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> 5290017accSMarcelo Ricardo Leitner #include <linux/sctp.h> 531da177e4SLinus Torvalds #include <linux/netdevice.h> 541da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 551da177e4SLinus Torvalds #include <net/pkt_sched.h> 561da177e4SLinus Torvalds #endif 571da177e4SLinus Torvalds #include <linux/string.h> 581da177e4SLinus Torvalds #include <linux/skbuff.h> 599c55e01cSJens Axboe #include <linux/splice.h> 601da177e4SLinus Torvalds #include <linux/cache.h> 611da177e4SLinus Torvalds #include <linux/rtnetlink.h> 621da177e4SLinus Torvalds #include <linux/init.h> 63716ea3a7SDavid Howells #include <linux/scatterlist.h> 64ac45f602SPatrick Ohly #include <linux/errqueue.h> 65268bb0ceSLinus Torvalds #include <linux/prefetch.h> 660d5501c1SVlad Yasevich #include <linux/if_vlan.h> 671da177e4SLinus Torvalds 681da177e4SLinus Torvalds #include <net/protocol.h> 691da177e4SLinus Torvalds #include <net/dst.h> 701da177e4SLinus Torvalds #include <net/sock.h> 711da177e4SLinus Torvalds #include <net/checksum.h> 72ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 731da177e4SLinus Torvalds #include <net/xfrm.h> 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds #include <asm/uaccess.h> 76ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7751c56b00SEric Dumazet #include <linux/highmem.h> 78b245be1fSWillem de Bruijn #include <linux/capability.h> 79b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 80a1f8e7f7SAl Viro 81d7e8883cSEric Dumazet struct kmem_cache *skbuff_head_cache __read_mostly; 82e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 835f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 845f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 851da177e4SLinus Torvalds 861da177e4SLinus Torvalds /** 87f05de73bSJean Sacren * skb_panic - private function for out-of-line support 881da177e4SLinus Torvalds * @skb: buffer 891da177e4SLinus Torvalds * @sz: size 90f05de73bSJean Sacren * @addr: address 9199d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 921da177e4SLinus Torvalds * 93f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 94f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 95f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 96f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 971da177e4SLinus Torvalds */ 98f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 9999d5851eSJames Hogan const char msg[]) 1001da177e4SLinus Torvalds { 101e005d193SJoe Perches pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 10299d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1034305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 10426095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1051da177e4SLinus Torvalds BUG(); 1061da177e4SLinus Torvalds } 1071da177e4SLinus Torvalds 108f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1091da177e4SLinus Torvalds { 110f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1111da177e4SLinus Torvalds } 1121da177e4SLinus Torvalds 113f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 114f05de73bSJean Sacren { 115f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 116f05de73bSJean Sacren } 117c93bdd0eSMel Gorman 118c93bdd0eSMel Gorman /* 119c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 120c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 121c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 122c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 123c93bdd0eSMel Gorman * memory is free 124c93bdd0eSMel Gorman */ 125c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 126c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 12761c5e88aSstephen hemminger 12861c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 12961c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 130c93bdd0eSMel Gorman { 131c93bdd0eSMel Gorman void *obj; 132c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 133c93bdd0eSMel Gorman 134c93bdd0eSMel Gorman /* 135c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 136c93bdd0eSMel Gorman * to the reserves, fail. 137c93bdd0eSMel Gorman */ 138c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 139c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 140c93bdd0eSMel Gorman node); 141c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 142c93bdd0eSMel Gorman goto out; 143c93bdd0eSMel Gorman 144c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 145c93bdd0eSMel Gorman ret_pfmemalloc = true; 146c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 147c93bdd0eSMel Gorman 148c93bdd0eSMel Gorman out: 149c93bdd0eSMel Gorman if (pfmemalloc) 150c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 151c93bdd0eSMel Gorman 152c93bdd0eSMel Gorman return obj; 153c93bdd0eSMel Gorman } 154c93bdd0eSMel Gorman 1551da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1561da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1571da177e4SLinus Torvalds * [BEEP] leaks. 1581da177e4SLinus Torvalds * 1591da177e4SLinus Torvalds */ 1601da177e4SLinus Torvalds 1610ebd0ac5SPatrick McHardy struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) 1620ebd0ac5SPatrick McHardy { 1630ebd0ac5SPatrick McHardy struct sk_buff *skb; 1640ebd0ac5SPatrick McHardy 1650ebd0ac5SPatrick McHardy /* Get the HEAD */ 1660ebd0ac5SPatrick McHardy skb = kmem_cache_alloc_node(skbuff_head_cache, 1670ebd0ac5SPatrick McHardy gfp_mask & ~__GFP_DMA, node); 1680ebd0ac5SPatrick McHardy if (!skb) 1690ebd0ac5SPatrick McHardy goto out; 1700ebd0ac5SPatrick McHardy 1710ebd0ac5SPatrick McHardy /* 1720ebd0ac5SPatrick McHardy * Only clear those fields we need to clear, not those that we will 1730ebd0ac5SPatrick McHardy * actually initialise below. Hence, don't put any more fields after 1740ebd0ac5SPatrick McHardy * the tail pointer in struct sk_buff! 1750ebd0ac5SPatrick McHardy */ 1760ebd0ac5SPatrick McHardy memset(skb, 0, offsetof(struct sk_buff, tail)); 1775e71d9d7SPablo Neira skb->head = NULL; 1780ebd0ac5SPatrick McHardy skb->truesize = sizeof(struct sk_buff); 1790ebd0ac5SPatrick McHardy atomic_set(&skb->users, 1); 1800ebd0ac5SPatrick McHardy 18135d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 1820ebd0ac5SPatrick McHardy out: 1830ebd0ac5SPatrick McHardy return skb; 1840ebd0ac5SPatrick McHardy } 1850ebd0ac5SPatrick McHardy 1861da177e4SLinus Torvalds /** 187d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1881da177e4SLinus Torvalds * @size: size to allocate 1891da177e4SLinus Torvalds * @gfp_mask: allocation mask 190c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 191c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 192c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 193c93bdd0eSMel Gorman * allocations in case the data is required for writeback 194b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1951da177e4SLinus Torvalds * 1961da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 19794b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 19894b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 1991da177e4SLinus Torvalds * 2001da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 2011da177e4SLinus Torvalds * %GFP_ATOMIC. 2021da177e4SLinus Torvalds */ 203dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 204c93bdd0eSMel Gorman int flags, int node) 2051da177e4SLinus Torvalds { 206e18b890bSChristoph Lameter struct kmem_cache *cache; 2074947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 2081da177e4SLinus Torvalds struct sk_buff *skb; 2091da177e4SLinus Torvalds u8 *data; 210c93bdd0eSMel Gorman bool pfmemalloc; 2111da177e4SLinus Torvalds 212c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 213c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 214c93bdd0eSMel Gorman 215c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 216c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 2178798b3fbSHerbert Xu 2181da177e4SLinus Torvalds /* Get the HEAD */ 219b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 2201da177e4SLinus Torvalds if (!skb) 2211da177e4SLinus Torvalds goto out; 222ec7d2f2cSEric Dumazet prefetchw(skb); 2231da177e4SLinus Torvalds 22487fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 22587fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 22687fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 22787fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 22887fb4b7bSEric Dumazet */ 229bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 23087fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 231c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2321da177e4SLinus Torvalds if (!data) 2331da177e4SLinus Torvalds goto nodata; 23487fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 23587fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 23687fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 23787fb4b7bSEric Dumazet */ 23887fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 239ec7d2f2cSEric Dumazet prefetchw(data + size); 2401da177e4SLinus Torvalds 241ca0605a7SArnaldo Carvalho de Melo /* 242c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 243c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 244c8005785SJohannes Berg * the tail pointer in struct sk_buff! 245ca0605a7SArnaldo Carvalho de Melo */ 246ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 24787fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 24887fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 249c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 2501da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2511da177e4SLinus Torvalds skb->head = data; 2521da177e4SLinus Torvalds skb->data = data; 25327a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2544305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 25535d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 25635d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 25719633e12SStephen Hemminger 2584947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2594947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 260ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2614947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 262c2aa3665SEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 2634947d3efSBenjamin LaHaise 264c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 265d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 2661da177e4SLinus Torvalds 267d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 268d0bf4a9eSEric Dumazet 269d0bf4a9eSEric Dumazet kmemcheck_annotate_bitfield(&fclones->skb2, flags1); 270d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 271d0bf4a9eSEric Dumazet atomic_set(&fclones->fclone_ref, 1); 272d179cd12SDavid S. Miller 2736ffe75ebSEric Dumazet fclones->skb2.fclone = SKB_FCLONE_CLONE; 274d0bf4a9eSEric Dumazet fclones->skb2.pfmemalloc = pfmemalloc; 275d179cd12SDavid S. Miller } 2761da177e4SLinus Torvalds out: 2771da177e4SLinus Torvalds return skb; 2781da177e4SLinus Torvalds nodata: 2798798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2801da177e4SLinus Torvalds skb = NULL; 2811da177e4SLinus Torvalds goto out; 2821da177e4SLinus Torvalds } 283b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds /** 2862ea2f62cSEric Dumazet * __build_skb - build a network buffer 287b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2882ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 289b2b5ce9dSEric Dumazet * 290b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 291deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2922ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2932ea2f62cSEric Dumazet * or vmalloc() 294b2b5ce9dSEric Dumazet * The return is the new skb buffer. 295b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 296b2b5ce9dSEric Dumazet * Notes : 297b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 298b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 299b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 300b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 301b2b5ce9dSEric Dumazet * before giving packet to stack. 302b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 303b2b5ce9dSEric Dumazet */ 3042ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 305b2b5ce9dSEric Dumazet { 306b2b5ce9dSEric Dumazet struct skb_shared_info *shinfo; 307b2b5ce9dSEric Dumazet struct sk_buff *skb; 308d3836f21SEric Dumazet unsigned int size = frag_size ? : ksize(data); 309b2b5ce9dSEric Dumazet 310b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 311b2b5ce9dSEric Dumazet if (!skb) 312b2b5ce9dSEric Dumazet return NULL; 313b2b5ce9dSEric Dumazet 314d3836f21SEric Dumazet size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 315b2b5ce9dSEric Dumazet 316b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 317b2b5ce9dSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 318b2b5ce9dSEric Dumazet atomic_set(&skb->users, 1); 319b2b5ce9dSEric Dumazet skb->head = data; 320b2b5ce9dSEric Dumazet skb->data = data; 321b2b5ce9dSEric Dumazet skb_reset_tail_pointer(skb); 322b2b5ce9dSEric Dumazet skb->end = skb->tail + size; 32335d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 32435d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 325b2b5ce9dSEric Dumazet 326b2b5ce9dSEric Dumazet /* make sure we initialize shinfo sequentially */ 327b2b5ce9dSEric Dumazet shinfo = skb_shinfo(skb); 328b2b5ce9dSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 329b2b5ce9dSEric Dumazet atomic_set(&shinfo->dataref, 1); 330b2b5ce9dSEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 331b2b5ce9dSEric Dumazet 332b2b5ce9dSEric Dumazet return skb; 333b2b5ce9dSEric Dumazet } 3342ea2f62cSEric Dumazet 3352ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3362ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3372ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 3382ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 3392ea2f62cSEric Dumazet */ 3402ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3412ea2f62cSEric Dumazet { 3422ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3432ea2f62cSEric Dumazet 3442ea2f62cSEric Dumazet if (skb && frag_size) { 3452ea2f62cSEric Dumazet skb->head_frag = 1; 3462f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 3472ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3482ea2f62cSEric Dumazet } 3492ea2f62cSEric Dumazet return skb; 3502ea2f62cSEric Dumazet } 351b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 352b2b5ce9dSEric Dumazet 353795bb1c0SJesper Dangaard Brouer #define NAPI_SKB_CACHE_SIZE 64 354795bb1c0SJesper Dangaard Brouer 355795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache { 356795bb1c0SJesper Dangaard Brouer struct page_frag_cache page; 357795bb1c0SJesper Dangaard Brouer size_t skb_count; 358795bb1c0SJesper Dangaard Brouer void *skb_cache[NAPI_SKB_CACHE_SIZE]; 359795bb1c0SJesper Dangaard Brouer }; 360795bb1c0SJesper Dangaard Brouer 361b63ae8caSAlexander Duyck static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 362795bb1c0SJesper Dangaard Brouer static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 363ffde7328SAlexander Duyck 364ffde7328SAlexander Duyck static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 365ffde7328SAlexander Duyck { 366b63ae8caSAlexander Duyck struct page_frag_cache *nc; 367ffde7328SAlexander Duyck unsigned long flags; 368ffde7328SAlexander Duyck void *data; 369ffde7328SAlexander Duyck 370ffde7328SAlexander Duyck local_irq_save(flags); 3719451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 3729451980aSAlexander Duyck data = __alloc_page_frag(nc, fragsz, gfp_mask); 3736f532612SEric Dumazet local_irq_restore(flags); 3746f532612SEric Dumazet return data; 3756f532612SEric Dumazet } 376c93bdd0eSMel Gorman 377c93bdd0eSMel Gorman /** 378c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 379c93bdd0eSMel Gorman * @fragsz: fragment size 380c93bdd0eSMel Gorman * 381c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 382c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 383c93bdd0eSMel Gorman */ 384c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 385c93bdd0eSMel Gorman { 386c93bdd0eSMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 387c93bdd0eSMel Gorman } 3886f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 3896f532612SEric Dumazet 390ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 391ffde7328SAlexander Duyck { 392795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 3939451980aSAlexander Duyck 394795bb1c0SJesper Dangaard Brouer return __alloc_page_frag(&nc->page, fragsz, gfp_mask); 395ffde7328SAlexander Duyck } 396ffde7328SAlexander Duyck 397ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 398ffde7328SAlexander Duyck { 399ffde7328SAlexander Duyck return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 400ffde7328SAlexander Duyck } 401ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 402ffde7328SAlexander Duyck 4036f532612SEric Dumazet /** 404fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 405fd11a83dSAlexander Duyck * @dev: network device to receive on 406d7499160SMasanari Iida * @len: length to allocate 407fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 408fd11a83dSAlexander Duyck * 409fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 410fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 411fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 412fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 413fd11a83dSAlexander Duyck * 414fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 415fd11a83dSAlexander Duyck */ 4169451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4179451980aSAlexander Duyck gfp_t gfp_mask) 418fd11a83dSAlexander Duyck { 419b63ae8caSAlexander Duyck struct page_frag_cache *nc; 4209451980aSAlexander Duyck unsigned long flags; 421fd11a83dSAlexander Duyck struct sk_buff *skb; 4229451980aSAlexander Duyck bool pfmemalloc; 4239451980aSAlexander Duyck void *data; 424fd11a83dSAlexander Duyck 4259451980aSAlexander Duyck len += NET_SKB_PAD; 426fd11a83dSAlexander Duyck 4279451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 428d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 429a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 430a080e7bdSAlexander Duyck if (!skb) 431a080e7bdSAlexander Duyck goto skb_fail; 432a080e7bdSAlexander Duyck goto skb_success; 433a080e7bdSAlexander Duyck } 4349451980aSAlexander Duyck 4359451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4369451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 4379451980aSAlexander Duyck 4389451980aSAlexander Duyck if (sk_memalloc_socks()) 4399451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 4409451980aSAlexander Duyck 4419451980aSAlexander Duyck local_irq_save(flags); 4429451980aSAlexander Duyck 4439451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 4449451980aSAlexander Duyck data = __alloc_page_frag(nc, len, gfp_mask); 4459451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 4469451980aSAlexander Duyck 4479451980aSAlexander Duyck local_irq_restore(flags); 4489451980aSAlexander Duyck 4499451980aSAlexander Duyck if (unlikely(!data)) 4509451980aSAlexander Duyck return NULL; 4519451980aSAlexander Duyck 4529451980aSAlexander Duyck skb = __build_skb(data, len); 4539451980aSAlexander Duyck if (unlikely(!skb)) { 454181edb2bSAlexander Duyck skb_free_frag(data); 4559451980aSAlexander Duyck return NULL; 4569451980aSAlexander Duyck } 4579451980aSAlexander Duyck 4589451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 4599451980aSAlexander Duyck if (pfmemalloc) 4609451980aSAlexander Duyck skb->pfmemalloc = 1; 4619451980aSAlexander Duyck skb->head_frag = 1; 4629451980aSAlexander Duyck 463a080e7bdSAlexander Duyck skb_success: 4648af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4657b2e497aSChristoph Hellwig skb->dev = dev; 466fd11a83dSAlexander Duyck 467a080e7bdSAlexander Duyck skb_fail: 4688af27456SChristoph Hellwig return skb; 4698af27456SChristoph Hellwig } 470b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4711da177e4SLinus Torvalds 472fd11a83dSAlexander Duyck /** 473fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 474fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 475d7499160SMasanari Iida * @len: length to allocate 476fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 477fd11a83dSAlexander Duyck * 478fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 479fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 480fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 481fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 482fd11a83dSAlexander Duyck * 483fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 484fd11a83dSAlexander Duyck */ 4859451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 4869451980aSAlexander Duyck gfp_t gfp_mask) 487fd11a83dSAlexander Duyck { 488795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 489fd11a83dSAlexander Duyck struct sk_buff *skb; 4909451980aSAlexander Duyck void *data; 491fd11a83dSAlexander Duyck 4929451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 493fd11a83dSAlexander Duyck 4949451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 495d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 496a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 497a080e7bdSAlexander Duyck if (!skb) 498a080e7bdSAlexander Duyck goto skb_fail; 499a080e7bdSAlexander Duyck goto skb_success; 500a080e7bdSAlexander Duyck } 5019451980aSAlexander Duyck 5029451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5039451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5049451980aSAlexander Duyck 5059451980aSAlexander Duyck if (sk_memalloc_socks()) 5069451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5079451980aSAlexander Duyck 508795bb1c0SJesper Dangaard Brouer data = __alloc_page_frag(&nc->page, len, gfp_mask); 5099451980aSAlexander Duyck if (unlikely(!data)) 5109451980aSAlexander Duyck return NULL; 5119451980aSAlexander Duyck 5129451980aSAlexander Duyck skb = __build_skb(data, len); 5139451980aSAlexander Duyck if (unlikely(!skb)) { 514181edb2bSAlexander Duyck skb_free_frag(data); 5159451980aSAlexander Duyck return NULL; 5169451980aSAlexander Duyck } 5179451980aSAlexander Duyck 5189451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 519795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5209451980aSAlexander Duyck skb->pfmemalloc = 1; 5219451980aSAlexander Duyck skb->head_frag = 1; 5229451980aSAlexander Duyck 523a080e7bdSAlexander Duyck skb_success: 524fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 525fd11a83dSAlexander Duyck skb->dev = napi->dev; 526fd11a83dSAlexander Duyck 527a080e7bdSAlexander Duyck skb_fail: 528fd11a83dSAlexander Duyck return skb; 529fd11a83dSAlexander Duyck } 530fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 531fd11a83dSAlexander Duyck 532654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 53350269e19SEric Dumazet int size, unsigned int truesize) 534654bed16SPeter Zijlstra { 535654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 536654bed16SPeter Zijlstra skb->len += size; 537654bed16SPeter Zijlstra skb->data_len += size; 53850269e19SEric Dumazet skb->truesize += truesize; 539654bed16SPeter Zijlstra } 540654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 541654bed16SPeter Zijlstra 542f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 543f8e617e1SJason Wang unsigned int truesize) 544f8e617e1SJason Wang { 545f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 546f8e617e1SJason Wang 547f8e617e1SJason Wang skb_frag_size_add(frag, size); 548f8e617e1SJason Wang skb->len += size; 549f8e617e1SJason Wang skb->data_len += size; 550f8e617e1SJason Wang skb->truesize += truesize; 551f8e617e1SJason Wang } 552f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 553f8e617e1SJason Wang 55427b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5551da177e4SLinus Torvalds { 556bd8a7036SEric Dumazet kfree_skb_list(*listp); 55727b437c8SHerbert Xu *listp = NULL; 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds 56027b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 56127b437c8SHerbert Xu { 56227b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 56327b437c8SHerbert Xu } 56427b437c8SHerbert Xu 5651da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5661da177e4SLinus Torvalds { 5671da177e4SLinus Torvalds struct sk_buff *list; 5681da177e4SLinus Torvalds 569fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5701da177e4SLinus Torvalds skb_get(list); 5711da177e4SLinus Torvalds } 5721da177e4SLinus Torvalds 573d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 574d3836f21SEric Dumazet { 575181edb2bSAlexander Duyck unsigned char *head = skb->head; 576181edb2bSAlexander Duyck 577d3836f21SEric Dumazet if (skb->head_frag) 578181edb2bSAlexander Duyck skb_free_frag(head); 579d3836f21SEric Dumazet else 580181edb2bSAlexander Duyck kfree(head); 581d3836f21SEric Dumazet } 582d3836f21SEric Dumazet 5835bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5841da177e4SLinus Torvalds { 585ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 5861da177e4SLinus Torvalds int i; 587ff04a771SEric Dumazet 588ff04a771SEric Dumazet if (skb->cloned && 589ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 590ff04a771SEric Dumazet &shinfo->dataref)) 591ff04a771SEric Dumazet return; 592ff04a771SEric Dumazet 593ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 594ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 5951da177e4SLinus Torvalds 596a6686f2fSShirley Ma /* 597a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 598a6686f2fSShirley Ma * the lower device DMA has done; 599a6686f2fSShirley Ma */ 600ff04a771SEric Dumazet if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) { 601a6686f2fSShirley Ma struct ubuf_info *uarg; 602a6686f2fSShirley Ma 603ff04a771SEric Dumazet uarg = shinfo->destructor_arg; 604a6686f2fSShirley Ma if (uarg->callback) 605e19d6763SMichael S. Tsirkin uarg->callback(uarg, true); 606a6686f2fSShirley Ma } 607a6686f2fSShirley Ma 608ff04a771SEric Dumazet if (shinfo->frag_list) 609ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6101da177e4SLinus Torvalds 611d3836f21SEric Dumazet skb_free_head(skb); 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds 6141da177e4SLinus Torvalds /* 6151da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6161da177e4SLinus Torvalds */ 6172d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6181da177e4SLinus Torvalds { 619d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 620d179cd12SDavid S. Miller 621d179cd12SDavid S. Miller switch (skb->fclone) { 622d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6231da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6246ffe75ebSEric Dumazet return; 625d179cd12SDavid S. Miller 626d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 627d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6286ffe75ebSEric Dumazet 6296ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6306ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6316ffe75ebSEric Dumazet * while here, branch prediction will be good. 6326ffe75ebSEric Dumazet */ 6336ffe75ebSEric Dumazet if (atomic_read(&fclones->fclone_ref) == 1) 6346ffe75ebSEric Dumazet goto fastpath; 635d179cd12SDavid S. Miller break; 636d179cd12SDavid S. Miller 6376ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 638d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 639d179cd12SDavid S. Miller break; 6403ff50b79SStephen Hemminger } 6416ffe75ebSEric Dumazet if (!atomic_dec_and_test(&fclones->fclone_ref)) 6426ffe75ebSEric Dumazet return; 6436ffe75ebSEric Dumazet fastpath: 6446ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6451da177e4SLinus Torvalds } 6461da177e4SLinus Torvalds 64704a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 6481da177e4SLinus Torvalds { 649adf30907SEric Dumazet skb_dst_drop(skb); 6501da177e4SLinus Torvalds #ifdef CONFIG_XFRM 6511da177e4SLinus Torvalds secpath_put(skb->sp); 6521da177e4SLinus Torvalds #endif 6531da177e4SLinus Torvalds if (skb->destructor) { 6549c2b3328SStephen Hemminger WARN_ON(in_irq()); 6551da177e4SLinus Torvalds skb->destructor(skb); 6561da177e4SLinus Torvalds } 657a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 6585f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 6592fc72c7bSKOVACS Krisztian #endif 6601109a90cSPablo Neira Ayuso #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 6611da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 6621da177e4SLinus Torvalds #endif 66304a4bb55SLennert Buytenhek } 66404a4bb55SLennert Buytenhek 66504a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 66604a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 66704a4bb55SLennert Buytenhek { 66804a4bb55SLennert Buytenhek skb_release_head_state(skb); 6695e71d9d7SPablo Neira if (likely(skb->head)) 6702d4baff8SHerbert Xu skb_release_data(skb); 6712d4baff8SHerbert Xu } 6721da177e4SLinus Torvalds 6732d4baff8SHerbert Xu /** 6742d4baff8SHerbert Xu * __kfree_skb - private function 6752d4baff8SHerbert Xu * @skb: buffer 6762d4baff8SHerbert Xu * 6772d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6782d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6792d4baff8SHerbert Xu * always call kfree_skb 6802d4baff8SHerbert Xu */ 6812d4baff8SHerbert Xu 6822d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6832d4baff8SHerbert Xu { 6842d4baff8SHerbert Xu skb_release_all(skb); 6851da177e4SLinus Torvalds kfree_skbmem(skb); 6861da177e4SLinus Torvalds } 687b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6881da177e4SLinus Torvalds 6891da177e4SLinus Torvalds /** 690231d06aeSJörn Engel * kfree_skb - free an sk_buff 691231d06aeSJörn Engel * @skb: buffer to free 692231d06aeSJörn Engel * 693231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 694231d06aeSJörn Engel * hit zero. 695231d06aeSJörn Engel */ 696231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 697231d06aeSJörn Engel { 698231d06aeSJörn Engel if (unlikely(!skb)) 699231d06aeSJörn Engel return; 700231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 701231d06aeSJörn Engel smp_rmb(); 702231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 703231d06aeSJörn Engel return; 704ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 705231d06aeSJörn Engel __kfree_skb(skb); 706231d06aeSJörn Engel } 707b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 708231d06aeSJörn Engel 709bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 710bd8a7036SEric Dumazet { 711bd8a7036SEric Dumazet while (segs) { 712bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 713bd8a7036SEric Dumazet 714bd8a7036SEric Dumazet kfree_skb(segs); 715bd8a7036SEric Dumazet segs = next; 716bd8a7036SEric Dumazet } 717bd8a7036SEric Dumazet } 718bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 719bd8a7036SEric Dumazet 720d1a203eaSStephen Hemminger /** 72125121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 72225121173SMichael S. Tsirkin * @skb: buffer that triggered an error 72325121173SMichael S. Tsirkin * 72425121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 72525121173SMichael S. Tsirkin * skb must be freed afterwards. 72625121173SMichael S. Tsirkin */ 72725121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 72825121173SMichael S. Tsirkin { 72925121173SMichael S. Tsirkin if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 73025121173SMichael S. Tsirkin struct ubuf_info *uarg; 73125121173SMichael S. Tsirkin 73225121173SMichael S. Tsirkin uarg = skb_shinfo(skb)->destructor_arg; 73325121173SMichael S. Tsirkin if (uarg->callback) 73425121173SMichael S. Tsirkin uarg->callback(uarg, false); 73525121173SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 73625121173SMichael S. Tsirkin } 73725121173SMichael S. Tsirkin } 73825121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 73925121173SMichael S. Tsirkin 74025121173SMichael S. Tsirkin /** 741ead2ceb0SNeil Horman * consume_skb - free an skbuff 742ead2ceb0SNeil Horman * @skb: buffer to free 743ead2ceb0SNeil Horman * 744ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 745ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 746ead2ceb0SNeil Horman * is being dropped after a failure and notes that 747ead2ceb0SNeil Horman */ 748ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 749ead2ceb0SNeil Horman { 750ead2ceb0SNeil Horman if (unlikely(!skb)) 751ead2ceb0SNeil Horman return; 752ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 753ead2ceb0SNeil Horman smp_rmb(); 754ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 755ead2ceb0SNeil Horman return; 75607dc22e7SKoki Sanagi trace_consume_skb(skb); 757ead2ceb0SNeil Horman __kfree_skb(skb); 758ead2ceb0SNeil Horman } 759ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 760ead2ceb0SNeil Horman 761795bb1c0SJesper Dangaard Brouer void __kfree_skb_flush(void) 762795bb1c0SJesper Dangaard Brouer { 763795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 764795bb1c0SJesper Dangaard Brouer 765795bb1c0SJesper Dangaard Brouer /* flush skb_cache if containing objects */ 766795bb1c0SJesper Dangaard Brouer if (nc->skb_count) { 767795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 768795bb1c0SJesper Dangaard Brouer nc->skb_cache); 769795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 770795bb1c0SJesper Dangaard Brouer } 771795bb1c0SJesper Dangaard Brouer } 772795bb1c0SJesper Dangaard Brouer 77315fad714SJesper Dangaard Brouer static inline void _kfree_skb_defer(struct sk_buff *skb) 774795bb1c0SJesper Dangaard Brouer { 775795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 776795bb1c0SJesper Dangaard Brouer 777795bb1c0SJesper Dangaard Brouer /* drop skb->head and call any destructors for packet */ 778795bb1c0SJesper Dangaard Brouer skb_release_all(skb); 779795bb1c0SJesper Dangaard Brouer 780795bb1c0SJesper Dangaard Brouer /* record skb to CPU local list */ 781795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 782795bb1c0SJesper Dangaard Brouer 783795bb1c0SJesper Dangaard Brouer #ifdef CONFIG_SLUB 784795bb1c0SJesper Dangaard Brouer /* SLUB writes into objects when freeing */ 785795bb1c0SJesper Dangaard Brouer prefetchw(skb); 786795bb1c0SJesper Dangaard Brouer #endif 787795bb1c0SJesper Dangaard Brouer 788795bb1c0SJesper Dangaard Brouer /* flush skb_cache if it is filled */ 789795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 790795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 791795bb1c0SJesper Dangaard Brouer nc->skb_cache); 792795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 793795bb1c0SJesper Dangaard Brouer } 794795bb1c0SJesper Dangaard Brouer } 79515fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 79615fad714SJesper Dangaard Brouer { 79715fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 79815fad714SJesper Dangaard Brouer } 799795bb1c0SJesper Dangaard Brouer 800795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 801795bb1c0SJesper Dangaard Brouer { 802795bb1c0SJesper Dangaard Brouer if (unlikely(!skb)) 803795bb1c0SJesper Dangaard Brouer return; 804795bb1c0SJesper Dangaard Brouer 805885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 806795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 807885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 808795bb1c0SJesper Dangaard Brouer return; 809795bb1c0SJesper Dangaard Brouer } 810795bb1c0SJesper Dangaard Brouer 811795bb1c0SJesper Dangaard Brouer if (likely(atomic_read(&skb->users) == 1)) 812795bb1c0SJesper Dangaard Brouer smp_rmb(); 813795bb1c0SJesper Dangaard Brouer else if (likely(!atomic_dec_and_test(&skb->users))) 814795bb1c0SJesper Dangaard Brouer return; 815795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 816795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 817795bb1c0SJesper Dangaard Brouer 818795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 819abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 820795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 821795bb1c0SJesper Dangaard Brouer return; 822795bb1c0SJesper Dangaard Brouer } 823795bb1c0SJesper Dangaard Brouer 82415fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 825795bb1c0SJesper Dangaard Brouer } 826795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 827795bb1c0SJesper Dangaard Brouer 828b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 829b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 830b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 831b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 832b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 833b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 834b1937227SEric Dumazet 835dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 836dec18810SHerbert Xu { 837dec18810SHerbert Xu new->tstamp = old->tstamp; 838b1937227SEric Dumazet /* We do not copy old->sk */ 839dec18810SHerbert Xu new->dev = old->dev; 840b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 8417fee226aSEric Dumazet skb_dst_copy(new, old); 842def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 843dec18810SHerbert Xu new->sp = secpath_get(old->sp); 844dec18810SHerbert Xu #endif 845b1937227SEric Dumazet __nf_copy(new, old, false); 8466aa895b0SPatrick McHardy 847b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 848b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 849b1937227SEric Dumazet */ 850b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 85106021292SEliezer Tamir 852b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 853b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 854b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 855b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 856b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 857b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 858b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 859b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 860b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 861b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 862b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 863b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 864b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 865b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 866b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 867b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 868b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 869b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 870b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 871b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 872b1937227SEric Dumazet #endif 873e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 874b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 87506021292SEliezer Tamir #endif 8762bd82484SEric Dumazet #ifdef CONFIG_XPS 8772bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 8782bd82484SEric Dumazet #endif 879b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 880b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 881b1937227SEric Dumazet #ifdef CONFIG_NET_CLS_ACT 882b1937227SEric Dumazet CHECK_SKB_FIELD(tc_verd); 883b1937227SEric Dumazet #endif 884b1937227SEric Dumazet #endif 885b1937227SEric Dumazet 886dec18810SHerbert Xu } 887dec18810SHerbert Xu 88882c49a35SHerbert Xu /* 88982c49a35SHerbert Xu * You should not add any new code to this function. Add it to 89082c49a35SHerbert Xu * __copy_skb_header above instead. 89182c49a35SHerbert Xu */ 892e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 8931da177e4SLinus Torvalds { 8941da177e4SLinus Torvalds #define C(x) n->x = skb->x 8951da177e4SLinus Torvalds 8961da177e4SLinus Torvalds n->next = n->prev = NULL; 8971da177e4SLinus Torvalds n->sk = NULL; 898dec18810SHerbert Xu __copy_skb_header(n, skb); 899dec18810SHerbert Xu 9001da177e4SLinus Torvalds C(len); 9011da177e4SLinus Torvalds C(data_len); 9023e6b3b2eSAlexey Dobriyan C(mac_len); 903334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 90402f1c89dSPaul Moore n->cloned = 1; 9051da177e4SLinus Torvalds n->nohdr = 0; 9061da177e4SLinus Torvalds n->destructor = NULL; 9071da177e4SLinus Torvalds C(tail); 9081da177e4SLinus Torvalds C(end); 90902f1c89dSPaul Moore C(head); 910d3836f21SEric Dumazet C(head_frag); 91102f1c89dSPaul Moore C(data); 91202f1c89dSPaul Moore C(truesize); 91302f1c89dSPaul Moore atomic_set(&n->users, 1); 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 9161da177e4SLinus Torvalds skb->cloned = 1; 9171da177e4SLinus Torvalds 9181da177e4SLinus Torvalds return n; 919e0053ec0SHerbert Xu #undef C 920e0053ec0SHerbert Xu } 921e0053ec0SHerbert Xu 922e0053ec0SHerbert Xu /** 923e0053ec0SHerbert Xu * skb_morph - morph one skb into another 924e0053ec0SHerbert Xu * @dst: the skb to receive the contents 925e0053ec0SHerbert Xu * @src: the skb to supply the contents 926e0053ec0SHerbert Xu * 927e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 928e0053ec0SHerbert Xu * supplied by the user. 929e0053ec0SHerbert Xu * 930e0053ec0SHerbert Xu * The target skb is returned upon exit. 931e0053ec0SHerbert Xu */ 932e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 933e0053ec0SHerbert Xu { 9342d4baff8SHerbert Xu skb_release_all(dst); 935e0053ec0SHerbert Xu return __skb_clone(dst, src); 936e0053ec0SHerbert Xu } 937e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 938e0053ec0SHerbert Xu 9392c53040fSBen Hutchings /** 9402c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 94148c83012SMichael S. Tsirkin * @skb: the skb to modify 94248c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 94348c83012SMichael S. Tsirkin * 94448c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 94548c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 94648c83012SMichael S. Tsirkin * to userspace pages. 94748c83012SMichael S. Tsirkin * 94848c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 94948c83012SMichael S. Tsirkin * %GFP_ATOMIC. 95048c83012SMichael S. Tsirkin * 95148c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 95248c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 95348c83012SMichael S. Tsirkin */ 95448c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 955a6686f2fSShirley Ma { 956a6686f2fSShirley Ma int i; 957a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 958a6686f2fSShirley Ma struct page *page, *head = NULL; 959a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 960a6686f2fSShirley Ma 961a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 962a6686f2fSShirley Ma u8 *vaddr; 963a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 964a6686f2fSShirley Ma 96502756ed4SKrishna Kumar page = alloc_page(gfp_mask); 966a6686f2fSShirley Ma if (!page) { 967a6686f2fSShirley Ma while (head) { 96840dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 969a6686f2fSShirley Ma put_page(head); 970a6686f2fSShirley Ma head = next; 971a6686f2fSShirley Ma } 972a6686f2fSShirley Ma return -ENOMEM; 973a6686f2fSShirley Ma } 97451c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 975a6686f2fSShirley Ma memcpy(page_address(page), 9769e903e08SEric Dumazet vaddr + f->page_offset, skb_frag_size(f)); 97751c56b00SEric Dumazet kunmap_atomic(vaddr); 97840dadff2SSunghan Suh set_page_private(page, (unsigned long)head); 979a6686f2fSShirley Ma head = page; 980a6686f2fSShirley Ma } 981a6686f2fSShirley Ma 982a6686f2fSShirley Ma /* skb frags release userspace buffers */ 98302756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 984a8605c60SIan Campbell skb_frag_unref(skb, i); 985a6686f2fSShirley Ma 986e19d6763SMichael S. Tsirkin uarg->callback(uarg, false); 987a6686f2fSShirley Ma 988a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 98902756ed4SKrishna Kumar for (i = num_frags - 1; i >= 0; i--) { 99002756ed4SKrishna Kumar __skb_fill_page_desc(skb, i, head, 0, 99102756ed4SKrishna Kumar skb_shinfo(skb)->frags[i].size); 99240dadff2SSunghan Suh head = (struct page *)page_private(head); 993a6686f2fSShirley Ma } 99448c83012SMichael S. Tsirkin 99548c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 996a6686f2fSShirley Ma return 0; 997a6686f2fSShirley Ma } 998dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 999a6686f2fSShirley Ma 1000e0053ec0SHerbert Xu /** 1001e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1002e0053ec0SHerbert Xu * @skb: buffer to clone 1003e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1004e0053ec0SHerbert Xu * 1005e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1006e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1007e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1008e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1009e0053ec0SHerbert Xu * 1010e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1011e0053ec0SHerbert Xu * %GFP_ATOMIC. 1012e0053ec0SHerbert Xu */ 1013e0053ec0SHerbert Xu 1014e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1015e0053ec0SHerbert Xu { 1016d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1017d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1018d0bf4a9eSEric Dumazet skb1); 10196ffe75ebSEric Dumazet struct sk_buff *n; 1020e0053ec0SHerbert Xu 102170008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1022a6686f2fSShirley Ma return NULL; 1023a6686f2fSShirley Ma 1024e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 10256ffe75ebSEric Dumazet atomic_read(&fclones->fclone_ref) == 1) { 10266ffe75ebSEric Dumazet n = &fclones->skb2; 10276ffe75ebSEric Dumazet atomic_set(&fclones->fclone_ref, 2); 1028e0053ec0SHerbert Xu } else { 1029c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1030c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1031c93bdd0eSMel Gorman 1032e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1033e0053ec0SHerbert Xu if (!n) 1034e0053ec0SHerbert Xu return NULL; 1035fe55f6d5SVegard Nossum 1036fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 1037e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1038e0053ec0SHerbert Xu } 1039e0053ec0SHerbert Xu 1040e0053ec0SHerbert Xu return __skb_clone(n, skb); 10411da177e4SLinus Torvalds } 1042b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 10431da177e4SLinus Torvalds 1044f5b17294SPravin B Shelar static void skb_headers_offset_update(struct sk_buff *skb, int off) 1045f5b17294SPravin B Shelar { 1046030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1047030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1048030737bcSEric Dumazet skb->csum_start += off; 1049f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1050f5b17294SPravin B Shelar skb->transport_header += off; 1051f5b17294SPravin B Shelar skb->network_header += off; 1052f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1053f5b17294SPravin B Shelar skb->mac_header += off; 1054f5b17294SPravin B Shelar skb->inner_transport_header += off; 1055f5b17294SPravin B Shelar skb->inner_network_header += off; 1056aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1057f5b17294SPravin B Shelar } 1058f5b17294SPravin B Shelar 10591da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 10601da177e4SLinus Torvalds { 1061dec18810SHerbert Xu __copy_skb_header(new, old); 1062dec18810SHerbert Xu 10637967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 10647967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 10657967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 10661da177e4SLinus Torvalds } 10671da177e4SLinus Torvalds 1068c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1069c93bdd0eSMel Gorman { 1070c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1071c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1072c93bdd0eSMel Gorman return 0; 1073c93bdd0eSMel Gorman } 1074c93bdd0eSMel Gorman 10751da177e4SLinus Torvalds /** 10761da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 10771da177e4SLinus Torvalds * @skb: buffer to copy 10781da177e4SLinus Torvalds * @gfp_mask: allocation priority 10791da177e4SLinus Torvalds * 10801da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 10811da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 10821da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 10831da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 10841da177e4SLinus Torvalds * 10851da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 10861da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 10871da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 10881da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 10891da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 10901da177e4SLinus Torvalds */ 10911da177e4SLinus Torvalds 1092dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 10931da177e4SLinus Torvalds { 10946602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1095ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1096c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1097c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 10986602cebbSEric Dumazet 10991da177e4SLinus Torvalds if (!n) 11001da177e4SLinus Torvalds return NULL; 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds /* Set the data pointer */ 11031da177e4SLinus Torvalds skb_reserve(n, headerlen); 11041da177e4SLinus Torvalds /* Set the tail pointer and length */ 11051da177e4SLinus Torvalds skb_put(n, skb->len); 11061da177e4SLinus Torvalds 11071da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 11081da177e4SLinus Torvalds BUG(); 11091da177e4SLinus Torvalds 11101da177e4SLinus Torvalds copy_skb_header(n, skb); 11111da177e4SLinus Torvalds return n; 11121da177e4SLinus Torvalds } 1113b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 11141da177e4SLinus Torvalds 11151da177e4SLinus Torvalds /** 1116bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 11171da177e4SLinus Torvalds * @skb: buffer to copy 1118117632e6SEric Dumazet * @headroom: headroom of new skb 11191da177e4SLinus Torvalds * @gfp_mask: allocation priority 1120bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1121bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1122bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 11231da177e4SLinus Torvalds * 11241da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 11251da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 11261da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 11271da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 11281da177e4SLinus Torvalds * or the pointer to the buffer on success. 11291da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 11301da177e4SLinus Torvalds */ 11311da177e4SLinus Torvalds 1132bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1133bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 11341da177e4SLinus Torvalds { 1135117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1136bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1137bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 11386602cebbSEric Dumazet 11391da177e4SLinus Torvalds if (!n) 11401da177e4SLinus Torvalds goto out; 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds /* Set the data pointer */ 1143117632e6SEric Dumazet skb_reserve(n, headroom); 11441da177e4SLinus Torvalds /* Set the tail pointer and length */ 11451da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 11461da177e4SLinus Torvalds /* Copy the bytes */ 1147d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 11481da177e4SLinus Torvalds 114925f484a6SHerbert Xu n->truesize += skb->data_len; 11501da177e4SLinus Torvalds n->data_len = skb->data_len; 11511da177e4SLinus Torvalds n->len = skb->len; 11521da177e4SLinus Torvalds 11531da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 11541da177e4SLinus Torvalds int i; 11551da177e4SLinus Torvalds 115670008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) { 11571511022cSDan Carpenter kfree_skb(n); 11581511022cSDan Carpenter n = NULL; 1159a6686f2fSShirley Ma goto out; 1160a6686f2fSShirley Ma } 11611da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 11621da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1163ea2ab693SIan Campbell skb_frag_ref(skb, i); 11641da177e4SLinus Torvalds } 11651da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds 116821dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 11691da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 11701da177e4SLinus Torvalds skb_clone_fraglist(n); 11711da177e4SLinus Torvalds } 11721da177e4SLinus Torvalds 11731da177e4SLinus Torvalds copy_skb_header(n, skb); 11741da177e4SLinus Torvalds out: 11751da177e4SLinus Torvalds return n; 11761da177e4SLinus Torvalds } 1177bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 11781da177e4SLinus Torvalds 11791da177e4SLinus Torvalds /** 11801da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 11811da177e4SLinus Torvalds * @skb: buffer to reallocate 11821da177e4SLinus Torvalds * @nhead: room to add at head 11831da177e4SLinus Torvalds * @ntail: room to add at tail 11841da177e4SLinus Torvalds * @gfp_mask: allocation priority 11851da177e4SLinus Torvalds * 1186bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1187bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 11881da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 11891da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 11901da177e4SLinus Torvalds * 11911da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 11921da177e4SLinus Torvalds * reloaded after call to this function. 11931da177e4SLinus Torvalds */ 11941da177e4SLinus Torvalds 119586a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1196dd0fc66fSAl Viro gfp_t gfp_mask) 11971da177e4SLinus Torvalds { 11981da177e4SLinus Torvalds int i; 11991da177e4SLinus Torvalds u8 *data; 1200ec47ea82SAlexander Duyck int size = nhead + skb_end_offset(skb) + ntail; 12011da177e4SLinus Torvalds long off; 12021da177e4SLinus Torvalds 12034edd87adSHerbert Xu BUG_ON(nhead < 0); 12044edd87adSHerbert Xu 12051da177e4SLinus Torvalds if (skb_shared(skb)) 12061da177e4SLinus Torvalds BUG(); 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 12091da177e4SLinus Torvalds 1210c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1211c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1212c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1213c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 12141da177e4SLinus Torvalds if (!data) 12151da177e4SLinus Torvalds goto nodata; 121687151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 12196602cebbSEric Dumazet * optimized for the cases when header is void. 12206602cebbSEric Dumazet */ 12216602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 12226602cebbSEric Dumazet 12236602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 12246602cebbSEric Dumazet skb_shinfo(skb), 1225fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 12261da177e4SLinus Torvalds 12273e24591aSAlexander Duyck /* 12283e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 12293e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 12303e24591aSAlexander Duyck * be since all we did is relocate the values 12313e24591aSAlexander Duyck */ 12323e24591aSAlexander Duyck if (skb_cloned(skb)) { 1233a6686f2fSShirley Ma /* copy this zero copy skb frags */ 123470008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1235a6686f2fSShirley Ma goto nofrags; 12361da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1237ea2ab693SIan Campbell skb_frag_ref(skb, i); 12381da177e4SLinus Torvalds 123921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 12401da177e4SLinus Torvalds skb_clone_fraglist(skb); 12411da177e4SLinus Torvalds 12421da177e4SLinus Torvalds skb_release_data(skb); 12433e24591aSAlexander Duyck } else { 12443e24591aSAlexander Duyck skb_free_head(skb); 12451fd63041SEric Dumazet } 12461da177e4SLinus Torvalds off = (data + nhead) - skb->head; 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds skb->head = data; 1249d3836f21SEric Dumazet skb->head_frag = 0; 12501da177e4SLinus Torvalds skb->data += off; 12514305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 12524305b541SArnaldo Carvalho de Melo skb->end = size; 125356eb8882SPatrick McHardy off = nhead; 12544305b541SArnaldo Carvalho de Melo #else 12554305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 125656eb8882SPatrick McHardy #endif 125727a884dcSArnaldo Carvalho de Melo skb->tail += off; 1258b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 12591da177e4SLinus Torvalds skb->cloned = 0; 1260334a8132SPatrick McHardy skb->hdr_len = 0; 12611da177e4SLinus Torvalds skb->nohdr = 0; 12621da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 12631da177e4SLinus Torvalds return 0; 12641da177e4SLinus Torvalds 1265a6686f2fSShirley Ma nofrags: 1266a6686f2fSShirley Ma kfree(data); 12671da177e4SLinus Torvalds nodata: 12681da177e4SLinus Torvalds return -ENOMEM; 12691da177e4SLinus Torvalds } 1270b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 12731da177e4SLinus Torvalds 12741da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 12751da177e4SLinus Torvalds { 12761da177e4SLinus Torvalds struct sk_buff *skb2; 12771da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds if (delta <= 0) 12801da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 12811da177e4SLinus Torvalds else { 12821da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 12831da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 12841da177e4SLinus Torvalds GFP_ATOMIC)) { 12851da177e4SLinus Torvalds kfree_skb(skb2); 12861da177e4SLinus Torvalds skb2 = NULL; 12871da177e4SLinus Torvalds } 12881da177e4SLinus Torvalds } 12891da177e4SLinus Torvalds return skb2; 12901da177e4SLinus Torvalds } 1291b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds /** 12941da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 12951da177e4SLinus Torvalds * @skb: buffer to copy 12961da177e4SLinus Torvalds * @newheadroom: new free bytes at head 12971da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 12981da177e4SLinus Torvalds * @gfp_mask: allocation priority 12991da177e4SLinus Torvalds * 13001da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 13011da177e4SLinus Torvalds * allocate additional space. 13021da177e4SLinus Torvalds * 13031da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 13041da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 13051da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 13061da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 13071da177e4SLinus Torvalds * 13081da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 13091da177e4SLinus Torvalds * is called from an interrupt. 13101da177e4SLinus Torvalds */ 13111da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 131286a76cafSVictor Fusco int newheadroom, int newtailroom, 1313dd0fc66fSAl Viro gfp_t gfp_mask) 13141da177e4SLinus Torvalds { 13151da177e4SLinus Torvalds /* 13161da177e4SLinus Torvalds * Allocate the copy buffer 13171da177e4SLinus Torvalds */ 1318c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1319c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1320c93bdd0eSMel Gorman NUMA_NO_NODE); 1321efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 13221da177e4SLinus Torvalds int head_copy_len, head_copy_off; 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds if (!n) 13251da177e4SLinus Torvalds return NULL; 13261da177e4SLinus Torvalds 13271da177e4SLinus Torvalds skb_reserve(n, newheadroom); 13281da177e4SLinus Torvalds 13291da177e4SLinus Torvalds /* Set the tail pointer and length */ 13301da177e4SLinus Torvalds skb_put(n, skb->len); 13311da177e4SLinus Torvalds 1332efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 13331da177e4SLinus Torvalds head_copy_off = 0; 13341da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 13351da177e4SLinus Torvalds head_copy_len = newheadroom; 13361da177e4SLinus Torvalds else 13371da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds /* Copy the linear header and data. */ 13401da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 13411da177e4SLinus Torvalds skb->len + head_copy_len)) 13421da177e4SLinus Torvalds BUG(); 13431da177e4SLinus Torvalds 13441da177e4SLinus Torvalds copy_skb_header(n, skb); 13451da177e4SLinus Torvalds 1346030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1347efd1e8d5SPatrick McHardy 13481da177e4SLinus Torvalds return n; 13491da177e4SLinus Torvalds } 1350b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds /** 13531da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 13541da177e4SLinus Torvalds * @skb: buffer to pad 13551da177e4SLinus Torvalds * @pad: space to pad 13561da177e4SLinus Torvalds * 13571da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 13581da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 13591da177e4SLinus Torvalds * beyond the buffer end onto the wire. 13601da177e4SLinus Torvalds * 13615b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 13621da177e4SLinus Torvalds */ 13631da177e4SLinus Torvalds 13645b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 13651da177e4SLinus Torvalds { 13665b057c6bSHerbert Xu int err; 13675b057c6bSHerbert Xu int ntail; 13681da177e4SLinus Torvalds 13691da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 13705b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 13711da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 13725b057c6bSHerbert Xu return 0; 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds 13754305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 13765b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 13775b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 13785b057c6bSHerbert Xu if (unlikely(err)) 13795b057c6bSHerbert Xu goto free_skb; 13805b057c6bSHerbert Xu } 13815b057c6bSHerbert Xu 13825b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 13835b057c6bSHerbert Xu * to be audited. 13845b057c6bSHerbert Xu */ 13855b057c6bSHerbert Xu err = skb_linearize(skb); 13865b057c6bSHerbert Xu if (unlikely(err)) 13875b057c6bSHerbert Xu goto free_skb; 13885b057c6bSHerbert Xu 13895b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 13905b057c6bSHerbert Xu return 0; 13915b057c6bSHerbert Xu 13925b057c6bSHerbert Xu free_skb: 13931da177e4SLinus Torvalds kfree_skb(skb); 13945b057c6bSHerbert Xu return err; 13951da177e4SLinus Torvalds } 1396b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 13971da177e4SLinus Torvalds 13980dde3e16SIlpo Järvinen /** 13990c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 14000c7ddf36SMathias Krause * @skb: start of the buffer to use 14010c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 14020c7ddf36SMathias Krause * @len: amount of data to add 14030c7ddf36SMathias Krause * 14040c7ddf36SMathias Krause * This function extends the used data area of the potentially 14050c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 14060c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 14070c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 14080c7ddf36SMathias Krause * returned. 14090c7ddf36SMathias Krause */ 14100c7ddf36SMathias Krause 14110c7ddf36SMathias Krause unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 14120c7ddf36SMathias Krause { 14130c7ddf36SMathias Krause if (tail != skb) { 14140c7ddf36SMathias Krause skb->data_len += len; 14150c7ddf36SMathias Krause skb->len += len; 14160c7ddf36SMathias Krause } 14170c7ddf36SMathias Krause return skb_put(tail, len); 14180c7ddf36SMathias Krause } 14190c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 14200c7ddf36SMathias Krause 14210c7ddf36SMathias Krause /** 14220dde3e16SIlpo Järvinen * skb_put - add data to a buffer 14230dde3e16SIlpo Järvinen * @skb: buffer to use 14240dde3e16SIlpo Järvinen * @len: amount of data to add 14250dde3e16SIlpo Järvinen * 14260dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 14270dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 14280dde3e16SIlpo Järvinen * first byte of the extra data is returned. 14290dde3e16SIlpo Järvinen */ 14300dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 14310dde3e16SIlpo Järvinen { 14320dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 14330dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 14340dde3e16SIlpo Järvinen skb->tail += len; 14350dde3e16SIlpo Järvinen skb->len += len; 14360dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 14370dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 14380dde3e16SIlpo Järvinen return tmp; 14390dde3e16SIlpo Järvinen } 14400dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 14410dde3e16SIlpo Järvinen 14426be8ac2fSIlpo Järvinen /** 1443c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1444c2aa270aSIlpo Järvinen * @skb: buffer to use 1445c2aa270aSIlpo Järvinen * @len: amount of data to add 1446c2aa270aSIlpo Järvinen * 1447c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1448c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1449c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1450c2aa270aSIlpo Järvinen */ 1451c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1452c2aa270aSIlpo Järvinen { 1453c2aa270aSIlpo Järvinen skb->data -= len; 1454c2aa270aSIlpo Järvinen skb->len += len; 1455c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1456c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1457c2aa270aSIlpo Järvinen return skb->data; 1458c2aa270aSIlpo Järvinen } 1459c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1460c2aa270aSIlpo Järvinen 1461c2aa270aSIlpo Järvinen /** 14626be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 14636be8ac2fSIlpo Järvinen * @skb: buffer to use 14646be8ac2fSIlpo Järvinen * @len: amount of data to remove 14656be8ac2fSIlpo Järvinen * 14666be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 14676be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 14686be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 14696be8ac2fSIlpo Järvinen * the old data. 14706be8ac2fSIlpo Järvinen */ 14716be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 14726be8ac2fSIlpo Järvinen { 147347d29646SDavid S. Miller return skb_pull_inline(skb, len); 14746be8ac2fSIlpo Järvinen } 14756be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 14766be8ac2fSIlpo Järvinen 1477419ae74eSIlpo Järvinen /** 1478419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1479419ae74eSIlpo Järvinen * @skb: buffer to alter 1480419ae74eSIlpo Järvinen * @len: new length 1481419ae74eSIlpo Järvinen * 1482419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1483419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1484419ae74eSIlpo Järvinen * The skb must be linear. 1485419ae74eSIlpo Järvinen */ 1486419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1487419ae74eSIlpo Järvinen { 1488419ae74eSIlpo Järvinen if (skb->len > len) 1489419ae74eSIlpo Järvinen __skb_trim(skb, len); 1490419ae74eSIlpo Järvinen } 1491419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1492419ae74eSIlpo Järvinen 14933cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 14941da177e4SLinus Torvalds */ 14951da177e4SLinus Torvalds 14963cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 14971da177e4SLinus Torvalds { 149827b437c8SHerbert Xu struct sk_buff **fragp; 149927b437c8SHerbert Xu struct sk_buff *frag; 15001da177e4SLinus Torvalds int offset = skb_headlen(skb); 15011da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 15021da177e4SLinus Torvalds int i; 150327b437c8SHerbert Xu int err; 150427b437c8SHerbert Xu 150527b437c8SHerbert Xu if (skb_cloned(skb) && 150627b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 150727b437c8SHerbert Xu return err; 15081da177e4SLinus Torvalds 1509f4d26fb3SHerbert Xu i = 0; 1510f4d26fb3SHerbert Xu if (offset >= len) 1511f4d26fb3SHerbert Xu goto drop_pages; 1512f4d26fb3SHerbert Xu 1513f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 15149e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 151527b437c8SHerbert Xu 151627b437c8SHerbert Xu if (end < len) { 15171da177e4SLinus Torvalds offset = end; 151827b437c8SHerbert Xu continue; 15191da177e4SLinus Torvalds } 15201da177e4SLinus Torvalds 15219e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 152227b437c8SHerbert Xu 1523f4d26fb3SHerbert Xu drop_pages: 152427b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 152527b437c8SHerbert Xu 152627b437c8SHerbert Xu for (; i < nfrags; i++) 1527ea2ab693SIan Campbell skb_frag_unref(skb, i); 152827b437c8SHerbert Xu 152921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 153027b437c8SHerbert Xu skb_drop_fraglist(skb); 1531f4d26fb3SHerbert Xu goto done; 153227b437c8SHerbert Xu } 153327b437c8SHerbert Xu 153427b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 153527b437c8SHerbert Xu fragp = &frag->next) { 153627b437c8SHerbert Xu int end = offset + frag->len; 153727b437c8SHerbert Xu 153827b437c8SHerbert Xu if (skb_shared(frag)) { 153927b437c8SHerbert Xu struct sk_buff *nfrag; 154027b437c8SHerbert Xu 154127b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 154227b437c8SHerbert Xu if (unlikely(!nfrag)) 154327b437c8SHerbert Xu return -ENOMEM; 154427b437c8SHerbert Xu 154527b437c8SHerbert Xu nfrag->next = frag->next; 154685bb2a60SEric Dumazet consume_skb(frag); 154727b437c8SHerbert Xu frag = nfrag; 154827b437c8SHerbert Xu *fragp = frag; 154927b437c8SHerbert Xu } 155027b437c8SHerbert Xu 155127b437c8SHerbert Xu if (end < len) { 155227b437c8SHerbert Xu offset = end; 155327b437c8SHerbert Xu continue; 155427b437c8SHerbert Xu } 155527b437c8SHerbert Xu 155627b437c8SHerbert Xu if (end > len && 155727b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 155827b437c8SHerbert Xu return err; 155927b437c8SHerbert Xu 156027b437c8SHerbert Xu if (frag->next) 156127b437c8SHerbert Xu skb_drop_list(&frag->next); 156227b437c8SHerbert Xu break; 156327b437c8SHerbert Xu } 156427b437c8SHerbert Xu 1565f4d26fb3SHerbert Xu done: 156627b437c8SHerbert Xu if (len > skb_headlen(skb)) { 15671da177e4SLinus Torvalds skb->data_len -= skb->len - len; 15681da177e4SLinus Torvalds skb->len = len; 15691da177e4SLinus Torvalds } else { 15701da177e4SLinus Torvalds skb->len = len; 15711da177e4SLinus Torvalds skb->data_len = 0; 157227a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 15731da177e4SLinus Torvalds } 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds return 0; 15761da177e4SLinus Torvalds } 1577b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds /** 15801da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 15811da177e4SLinus Torvalds * @skb: buffer to reallocate 15821da177e4SLinus Torvalds * @delta: number of bytes to advance tail 15831da177e4SLinus Torvalds * 15841da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 15851da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 15861da177e4SLinus Torvalds * data from fragmented part. 15871da177e4SLinus Torvalds * 15881da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 15891da177e4SLinus Torvalds * 15901da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 15911da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 15921da177e4SLinus Torvalds * 15931da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 15941da177e4SLinus Torvalds * reloaded after call to this function. 15951da177e4SLinus Torvalds */ 15961da177e4SLinus Torvalds 15971da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 15981da177e4SLinus Torvalds * when it is necessary. 15991da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 16001da177e4SLinus Torvalds * 2. It may change skb pointers. 16011da177e4SLinus Torvalds * 16021da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 16031da177e4SLinus Torvalds */ 16041da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 16051da177e4SLinus Torvalds { 16061da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 16071da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 16081da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 16091da177e4SLinus Torvalds */ 16104305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 16111da177e4SLinus Torvalds 16121da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 16131da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 16141da177e4SLinus Torvalds GFP_ATOMIC)) 16151da177e4SLinus Torvalds return NULL; 16161da177e4SLinus Torvalds } 16171da177e4SLinus Torvalds 161827a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 16191da177e4SLinus Torvalds BUG(); 16201da177e4SLinus Torvalds 16211da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 16221da177e4SLinus Torvalds * size of pulled pages. Superb. 16231da177e4SLinus Torvalds */ 162421dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 16251da177e4SLinus Torvalds goto pull_pages; 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 16281da177e4SLinus Torvalds eat = delta; 16291da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16309e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 16319e903e08SEric Dumazet 16329e903e08SEric Dumazet if (size >= eat) 16331da177e4SLinus Torvalds goto pull_pages; 16349e903e08SEric Dumazet eat -= size; 16351da177e4SLinus Torvalds } 16361da177e4SLinus Torvalds 16371da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 16381da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 16391da177e4SLinus Torvalds * but taking into account that pulling is expected to 16401da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 16411da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 16421da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 16431da177e4SLinus Torvalds */ 16441da177e4SLinus Torvalds if (eat) { 16451da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 16461da177e4SLinus Torvalds struct sk_buff *clone = NULL; 16471da177e4SLinus Torvalds struct sk_buff *insp = NULL; 16481da177e4SLinus Torvalds 16491da177e4SLinus Torvalds do { 165009a62660SKris Katterjohn BUG_ON(!list); 16511da177e4SLinus Torvalds 16521da177e4SLinus Torvalds if (list->len <= eat) { 16531da177e4SLinus Torvalds /* Eaten as whole. */ 16541da177e4SLinus Torvalds eat -= list->len; 16551da177e4SLinus Torvalds list = list->next; 16561da177e4SLinus Torvalds insp = list; 16571da177e4SLinus Torvalds } else { 16581da177e4SLinus Torvalds /* Eaten partially. */ 16591da177e4SLinus Torvalds 16601da177e4SLinus Torvalds if (skb_shared(list)) { 16611da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 16621da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 16631da177e4SLinus Torvalds if (!clone) 16641da177e4SLinus Torvalds return NULL; 16651da177e4SLinus Torvalds insp = list->next; 16661da177e4SLinus Torvalds list = clone; 16671da177e4SLinus Torvalds } else { 16681da177e4SLinus Torvalds /* This may be pulled without 16691da177e4SLinus Torvalds * problems. */ 16701da177e4SLinus Torvalds insp = list; 16711da177e4SLinus Torvalds } 16721da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 16731da177e4SLinus Torvalds kfree_skb(clone); 16741da177e4SLinus Torvalds return NULL; 16751da177e4SLinus Torvalds } 16761da177e4SLinus Torvalds break; 16771da177e4SLinus Torvalds } 16781da177e4SLinus Torvalds } while (eat); 16791da177e4SLinus Torvalds 16801da177e4SLinus Torvalds /* Free pulled out fragments. */ 16811da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 16821da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 16831da177e4SLinus Torvalds kfree_skb(list); 16841da177e4SLinus Torvalds } 16851da177e4SLinus Torvalds /* And insert new clone at head. */ 16861da177e4SLinus Torvalds if (clone) { 16871da177e4SLinus Torvalds clone->next = list; 16881da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 16891da177e4SLinus Torvalds } 16901da177e4SLinus Torvalds } 16911da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 16921da177e4SLinus Torvalds 16931da177e4SLinus Torvalds pull_pages: 16941da177e4SLinus Torvalds eat = delta; 16951da177e4SLinus Torvalds k = 0; 16961da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16979e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 16989e903e08SEric Dumazet 16999e903e08SEric Dumazet if (size <= eat) { 1700ea2ab693SIan Campbell skb_frag_unref(skb, i); 17019e903e08SEric Dumazet eat -= size; 17021da177e4SLinus Torvalds } else { 17031da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 17041da177e4SLinus Torvalds if (eat) { 17051da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 17069e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 17071da177e4SLinus Torvalds eat = 0; 17081da177e4SLinus Torvalds } 17091da177e4SLinus Torvalds k++; 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds } 17121da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 17131da177e4SLinus Torvalds 17141da177e4SLinus Torvalds skb->tail += delta; 17151da177e4SLinus Torvalds skb->data_len -= delta; 17161da177e4SLinus Torvalds 171727a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 17181da177e4SLinus Torvalds } 1719b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 17201da177e4SLinus Torvalds 172122019b17SEric Dumazet /** 172222019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 172322019b17SEric Dumazet * @skb: source skb 172422019b17SEric Dumazet * @offset: offset in source 172522019b17SEric Dumazet * @to: destination buffer 172622019b17SEric Dumazet * @len: number of bytes to copy 172722019b17SEric Dumazet * 172822019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 172922019b17SEric Dumazet * destination buffer. 173022019b17SEric Dumazet * 173122019b17SEric Dumazet * CAUTION ! : 173222019b17SEric Dumazet * If its prototype is ever changed, 173322019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 173422019b17SEric Dumazet * since it is called from BPF assembly code. 173522019b17SEric Dumazet */ 17361da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 17371da177e4SLinus Torvalds { 17381a028e50SDavid S. Miller int start = skb_headlen(skb); 1739fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1740fbb398a8SDavid S. Miller int i, copy; 17411da177e4SLinus Torvalds 17421da177e4SLinus Torvalds if (offset > (int)skb->len - len) 17431da177e4SLinus Torvalds goto fault; 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* Copy header. */ 17461a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 17471da177e4SLinus Torvalds if (copy > len) 17481da177e4SLinus Torvalds copy = len; 1749d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 17501da177e4SLinus Torvalds if ((len -= copy) == 0) 17511da177e4SLinus Torvalds return 0; 17521da177e4SLinus Torvalds offset += copy; 17531da177e4SLinus Torvalds to += copy; 17541da177e4SLinus Torvalds } 17551da177e4SLinus Torvalds 17561da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 17571a028e50SDavid S. Miller int end; 175851c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 17591da177e4SLinus Torvalds 1760547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17611a028e50SDavid S. Miller 176251c56b00SEric Dumazet end = start + skb_frag_size(f); 17631da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17641da177e4SLinus Torvalds u8 *vaddr; 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds if (copy > len) 17671da177e4SLinus Torvalds copy = len; 17681da177e4SLinus Torvalds 176951c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 17701da177e4SLinus Torvalds memcpy(to, 177151c56b00SEric Dumazet vaddr + f->page_offset + offset - start, 177251c56b00SEric Dumazet copy); 177351c56b00SEric Dumazet kunmap_atomic(vaddr); 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds if ((len -= copy) == 0) 17761da177e4SLinus Torvalds return 0; 17771da177e4SLinus Torvalds offset += copy; 17781da177e4SLinus Torvalds to += copy; 17791da177e4SLinus Torvalds } 17801a028e50SDavid S. Miller start = end; 17811da177e4SLinus Torvalds } 17821da177e4SLinus Torvalds 1783fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 17841a028e50SDavid S. Miller int end; 17851da177e4SLinus Torvalds 1786547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17871a028e50SDavid S. Miller 1788fbb398a8SDavid S. Miller end = start + frag_iter->len; 17891da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17901da177e4SLinus Torvalds if (copy > len) 17911da177e4SLinus Torvalds copy = len; 1792fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 17931da177e4SLinus Torvalds goto fault; 17941da177e4SLinus Torvalds if ((len -= copy) == 0) 17951da177e4SLinus Torvalds return 0; 17961da177e4SLinus Torvalds offset += copy; 17971da177e4SLinus Torvalds to += copy; 17981da177e4SLinus Torvalds } 17991a028e50SDavid S. Miller start = end; 18001da177e4SLinus Torvalds } 1801a6686f2fSShirley Ma 18021da177e4SLinus Torvalds if (!len) 18031da177e4SLinus Torvalds return 0; 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds fault: 18061da177e4SLinus Torvalds return -EFAULT; 18071da177e4SLinus Torvalds } 1808b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 18091da177e4SLinus Torvalds 18109c55e01cSJens Axboe /* 18119c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 18129c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 18139c55e01cSJens Axboe */ 18149c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 18159c55e01cSJens Axboe { 18168b9d3728SJarek Poplawski put_page(spd->pages[i]); 18178b9d3728SJarek Poplawski } 18189c55e01cSJens Axboe 1819a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 18204fb66994SJarek Poplawski unsigned int *offset, 182118aafc62SEric Dumazet struct sock *sk) 18228b9d3728SJarek Poplawski { 18235640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 18248b9d3728SJarek Poplawski 18255640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 18268b9d3728SJarek Poplawski return NULL; 18274fb66994SJarek Poplawski 18285640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 18294fb66994SJarek Poplawski 18305640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 18315640f768SEric Dumazet page_address(page) + *offset, *len); 18325640f768SEric Dumazet *offset = pfrag->offset; 18335640f768SEric Dumazet pfrag->offset += *len; 18344fb66994SJarek Poplawski 18355640f768SEric Dumazet return pfrag->page; 18369c55e01cSJens Axboe } 18379c55e01cSJens Axboe 183841c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 183941c73a0dSEric Dumazet struct page *page, 184041c73a0dSEric Dumazet unsigned int offset) 184141c73a0dSEric Dumazet { 184241c73a0dSEric Dumazet return spd->nr_pages && 184341c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 184441c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 184541c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 184641c73a0dSEric Dumazet } 184741c73a0dSEric Dumazet 18489c55e01cSJens Axboe /* 18499c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 18509c55e01cSJens Axboe */ 1851a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 185235f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 18534fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 185418aafc62SEric Dumazet bool linear, 18557a67e56fSJarek Poplawski struct sock *sk) 18569c55e01cSJens Axboe { 185741c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1858a108d5f3SDavid S. Miller return true; 18599c55e01cSJens Axboe 18608b9d3728SJarek Poplawski if (linear) { 186118aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 18628b9d3728SJarek Poplawski if (!page) 1863a108d5f3SDavid S. Miller return true; 186441c73a0dSEric Dumazet } 186541c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 186641c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 1867a108d5f3SDavid S. Miller return false; 186841c73a0dSEric Dumazet } 18698b9d3728SJarek Poplawski get_page(page); 18709c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 18714fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 18729c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 18739c55e01cSJens Axboe spd->nr_pages++; 18748b9d3728SJarek Poplawski 1875a108d5f3SDavid S. Miller return false; 18769c55e01cSJens Axboe } 18779c55e01cSJens Axboe 1878a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 18792870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 188018aafc62SEric Dumazet unsigned int *len, 1881d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 188235f3d14dSJens Axboe struct sock *sk, 188335f3d14dSJens Axboe struct pipe_inode_info *pipe) 18849c55e01cSJens Axboe { 18852870c43dSOctavian Purdila if (!*len) 1886a108d5f3SDavid S. Miller return true; 18879c55e01cSJens Axboe 18882870c43dSOctavian Purdila /* skip this segment if already processed */ 18892870c43dSOctavian Purdila if (*off >= plen) { 18902870c43dSOctavian Purdila *off -= plen; 1891a108d5f3SDavid S. Miller return false; 18922870c43dSOctavian Purdila } 18932870c43dSOctavian Purdila 18942870c43dSOctavian Purdila /* ignore any bits we already processed */ 18959ca1b22dSEric Dumazet poff += *off; 18969ca1b22dSEric Dumazet plen -= *off; 18972870c43dSOctavian Purdila *off = 0; 18982870c43dSOctavian Purdila 189918aafc62SEric Dumazet do { 190018aafc62SEric Dumazet unsigned int flen = min(*len, plen); 19012870c43dSOctavian Purdila 190218aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 190318aafc62SEric Dumazet linear, sk)) 1904a108d5f3SDavid S. Miller return true; 190518aafc62SEric Dumazet poff += flen; 190618aafc62SEric Dumazet plen -= flen; 19072870c43dSOctavian Purdila *len -= flen; 190818aafc62SEric Dumazet } while (*len && plen); 19092870c43dSOctavian Purdila 1910a108d5f3SDavid S. Miller return false; 1911db43a282SOctavian Purdila } 19129c55e01cSJens Axboe 19139c55e01cSJens Axboe /* 1914a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 19152870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 19169c55e01cSJens Axboe */ 1917a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 191835f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 191935f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 19202870c43dSOctavian Purdila { 19212870c43dSOctavian Purdila int seg; 1922fa9835e5STom Herbert struct sk_buff *iter; 19239c55e01cSJens Axboe 19241d0c0b32SEric Dumazet /* map the linear part : 19252996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 19262996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 19272996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 19289c55e01cSJens Axboe */ 19292870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 19302870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 19312870c43dSOctavian Purdila skb_headlen(skb), 193218aafc62SEric Dumazet offset, len, spd, 19333a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 19341d0c0b32SEric Dumazet sk, pipe)) 1935a108d5f3SDavid S. Miller return true; 19369c55e01cSJens Axboe 19379c55e01cSJens Axboe /* 19389c55e01cSJens Axboe * then map the fragments 19399c55e01cSJens Axboe */ 19409c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 19419c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 19429c55e01cSJens Axboe 1943ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 19449e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 194518aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 1946a108d5f3SDavid S. Miller return true; 19479c55e01cSJens Axboe } 19489c55e01cSJens Axboe 1949fa9835e5STom Herbert skb_walk_frags(skb, iter) { 1950fa9835e5STom Herbert if (*offset >= iter->len) { 1951fa9835e5STom Herbert *offset -= iter->len; 1952fa9835e5STom Herbert continue; 1953fa9835e5STom Herbert } 1954fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 1955fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 1956fa9835e5STom Herbert * case. 1957fa9835e5STom Herbert */ 1958fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 1959fa9835e5STom Herbert return true; 1960fa9835e5STom Herbert } 1961fa9835e5STom Herbert 1962a108d5f3SDavid S. Miller return false; 19639c55e01cSJens Axboe } 19649c55e01cSJens Axboe 1965a60e3cc7SHannes Frederic Sowa ssize_t skb_socket_splice(struct sock *sk, 1966a60e3cc7SHannes Frederic Sowa struct pipe_inode_info *pipe, 1967a60e3cc7SHannes Frederic Sowa struct splice_pipe_desc *spd) 1968a60e3cc7SHannes Frederic Sowa { 1969a60e3cc7SHannes Frederic Sowa int ret; 1970a60e3cc7SHannes Frederic Sowa 1971a60e3cc7SHannes Frederic Sowa /* Drop the socket lock, otherwise we have reverse 1972a60e3cc7SHannes Frederic Sowa * locking dependencies between sk_lock and i_mutex 1973a60e3cc7SHannes Frederic Sowa * here as compared to sendfile(). We enter here 1974a60e3cc7SHannes Frederic Sowa * with the socket lock held, and splice_to_pipe() will 1975a60e3cc7SHannes Frederic Sowa * grab the pipe inode lock. For sendfile() emulation, 1976a60e3cc7SHannes Frederic Sowa * we call into ->sendpage() with the i_mutex lock held 1977a60e3cc7SHannes Frederic Sowa * and networking will grab the socket lock. 1978a60e3cc7SHannes Frederic Sowa */ 1979a60e3cc7SHannes Frederic Sowa release_sock(sk); 1980a60e3cc7SHannes Frederic Sowa ret = splice_to_pipe(pipe, spd); 1981a60e3cc7SHannes Frederic Sowa lock_sock(sk); 1982a60e3cc7SHannes Frederic Sowa 1983a60e3cc7SHannes Frederic Sowa return ret; 1984a60e3cc7SHannes Frederic Sowa } 1985a60e3cc7SHannes Frederic Sowa 19869c55e01cSJens Axboe /* 19879c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 1988fa9835e5STom Herbert * the fragments, and the frag list. 19899c55e01cSJens Axboe */ 1990a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 19919c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 1992a60e3cc7SHannes Frederic Sowa unsigned int flags, 1993a60e3cc7SHannes Frederic Sowa ssize_t (*splice_cb)(struct sock *, 1994a60e3cc7SHannes Frederic Sowa struct pipe_inode_info *, 1995a60e3cc7SHannes Frederic Sowa struct splice_pipe_desc *)) 19969c55e01cSJens Axboe { 199741c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 199841c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 19999c55e01cSJens Axboe struct splice_pipe_desc spd = { 20009c55e01cSJens Axboe .pages = pages, 20019c55e01cSJens Axboe .partial = partial, 2002047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 20039c55e01cSJens Axboe .flags = flags, 200428a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 20059c55e01cSJens Axboe .spd_release = sock_spd_release, 20069c55e01cSJens Axboe }; 200735f3d14dSJens Axboe int ret = 0; 200835f3d14dSJens Axboe 2009fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 20109c55e01cSJens Axboe 2011a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 2012a60e3cc7SHannes Frederic Sowa ret = splice_cb(sk, pipe, &spd); 20139c55e01cSJens Axboe 201435f3d14dSJens Axboe return ret; 20159c55e01cSJens Axboe } 20162b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 20179c55e01cSJens Axboe 2018357b40a1SHerbert Xu /** 2019357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2020357b40a1SHerbert Xu * @skb: destination buffer 2021357b40a1SHerbert Xu * @offset: offset in destination 2022357b40a1SHerbert Xu * @from: source buffer 2023357b40a1SHerbert Xu * @len: number of bytes to copy 2024357b40a1SHerbert Xu * 2025357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2026357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2027357b40a1SHerbert Xu * traversing fragment lists and such. 2028357b40a1SHerbert Xu */ 2029357b40a1SHerbert Xu 20300c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2031357b40a1SHerbert Xu { 20321a028e50SDavid S. Miller int start = skb_headlen(skb); 2033fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2034fbb398a8SDavid S. Miller int i, copy; 2035357b40a1SHerbert Xu 2036357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2037357b40a1SHerbert Xu goto fault; 2038357b40a1SHerbert Xu 20391a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2040357b40a1SHerbert Xu if (copy > len) 2041357b40a1SHerbert Xu copy = len; 204227d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2043357b40a1SHerbert Xu if ((len -= copy) == 0) 2044357b40a1SHerbert Xu return 0; 2045357b40a1SHerbert Xu offset += copy; 2046357b40a1SHerbert Xu from += copy; 2047357b40a1SHerbert Xu } 2048357b40a1SHerbert Xu 2049357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2050357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 20511a028e50SDavid S. Miller int end; 2052357b40a1SHerbert Xu 2053547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20541a028e50SDavid S. Miller 20559e903e08SEric Dumazet end = start + skb_frag_size(frag); 2056357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2057357b40a1SHerbert Xu u8 *vaddr; 2058357b40a1SHerbert Xu 2059357b40a1SHerbert Xu if (copy > len) 2060357b40a1SHerbert Xu copy = len; 2061357b40a1SHerbert Xu 206251c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 20631a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 20641a028e50SDavid S. Miller from, copy); 206551c56b00SEric Dumazet kunmap_atomic(vaddr); 2066357b40a1SHerbert Xu 2067357b40a1SHerbert Xu if ((len -= copy) == 0) 2068357b40a1SHerbert Xu return 0; 2069357b40a1SHerbert Xu offset += copy; 2070357b40a1SHerbert Xu from += copy; 2071357b40a1SHerbert Xu } 20721a028e50SDavid S. Miller start = end; 2073357b40a1SHerbert Xu } 2074357b40a1SHerbert Xu 2075fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 20761a028e50SDavid S. Miller int end; 2077357b40a1SHerbert Xu 2078547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20791a028e50SDavid S. Miller 2080fbb398a8SDavid S. Miller end = start + frag_iter->len; 2081357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2082357b40a1SHerbert Xu if (copy > len) 2083357b40a1SHerbert Xu copy = len; 2084fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 20851a028e50SDavid S. Miller from, copy)) 2086357b40a1SHerbert Xu goto fault; 2087357b40a1SHerbert Xu if ((len -= copy) == 0) 2088357b40a1SHerbert Xu return 0; 2089357b40a1SHerbert Xu offset += copy; 2090357b40a1SHerbert Xu from += copy; 2091357b40a1SHerbert Xu } 20921a028e50SDavid S. Miller start = end; 2093357b40a1SHerbert Xu } 2094357b40a1SHerbert Xu if (!len) 2095357b40a1SHerbert Xu return 0; 2096357b40a1SHerbert Xu 2097357b40a1SHerbert Xu fault: 2098357b40a1SHerbert Xu return -EFAULT; 2099357b40a1SHerbert Xu } 2100357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2101357b40a1SHerbert Xu 21021da177e4SLinus Torvalds /* Checksum skb data. */ 21032817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 21042817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 21051da177e4SLinus Torvalds { 21061a028e50SDavid S. Miller int start = skb_headlen(skb); 21071a028e50SDavid S. Miller int i, copy = start - offset; 2108fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 21091da177e4SLinus Torvalds int pos = 0; 21101da177e4SLinus Torvalds 21111da177e4SLinus Torvalds /* Checksum header. */ 21121da177e4SLinus Torvalds if (copy > 0) { 21131da177e4SLinus Torvalds if (copy > len) 21141da177e4SLinus Torvalds copy = len; 21152817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 21161da177e4SLinus Torvalds if ((len -= copy) == 0) 21171da177e4SLinus Torvalds return csum; 21181da177e4SLinus Torvalds offset += copy; 21191da177e4SLinus Torvalds pos = copy; 21201da177e4SLinus Torvalds } 21211da177e4SLinus Torvalds 21221da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21231a028e50SDavid S. Miller int end; 212451c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 21251da177e4SLinus Torvalds 2126547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21271a028e50SDavid S. Miller 212851c56b00SEric Dumazet end = start + skb_frag_size(frag); 21291da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 213044bb9363SAl Viro __wsum csum2; 21311da177e4SLinus Torvalds u8 *vaddr; 21321da177e4SLinus Torvalds 21331da177e4SLinus Torvalds if (copy > len) 21341da177e4SLinus Torvalds copy = len; 213551c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 21362817a336SDaniel Borkmann csum2 = ops->update(vaddr + frag->page_offset + 21371a028e50SDavid S. Miller offset - start, copy, 0); 213851c56b00SEric Dumazet kunmap_atomic(vaddr); 21392817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 21401da177e4SLinus Torvalds if (!(len -= copy)) 21411da177e4SLinus Torvalds return csum; 21421da177e4SLinus Torvalds offset += copy; 21431da177e4SLinus Torvalds pos += copy; 21441da177e4SLinus Torvalds } 21451a028e50SDavid S. Miller start = end; 21461da177e4SLinus Torvalds } 21471da177e4SLinus Torvalds 2148fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 21491a028e50SDavid S. Miller int end; 21501da177e4SLinus Torvalds 2151547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21521a028e50SDavid S. Miller 2153fbb398a8SDavid S. Miller end = start + frag_iter->len; 21541da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21555f92a738SAl Viro __wsum csum2; 21561da177e4SLinus Torvalds if (copy > len) 21571da177e4SLinus Torvalds copy = len; 21582817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 21592817a336SDaniel Borkmann copy, 0, ops); 21602817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 21611da177e4SLinus Torvalds if ((len -= copy) == 0) 21621da177e4SLinus Torvalds return csum; 21631da177e4SLinus Torvalds offset += copy; 21641da177e4SLinus Torvalds pos += copy; 21651da177e4SLinus Torvalds } 21661a028e50SDavid S. Miller start = end; 21671da177e4SLinus Torvalds } 216809a62660SKris Katterjohn BUG_ON(len); 21691da177e4SLinus Torvalds 21701da177e4SLinus Torvalds return csum; 21711da177e4SLinus Torvalds } 21722817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 21732817a336SDaniel Borkmann 21742817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 21752817a336SDaniel Borkmann int len, __wsum csum) 21762817a336SDaniel Borkmann { 21772817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2178cea80ea8SDaniel Borkmann .update = csum_partial_ext, 21792817a336SDaniel Borkmann .combine = csum_block_add_ext, 21802817a336SDaniel Borkmann }; 21812817a336SDaniel Borkmann 21822817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 21832817a336SDaniel Borkmann } 2184b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 21851da177e4SLinus Torvalds 21861da177e4SLinus Torvalds /* Both of above in one bottle. */ 21871da177e4SLinus Torvalds 218881d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 218981d77662SAl Viro u8 *to, int len, __wsum csum) 21901da177e4SLinus Torvalds { 21911a028e50SDavid S. Miller int start = skb_headlen(skb); 21921a028e50SDavid S. Miller int i, copy = start - offset; 2193fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 21941da177e4SLinus Torvalds int pos = 0; 21951da177e4SLinus Torvalds 21961da177e4SLinus Torvalds /* Copy header. */ 21971da177e4SLinus Torvalds if (copy > 0) { 21981da177e4SLinus Torvalds if (copy > len) 21991da177e4SLinus Torvalds copy = len; 22001da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 22011da177e4SLinus Torvalds copy, csum); 22021da177e4SLinus Torvalds if ((len -= copy) == 0) 22031da177e4SLinus Torvalds return csum; 22041da177e4SLinus Torvalds offset += copy; 22051da177e4SLinus Torvalds to += copy; 22061da177e4SLinus Torvalds pos = copy; 22071da177e4SLinus Torvalds } 22081da177e4SLinus Torvalds 22091da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22101a028e50SDavid S. Miller int end; 22111da177e4SLinus Torvalds 2212547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22131a028e50SDavid S. Miller 22149e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 22151da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 22165084205fSAl Viro __wsum csum2; 22171da177e4SLinus Torvalds u8 *vaddr; 22181da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 22191da177e4SLinus Torvalds 22201da177e4SLinus Torvalds if (copy > len) 22211da177e4SLinus Torvalds copy = len; 222251c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 22231da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 22241a028e50SDavid S. Miller frag->page_offset + 22251a028e50SDavid S. Miller offset - start, to, 22261a028e50SDavid S. Miller copy, 0); 222751c56b00SEric Dumazet kunmap_atomic(vaddr); 22281da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 22291da177e4SLinus Torvalds if (!(len -= copy)) 22301da177e4SLinus Torvalds return csum; 22311da177e4SLinus Torvalds offset += copy; 22321da177e4SLinus Torvalds to += copy; 22331da177e4SLinus Torvalds pos += copy; 22341da177e4SLinus Torvalds } 22351a028e50SDavid S. Miller start = end; 22361da177e4SLinus Torvalds } 22371da177e4SLinus Torvalds 2238fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 223981d77662SAl Viro __wsum csum2; 22401a028e50SDavid S. Miller int end; 22411da177e4SLinus Torvalds 2242547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22431a028e50SDavid S. Miller 2244fbb398a8SDavid S. Miller end = start + frag_iter->len; 22451da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 22461da177e4SLinus Torvalds if (copy > len) 22471da177e4SLinus Torvalds copy = len; 2248fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 22491a028e50SDavid S. Miller offset - start, 22501da177e4SLinus Torvalds to, copy, 0); 22511da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 22521da177e4SLinus Torvalds if ((len -= copy) == 0) 22531da177e4SLinus Torvalds return csum; 22541da177e4SLinus Torvalds offset += copy; 22551da177e4SLinus Torvalds to += copy; 22561da177e4SLinus Torvalds pos += copy; 22571da177e4SLinus Torvalds } 22581a028e50SDavid S. Miller start = end; 22591da177e4SLinus Torvalds } 226009a62660SKris Katterjohn BUG_ON(len); 22611da177e4SLinus Torvalds return csum; 22621da177e4SLinus Torvalds } 2263b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 22641da177e4SLinus Torvalds 2265af2806f8SThomas Graf /** 2266af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2267af2806f8SThomas Graf * @from: source buffer 2268af2806f8SThomas Graf * 2269af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2270af2806f8SThomas Graf * into skb_zerocopy(). 2271af2806f8SThomas Graf */ 2272af2806f8SThomas Graf unsigned int 2273af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2274af2806f8SThomas Graf { 2275af2806f8SThomas Graf unsigned int hlen = 0; 2276af2806f8SThomas Graf 2277af2806f8SThomas Graf if (!from->head_frag || 2278af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2279af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2280af2806f8SThomas Graf hlen = skb_headlen(from); 2281af2806f8SThomas Graf 2282af2806f8SThomas Graf if (skb_has_frag_list(from)) 2283af2806f8SThomas Graf hlen = from->len; 2284af2806f8SThomas Graf 2285af2806f8SThomas Graf return hlen; 2286af2806f8SThomas Graf } 2287af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2288af2806f8SThomas Graf 2289af2806f8SThomas Graf /** 2290af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2291af2806f8SThomas Graf * @to: destination buffer 22927fceb4deSMasanari Iida * @from: source buffer 2293af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2294af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2295af2806f8SThomas Graf * 2296af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2297af2806f8SThomas Graf * to the frags in the source buffer. 2298af2806f8SThomas Graf * 2299af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2300af2806f8SThomas Graf * headroom in the `to` buffer. 230136d5fe6aSZoltan Kiss * 230236d5fe6aSZoltan Kiss * Return value: 230336d5fe6aSZoltan Kiss * 0: everything is OK 230436d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 230536d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2306af2806f8SThomas Graf */ 230736d5fe6aSZoltan Kiss int 230836d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2309af2806f8SThomas Graf { 2310af2806f8SThomas Graf int i, j = 0; 2311af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 231236d5fe6aSZoltan Kiss int ret; 2313af2806f8SThomas Graf struct page *page; 2314af2806f8SThomas Graf unsigned int offset; 2315af2806f8SThomas Graf 2316af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2317af2806f8SThomas Graf 2318af2806f8SThomas Graf /* dont bother with small payloads */ 231936d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 232036d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2321af2806f8SThomas Graf 2322af2806f8SThomas Graf if (hlen) { 232336d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 232436d5fe6aSZoltan Kiss if (unlikely(ret)) 232536d5fe6aSZoltan Kiss return ret; 2326af2806f8SThomas Graf len -= hlen; 2327af2806f8SThomas Graf } else { 2328af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2329af2806f8SThomas Graf if (plen) { 2330af2806f8SThomas Graf page = virt_to_head_page(from->head); 2331af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2332af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2333af2806f8SThomas Graf get_page(page); 2334af2806f8SThomas Graf j = 1; 2335af2806f8SThomas Graf len -= plen; 2336af2806f8SThomas Graf } 2337af2806f8SThomas Graf } 2338af2806f8SThomas Graf 2339af2806f8SThomas Graf to->truesize += len + plen; 2340af2806f8SThomas Graf to->len += len + plen; 2341af2806f8SThomas Graf to->data_len += len + plen; 2342af2806f8SThomas Graf 234336d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 234436d5fe6aSZoltan Kiss skb_tx_error(from); 234536d5fe6aSZoltan Kiss return -ENOMEM; 234636d5fe6aSZoltan Kiss } 234736d5fe6aSZoltan Kiss 2348af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2349af2806f8SThomas Graf if (!len) 2350af2806f8SThomas Graf break; 2351af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2352af2806f8SThomas Graf skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2353af2806f8SThomas Graf len -= skb_shinfo(to)->frags[j].size; 2354af2806f8SThomas Graf skb_frag_ref(to, j); 2355af2806f8SThomas Graf j++; 2356af2806f8SThomas Graf } 2357af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 235836d5fe6aSZoltan Kiss 235936d5fe6aSZoltan Kiss return 0; 2360af2806f8SThomas Graf } 2361af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2362af2806f8SThomas Graf 23631da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 23641da177e4SLinus Torvalds { 2365d3bc23e7SAl Viro __wsum csum; 23661da177e4SLinus Torvalds long csstart; 23671da177e4SLinus Torvalds 236884fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 236955508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 23701da177e4SLinus Torvalds else 23711da177e4SLinus Torvalds csstart = skb_headlen(skb); 23721da177e4SLinus Torvalds 237309a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 23741da177e4SLinus Torvalds 2375d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 23761da177e4SLinus Torvalds 23771da177e4SLinus Torvalds csum = 0; 23781da177e4SLinus Torvalds if (csstart != skb->len) 23791da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 23801da177e4SLinus Torvalds skb->len - csstart, 0); 23811da177e4SLinus Torvalds 238284fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2383ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 23841da177e4SLinus Torvalds 2385d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 23861da177e4SLinus Torvalds } 23871da177e4SLinus Torvalds } 2388b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 23891da177e4SLinus Torvalds 23901da177e4SLinus Torvalds /** 23911da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 23921da177e4SLinus Torvalds * @list: list to dequeue from 23931da177e4SLinus Torvalds * 23941da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 23951da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 23961da177e4SLinus Torvalds * returned or %NULL if the list is empty. 23971da177e4SLinus Torvalds */ 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 24001da177e4SLinus Torvalds { 24011da177e4SLinus Torvalds unsigned long flags; 24021da177e4SLinus Torvalds struct sk_buff *result; 24031da177e4SLinus Torvalds 24041da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24051da177e4SLinus Torvalds result = __skb_dequeue(list); 24061da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24071da177e4SLinus Torvalds return result; 24081da177e4SLinus Torvalds } 2409b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 24101da177e4SLinus Torvalds 24111da177e4SLinus Torvalds /** 24121da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 24131da177e4SLinus Torvalds * @list: list to dequeue from 24141da177e4SLinus Torvalds * 24151da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 24161da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 24171da177e4SLinus Torvalds * returned or %NULL if the list is empty. 24181da177e4SLinus Torvalds */ 24191da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 24201da177e4SLinus Torvalds { 24211da177e4SLinus Torvalds unsigned long flags; 24221da177e4SLinus Torvalds struct sk_buff *result; 24231da177e4SLinus Torvalds 24241da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24251da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 24261da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24271da177e4SLinus Torvalds return result; 24281da177e4SLinus Torvalds } 2429b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 24301da177e4SLinus Torvalds 24311da177e4SLinus Torvalds /** 24321da177e4SLinus Torvalds * skb_queue_purge - empty a list 24331da177e4SLinus Torvalds * @list: list to empty 24341da177e4SLinus Torvalds * 24351da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 24361da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 24371da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 24381da177e4SLinus Torvalds */ 24391da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 24401da177e4SLinus Torvalds { 24411da177e4SLinus Torvalds struct sk_buff *skb; 24421da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 24431da177e4SLinus Torvalds kfree_skb(skb); 24441da177e4SLinus Torvalds } 2445b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 24461da177e4SLinus Torvalds 24471da177e4SLinus Torvalds /** 24489f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 24499f5afeaeSYaogong Wang * @root: root of the rbtree to empty 24509f5afeaeSYaogong Wang * 24519f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 24529f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 24539f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 24549f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 24559f5afeaeSYaogong Wang */ 24569f5afeaeSYaogong Wang void skb_rbtree_purge(struct rb_root *root) 24579f5afeaeSYaogong Wang { 24589f5afeaeSYaogong Wang struct sk_buff *skb, *next; 24599f5afeaeSYaogong Wang 24609f5afeaeSYaogong Wang rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode) 24619f5afeaeSYaogong Wang kfree_skb(skb); 24629f5afeaeSYaogong Wang 24639f5afeaeSYaogong Wang *root = RB_ROOT; 24649f5afeaeSYaogong Wang } 24659f5afeaeSYaogong Wang 24669f5afeaeSYaogong Wang /** 24671da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 24681da177e4SLinus Torvalds * @list: list to use 24691da177e4SLinus Torvalds * @newsk: buffer to queue 24701da177e4SLinus Torvalds * 24711da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 24721da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 24731da177e4SLinus Torvalds * safely. 24741da177e4SLinus Torvalds * 24751da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24761da177e4SLinus Torvalds */ 24771da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 24781da177e4SLinus Torvalds { 24791da177e4SLinus Torvalds unsigned long flags; 24801da177e4SLinus Torvalds 24811da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24821da177e4SLinus Torvalds __skb_queue_head(list, newsk); 24831da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24841da177e4SLinus Torvalds } 2485b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 24861da177e4SLinus Torvalds 24871da177e4SLinus Torvalds /** 24881da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 24891da177e4SLinus Torvalds * @list: list to use 24901da177e4SLinus Torvalds * @newsk: buffer to queue 24911da177e4SLinus Torvalds * 24921da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 24931da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 24941da177e4SLinus Torvalds * safely. 24951da177e4SLinus Torvalds * 24961da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24971da177e4SLinus Torvalds */ 24981da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 24991da177e4SLinus Torvalds { 25001da177e4SLinus Torvalds unsigned long flags; 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 25031da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 25041da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 25051da177e4SLinus Torvalds } 2506b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 25078728b834SDavid S. Miller 25081da177e4SLinus Torvalds /** 25091da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 25101da177e4SLinus Torvalds * @skb: buffer to remove 25118728b834SDavid S. Miller * @list: list to use 25121da177e4SLinus Torvalds * 25138728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 25148728b834SDavid S. Miller * function is atomic with respect to other list locked calls 25151da177e4SLinus Torvalds * 25168728b834SDavid S. Miller * You must know what list the SKB is on. 25171da177e4SLinus Torvalds */ 25188728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 25191da177e4SLinus Torvalds { 25201da177e4SLinus Torvalds unsigned long flags; 25211da177e4SLinus Torvalds 25221da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 25238728b834SDavid S. Miller __skb_unlink(skb, list); 25241da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 25251da177e4SLinus Torvalds } 2526b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 25271da177e4SLinus Torvalds 25281da177e4SLinus Torvalds /** 25291da177e4SLinus Torvalds * skb_append - append a buffer 25301da177e4SLinus Torvalds * @old: buffer to insert after 25311da177e4SLinus Torvalds * @newsk: buffer to insert 25328728b834SDavid S. Miller * @list: list to use 25331da177e4SLinus Torvalds * 25341da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 25351da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 25361da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 25371da177e4SLinus Torvalds */ 25388728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 25391da177e4SLinus Torvalds { 25401da177e4SLinus Torvalds unsigned long flags; 25411da177e4SLinus Torvalds 25428728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 25437de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 25448728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 25451da177e4SLinus Torvalds } 2546b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 25471da177e4SLinus Torvalds 25481da177e4SLinus Torvalds /** 25491da177e4SLinus Torvalds * skb_insert - insert a buffer 25501da177e4SLinus Torvalds * @old: buffer to insert before 25511da177e4SLinus Torvalds * @newsk: buffer to insert 25528728b834SDavid S. Miller * @list: list to use 25531da177e4SLinus Torvalds * 25548728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 25558728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 25568728b834SDavid S. Miller * calls. 25578728b834SDavid S. Miller * 25581da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 25591da177e4SLinus Torvalds */ 25608728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 25611da177e4SLinus Torvalds { 25621da177e4SLinus Torvalds unsigned long flags; 25631da177e4SLinus Torvalds 25648728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 25658728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 25668728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 25671da177e4SLinus Torvalds } 2568b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 25691da177e4SLinus Torvalds 25701da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 25711da177e4SLinus Torvalds struct sk_buff* skb1, 25721da177e4SLinus Torvalds const u32 len, const int pos) 25731da177e4SLinus Torvalds { 25741da177e4SLinus Torvalds int i; 25751da177e4SLinus Torvalds 2576d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2577d626f62bSArnaldo Carvalho de Melo pos - len); 25781da177e4SLinus Torvalds /* And move data appendix as is. */ 25791da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 25801da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 25811da177e4SLinus Torvalds 25821da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 25831da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 25841da177e4SLinus Torvalds skb1->data_len = skb->data_len; 25851da177e4SLinus Torvalds skb1->len += skb1->data_len; 25861da177e4SLinus Torvalds skb->data_len = 0; 25871da177e4SLinus Torvalds skb->len = len; 258827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 25891da177e4SLinus Torvalds } 25901da177e4SLinus Torvalds 25911da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 25921da177e4SLinus Torvalds struct sk_buff* skb1, 25931da177e4SLinus Torvalds const u32 len, int pos) 25941da177e4SLinus Torvalds { 25951da177e4SLinus Torvalds int i, k = 0; 25961da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 25971da177e4SLinus Torvalds 25981da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 25991da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 26001da177e4SLinus Torvalds skb->len = len; 26011da177e4SLinus Torvalds skb->data_len = len - pos; 26021da177e4SLinus Torvalds 26031da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 26049e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 26051da177e4SLinus Torvalds 26061da177e4SLinus Torvalds if (pos + size > len) { 26071da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 26081da177e4SLinus Torvalds 26091da177e4SLinus Torvalds if (pos < len) { 26101da177e4SLinus Torvalds /* Split frag. 26111da177e4SLinus Torvalds * We have two variants in this case: 26121da177e4SLinus Torvalds * 1. Move all the frag to the second 26131da177e4SLinus Torvalds * part, if it is possible. F.e. 26141da177e4SLinus Torvalds * this approach is mandatory for TUX, 26151da177e4SLinus Torvalds * where splitting is expensive. 26161da177e4SLinus Torvalds * 2. Split is accurately. We make this. 26171da177e4SLinus Torvalds */ 2618ea2ab693SIan Campbell skb_frag_ref(skb, i); 26191da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 26209e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 26219e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 26221da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 26231da177e4SLinus Torvalds } 26241da177e4SLinus Torvalds k++; 26251da177e4SLinus Torvalds } else 26261da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 26271da177e4SLinus Torvalds pos += size; 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 26301da177e4SLinus Torvalds } 26311da177e4SLinus Torvalds 26321da177e4SLinus Torvalds /** 26331da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 26341da177e4SLinus Torvalds * @skb: the buffer to split 26351da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 26361da177e4SLinus Torvalds * @len: new length for skb 26371da177e4SLinus Torvalds */ 26381da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 26391da177e4SLinus Torvalds { 26401da177e4SLinus Torvalds int pos = skb_headlen(skb); 26411da177e4SLinus Torvalds 264268534c68SAmerigo Wang skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 26431da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 26441da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 26451da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 26461da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 26471da177e4SLinus Torvalds } 2648b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 26491da177e4SLinus Torvalds 26509f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 26519f782db3SIlpo Järvinen * 26529f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 26539f782db3SIlpo Järvinen */ 2654832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2655832d11c5SIlpo Järvinen { 26560ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2657832d11c5SIlpo Järvinen } 2658832d11c5SIlpo Järvinen 2659832d11c5SIlpo Järvinen /** 2660832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2661832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2662832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2663832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2664832d11c5SIlpo Järvinen * 2665832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 266620e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 2667832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2668832d11c5SIlpo Järvinen * 2669832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2670832d11c5SIlpo Järvinen * 2671832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2672832d11c5SIlpo Järvinen * to have non-paged data as well. 2673832d11c5SIlpo Järvinen * 2674832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2675832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2676832d11c5SIlpo Järvinen */ 2677832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2678832d11c5SIlpo Järvinen { 2679832d11c5SIlpo Järvinen int from, to, merge, todo; 2680832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2681832d11c5SIlpo Järvinen 2682832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2683832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2684832d11c5SIlpo Järvinen 2685832d11c5SIlpo Järvinen todo = shiftlen; 2686832d11c5SIlpo Järvinen from = 0; 2687832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2688832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2689832d11c5SIlpo Järvinen 2690832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2691832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2692832d11c5SIlpo Järvinen */ 2693832d11c5SIlpo Järvinen if (!to || 2694ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2695ea2ab693SIan Campbell fragfrom->page_offset)) { 2696832d11c5SIlpo Järvinen merge = -1; 2697832d11c5SIlpo Järvinen } else { 2698832d11c5SIlpo Järvinen merge = to - 1; 2699832d11c5SIlpo Järvinen 27009e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2701832d11c5SIlpo Järvinen if (todo < 0) { 2702832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2703832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2704832d11c5SIlpo Järvinen return 0; 2705832d11c5SIlpo Järvinen 27069f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 27079f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2708832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2709832d11c5SIlpo Järvinen 27109e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 27119e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 2712832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2713832d11c5SIlpo Järvinen 2714832d11c5SIlpo Järvinen goto onlymerged; 2715832d11c5SIlpo Järvinen } 2716832d11c5SIlpo Järvinen 2717832d11c5SIlpo Järvinen from++; 2718832d11c5SIlpo Järvinen } 2719832d11c5SIlpo Järvinen 2720832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2721832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2722832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2723832d11c5SIlpo Järvinen return 0; 2724832d11c5SIlpo Järvinen 2725832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2726832d11c5SIlpo Järvinen return 0; 2727832d11c5SIlpo Järvinen 2728832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2729832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2730832d11c5SIlpo Järvinen return 0; 2731832d11c5SIlpo Järvinen 2732832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2733832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2734832d11c5SIlpo Järvinen 27359e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 2736832d11c5SIlpo Järvinen *fragto = *fragfrom; 27379e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2738832d11c5SIlpo Järvinen from++; 2739832d11c5SIlpo Järvinen to++; 2740832d11c5SIlpo Järvinen 2741832d11c5SIlpo Järvinen } else { 2742ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2743832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2744832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 27459e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 2746832d11c5SIlpo Järvinen 2747832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 27489e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 2749832d11c5SIlpo Järvinen todo = 0; 2750832d11c5SIlpo Järvinen 2751832d11c5SIlpo Järvinen to++; 2752832d11c5SIlpo Järvinen break; 2753832d11c5SIlpo Järvinen } 2754832d11c5SIlpo Järvinen } 2755832d11c5SIlpo Järvinen 2756832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2757832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2758832d11c5SIlpo Järvinen 2759832d11c5SIlpo Järvinen if (merge >= 0) { 2760832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2761832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2762832d11c5SIlpo Järvinen 27639e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2764ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2765832d11c5SIlpo Järvinen } 2766832d11c5SIlpo Järvinen 2767832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2768832d11c5SIlpo Järvinen to = 0; 2769832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2770832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2771832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2772832d11c5SIlpo Järvinen 2773832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2774832d11c5SIlpo Järvinen 2775832d11c5SIlpo Järvinen onlymerged: 2776832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2777832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2778832d11c5SIlpo Järvinen */ 2779832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2780832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2781832d11c5SIlpo Järvinen 2782832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2783832d11c5SIlpo Järvinen skb->len -= shiftlen; 2784832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2785832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2786832d11c5SIlpo Järvinen tgt->len += shiftlen; 2787832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2788832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2789832d11c5SIlpo Järvinen 2790832d11c5SIlpo Järvinen return shiftlen; 2791832d11c5SIlpo Järvinen } 2792832d11c5SIlpo Järvinen 2793677e90edSThomas Graf /** 2794677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2795677e90edSThomas Graf * @skb: the buffer to read 2796677e90edSThomas Graf * @from: lower offset of data to be read 2797677e90edSThomas Graf * @to: upper offset of data to be read 2798677e90edSThomas Graf * @st: state variable 2799677e90edSThomas Graf * 2800677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2801677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2802677e90edSThomas Graf */ 2803677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2804677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2805677e90edSThomas Graf { 2806677e90edSThomas Graf st->lower_offset = from; 2807677e90edSThomas Graf st->upper_offset = to; 2808677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2809677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2810677e90edSThomas Graf st->frag_data = NULL; 2811677e90edSThomas Graf } 2812b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2813677e90edSThomas Graf 2814677e90edSThomas Graf /** 2815677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2816677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2817677e90edSThomas Graf * @data: destination pointer for data to be returned 2818677e90edSThomas Graf * @st: state variable 2819677e90edSThomas Graf * 2820bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 2821677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2822bc32383cSMathias Krause * the head of the data block to @data and returns the length 2823677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2824677e90edSThomas Graf * offset has been reached. 2825677e90edSThomas Graf * 2826677e90edSThomas Graf * The caller is not required to consume all of the data 2827bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 2828677e90edSThomas Graf * of bytes already consumed and the next call to 2829677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2830677e90edSThomas Graf * 283125985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2832e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 2833677e90edSThomas Graf * reads of potentially non linear data. 2834677e90edSThomas Graf * 2835bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2836677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2837677e90edSThomas Graf * a stack for this purpose. 2838677e90edSThomas Graf */ 2839677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2840677e90edSThomas Graf struct skb_seq_state *st) 2841677e90edSThomas Graf { 2842677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2843677e90edSThomas Graf skb_frag_t *frag; 2844677e90edSThomas Graf 2845aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 2846aeb193eaSWedson Almeida Filho if (st->frag_data) { 2847aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 2848aeb193eaSWedson Almeida Filho st->frag_data = NULL; 2849aeb193eaSWedson Almeida Filho } 2850677e90edSThomas Graf return 0; 2851aeb193eaSWedson Almeida Filho } 2852677e90edSThomas Graf 2853677e90edSThomas Graf next_skb: 285495e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2855677e90edSThomas Graf 2856995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 285795e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2858677e90edSThomas Graf return block_limit - abs_offset; 2859677e90edSThomas Graf } 2860677e90edSThomas Graf 2861677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2862677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2863677e90edSThomas Graf 2864677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2865677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 28669e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 2867677e90edSThomas Graf 2868677e90edSThomas Graf if (abs_offset < block_limit) { 2869677e90edSThomas Graf if (!st->frag_data) 287051c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 2871677e90edSThomas Graf 2872677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2873677e90edSThomas Graf (abs_offset - st->stepped_offset); 2874677e90edSThomas Graf 2875677e90edSThomas Graf return block_limit - abs_offset; 2876677e90edSThomas Graf } 2877677e90edSThomas Graf 2878677e90edSThomas Graf if (st->frag_data) { 287951c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2880677e90edSThomas Graf st->frag_data = NULL; 2881677e90edSThomas Graf } 2882677e90edSThomas Graf 2883677e90edSThomas Graf st->frag_idx++; 28849e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 2885677e90edSThomas Graf } 2886677e90edSThomas Graf 28875b5a60daSOlaf Kirch if (st->frag_data) { 288851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 28895b5a60daSOlaf Kirch st->frag_data = NULL; 28905b5a60daSOlaf Kirch } 28915b5a60daSOlaf Kirch 289221dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2893677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 289495e3b24cSHerbert Xu st->frag_idx = 0; 2895677e90edSThomas Graf goto next_skb; 289671b3346dSShyam Iyer } else if (st->cur_skb->next) { 289771b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 289871b3346dSShyam Iyer st->frag_idx = 0; 2899677e90edSThomas Graf goto next_skb; 2900677e90edSThomas Graf } 2901677e90edSThomas Graf 2902677e90edSThomas Graf return 0; 2903677e90edSThomas Graf } 2904b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2905677e90edSThomas Graf 2906677e90edSThomas Graf /** 2907677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2908677e90edSThomas Graf * @st: state variable 2909677e90edSThomas Graf * 2910677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2911677e90edSThomas Graf * returned 0. 2912677e90edSThomas Graf */ 2913677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2914677e90edSThomas Graf { 2915677e90edSThomas Graf if (st->frag_data) 291651c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2917677e90edSThomas Graf } 2918b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2919677e90edSThomas Graf 29203fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 29213fc7e8a6SThomas Graf 29223fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 29233fc7e8a6SThomas Graf struct ts_config *conf, 29243fc7e8a6SThomas Graf struct ts_state *state) 29253fc7e8a6SThomas Graf { 29263fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 29273fc7e8a6SThomas Graf } 29283fc7e8a6SThomas Graf 29293fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 29303fc7e8a6SThomas Graf { 29313fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 29323fc7e8a6SThomas Graf } 29333fc7e8a6SThomas Graf 29343fc7e8a6SThomas Graf /** 29353fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 29363fc7e8a6SThomas Graf * @skb: the buffer to look in 29373fc7e8a6SThomas Graf * @from: search offset 29383fc7e8a6SThomas Graf * @to: search limit 29393fc7e8a6SThomas Graf * @config: textsearch configuration 29403fc7e8a6SThomas Graf * 29413fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 29423fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 29433fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 29443fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 29453fc7e8a6SThomas Graf */ 29463fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 2947059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 29483fc7e8a6SThomas Graf { 2949059a2440SBojan Prtvar struct ts_state state; 2950f72b948dSPhil Oester unsigned int ret; 2951f72b948dSPhil Oester 29523fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 29533fc7e8a6SThomas Graf config->finish = skb_ts_finish; 29543fc7e8a6SThomas Graf 2955059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 29563fc7e8a6SThomas Graf 2957059a2440SBojan Prtvar ret = textsearch_find(config, &state); 2958f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 29593fc7e8a6SThomas Graf } 2960b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 29613fc7e8a6SThomas Graf 2962e89e9cf5SAnanda Raju /** 29632c53040fSBen Hutchings * skb_append_datato_frags - append the user data to a skb 2964e89e9cf5SAnanda Raju * @sk: sock structure 2965e793c0f7SMasanari Iida * @skb: skb structure to be appended with user data. 2966e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2967e89e9cf5SAnanda Raju * @from: pointer to user message iov 2968e89e9cf5SAnanda Raju * @length: length of the iov message 2969e89e9cf5SAnanda Raju * 2970e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2971e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2972e89e9cf5SAnanda Raju */ 2973e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2974dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2975e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2976e89e9cf5SAnanda Raju void *from, int length) 2977e89e9cf5SAnanda Raju { 2978b2111724SEric Dumazet int frg_cnt = skb_shinfo(skb)->nr_frags; 2979b2111724SEric Dumazet int copy; 2980e89e9cf5SAnanda Raju int offset = 0; 2981e89e9cf5SAnanda Raju int ret; 2982b2111724SEric Dumazet struct page_frag *pfrag = ¤t->task_frag; 2983e89e9cf5SAnanda Raju 2984e89e9cf5SAnanda Raju do { 2985e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2986e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2987b2111724SEric Dumazet return -EMSGSIZE; 2988e89e9cf5SAnanda Raju 2989b2111724SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 2990e89e9cf5SAnanda Raju return -ENOMEM; 2991e89e9cf5SAnanda Raju 2992e89e9cf5SAnanda Raju /* copy the user data to page */ 2993b2111724SEric Dumazet copy = min_t(int, length, pfrag->size - pfrag->offset); 2994e89e9cf5SAnanda Raju 2995b2111724SEric Dumazet ret = getfrag(from, page_address(pfrag->page) + pfrag->offset, 2996e89e9cf5SAnanda Raju offset, copy, 0, skb); 2997e89e9cf5SAnanda Raju if (ret < 0) 2998e89e9cf5SAnanda Raju return -EFAULT; 2999e89e9cf5SAnanda Raju 3000e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 3001b2111724SEric Dumazet skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset, 3002b2111724SEric Dumazet copy); 3003b2111724SEric Dumazet frg_cnt++; 3004b2111724SEric Dumazet pfrag->offset += copy; 3005b2111724SEric Dumazet get_page(pfrag->page); 3006b2111724SEric Dumazet 3007b2111724SEric Dumazet skb->truesize += copy; 3008b2111724SEric Dumazet atomic_add(copy, &sk->sk_wmem_alloc); 3009e89e9cf5SAnanda Raju skb->len += copy; 3010e89e9cf5SAnanda Raju skb->data_len += copy; 3011e89e9cf5SAnanda Raju offset += copy; 3012e89e9cf5SAnanda Raju length -= copy; 3013e89e9cf5SAnanda Raju 3014e89e9cf5SAnanda Raju } while (length > 0); 3015e89e9cf5SAnanda Raju 3016e89e9cf5SAnanda Raju return 0; 3017e89e9cf5SAnanda Raju } 3018b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 3019e89e9cf5SAnanda Raju 3020be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3021be12a1feSHannes Frederic Sowa int offset, size_t size) 3022be12a1feSHannes Frederic Sowa { 3023be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3024be12a1feSHannes Frederic Sowa 3025be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3026be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3027be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3028be12a1feSHannes Frederic Sowa get_page(page); 3029be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3030be12a1feSHannes Frederic Sowa } else { 3031be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3032be12a1feSHannes Frederic Sowa } 3033be12a1feSHannes Frederic Sowa 3034be12a1feSHannes Frederic Sowa return 0; 3035be12a1feSHannes Frederic Sowa } 3036be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3037be12a1feSHannes Frederic Sowa 3038cbb042f9SHerbert Xu /** 3039cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3040cbb042f9SHerbert Xu * @skb: buffer to update 3041cbb042f9SHerbert Xu * @len: length of data pulled 3042cbb042f9SHerbert Xu * 3043cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3044fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 304584fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 304684fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 304784fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3048cbb042f9SHerbert Xu */ 3049cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3050cbb042f9SHerbert Xu { 305131b33dfbSPravin B Shelar unsigned char *data = skb->data; 305231b33dfbSPravin B Shelar 3053cbb042f9SHerbert Xu BUG_ON(len > skb->len); 305431b33dfbSPravin B Shelar __skb_pull(skb, len); 305531b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 305631b33dfbSPravin B Shelar return skb->data; 3057cbb042f9SHerbert Xu } 3058f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3059f94691acSArnaldo Carvalho de Melo 3060f4c50d99SHerbert Xu /** 3061f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3062df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3063576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3064f4c50d99SHerbert Xu * 3065f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 30664c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 30674c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3068f4c50d99SHerbert Xu */ 3069df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3070df5771ffSMichael S. Tsirkin netdev_features_t features) 3071f4c50d99SHerbert Xu { 3072f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3073f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 30741a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3075df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3076df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3077df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 30781fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3079f4c50d99SHerbert Xu unsigned int offset = doffset; 3080df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3081802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3082f4c50d99SHerbert Xu unsigned int headroom; 3083802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3084ec5f0615SPravin B Shelar __be16 proto; 308536c98382SAlexander Duyck bool csum, sg; 3086df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3087f4c50d99SHerbert Xu int err = -ENOMEM; 3088f4c50d99SHerbert Xu int i = 0; 3089f4c50d99SHerbert Xu int pos; 309053d6471cSVlad Yasevich int dummy; 3091f4c50d99SHerbert Xu 30925882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 309353d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 3094ec5f0615SPravin B Shelar if (unlikely(!proto)) 3095ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3096ec5f0615SPravin B Shelar 309736c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3098f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 30997e2b10c1STom Herbert 3100802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 3101802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 3102802ab55aSAlexander Duyck * now. 3103802ab55aSAlexander Duyck */ 310436c98382SAlexander Duyck if (sg && csum && (features & NETIF_F_GSO_PARTIAL)) { 3105802ab55aSAlexander Duyck partial_segs = len / mss; 3106d7fb5a80SAlexander Duyck if (partial_segs > 1) 3107802ab55aSAlexander Duyck mss *= partial_segs; 3108d7fb5a80SAlexander Duyck else 3109d7fb5a80SAlexander Duyck partial_segs = 0; 3110802ab55aSAlexander Duyck } 3111802ab55aSAlexander Duyck 3112df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3113df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3114f4c50d99SHerbert Xu 3115f4c50d99SHerbert Xu do { 3116f4c50d99SHerbert Xu struct sk_buff *nskb; 31178cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3118c8884eddSHerbert Xu int hsize; 3119f4c50d99SHerbert Xu int size; 3120f4c50d99SHerbert Xu 31213953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 31223953c46cSMarcelo Ricardo Leitner len = list_skb->len; 31233953c46cSMarcelo Ricardo Leitner } else { 3124df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3125f4c50d99SHerbert Xu if (len > mss) 3126f4c50d99SHerbert Xu len = mss; 31273953c46cSMarcelo Ricardo Leitner } 3128f4c50d99SHerbert Xu 3129df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3130f4c50d99SHerbert Xu if (hsize < 0) 3131f4c50d99SHerbert Xu hsize = 0; 3132c8884eddSHerbert Xu if (hsize > len || !sg) 3133c8884eddSHerbert Xu hsize = len; 3134f4c50d99SHerbert Xu 31351a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 31361a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 31371a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 313889319d38SHerbert Xu 31399d8506ccSHerbert Xu i = 0; 31401a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 31411a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 31421fd819ecSMichael S. Tsirkin frag_skb = list_skb; 31431a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 31449d8506ccSHerbert Xu 31459d8506ccSHerbert Xu while (pos < offset + len) { 31469d8506ccSHerbert Xu BUG_ON(i >= nfrags); 31479d8506ccSHerbert Xu 31484e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 31499d8506ccSHerbert Xu if (pos + size > offset + len) 31509d8506ccSHerbert Xu break; 31519d8506ccSHerbert Xu 31529d8506ccSHerbert Xu i++; 31539d8506ccSHerbert Xu pos += size; 31544e1beba1SMichael S. Tsirkin frag++; 31559d8506ccSHerbert Xu } 31569d8506ccSHerbert Xu 31571a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 31581a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 315989319d38SHerbert Xu 3160f4c50d99SHerbert Xu if (unlikely(!nskb)) 3161f4c50d99SHerbert Xu goto err; 3162f4c50d99SHerbert Xu 31639d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 31649d8506ccSHerbert Xu kfree_skb(nskb); 31659d8506ccSHerbert Xu goto err; 31669d8506ccSHerbert Xu } 31679d8506ccSHerbert Xu 3168ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 316989319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 317089319d38SHerbert Xu kfree_skb(nskb); 317189319d38SHerbert Xu goto err; 317289319d38SHerbert Xu } 317389319d38SHerbert Xu 3174ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 317589319d38SHerbert Xu skb_release_head_state(nskb); 317689319d38SHerbert Xu __skb_push(nskb, doffset); 317789319d38SHerbert Xu } else { 3178c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3179df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3180c93bdd0eSMel Gorman NUMA_NO_NODE); 318189319d38SHerbert Xu 318289319d38SHerbert Xu if (unlikely(!nskb)) 318389319d38SHerbert Xu goto err; 318489319d38SHerbert Xu 318589319d38SHerbert Xu skb_reserve(nskb, headroom); 318689319d38SHerbert Xu __skb_put(nskb, doffset); 318789319d38SHerbert Xu } 318889319d38SHerbert Xu 3189f4c50d99SHerbert Xu if (segs) 3190f4c50d99SHerbert Xu tail->next = nskb; 3191f4c50d99SHerbert Xu else 3192f4c50d99SHerbert Xu segs = nskb; 3193f4c50d99SHerbert Xu tail = nskb; 3194f4c50d99SHerbert Xu 3195df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3196f4c50d99SHerbert Xu 3197030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3198fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 319968c33163SPravin B Shelar 3200df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 320168c33163SPravin B Shelar nskb->data - tnl_hlen, 320268c33163SPravin B Shelar doffset + tnl_hlen); 320389319d38SHerbert Xu 32049d8506ccSHerbert Xu if (nskb->len == len + doffset) 32051cdbcb79SSimon Horman goto perform_csum_check; 320689319d38SHerbert Xu 32077fbeffedSAlexander Duyck if (!sg) { 32087fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 32096f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 321076443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 321176443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 3212f4c50d99SHerbert Xu skb_put(nskb, len), 3213f4c50d99SHerbert Xu len, 0); 32147e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3215de843723STom Herbert skb_headroom(nskb) + doffset; 3216f4c50d99SHerbert Xu continue; 3217f4c50d99SHerbert Xu } 3218f4c50d99SHerbert Xu 32198cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3220f4c50d99SHerbert Xu 3221df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3222d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3223f4c50d99SHerbert Xu 3224df5771ffSMichael S. Tsirkin skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags & 3225df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3226cef401deSEric Dumazet 32279d8506ccSHerbert Xu while (pos < offset + len) { 32289d8506ccSHerbert Xu if (i >= nfrags) { 32291a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb)); 32309d8506ccSHerbert Xu 32319d8506ccSHerbert Xu i = 0; 32321a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 32331a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 32341fd819ecSMichael S. Tsirkin frag_skb = list_skb; 32359d8506ccSHerbert Xu 32369d8506ccSHerbert Xu BUG_ON(!nfrags); 32379d8506ccSHerbert Xu 32381a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 32399d8506ccSHerbert Xu } 32409d8506ccSHerbert Xu 32419d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 32429d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 32439d8506ccSHerbert Xu net_warn_ratelimited( 32449d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 32459d8506ccSHerbert Xu pos, mss); 32469d8506ccSHerbert Xu goto err; 32479d8506ccSHerbert Xu } 32489d8506ccSHerbert Xu 32491fd819ecSMichael S. Tsirkin if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC))) 32501fd819ecSMichael S. Tsirkin goto err; 32511fd819ecSMichael S. Tsirkin 32524e1beba1SMichael S. Tsirkin *nskb_frag = *frag; 32538cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 32548cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3255f4c50d99SHerbert Xu 3256f4c50d99SHerbert Xu if (pos < offset) { 32578cb19905SMichael S. Tsirkin nskb_frag->page_offset += offset - pos; 32588cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3259f4c50d99SHerbert Xu } 3260f4c50d99SHerbert Xu 326189319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3262f4c50d99SHerbert Xu 3263f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3264f4c50d99SHerbert Xu i++; 32654e1beba1SMichael S. Tsirkin frag++; 3266f4c50d99SHerbert Xu pos += size; 3267f4c50d99SHerbert Xu } else { 32688cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 326989319d38SHerbert Xu goto skip_fraglist; 3270f4c50d99SHerbert Xu } 3271f4c50d99SHerbert Xu 32728cb19905SMichael S. Tsirkin nskb_frag++; 3273f4c50d99SHerbert Xu } 3274f4c50d99SHerbert Xu 327589319d38SHerbert Xu skip_fraglist: 3276f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3277f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3278f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3279ec5f0615SPravin B Shelar 32801cdbcb79SSimon Horman perform_csum_check: 32817fbeffedSAlexander Duyck if (!csum) { 3282ddff00d4SAlexander Duyck if (skb_has_shared_frag(nskb)) { 3283ddff00d4SAlexander Duyck err = __skb_linearize(nskb); 3284ddff00d4SAlexander Duyck if (err) 3285ddff00d4SAlexander Duyck goto err; 3286ddff00d4SAlexander Duyck } 32877fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 3288ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 328976443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 329076443456SAlexander Duyck skb_checksum(nskb, doffset, 329176443456SAlexander Duyck nskb->len - doffset, 0); 32927e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 32937e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3294ec5f0615SPravin B Shelar } 3295df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3296f4c50d99SHerbert Xu 3297bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3298bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3299bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3300bec3cfdcSEric Dumazet */ 3301bec3cfdcSEric Dumazet segs->prev = tail; 3302432c856fSToshiaki Makita 3303802ab55aSAlexander Duyck /* Update GSO info on first skb in partial sequence. */ 3304802ab55aSAlexander Duyck if (partial_segs) { 3305802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 3306802ab55aSAlexander Duyck 3307802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 3308802ab55aSAlexander Duyck type |= SKB_GSO_PARTIAL; 3309802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 3310802ab55aSAlexander Duyck 3311802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 3312802ab55aSAlexander Duyck * our way back down the stack of protocols. 3313802ab55aSAlexander Duyck */ 3314802ab55aSAlexander Duyck skb_shinfo(segs)->gso_size = skb_shinfo(head_skb)->gso_size; 3315802ab55aSAlexander Duyck skb_shinfo(segs)->gso_segs = partial_segs; 3316802ab55aSAlexander Duyck skb_shinfo(segs)->gso_type = type; 3317802ab55aSAlexander Duyck SKB_GSO_CB(segs)->data_offset = skb_headroom(segs) + doffset; 3318802ab55aSAlexander Duyck } 3319802ab55aSAlexander Duyck 3320432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 3321432c856fSToshiaki Makita * using skb_set_owner_w(). 3322432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 3323432c856fSToshiaki Makita */ 3324432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 3325432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 3326432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 3327432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 3328432c856fSToshiaki Makita } 3329f4c50d99SHerbert Xu return segs; 3330f4c50d99SHerbert Xu 3331f4c50d99SHerbert Xu err: 3332289dccbeSEric Dumazet kfree_skb_list(segs); 3333f4c50d99SHerbert Xu return ERR_PTR(err); 3334f4c50d99SHerbert Xu } 3335f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3336f4c50d99SHerbert Xu 333771d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 333871d93b39SHerbert Xu { 33398a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 334067147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 334167147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 33428a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 334358025e46SEric Dumazet struct sk_buff *lp, *p = *head; 3344715dc1f3SEric Dumazet unsigned int delta_truesize; 334571d93b39SHerbert Xu 33468a29111cSEric Dumazet if (unlikely(p->len + len >= 65536)) 334771d93b39SHerbert Xu return -E2BIG; 334871d93b39SHerbert Xu 334929e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 33508a29111cSEric Dumazet pinfo = skb_shinfo(lp); 33518a29111cSEric Dumazet 33528a29111cSEric Dumazet if (headlen <= offset) { 335342da6994SHerbert Xu skb_frag_t *frag; 335466e92fcfSHerbert Xu skb_frag_t *frag2; 33559aaa156cSHerbert Xu int i = skbinfo->nr_frags; 33569aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 335742da6994SHerbert Xu 335866e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 33598a29111cSEric Dumazet goto merge; 336081705ad1SHerbert Xu 33618a29111cSEric Dumazet offset -= headlen; 33629aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 33639aaa156cSHerbert Xu skbinfo->nr_frags = 0; 3364f5572068SHerbert Xu 33659aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 33669aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 336766e92fcfSHerbert Xu do { 336866e92fcfSHerbert Xu *--frag = *--frag2; 336966e92fcfSHerbert Xu } while (--i); 337066e92fcfSHerbert Xu 337166e92fcfSHerbert Xu frag->page_offset += offset; 33729e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 337366e92fcfSHerbert Xu 3374715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3375ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3376ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3377715dc1f3SEric Dumazet 3378f5572068SHerbert Xu skb->truesize -= skb->data_len; 3379f5572068SHerbert Xu skb->len -= skb->data_len; 3380f5572068SHerbert Xu skb->data_len = 0; 3381f5572068SHerbert Xu 3382715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 33835d38a079SHerbert Xu goto done; 3384d7e8883cSEric Dumazet } else if (skb->head_frag) { 3385d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3386d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3387d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3388d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3389d7e8883cSEric Dumazet unsigned int first_offset; 3390d7e8883cSEric Dumazet 3391d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 33928a29111cSEric Dumazet goto merge; 3393d7e8883cSEric Dumazet 3394d7e8883cSEric Dumazet first_offset = skb->data - 3395d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3396d7e8883cSEric Dumazet offset; 3397d7e8883cSEric Dumazet 3398d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3399d7e8883cSEric Dumazet 3400d7e8883cSEric Dumazet frag->page.p = page; 3401d7e8883cSEric Dumazet frag->page_offset = first_offset; 3402d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3403d7e8883cSEric Dumazet 3404d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3405d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3406d7e8883cSEric Dumazet 3407715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3408d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3409d7e8883cSEric Dumazet goto done; 34108a29111cSEric Dumazet } 341171d93b39SHerbert Xu 341271d93b39SHerbert Xu merge: 3413715dc1f3SEric Dumazet delta_truesize = skb->truesize; 341467147ba9SHerbert Xu if (offset > headlen) { 3415d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3416d1dc7abfSMichal Schmidt 3417d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 34189e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3419d1dc7abfSMichal Schmidt skb->data_len -= eat; 3420d1dc7abfSMichal Schmidt skb->len -= eat; 342167147ba9SHerbert Xu offset = headlen; 342256035022SHerbert Xu } 342356035022SHerbert Xu 342467147ba9SHerbert Xu __skb_pull(skb, offset); 342556035022SHerbert Xu 342629e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 34278a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 34288a29111cSEric Dumazet else 3429c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3430c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 3431f4a775d1SEric Dumazet __skb_header_release(skb); 34328a29111cSEric Dumazet lp = p; 343371d93b39SHerbert Xu 34345d38a079SHerbert Xu done: 34355d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 343637fe4732SHerbert Xu p->data_len += len; 3437715dc1f3SEric Dumazet p->truesize += delta_truesize; 343837fe4732SHerbert Xu p->len += len; 34398a29111cSEric Dumazet if (lp != p) { 34408a29111cSEric Dumazet lp->data_len += len; 34418a29111cSEric Dumazet lp->truesize += delta_truesize; 34428a29111cSEric Dumazet lp->len += len; 34438a29111cSEric Dumazet } 344471d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 344571d93b39SHerbert Xu return 0; 344671d93b39SHerbert Xu } 344757c05650SMarcelo Ricardo Leitner EXPORT_SYMBOL_GPL(skb_gro_receive); 344871d93b39SHerbert Xu 34491da177e4SLinus Torvalds void __init skb_init(void) 34501da177e4SLinus Torvalds { 34511da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 34521da177e4SLinus Torvalds sizeof(struct sk_buff), 34531da177e4SLinus Torvalds 0, 3454e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 345520c2df83SPaul Mundt NULL); 3456d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3457d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 3458d179cd12SDavid S. Miller 0, 3459e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 346020c2df83SPaul Mundt NULL); 34611da177e4SLinus Torvalds } 34621da177e4SLinus Torvalds 3463716ea3a7SDavid Howells /** 3464716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3465716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 3466716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 3467716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 3468716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 3469716ea3a7SDavid Howells * 3470716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 3471716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 3472716ea3a7SDavid Howells */ 347351c739d1SDavid S. Miller static int 347451c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3475716ea3a7SDavid Howells { 34761a028e50SDavid S. Miller int start = skb_headlen(skb); 34771a028e50SDavid S. Miller int i, copy = start - offset; 3478fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3479716ea3a7SDavid Howells int elt = 0; 3480716ea3a7SDavid Howells 3481716ea3a7SDavid Howells if (copy > 0) { 3482716ea3a7SDavid Howells if (copy > len) 3483716ea3a7SDavid Howells copy = len; 3484642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 3485716ea3a7SDavid Howells elt++; 3486716ea3a7SDavid Howells if ((len -= copy) == 0) 3487716ea3a7SDavid Howells return elt; 3488716ea3a7SDavid Howells offset += copy; 3489716ea3a7SDavid Howells } 3490716ea3a7SDavid Howells 3491716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 34921a028e50SDavid S. Miller int end; 3493716ea3a7SDavid Howells 3494547b792cSIlpo Järvinen WARN_ON(start > offset + len); 34951a028e50SDavid S. Miller 34969e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3497716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3498716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3499716ea3a7SDavid Howells 3500716ea3a7SDavid Howells if (copy > len) 3501716ea3a7SDavid Howells copy = len; 3502ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3503642f1490SJens Axboe frag->page_offset+offset-start); 3504716ea3a7SDavid Howells elt++; 3505716ea3a7SDavid Howells if (!(len -= copy)) 3506716ea3a7SDavid Howells return elt; 3507716ea3a7SDavid Howells offset += copy; 3508716ea3a7SDavid Howells } 35091a028e50SDavid S. Miller start = end; 3510716ea3a7SDavid Howells } 3511716ea3a7SDavid Howells 3512fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 35131a028e50SDavid S. Miller int end; 3514716ea3a7SDavid Howells 3515547b792cSIlpo Järvinen WARN_ON(start > offset + len); 35161a028e50SDavid S. Miller 3517fbb398a8SDavid S. Miller end = start + frag_iter->len; 3518716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3519716ea3a7SDavid Howells if (copy > len) 3520716ea3a7SDavid Howells copy = len; 3521fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 352251c739d1SDavid S. Miller copy); 3523716ea3a7SDavid Howells if ((len -= copy) == 0) 3524716ea3a7SDavid Howells return elt; 3525716ea3a7SDavid Howells offset += copy; 3526716ea3a7SDavid Howells } 35271a028e50SDavid S. Miller start = end; 3528716ea3a7SDavid Howells } 3529716ea3a7SDavid Howells BUG_ON(len); 3530716ea3a7SDavid Howells return elt; 3531716ea3a7SDavid Howells } 3532716ea3a7SDavid Howells 353325a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 353425a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 353525a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 353625a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 353725a91d8dSFan Du * 353825a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 353925a91d8dSFan Du * 1. sg_init_table 354025a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 354125a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 354225a91d8dSFan Du * 354325a91d8dSFan Du * This is equivalent to: 354425a91d8dSFan Du * 1. sg_init_table 354525a91d8dSFan Du * 2. skb_to_sgvec(payload1) 354625a91d8dSFan Du * 3. sg_unmark_end 354725a91d8dSFan Du * 4. skb_to_sgvec(payload2) 354825a91d8dSFan Du * 354925a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 355025a91d8dSFan Du * is more preferable. 355125a91d8dSFan Du */ 355225a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 355325a91d8dSFan Du int offset, int len) 355425a91d8dSFan Du { 355525a91d8dSFan Du return __skb_to_sgvec(skb, sg, offset, len); 355625a91d8dSFan Du } 355725a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 355825a91d8dSFan Du 355951c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 356051c739d1SDavid S. Miller { 356151c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 356251c739d1SDavid S. Miller 3563c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 356451c739d1SDavid S. Miller 356551c739d1SDavid S. Miller return nsg; 356651c739d1SDavid S. Miller } 3567b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 356851c739d1SDavid S. Miller 3569716ea3a7SDavid Howells /** 3570716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 3571716ea3a7SDavid Howells * @skb: The socket buffer to check. 3572716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 3573716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 3574716ea3a7SDavid Howells * 3575716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 3576716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 3577716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 3578716ea3a7SDavid Howells * 3579716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 3580716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 3581716ea3a7SDavid Howells * set to point to the skb in which this space begins. 3582716ea3a7SDavid Howells * 3583716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3584716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3585716ea3a7SDavid Howells */ 3586716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3587716ea3a7SDavid Howells { 3588716ea3a7SDavid Howells int copyflag; 3589716ea3a7SDavid Howells int elt; 3590716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3591716ea3a7SDavid Howells 3592716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3593716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3594716ea3a7SDavid Howells * at the moment even if they are anonymous). 3595716ea3a7SDavid Howells */ 3596716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3597716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3598716ea3a7SDavid Howells return -ENOMEM; 3599716ea3a7SDavid Howells 3600716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 360121dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3602716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3603716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3604716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3605716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3606716ea3a7SDavid Howells 3607716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3608716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3609716ea3a7SDavid Howells return -ENOMEM; 3610716ea3a7SDavid Howells 3611716ea3a7SDavid Howells /* Voila! */ 3612716ea3a7SDavid Howells *trailer = skb; 3613716ea3a7SDavid Howells return 1; 3614716ea3a7SDavid Howells } 3615716ea3a7SDavid Howells 3616716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3617716ea3a7SDavid Howells 3618716ea3a7SDavid Howells elt = 1; 3619716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3620716ea3a7SDavid Howells copyflag = 0; 3621716ea3a7SDavid Howells 3622716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3623716ea3a7SDavid Howells int ntail = 0; 3624716ea3a7SDavid Howells 3625716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3626716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3627716ea3a7SDavid Howells * after it. */ 3628716ea3a7SDavid Howells 3629716ea3a7SDavid Howells if (skb_shared(skb1)) 3630716ea3a7SDavid Howells copyflag = 1; 3631716ea3a7SDavid Howells 3632716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3633716ea3a7SDavid Howells 3634716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3635716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 363621dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3637716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3638716ea3a7SDavid Howells ntail = tailbits + 128; 3639716ea3a7SDavid Howells } 3640716ea3a7SDavid Howells 3641716ea3a7SDavid Howells if (copyflag || 3642716ea3a7SDavid Howells skb_cloned(skb1) || 3643716ea3a7SDavid Howells ntail || 3644716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 364521dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3646716ea3a7SDavid Howells struct sk_buff *skb2; 3647716ea3a7SDavid Howells 3648716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3649716ea3a7SDavid Howells if (ntail == 0) 3650716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3651716ea3a7SDavid Howells else 3652716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3653716ea3a7SDavid Howells skb_headroom(skb1), 3654716ea3a7SDavid Howells ntail, 3655716ea3a7SDavid Howells GFP_ATOMIC); 3656716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3657716ea3a7SDavid Howells return -ENOMEM; 3658716ea3a7SDavid Howells 3659716ea3a7SDavid Howells if (skb1->sk) 3660716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3661716ea3a7SDavid Howells 3662716ea3a7SDavid Howells /* Looking around. Are we still alive? 3663716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3664716ea3a7SDavid Howells 3665716ea3a7SDavid Howells skb2->next = skb1->next; 3666716ea3a7SDavid Howells *skb_p = skb2; 3667716ea3a7SDavid Howells kfree_skb(skb1); 3668716ea3a7SDavid Howells skb1 = skb2; 3669716ea3a7SDavid Howells } 3670716ea3a7SDavid Howells elt++; 3671716ea3a7SDavid Howells *trailer = skb1; 3672716ea3a7SDavid Howells skb_p = &skb1->next; 3673716ea3a7SDavid Howells } 3674716ea3a7SDavid Howells 3675716ea3a7SDavid Howells return elt; 3676716ea3a7SDavid Howells } 3677b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3678716ea3a7SDavid Howells 3679b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3680b1faf566SEric Dumazet { 3681b1faf566SEric Dumazet struct sock *sk = skb->sk; 3682b1faf566SEric Dumazet 3683b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3684b1faf566SEric Dumazet } 3685b1faf566SEric Dumazet 3686b1faf566SEric Dumazet /* 3687b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3688b1faf566SEric Dumazet */ 3689b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3690b1faf566SEric Dumazet { 3691b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 369295c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 3693b1faf566SEric Dumazet return -ENOMEM; 3694b1faf566SEric Dumazet 3695b1faf566SEric Dumazet skb_orphan(skb); 3696b1faf566SEric Dumazet skb->sk = sk; 3697b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3698b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3699b1faf566SEric Dumazet 3700abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3701abb57ea4SEric Dumazet skb_dst_force(skb); 3702abb57ea4SEric Dumazet 3703b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3704b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3705676d2369SDavid S. Miller sk->sk_data_ready(sk); 3706b1faf566SEric Dumazet return 0; 3707b1faf566SEric Dumazet } 3708b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3709b1faf566SEric Dumazet 3710364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 3711364a9e93SWillem de Bruijn { 3712364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 3713364a9e93SWillem de Bruijn struct sk_buff *skb, *skb_next; 3714997d5c3fSEric Dumazet unsigned long flags; 3715364a9e93SWillem de Bruijn int err = 0; 3716364a9e93SWillem de Bruijn 3717997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 3718364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 3719364a9e93SWillem de Bruijn if (skb && (skb_next = skb_peek(q))) 3720364a9e93SWillem de Bruijn err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 3721997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 3722364a9e93SWillem de Bruijn 3723364a9e93SWillem de Bruijn sk->sk_err = err; 3724364a9e93SWillem de Bruijn if (err) 3725364a9e93SWillem de Bruijn sk->sk_error_report(sk); 3726364a9e93SWillem de Bruijn 3727364a9e93SWillem de Bruijn return skb; 3728364a9e93SWillem de Bruijn } 3729364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 3730364a9e93SWillem de Bruijn 3731cab41c47SAlexander Duyck /** 3732cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 3733cab41c47SAlexander Duyck * @skb: the skb to clone 3734cab41c47SAlexander Duyck * 3735cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 3736cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 3737cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 3738cab41c47SAlexander Duyck * 3739cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 3740cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 3741cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 3742cab41c47SAlexander Duyck * the sk_error_queue. 3743cab41c47SAlexander Duyck */ 374462bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 374562bccb8cSAlexander Duyck { 374662bccb8cSAlexander Duyck struct sock *sk = skb->sk; 374762bccb8cSAlexander Duyck struct sk_buff *clone; 374862bccb8cSAlexander Duyck 374962bccb8cSAlexander Duyck if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt)) 375062bccb8cSAlexander Duyck return NULL; 375162bccb8cSAlexander Duyck 375262bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 375362bccb8cSAlexander Duyck if (!clone) { 375462bccb8cSAlexander Duyck sock_put(sk); 375562bccb8cSAlexander Duyck return NULL; 375662bccb8cSAlexander Duyck } 375762bccb8cSAlexander Duyck 375862bccb8cSAlexander Duyck clone->sk = sk; 375962bccb8cSAlexander Duyck clone->destructor = sock_efree; 376062bccb8cSAlexander Duyck 376162bccb8cSAlexander Duyck return clone; 376262bccb8cSAlexander Duyck } 376362bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 376462bccb8cSAlexander Duyck 376537846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 376637846ef0SAlexander Duyck struct sock *sk, 376737846ef0SAlexander Duyck int tstype) 3768ac45f602SPatrick Ohly { 3769ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3770ac45f602SPatrick Ohly int err; 3771ac45f602SPatrick Ohly 3772ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3773ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3774ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3775ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 3776e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 37774ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 377809c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 3779ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 3780ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 37814ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 37824ed2d765SWillem de Bruijn } 378329030374SEric Dumazet 3784ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 378529030374SEric Dumazet 3786ac45f602SPatrick Ohly if (err) 3787ac45f602SPatrick Ohly kfree_skb(skb); 3788ac45f602SPatrick Ohly } 378937846ef0SAlexander Duyck 3790b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 3791b245be1fSWillem de Bruijn { 3792b245be1fSWillem de Bruijn bool ret; 3793b245be1fSWillem de Bruijn 3794b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 3795b245be1fSWillem de Bruijn return true; 3796b245be1fSWillem de Bruijn 3797b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 3798b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 3799b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 3800b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 3801b245be1fSWillem de Bruijn return ret; 3802b245be1fSWillem de Bruijn } 3803b245be1fSWillem de Bruijn 380437846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 380537846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 380637846ef0SAlexander Duyck { 380737846ef0SAlexander Duyck struct sock *sk = skb->sk; 380837846ef0SAlexander Duyck 3809b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 3810b245be1fSWillem de Bruijn return; 3811b245be1fSWillem de Bruijn 381262bccb8cSAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 381362bccb8cSAlexander Duyck sock_hold(sk); 381437846ef0SAlexander Duyck 381537846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 381637846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND); 381737846ef0SAlexander Duyck 381837846ef0SAlexander Duyck sock_put(sk); 381937846ef0SAlexander Duyck } 382037846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 382137846ef0SAlexander Duyck 382237846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 382337846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 382437846ef0SAlexander Duyck struct sock *sk, int tstype) 382537846ef0SAlexander Duyck { 382637846ef0SAlexander Duyck struct sk_buff *skb; 38273a8dd971SWillem de Bruijn bool tsonly; 382837846ef0SAlexander Duyck 38293a8dd971SWillem de Bruijn if (!sk) 38303a8dd971SWillem de Bruijn return; 38313a8dd971SWillem de Bruijn 38323a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 38333a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 383437846ef0SAlexander Duyck return; 383537846ef0SAlexander Duyck 383649ca0d8bSWillem de Bruijn if (tsonly) 383749ca0d8bSWillem de Bruijn skb = alloc_skb(0, GFP_ATOMIC); 383837846ef0SAlexander Duyck else 383937846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 384037846ef0SAlexander Duyck if (!skb) 384137846ef0SAlexander Duyck return; 384237846ef0SAlexander Duyck 384349ca0d8bSWillem de Bruijn if (tsonly) { 384449ca0d8bSWillem de Bruijn skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags; 384549ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 384649ca0d8bSWillem de Bruijn } 384749ca0d8bSWillem de Bruijn 384849ca0d8bSWillem de Bruijn if (hwtstamps) 384949ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 385049ca0d8bSWillem de Bruijn else 385149ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 385249ca0d8bSWillem de Bruijn 385337846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, tstype); 385437846ef0SAlexander Duyck } 3855e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 3856e7fd2885SWillem de Bruijn 3857e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 3858e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 3859e7fd2885SWillem de Bruijn { 3860e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 3861e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 3862e7fd2885SWillem de Bruijn } 3863ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3864ac45f602SPatrick Ohly 38656e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 38666e3e939fSJohannes Berg { 38676e3e939fSJohannes Berg struct sock *sk = skb->sk; 38686e3e939fSJohannes Berg struct sock_exterr_skb *serr; 38696e3e939fSJohannes Berg int err; 38706e3e939fSJohannes Berg 38716e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 38726e3e939fSJohannes Berg skb->wifi_acked = acked; 38736e3e939fSJohannes Berg 38746e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 38756e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 38766e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 38776e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 38786e3e939fSJohannes Berg 3879bf7fa551SAlexander Duyck /* take a reference to prevent skb_orphan() from freeing the socket */ 3880bf7fa551SAlexander Duyck sock_hold(sk); 3881bf7fa551SAlexander Duyck 38826e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 38836e3e939fSJohannes Berg if (err) 38846e3e939fSJohannes Berg kfree_skb(skb); 3885bf7fa551SAlexander Duyck 3886bf7fa551SAlexander Duyck sock_put(sk); 38876e3e939fSJohannes Berg } 38886e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 38896e3e939fSJohannes Berg 3890f35d9d8aSRusty Russell /** 3891f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3892f35d9d8aSRusty Russell * @skb: the skb to set 3893f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3894f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3895f35d9d8aSRusty Russell * 3896f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3897f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3898f35d9d8aSRusty Russell * 3899f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3900f35d9d8aSRusty Russell * returns false you should drop the packet. 3901f35d9d8aSRusty Russell */ 3902f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3903f35d9d8aSRusty Russell { 39045ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 39055ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3906e87cc472SJoe Perches net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 39075ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3908f35d9d8aSRusty Russell return false; 3909f35d9d8aSRusty Russell } 3910f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3911f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3912f35d9d8aSRusty Russell skb->csum_offset = off; 3913e5d5decaSJason Wang skb_set_transport_header(skb, start); 3914f35d9d8aSRusty Russell return true; 3915f35d9d8aSRusty Russell } 3916b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3917f35d9d8aSRusty Russell 3918ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 3919ed1f50c3SPaul Durrant unsigned int max) 3920ed1f50c3SPaul Durrant { 3921ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 3922ed1f50c3SPaul Durrant return 0; 3923ed1f50c3SPaul Durrant 3924ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 3925ed1f50c3SPaul Durrant * won't need to do it again. 3926ed1f50c3SPaul Durrant */ 3927ed1f50c3SPaul Durrant if (max > skb->len) 3928ed1f50c3SPaul Durrant max = skb->len; 3929ed1f50c3SPaul Durrant 3930ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 3931ed1f50c3SPaul Durrant return -ENOMEM; 3932ed1f50c3SPaul Durrant 3933ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 3934ed1f50c3SPaul Durrant return -EPROTO; 3935ed1f50c3SPaul Durrant 3936ed1f50c3SPaul Durrant return 0; 3937ed1f50c3SPaul Durrant } 3938ed1f50c3SPaul Durrant 3939f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 3940f9708b43SJan Beulich 3941f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 3942f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 3943f9708b43SJan Beulich unsigned int off) 3944f9708b43SJan Beulich { 3945f9708b43SJan Beulich switch (proto) { 3946f9708b43SJan Beulich int err; 3947f9708b43SJan Beulich 3948f9708b43SJan Beulich case IPPROTO_TCP: 3949f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 3950f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 3951f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3952f9708b43SJan Beulich offsetof(struct tcphdr, 3953f9708b43SJan Beulich check))) 3954f9708b43SJan Beulich err = -EPROTO; 3955f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 3956f9708b43SJan Beulich 3957f9708b43SJan Beulich case IPPROTO_UDP: 3958f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 3959f9708b43SJan Beulich off + sizeof(struct udphdr)); 3960f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3961f9708b43SJan Beulich offsetof(struct udphdr, 3962f9708b43SJan Beulich check))) 3963f9708b43SJan Beulich err = -EPROTO; 3964f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 3965f9708b43SJan Beulich } 3966f9708b43SJan Beulich 3967f9708b43SJan Beulich return ERR_PTR(-EPROTO); 3968f9708b43SJan Beulich } 3969f9708b43SJan Beulich 3970ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 3971ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 3972ed1f50c3SPaul Durrant */ 3973ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 3974ed1f50c3SPaul Durrant 3975f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 3976ed1f50c3SPaul Durrant { 3977ed1f50c3SPaul Durrant unsigned int off; 3978ed1f50c3SPaul Durrant bool fragment; 3979f9708b43SJan Beulich __sum16 *csum; 3980ed1f50c3SPaul Durrant int err; 3981ed1f50c3SPaul Durrant 3982ed1f50c3SPaul Durrant fragment = false; 3983ed1f50c3SPaul Durrant 3984ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 3985ed1f50c3SPaul Durrant sizeof(struct iphdr), 3986ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 3987ed1f50c3SPaul Durrant if (err < 0) 3988ed1f50c3SPaul Durrant goto out; 3989ed1f50c3SPaul Durrant 3990ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 3991ed1f50c3SPaul Durrant fragment = true; 3992ed1f50c3SPaul Durrant 3993ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 3994ed1f50c3SPaul Durrant 3995ed1f50c3SPaul Durrant err = -EPROTO; 3996ed1f50c3SPaul Durrant 3997ed1f50c3SPaul Durrant if (fragment) 3998ed1f50c3SPaul Durrant goto out; 3999ed1f50c3SPaul Durrant 4000f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4001f9708b43SJan Beulich if (IS_ERR(csum)) 4002f9708b43SJan Beulich return PTR_ERR(csum); 4003ed1f50c3SPaul Durrant 4004ed1f50c3SPaul Durrant if (recalculate) 4005f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4006ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4007ed1f50c3SPaul Durrant skb->len - off, 4008f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4009ed1f50c3SPaul Durrant err = 0; 4010ed1f50c3SPaul Durrant 4011ed1f50c3SPaul Durrant out: 4012ed1f50c3SPaul Durrant return err; 4013ed1f50c3SPaul Durrant } 4014ed1f50c3SPaul Durrant 4015ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4016ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4017ed1f50c3SPaul Durrant */ 4018ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4019ed1f50c3SPaul Durrant 4020ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4021ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4022ed1f50c3SPaul Durrant 4023ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4024ed1f50c3SPaul Durrant { 4025ed1f50c3SPaul Durrant int err; 4026ed1f50c3SPaul Durrant u8 nexthdr; 4027ed1f50c3SPaul Durrant unsigned int off; 4028ed1f50c3SPaul Durrant unsigned int len; 4029ed1f50c3SPaul Durrant bool fragment; 4030ed1f50c3SPaul Durrant bool done; 4031f9708b43SJan Beulich __sum16 *csum; 4032ed1f50c3SPaul Durrant 4033ed1f50c3SPaul Durrant fragment = false; 4034ed1f50c3SPaul Durrant done = false; 4035ed1f50c3SPaul Durrant 4036ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4037ed1f50c3SPaul Durrant 4038ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4039ed1f50c3SPaul Durrant if (err < 0) 4040ed1f50c3SPaul Durrant goto out; 4041ed1f50c3SPaul Durrant 4042ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4043ed1f50c3SPaul Durrant 4044ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4045ed1f50c3SPaul Durrant while (off <= len && !done) { 4046ed1f50c3SPaul Durrant switch (nexthdr) { 4047ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4048ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4049ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4050ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 4051ed1f50c3SPaul Durrant 4052ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4053ed1f50c3SPaul Durrant off + 4054ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 4055ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4056ed1f50c3SPaul Durrant if (err < 0) 4057ed1f50c3SPaul Durrant goto out; 4058ed1f50c3SPaul Durrant 4059ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4060ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4061ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 4062ed1f50c3SPaul Durrant break; 4063ed1f50c3SPaul Durrant } 4064ed1f50c3SPaul Durrant case IPPROTO_AH: { 4065ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 4066ed1f50c3SPaul Durrant 4067ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4068ed1f50c3SPaul Durrant off + 4069ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 4070ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4071ed1f50c3SPaul Durrant if (err < 0) 4072ed1f50c3SPaul Durrant goto out; 4073ed1f50c3SPaul Durrant 4074ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4075ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4076ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 4077ed1f50c3SPaul Durrant break; 4078ed1f50c3SPaul Durrant } 4079ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 4080ed1f50c3SPaul Durrant struct frag_hdr *hp; 4081ed1f50c3SPaul Durrant 4082ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4083ed1f50c3SPaul Durrant off + 4084ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 4085ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4086ed1f50c3SPaul Durrant if (err < 0) 4087ed1f50c3SPaul Durrant goto out; 4088ed1f50c3SPaul Durrant 4089ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 4090ed1f50c3SPaul Durrant 4091ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 4092ed1f50c3SPaul Durrant fragment = true; 4093ed1f50c3SPaul Durrant 4094ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4095ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 4096ed1f50c3SPaul Durrant break; 4097ed1f50c3SPaul Durrant } 4098ed1f50c3SPaul Durrant default: 4099ed1f50c3SPaul Durrant done = true; 4100ed1f50c3SPaul Durrant break; 4101ed1f50c3SPaul Durrant } 4102ed1f50c3SPaul Durrant } 4103ed1f50c3SPaul Durrant 4104ed1f50c3SPaul Durrant err = -EPROTO; 4105ed1f50c3SPaul Durrant 4106ed1f50c3SPaul Durrant if (!done || fragment) 4107ed1f50c3SPaul Durrant goto out; 4108ed1f50c3SPaul Durrant 4109f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 4110f9708b43SJan Beulich if (IS_ERR(csum)) 4111f9708b43SJan Beulich return PTR_ERR(csum); 4112ed1f50c3SPaul Durrant 4113ed1f50c3SPaul Durrant if (recalculate) 4114f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 4115ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 4116f9708b43SJan Beulich skb->len - off, nexthdr, 0); 4117ed1f50c3SPaul Durrant err = 0; 4118ed1f50c3SPaul Durrant 4119ed1f50c3SPaul Durrant out: 4120ed1f50c3SPaul Durrant return err; 4121ed1f50c3SPaul Durrant } 4122ed1f50c3SPaul Durrant 4123ed1f50c3SPaul Durrant /** 4124ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 4125ed1f50c3SPaul Durrant * @skb: the skb to set up 4126ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 4127ed1f50c3SPaul Durrant */ 4128ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 4129ed1f50c3SPaul Durrant { 4130ed1f50c3SPaul Durrant int err; 4131ed1f50c3SPaul Durrant 4132ed1f50c3SPaul Durrant switch (skb->protocol) { 4133ed1f50c3SPaul Durrant case htons(ETH_P_IP): 4134f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 4135ed1f50c3SPaul Durrant break; 4136ed1f50c3SPaul Durrant 4137ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 4138ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 4139ed1f50c3SPaul Durrant break; 4140ed1f50c3SPaul Durrant 4141ed1f50c3SPaul Durrant default: 4142ed1f50c3SPaul Durrant err = -EPROTO; 4143ed1f50c3SPaul Durrant break; 4144ed1f50c3SPaul Durrant } 4145ed1f50c3SPaul Durrant 4146ed1f50c3SPaul Durrant return err; 4147ed1f50c3SPaul Durrant } 4148ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 4149ed1f50c3SPaul Durrant 41509afd85c9SLinus Lüssing /** 41519afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 41529afd85c9SLinus Lüssing * @skb: the skb to check 41539afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 41549afd85c9SLinus Lüssing * 41559afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 41569afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 41579afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 41589afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 41599afd85c9SLinus Lüssing * 4160a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4161a516993fSLinus Lüssing * differs from the provided skb. 41629afd85c9SLinus Lüssing */ 41639afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 41649afd85c9SLinus Lüssing unsigned int transport_len) 41659afd85c9SLinus Lüssing { 41669afd85c9SLinus Lüssing struct sk_buff *skb_chk; 41679afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 41689afd85c9SLinus Lüssing int ret; 41699afd85c9SLinus Lüssing 4170a516993fSLinus Lüssing if (skb->len < len) 41719afd85c9SLinus Lüssing return NULL; 4172a516993fSLinus Lüssing else if (skb->len == len) 41739afd85c9SLinus Lüssing return skb; 41749afd85c9SLinus Lüssing 41759afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 41769afd85c9SLinus Lüssing if (!skb_chk) 41779afd85c9SLinus Lüssing return NULL; 41789afd85c9SLinus Lüssing 41799afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 41809afd85c9SLinus Lüssing if (ret) { 41819afd85c9SLinus Lüssing kfree_skb(skb_chk); 41829afd85c9SLinus Lüssing return NULL; 41839afd85c9SLinus Lüssing } 41849afd85c9SLinus Lüssing 41859afd85c9SLinus Lüssing return skb_chk; 41869afd85c9SLinus Lüssing } 41879afd85c9SLinus Lüssing 41889afd85c9SLinus Lüssing /** 41899afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 41909afd85c9SLinus Lüssing * @skb: the skb to check 41919afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 41929afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 41939afd85c9SLinus Lüssing * 41949afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 41959afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 41969afd85c9SLinus Lüssing * 41979afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 41989afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 41999afd85c9SLinus Lüssing * 4200a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4201a516993fSLinus Lüssing * differs from the provided skb. 42029afd85c9SLinus Lüssing */ 42039afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 42049afd85c9SLinus Lüssing unsigned int transport_len, 42059afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 42069afd85c9SLinus Lüssing { 42079afd85c9SLinus Lüssing struct sk_buff *skb_chk; 42089afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 4209fcba67c9SLinus Lüssing __sum16 ret; 42109afd85c9SLinus Lüssing 42119afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 42129afd85c9SLinus Lüssing if (!skb_chk) 4213a516993fSLinus Lüssing goto err; 42149afd85c9SLinus Lüssing 4215a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 4216a516993fSLinus Lüssing goto err; 42179afd85c9SLinus Lüssing 42189b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 42199afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 42209b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 42219afd85c9SLinus Lüssing 4222a516993fSLinus Lüssing if (ret) 4223a516993fSLinus Lüssing goto err; 42249afd85c9SLinus Lüssing 42259afd85c9SLinus Lüssing return skb_chk; 4226a516993fSLinus Lüssing 4227a516993fSLinus Lüssing err: 4228a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 4229a516993fSLinus Lüssing kfree_skb(skb_chk); 4230a516993fSLinus Lüssing 4231a516993fSLinus Lüssing return NULL; 4232a516993fSLinus Lüssing 42339afd85c9SLinus Lüssing } 42349afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 42359afd85c9SLinus Lüssing 42364497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 42374497b076SBen Hutchings { 4238e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 4239e87cc472SJoe Perches skb->dev->name); 42404497b076SBen Hutchings } 42414497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 4242bad43ca8SEric Dumazet 4243bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 4244bad43ca8SEric Dumazet { 42453d861f66SEric Dumazet if (head_stolen) { 42463d861f66SEric Dumazet skb_release_head_state(skb); 4247bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 42483d861f66SEric Dumazet } else { 4249bad43ca8SEric Dumazet __kfree_skb(skb); 4250bad43ca8SEric Dumazet } 42513d861f66SEric Dumazet } 4252bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 4253bad43ca8SEric Dumazet 4254bad43ca8SEric Dumazet /** 4255bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 4256bad43ca8SEric Dumazet * @to: prior buffer 4257bad43ca8SEric Dumazet * @from: buffer to add 4258bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 4259c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 4260bad43ca8SEric Dumazet */ 4261bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 4262bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 4263bad43ca8SEric Dumazet { 4264bad43ca8SEric Dumazet int i, delta, len = from->len; 4265bad43ca8SEric Dumazet 4266bad43ca8SEric Dumazet *fragstolen = false; 4267bad43ca8SEric Dumazet 4268bad43ca8SEric Dumazet if (skb_cloned(to)) 4269bad43ca8SEric Dumazet return false; 4270bad43ca8SEric Dumazet 4271bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 4272e93a0435SEric Dumazet if (len) 4273bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 4274bad43ca8SEric Dumazet *delta_truesize = 0; 4275bad43ca8SEric Dumazet return true; 4276bad43ca8SEric Dumazet } 4277bad43ca8SEric Dumazet 4278bad43ca8SEric Dumazet if (skb_has_frag_list(to) || skb_has_frag_list(from)) 4279bad43ca8SEric Dumazet return false; 4280bad43ca8SEric Dumazet 4281bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 4282bad43ca8SEric Dumazet struct page *page; 4283bad43ca8SEric Dumazet unsigned int offset; 4284bad43ca8SEric Dumazet 4285bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4286bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 4287bad43ca8SEric Dumazet return false; 4288bad43ca8SEric Dumazet 4289bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 4290bad43ca8SEric Dumazet return false; 4291bad43ca8SEric Dumazet 4292bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4293bad43ca8SEric Dumazet 4294bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 4295bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 4296bad43ca8SEric Dumazet 4297bad43ca8SEric Dumazet skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 4298bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 4299bad43ca8SEric Dumazet *fragstolen = true; 4300bad43ca8SEric Dumazet } else { 4301bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4302bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 4303bad43ca8SEric Dumazet return false; 4304bad43ca8SEric Dumazet 4305f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 4306bad43ca8SEric Dumazet } 4307bad43ca8SEric Dumazet 4308bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 4309bad43ca8SEric Dumazet 4310bad43ca8SEric Dumazet memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 4311bad43ca8SEric Dumazet skb_shinfo(from)->frags, 4312bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 4313bad43ca8SEric Dumazet skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 4314bad43ca8SEric Dumazet 4315bad43ca8SEric Dumazet if (!skb_cloned(from)) 4316bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags = 0; 4317bad43ca8SEric Dumazet 43188ea853fdSLi RongQing /* if the skb is not cloned this does nothing 43198ea853fdSLi RongQing * since we set nr_frags to 0. 43208ea853fdSLi RongQing */ 4321bad43ca8SEric Dumazet for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 4322bad43ca8SEric Dumazet skb_frag_ref(from, i); 4323bad43ca8SEric Dumazet 4324bad43ca8SEric Dumazet to->truesize += delta; 4325bad43ca8SEric Dumazet to->len += len; 4326bad43ca8SEric Dumazet to->data_len += len; 4327bad43ca8SEric Dumazet 4328bad43ca8SEric Dumazet *delta_truesize = delta; 4329bad43ca8SEric Dumazet return true; 4330bad43ca8SEric Dumazet } 4331bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 4332621e84d6SNicolas Dichtel 4333621e84d6SNicolas Dichtel /** 43348b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 4335621e84d6SNicolas Dichtel * 4336621e84d6SNicolas Dichtel * @skb: buffer to clean 43378b27f277SNicolas Dichtel * @xnet: packet is crossing netns 4338621e84d6SNicolas Dichtel * 43398b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 43408b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 43418b27f277SNicolas Dichtel * operations. 43428b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 43438b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 43448b27f277SNicolas Dichtel * skb that could impact namespace isolation. 4345621e84d6SNicolas Dichtel */ 43468b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4347621e84d6SNicolas Dichtel { 4348621e84d6SNicolas Dichtel skb->tstamp.tv64 = 0; 4349621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 4350621e84d6SNicolas Dichtel skb->skb_iif = 0; 435160ff7467SWANG Cong skb->ignore_df = 0; 4352621e84d6SNicolas Dichtel skb_dst_drop(skb); 4353621e84d6SNicolas Dichtel secpath_reset(skb); 4354621e84d6SNicolas Dichtel nf_reset(skb); 4355621e84d6SNicolas Dichtel nf_reset_trace(skb); 4356213dd74aSHerbert Xu 4357213dd74aSHerbert Xu if (!xnet) 4358213dd74aSHerbert Xu return; 4359213dd74aSHerbert Xu 4360213dd74aSHerbert Xu skb_orphan(skb); 4361213dd74aSHerbert Xu skb->mark = 0; 4362621e84d6SNicolas Dichtel } 4363621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 4364de960aa9SFlorian Westphal 4365de960aa9SFlorian Westphal /** 4366de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4367de960aa9SFlorian Westphal * 4368de960aa9SFlorian Westphal * @skb: GSO skb 4369de960aa9SFlorian Westphal * 4370de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 4371de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 4372de960aa9SFlorian Westphal * 4373de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4374de960aa9SFlorian Westphal */ 4375de960aa9SFlorian Westphal unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4376de960aa9SFlorian Westphal { 4377de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 4378f993bc25SFlorian Westphal unsigned int thlen = 0; 4379f993bc25SFlorian Westphal 4380f993bc25SFlorian Westphal if (skb->encapsulation) { 4381f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 4382f993bc25SFlorian Westphal skb_transport_header(skb); 4383de960aa9SFlorian Westphal 4384de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 4385f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 4386f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 4387f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 438890017accSMarcelo Ricardo Leitner } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) { 438990017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 4390f993bc25SFlorian Westphal } 43916d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 43926d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 43936d39d589SFlorian Westphal * accounted for. 43946d39d589SFlorian Westphal */ 4395f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 4396de960aa9SFlorian Westphal } 4397de960aa9SFlorian Westphal EXPORT_SYMBOL_GPL(skb_gso_transport_seglen); 43980d5501c1SVlad Yasevich 4399ae7ef81eSMarcelo Ricardo Leitner /** 4400ae7ef81eSMarcelo Ricardo Leitner * skb_gso_validate_mtu - Return in case such skb fits a given MTU 4401ae7ef81eSMarcelo Ricardo Leitner * 4402ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 440376f21b99SDavid S. Miller * @mtu: MTU to validate against 4404ae7ef81eSMarcelo Ricardo Leitner * 4405ae7ef81eSMarcelo Ricardo Leitner * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU 4406ae7ef81eSMarcelo Ricardo Leitner * once split. 4407ae7ef81eSMarcelo Ricardo Leitner */ 4408ae7ef81eSMarcelo Ricardo Leitner bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu) 4409ae7ef81eSMarcelo Ricardo Leitner { 4410ae7ef81eSMarcelo Ricardo Leitner const struct skb_shared_info *shinfo = skb_shinfo(skb); 4411ae7ef81eSMarcelo Ricardo Leitner const struct sk_buff *iter; 4412ae7ef81eSMarcelo Ricardo Leitner unsigned int hlen; 4413ae7ef81eSMarcelo Ricardo Leitner 4414ae7ef81eSMarcelo Ricardo Leitner hlen = skb_gso_network_seglen(skb); 4415ae7ef81eSMarcelo Ricardo Leitner 4416ae7ef81eSMarcelo Ricardo Leitner if (shinfo->gso_size != GSO_BY_FRAGS) 4417ae7ef81eSMarcelo Ricardo Leitner return hlen <= mtu; 4418ae7ef81eSMarcelo Ricardo Leitner 4419ae7ef81eSMarcelo Ricardo Leitner /* Undo this so we can re-use header sizes */ 4420ae7ef81eSMarcelo Ricardo Leitner hlen -= GSO_BY_FRAGS; 4421ae7ef81eSMarcelo Ricardo Leitner 4422ae7ef81eSMarcelo Ricardo Leitner skb_walk_frags(skb, iter) { 4423ae7ef81eSMarcelo Ricardo Leitner if (hlen + skb_headlen(iter) > mtu) 4424ae7ef81eSMarcelo Ricardo Leitner return false; 4425ae7ef81eSMarcelo Ricardo Leitner } 4426ae7ef81eSMarcelo Ricardo Leitner 4427ae7ef81eSMarcelo Ricardo Leitner return true; 4428ae7ef81eSMarcelo Ricardo Leitner } 4429ae7ef81eSMarcelo Ricardo Leitner EXPORT_SYMBOL_GPL(skb_gso_validate_mtu); 4430ae7ef81eSMarcelo Ricardo Leitner 44310d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 44320d5501c1SVlad Yasevich { 44330d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 44340d5501c1SVlad Yasevich kfree_skb(skb); 44350d5501c1SVlad Yasevich return NULL; 44360d5501c1SVlad Yasevich } 44370d5501c1SVlad Yasevich 4438f6548615SVlad Yasevich memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN, 4439a6e18ff1SVlad Yasevich 2 * ETH_ALEN); 44400d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 44410d5501c1SVlad Yasevich return skb; 44420d5501c1SVlad Yasevich } 44430d5501c1SVlad Yasevich 44440d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 44450d5501c1SVlad Yasevich { 44460d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 44470d5501c1SVlad Yasevich u16 vlan_tci; 44480d5501c1SVlad Yasevich 4449df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 44500d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 44510d5501c1SVlad Yasevich return skb; 44520d5501c1SVlad Yasevich } 44530d5501c1SVlad Yasevich 44540d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 44550d5501c1SVlad Yasevich if (unlikely(!skb)) 44560d5501c1SVlad Yasevich goto err_free; 44570d5501c1SVlad Yasevich 44580d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 44590d5501c1SVlad Yasevich goto err_free; 44600d5501c1SVlad Yasevich 44610d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 44620d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 44630d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 44640d5501c1SVlad Yasevich 44650d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 44660d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 44670d5501c1SVlad Yasevich 44680d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 44690d5501c1SVlad Yasevich if (unlikely(!skb)) 44700d5501c1SVlad Yasevich goto err_free; 44710d5501c1SVlad Yasevich 44720d5501c1SVlad Yasevich skb_reset_network_header(skb); 44730d5501c1SVlad Yasevich skb_reset_transport_header(skb); 44740d5501c1SVlad Yasevich skb_reset_mac_len(skb); 44750d5501c1SVlad Yasevich 44760d5501c1SVlad Yasevich return skb; 44770d5501c1SVlad Yasevich 44780d5501c1SVlad Yasevich err_free: 44790d5501c1SVlad Yasevich kfree_skb(skb); 44800d5501c1SVlad Yasevich return NULL; 44810d5501c1SVlad Yasevich } 44820d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 44832e4e4410SEric Dumazet 4484e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 4485e2195121SJiri Pirko { 4486e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 4487e2195121SJiri Pirko return -ENOMEM; 4488e2195121SJiri Pirko 4489e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 4490e2195121SJiri Pirko return 0; 4491e2195121SJiri Pirko 4492e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 4493e2195121SJiri Pirko } 4494e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 4495e2195121SJiri Pirko 449693515d53SJiri Pirko /* remove VLAN header from packet and update csum accordingly. */ 449793515d53SJiri Pirko static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 449893515d53SJiri Pirko { 449993515d53SJiri Pirko struct vlan_hdr *vhdr; 450093515d53SJiri Pirko unsigned int offset = skb->data - skb_mac_header(skb); 450193515d53SJiri Pirko int err; 450293515d53SJiri Pirko 450393515d53SJiri Pirko __skb_push(skb, offset); 450493515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 450593515d53SJiri Pirko if (unlikely(err)) 450693515d53SJiri Pirko goto pull; 450793515d53SJiri Pirko 450893515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 450993515d53SJiri Pirko 451093515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 451193515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 451293515d53SJiri Pirko 451393515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 451493515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 451593515d53SJiri Pirko 451693515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 451793515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 451893515d53SJiri Pirko 451993515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 452093515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 452193515d53SJiri Pirko 452293515d53SJiri Pirko skb_reset_mac_len(skb); 452393515d53SJiri Pirko pull: 452493515d53SJiri Pirko __skb_pull(skb, offset); 452593515d53SJiri Pirko 452693515d53SJiri Pirko return err; 452793515d53SJiri Pirko } 452893515d53SJiri Pirko 452993515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 453093515d53SJiri Pirko { 453193515d53SJiri Pirko u16 vlan_tci; 453293515d53SJiri Pirko __be16 vlan_proto; 453393515d53SJiri Pirko int err; 453493515d53SJiri Pirko 4535df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 453693515d53SJiri Pirko skb->vlan_tci = 0; 453793515d53SJiri Pirko } else { 453893515d53SJiri Pirko if (unlikely((skb->protocol != htons(ETH_P_8021Q) && 453993515d53SJiri Pirko skb->protocol != htons(ETH_P_8021AD)) || 454093515d53SJiri Pirko skb->len < VLAN_ETH_HLEN)) 454193515d53SJiri Pirko return 0; 454293515d53SJiri Pirko 454393515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 454493515d53SJiri Pirko if (err) 454593515d53SJiri Pirko return err; 454693515d53SJiri Pirko } 454793515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 454893515d53SJiri Pirko if (likely((skb->protocol != htons(ETH_P_8021Q) && 454993515d53SJiri Pirko skb->protocol != htons(ETH_P_8021AD)) || 455093515d53SJiri Pirko skb->len < VLAN_ETH_HLEN)) 455193515d53SJiri Pirko return 0; 455293515d53SJiri Pirko 455393515d53SJiri Pirko vlan_proto = skb->protocol; 455493515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 455593515d53SJiri Pirko if (unlikely(err)) 455693515d53SJiri Pirko return err; 455793515d53SJiri Pirko 455893515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 455993515d53SJiri Pirko return 0; 456093515d53SJiri Pirko } 456193515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 456293515d53SJiri Pirko 456393515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 456493515d53SJiri Pirko { 4565df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 456693515d53SJiri Pirko unsigned int offset = skb->data - skb_mac_header(skb); 456793515d53SJiri Pirko int err; 456893515d53SJiri Pirko 456993515d53SJiri Pirko /* __vlan_insert_tag expect skb->data pointing to mac header. 457093515d53SJiri Pirko * So change skb->data before calling it and change back to 457193515d53SJiri Pirko * original position later 457293515d53SJiri Pirko */ 457393515d53SJiri Pirko __skb_push(skb, offset); 457493515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 4575df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 45769241e2dfSDaniel Borkmann if (err) { 45779241e2dfSDaniel Borkmann __skb_pull(skb, offset); 457893515d53SJiri Pirko return err; 45799241e2dfSDaniel Borkmann } 45809241e2dfSDaniel Borkmann 458193515d53SJiri Pirko skb->protocol = skb->vlan_proto; 458293515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 458393515d53SJiri Pirko 45846b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 45859241e2dfSDaniel Borkmann __skb_pull(skb, offset); 458693515d53SJiri Pirko } 458793515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 458893515d53SJiri Pirko return 0; 458993515d53SJiri Pirko } 459093515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 459193515d53SJiri Pirko 45922e4e4410SEric Dumazet /** 45932e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 45942e4e4410SEric Dumazet * 4595de3f0d0eSMasanari Iida * @header_len: size of linear part 4596de3f0d0eSMasanari Iida * @data_len: needed length in frags 4597de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 4598de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 4599de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 46002e4e4410SEric Dumazet * 46012e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 46022e4e4410SEric Dumazet */ 46032e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 46042e4e4410SEric Dumazet unsigned long data_len, 46052e4e4410SEric Dumazet int max_page_order, 46062e4e4410SEric Dumazet int *errcode, 46072e4e4410SEric Dumazet gfp_t gfp_mask) 46082e4e4410SEric Dumazet { 46092e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 46102e4e4410SEric Dumazet unsigned long chunk; 46112e4e4410SEric Dumazet struct sk_buff *skb; 46122e4e4410SEric Dumazet struct page *page; 46132e4e4410SEric Dumazet gfp_t gfp_head; 46142e4e4410SEric Dumazet int i; 46152e4e4410SEric Dumazet 46162e4e4410SEric Dumazet *errcode = -EMSGSIZE; 46172e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 46182e4e4410SEric Dumazet * high order pages... 46192e4e4410SEric Dumazet */ 46202e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 46212e4e4410SEric Dumazet return NULL; 46222e4e4410SEric Dumazet 46232e4e4410SEric Dumazet gfp_head = gfp_mask; 4624d0164adcSMel Gorman if (gfp_head & __GFP_DIRECT_RECLAIM) 46252e4e4410SEric Dumazet gfp_head |= __GFP_REPEAT; 46262e4e4410SEric Dumazet 46272e4e4410SEric Dumazet *errcode = -ENOBUFS; 46282e4e4410SEric Dumazet skb = alloc_skb(header_len, gfp_head); 46292e4e4410SEric Dumazet if (!skb) 46302e4e4410SEric Dumazet return NULL; 46312e4e4410SEric Dumazet 46322e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 46332e4e4410SEric Dumazet 46342e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 46352e4e4410SEric Dumazet int order = max_page_order; 46362e4e4410SEric Dumazet 46372e4e4410SEric Dumazet while (order) { 46382e4e4410SEric Dumazet if (npages >= 1 << order) { 4639d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 46402e4e4410SEric Dumazet __GFP_COMP | 46412e4e4410SEric Dumazet __GFP_NOWARN | 46422e4e4410SEric Dumazet __GFP_NORETRY, 46432e4e4410SEric Dumazet order); 46442e4e4410SEric Dumazet if (page) 46452e4e4410SEric Dumazet goto fill_page; 46462e4e4410SEric Dumazet /* Do not retry other high order allocations */ 46472e4e4410SEric Dumazet order = 1; 46482e4e4410SEric Dumazet max_page_order = 0; 46492e4e4410SEric Dumazet } 46502e4e4410SEric Dumazet order--; 46512e4e4410SEric Dumazet } 46522e4e4410SEric Dumazet page = alloc_page(gfp_mask); 46532e4e4410SEric Dumazet if (!page) 46542e4e4410SEric Dumazet goto failure; 46552e4e4410SEric Dumazet fill_page: 46562e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 46572e4e4410SEric Dumazet PAGE_SIZE << order); 46582e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 46592e4e4410SEric Dumazet data_len -= chunk; 46602e4e4410SEric Dumazet npages -= 1 << order; 46612e4e4410SEric Dumazet } 46622e4e4410SEric Dumazet return skb; 46632e4e4410SEric Dumazet 46642e4e4410SEric Dumazet failure: 46652e4e4410SEric Dumazet kfree_skb(skb); 46662e4e4410SEric Dumazet return NULL; 46672e4e4410SEric Dumazet } 46682e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 46696fa01ccdSSowmini Varadhan 46706fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 46716fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 46726fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 46736fa01ccdSSowmini Varadhan { 46746fa01ccdSSowmini Varadhan int i; 46756fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 46766fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 46776fa01ccdSSowmini Varadhan u8 *data; 46786fa01ccdSSowmini Varadhan 46796fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 46806fa01ccdSSowmini Varadhan 46816fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 46826fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 46836fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 46846fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 46856fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 46866fa01ccdSSowmini Varadhan if (!data) 46876fa01ccdSSowmini Varadhan return -ENOMEM; 46886fa01ccdSSowmini Varadhan 46896fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 46906fa01ccdSSowmini Varadhan 46916fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 46926fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 46936fa01ccdSSowmini Varadhan skb->len -= off; 46946fa01ccdSSowmini Varadhan 46956fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 46966fa01ccdSSowmini Varadhan skb_shinfo(skb), 46976fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 46986fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 46996fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 47006fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 47016fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 47026fa01ccdSSowmini Varadhan kfree(data); 47036fa01ccdSSowmini Varadhan return -ENOMEM; 47046fa01ccdSSowmini Varadhan } 47056fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 47066fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 47076fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 47086fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 47096fa01ccdSSowmini Varadhan skb_release_data(skb); 47106fa01ccdSSowmini Varadhan } else { 47116fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 47126fa01ccdSSowmini Varadhan * relocate values 47136fa01ccdSSowmini Varadhan */ 47146fa01ccdSSowmini Varadhan skb_free_head(skb); 47156fa01ccdSSowmini Varadhan } 47166fa01ccdSSowmini Varadhan 47176fa01ccdSSowmini Varadhan skb->head = data; 47186fa01ccdSSowmini Varadhan skb->data = data; 47196fa01ccdSSowmini Varadhan skb->head_frag = 0; 47206fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 47216fa01ccdSSowmini Varadhan skb->end = size; 47226fa01ccdSSowmini Varadhan #else 47236fa01ccdSSowmini Varadhan skb->end = skb->head + size; 47246fa01ccdSSowmini Varadhan #endif 47256fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 47266fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 47276fa01ccdSSowmini Varadhan skb->cloned = 0; 47286fa01ccdSSowmini Varadhan skb->hdr_len = 0; 47296fa01ccdSSowmini Varadhan skb->nohdr = 0; 47306fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 47316fa01ccdSSowmini Varadhan 47326fa01ccdSSowmini Varadhan return 0; 47336fa01ccdSSowmini Varadhan } 47346fa01ccdSSowmini Varadhan 47356fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 47366fa01ccdSSowmini Varadhan 47376fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 47386fa01ccdSSowmini Varadhan * pskb_carve() 47396fa01ccdSSowmini Varadhan */ 47406fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 47416fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 47426fa01ccdSSowmini Varadhan gfp_t gfp_mask) 47436fa01ccdSSowmini Varadhan { 47446fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 47456fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 47466fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 47476fa01ccdSSowmini Varadhan 47486fa01ccdSSowmini Varadhan do { 47496fa01ccdSSowmini Varadhan if (!list) { 47506fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 47516fa01ccdSSowmini Varadhan return -EFAULT; 47526fa01ccdSSowmini Varadhan } 47536fa01ccdSSowmini Varadhan if (list->len <= eat) { 47546fa01ccdSSowmini Varadhan /* Eaten as whole. */ 47556fa01ccdSSowmini Varadhan eat -= list->len; 47566fa01ccdSSowmini Varadhan list = list->next; 47576fa01ccdSSowmini Varadhan insp = list; 47586fa01ccdSSowmini Varadhan } else { 47596fa01ccdSSowmini Varadhan /* Eaten partially. */ 47606fa01ccdSSowmini Varadhan if (skb_shared(list)) { 47616fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 47626fa01ccdSSowmini Varadhan if (!clone) 47636fa01ccdSSowmini Varadhan return -ENOMEM; 47646fa01ccdSSowmini Varadhan insp = list->next; 47656fa01ccdSSowmini Varadhan list = clone; 47666fa01ccdSSowmini Varadhan } else { 47676fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 47686fa01ccdSSowmini Varadhan insp = list; 47696fa01ccdSSowmini Varadhan } 47706fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 47716fa01ccdSSowmini Varadhan kfree_skb(clone); 47726fa01ccdSSowmini Varadhan return -ENOMEM; 47736fa01ccdSSowmini Varadhan } 47746fa01ccdSSowmini Varadhan break; 47756fa01ccdSSowmini Varadhan } 47766fa01ccdSSowmini Varadhan } while (eat); 47776fa01ccdSSowmini Varadhan 47786fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 47796fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 47806fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 47816fa01ccdSSowmini Varadhan kfree_skb(list); 47826fa01ccdSSowmini Varadhan } 47836fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 47846fa01ccdSSowmini Varadhan if (clone) { 47856fa01ccdSSowmini Varadhan clone->next = list; 47866fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 47876fa01ccdSSowmini Varadhan } 47886fa01ccdSSowmini Varadhan return 0; 47896fa01ccdSSowmini Varadhan } 47906fa01ccdSSowmini Varadhan 47916fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 47926fa01ccdSSowmini Varadhan * non-linear part of skb 47936fa01ccdSSowmini Varadhan */ 47946fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 47956fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 47966fa01ccdSSowmini Varadhan { 47976fa01ccdSSowmini Varadhan int i, k = 0; 47986fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 47996fa01ccdSSowmini Varadhan u8 *data; 48006fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 48016fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 48026fa01ccdSSowmini Varadhan 48036fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 48046fa01ccdSSowmini Varadhan 48056fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 48066fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 48076fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 48086fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 48096fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 48106fa01ccdSSowmini Varadhan if (!data) 48116fa01ccdSSowmini Varadhan return -ENOMEM; 48126fa01ccdSSowmini Varadhan 48136fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 48146fa01ccdSSowmini Varadhan 48156fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 48166fa01ccdSSowmini Varadhan skb_shinfo(skb), offsetof(struct skb_shared_info, 48176fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 48186fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 48196fa01ccdSSowmini Varadhan kfree(data); 48206fa01ccdSSowmini Varadhan return -ENOMEM; 48216fa01ccdSSowmini Varadhan } 48226fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 48236fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 48246fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 48256fa01ccdSSowmini Varadhan 48266fa01ccdSSowmini Varadhan if (pos + fsize > off) { 48276fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 48286fa01ccdSSowmini Varadhan 48296fa01ccdSSowmini Varadhan if (pos < off) { 48306fa01ccdSSowmini Varadhan /* Split frag. 48316fa01ccdSSowmini Varadhan * We have two variants in this case: 48326fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 48336fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 48346fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 48356fa01ccdSSowmini Varadhan * where splitting is expensive. 48366fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 48376fa01ccdSSowmini Varadhan */ 48386fa01ccdSSowmini Varadhan shinfo->frags[0].page_offset += off - pos; 48396fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 48406fa01ccdSSowmini Varadhan } 48416fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 48426fa01ccdSSowmini Varadhan k++; 48436fa01ccdSSowmini Varadhan } 48446fa01ccdSSowmini Varadhan pos += fsize; 48456fa01ccdSSowmini Varadhan } 48466fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 48476fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 48486fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 48496fa01ccdSSowmini Varadhan 48506fa01ccdSSowmini Varadhan if (k == 0) { 48516fa01ccdSSowmini Varadhan /* split line is in frag list */ 48526fa01ccdSSowmini Varadhan pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask); 48536fa01ccdSSowmini Varadhan } 48546fa01ccdSSowmini Varadhan skb_release_data(skb); 48556fa01ccdSSowmini Varadhan 48566fa01ccdSSowmini Varadhan skb->head = data; 48576fa01ccdSSowmini Varadhan skb->head_frag = 0; 48586fa01ccdSSowmini Varadhan skb->data = data; 48596fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 48606fa01ccdSSowmini Varadhan skb->end = size; 48616fa01ccdSSowmini Varadhan #else 48626fa01ccdSSowmini Varadhan skb->end = skb->head + size; 48636fa01ccdSSowmini Varadhan #endif 48646fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 48656fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 48666fa01ccdSSowmini Varadhan skb->cloned = 0; 48676fa01ccdSSowmini Varadhan skb->hdr_len = 0; 48686fa01ccdSSowmini Varadhan skb->nohdr = 0; 48696fa01ccdSSowmini Varadhan skb->len -= off; 48706fa01ccdSSowmini Varadhan skb->data_len = skb->len; 48716fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 48726fa01ccdSSowmini Varadhan return 0; 48736fa01ccdSSowmini Varadhan } 48746fa01ccdSSowmini Varadhan 48756fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 48766fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 48776fa01ccdSSowmini Varadhan { 48786fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 48796fa01ccdSSowmini Varadhan 48806fa01ccdSSowmini Varadhan if (len < headlen) 48816fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 48826fa01ccdSSowmini Varadhan else 48836fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 48846fa01ccdSSowmini Varadhan } 48856fa01ccdSSowmini Varadhan 48866fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 48876fa01ccdSSowmini Varadhan * a new skb 48886fa01ccdSSowmini Varadhan */ 48896fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 48906fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 48916fa01ccdSSowmini Varadhan { 48926fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 48936fa01ccdSSowmini Varadhan 48946fa01ccdSSowmini Varadhan if (!clone) 48956fa01ccdSSowmini Varadhan return NULL; 48966fa01ccdSSowmini Varadhan 48976fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 48986fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 48996fa01ccdSSowmini Varadhan kfree_skb(clone); 49006fa01ccdSSowmini Varadhan return NULL; 49016fa01ccdSSowmini Varadhan } 49026fa01ccdSSowmini Varadhan return clone; 49036fa01ccdSSowmini Varadhan } 49046fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 4905