus Torvalds n->sk = NULL; 792dec18810SHerbert Xu __copy_skb_header(n, skb); 793dec18810SHerbert Xu 7941da177e4SLinus Torvalds C(len); 7951da177e4SLinus Torvalds C(data_len); 7963e6b3b2eSAlexey Dobriyan C(mac_len); 797334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 79802f1c89dSPaul Moore n->cloned = 1; 7991da177e4SLinus Torvalds n->nohdr = 0; 8001da177e4SLinus Torvalds n->destructor = NULL; 8011da177e4SLinus Torvalds C(tail); 8021da177e4SLinus Torvalds C(end); 80302f1c89dSPaul Moore C(head); 804d3836f21SEric Dumazet C(head_frag); 80502f1c89dSPaul Moore C(data); 80602f1c89dSPaul Moore C(truesize); 80702f1c89dSPaul Moore atomic_set(&n->users, 1); 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 8101da177e4SLinus Torvalds skb->cloned = 1; 8111da177e4SLinus Torvalds 8121da177e4SLinus Torvalds return n; 813e0053ec0SHerbert Xu #undef C 814e0053ec0SHerbert Xu } 815e0053ec0SHerbert Xu 816e0053ec0SHerbert Xu /** 817e0053ec0SHerbert Xu * skb_morph - morph one skb into another 818e0053ec0SHerbert Xu * @dst: the skb to receive the contents 819e0053ec0SHerbert Xu * @src: the skb to supply the contents 820e0053ec0SHerbert Xu * 821e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 822e0053ec0SHerbert Xu * supplied by the user. 823e0053ec0SHerbert Xu * 824e0053ec0SHerbert Xu * The target skb is returned upon exit. 825e0053ec0SHerbert Xu */ 826e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 827e0053ec0SHerbert Xu { 8282d4baff8SHerbert Xu skb_release_all(dst); 829e0053ec0SHerbert Xu return __skb_clone(dst, src); 830e0053ec0SHerbert Xu } 831e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 832e0053ec0SHerbert Xu 8332c53040fSBen Hutchings /** 8342c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 83548c83012SMichael S. Tsirkin * @skb: the skb to modify 83648c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 83748c83012SMichael S. Tsirkin * 83848c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 83948c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 84048c83012SMichael S. Tsirkin * to userspace pages. 84148c83012SMichael S. Tsirkin * 84248c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 84348c83012SMichael S. Tsirkin * %GFP_ATOMIC. 84448c83012SMichael S. Tsirkin * 84548c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 84648c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 84748c83012SMichael S. Tsirkin */ 84848c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 849a6686f2fSShirley Ma { 850a6686f2fSShirley Ma int i; 851a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 852a6686f2fSShirley Ma struct page *page, *head = NULL; 853a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 854a6686f2fSShirley Ma 855a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 856a6686f2fSShirley Ma u8 *vaddr; 857a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 858a6686f2fSShirley Ma 85902756ed4SKrishna Kumar page = alloc_page(gfp_mask); 860a6686f2fSShirley Ma if (!page) { 861a6686f2fSShirley Ma while (head) { 86240dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 863a6686f2fSShirley Ma put_page(head); 864a6686f2fSShirley Ma head = next; 865a6686f2fSShirley Ma } 866a6686f2fSShirley Ma return -ENOMEM; 867a6686f2fSShirley Ma } 86851c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 869a6686f2fSShirley Ma memcpy(page_address(page), 8709e903e08SEric Dumazet vaddr + f->page_offset, skb_frag_size(f)); 87151c56b00SEric Dumazet kunmap_atomic(vaddr); 87240dadff2SSunghan Suh set_page_private(page, (unsigned long)head); 873a6686f2fSShirley Ma head = page; 874a6686f2fSShirley Ma } 875a6686f2fSShirley Ma 876a6686f2fSShirley Ma /* skb frags release userspace buffers */ 87702756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 878a8605c60SIan Campbell skb_frag_unref(skb, i); 879a6686f2fSShirley Ma 880e19d6763SMichael S. Tsirkin uarg->callback(uarg, false); 881a6686f2fSShirley Ma 882a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 88302756ed4SKrishna Kumar for (i = num_frags - 1; i >= 0; i--) { 88402756ed4SKrishna Kumar __skb_fill_page_desc(skb, i, head, 0, 88502756ed4SKrishna Kumar skb_shinfo(skb)->frags[i].size); 88640dadff2SSunghan Suh head = (struct page *)page_private(head); 887a6686f2fSShirley Ma } 88848c83012SMichael S. Tsirkin 88948c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 890a6686f2fSShirley Ma return 0; 891a6686f2fSShirley Ma } 892dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 893a6686f2fSShirley Ma 894e0053ec0SHerbert Xu /** 895e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 896e0053ec0SHerbert Xu * @skb: buffer to clone 897e0053ec0SHerbert Xu * @gfp_mask: allocation priority 898e0053ec0SHerbert Xu * 899e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 900e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 901e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 902e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 903e0053ec0SHerbert Xu * 904e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 905e0053ec0SHerbert Xu * %GFP_ATOMIC. 906e0053ec0SHerbert Xu */ 907e0053ec0SHerbert Xu 908e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 909e0053ec0SHerbert Xu { 910d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 911d0bf4a9eSEric Dumazet struct sk_buff_fclones, 912d0bf4a9eSEric Dumazet skb1); 9136ffe75ebSEric Dumazet struct sk_buff *n; 914e0053ec0SHerbert Xu 91570008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 916a6686f2fSShirley Ma return NULL; 917a6686f2fSShirley Ma 918e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 9196ffe75ebSEric Dumazet atomic_read(&fclones->fclone_ref) == 1) { 9206ffe75ebSEric Dumazet n = &fclones->skb2; 9216ffe75ebSEric Dumazet atomic_set(&fclones->fclone_ref, 2); 922e0053ec0SHerbert Xu } else { 923c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 924c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 925c93bdd0eSMel Gorman 926e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 927e0053ec0SHerbert Xu if (!n) 928e0053ec0SHerbert Xu return NULL; 929fe55f6d5SVegard Nossum 930fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 931e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 932e0053ec0SHerbert Xu } 933e0053ec0SHerbert Xu 934e0053ec0SHerbert Xu return __skb_clone(n, skb); 9351da177e4SLinus Torvalds } 936b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 9371da177e4SLinus Torvalds 938f5b17294SPravin B Shelar static void skb_headers_offset_update(struct sk_buff *skb, int off) 939f5b17294SPravin B Shelar { 940030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 941030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 942030737bcSEric Dumazet skb->csum_start += off; 943f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 944f5b17294SPravin B Shelar skb->transport_header += off; 945f5b17294SPravin B Shelar skb->network_header += off; 946f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 947f5b17294SPravin B Shelar skb->mac_header += off; 948f5b17294SPravin B Shelar skb->inner_transport_header += off; 949f5b17294SPravin B Shelar skb->inner_network_header += off; 950aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 951f5b17294SPravin B Shelar } 952f5b17294SPravin B Shelar 9531da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 9541da177e4SLinus Torvalds { 955dec18810SHerbert Xu __copy_skb_header(new, old); 956dec18810SHerbert Xu 9577967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 9587967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 9597967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 9601da177e4SLinus Torvalds } 9611da177e4SLinus Torvalds 962c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 963c93bdd0eSMel Gorman { 964c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 965c93bdd0eSMel Gorman return SKB_ALLOC_RX; 966c93bdd0eSMel Gorman return 0; 967c93bdd0eSMel Gorman } 968c93bdd0eSMel Gorman 9691da177e4SLinus Torvalds /** 9701da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 9711da177e4SLinus Torvalds * @skb: buffer to copy 9721da177e4SLinus Torvalds * @gfp_mask: allocation priority 9731da177e4SLinus Torvalds * 9741da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 9751da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 9761da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 9771da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 9781da177e4SLinus Torvalds * 9791da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 9801da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 9811da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 9821da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 9831da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 9841da177e4SLinus Torvalds */ 9851da177e4SLinus Torvalds 986dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 9871da177e4SLinus Torvalds { 9886602cebbSEric Dumazet int headerlen = skb_headroom(skb); 989ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 990c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 991c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 9926602cebbSEric Dumazet 9931da177e4SLinus Torvalds if (!n) 9941da177e4SLinus Torvalds return NULL; 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds /* Set the data pointer */ 9971da177e4SLinus Torvalds skb_reserve(n, headerlen); 9981da177e4SLinus Torvalds /* Set the tail pointer and length */ 9991da177e4SLinus Torvalds skb_put(n, skb->len); 10001da177e4SLinus Torvalds 10011da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 10021da177e4SLinus Torvalds BUG(); 10031da177e4SLinus Torvalds 10041da177e4SLinus Torvalds copy_skb_header(n, skb); 10051da177e4SLinus Torvalds return n; 10061da177e4SLinus Torvalds } 1007b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds /** 1010bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 10111da177e4SLinus Torvalds * @skb: buffer to copy 1012117632e6SEric Dumazet * @headroom: headroom of new skb 10131da177e4SLinus Torvalds * @gfp_mask: allocation priority 1014bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1015bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1016bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 10171da177e4SLinus Torvalds * 10181da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 10191da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 10201da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 10211da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 10221da177e4SLinus Torvalds * or the pointer to the buffer on success. 10231da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 10241da177e4SLinus Torvalds */ 10251da177e4SLinus Torvalds 1026bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1027bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 10281da177e4SLinus Torvalds { 1029117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1030bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1031bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 10326602cebbSEric Dumazet 10331da177e4SLinus Torvalds if (!n) 10341da177e4SLinus Torvalds goto out; 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds /* Set the data pointer */ 1037117632e6SEric Dumazet skb_reserve(n, headroom); 10381da177e4SLinus Torvalds /* Set the tail pointer and length */ 10391da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 10401da177e4SLinus Torvalds /* Copy the bytes */ 1041d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 10421da177e4SLinus Torvalds 104325f484a6SHerbert Xu n->truesize += skb->data_len; 10441da177e4SLinus Torvalds n->data_len = skb->data_len; 10451da177e4SLinus Torvalds n->len = skb->len; 10461da177e4SLinus Torvalds 10471da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 10481da177e4SLinus Torvalds int i; 10491da177e4SLinus Torvalds 105070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) { 10511511022cSDan Carpenter kfree_skb(n); 10521511022cSDan Carpenter n = NULL; 1053a6686f2fSShirley Ma goto out; 1054a6686f2fSShirley Ma } 10551da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 10561da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1057ea2ab693SIan Campbell skb_frag_ref(skb, i); 10581da177e4SLinus Torvalds } 10591da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 10601da177e4SLinus Torvalds } 10611da177e4SLinus Torvalds 106221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 10631da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 10641da177e4SLinus Torvalds skb_clone_fraglist(n); 10651da177e4SLinus Torvalds } 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds copy_skb_header(n, skb); 10681da177e4SLinus Torvalds out: 10691da177e4SLinus Torvalds return n; 10701da177e4SLinus Torvalds } 1071bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds /** 10741da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 10751da177e4SLinus Torvalds * @skb: buffer to reallocate 10761da177e4SLinus Torvalds * @nhead: room to add at head 10771da177e4SLinus Torvalds * @ntail: room to add at tail 10781da177e4SLinus Torvalds * @gfp_mask: allocation priority 10791da177e4SLinus Torvalds * 1080bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1081bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 10821da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 10831da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 10841da177e4SLinus Torvalds * 10851da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 10861da177e4SLinus Torvalds * reloaded after call to this function. 10871da177e4SLinus Torvalds */ 10881da177e4SLinus Torvalds 108986a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1090dd0fc66fSAl Viro gfp_t gfp_mask) 10911da177e4SLinus Torvalds { 10921da177e4SLinus Torvalds int i; 10931da177e4SLinus Torvalds u8 *data; 1094ec47ea82SAlexander Duyck int size = nhead + skb_end_offset(skb) + ntail; 10951da177e4SLinus Torvalds long off; 10961da177e4SLinus Torvalds 10974edd87adSHerbert Xu BUG_ON(nhead < 0); 10984edd87adSHerbert Xu 10991da177e4SLinus Torvalds if (skb_shared(skb)) 11001da177e4SLinus Torvalds BUG(); 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 11031da177e4SLinus Torvalds 1104c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1105c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1106c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1107c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 11081da177e4SLinus Torvalds if (!data) 11091da177e4SLinus Torvalds goto nodata; 111087151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 11111da177e4SLinus Torvalds 11121da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 11136602cebbSEric Dumazet * optimized for the cases when header is void. 11146602cebbSEric Dumazet */ 11156602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 11166602cebbSEric Dumazet 11176602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 11186602cebbSEric Dumazet skb_shinfo(skb), 1119fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 11201da177e4SLinus Torvalds 11213e24591aSAlexander Duyck /* 11223e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 11233e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 11243e24591aSAlexander Duyck * be since all we did is relocate the values 11253e24591aSAlexander Duyck */ 11263e24591aSAlexander Duyck if (skb_cloned(skb)) { 1127a6686f2fSShirley Ma /* copy this zero copy skb frags */ 112870008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1129a6686f2fSShirley Ma goto nofrags; 11301da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1131ea2ab693SIan Campbell skb_frag_ref(skb, i); 11321da177e4SLinus Torvalds 113321dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 11341da177e4SLinus Torvalds skb_clone_fraglist(skb); 11351da177e4SLinus Torvalds 11361da177e4SLinus Torvalds skb_release_data(skb); 11373e24591aSAlexander Duyck } else { 11383e24591aSAlexander Duyck skb_free_head(skb); 11391fd63041SEric Dumazet } 11401da177e4SLinus Torvalds off = (data + nhead) - skb->head; 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds skb->head = data; 1143d3836f21SEric Dumazet skb->head_frag = 0; 11441da177e4SLinus Torvalds skb->data += off; 11454305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 11464305b541SArnaldo Carvalho de Melo skb->end = size; 114756eb8882SPatrick McHardy off = nhead; 11484305b541SArnaldo Carvalho de Melo #else 11494305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 115056eb8882SPatrick McHardy #endif 115127a884dcSArnaldo Carvalho de Melo skb->tail += off; 1152b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 11531da177e4SLinus Torvalds skb->cloned = 0; 1154334a8132SPatrick McHardy skb->hdr_len = 0; 11551da177e4SLinus Torvalds skb->nohdr = 0; 11561da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 11571da177e4SLinus Torvalds return 0; 11581da177e4SLinus Torvalds 1159a6686f2fSShirley Ma nofrags: 1160a6686f2fSShirley Ma kfree(data); 11611da177e4SLinus Torvalds nodata: 11621da177e4SLinus Torvalds return -ENOMEM; 11631da177e4SLinus Torvalds } 1164b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 11691da177e4SLinus Torvalds { 11701da177e4SLinus Torvalds struct sk_buff *skb2; 11711da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 11721da177e4SLinus Torvalds 11731da177e4SLinus Torvalds if (delta <= 0) 11741da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 11751da177e4SLinus Torvalds else { 11761da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 11771da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 11781da177e4SLinus Torvalds GFP_ATOMIC)) { 11791da177e4SLinus Torvalds kfree_skb(skb2); 11801da177e4SLinus Torvalds skb2 = NULL; 11811da177e4SLinus Torvalds } 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds return skb2; 11841da177e4SLinus Torvalds } 1185b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 11861da177e4SLinus Torvalds 11871da177e4SLinus Torvalds /** 11881da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 11891da177e4SLinus Torvalds * @skb: buffer to copy 11901da177e4SLinus Torvalds * @newheadroom: new free bytes at head 11911da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 11921da177e4SLinus Torvalds * @gfp_mask: allocation priority 11931da177e4SLinus Torvalds * 11941da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 11951da177e4SLinus Torvalds * allocate additional space. 11961da177e4SLinus Torvalds * 11971da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 11981da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 11991da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 12001da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 12011da177e4SLinus Torvalds * 12021da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 12031da177e4SLinus Torvalds * is called from an interrupt. 12041da177e4SLinus Torvalds */ 12051da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 120686a76cafSVictor Fusco int newheadroom, int newtailroom, 1207dd0fc66fSAl Viro gfp_t gfp_mask) 12081da177e4SLinus Torvalds { 12091da177e4SLinus Torvalds /* 12101da177e4SLinus Torvalds * Allocate the copy buffer 12111da177e4SLinus Torvalds */ 1212c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1213c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1214c93bdd0eSMel Gorman NUMA_NO_NODE); 1215efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 12161da177e4SLinus Torvalds int head_copy_len, head_copy_off; 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds if (!n) 12191da177e4SLinus Torvalds return NULL; 12201da177e4SLinus Torvalds 12211da177e4SLinus Torvalds skb_reserve(n, newheadroom); 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds /* Set the tail pointer and length */ 12241da177e4SLinus Torvalds skb_put(n, skb->len); 12251da177e4SLinus Torvalds 1226efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 12271da177e4SLinus Torvalds head_copy_off = 0; 12281da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 12291da177e4SLinus Torvalds head_copy_len = newheadroom; 12301da177e4SLinus Torvalds else 12311da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds /* Copy the linear header and data. */ 12341da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 12351da177e4SLinus Torvalds skb->len + head_copy_len)) 12361da177e4SLinus Torvalds BUG(); 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds copy_skb_header(n, skb); 12391da177e4SLinus Torvalds 1240030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1241efd1e8d5SPatrick McHardy 12421da177e4SLinus Torvalds return n; 12431da177e4SLinus Torvalds } 1244b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 12451da177e4SLinus Torvalds 12461da177e4SLinus Torvalds /** 12471da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 12481da177e4SLinus Torvalds * @skb: buffer to pad 12491da177e4SLinus Torvalds * @pad: space to pad 12501da177e4SLinus Torvalds * 12511da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 12521da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 12531da177e4SLinus Torvalds * beyond the buffer end onto the wire. 12541da177e4SLinus Torvalds * 12555b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 12561da177e4SLinus Torvalds */ 12571da177e4SLinus Torvalds 12585b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 12591da177e4SLinus Torvalds { 12605b057c6bSHerbert Xu int err; 12615b057c6bSHerbert Xu int ntail; 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 12645b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 12651da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 12665b057c6bSHerbert Xu return 0; 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds 12694305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 12705b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 12715b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 12725b057c6bSHerbert Xu if (unlikely(err)) 12735b057c6bSHerbert Xu goto free_skb; 12745b057c6bSHerbert Xu } 12755b057c6bSHerbert Xu 12765b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 12775b057c6bSHerbert Xu * to be audited. 12785b057c6bSHerbert Xu */ 12795b057c6bSHerbert Xu err = skb_linearize(skb); 12805b057c6bSHerbert Xu if (unlikely(err)) 12815b057c6bSHerbert Xu goto free_skb; 12825b057c6bSHerbert Xu 12835b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 12845b057c6bSHerbert Xu return 0; 12855b057c6bSHerbert Xu 12865b057c6bSHerbert Xu free_skb: 12871da177e4SLinus Torvalds kfree_skb(skb); 12885b057c6bSHerbert Xu return err; 12891da177e4SLinus Torvalds } 1290b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 12911da177e4SLinus Torvalds 12920dde3e16SIlpo Järvinen /** 12930c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 12940c7ddf36SMathias Krause * @skb: start of the buffer to use 12950c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 12960c7ddf36SMathias Krause * @len: amount of data to add 12970c7ddf36SMathias Krause * 12980c7ddf36SMathias Krause * This function extends the used data area of the potentially 12990c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 13000c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 13010c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 13020c7ddf36SMathias Krause * returned. 13030c7ddf36SMathias Krause */ 13040c7ddf36SMathias Krause 13050c7ddf36SMathias Krause unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 13060c7ddf36SMathias Krause { 13070c7ddf36SMathias Krause if (tail != skb) { 13080c7ddf36SMathias Krause skb->data_len += len; 13090c7ddf36SMathias Krause skb->len += len; 13100c7ddf36SMathias Krause } 13110c7ddf36SMathias Krause return skb_put(tail, len); 13120c7ddf36SMathias Krause } 13130c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 13140c7ddf36SMathias Krause 13150c7ddf36SMathias Krause /** 13160dde3e16SIlpo Järvinen * skb_put - add data to a buffer 13170dde3e16SIlpo Järvinen * @skb: buffer to use 13180dde3e16SIlpo Järvinen * @len: amount of data to add 13190dde3e16SIlpo Järvinen * 13200dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 13210dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 13220dde3e16SIlpo Järvinen * first byte of the extra data is returned. 13230dde3e16SIlpo Järvinen */ 13240dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 13250dde3e16SIlpo Järvinen { 13260dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 13270dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 13280dde3e16SIlpo Järvinen skb->tail += len; 13290dde3e16SIlpo Järvinen skb->len += len; 13300dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 13310dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 13320dde3e16SIlpo Järvinen return tmp; 13330dde3e16SIlpo Järvinen } 13340dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 13350dde3e16SIlpo Järvinen 13366be8ac2fSIlpo Järvinen /** 1337c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1338c2aa270aSIlpo Järvinen * @skb: buffer to use 1339c2aa270aSIlpo Järvinen * @len: amount of data to add 1340c2aa270aSIlpo Järvinen * 1341c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1342c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1343c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1344c2aa270aSIlpo Järvinen */ 1345c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1346c2aa270aSIlpo Järvinen { 1347c2aa270aSIlpo Järvinen skb->data -= len; 1348c2aa270aSIlpo Järvinen skb->len += len; 1349c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1350c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1351c2aa270aSIlpo Järvinen return skb->data; 1352c2aa270aSIlpo Järvinen } 1353c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1354c2aa270aSIlpo Järvinen 1355c2aa270aSIlpo Järvinen /** 13566be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 13576be8ac2fSIlpo Järvinen * @skb: buffer to use 13586be8ac2fSIlpo Järvinen * @len: amount of data to remove 13596be8ac2fSIlpo Järvinen * 13606be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 13616be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 13626be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 13636be8ac2fSIlpo Järvinen * the old data. 13646be8ac2fSIlpo Järvinen */ 13656be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 13666be8ac2fSIlpo Järvinen { 136747d29646SDavid S. Miller return skb_pull_inline(skb, len); 13686be8ac2fSIlpo Järvinen } 13696be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 13706be8ac2fSIlpo Järvinen 1371419ae74eSIlpo Järvinen /** 1372419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1373419ae74eSIlpo Järvinen * @skb: buffer to alter 1374419ae74eSIlpo Järvinen * @len: new length 1375419ae74eSIlpo Järvinen * 1376419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1377419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1378419ae74eSIlpo Järvinen * The skb must be linear. 1379419ae74eSIlpo Järvinen */ 1380419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1381419ae74eSIlpo Järvinen { 1382419ae74eSIlpo Järvinen if (skb->len > len) 1383419ae74eSIlpo Järvinen __skb_trim(skb, len); 1384419ae74eSIlpo Järvinen } 1385419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1386419ae74eSIlpo Järvinen 13873cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 13881da177e4SLinus Torvalds */ 13891da177e4SLinus Torvalds 13903cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 13911da177e4SLinus Torvalds { 139227b437c8SHerbert Xu struct sk_buff **fragp; 139327b437c8SHerbert Xu struct sk_buff *frag; 13941da177e4SLinus Torvalds int offset = skb_headlen(skb); 13951da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 13961da177e4SLinus Torvalds int i; 139727b437c8SHerbert Xu int err; 139827b437c8SHerbert Xu 139927b437c8SHerbert Xu if (skb_cloned(skb) && 140027b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 140127b437c8SHerbert Xu return err; 14021da177e4SLinus Torvalds 1403f4d26fb3SHerbert Xu i = 0; 1404f4d26fb3SHerbert Xu if (offset >= len) 1405f4d26fb3SHerbert Xu goto drop_pages; 1406f4d26fb3SHerbert Xu 1407f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 14089e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 140927b437c8SHerbert Xu 141027b437c8SHerbert Xu if (end < len) { 14111da177e4SLinus Torvalds offset = end; 141227b437c8SHerbert Xu continue; 14131da177e4SLinus Torvalds } 14141da177e4SLinus Torvalds 14159e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 141627b437c8SHerbert Xu 1417f4d26fb3SHerbert Xu drop_pages: 141827b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 141927b437c8SHerbert Xu 142027b437c8SHerbert Xu for (; i < nfrags; i++) 1421ea2ab693SIan Campbell skb_frag_unref(skb, i); 142227b437c8SHerbert Xu 142321dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 142427b437c8SHerbert Xu skb_drop_fraglist(skb); 1425f4d26fb3SHerbert Xu goto done; 142627b437c8SHerbert Xu } 142727b437c8SHerbert Xu 142827b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 142927b437c8SHerbert Xu fragp = &frag->next) { 143027b437c8SHerbert Xu int end = offset + frag->len; 143127b437c8SHerbert Xu 143227b437c8SHerbert Xu if (skb_shared(frag)) { 143327b437c8SHerbert Xu struct sk_buff *nfrag; 143427b437c8SHerbert Xu 143527b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 143627b437c8SHerbert Xu if (unlikely(!nfrag)) 143727b437c8SHerbert Xu return -ENOMEM; 143827b437c8SHerbert Xu 143927b437c8SHerbert Xu nfrag->next = frag->next; 144085bb2a60SEric Dumazet consume_skb(frag); 144127b437c8SHerbert Xu frag = nfrag; 144227b437c8SHerbert Xu *fragp = frag; 144327b437c8SHerbert Xu } 144427b437c8SHerbert Xu 144527b437c8SHerbert Xu if (end < len) { 144627b437c8SHerbert Xu offset = end; 144727b437c8SHerbert Xu continue; 144827b437c8SHerbert Xu } 144927b437c8SHerbert Xu 145027b437c8SHerbert Xu if (end > len && 145127b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 145227b437c8SHerbert Xu return err; 145327b437c8SHerbert Xu 145427b437c8SHerbert Xu if (frag->next) 145527b437c8SHerbert Xu skb_drop_list(&frag->next); 145627b437c8SHerbert Xu break; 145727b437c8SHerbert Xu } 145827b437c8SHerbert Xu 1459f4d26fb3SHerbert Xu done: 146027b437c8SHerbert Xu if (len > skb_headlen(skb)) { 14611da177e4SLinus Torvalds skb->data_len -= skb->len - len; 14621da177e4SLinus Torvalds skb->len = len; 14631da177e4SLinus Torvalds } else { 14641da177e4SLinus Torvalds skb->len = len; 14651da177e4SLinus Torvalds skb->data_len = 0; 146627a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 14671da177e4SLinus Torvalds } 14681da177e4SLinus Torvalds 14691da177e4SLinus Torvalds return 0; 14701da177e4SLinus Torvalds } 1471b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds /** 14741da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 14751da177e4SLinus Torvalds * @skb: buffer to reallocate 14761da177e4SLinus Torvalds * @delta: number of bytes to advance tail 14771da177e4SLinus Torvalds * 14781da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 14791da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 14801da177e4SLinus Torvalds * data from fragmented part. 14811da177e4SLinus Torvalds * 14821da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 14831da177e4SLinus Torvalds * 14841da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 14851da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 14861da177e4SLinus Torvalds * 14871da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 14881da177e4SLinus Torvalds * reloaded after call to this function. 14891da177e4SLinus Torvalds */ 14901da177e4SLinus Torvalds 14911da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 14921da177e4SLinus Torvalds * when it is necessary. 14931da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 14941da177e4SLinus Torvalds * 2. It may change skb pointers. 14951da177e4SLinus Torvalds * 14961da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 14971da177e4SLinus Torvalds */ 14981da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 14991da177e4SLinus Torvalds { 15001da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 15011da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 15021da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 15031da177e4SLinus Torvalds */ 15044305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 15051da177e4SLinus Torvalds 15061da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 15071da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 15081da177e4SLinus Torvalds GFP_ATOMIC)) 15091da177e4SLinus Torvalds return NULL; 15101da177e4SLinus Torvalds } 15111da177e4SLinus Torvalds 151227a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 15131da177e4SLinus Torvalds BUG(); 15141da177e4SLinus Torvalds 15151da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 15161da177e4SLinus Torvalds * size of pulled pages. Superb. 15171da177e4SLinus Torvalds */ 151821dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 15191da177e4SLinus Torvalds goto pull_pages; 15201da177e4SLinus Torvalds 15211da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 15221da177e4SLinus Torvalds eat = delta; 15231da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 15249e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 15259e903e08SEric Dumazet 15269e903e08SEric Dumazet if (size >= eat) 15271da177e4SLinus Torvalds goto pull_pages; 15289e903e08SEric Dumazet eat -= size; 15291da177e4SLinus Torvalds } 15301da177e4SLinus Torvalds 15311da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 15321da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 15331da177e4SLinus Torvalds * but taking into account that pulling is expected to 15341da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 15351da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 15361da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 15371da177e4SLinus Torvalds */ 15381da177e4SLinus Torvalds if (eat) { 15391da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 15401da177e4SLinus Torvalds struct sk_buff *clone = NULL; 15411da177e4SLinus Torvalds struct sk_buff *insp = NULL; 15421da177e4SLinus Torvalds 15431da177e4SLinus Torvalds do { 154409a62660SKris Katterjohn BUG_ON(!list); 15451da177e4SLinus Torvalds 15461da177e4SLinus Torvalds if (list->len <= eat) { 15471da177e4SLinus Torvalds /* Eaten as whole. */ 15481da177e4SLinus Torvalds eat -= list->len; 15491da177e4SLinus Torvalds list = list->next; 15501da177e4SLinus Torvalds insp = list; 15511da177e4SLinus Torvalds } else { 15521da177e4SLinus Torvalds /* Eaten partially. */ 15531da177e4SLinus Torvalds 15541da177e4SLinus Torvalds if (skb_shared(list)) { 15551da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 15561da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 15571da177e4SLinus Torvalds if (!clone) 15581da177e4SLinus Torvalds return NULL; 15591da177e4SLinus Torvalds insp = list->next; 15601da177e4SLinus Torvalds list = clone; 15611da177e4SLinus Torvalds } else { 15621da177e4SLinus Torvalds /* This may be pulled without 15631da177e4SLinus Torvalds * problems. */ 15641da177e4SLinus Torvalds insp = list; 15651da177e4SLinus Torvalds } 15661da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 15671da177e4SLinus Torvalds kfree_skb(clone); 15681da177e4SLinus Torvalds return NULL; 15691da177e4SLinus Torvalds } 15701da177e4SLinus Torvalds break; 15711da177e4SLinus Torvalds } 15721da177e4SLinus Torvalds } while (eat); 15731da177e4SLinus Torvalds 15741da177e4SLinus Torvalds /* Free pulled out fragments. */ 15751da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 15761da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 15771da177e4SLinus Torvalds kfree_skb(list); 15781da177e4SLinus Torvalds } 15791da177e4SLinus Torvalds /* And insert new clone at head. */ 15801da177e4SLinus Torvalds if (clone) { 15811da177e4SLinus Torvalds clone->next = list; 15821da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 15831da177e4SLinus Torvalds } 15841da177e4SLinus Torvalds } 15851da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 15861da177e4SLinus Torvalds 15871da177e4SLinus Torvalds pull_pages: 15881da177e4SLinus Torvalds eat = delta; 15891da177e4SLinus Torvalds k = 0; 15901da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 15919e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 15929e903e08SEric Dumazet 15939e903e08SEric Dumazet if (size <= eat) { 1594ea2ab693SIan Campbell skb_frag_unref(skb, i); 15959e903e08SEric Dumazet eat -= size; 15961da177e4SLinus Torvalds } else { 15971da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 15981da177e4SLinus Torvalds if (eat) { 15991da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 16009e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 16011da177e4SLinus Torvalds eat = 0; 16021da177e4SLinus Torvalds } 16031da177e4SLinus Torvalds k++; 16041da177e4SLinus Torvalds } 16051da177e4SLinus Torvalds } 16061da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds skb->tail += delta; 16091da177e4SLinus Torvalds skb->data_len -= delta; 16101da177e4SLinus Torvalds 161127a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 16121da177e4SLinus Torvalds } 1613b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 16141da177e4SLinus Torvalds 161522019b17SEric Dumazet /** 161622019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 161722019b17SEric Dumazet * @skb: source skb 161822019b17SEric Dumazet * @offset: offset in source 161922019b17SEric Dumazet * @to: destination buffer 162022019b17SEric Dumazet * @len: number of bytes to copy 162122019b17SEric Dumazet * 162222019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 162322019b17SEric Dumazet * destination buffer. 162422019b17SEric Dumazet * 162522019b17SEric Dumazet * CAUTION ! : 162622019b17SEric Dumazet * If its prototype is ever changed, 162722019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 162822019b17SEric Dumazet * since it is called from BPF assembly code. 162922019b17SEric Dumazet */ 16301da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 16311da177e4SLinus Torvalds { 16321a028e50SDavid S. Miller int start = skb_headlen(skb); 1633fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1634fbb398a8SDavid S. Miller int i, copy; 16351da177e4SLinus Torvalds 16361da177e4SLinus Torvalds if (offset > (int)skb->len - len) 16371da177e4SLinus Torvalds goto fault; 16381da177e4SLinus Torvalds 16391da177e4SLinus Torvalds /* Copy header. */ 16401a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 16411da177e4SLinus Torvalds if (copy > len) 16421da177e4SLinus Torvalds copy = len; 1643d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 16441da177e4SLinus Torvalds if ((len -= copy) == 0) 16451da177e4SLinus Torvalds return 0; 16461da177e4SLinus Torvalds offset += copy; 16471da177e4SLinus Torvalds to += copy; 16481da177e4SLinus Torvalds } 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16511a028e50SDavid S. Miller int end; 165251c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 16531da177e4SLinus Torvalds 1654547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16551a028e50SDavid S. Miller 165651c56b00SEric Dumazet end = start + skb_frag_size(f); 16571da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 16581da177e4SLinus Torvalds u8 *vaddr; 16591da177e4SLinus Torvalds 16601da177e4SLinus Torvalds if (copy > len) 16611da177e4SLinus Torvalds copy = len; 16621da177e4SLinus Torvalds 166351c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 16641da177e4SLinus Torvalds memcpy(to, 166551c56b00SEric Dumazet vaddr + f->page_offset + offset - start, 166651c56b00SEric Dumazet copy); 166751c56b00SEric Dumazet kunmap_atomic(vaddr); 16681da177e4SLinus Torvalds 16691da177e4SLinus Torvalds if ((len -= copy) == 0) 16701da177e4SLinus Torvalds return 0; 16711da177e4SLinus Torvalds offset += copy; 16721da177e4SLinus Torvalds to += copy; 16731da177e4SLinus Torvalds } 16741a028e50SDavid S. Miller start = end; 16751da177e4SLinus Torvalds } 16761da177e4SLinus Torvalds 1677fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 16781a028e50SDavid S. Miller int end; 16791da177e4SLinus Torvalds 1680547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16811a028e50SDavid S. Miller 1682fbb398a8SDavid S. Miller end = start + frag_iter->len; 16831da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 16841da177e4SLinus Torvalds if (copy > len) 16851da177e4SLinus Torvalds copy = len; 1686fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 16871da177e4SLinus Torvalds goto fault; 16881da177e4SLinus Torvalds if ((len -= copy) == 0) 16891da177e4SLinus Torvalds return 0; 16901da177e4SLinus Torvalds offset += copy; 16911da177e4SLinus Torvalds to += copy; 16921da177e4SLinus Torvalds } 16931a028e50SDavid S. Miller start = end; 16941da177e4SLinus Torvalds } 1695a6686f2fSShirley Ma 16961da177e4SLinus Torvalds if (!len) 16971da177e4SLinus Torvalds return 0; 16981da177e4SLinus Torvalds 16991da177e4SLinus Torvalds fault: 17001da177e4SLinus Torvalds return -EFAULT; 17011da177e4SLinus Torvalds } 1702b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 17031da177e4SLinus Torvalds 17049c55e01cSJens Axboe /* 17059c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 17069c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 17079c55e01cSJens Axboe */ 17089c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 17099c55e01cSJens Axboe { 17108b9d3728SJarek Poplawski put_page(spd->pages[i]); 17118b9d3728SJarek Poplawski } 17129c55e01cSJens Axboe 1713a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 17144fb66994SJarek Poplawski unsigned int *offset, 171518aafc62SEric Dumazet struct sock *sk) 17168b9d3728SJarek Poplawski { 17175640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 17188b9d3728SJarek Poplawski 17195640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 17208b9d3728SJarek Poplawski return NULL; 17214fb66994SJarek Poplawski 17225640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 17234fb66994SJarek Poplawski 17245640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 17255640f768SEric Dumazet page_address(page) + *offset, *len); 17265640f768SEric Dumazet *offset = pfrag->offset; 17275640f768SEric Dumazet pfrag->offset += *len; 17284fb66994SJarek Poplawski 17295640f768SEric Dumazet return pfrag->page; 17309c55e01cSJens Axboe } 17319c55e01cSJens Axboe 173241c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 173341c73a0dSEric Dumazet struct page *page, 173441c73a0dSEric Dumazet unsigned int offset) 173541c73a0dSEric Dumazet { 173641c73a0dSEric Dumazet return spd->nr_pages && 173741c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 173841c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 173941c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 174041c73a0dSEric Dumazet } 174141c73a0dSEric Dumazet 17429c55e01cSJens Axboe /* 17439c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 17449c55e01cSJens Axboe */ 1745a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 174635f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 17474fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 174818aafc62SEric Dumazet bool linear, 17497a67e56fSJarek Poplawski struct sock *sk) 17509c55e01cSJens Axboe { 175141c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1752a108d5f3SDavid S. Miller return true; 17539c55e01cSJens Axboe 17548b9d3728SJarek Poplawski if (linear) { 175518aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 17568b9d3728SJarek Poplawski if (!page) 1757a108d5f3SDavid S. Miller return true; 175841c73a0dSEric Dumazet } 175941c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 176041c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 1761a108d5f3SDavid S. Miller return false; 176241c73a0dSEric Dumazet } 17638b9d3728SJarek Poplawski get_page(page); 17649c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 17654fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 17669c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 17679c55e01cSJens Axboe spd->nr_pages++; 17688b9d3728SJarek Poplawski 1769a108d5f3SDavid S. Miller return false; 17709c55e01cSJens Axboe } 17719c55e01cSJens Axboe 1772a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 17732870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 177418aafc62SEric Dumazet unsigned int *len, 1775d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 177635f3d14dSJens Axboe struct sock *sk, 177735f3d14dSJens Axboe struct pipe_inode_info *pipe) 17789c55e01cSJens Axboe { 17792870c43dSOctavian Purdila if (!*len) 1780a108d5f3SDavid S. Miller return true; 17819c55e01cSJens Axboe 17822870c43dSOctavian Purdila /* skip this segment if already processed */ 17832870c43dSOctavian Purdila if (*off >= plen) { 17842870c43dSOctavian Purdila *off -= plen; 1785a108d5f3SDavid S. Miller return false; 17862870c43dSOctavian Purdila } 17872870c43dSOctavian Purdila 17882870c43dSOctavian Purdila /* ignore any bits we already processed */ 17899ca1b22dSEric Dumazet poff += *off; 17909ca1b22dSEric Dumazet plen -= *off; 17912870c43dSOctavian Purdila *off = 0; 17922870c43dSOctavian Purdila 179318aafc62SEric Dumazet do { 179418aafc62SEric Dumazet unsigned int flen = min(*len, plen); 17952870c43dSOctavian Purdila 179618aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 179718aafc62SEric Dumazet linear, sk)) 1798a108d5f3SDavid S. Miller return true; 179918aafc62SEric Dumazet poff += flen; 180018aafc62SEric Dumazet plen -= flen; 18012870c43dSOctavian Purdila *len -= flen; 180218aafc62SEric Dumazet } while (*len && plen); 18032870c43dSOctavian Purdila 1804a108d5f3SDavid S. Miller return false; 1805db43a282SOctavian Purdila } 18069c55e01cSJens Axboe 18079c55e01cSJens Axboe /* 1808a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 18092870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 18109c55e01cSJens Axboe */ 1811a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 181235f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 181335f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 18142870c43dSOctavian Purdila { 18152870c43dSOctavian Purdila int seg; 18169c55e01cSJens Axboe 18171d0c0b32SEric Dumazet /* map the linear part : 18182996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 18192996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 18202996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 18219c55e01cSJens Axboe */ 18222870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 18232870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 18242870c43dSOctavian Purdila skb_headlen(skb), 182518aafc62SEric Dumazet offset, len, spd, 18263a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 18271d0c0b32SEric Dumazet sk, pipe)) 1828a108d5f3SDavid S. Miller return true; 18299c55e01cSJens Axboe 18309c55e01cSJens Axboe /* 18319c55e01cSJens Axboe * then map the fragments 18329c55e01cSJens Axboe */ 18339c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 18349c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 18359c55e01cSJens Axboe 1836ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 18379e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 183818aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 1839a108d5f3SDavid S. Miller return true; 18409c55e01cSJens Axboe } 18419c55e01cSJens Axboe 1842a108d5f3SDavid S. Miller return false; 18439c55e01cSJens Axboe } 18449c55e01cSJens Axboe 18459c55e01cSJens Axboe /* 18469c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 18479c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 18489c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 18499c55e01cSJens Axboe * handle that cleanly. 18509c55e01cSJens Axboe */ 18518b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 18529c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 18539c55e01cSJens Axboe unsigned int flags) 18549c55e01cSJens Axboe { 185541c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 185641c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 18579c55e01cSJens Axboe struct splice_pipe_desc spd = { 18589c55e01cSJens Axboe .pages = pages, 18599c55e01cSJens Axboe .partial = partial, 1860047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 18619c55e01cSJens Axboe .flags = flags, 186228a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 18639c55e01cSJens Axboe .spd_release = sock_spd_release, 18649c55e01cSJens Axboe }; 1865fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 18667a67e56fSJarek Poplawski struct sock *sk = skb->sk; 186735f3d14dSJens Axboe int ret = 0; 186835f3d14dSJens Axboe 18699c55e01cSJens Axboe /* 18709c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 18719c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 18729c55e01cSJens Axboe */ 187335f3d14dSJens Axboe if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 18749c55e01cSJens Axboe goto done; 18759c55e01cSJens Axboe else if (!tlen) 18769c55e01cSJens Axboe goto done; 18779c55e01cSJens Axboe 18789c55e01cSJens Axboe /* 18799c55e01cSJens Axboe * now see if we have a frag_list to map 18809c55e01cSJens Axboe */ 1881fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 1882fbb398a8SDavid S. Miller if (!tlen) 18839c55e01cSJens Axboe break; 188435f3d14dSJens Axboe if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1885fbb398a8SDavid S. Miller break; 18869c55e01cSJens Axboe } 18879c55e01cSJens Axboe 18889c55e01cSJens Axboe done: 18899c55e01cSJens Axboe if (spd.nr_pages) { 18909c55e01cSJens Axboe /* 18919c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 18929c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 18939c55e01cSJens Axboe * here as compared to sendfile(). We enter here 18949c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 18959c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 18969c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 18979c55e01cSJens Axboe * and networking will grab the socket lock. 18989c55e01cSJens Axboe */ 1899293ad604SOctavian Purdila release_sock(sk); 19009c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1901293ad604SOctavian Purdila lock_sock(sk); 19029c55e01cSJens Axboe } 19039c55e01cSJens Axboe 190435f3d14dSJens Axboe return ret; 19059c55e01cSJens Axboe } 19069c55e01cSJens Axboe 1907357b40a1SHerbert Xu /** 1908357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1909357b40a1SHerbert Xu * @skb: destination buffer 1910357b40a1SHerbert Xu * @offset: offset in destination 1911357b40a1SHerbert Xu * @from: source buffer 1912357b40a1SHerbert Xu * @len: number of bytes to copy 1913357b40a1SHerbert Xu * 1914357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1915357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1916357b40a1SHerbert Xu * traversing fragment lists and such. 1917357b40a1SHerbert Xu */ 1918357b40a1SHerbert Xu 19190c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1920357b40a1SHerbert Xu { 19211a028e50SDavid S. Miller int start = skb_headlen(skb); 1922fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1923fbb398a8SDavid S. Miller int i, copy; 1924357b40a1SHerbert Xu 1925357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1926357b40a1SHerbert Xu goto fault; 1927357b40a1SHerbert Xu 19281a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1929357b40a1SHerbert Xu if (copy > len) 1930357b40a1SHerbert Xu copy = len; 193127d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1932357b40a1SHerbert Xu if ((len -= copy) == 0) 1933357b40a1SHerbert Xu return 0; 1934357b40a1SHerbert Xu offset += copy; 1935357b40a1SHerbert Xu from += copy; 1936357b40a1SHerbert Xu } 1937357b40a1SHerbert Xu 1938357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1939357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 19401a028e50SDavid S. Miller int end; 1941357b40a1SHerbert Xu 1942547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19431a028e50SDavid S. Miller 19449e903e08SEric Dumazet end = start + skb_frag_size(frag); 1945357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1946357b40a1SHerbert Xu u8 *vaddr; 1947357b40a1SHerbert Xu 1948357b40a1SHerbert Xu if (copy > len) 1949357b40a1SHerbert Xu copy = len; 1950357b40a1SHerbert Xu 195151c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 19521a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 19531a028e50SDavid S. Miller from, copy); 195451c56b00SEric Dumazet kunmap_atomic(vaddr); 1955357b40a1SHerbert Xu 1956357b40a1SHerbert Xu if ((len -= copy) == 0) 1957357b40a1SHerbert Xu return 0; 1958357b40a1SHerbert Xu offset += copy; 1959357b40a1SHerbert Xu from += copy; 1960357b40a1SHerbert Xu } 19611a028e50SDavid S. Miller start = end; 1962357b40a1SHerbert Xu } 1963357b40a1SHerbert Xu 1964fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 19651a028e50SDavid S. Miller int end; 1966357b40a1SHerbert Xu 1967547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19681a028e50SDavid S. Miller 1969fbb398a8SDavid S. Miller end = start + frag_iter->len; 1970357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1971357b40a1SHerbert Xu if (copy > len) 1972357b40a1SHerbert Xu copy = len; 1973fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 19741a028e50SDavid S. Miller from, copy)) 1975357b40a1SHerbert Xu goto fault; 1976357b40a1SHerbert Xu if ((len -= copy) == 0) 1977357b40a1SHerbert Xu return 0; 1978357b40a1SHerbert Xu offset += copy; 1979357b40a1SHerbert Xu from += copy; 1980357b40a1SHerbert Xu } 19811a028e50SDavid S. Miller start = end; 1982357b40a1SHerbert Xu } 1983357b40a1SHerbert Xu if (!len) 1984357b40a1SHerbert Xu return 0; 1985357b40a1SHerbert Xu 1986357b40a1SHerbert Xu fault: 1987357b40a1SHerbert Xu return -EFAULT; 1988357b40a1SHerbert Xu } 1989357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1990357b40a1SHerbert Xu 19911da177e4SLinus Torvalds /* Checksum skb data. */ 19922817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 19932817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 19941da177e4SLinus Torvalds { 19951a028e50SDavid S. Miller int start = skb_headlen(skb); 19961a028e50SDavid S. Miller int i, copy = start - offset; 1997fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 19981da177e4SLinus Torvalds int pos = 0; 19991da177e4SLinus Torvalds 20001da177e4SLinus Torvalds /* Checksum header. */ 20011da177e4SLinus Torvalds if (copy > 0) { 20021da177e4SLinus Torvalds if (copy > len) 20031da177e4SLinus Torvalds copy = len; 20042817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 20051da177e4SLinus Torvalds if ((len -= copy) == 0) 20061da177e4SLinus Torvalds return csum; 20071da177e4SLinus Torvalds offset += copy; 20081da177e4SLinus Torvalds pos = copy; 20091da177e4SLinus Torvalds } 20101da177e4SLinus Torvalds 20111da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20121a028e50SDavid S. Miller int end; 201351c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 20141da177e4SLinus Torvalds 2015547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20161a028e50SDavid S. Miller 201751c56b00SEric Dumazet end = start + skb_frag_size(frag); 20181da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 201944bb9363SAl Viro __wsum csum2; 20201da177e4SLinus Torvalds u8 *vaddr; 20211da177e4SLinus Torvalds 20221da177e4SLinus Torvalds if (copy > len) 20231da177e4SLinus Torvalds copy = len; 202451c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 20252817a336SDaniel Borkmann csum2 = ops->update(vaddr + frag->page_offset + 20261a028e50SDavid S. Miller offset - start, copy, 0); 202751c56b00SEric Dumazet kunmap_atomic(vaddr); 20282817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 20291da177e4SLinus Torvalds if (!(len -= copy)) 20301da177e4SLinus Torvalds return csum; 20311da177e4SLinus Torvalds offset += copy; 20321da177e4SLinus Torvalds pos += copy; 20331da177e4SLinus Torvalds } 20341a028e50SDavid S. Miller start = end; 20351da177e4SLinus Torvalds } 20361da177e4SLinus Torvalds 2037fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 20381a028e50SDavid S. Miller int end; 20391da177e4SLinus Torvalds 2040547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20411a028e50SDavid S. Miller 2042fbb398a8SDavid S. Miller end = start + frag_iter->len; 20431da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 20445f92a738SAl Viro __wsum csum2; 20451da177e4SLinus Torvalds if (copy > len) 20461da177e4SLinus Torvalds copy = len; 20472817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 20482817a336SDaniel Borkmann copy, 0, ops); 20492817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 20501da177e4SLinus Torvalds if ((len -= copy) == 0) 20511da177e4SLinus Torvalds return csum; 20521da177e4SLinus Torvalds offset += copy; 20531da177e4SLinus Torvalds pos += copy; 20541da177e4SLinus Torvalds } 20551a028e50SDavid S. Miller start = end; 20561da177e4SLinus Torvalds } 205709a62660SKris Katterjohn BUG_ON(len); 20581da177e4SLinus Torvalds 20591da177e4SLinus Torvalds return csum; 20601da177e4SLinus Torvalds } 20612817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 20622817a336SDaniel Borkmann 20632817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 20642817a336SDaniel Borkmann int len, __wsum csum) 20652817a336SDaniel Borkmann { 20662817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2067cea80ea8SDaniel Borkmann .update = csum_partial_ext, 20682817a336SDaniel Borkmann .combine = csum_block_add_ext, 20692817a336SDaniel Borkmann }; 20702817a336SDaniel Borkmann 20712817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 20722817a336SDaniel Borkmann } 2073b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 20741da177e4SLinus Torvalds 20751da177e4SLinus Torvalds /* Both of above in one bottle. */ 20761da177e4SLinus Torvalds 207781d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 207881d77662SAl Viro u8 *to, int len, __wsum csum) 20791da177e4SLinus Torvalds { 20801a028e50SDavid S. Miller int start = skb_headlen(skb); 20811a028e50SDavid S. Miller int i, copy = start - offset; 2082fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 20831da177e4SLinus Torvalds int pos = 0; 20841da177e4SLinus Torvalds 20851da177e4SLinus Torvalds /* Copy header. */ 20861da177e4SLinus Torvalds if (copy > 0) { 20871da177e4SLinus Torvalds if (copy > len) 20881da177e4SLinus Torvalds copy = len; 20891da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 20901da177e4SLinus Torvalds copy, csum); 20911da177e4SLinus Torvalds if ((len -= copy) == 0) 20921da177e4SLinus Torvalds return csum; 20931da177e4SLinus Torvalds offset += copy; 20941da177e4SLinus Torvalds to += copy; 20951da177e4SLinus Torvalds pos = copy; 20961da177e4SLinus Torvalds } 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20991a028e50SDavid S. Miller int end; 21001da177e4SLinus Torvalds 2101547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21021a028e50SDavid S. Miller 21039e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 21041da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21055084205fSAl Viro __wsum csum2; 21061da177e4SLinus Torvalds u8 *vaddr; 21071da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 21081da177e4SLinus Torvalds 21091da177e4SLinus Torvalds if (copy > len) 21101da177e4SLinus Torvalds copy = len; 211151c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 21121da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 21131a028e50SDavid S. Miller frag->page_offset + 21141a028e50SDavid S. Miller offset - start, to, 21151a028e50SDavid S. Miller copy, 0); 211651c56b00SEric Dumazet kunmap_atomic(vaddr); 21171da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 21181da177e4SLinus Torvalds if (!(len -= copy)) 21191da177e4SLinus Torvalds return csum; 21201da177e4SLinus Torvalds offset += copy; 21211da177e4SLinus Torvalds to += copy; 21221da177e4SLinus Torvalds pos += copy; 21231da177e4SLinus Torvalds } 21241a028e50SDavid S. Miller start = end; 21251da177e4SLinus Torvalds } 21261da177e4SLinus Torvalds 2127fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 212881d77662SAl Viro __wsum csum2; 21291a028e50SDavid S. Miller int end; 21301da177e4SLinus Torvalds 2131547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21321a028e50SDavid S. Miller 2133fbb398a8SDavid S. Miller end = start + frag_iter->len; 21341da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21351da177e4SLinus Torvalds if (copy > len) 21361da177e4SLinus Torvalds copy = len; 2137fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 21381a028e50SDavid S. Miller offset - start, 21391da177e4SLinus Torvalds to, copy, 0); 21401da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 21411da177e4SLinus Torvalds if ((len -= copy) == 0) 21421da177e4SLinus Torvalds return csum; 21431da177e4SLinus Torvalds offset += copy; 21441da177e4SLinus Torvalds to += copy; 21451da177e4SLinus Torvalds pos += copy; 21461da177e4SLinus Torvalds } 21471a028e50SDavid S. Miller start = end; 21481da177e4SLinus Torvalds } 214909a62660SKris Katterjohn BUG_ON(len); 21501da177e4SLinus Torvalds return csum; 21511da177e4SLinus Torvalds } 2152b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 21531da177e4SLinus Torvalds 2154af2806f8SThomas Graf /** 2155af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2156af2806f8SThomas Graf * @from: source buffer 2157af2806f8SThomas Graf * 2158af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2159af2806f8SThomas Graf * into skb_zerocopy(). 2160af2806f8SThomas Graf */ 2161af2806f8SThomas Graf unsigned int 2162af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2163af2806f8SThomas Graf { 2164af2806f8SThomas Graf unsigned int hlen = 0; 2165af2806f8SThomas Graf 2166af2806f8SThomas Graf if (!from->head_frag || 2167af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2168af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2169af2806f8SThomas Graf hlen = skb_headlen(from); 2170af2806f8SThomas Graf 2171af2806f8SThomas Graf if (skb_has_frag_list(from)) 2172af2806f8SThomas Graf hlen = from->len; 2173af2806f8SThomas Graf 2174af2806f8SThomas Graf return hlen; 2175af2806f8SThomas Graf } 2176af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2177af2806f8SThomas Graf 2178af2806f8SThomas Graf /** 2179af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2180af2806f8SThomas Graf * @to: destination buffer 21817fceb4deSMasanari Iida * @from: source buffer 2182af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2183af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2184af2806f8SThomas Graf * 2185af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2186af2806f8SThomas Graf * to the frags in the source buffer. 2187af2806f8SThomas Graf * 2188af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2189af2806f8SThomas Graf * headroom in the `to` buffer. 219036d5fe6aSZoltan Kiss * 219136d5fe6aSZoltan Kiss * Return value: 219236d5fe6aSZoltan Kiss * 0: everything is OK 219336d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 219436d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2195af2806f8SThomas Graf */ 219636d5fe6aSZoltan Kiss int 219736d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2198af2806f8SThomas Graf { 2199af2806f8SThomas Graf int i, j = 0; 2200af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 220136d5fe6aSZoltan Kiss int ret; 2202af2806f8SThomas Graf struct page *page; 2203af2806f8SThomas Graf unsigned int offset; 2204af2806f8SThomas Graf 2205af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2206af2806f8SThomas Graf 2207af2806f8SThomas Graf /* dont bother with small payloads */ 220836d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 220936d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2210af2806f8SThomas Graf 2211af2806f8SThomas Graf if (hlen) { 221236d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 221336d5fe6aSZoltan Kiss if (unlikely(ret)) 221436d5fe6aSZoltan Kiss return ret; 2215af2806f8SThomas Graf len -= hlen; 2216af2806f8SThomas Graf } else { 2217af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2218af2806f8SThomas Graf if (plen) { 2219af2806f8SThomas Graf page = virt_to_head_page(from->head); 2220af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2221af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2222af2806f8SThomas Graf get_page(page); 2223af2806f8SThomas Graf j = 1; 2224af2806f8SThomas Graf len -= plen; 2225af2806f8SThomas Graf } 2226af2806f8SThomas Graf } 2227af2806f8SThomas Graf 2228af2806f8SThomas Graf to->truesize += len + plen; 2229af2806f8SThomas Graf to->len += len + plen; 2230af2806f8SThomas Graf to->data_len += len + plen; 2231af2806f8SThomas Graf 223236d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 223336d5fe6aSZoltan Kiss skb_tx_error(from); 223436d5fe6aSZoltan Kiss return -ENOMEM; 223536d5fe6aSZoltan Kiss } 223636d5fe6aSZoltan Kiss 2237af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2238af2806f8SThomas Graf if (!len) 2239af2806f8SThomas Graf break; 2240af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2241af2806f8SThomas Graf skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2242af2806f8SThomas Graf len -= skb_shinfo(to)->frags[j].size; 2243af2806f8SThomas Graf skb_frag_ref(to, j); 2244af2806f8SThomas Graf j++; 2245af2806f8SThomas Graf } 2246af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 224736d5fe6aSZoltan Kiss 224836d5fe6aSZoltan Kiss return 0; 2249af2806f8SThomas Graf } 2250af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2251af2806f8SThomas Graf 22521da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 22531da177e4SLinus Torvalds { 2254d3bc23e7SAl Viro __wsum csum; 22551da177e4SLinus Torvalds long csstart; 22561da177e4SLinus Torvalds 225784fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 225855508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 22591da177e4SLinus Torvalds else 22601da177e4SLinus Torvalds csstart = skb_headlen(skb); 22611da177e4SLinus Torvalds 226209a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 22631da177e4SLinus Torvalds 2264d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 22651da177e4SLinus Torvalds 22661da177e4SLinus Torvalds csum = 0; 22671da177e4SLinus Torvalds if (csstart != skb->len) 22681da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 22691da177e4SLinus Torvalds skb->len - csstart, 0); 22701da177e4SLinus Torvalds 227184fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2272ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 22731da177e4SLinus Torvalds 2274d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 22751da177e4SLinus Torvalds } 22761da177e4SLinus Torvalds } 2277b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 22781da177e4SLinus Torvalds 22791da177e4SLinus Torvalds /** 22801da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 22811da177e4SLinus Torvalds * @list: list to dequeue from 22821da177e4SLinus Torvalds * 22831da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 22841da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 22851da177e4SLinus Torvalds * returned or %NULL if the list is empty. 22861da177e4SLinus Torvalds */ 22871da177e4SLinus Torvalds 22881da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 22891da177e4SLinus Torvalds { 22901da177e4SLinus Torvalds unsigned long flags; 22911da177e4SLinus Torvalds struct sk_buff *result; 22921da177e4SLinus Torvalds 22931da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 22941da177e4SLinus Torvalds result = __skb_dequeue(list); 22951da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 22961da177e4SLinus Torvalds return result; 22971da177e4SLinus Torvalds } 2298b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 22991da177e4SLinus Torvalds 23001da177e4SLinus Torvalds /** 23011da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 23021da177e4SLinus Torvalds * @list: list to dequeue from 23031da177e4SLinus Torvalds * 23041da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 23051da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 23061da177e4SLinus Torvalds * returned or %NULL if the list is empty. 23071da177e4SLinus Torvalds */ 23081da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 23091da177e4SLinus Torvalds { 23101da177e4SLinus Torvalds unsigned long flags; 23111da177e4SLinus Torvalds struct sk_buff *result; 23121da177e4SLinus Torvalds 23131da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23141da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 23151da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23161da177e4SLinus Torvalds return result; 23171da177e4SLinus Torvalds } 2318b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 23191da177e4SLinus Torvalds 23201da177e4SLinus Torvalds /** 23211da177e4SLinus Torvalds * skb_queue_purge - empty a list 23221da177e4SLinus Torvalds * @list: list to empty 23231da177e4SLinus Torvalds * 23241da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 23251da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 23261da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 23271da177e4SLinus Torvalds */ 23281da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 23291da177e4SLinus Torvalds { 23301da177e4SLinus Torvalds struct sk_buff *skb; 23311da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 23321da177e4SLinus Torvalds kfree_skb(skb); 23331da177e4SLinus Torvalds } 2334b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 23351da177e4SLinus Torvalds 23361da177e4SLinus Torvalds /** 23371da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 23381da177e4SLinus Torvalds * @list: list to use 23391da177e4SLinus Torvalds * @newsk: buffer to queue 23401da177e4SLinus Torvalds * 23411da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 23421da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 23431da177e4SLinus Torvalds * safely. 23441da177e4SLinus Torvalds * 23451da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 23461da177e4SLinus Torvalds */ 23471da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 23481da177e4SLinus Torvalds { 23491da177e4SLinus Torvalds unsigned long flags; 23501da177e4SLinus Torvalds 23511da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23521da177e4SLinus Torvalds __skb_queue_head(list, newsk); 23531da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23541da177e4SLinus Torvalds } 2355b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 23561da177e4SLinus Torvalds 23571da177e4SLinus Torvalds /** 23581da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 23591da177e4SLinus Torvalds * @list: list to use 23601da177e4SLinus Torvalds * @newsk: buffer to queue 23611da177e4SLinus Torvalds * 23621da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 23631da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 23641da177e4SLinus Torvalds * safely. 23651da177e4SLinus Torvalds * 23661da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 23671da177e4SLinus Torvalds */ 23681da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 23691da177e4SLinus Torvalds { 23701da177e4SLinus Torvalds unsigned long flags; 23711da177e4SLinus Torvalds 23721da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23731da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 23741da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23751da177e4SLinus Torvalds } 2376b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 23778728b834SDavid S. Miller 23781da177e4SLinus Torvalds /** 23791da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 23801da177e4SLinus Torvalds * @skb: buffer to remove 23818728b834SDavid S. Miller * @list: list to use 23821da177e4SLinus Torvalds * 23838728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 23848728b834SDavid S. Miller * function is atomic with respect to other list locked calls 23851da177e4SLinus Torvalds * 23868728b834SDavid S. Miller * You must know what list the SKB is on. 23871da177e4SLinus Torvalds */ 23888728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 23891da177e4SLinus Torvalds { 23901da177e4SLinus Torvalds unsigned long flags; 23911da177e4SLinus Torvalds 23921da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23938728b834SDavid S. Miller __skb_unlink(skb, list); 23941da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23951da177e4SLinus Torvalds } 2396b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 23971da177e4SLinus Torvalds 23981da177e4SLinus Torvalds /** 23991da177e4SLinus Torvalds * skb_append - append a buffer 24001da177e4SLinus Torvalds * @old: buffer to insert after 24011da177e4SLinus Torvalds * @newsk: buffer to insert 24028728b834SDavid S. Miller * @list: list to use 24031da177e4SLinus Torvalds * 24041da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 24051da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 24061da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24071da177e4SLinus Torvalds */ 24088728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 24091da177e4SLinus Torvalds { 24101da177e4SLinus Torvalds unsigned long flags; 24111da177e4SLinus Torvalds 24128728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 24137de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 24148728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 24151da177e4SLinus Torvalds } 2416b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 24171da177e4SLinus Torvalds 24181da177e4SLinus Torvalds /** 24191da177e4SLinus Torvalds * skb_insert - insert a buffer 24201da177e4SLinus Torvalds * @old: buffer to insert before 24211da177e4SLinus Torvalds * @newsk: buffer to insert 24228728b834SDavid S. Miller * @list: list to use 24231da177e4SLinus Torvalds * 24248728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 24258728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 24268728b834SDavid S. Miller * calls. 24278728b834SDavid S. Miller * 24281da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24291da177e4SLinus Torvalds */ 24308728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 24311da177e4SLinus Torvalds { 24321da177e4SLinus Torvalds unsigned long flags; 24331da177e4SLinus Torvalds 24348728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 24358728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 24368728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 24371da177e4SLinus Torvalds } 2438b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 24391da177e4SLinus Torvalds 24401da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 24411da177e4SLinus Torvalds struct sk_buff* skb1, 24421da177e4SLinus Torvalds const u32 len, const int pos) 24431da177e4SLinus Torvalds { 24441da177e4SLinus Torvalds int i; 24451da177e4SLinus Torvalds 2446d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2447d626f62bSArnaldo Carvalho de Melo pos - len); 24481da177e4SLinus Torvalds /* And move data appendix as is. */ 24491da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 24501da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 24511da177e4SLinus Torvalds 24521da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 24531da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 24541da177e4SLinus Torvalds skb1->data_len = skb->data_len; 24551da177e4SLinus Torvalds skb1->len += skb1->data_len; 24561da177e4SLinus Torvalds skb->data_len = 0; 24571da177e4SLinus Torvalds skb->len = len; 245827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 24591da177e4SLinus Torvalds } 24601da177e4SLinus Torvalds 24611da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 24621da177e4SLinus Torvalds struct sk_buff* skb1, 24631da177e4SLinus Torvalds const u32 len, int pos) 24641da177e4SLinus Torvalds { 24651da177e4SLinus Torvalds int i, k = 0; 24661da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 24671da177e4SLinus Torvalds 24681da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 24691da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 24701da177e4SLinus Torvalds skb->len = len; 24711da177e4SLinus Torvalds skb->data_len = len - pos; 24721da177e4SLinus Torvalds 24731da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 24749e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 24751da177e4SLinus Torvalds 24761da177e4SLinus Torvalds if (pos + size > len) { 24771da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 24781da177e4SLinus Torvalds 24791da177e4SLinus Torvalds if (pos < len) { 24801da177e4SLinus Torvalds /* Split frag. 24811da177e4SLinus Torvalds * We have two variants in this case: 24821da177e4SLinus Torvalds * 1. Move all the frag to the second 24831da177e4SLinus Torvalds * part, if it is possible. F.e. 24841da177e4SLinus Torvalds * this approach is mandatory for TUX, 24851da177e4SLinus Torvalds * where splitting is expensive. 24861da177e4SLinus Torvalds * 2. Split is accurately. We make this. 24871da177e4SLinus Torvalds */ 2488ea2ab693SIan Campbell skb_frag_ref(skb, i); 24891da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 24909e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 24919e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 24921da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 24931da177e4SLinus Torvalds } 24941da177e4SLinus Torvalds k++; 24951da177e4SLinus Torvalds } else 24961da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 24971da177e4SLinus Torvalds pos += size; 24981da177e4SLinus Torvalds } 24991da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 25001da177e4SLinus Torvalds } 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds /** 25031da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 25041da177e4SLinus Torvalds * @skb: the buffer to split 25051da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 25061da177e4SLinus Torvalds * @len: new length for skb 25071da177e4SLinus Torvalds */ 25081da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 25091da177e4SLinus Torvalds { 25101da177e4SLinus Torvalds int pos = skb_headlen(skb); 25111da177e4SLinus Torvalds 251268534c68SAmerigo Wang skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 25131da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 25141da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 25151da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 25161da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 25171da177e4SLinus Torvalds } 2518b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 25191da177e4SLinus Torvalds 25209f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 25219f782db3SIlpo Järvinen * 25229f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 25239f782db3SIlpo Järvinen */ 2524832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2525832d11c5SIlpo Järvinen { 25260ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2527832d11c5SIlpo Järvinen } 2528832d11c5SIlpo Järvinen 2529832d11c5SIlpo Järvinen /** 2530832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2531832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2532832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2533832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2534832d11c5SIlpo Järvinen * 2535832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 253620e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 2537832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2538832d11c5SIlpo Järvinen * 2539832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2540832d11c5SIlpo Järvinen * 2541832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2542832d11c5SIlpo Järvinen * to have non-paged data as well. 2543832d11c5SIlpo Järvinen * 2544832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2545832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2546832d11c5SIlpo Järvinen */ 2547832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2548832d11c5SIlpo Järvinen { 2549832d11c5SIlpo Järvinen int from, to, merge, todo; 2550832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2551832d11c5SIlpo Järvinen 2552832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2553832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2554832d11c5SIlpo Järvinen 2555832d11c5SIlpo Järvinen todo = shiftlen; 2556832d11c5SIlpo Järvinen from = 0; 2557832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2558832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2559832d11c5SIlpo Järvinen 2560832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2561832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2562832d11c5SIlpo Järvinen */ 2563832d11c5SIlpo Järvinen if (!to || 2564ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2565ea2ab693SIan Campbell fragfrom->page_offset)) { 2566832d11c5SIlpo Järvinen merge = -1; 2567832d11c5SIlpo Järvinen } else { 2568832d11c5SIlpo Järvinen merge = to - 1; 2569832d11c5SIlpo Järvinen 25709e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2571832d11c5SIlpo Järvinen if (todo < 0) { 2572832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2573832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2574832d11c5SIlpo Järvinen return 0; 2575832d11c5SIlpo Järvinen 25769f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 25779f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2578832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2579832d11c5SIlpo Järvinen 25809e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 25819e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 2582832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2583832d11c5SIlpo Järvinen 2584832d11c5SIlpo Järvinen goto onlymerged; 2585832d11c5SIlpo Järvinen } 2586832d11c5SIlpo Järvinen 2587832d11c5SIlpo Järvinen from++; 2588832d11c5SIlpo Järvinen } 2589832d11c5SIlpo Järvinen 2590832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2591832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2592832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2593832d11c5SIlpo Järvinen return 0; 2594832d11c5SIlpo Järvinen 2595832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2596832d11c5SIlpo Järvinen return 0; 2597832d11c5SIlpo Järvinen 2598832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2599832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2600832d11c5SIlpo Järvinen return 0; 2601832d11c5SIlpo Järvinen 2602832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2603832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2604832d11c5SIlpo Järvinen 26059e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 2606832d11c5SIlpo Järvinen *fragto = *fragfrom; 26079e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2608832d11c5SIlpo Järvinen from++; 2609832d11c5SIlpo Järvinen to++; 2610832d11c5SIlpo Järvinen 2611832d11c5SIlpo Järvinen } else { 2612ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2613832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2614832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 26159e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 2616832d11c5SIlpo Järvinen 2617832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 26189e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 2619832d11c5SIlpo Järvinen todo = 0; 2620832d11c5SIlpo Järvinen 2621832d11c5SIlpo Järvinen to++; 2622832d11c5SIlpo Järvinen break; 2623832d11c5SIlpo Järvinen } 2624832d11c5SIlpo Järvinen } 2625832d11c5SIlpo Järvinen 2626832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2627832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2628832d11c5SIlpo Järvinen 2629832d11c5SIlpo Järvinen if (merge >= 0) { 2630832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2631832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2632832d11c5SIlpo Järvinen 26339e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2634ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2635832d11c5SIlpo Järvinen } 2636832d11c5SIlpo Järvinen 2637832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2638832d11c5SIlpo Järvinen to = 0; 2639832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2640832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2641832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2642832d11c5SIlpo Järvinen 2643832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2644832d11c5SIlpo Järvinen 2645832d11c5SIlpo Järvinen onlymerged: 2646832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2647832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2648832d11c5SIlpo Järvinen */ 2649832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2650832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2651832d11c5SIlpo Järvinen 2652832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2653832d11c5SIlpo Järvinen skb->len -= shiftlen; 2654832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2655832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2656832d11c5SIlpo Järvinen tgt->len += shiftlen; 2657832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2658832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2659832d11c5SIlpo Järvinen 2660832d11c5SIlpo Järvinen return shiftlen; 2661832d11c5SIlpo Järvinen } 2662832d11c5SIlpo Järvinen 2663677e90edSThomas Graf /** 2664677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2665677e90edSThomas Graf * @skb: the buffer to read 2666677e90edSThomas Graf * @from: lower offset of data to be read 2667677e90edSThomas Graf * @to: upper offset of data to be read 2668677e90edSThomas Graf * @st: state variable 2669677e90edSThomas Graf * 2670677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2671677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2672677e90edSThomas Graf */ 2673677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2674677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2675677e90edSThomas Graf { 2676677e90edSThomas Graf st->lower_offset = from; 2677677e90edSThomas Graf st->upper_offset = to; 2678677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2679677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2680677e90edSThomas Graf st->frag_data = NULL; 2681677e90edSThomas Graf } 2682b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2683677e90edSThomas Graf 2684677e90edSThomas Graf /** 2685677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2686677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2687677e90edSThomas Graf * @data: destination pointer for data to be returned 2688677e90edSThomas Graf * @st: state variable 2689677e90edSThomas Graf * 2690bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 2691677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2692bc32383cSMathias Krause * the head of the data block to @data and returns the length 2693677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2694677e90edSThomas Graf * offset has been reached. 2695677e90edSThomas Graf * 2696677e90edSThomas Graf * The caller is not required to consume all of the data 2697bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 2698677e90edSThomas Graf * of bytes already consumed and the next call to 2699677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2700677e90edSThomas Graf * 270125985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2702e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 2703677e90edSThomas Graf * reads of potentially non linear data. 2704677e90edSThomas Graf * 2705bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2706677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2707677e90edSThomas Graf * a stack for this purpose. 2708677e90edSThomas Graf */ 2709677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2710677e90edSThomas Graf struct skb_seq_state *st) 2711677e90edSThomas Graf { 2712677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2713677e90edSThomas Graf skb_frag_t *frag; 2714677e90edSThomas Graf 2715aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 2716aeb193eaSWedson Almeida Filho if (st->frag_data) { 2717aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 2718aeb193eaSWedson Almeida Filho st->frag_data = NULL; 2719aeb193eaSWedson Almeida Filho } 2720677e90edSThomas Graf return 0; 2721aeb193eaSWedson Almeida Filho } 2722677e90edSThomas Graf 2723677e90edSThomas Graf next_skb: 272495e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2725677e90edSThomas Graf 2726995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 272795e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2728677e90edSThomas Graf return block_limit - abs_offset; 2729677e90edSThomas Graf } 2730677e90edSThomas Graf 2731677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2732677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2733677e90edSThomas Graf 2734677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2735677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 27369e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 2737677e90edSThomas Graf 2738677e90edSThomas Graf if (abs_offset < block_limit) { 2739677e90edSThomas Graf if (!st->frag_data) 274051c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 2741677e90edSThomas Graf 2742677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2743677e90edSThomas Graf (abs_offset - st->stepped_offset); 2744677e90edSThomas Graf 2745677e90edSThomas Graf return block_limit - abs_offset; 2746677e90edSThomas Graf } 2747677e90edSThomas Graf 2748677e90edSThomas Graf if (st->frag_data) { 274951c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2750677e90edSThomas Graf st->frag_data = NULL; 2751677e90edSThomas Graf } 2752677e90edSThomas Graf 2753677e90edSThomas Graf st->frag_idx++; 27549e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 2755677e90edSThomas Graf } 2756677e90edSThomas Graf 27575b5a60daSOlaf Kirch if (st->frag_data) { 275851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 27595b5a60daSOlaf Kirch st->frag_data = NULL; 27605b5a60daSOlaf Kirch } 27615b5a60daSOlaf Kirch 276221dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2763677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 276495e3b24cSHerbert Xu st->frag_idx = 0; 2765677e90edSThomas Graf goto next_skb; 276671b3346dSShyam Iyer } else if (st->cur_skb->next) { 276771b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 276871b3346dSShyam Iyer st->frag_idx = 0; 2769677e90edSThomas Graf goto next_skb; 2770677e90edSThomas Graf } 2771677e90edSThomas Graf 2772677e90edSThomas Graf return 0; 2773677e90edSThomas Graf } 2774b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2775677e90edSThomas Graf 2776677e90edSThomas Graf /** 2777677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2778677e90edSThomas Graf * @st: state variable 2779677e90edSThomas Graf * 2780677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2781677e90edSThomas Graf * returned 0. 2782677e90edSThomas Graf */ 2783677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2784677e90edSThomas Graf { 2785677e90edSThomas Graf if (st->frag_data) 278651c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2787677e90edSThomas Graf } 2788b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2789677e90edSThomas Graf 27903fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 27913fc7e8a6SThomas Graf 27923fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 27933fc7e8a6SThomas Graf struct ts_config *conf, 27943fc7e8a6SThomas Graf struct ts_state *state) 27953fc7e8a6SThomas Graf { 27963fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 27973fc7e8a6SThomas Graf } 27983fc7e8a6SThomas Graf 27993fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 28003fc7e8a6SThomas Graf { 28013fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 28023fc7e8a6SThomas Graf } 28033fc7e8a6SThomas Graf 28043fc7e8a6SThomas Graf /** 28053fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 28063fc7e8a6SThomas Graf * @skb: the buffer to look in 28073fc7e8a6SThomas Graf * @from: search offset 28083fc7e8a6SThomas Graf * @to: search limit 28093fc7e8a6SThomas Graf * @config: textsearch configuration 28103fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 28113fc7e8a6SThomas Graf * 28123fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 28133fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 28143fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 28153fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 28163fc7e8a6SThomas Graf */ 28173fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 28183fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 28193fc7e8a6SThomas Graf struct ts_state *state) 28203fc7e8a6SThomas Graf { 2821f72b948dSPhil Oester unsigned int ret; 2822f72b948dSPhil Oester 28233fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 28243fc7e8a6SThomas Graf config->finish = skb_ts_finish; 28253fc7e8a6SThomas Graf 28263fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 28273fc7e8a6SThomas Graf 2828f72b948dSPhil Oester ret = textsearch_find(config, state); 2829f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 28303fc7e8a6SThomas Graf } 2831b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 28323fc7e8a6SThomas Graf 2833e89e9cf5SAnanda Raju /** 28342c53040fSBen Hutchings * skb_append_datato_frags - append the user data to a skb 2835e89e9cf5SAnanda Raju * @sk: sock structure 2836e793c0f7SMasanari Iida * @skb: skb structure to be appended with user data. 2837e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2838e89e9cf5SAnanda Raju * @from: pointer to user message iov 2839e89e9cf5SAnanda Raju * @length: length of the iov message 2840e89e9cf5SAnanda Raju * 2841e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2842e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2843e89e9cf5SAnanda Raju */ 2844e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2845dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2846e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2847e89e9cf5SAnanda Raju void *from, int length) 2848e89e9cf5SAnanda Raju { 2849b2111724SEric Dumazet int frg_cnt = skb_shinfo(skb)->nr_frags; 2850b2111724SEric Dumazet int copy; 2851e89e9cf5SAnanda Raju int offset = 0; 2852e89e9cf5SAnanda Raju int ret; 2853b2111724SEric Dumazet struct page_frag *pfrag = ¤t->task_frag; 2854e89e9cf5SAnanda Raju 2855e89e9cf5SAnanda Raju do { 2856e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2857e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2858b2111724SEric Dumazet return -EMSGSIZE; 2859e89e9cf5SAnanda Raju 2860b2111724SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 2861e89e9cf5SAnanda Raju return -ENOMEM; 2862e89e9cf5SAnanda Raju 2863e89e9cf5SAnanda Raju /* copy the user data to page */ 2864b2111724SEric Dumazet copy = min_t(int, length, pfrag->size - pfrag->offset); 2865e89e9cf5SAnanda Raju 2866b2111724SEric Dumazet ret = getfrag(from, page_address(pfrag->page) + pfrag->offset, 2867e89e9cf5SAnanda Raju offset, copy, 0, skb); 2868e89e9cf5SAnanda Raju if (ret < 0) 2869e89e9cf5SAnanda Raju return -EFAULT; 2870e89e9cf5SAnanda Raju 2871e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2872b2111724SEric Dumazet skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset, 2873b2111724SEric Dumazet copy); 2874b2111724SEric Dumazet frg_cnt++; 2875b2111724SEric Dumazet pfrag->offset += copy; 2876b2111724SEric Dumazet get_page(pfrag->page); 2877b2111724SEric Dumazet 2878b2111724SEric Dumazet skb->truesize += copy; 2879b2111724SEric Dumazet atomic_add(copy, &sk->sk_wmem_alloc); 2880e89e9cf5SAnanda Raju skb->len += copy; 2881e89e9cf5SAnanda Raju skb->data_len += copy; 2882e89e9cf5SAnanda Raju offset += copy; 2883e89e9cf5SAnanda Raju length -= copy; 2884e89e9cf5SAnanda Raju 2885e89e9cf5SAnanda Raju } while (length > 0); 2886e89e9cf5SAnanda Raju 2887e89e9cf5SAnanda Raju return 0; 2888e89e9cf5SAnanda Raju } 2889b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2890e89e9cf5SAnanda Raju 2891cbb042f9SHerbert Xu /** 2892cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2893cbb042f9SHerbert Xu * @skb: buffer to update 2894cbb042f9SHerbert Xu * @len: length of data pulled 2895cbb042f9SHerbert Xu * 2896cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2897fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 289884fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 289984fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 290084fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2901cbb042f9SHerbert Xu */ 2902cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2903cbb042f9SHerbert Xu { 2904cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2905cbb042f9SHerbert Xu skb->len -= len; 2906cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2907cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2908cbb042f9SHerbert Xu return skb->data += len; 2909cbb042f9SHerbert Xu } 2910f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2911f94691acSArnaldo Carvalho de Melo 2912f4c50d99SHerbert Xu /** 2913f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2914df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 2915576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2916f4c50d99SHerbert Xu * 2917f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 29184c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 29194c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2920f4c50d99SHerbert Xu */ 2921df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 2922df5771ffSMichael S. Tsirkin netdev_features_t features) 2923f4c50d99SHerbert Xu { 2924f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2925f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 29261a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 2927df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 2928df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 2929df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 29301fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 2931f4c50d99SHerbert Xu unsigned int offset = doffset; 2932df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 2933f4c50d99SHerbert Xu unsigned int headroom; 2934f4c50d99SHerbert Xu unsigned int len; 2935ec5f0615SPravin B Shelar __be16 proto; 2936ec5f0615SPravin B Shelar bool csum; 293704ed3e74SMichał Mirosław int sg = !!(features & NETIF_F_SG); 2938df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 2939f4c50d99SHerbert Xu int err = -ENOMEM; 2940f4c50d99SHerbert Xu int i = 0; 2941f4c50d99SHerbert Xu int pos; 294253d6471cSVlad Yasevich int dummy; 2943f4c50d99SHerbert Xu 29445882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 294553d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 2946ec5f0615SPravin B Shelar if (unlikely(!proto)) 2947ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 2948ec5f0615SPravin B Shelar 29497e2b10c1STom Herbert csum = !head_skb->encap_hdr_csum && 29507e2b10c1STom Herbert !!can_checksum_protocol(features, proto); 29517e2b10c1STom Herbert 2952df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 2953df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 2954f4c50d99SHerbert Xu 2955f4c50d99SHerbert Xu do { 2956f4c50d99SHerbert Xu struct sk_buff *nskb; 29578cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 2958c8884eddSHerbert Xu int hsize; 2959f4c50d99SHerbert Xu int size; 2960f4c50d99SHerbert Xu 2961df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 2962f4c50d99SHerbert Xu if (len > mss) 2963f4c50d99SHerbert Xu len = mss; 2964f4c50d99SHerbert Xu 2965df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 2966f4c50d99SHerbert Xu if (hsize < 0) 2967f4c50d99SHerbert Xu hsize = 0; 2968c8884eddSHerbert Xu if (hsize > len || !sg) 2969c8884eddSHerbert Xu hsize = len; 2970f4c50d99SHerbert Xu 29711a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 29721a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 29731a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 297489319d38SHerbert Xu 29759d8506ccSHerbert Xu i = 0; 29761a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 29771a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 29781fd819ecSMichael S. Tsirkin frag_skb = list_skb; 29791a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 29809d8506ccSHerbert Xu 29819d8506ccSHerbert Xu while (pos < offset + len) { 29829d8506ccSHerbert Xu BUG_ON(i >= nfrags); 29839d8506ccSHerbert Xu 29844e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 29859d8506ccSHerbert Xu if (pos + size > offset + len) 29869d8506ccSHerbert Xu break; 29879d8506ccSHerbert Xu 29889d8506ccSHerbert Xu i++; 29899d8506ccSHerbert Xu pos += size; 29904e1beba1SMichael S. Tsirkin frag++; 29919d8506ccSHerbert Xu } 29929d8506ccSHerbert Xu 29931a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 29941a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 299589319d38SHerbert Xu 2996f4c50d99SHerbert Xu if (unlikely(!nskb)) 2997f4c50d99SHerbert Xu goto err; 2998f4c50d99SHerbert Xu 29999d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 30009d8506ccSHerbert Xu kfree_skb(nskb); 30019d8506ccSHerbert Xu goto err; 30029d8506ccSHerbert Xu } 30039d8506ccSHerbert Xu 3004ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 300589319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 300689319d38SHerbert Xu kfree_skb(nskb); 300789319d38SHerbert Xu goto err; 300889319d38SHerbert Xu } 300989319d38SHerbert Xu 3010ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 301189319d38SHerbert Xu skb_release_head_state(nskb); 301289319d38SHerbert Xu __skb_push(nskb, doffset); 301389319d38SHerbert Xu } else { 3014c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3015df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3016c93bdd0eSMel Gorman NUMA_NO_NODE); 301789319d38SHerbert Xu 301889319d38SHerbert Xu if (unlikely(!nskb)) 301989319d38SHerbert Xu goto err; 302089319d38SHerbert Xu 302189319d38SHerbert Xu skb_reserve(nskb, headroom); 302289319d38SHerbert Xu __skb_put(nskb, doffset); 302389319d38SHerbert Xu } 302489319d38SHerbert Xu 3025f4c50d99SHerbert Xu if (segs) 3026f4c50d99SHerbert Xu tail->next = nskb; 3027f4c50d99SHerbert Xu else 3028f4c50d99SHerbert Xu segs = nskb; 3029f4c50d99SHerbert Xu tail = nskb; 3030f4c50d99SHerbert Xu 3031df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3032f4c50d99SHerbert Xu 3033030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3034fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 303568c33163SPravin B Shelar 3036df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 303768c33163SPravin B Shelar nskb->data - tnl_hlen, 303868c33163SPravin B Shelar doffset + tnl_hlen); 303989319d38SHerbert Xu 30409d8506ccSHerbert Xu if (nskb->len == len + doffset) 30411cdbcb79SSimon Horman goto perform_csum_check; 304289319d38SHerbert Xu 3043e585f236STom Herbert if (!sg && !nskb->remcsum_offload) { 30446f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 3045df5771ffSMichael S. Tsirkin nskb->csum = skb_copy_and_csum_bits(head_skb, offset, 3046f4c50d99SHerbert Xu skb_put(nskb, len), 3047f4c50d99SHerbert Xu len, 0); 30487e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3049de843723STom Herbert skb_headroom(nskb) + doffset; 3050f4c50d99SHerbert Xu continue; 3051f4c50d99SHerbert Xu } 3052f4c50d99SHerbert Xu 30538cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3054f4c50d99SHerbert Xu 3055df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3056d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3057f4c50d99SHerbert Xu 3058df5771ffSMichael S. Tsirkin skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags & 3059df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3060cef401deSEric Dumazet 30619d8506ccSHerbert Xu while (pos < offset + len) { 30629d8506ccSHerbert Xu if (i >= nfrags) { 30631a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb)); 30649d8506ccSHerbert Xu 30659d8506ccSHerbert Xu i = 0; 30661a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 30671a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 30681fd819ecSMichael S. Tsirkin frag_skb = list_skb; 30699d8506ccSHerbert Xu 30709d8506ccSHerbert Xu BUG_ON(!nfrags); 30719d8506ccSHerbert Xu 30721a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 30739d8506ccSHerbert Xu } 30749d8506ccSHerbert Xu 30759d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 30769d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 30779d8506ccSHerbert Xu net_warn_ratelimited( 30789d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 30799d8506ccSHerbert Xu pos, mss); 30809d8506ccSHerbert Xu goto err; 30819d8506ccSHerbert Xu } 30829d8506ccSHerbert Xu 30831fd819ecSMichael S. Tsirkin if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC))) 30841fd819ecSMichael S. Tsirkin goto err; 30851fd819ecSMichael S. Tsirkin 30864e1beba1SMichael S. Tsirkin *nskb_frag = *frag; 30878cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 30888cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3089f4c50d99SHerbert Xu 3090f4c50d99SHerbert Xu if (pos < offset) { 30918cb19905SMichael S. Tsirkin nskb_frag->page_offset += offset - pos; 30928cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3093f4c50d99SHerbert Xu } 3094f4c50d99SHerbert Xu 309589319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3096f4c50d99SHerbert Xu 3097f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3098f4c50d99SHerbert Xu i++; 30994e1beba1SMichael S. Tsirkin frag++; 3100f4c50d99SHerbert Xu pos += size; 3101f4c50d99SHerbert Xu } else { 31028cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 310389319d38SHerbert Xu goto skip_fraglist; 3104f4c50d99SHerbert Xu } 3105f4c50d99SHerbert Xu 31068cb19905SMichael S. Tsirkin nskb_frag++; 3107f4c50d99SHerbert Xu } 3108f4c50d99SHerbert Xu 310989319d38SHerbert Xu skip_fraglist: 3110f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3111f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3112f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3113ec5f0615SPravin B Shelar 31141cdbcb79SSimon Horman perform_csum_check: 3115e585f236STom Herbert if (!csum && !nskb->remcsum_offload) { 3116ec5f0615SPravin B Shelar nskb->csum = skb_checksum(nskb, doffset, 3117ec5f0615SPravin B Shelar nskb->len - doffset, 0); 3118ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 31197e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 31207e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3121ec5f0615SPravin B Shelar } 3122df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3123f4c50d99SHerbert Xu 3124bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3125bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3126bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3127bec3cfdcSEric Dumazet */ 3128bec3cfdcSEric Dumazet segs->prev = tail; 3129432c856fSToshiaki Makita 3130432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 3131432c856fSToshiaki Makita * using skb_set_owner_w(). 3132432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 3133432c856fSToshiaki Makita */ 3134432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 3135432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 3136432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 3137432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 3138432c856fSToshiaki Makita } 3139f4c50d99SHerbert Xu return segs; 3140f4c50d99SHerbert Xu 3141f4c50d99SHerbert Xu err: 3142289dccbeSEric Dumazet kfree_skb_list(segs); 3143f4c50d99SHerbert Xu return ERR_PTR(err); 3144f4c50d99SHerbert Xu } 3145f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3146f4c50d99SHerbert Xu 314771d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 314871d93b39SHerbert Xu { 31498a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 315067147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 315167147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 31528a29111cSEric Dumazet struct sk_buff *nskb, *lp, *p = *head; 31538a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 3154715dc1f3SEric Dumazet unsigned int delta_truesize; 31558a29111cSEric Dumazet unsigned int headroom; 315671d93b39SHerbert Xu 31578a29111cSEric Dumazet if (unlikely(p->len + len >= 65536)) 315871d93b39SHerbert Xu return -E2BIG; 315971d93b39SHerbert Xu 316029e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 31618a29111cSEric Dumazet pinfo = skb_shinfo(lp); 31628a29111cSEric Dumazet 31638a29111cSEric Dumazet if (headlen <= offset) { 316442da6994SHerbert Xu skb_frag_t *frag; 316566e92fcfSHerbert Xu skb_frag_t *frag2; 31669aaa156cSHerbert Xu int i = skbinfo->nr_frags; 31679aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 316842da6994SHerbert Xu 316966e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 31708a29111cSEric Dumazet goto merge; 317181705ad1SHerbert Xu 31728a29111cSEric Dumazet offset -= headlen; 31739aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 31749aaa156cSHerbert Xu skbinfo->nr_frags = 0; 3175f5572068SHerbert Xu 31769aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 31779aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 317866e92fcfSHerbert Xu do { 317966e92fcfSHerbert Xu *--frag = *--frag2; 318066e92fcfSHerbert Xu } while (--i); 318166e92fcfSHerbert Xu 318266e92fcfSHerbert Xu frag->page_offset += offset; 31839e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 318466e92fcfSHerbert Xu 3185715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3186ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3187ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3188715dc1f3SEric Dumazet 3189f5572068SHerbert Xu skb->truesize -= skb->data_len; 3190f5572068SHerbert Xu skb->len -= skb->data_len; 3191f5572068SHerbert Xu skb->data_len = 0; 3192f5572068SHerbert Xu 3193715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 31945d38a079SHerbert Xu goto done; 3195d7e8883cSEric Dumazet } else if (skb->head_frag) { 3196d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3197d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3198d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3199d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3200d7e8883cSEric Dumazet unsigned int first_offset; 3201d7e8883cSEric Dumazet 3202d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 32038a29111cSEric Dumazet goto merge; 3204d7e8883cSEric Dumazet 3205d7e8883cSEric Dumazet first_offset = skb->data - 3206d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3207d7e8883cSEric Dumazet offset; 3208d7e8883cSEric Dumazet 3209d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3210d7e8883cSEric Dumazet 3211d7e8883cSEric Dumazet frag->page.p = page; 3212d7e8883cSEric Dumazet frag->page_offset = first_offset; 3213d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3214d7e8883cSEric Dumazet 3215d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3216d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3217d7e8883cSEric Dumazet 3218715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3219d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3220d7e8883cSEric Dumazet goto done; 32218a29111cSEric Dumazet } 322273d3fe6dSEric Dumazet /* switch back to head shinfo */ 322373d3fe6dSEric Dumazet pinfo = skb_shinfo(p); 322473d3fe6dSEric Dumazet 32258a29111cSEric Dumazet if (pinfo->frag_list) 32268a29111cSEric Dumazet goto merge; 32278a29111cSEric Dumazet if (skb_gro_len(p) != pinfo->gso_size) 322869c0cab1SHerbert Xu return -E2BIG; 322971d93b39SHerbert Xu 323071d93b39SHerbert Xu headroom = skb_headroom(p); 32313d3be433SEric Dumazet nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); 323271d93b39SHerbert Xu if (unlikely(!nskb)) 323371d93b39SHerbert Xu return -ENOMEM; 323471d93b39SHerbert Xu 323571d93b39SHerbert Xu __copy_skb_header(nskb, p); 323671d93b39SHerbert Xu nskb->mac_len = p->mac_len; 323771d93b39SHerbert Xu 323871d93b39SHerbert Xu skb_reserve(nskb, headroom); 323986911732SHerbert Xu __skb_put(nskb, skb_gro_offset(p)); 324071d93b39SHerbert Xu 324186911732SHerbert Xu skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 324271d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 324371d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 324471d93b39SHerbert Xu 324586911732SHerbert Xu __skb_pull(p, skb_gro_offset(p)); 324686911732SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), 324786911732SHerbert Xu p->data - skb_mac_header(p)); 324871d93b39SHerbert Xu 324971d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 32509aaa156cSHerbert Xu skb_shinfo(nskb)->gso_size = pinfo->gso_size; 3251622e0ca1SHerbert Xu pinfo->gso_size = 0; 3252f4a775d1SEric Dumazet __skb_header_release(p); 3253c3c7c254SEric Dumazet NAPI_GRO_CB(nskb)->last = p; 325471d93b39SHerbert Xu 325571d93b39SHerbert Xu nskb->data_len += p->len; 3256de8261c2SEric Dumazet nskb->truesize += p->truesize; 325771d93b39SHerbert Xu nskb->len += p->len; 325871d93b39SHerbert Xu 325971d93b39SHerbert Xu *head = nskb; 326071d93b39SHerbert Xu nskb->next = p->next; 326171d93b39SHerbert Xu p->next = NULL; 326271d93b39SHerbert Xu 326371d93b39SHerbert Xu p = nskb; 326471d93b39SHerbert Xu 326571d93b39SHerbert Xu merge: 3266715dc1f3SEric Dumazet delta_truesize = skb->truesize; 326767147ba9SHerbert Xu if (offset > headlen) { 3268d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3269d1dc7abfSMichal Schmidt 3270d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 32719e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3272d1dc7abfSMichal Schmidt skb->data_len -= eat; 3273d1dc7abfSMichal Schmidt skb->len -= eat; 327467147ba9SHerbert Xu offset = headlen; 327556035022SHerbert Xu } 327656035022SHerbert Xu 327767147ba9SHerbert Xu __skb_pull(skb, offset); 327856035022SHerbert Xu 327929e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 32808a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 32818a29111cSEric Dumazet else 3282c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3283c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 3284f4a775d1SEric Dumazet __skb_header_release(skb); 32858a29111cSEric Dumazet lp = p; 328671d93b39SHerbert Xu 32875d38a079SHerbert Xu done: 32885d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 328937fe4732SHerbert Xu p->data_len += len; 3290715dc1f3SEric Dumazet p->truesize += delta_truesize; 329137fe4732SHerbert Xu p->len += len; 32928a29111cSEric Dumazet if (lp != p) { 32938a29111cSEric Dumazet lp->data_len += len; 32948a29111cSEric Dumazet lp->truesize += delta_truesize; 32958a29111cSEric Dumazet lp->len += len; 32968a29111cSEric Dumazet } 329771d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 329871d93b39SHerbert Xu return 0; 329971d93b39SHerbert Xu } 330071d93b39SHerbert Xu 33011da177e4SLinus Torvalds void __init skb_init(void) 33021da177e4SLinus Torvalds { 33031da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 33041da177e4SLinus Torvalds sizeof(struct sk_buff), 33051da177e4SLinus Torvalds 0, 3306e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 330720c2df83SPaul Mundt NULL); 3308d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3309d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 3310d179cd12SDavid S. Miller 0, 3311e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 331220c2df83SPaul Mundt NULL); 33131da177e4SLinus Torvalds } 33141da177e4SLinus Torvalds 3315716ea3a7SDavid Howells /** 3316716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3317716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 3318716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 3319716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 3320716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 3321716ea3a7SDavid Howells * 3322716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 3323716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 3324716ea3a7SDavid Howells */ 332551c739d1SDavid S. Miller static int 332651c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3327716ea3a7SDavid Howells { 33281a028e50SDavid S. Miller int start = skb_headlen(skb); 33291a028e50SDavid S. Miller int i, copy = start - offset; 3330fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3331716ea3a7SDavid Howells int elt = 0; 3332716ea3a7SDavid Howells 3333716ea3a7SDavid Howells if (copy > 0) { 3334716ea3a7SDavid Howells if (copy > len) 3335716ea3a7SDavid Howells copy = len; 3336642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 3337716ea3a7SDavid Howells elt++; 3338716ea3a7SDavid Howells if ((len -= copy) == 0) 3339716ea3a7SDavid Howells return elt; 3340716ea3a7SDavid Howells offset += copy; 3341716ea3a7SDavid Howells } 3342716ea3a7SDavid Howells 3343716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 33441a028e50SDavid S. Miller int end; 3345716ea3a7SDavid Howells 3346547b792cSIlpo Järvinen WARN_ON(start > offset + len); 33471a028e50SDavid S. Miller 33489e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3349716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3350716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3351716ea3a7SDavid Howells 3352716ea3a7SDavid Howells if (copy > len) 3353716ea3a7SDavid Howells copy = len; 3354ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3355642f1490SJens Axboe frag->page_offset+offset-start); 3356716ea3a7SDavid Howells elt++; 3357716ea3a7SDavid Howells if (!(len -= copy)) 3358716ea3a7SDavid Howells return elt; 3359716ea3a7SDavid Howells offset += copy; 3360716ea3a7SDavid Howells } 33611a028e50SDavid S. Miller start = end; 3362716ea3a7SDavid Howells } 3363716ea3a7SDavid Howells 3364fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 33651a028e50SDavid S. Miller int end; 3366716ea3a7SDavid Howells 3367547b792cSIlpo Järvinen WARN_ON(start > offset + len); 33681a028e50SDavid S. Miller 3369fbb398a8SDavid S. Miller end = start + frag_iter->len; 3370716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3371716ea3a7SDavid Howells if (copy > len) 3372716ea3a7SDavid Howells copy = len; 3373fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 337451c739d1SDavid S. Miller copy); 3375716ea3a7SDavid Howells if ((len -= copy) == 0) 3376716ea3a7SDavid Howells return elt; 3377716ea3a7SDavid Howells offset += copy; 3378716ea3a7SDavid Howells } 33791a028e50SDavid S. Miller start = end; 3380716ea3a7SDavid Howells } 3381716ea3a7SDavid Howells BUG_ON(len); 3382716ea3a7SDavid Howells return elt; 3383716ea3a7SDavid Howells } 3384716ea3a7SDavid Howells 338525a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 338625a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 338725a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 338825a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 338925a91d8dSFan Du * 339025a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 339125a91d8dSFan Du * 1. sg_init_table 339225a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 339325a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 339425a91d8dSFan Du * 339525a91d8dSFan Du * This is equivalent to: 339625a91d8dSFan Du * 1. sg_init_table 339725a91d8dSFan Du * 2. skb_to_sgvec(payload1) 339825a91d8dSFan Du * 3. sg_unmark_end 339925a91d8dSFan Du * 4. skb_to_sgvec(payload2) 340025a91d8dSFan Du * 340125a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 340225a91d8dSFan Du * is more preferable. 340325a91d8dSFan Du */ 340425a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 340525a91d8dSFan Du int offset, int len) 340625a91d8dSFan Du { 340725a91d8dSFan Du return __skb_to_sgvec(skb, sg, offset, len); 340825a91d8dSFan Du } 340925a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 341025a91d8dSFan Du 341151c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 341251c739d1SDavid S. Miller { 341351c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 341451c739d1SDavid S. Miller 3415c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 341651c739d1SDavid S. Miller 341751c739d1SDavid S. Miller return nsg; 341851c739d1SDavid S. Miller } 3419b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 342051c739d1SDavid S. Miller 3421716ea3a7SDavid Howells /** 3422716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 3423716ea3a7SDavid Howells * @skb: The socket buffer to check. 3424716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 3425716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 3426716ea3a7SDavid Howells * 3427716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 3428716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 3429716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 3430716ea3a7SDavid Howells * 3431716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 3432716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 3433716ea3a7SDavid Howells * set to point to the skb in which this space begins. 3434716ea3a7SDavid Howells * 3435716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3436716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3437716ea3a7SDavid Howells */ 3438716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3439716ea3a7SDavid Howells { 3440716ea3a7SDavid Howells int copyflag; 3441716ea3a7SDavid Howells int elt; 3442716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3443716ea3a7SDavid Howells 3444716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3445716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3446716ea3a7SDavid Howells * at the moment even if they are anonymous). 3447716ea3a7SDavid Howells */ 3448716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3449716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3450716ea3a7SDavid Howells return -ENOMEM; 3451716ea3a7SDavid Howells 3452716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 345321dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3454716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3455716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3456716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3457716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3458716ea3a7SDavid Howells 3459716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3460716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3461716ea3a7SDavid Howells return -ENOMEM; 3462716ea3a7SDavid Howells 3463716ea3a7SDavid Howells /* Voila! */ 3464716ea3a7SDavid Howells *trailer = skb; 3465716ea3a7SDavid Howells return 1; 3466716ea3a7SDavid Howells } 3467716ea3a7SDavid Howells 3468716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3469716ea3a7SDavid Howells 3470716ea3a7SDavid Howells elt = 1; 3471716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3472716ea3a7SDavid Howells copyflag = 0; 3473716ea3a7SDavid Howells 3474716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3475716ea3a7SDavid Howells int ntail = 0; 3476716ea3a7SDavid Howells 3477716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3478716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3479716ea3a7SDavid Howells * after it. */ 3480716ea3a7SDavid Howells 3481716ea3a7SDavid Howells if (skb_shared(skb1)) 3482716ea3a7SDavid Howells copyflag = 1; 3483716ea3a7SDavid Howells 3484716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3485716ea3a7SDavid Howells 3486716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3487716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 348821dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3489716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3490716ea3a7SDavid Howells ntail = tailbits + 128; 3491716ea3a7SDavid Howells } 3492716ea3a7SDavid Howells 3493716ea3a7SDavid Howells if (copyflag || 3494716ea3a7SDavid Howells skb_cloned(skb1) || 3495716ea3a7SDavid Howells ntail || 3496716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 349721dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3498716ea3a7SDavid Howells struct sk_buff *skb2; 3499716ea3a7SDavid Howells 3500716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3501716ea3a7SDavid Howells if (ntail == 0) 3502716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3503716ea3a7SDavid Howells else 3504716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3505716ea3a7SDavid Howells skb_headroom(skb1), 3506716ea3a7SDavid Howells ntail, 3507716ea3a7SDavid Howells GFP_ATOMIC); 3508716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3509716ea3a7SDavid Howells return -ENOMEM; 3510716ea3a7SDavid Howells 3511716ea3a7SDavid Howells if (skb1->sk) 3512716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3513716ea3a7SDavid Howells 3514716ea3a7SDavid Howells /* Looking around. Are we still alive? 3515716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3516716ea3a7SDavid Howells 3517716ea3a7SDavid Howells skb2->next = skb1->next; 3518716ea3a7SDavid Howells *skb_p = skb2; 3519716ea3a7SDavid Howells kfree_skb(skb1); 3520716ea3a7SDavid Howells skb1 = skb2; 3521716ea3a7SDavid Howells } 3522716ea3a7SDavid Howells elt++; 3523716ea3a7SDavid Howells *trailer = skb1; 3524716ea3a7SDavid Howells skb_p = &skb1->next; 3525716ea3a7SDavid Howells } 3526716ea3a7SDavid Howells 3527716ea3a7SDavid Howells return elt; 3528716ea3a7SDavid Howells } 3529b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3530716ea3a7SDavid Howells 3531b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3532b1faf566SEric Dumazet { 3533b1faf566SEric Dumazet struct sock *sk = skb->sk; 3534b1faf566SEric Dumazet 3535b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3536b1faf566SEric Dumazet } 3537b1faf566SEric Dumazet 3538b1faf566SEric Dumazet /* 3539b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3540b1faf566SEric Dumazet */ 3541b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3542b1faf566SEric Dumazet { 3543b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 354495c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 3545b1faf566SEric Dumazet return -ENOMEM; 3546b1faf566SEric Dumazet 3547b1faf566SEric Dumazet skb_orphan(skb); 3548b1faf566SEric Dumazet skb->sk = sk; 3549b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3550b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3551b1faf566SEric Dumazet 3552abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3553abb57ea4SEric Dumazet skb_dst_force(skb); 3554abb57ea4SEric Dumazet 3555b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3556b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3557676d2369SDavid S. Miller sk->sk_data_ready(sk); 3558b1faf566SEric Dumazet return 0; 3559b1faf566SEric Dumazet } 3560b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3561b1faf566SEric Dumazet 3562364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 3563364a9e93SWillem de Bruijn { 3564364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 3565364a9e93SWillem de Bruijn struct sk_buff *skb, *skb_next; 3566364a9e93SWillem de Bruijn int err = 0; 3567364a9e93SWillem de Bruijn 3568364a9e93SWillem de Bruijn spin_lock_bh(&q->lock); 3569364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 3570364a9e93SWillem de Bruijn if (skb && (skb_next = skb_peek(q))) 3571364a9e93SWillem de Bruijn err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 3572364a9e93SWillem de Bruijn spin_unlock_bh(&q->lock); 3573364a9e93SWillem de Bruijn 3574364a9e93SWillem de Bruijn sk->sk_err = err; 3575364a9e93SWillem de Bruijn if (err) 3576364a9e93SWillem de Bruijn sk->sk_error_report(sk); 3577364a9e93SWillem de Bruijn 3578364a9e93SWillem de Bruijn return skb; 3579364a9e93SWillem de Bruijn } 3580364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 3581364a9e93SWillem de Bruijn 3582cab41c47SAlexander Duyck /** 3583cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 3584cab41c47SAlexander Duyck * @skb: the skb to clone 3585cab41c47SAlexander Duyck * 3586cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 3587cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 3588cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 3589cab41c47SAlexander Duyck * 3590cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 3591cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 3592cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 3593cab41c47SAlexander Duyck * the sk_error_queue. 3594cab41c47SAlexander Duyck */ 359562bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 359662bccb8cSAlexander Duyck { 359762bccb8cSAlexander Duyck struct sock *sk = skb->sk; 359862bccb8cSAlexander Duyck struct sk_buff *clone; 359962bccb8cSAlexander Duyck 360062bccb8cSAlexander Duyck if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt)) 360162bccb8cSAlexander Duyck return NULL; 360262bccb8cSAlexander Duyck 360362bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 360462bccb8cSAlexander Duyck if (!clone) { 360562bccb8cSAlexander Duyck sock_put(sk); 360662bccb8cSAlexander Duyck return NULL; 360762bccb8cSAlexander Duyck } 360862bccb8cSAlexander Duyck 360962bccb8cSAlexander Duyck clone->sk = sk; 361062bccb8cSAlexander Duyck clone->destructor = sock_efree; 361162bccb8cSAlexander Duyck 361262bccb8cSAlexander Duyck return clone; 361362bccb8cSAlexander Duyck } 361462bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 361562bccb8cSAlexander Duyck 361637846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 361737846ef0SAlexander Duyck struct sock *sk, 361837846ef0SAlexander Duyck int tstype) 3619ac45f602SPatrick Ohly { 3620ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3621ac45f602SPatrick Ohly int err; 3622ac45f602SPatrick Ohly 3623ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3624ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3625ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3626ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 3627e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 36284ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 362909c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 36304ed2d765SWillem de Bruijn if (sk->sk_protocol == IPPROTO_TCP) 36314ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 36324ed2d765SWillem de Bruijn } 363329030374SEric Dumazet 3634ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 363529030374SEric Dumazet 3636ac45f602SPatrick Ohly if (err) 3637ac45f602SPatrick Ohly kfree_skb(skb); 3638ac45f602SPatrick Ohly } 363937846ef0SAlexander Duyck 364037846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 364137846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 364237846ef0SAlexander Duyck { 364337846ef0SAlexander Duyck struct sock *sk = skb->sk; 364437846ef0SAlexander Duyck 364562bccb8cSAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 364662bccb8cSAlexander Duyck sock_hold(sk); 364737846ef0SAlexander Duyck 364837846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 364937846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND); 365037846ef0SAlexander Duyck 365137846ef0SAlexander Duyck sock_put(sk); 365237846ef0SAlexander Duyck } 365337846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 365437846ef0SAlexander Duyck 365537846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 365637846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 365737846ef0SAlexander Duyck struct sock *sk, int tstype) 365837846ef0SAlexander Duyck { 365937846ef0SAlexander Duyck struct sk_buff *skb; 366037846ef0SAlexander Duyck 366137846ef0SAlexander Duyck if (!sk) 366237846ef0SAlexander Duyck return; 366337846ef0SAlexander Duyck 366437846ef0SAlexander Duyck if (hwtstamps) 366537846ef0SAlexander Duyck *skb_hwtstamps(orig_skb) = *hwtstamps; 366637846ef0SAlexander Duyck else 366737846ef0SAlexander Duyck orig_skb->tstamp = ktime_get_real(); 366837846ef0SAlexander Duyck 366937846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 367037846ef0SAlexander Duyck if (!skb) 367137846ef0SAlexander Duyck return; 367237846ef0SAlexander Duyck 367337846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, tstype); 367437846ef0SAlexander Duyck } 3675e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 3676e7fd2885SWillem de Bruijn 3677e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 3678e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 3679e7fd2885SWillem de Bruijn { 3680e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 3681e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 3682e7fd2885SWillem de Bruijn } 3683ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3684ac45f602SPatrick Ohly 36856e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 36866e3e939fSJohannes Berg { 36876e3e939fSJohannes Berg struct sock *sk = skb->sk; 36886e3e939fSJohannes Berg struct sock_exterr_skb *serr; 36896e3e939fSJohannes Berg int err; 36906e3e939fSJohannes Berg 36916e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 36926e3e939fSJohannes Berg skb->wifi_acked = acked; 36936e3e939fSJohannes Berg 36946e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 36956e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 36966e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 36976e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 36986e3e939fSJohannes Berg 3699bf7fa551SAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 3700bf7fa551SAlexander Duyck sock_hold(sk); 3701bf7fa551SAlexander Duyck 37026e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 37036e3e939fSJohannes Berg if (err) 37046e3e939fSJohannes Berg kfree_skb(skb); 3705bf7fa551SAlexander Duyck 3706bf7fa551SAlexander Duyck sock_put(sk); 37076e3e939fSJohannes Berg } 37086e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 37096e3e939fSJohannes Berg 3710ac45f602SPatrick Ohly 3711f35d9d8aSRusty Russell /** 3712f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3713f35d9d8aSRusty Russell * @skb: the skb to set 3714f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3715f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3716f35d9d8aSRusty Russell * 3717f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3718f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3719f35d9d8aSRusty Russell * 3720f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3721f35d9d8aSRusty Russell * returns false you should drop the packet. 3722f35d9d8aSRusty Russell */ 3723f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3724f35d9d8aSRusty Russell { 37255ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 37265ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3727e87cc472SJoe Perches net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 37285ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3729f35d9d8aSRusty Russell return false; 3730f35d9d8aSRusty Russell } 3731f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3732f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3733f35d9d8aSRusty Russell skb->csum_offset = off; 3734e5d5decaSJason Wang skb_set_transport_header(skb, start); 3735f35d9d8aSRusty Russell return true; 3736f35d9d8aSRusty Russell } 3737b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3738f35d9d8aSRusty Russell 3739ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 3740ed1f50c3SPaul Durrant unsigned int max) 3741ed1f50c3SPaul Durrant { 3742ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 3743ed1f50c3SPaul Durrant return 0; 3744ed1f50c3SPaul Durrant 3745ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 3746ed1f50c3SPaul Durrant * won't need to do it again. 3747ed1f50c3SPaul Durrant */ 3748ed1f50c3SPaul Durrant if (max > skb->len) 3749ed1f50c3SPaul Durrant max = skb->len; 3750ed1f50c3SPaul Durrant 3751ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 3752ed1f50c3SPaul Durrant return -ENOMEM; 3753ed1f50c3SPaul Durrant 3754ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 3755ed1f50c3SPaul Durrant return -EPROTO; 3756ed1f50c3SPaul Durrant 3757ed1f50c3SPaul Durrant return 0; 3758ed1f50c3SPaul Durrant } 3759ed1f50c3SPaul Durrant 3760f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 3761f9708b43SJan Beulich 3762f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 3763f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 3764f9708b43SJan Beulich unsigned int off) 3765f9708b43SJan Beulich { 3766f9708b43SJan Beulich switch (proto) { 3767f9708b43SJan Beulich int err; 3768f9708b43SJan Beulich 3769f9708b43SJan Beulich case IPPROTO_TCP: 3770f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 3771f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 3772f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3773f9708b43SJan Beulich offsetof(struct tcphdr, 3774f9708b43SJan Beulich check))) 3775f9708b43SJan Beulich err = -EPROTO; 3776f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 3777f9708b43SJan Beulich 3778f9708b43SJan Beulich case IPPROTO_UDP: 3779f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 3780f9708b43SJan Beulich off + sizeof(struct udphdr)); 3781f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3782f9708b43SJan Beulich offsetof(struct udphdr, 3783f9708b43SJan Beulich check))) 3784f9708b43SJan Beulich err = -EPROTO; 3785f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 3786f9708b43SJan Beulich } 3787f9708b43SJan Beulich 3788f9708b43SJan Beulich return ERR_PTR(-EPROTO); 3789f9708b43SJan Beulich } 3790f9708b43SJan Beulich 3791ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 3792ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 3793ed1f50c3SPaul Durrant */ 3794ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 3795ed1f50c3SPaul Durrant 3796f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 3797ed1f50c3SPaul Durrant { 3798ed1f50c3SPaul Durrant unsigned int off; 3799ed1f50c3SPaul Durrant bool fragment; 3800f9708b43SJan Beulich __sum16 *csum; 3801ed1f50c3SPaul Durrant int err; 3802ed1f50c3SPaul Durrant 3803ed1f50c3SPaul Durrant fragment = false; 3804ed1f50c3SPaul Durrant 3805ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3806ed1f50c3SPaul Durrant sizeof(struct iphdr), 3807ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 3808ed1f50c3SPaul Durrant if (err < 0) 3809ed1f50c3SPaul Durrant goto out; 3810ed1f50c3SPaul Durrant 3811ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 3812ed1f50c3SPaul Durrant fragment = true; 3813ed1f50c3SPaul Durrant 3814ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 3815ed1f50c3SPaul Durrant 3816ed1f50c3SPaul Durrant err = -EPROTO; 3817ed1f50c3SPaul Durrant 3818ed1f50c3SPaul Durrant if (fragment) 3819ed1f50c3SPaul Durrant goto out; 3820ed1f50c3SPaul Durrant 3821f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 3822f9708b43SJan Beulich if (IS_ERR(csum)) 3823f9708b43SJan Beulich return PTR_ERR(csum); 3824ed1f50c3SPaul Durrant 3825ed1f50c3SPaul Durrant if (recalculate) 3826f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 3827ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 3828ed1f50c3SPaul Durrant skb->len - off, 3829f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 3830ed1f50c3SPaul Durrant err = 0; 3831ed1f50c3SPaul Durrant 3832ed1f50c3SPaul Durrant out: 3833ed1f50c3SPaul Durrant return err; 3834ed1f50c3SPaul Durrant } 3835ed1f50c3SPaul Durrant 3836ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 3837ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 3838ed1f50c3SPaul Durrant */ 3839ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 3840ed1f50c3SPaul Durrant 3841ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 3842ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 3843ed1f50c3SPaul Durrant 3844ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 3845ed1f50c3SPaul Durrant { 3846ed1f50c3SPaul Durrant int err; 3847ed1f50c3SPaul Durrant u8 nexthdr; 3848ed1f50c3SPaul Durrant unsigned int off; 3849ed1f50c3SPaul Durrant unsigned int len; 3850ed1f50c3SPaul Durrant bool fragment; 3851ed1f50c3SPaul Durrant bool done; 3852f9708b43SJan Beulich __sum16 *csum; 3853ed1f50c3SPaul Durrant 3854ed1f50c3SPaul Durrant fragment = false; 3855ed1f50c3SPaul Durrant done = false; 3856ed1f50c3SPaul Durrant 3857ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 3858ed1f50c3SPaul Durrant 3859ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 3860ed1f50c3SPaul Durrant if (err < 0) 3861ed1f50c3SPaul Durrant goto out; 3862ed1f50c3SPaul Durrant 3863ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 3864ed1f50c3SPaul Durrant 3865ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 3866ed1f50c3SPaul Durrant while (off <= len && !done) { 3867ed1f50c3SPaul Durrant switch (nexthdr) { 3868ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 3869ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 3870ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 3871ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 3872ed1f50c3SPaul Durrant 3873ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3874ed1f50c3SPaul Durrant off + 3875ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 3876ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3877ed1f50c3SPaul Durrant if (err < 0) 3878ed1f50c3SPaul Durrant goto out; 3879ed1f50c3SPaul Durrant 3880ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 3881ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3882ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 3883ed1f50c3SPaul Durrant break; 3884ed1f50c3SPaul Durrant } 3885ed1f50c3SPaul Durrant case IPPROTO_AH: { 3886ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 3887ed1f50c3SPaul Durrant 3888ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3889ed1f50c3SPaul Durrant off + 3890ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 3891ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3892ed1f50c3SPaul Durrant if (err < 0) 3893ed1f50c3SPaul Durrant goto out; 3894ed1f50c3SPaul Durrant 3895ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 3896ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3897ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 3898ed1f50c3SPaul Durrant break; 3899ed1f50c3SPaul Durrant } 3900ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 3901ed1f50c3SPaul Durrant struct frag_hdr *hp; 3902ed1f50c3SPaul Durrant 3903ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3904ed1f50c3SPaul Durrant off + 3905ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 3906ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 3907ed1f50c3SPaul Durrant if (err < 0) 3908ed1f50c3SPaul Durrant goto out; 3909ed1f50c3SPaul Durrant 3910ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 3911ed1f50c3SPaul Durrant 3912ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 3913ed1f50c3SPaul Durrant fragment = true; 3914ed1f50c3SPaul Durrant 3915ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 3916ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 3917ed1f50c3SPaul Durrant break; 3918ed1f50c3SPaul Durrant } 3919ed1f50c3SPaul Durrant default: 3920ed1f50c3SPaul Durrant done = true; 3921ed1f50c3SPaul Durrant break; 3922ed1f50c3SPaul Durrant } 3923ed1f50c3SPaul Durrant } 3924ed1f50c3SPaul Durrant 3925ed1f50c3SPaul Durrant err = -EPROTO; 3926ed1f50c3SPaul Durrant 3927ed1f50c3SPaul Durrant if (!done || fragment) 3928ed1f50c3SPaul Durrant goto out; 3929ed1f50c3SPaul Durrant 3930f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 3931f9708b43SJan Beulich if (IS_ERR(csum)) 3932f9708b43SJan Beulich return PTR_ERR(csum); 3933ed1f50c3SPaul Durrant 3934ed1f50c3SPaul Durrant if (recalculate) 3935f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 3936ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 3937f9708b43SJan Beulich skb->len - off, nexthdr, 0); 3938ed1f50c3SPaul Durrant err = 0; 3939ed1f50c3SPaul Durrant 3940ed1f50c3SPaul Durrant out: 3941ed1f50c3SPaul Durrant return err; 3942ed1f50c3SPaul Durrant } 3943ed1f50c3SPaul Durrant 3944ed1f50c3SPaul Durrant /** 3945ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 3946ed1f50c3SPaul Durrant * @skb: the skb to set up 3947ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 3948ed1f50c3SPaul Durrant */ 3949ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 3950ed1f50c3SPaul Durrant { 3951ed1f50c3SPaul Durrant int err; 3952ed1f50c3SPaul Durrant 3953ed1f50c3SPaul Durrant switch (skb->protocol) { 3954ed1f50c3SPaul Durrant case htons(ETH_P_IP): 3955f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 3956ed1f50c3SPaul Durrant break; 3957ed1f50c3SPaul Durrant 3958ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 3959ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 3960ed1f50c3SPaul Durrant break; 3961ed1f50c3SPaul Durrant 3962ed1f50c3SPaul Durrant default: 3963ed1f50c3SPaul Durrant err = -EPROTO; 3964ed1f50c3SPaul Durrant break; 3965ed1f50c3SPaul Durrant } 3966ed1f50c3SPaul Durrant 3967ed1f50c3SPaul Durrant return err; 3968ed1f50c3SPaul Durrant } 3969ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 3970ed1f50c3SPaul Durrant 39714497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 39724497b076SBen Hutchings { 3973e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 3974e87cc472SJoe Perches skb->dev->name); 39754497b076SBen Hutchings } 39764497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 3977bad43ca8SEric Dumazet 3978bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 3979bad43ca8SEric Dumazet { 39803d861f66SEric Dumazet if (head_stolen) { 39813d861f66SEric Dumazet skb_release_head_state(skb); 3982bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 39833d861f66SEric Dumazet } else { 3984bad43ca8SEric Dumazet __kfree_skb(skb); 3985bad43ca8SEric Dumazet } 39863d861f66SEric Dumazet } 3987bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 3988bad43ca8SEric Dumazet 3989bad43ca8SEric Dumazet /** 3990bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 3991bad43ca8SEric Dumazet * @to: prior buffer 3992bad43ca8SEric Dumazet * @from: buffer to add 3993bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 3994c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 3995bad43ca8SEric Dumazet */ 3996bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 3997bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 3998bad43ca8SEric Dumazet { 3999bad43ca8SEric Dumazet int i, delta, len = from->len; 4000bad43ca8SEric Dumazet 4001bad43ca8SEric Dumazet *fragstolen = false; 4002bad43ca8SEric Dumazet 4003bad43ca8SEric Dumazet if (skb_cloned(to)) 4004bad43ca8SEric Dumazet return false; 4005bad43ca8SEric Dumazet 4006bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 4007e93a0435SEric Dumazet if (len) 4008bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 4009bad43ca8SEric Dumazet *delta_truesize = 0; 4010bad43ca8SEric Dumazet return true; 4011bad43ca8SEric Dumazet } 4012bad43ca8SEric Dumazet 4013bad43ca8SEric Dumazet if (skb_has_frag_list(to) || skb_has_frag_list(from)) 4014bad43ca8SEric Dumazet return false; 4015bad43ca8SEric Dumazet 4016bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 4017bad43ca8SEric Dumazet struct page *page; 4018bad43ca8SEric Dumazet unsigned int offset; 4019bad43ca8SEric Dumazet 4020bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4021bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 4022bad43ca8SEric Dumazet return false; 4023bad43ca8SEric Dumazet 4024bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 4025bad43ca8SEric Dumazet return false; 4026bad43ca8SEric Dumazet 4027bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4028bad43ca8SEric Dumazet 4029bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 4030bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 4031bad43ca8SEric Dumazet 4032bad43ca8SEric Dumazet skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 4033bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 4034bad43ca8SEric Dumazet *fragstolen = true; 4035bad43ca8SEric Dumazet } else { 4036bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4037bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 4038bad43ca8SEric Dumazet return false; 4039bad43ca8SEric Dumazet 4040f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 4041bad43ca8SEric Dumazet } 4042bad43ca8SEric Dumazet 4043bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 4044bad43ca8SEric Dumazet 4045bad43ca8SEric Dumazet memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 4046bad43ca8SEric Dumazet skb_shinfo(from)->frags, 4047bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 4048bad43ca8SEric Dumazet skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 4049bad43ca8SEric Dumazet 4050bad43ca8SEric Dumazet if (!skb_cloned(from)) 4051bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags = 0; 4052bad43ca8SEric Dumazet 40538ea853fdSLi RongQing /* if the skb is not cloned this does nothing 40548ea853fdSLi RongQing * since we set nr_frags to 0. 40558ea853fdSLi RongQing */ 4056bad43ca8SEric Dumazet for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 4057bad43ca8SEric Dumazet skb_frag_ref(from, i); 4058bad43ca8SEric Dumazet 4059bad43ca8SEric Dumazet to->truesize += delta; 4060bad43ca8SEric Dumazet to->len += len; 4061bad43ca8SEric Dumazet to->data_len += len; 4062bad43ca8SEric Dumazet 4063bad43ca8SEric Dumazet *delta_truesize = delta; 4064bad43ca8SEric Dumazet return true; 4065bad43ca8SEric Dumazet } 4066bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 4067621e84d6SNicolas Dichtel 4068621e84d6SNicolas Dichtel /** 40698b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 4070621e84d6SNicolas Dichtel * 4071621e84d6SNicolas Dichtel * @skb: buffer to clean 40728b27f277SNicolas Dichtel * @xnet: packet is crossing netns 4073621e84d6SNicolas Dichtel * 40748b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 40758b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 40768b27f277SNicolas Dichtel * operations. 40778b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 40788b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 40798b27f277SNicolas Dichtel * skb that could impact namespace isolation. 4080621e84d6SNicolas Dichtel */ 40818b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4082621e84d6SNicolas Dichtel { 40838b27f277SNicolas Dichtel if (xnet) 4084621e84d6SNicolas Dichtel skb_orphan(skb); 4085621e84d6SNicolas Dichtel skb->tstamp.tv64 = 0; 4086621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 4087621e84d6SNicolas Dichtel skb->skb_iif = 0; 408860ff7467SWANG Cong skb->ignore_df = 0; 4089621e84d6SNicolas Dichtel skb_dst_drop(skb); 4090621e84d6SNicolas Dichtel skb->mark = 0; 4091621e84d6SNicolas Dichtel secpath_reset(skb); 4092621e84d6SNicolas Dichtel nf_reset(skb); 4093621e84d6SNicolas Dichtel nf_reset_trace(skb); 4094621e84d6SNicolas Dichtel } 4095621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 4096de960aa9SFlorian Westphal 4097de960aa9SFlorian Westphal /** 4098de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4099de960aa9SFlorian Westphal * 4100de960aa9SFlorian Westphal * @skb: GSO skb 4101de960aa9SFlorian Westphal * 4102de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 4103de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 4104de960aa9SFlorian Westphal * 4105de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4106de960aa9SFlorian Westphal */ 4107de960aa9SFlorian Westphal unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4108de960aa9SFlorian Westphal { 4109de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 4110f993bc25SFlorian Westphal unsigned int thlen = 0; 4111f993bc25SFlorian Westphal 4112f993bc25SFlorian Westphal if (skb->encapsulation) { 4113f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 4114f993bc25SFlorian Westphal skb_transport_header(skb); 4115de960aa9SFlorian Westphal 4116de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 4117f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 4118f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 4119f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 4120f993bc25SFlorian Westphal } 41216d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 41226d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 41236d39d589SFlorian Westphal * accounted for. 41246d39d589SFlorian Westphal */ 4125f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 4126de960aa9SFlorian Westphal } 4127de960aa9SFlorian Westphal EXPORT_SYMBOL_GPL(skb_gso_transport_seglen); 41280d5501c1SVlad Yasevich 41290d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 41300d5501c1SVlad Yasevich { 41310d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 41320d5501c1SVlad Yasevich kfree_skb(skb); 41330d5501c1SVlad Yasevich return NULL; 41340d5501c1SVlad Yasevich } 41350d5501c1SVlad Yasevich 41360d5501c1SVlad Yasevich memmove(skb->data - ETH_HLEN, skb->data - VLAN_ETH_HLEN, 2 * ETH_ALEN); 41370d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 41380d5501c1SVlad Yasevich return skb; 41390d5501c1SVlad Yasevich } 41400d5501c1SVlad Yasevich 41410d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 41420d5501c1SVlad Yasevich { 41430d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 41440d5501c1SVlad Yasevich u16 vlan_tci; 41450d5501c1SVlad Yasevich 41460d5501c1SVlad Yasevich if (unlikely(vlan_tx_tag_present(skb))) { 41470d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 41480d5501c1SVlad Yasevich return skb; 41490d5501c1SVlad Yasevich } 41500d5501c1SVlad Yasevich 41510d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 41520d5501c1SVlad Yasevich if (unlikely(!skb)) 41530d5501c1SVlad Yasevich goto err_free; 41540d5501c1SVlad Yasevich 41550d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 41560d5501c1SVlad Yasevich goto err_free; 41570d5501c1SVlad Yasevich 41580d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 41590d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 41600d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 41610d5501c1SVlad Yasevich 41620d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 41630d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 41640d5501c1SVlad Yasevich 41650d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 41660d5501c1SVlad Yasevich if (unlikely(!skb)) 41670d5501c1SVlad Yasevich goto err_free; 41680d5501c1SVlad Yasevich 41690d5501c1SVlad Yasevich skb_reset_network_header(skb); 41700d5501c1SVlad Yasevich skb_reset_transport_header(skb); 41710d5501c1SVlad Yasevich skb_reset_mac_len(skb); 41720d5501c1SVlad Yasevich 41730d5501c1SVlad Yasevich return skb; 41740d5501c1SVlad Yasevich 41750d5501c1SVlad Yasevich err_free: 41760d5501c1SVlad Yasevich kfree_skb(skb); 41770d5501c1SVlad Yasevich return NULL; 41780d5501c1SVlad Yasevich } 41790d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 41802e4e4410SEric Dumazet 4181e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 4182e2195121SJiri Pirko { 4183e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 4184e2195121SJiri Pirko return -ENOMEM; 4185e2195121SJiri Pirko 4186e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 4187e2195121SJiri Pirko return 0; 4188e2195121SJiri Pirko 4189e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 4190e2195121SJiri Pirko } 4191e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 4192e2195121SJiri Pirko 419393515d53SJiri Pirko /* remove VLAN header from packet and update csum accordingly. */ 419493515d53SJiri Pirko static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 419593515d53SJiri Pirko { 419693515d53SJiri Pirko struct vlan_hdr *vhdr; 419793515d53SJiri Pirko unsigned int offset = skb->data - skb_mac_header(skb); 419893515d53SJiri Pirko int err; 419993515d53SJiri Pirko 420093515d53SJiri Pirko __skb_push(skb, offset); 420193515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 420293515d53SJiri Pirko if (unlikely(err)) 420393515d53SJiri Pirko goto pull; 420493515d53SJiri Pirko 420593515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 420693515d53SJiri Pirko 420793515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 420893515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 420993515d53SJiri Pirko 421093515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 421193515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 421293515d53SJiri Pirko 421393515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 421493515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 421593515d53SJiri Pirko 421693515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 421793515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 421893515d53SJiri Pirko 421993515d53SJiri Pirko skb_reset_mac_len(skb); 422093515d53SJiri Pirko pull: 422193515d53SJiri Pirko __skb_pull(skb, offset); 422293515d53SJiri Pirko 422393515d53SJiri Pirko return err; 422493515d53SJiri Pirko } 422593515d53SJiri Pirko 422693515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 422793515d53SJiri Pirko { 422893515d53SJiri Pirko u16 vlan_tci; 422993515d53SJiri Pirko __be16 vlan_proto; 423093515d53SJiri Pirko int err; 423193515d53SJiri Pirko 423293515d53SJiri Pirko if (likely(vlan_tx_tag_present(skb))) { 423393515d53SJiri Pirko skb->vlan_tci = 0; 423493515d53SJiri Pirko } else { 423593515d53SJiri Pirko if (unlikely((skb->protocol != htons(ETH_P_8021Q) && 423693515d53SJiri Pirko skb->protocol != htons(ETH_P_8021AD)) || 423793515d53SJiri Pirko skb->len < VLAN_ETH_HLEN)) 423893515d53SJiri Pirko return 0; 423993515d53SJiri Pirko 424093515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 424193515d53SJiri Pirko if (err) 424293515d53SJiri Pirko return err; 424393515d53SJiri Pirko } 424493515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 424593515d53SJiri Pirko if (likely((skb->protocol != htons(ETH_P_8021Q) && 424693515d53SJiri Pirko skb->protocol != htons(ETH_P_8021AD)) || 424793515d53SJiri Pirko skb->len < VLAN_ETH_HLEN)) 424893515d53SJiri Pirko return 0; 424993515d53SJiri Pirko 425093515d53SJiri Pirko vlan_proto = skb->protocol; 425193515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 425293515d53SJiri Pirko if (unlikely(err)) 425393515d53SJiri Pirko return err; 425493515d53SJiri Pirko 425593515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 425693515d53SJiri Pirko return 0; 425793515d53SJiri Pirko } 425893515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 425993515d53SJiri Pirko 426093515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 426193515d53SJiri Pirko { 426293515d53SJiri Pirko if (vlan_tx_tag_present(skb)) { 426393515d53SJiri Pirko unsigned int offset = skb->data - skb_mac_header(skb); 426493515d53SJiri Pirko int err; 426593515d53SJiri Pirko 426693515d53SJiri Pirko /* __vlan_insert_tag expect skb->data pointing to mac header. 426793515d53SJiri Pirko * So change skb->data before calling it and change back to 426893515d53SJiri Pirko * original position later 426993515d53SJiri Pirko */ 427093515d53SJiri Pirko __skb_push(skb, offset); 427193515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 427293515d53SJiri Pirko vlan_tx_tag_get(skb)); 427393515d53SJiri Pirko if (err) 427493515d53SJiri Pirko return err; 427593515d53SJiri Pirko skb->protocol = skb->vlan_proto; 427693515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 427793515d53SJiri Pirko __skb_pull(skb, offset); 427893515d53SJiri Pirko 427993515d53SJiri Pirko if (skb->ip_summed == CHECKSUM_COMPLETE) 428093515d53SJiri Pirko skb->csum = csum_add(skb->csum, csum_partial(skb->data 428193515d53SJiri Pirko + (2 * ETH_ALEN), VLAN_HLEN, 0)); 428293515d53SJiri Pirko } 428393515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 428493515d53SJiri Pirko return 0; 428593515d53SJiri Pirko } 428693515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 428793515d53SJiri Pirko 42882e4e4410SEric Dumazet /** 42892e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 42902e4e4410SEric Dumazet * 4291de3f0d0eSMasanari Iida * @header_len: size of linear part 4292de3f0d0eSMasanari Iida * @data_len: needed length in frags 4293de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 4294de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 4295de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 42962e4e4410SEric Dumazet * 42972e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 42982e4e4410SEric Dumazet */ 42992e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 43002e4e4410SEric Dumazet unsigned long data_len, 43012e4e4410SEric Dumazet int max_page_order, 43022e4e4410SEric Dumazet int *errcode, 43032e4e4410SEric Dumazet gfp_t gfp_mask) 43042e4e4410SEric Dumazet { 43052e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 43062e4e4410SEric Dumazet unsigned long chunk; 43072e4e4410SEric Dumazet struct sk_buff *skb; 43082e4e4410SEric Dumazet struct page *page; 43092e4e4410SEric Dumazet gfp_t gfp_head; 43102e4e4410SEric Dumazet int i; 43112e4e4410SEric Dumazet 43122e4e4410SEric Dumazet *errcode = -EMSGSIZE; 43132e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 43142e4e4410SEric Dumazet * high order pages... 43152e4e4410SEric Dumazet */ 43162e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 43172e4e4410SEric Dumazet return NULL; 43182e4e4410SEric Dumazet 43192e4e4410SEric Dumazet gfp_head = gfp_mask; 43202e4e4410SEric Dumazet if (gfp_head & __GFP_WAIT) 43212e4e4410SEric Dumazet gfp_head |= __GFP_REPEAT; 43222e4e4410SEric Dumazet 43232e4e4410SEric Dumazet *errcode = -ENOBUFS; 43242e4e4410SEric Dumazet skb = alloc_skb(header_len, gfp_head); 43252e4e4410SEric Dumazet if (!skb) 43262e4e4410SEric Dumazet return NULL; 43272e4e4410SEric Dumazet 43282e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 43292e4e4410SEric Dumazet 43302e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 43312e4e4410SEric Dumazet int order = max_page_order; 43322e4e4410SEric Dumazet 43332e4e4410SEric Dumazet while (order) { 43342e4e4410SEric Dumazet if (npages >= 1 << order) { 43352e4e4410SEric Dumazet page = alloc_pages(gfp_mask | 43362e4e4410SEric Dumazet __GFP_COMP | 43372e4e4410SEric Dumazet __GFP_NOWARN | 43382e4e4410SEric Dumazet __GFP_NORETRY, 43392e4e4410SEric Dumazet order); 43402e4e4410SEric Dumazet if (page) 43412e4e4410SEric Dumazet goto fill_page; 43422e4e4410SEric Dumazet /* Do not retry other high order allocations */ 43432e4e4410SEric Dumazet order = 1; 43442e4e4410SEric Dumazet max_page_order = 0; 43452e4e4410SEric Dumazet } 43462e4e4410SEric Dumazet order--; 43472e4e4410SEric Dumazet } 43482e4e4410SEric Dumazet page = alloc_page(gfp_mask); 43492e4e4410SEric Dumazet if (!page) 43502e4e4410SEric Dumazet goto failure; 43512e4e4410SEric Dumazet fill_page: 43522e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 43532e4e4410SEric Dumazet PAGE_SIZE << order); 43542e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 43552e4e4410SEric Dumazet data_len -= chunk; 43562e4e4410SEric Dumazet npages -= 1 << order; 43572e4e4410SEric Dumazet } 43582e4e4410SEric Dumazet return skb; 43592e4e4410SEric Dumazet 43602e4e4410SEric Dumazet failure: 43612e4e4410SEric Dumazet kfree_skb(skb); 43622e4e4410SEric Dumazet return NULL; 43632e4e4410SEric Dumazet } 43642e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 4365