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 757c0f6ba6SLinus Torvalds #include <linux/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; 274d179cd12SDavid S. Miller } 2751da177e4SLinus Torvalds out: 2761da177e4SLinus Torvalds return skb; 2771da177e4SLinus Torvalds nodata: 2788798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2791da177e4SLinus Torvalds skb = NULL; 2801da177e4SLinus Torvalds goto out; 2811da177e4SLinus Torvalds } 282b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2831da177e4SLinus Torvalds 2841da177e4SLinus Torvalds /** 2852ea2f62cSEric Dumazet * __build_skb - build a network buffer 286b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2872ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 288b2b5ce9dSEric Dumazet * 289b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 290deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2912ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2922ea2f62cSEric Dumazet * or vmalloc() 293b2b5ce9dSEric Dumazet * The return is the new skb buffer. 294b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 295b2b5ce9dSEric Dumazet * Notes : 296b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 297b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 298b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 299b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 300b2b5ce9dSEric Dumazet * before giving packet to stack. 301b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 302b2b5ce9dSEric Dumazet */ 3032ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 304b2b5ce9dSEric Dumazet { 305b2b5ce9dSEric Dumazet struct skb_shared_info *shinfo; 306b2b5ce9dSEric Dumazet struct sk_buff *skb; 307d3836f21SEric Dumazet unsigned int size = frag_size ? : ksize(data); 308b2b5ce9dSEric Dumazet 309b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 310b2b5ce9dSEric Dumazet if (!skb) 311b2b5ce9dSEric Dumazet return NULL; 312b2b5ce9dSEric Dumazet 313d3836f21SEric Dumazet size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 314b2b5ce9dSEric Dumazet 315b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 316b2b5ce9dSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 317b2b5ce9dSEric Dumazet atomic_set(&skb->users, 1); 318b2b5ce9dSEric Dumazet skb->head = data; 319b2b5ce9dSEric Dumazet skb->data = data; 320b2b5ce9dSEric Dumazet skb_reset_tail_pointer(skb); 321b2b5ce9dSEric Dumazet skb->end = skb->tail + size; 32235d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 32335d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 324b2b5ce9dSEric Dumazet 325b2b5ce9dSEric Dumazet /* make sure we initialize shinfo sequentially */ 326b2b5ce9dSEric Dumazet shinfo = skb_shinfo(skb); 327b2b5ce9dSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 328b2b5ce9dSEric Dumazet atomic_set(&shinfo->dataref, 1); 329b2b5ce9dSEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 330b2b5ce9dSEric Dumazet 331b2b5ce9dSEric Dumazet return skb; 332b2b5ce9dSEric Dumazet } 3332ea2f62cSEric Dumazet 3342ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3352ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3362ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 3372ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 3382ea2f62cSEric Dumazet */ 3392ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3402ea2f62cSEric Dumazet { 3412ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3422ea2f62cSEric Dumazet 3432ea2f62cSEric Dumazet if (skb && frag_size) { 3442ea2f62cSEric Dumazet skb->head_frag = 1; 3452f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 3462ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3472ea2f62cSEric Dumazet } 3482ea2f62cSEric Dumazet return skb; 3492ea2f62cSEric Dumazet } 350b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 351b2b5ce9dSEric Dumazet 352795bb1c0SJesper Dangaard Brouer #define NAPI_SKB_CACHE_SIZE 64 353795bb1c0SJesper Dangaard Brouer 354795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache { 355795bb1c0SJesper Dangaard Brouer struct page_frag_cache page; 356e0d7924aSAlexey Dobriyan unsigned int skb_count; 357795bb1c0SJesper Dangaard Brouer void *skb_cache[NAPI_SKB_CACHE_SIZE]; 358795bb1c0SJesper Dangaard Brouer }; 359795bb1c0SJesper Dangaard Brouer 360b63ae8caSAlexander Duyck static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 361795bb1c0SJesper Dangaard Brouer static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 362ffde7328SAlexander Duyck 363ffde7328SAlexander Duyck static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 364ffde7328SAlexander Duyck { 365b63ae8caSAlexander Duyck struct page_frag_cache *nc; 366ffde7328SAlexander Duyck unsigned long flags; 367ffde7328SAlexander Duyck void *data; 368ffde7328SAlexander Duyck 369ffde7328SAlexander Duyck local_irq_save(flags); 3709451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 3718c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, fragsz, gfp_mask); 3726f532612SEric Dumazet local_irq_restore(flags); 3736f532612SEric Dumazet return data; 3746f532612SEric Dumazet } 375c93bdd0eSMel Gorman 376c93bdd0eSMel Gorman /** 377c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 378c93bdd0eSMel Gorman * @fragsz: fragment size 379c93bdd0eSMel Gorman * 380c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 381c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 382c93bdd0eSMel Gorman */ 383c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 384c93bdd0eSMel Gorman { 385c93bdd0eSMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 386c93bdd0eSMel Gorman } 3876f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 3886f532612SEric Dumazet 389ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 390ffde7328SAlexander Duyck { 391795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 3929451980aSAlexander Duyck 3938c2dd3e4SAlexander Duyck return page_frag_alloc(&nc->page, fragsz, gfp_mask); 394ffde7328SAlexander Duyck } 395ffde7328SAlexander Duyck 396ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 397ffde7328SAlexander Duyck { 398ffde7328SAlexander Duyck return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 399ffde7328SAlexander Duyck } 400ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 401ffde7328SAlexander Duyck 4026f532612SEric Dumazet /** 403fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 404fd11a83dSAlexander Duyck * @dev: network device to receive on 405d7499160SMasanari Iida * @len: length to allocate 406fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 407fd11a83dSAlexander Duyck * 408fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 409fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 410fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 411fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 412fd11a83dSAlexander Duyck * 413fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 414fd11a83dSAlexander Duyck */ 4159451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4169451980aSAlexander Duyck gfp_t gfp_mask) 417fd11a83dSAlexander Duyck { 418b63ae8caSAlexander Duyck struct page_frag_cache *nc; 4199451980aSAlexander Duyck unsigned long flags; 420fd11a83dSAlexander Duyck struct sk_buff *skb; 4219451980aSAlexander Duyck bool pfmemalloc; 4229451980aSAlexander Duyck void *data; 423fd11a83dSAlexander Duyck 4249451980aSAlexander Duyck len += NET_SKB_PAD; 425fd11a83dSAlexander Duyck 4269451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 427d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 428a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 429a080e7bdSAlexander Duyck if (!skb) 430a080e7bdSAlexander Duyck goto skb_fail; 431a080e7bdSAlexander Duyck goto skb_success; 432a080e7bdSAlexander Duyck } 4339451980aSAlexander Duyck 4349451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4359451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 4369451980aSAlexander Duyck 4379451980aSAlexander Duyck if (sk_memalloc_socks()) 4389451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 4399451980aSAlexander Duyck 4409451980aSAlexander Duyck local_irq_save(flags); 4419451980aSAlexander Duyck 4429451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 4438c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 4449451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 4459451980aSAlexander Duyck 4469451980aSAlexander Duyck local_irq_restore(flags); 4479451980aSAlexander Duyck 4489451980aSAlexander Duyck if (unlikely(!data)) 4499451980aSAlexander Duyck return NULL; 4509451980aSAlexander Duyck 4519451980aSAlexander Duyck skb = __build_skb(data, len); 4529451980aSAlexander Duyck if (unlikely(!skb)) { 453181edb2bSAlexander Duyck skb_free_frag(data); 4549451980aSAlexander Duyck return NULL; 4559451980aSAlexander Duyck } 4569451980aSAlexander Duyck 4579451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 4589451980aSAlexander Duyck if (pfmemalloc) 4599451980aSAlexander Duyck skb->pfmemalloc = 1; 4609451980aSAlexander Duyck skb->head_frag = 1; 4619451980aSAlexander Duyck 462a080e7bdSAlexander Duyck skb_success: 4638af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4647b2e497aSChristoph Hellwig skb->dev = dev; 465fd11a83dSAlexander Duyck 466a080e7bdSAlexander Duyck skb_fail: 4678af27456SChristoph Hellwig return skb; 4688af27456SChristoph Hellwig } 469b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4701da177e4SLinus Torvalds 471fd11a83dSAlexander Duyck /** 472fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 473fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 474d7499160SMasanari Iida * @len: length to allocate 475fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 476fd11a83dSAlexander Duyck * 477fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 478fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 479fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 480fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 481fd11a83dSAlexander Duyck * 482fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 483fd11a83dSAlexander Duyck */ 4849451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 4859451980aSAlexander Duyck gfp_t gfp_mask) 486fd11a83dSAlexander Duyck { 487795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 488fd11a83dSAlexander Duyck struct sk_buff *skb; 4899451980aSAlexander Duyck void *data; 490fd11a83dSAlexander Duyck 4919451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 492fd11a83dSAlexander Duyck 4939451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 494d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 495a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 496a080e7bdSAlexander Duyck if (!skb) 497a080e7bdSAlexander Duyck goto skb_fail; 498a080e7bdSAlexander Duyck goto skb_success; 499a080e7bdSAlexander Duyck } 5009451980aSAlexander Duyck 5019451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5029451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5039451980aSAlexander Duyck 5049451980aSAlexander Duyck if (sk_memalloc_socks()) 5059451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5069451980aSAlexander Duyck 5078c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5089451980aSAlexander Duyck if (unlikely(!data)) 5099451980aSAlexander Duyck return NULL; 5109451980aSAlexander Duyck 5119451980aSAlexander Duyck skb = __build_skb(data, len); 5129451980aSAlexander Duyck if (unlikely(!skb)) { 513181edb2bSAlexander Duyck skb_free_frag(data); 5149451980aSAlexander Duyck return NULL; 5159451980aSAlexander Duyck } 5169451980aSAlexander Duyck 5179451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 518795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5199451980aSAlexander Duyck skb->pfmemalloc = 1; 5209451980aSAlexander Duyck skb->head_frag = 1; 5219451980aSAlexander Duyck 522a080e7bdSAlexander Duyck skb_success: 523fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 524fd11a83dSAlexander Duyck skb->dev = napi->dev; 525fd11a83dSAlexander Duyck 526a080e7bdSAlexander Duyck skb_fail: 527fd11a83dSAlexander Duyck return skb; 528fd11a83dSAlexander Duyck } 529fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 530fd11a83dSAlexander Duyck 531654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 53250269e19SEric Dumazet int size, unsigned int truesize) 533654bed16SPeter Zijlstra { 534654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 535654bed16SPeter Zijlstra skb->len += size; 536654bed16SPeter Zijlstra skb->data_len += size; 53750269e19SEric Dumazet skb->truesize += truesize; 538654bed16SPeter Zijlstra } 539654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 540654bed16SPeter Zijlstra 541f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 542f8e617e1SJason Wang unsigned int truesize) 543f8e617e1SJason Wang { 544f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 545f8e617e1SJason Wang 546f8e617e1SJason Wang skb_frag_size_add(frag, size); 547f8e617e1SJason Wang skb->len += size; 548f8e617e1SJason Wang skb->data_len += size; 549f8e617e1SJason Wang skb->truesize += truesize; 550f8e617e1SJason Wang } 551f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 552f8e617e1SJason Wang 55327b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5541da177e4SLinus Torvalds { 555bd8a7036SEric Dumazet kfree_skb_list(*listp); 55627b437c8SHerbert Xu *listp = NULL; 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds 55927b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 56027b437c8SHerbert Xu { 56127b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 56227b437c8SHerbert Xu } 56327b437c8SHerbert Xu 5641da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5651da177e4SLinus Torvalds { 5661da177e4SLinus Torvalds struct sk_buff *list; 5671da177e4SLinus Torvalds 568fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5691da177e4SLinus Torvalds skb_get(list); 5701da177e4SLinus Torvalds } 5711da177e4SLinus Torvalds 572d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 573d3836f21SEric Dumazet { 574181edb2bSAlexander Duyck unsigned char *head = skb->head; 575181edb2bSAlexander Duyck 576d3836f21SEric Dumazet if (skb->head_frag) 577181edb2bSAlexander Duyck skb_free_frag(head); 578d3836f21SEric Dumazet else 579181edb2bSAlexander Duyck kfree(head); 580d3836f21SEric Dumazet } 581d3836f21SEric Dumazet 5825bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5831da177e4SLinus Torvalds { 584ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 5851da177e4SLinus Torvalds int i; 586ff04a771SEric Dumazet 587ff04a771SEric Dumazet if (skb->cloned && 588ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 589ff04a771SEric Dumazet &shinfo->dataref)) 590ff04a771SEric Dumazet return; 591ff04a771SEric Dumazet 592ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 593ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 5941da177e4SLinus Torvalds 595a6686f2fSShirley Ma /* 596a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 597a6686f2fSShirley Ma * the lower device DMA has done; 598a6686f2fSShirley Ma */ 599ff04a771SEric Dumazet if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) { 600a6686f2fSShirley Ma struct ubuf_info *uarg; 601a6686f2fSShirley Ma 602ff04a771SEric Dumazet uarg = shinfo->destructor_arg; 603a6686f2fSShirley Ma if (uarg->callback) 604e19d6763SMichael S. Tsirkin uarg->callback(uarg, true); 605a6686f2fSShirley Ma } 606a6686f2fSShirley Ma 607ff04a771SEric Dumazet if (shinfo->frag_list) 608ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6091da177e4SLinus Torvalds 610d3836f21SEric Dumazet skb_free_head(skb); 6111da177e4SLinus Torvalds } 6121da177e4SLinus Torvalds 6131da177e4SLinus Torvalds /* 6141da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6151da177e4SLinus Torvalds */ 6162d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6171da177e4SLinus Torvalds { 618d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 619d179cd12SDavid S. Miller 620d179cd12SDavid S. Miller switch (skb->fclone) { 621d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6221da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6236ffe75ebSEric Dumazet return; 624d179cd12SDavid S. Miller 625d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 626d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6276ffe75ebSEric Dumazet 6286ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6296ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6306ffe75ebSEric Dumazet * while here, branch prediction will be good. 6316ffe75ebSEric Dumazet */ 6326ffe75ebSEric Dumazet if (atomic_read(&fclones->fclone_ref) == 1) 6336ffe75ebSEric Dumazet goto fastpath; 634d179cd12SDavid S. Miller break; 635d179cd12SDavid S. Miller 6366ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 637d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 638d179cd12SDavid S. Miller break; 6393ff50b79SStephen Hemminger } 6406ffe75ebSEric Dumazet if (!atomic_dec_and_test(&fclones->fclone_ref)) 6416ffe75ebSEric Dumazet return; 6426ffe75ebSEric Dumazet fastpath: 6436ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds 64604a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 6471da177e4SLinus Torvalds { 648adf30907SEric Dumazet skb_dst_drop(skb); 6491da177e4SLinus Torvalds #ifdef CONFIG_XFRM 6501da177e4SLinus Torvalds secpath_put(skb->sp); 6511da177e4SLinus Torvalds #endif 6521da177e4SLinus Torvalds if (skb->destructor) { 6539c2b3328SStephen Hemminger WARN_ON(in_irq()); 6541da177e4SLinus Torvalds skb->destructor(skb); 6551da177e4SLinus Torvalds } 656a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 657cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 6582fc72c7bSKOVACS Krisztian #endif 6591109a90cSPablo Neira Ayuso #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 6601da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 6611da177e4SLinus Torvalds #endif 66204a4bb55SLennert Buytenhek } 66304a4bb55SLennert Buytenhek 66404a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 66504a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 66604a4bb55SLennert Buytenhek { 66704a4bb55SLennert Buytenhek skb_release_head_state(skb); 6685e71d9d7SPablo Neira if (likely(skb->head)) 6692d4baff8SHerbert Xu skb_release_data(skb); 6702d4baff8SHerbert Xu } 6711da177e4SLinus Torvalds 6722d4baff8SHerbert Xu /** 6732d4baff8SHerbert Xu * __kfree_skb - private function 6742d4baff8SHerbert Xu * @skb: buffer 6752d4baff8SHerbert Xu * 6762d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6772d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6782d4baff8SHerbert Xu * always call kfree_skb 6792d4baff8SHerbert Xu */ 6802d4baff8SHerbert Xu 6812d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6822d4baff8SHerbert Xu { 6832d4baff8SHerbert Xu skb_release_all(skb); 6841da177e4SLinus Torvalds kfree_skbmem(skb); 6851da177e4SLinus Torvalds } 686b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds /** 689231d06aeSJörn Engel * kfree_skb - free an sk_buff 690231d06aeSJörn Engel * @skb: buffer to free 691231d06aeSJörn Engel * 692231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 693231d06aeSJörn Engel * hit zero. 694231d06aeSJörn Engel */ 695231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 696231d06aeSJörn Engel { 697231d06aeSJörn Engel if (unlikely(!skb)) 698231d06aeSJörn Engel return; 699231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 700231d06aeSJörn Engel smp_rmb(); 701231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 702231d06aeSJörn Engel return; 703ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 704231d06aeSJörn Engel __kfree_skb(skb); 705231d06aeSJörn Engel } 706b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 707231d06aeSJörn Engel 708bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 709bd8a7036SEric Dumazet { 710bd8a7036SEric Dumazet while (segs) { 711bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 712bd8a7036SEric Dumazet 713bd8a7036SEric Dumazet kfree_skb(segs); 714bd8a7036SEric Dumazet segs = next; 715bd8a7036SEric Dumazet } 716bd8a7036SEric Dumazet } 717bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 718bd8a7036SEric Dumazet 719d1a203eaSStephen Hemminger /** 72025121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 72125121173SMichael S. Tsirkin * @skb: buffer that triggered an error 72225121173SMichael S. Tsirkin * 72325121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 72425121173SMichael S. Tsirkin * skb must be freed afterwards. 72525121173SMichael S. Tsirkin */ 72625121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 72725121173SMichael S. Tsirkin { 72825121173SMichael S. Tsirkin if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 72925121173SMichael S. Tsirkin struct ubuf_info *uarg; 73025121173SMichael S. Tsirkin 73125121173SMichael S. Tsirkin uarg = skb_shinfo(skb)->destructor_arg; 73225121173SMichael S. Tsirkin if (uarg->callback) 73325121173SMichael S. Tsirkin uarg->callback(uarg, false); 73425121173SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 73525121173SMichael S. Tsirkin } 73625121173SMichael S. Tsirkin } 73725121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 73825121173SMichael S. Tsirkin 73925121173SMichael S. Tsirkin /** 740ead2ceb0SNeil Horman * consume_skb - free an skbuff 741ead2ceb0SNeil Horman * @skb: buffer to free 742ead2ceb0SNeil Horman * 743ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 744ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 745ead2ceb0SNeil Horman * is being dropped after a failure and notes that 746ead2ceb0SNeil Horman */ 747ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 748ead2ceb0SNeil Horman { 749ead2ceb0SNeil Horman if (unlikely(!skb)) 750ead2ceb0SNeil Horman return; 751ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 752ead2ceb0SNeil Horman smp_rmb(); 753ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 754ead2ceb0SNeil Horman return; 75507dc22e7SKoki Sanagi trace_consume_skb(skb); 756ead2ceb0SNeil Horman __kfree_skb(skb); 757ead2ceb0SNeil Horman } 758ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 759ead2ceb0SNeil Horman 760795bb1c0SJesper Dangaard Brouer void __kfree_skb_flush(void) 761795bb1c0SJesper Dangaard Brouer { 762795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 763795bb1c0SJesper Dangaard Brouer 764795bb1c0SJesper Dangaard Brouer /* flush skb_cache if containing objects */ 765795bb1c0SJesper Dangaard Brouer if (nc->skb_count) { 766795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 767795bb1c0SJesper Dangaard Brouer nc->skb_cache); 768795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 769795bb1c0SJesper Dangaard Brouer } 770795bb1c0SJesper Dangaard Brouer } 771795bb1c0SJesper Dangaard Brouer 77215fad714SJesper Dangaard Brouer static inline void _kfree_skb_defer(struct sk_buff *skb) 773795bb1c0SJesper Dangaard Brouer { 774795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 775795bb1c0SJesper Dangaard Brouer 776795bb1c0SJesper Dangaard Brouer /* drop skb->head and call any destructors for packet */ 777795bb1c0SJesper Dangaard Brouer skb_release_all(skb); 778795bb1c0SJesper Dangaard Brouer 779795bb1c0SJesper Dangaard Brouer /* record skb to CPU local list */ 780795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 781795bb1c0SJesper Dangaard Brouer 782795bb1c0SJesper Dangaard Brouer #ifdef CONFIG_SLUB 783795bb1c0SJesper Dangaard Brouer /* SLUB writes into objects when freeing */ 784795bb1c0SJesper Dangaard Brouer prefetchw(skb); 785795bb1c0SJesper Dangaard Brouer #endif 786795bb1c0SJesper Dangaard Brouer 787795bb1c0SJesper Dangaard Brouer /* flush skb_cache if it is filled */ 788795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 789795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 790795bb1c0SJesper Dangaard Brouer nc->skb_cache); 791795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 792795bb1c0SJesper Dangaard Brouer } 793795bb1c0SJesper Dangaard Brouer } 79415fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 79515fad714SJesper Dangaard Brouer { 79615fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 79715fad714SJesper Dangaard Brouer } 798795bb1c0SJesper Dangaard Brouer 799795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 800795bb1c0SJesper Dangaard Brouer { 801795bb1c0SJesper Dangaard Brouer if (unlikely(!skb)) 802795bb1c0SJesper Dangaard Brouer return; 803795bb1c0SJesper Dangaard Brouer 804885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 805795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 806885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 807795bb1c0SJesper Dangaard Brouer return; 808795bb1c0SJesper Dangaard Brouer } 809795bb1c0SJesper Dangaard Brouer 810795bb1c0SJesper Dangaard Brouer if (likely(atomic_read(&skb->users) == 1)) 811795bb1c0SJesper Dangaard Brouer smp_rmb(); 812795bb1c0SJesper Dangaard Brouer else if (likely(!atomic_dec_and_test(&skb->users))) 813795bb1c0SJesper Dangaard Brouer return; 814795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 815795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 816795bb1c0SJesper Dangaard Brouer 817795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 818abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 819795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 820795bb1c0SJesper Dangaard Brouer return; 821795bb1c0SJesper Dangaard Brouer } 822795bb1c0SJesper Dangaard Brouer 82315fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 824795bb1c0SJesper Dangaard Brouer } 825795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 826795bb1c0SJesper Dangaard Brouer 827b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 828b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 829b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 830b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 831b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 832b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 833b1937227SEric Dumazet 834dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 835dec18810SHerbert Xu { 836dec18810SHerbert Xu new->tstamp = old->tstamp; 837b1937227SEric Dumazet /* We do not copy old->sk */ 838dec18810SHerbert Xu new->dev = old->dev; 839b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 8407fee226aSEric Dumazet skb_dst_copy(new, old); 841def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 842dec18810SHerbert Xu new->sp = secpath_get(old->sp); 843dec18810SHerbert Xu #endif 844b1937227SEric Dumazet __nf_copy(new, old, false); 8456aa895b0SPatrick McHardy 846b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 847b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 848b1937227SEric Dumazet */ 849b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 85006021292SEliezer Tamir 851b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 852b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 853b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 854b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 855b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 856b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 857b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 858b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 859b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 860b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 861b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 862b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 863b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 864b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 865b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 866b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 867b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 868b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 869b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 870b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 871b1937227SEric Dumazet #endif 872e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 873b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 87406021292SEliezer Tamir #endif 8752bd82484SEric Dumazet #ifdef CONFIG_XPS 8762bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 8772bd82484SEric Dumazet #endif 878b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 879b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 880b1937227SEric Dumazet #endif 881b1937227SEric Dumazet 882dec18810SHerbert Xu } 883dec18810SHerbert Xu 88482c49a35SHerbert Xu /* 88582c49a35SHerbert Xu * You should not add any new code to this function. Add it to 88682c49a35SHerbert Xu * __copy_skb_header above instead. 88782c49a35SHerbert Xu */ 888e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 8891da177e4SLinus Torvalds { 8901da177e4SLinus Torvalds #define C(x) n->x = skb->x 8911da177e4SLinus Torvalds 8921da177e4SLinus Torvalds n->next = n->prev = NULL; 8931da177e4SLinus Torvalds n->sk = NULL; 894dec18810SHerbert Xu __copy_skb_header(n, skb); 895dec18810SHerbert Xu 8961da177e4SLinus Torvalds C(len); 8971da177e4SLinus Torvalds C(data_len); 8983e6b3b2eSAlexey Dobriyan C(mac_len); 899334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 90002f1c89dSPaul Moore n->cloned = 1; 9011da177e4SLinus Torvalds n->nohdr = 0; 9021da177e4SLinus Torvalds n->destructor = NULL; 9031da177e4SLinus Torvalds C(tail); 9041da177e4SLinus Torvalds C(end); 90502f1c89dSPaul Moore C(head); 906d3836f21SEric Dumazet C(head_frag); 90702f1c89dSPaul Moore C(data); 90802f1c89dSPaul Moore C(truesize); 90902f1c89dSPaul Moore atomic_set(&n->users, 1); 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 9121da177e4SLinus Torvalds skb->cloned = 1; 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds return n; 915e0053ec0SHerbert Xu #undef C 916e0053ec0SHerbert Xu } 917e0053ec0SHerbert Xu 918e0053ec0SHerbert Xu /** 919e0053ec0SHerbert Xu * skb_morph - morph one skb into another 920e0053ec0SHerbert Xu * @dst: the skb to receive the contents 921e0053ec0SHerbert Xu * @src: the skb to supply the contents 922e0053ec0SHerbert Xu * 923e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 924e0053ec0SHerbert Xu * supplied by the user. 925e0053ec0SHerbert Xu * 926e0053ec0SHerbert Xu * The target skb is returned upon exit. 927e0053ec0SHerbert Xu */ 928e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 929e0053ec0SHerbert Xu { 9302d4baff8SHerbert Xu skb_release_all(dst); 931e0053ec0SHerbert Xu return __skb_clone(dst, src); 932e0053ec0SHerbert Xu } 933e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 934e0053ec0SHerbert Xu 9352c53040fSBen Hutchings /** 9362c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 93748c83012SMichael S. Tsirkin * @skb: the skb to modify 93848c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 93948c83012SMichael S. Tsirkin * 94048c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 94148c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 94248c83012SMichael S. Tsirkin * to userspace pages. 94348c83012SMichael S. Tsirkin * 94448c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 94548c83012SMichael S. Tsirkin * %GFP_ATOMIC. 94648c83012SMichael S. Tsirkin * 94748c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 94848c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 94948c83012SMichael S. Tsirkin */ 95048c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 951a6686f2fSShirley Ma { 952a6686f2fSShirley Ma int i; 953a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 954a6686f2fSShirley Ma struct page *page, *head = NULL; 955a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 956a6686f2fSShirley Ma 957a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 958a6686f2fSShirley Ma u8 *vaddr; 959a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 960a6686f2fSShirley Ma 96102756ed4SKrishna Kumar page = alloc_page(gfp_mask); 962a6686f2fSShirley Ma if (!page) { 963a6686f2fSShirley Ma while (head) { 96440dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 965a6686f2fSShirley Ma put_page(head); 966a6686f2fSShirley Ma head = next; 967a6686f2fSShirley Ma } 968a6686f2fSShirley Ma return -ENOMEM; 969a6686f2fSShirley Ma } 97051c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 971a6686f2fSShirley Ma memcpy(page_address(page), 9729e903e08SEric Dumazet vaddr + f->page_offset, skb_frag_size(f)); 97351c56b00SEric Dumazet kunmap_atomic(vaddr); 97440dadff2SSunghan Suh set_page_private(page, (unsigned long)head); 975a6686f2fSShirley Ma head = page; 976a6686f2fSShirley Ma } 977a6686f2fSShirley Ma 978a6686f2fSShirley Ma /* skb frags release userspace buffers */ 97902756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 980a8605c60SIan Campbell skb_frag_unref(skb, i); 981a6686f2fSShirley Ma 982e19d6763SMichael S. Tsirkin uarg->callback(uarg, false); 983a6686f2fSShirley Ma 984a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 98502756ed4SKrishna Kumar for (i = num_frags - 1; i >= 0; i--) { 98602756ed4SKrishna Kumar __skb_fill_page_desc(skb, i, head, 0, 98702756ed4SKrishna Kumar skb_shinfo(skb)->frags[i].size); 98840dadff2SSunghan Suh head = (struct page *)page_private(head); 989a6686f2fSShirley Ma } 99048c83012SMichael S. Tsirkin 99148c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 992a6686f2fSShirley Ma return 0; 993a6686f2fSShirley Ma } 994dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 995a6686f2fSShirley Ma 996e0053ec0SHerbert Xu /** 997e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 998e0053ec0SHerbert Xu * @skb: buffer to clone 999e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1000e0053ec0SHerbert Xu * 1001e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1002e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1003e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1004e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1005e0053ec0SHerbert Xu * 1006e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1007e0053ec0SHerbert Xu * %GFP_ATOMIC. 1008e0053ec0SHerbert Xu */ 1009e0053ec0SHerbert Xu 1010e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1011e0053ec0SHerbert Xu { 1012d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1013d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1014d0bf4a9eSEric Dumazet skb1); 10156ffe75ebSEric Dumazet struct sk_buff *n; 1016e0053ec0SHerbert Xu 101770008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1018a6686f2fSShirley Ma return NULL; 1019a6686f2fSShirley Ma 1020e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 10216ffe75ebSEric Dumazet atomic_read(&fclones->fclone_ref) == 1) { 10226ffe75ebSEric Dumazet n = &fclones->skb2; 10236ffe75ebSEric Dumazet atomic_set(&fclones->fclone_ref, 2); 1024e0053ec0SHerbert Xu } else { 1025c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1026c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1027c93bdd0eSMel Gorman 1028e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1029e0053ec0SHerbert Xu if (!n) 1030e0053ec0SHerbert Xu return NULL; 1031fe55f6d5SVegard Nossum 1032fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 1033e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1034e0053ec0SHerbert Xu } 1035e0053ec0SHerbert Xu 1036e0053ec0SHerbert Xu return __skb_clone(n, skb); 10371da177e4SLinus Torvalds } 1038b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 10391da177e4SLinus Torvalds 1040f5b17294SPravin B Shelar static void skb_headers_offset_update(struct sk_buff *skb, int off) 1041f5b17294SPravin B Shelar { 1042030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1043030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1044030737bcSEric Dumazet skb->csum_start += off; 1045f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1046f5b17294SPravin B Shelar skb->transport_header += off; 1047f5b17294SPravin B Shelar skb->network_header += off; 1048f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1049f5b17294SPravin B Shelar skb->mac_header += off; 1050f5b17294SPravin B Shelar skb->inner_transport_header += off; 1051f5b17294SPravin B Shelar skb->inner_network_header += off; 1052aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1053f5b17294SPravin B Shelar } 1054f5b17294SPravin B Shelar 10551da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 10561da177e4SLinus Torvalds { 1057dec18810SHerbert Xu __copy_skb_header(new, old); 1058dec18810SHerbert Xu 10597967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 10607967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 10617967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 10621da177e4SLinus Torvalds } 10631da177e4SLinus Torvalds 1064c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1065c93bdd0eSMel Gorman { 1066c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1067c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1068c93bdd0eSMel Gorman return 0; 1069c93bdd0eSMel Gorman } 1070c93bdd0eSMel Gorman 10711da177e4SLinus Torvalds /** 10721da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 10731da177e4SLinus Torvalds * @skb: buffer to copy 10741da177e4SLinus Torvalds * @gfp_mask: allocation priority 10751da177e4SLinus Torvalds * 10761da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 10771da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 10781da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 10791da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 10801da177e4SLinus Torvalds * 10811da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 10821da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 10831da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 10841da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 10851da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 10861da177e4SLinus Torvalds */ 10871da177e4SLinus Torvalds 1088dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 10891da177e4SLinus Torvalds { 10906602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1091ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1092c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1093c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 10946602cebbSEric Dumazet 10951da177e4SLinus Torvalds if (!n) 10961da177e4SLinus Torvalds return NULL; 10971da177e4SLinus Torvalds 10981da177e4SLinus Torvalds /* Set the data pointer */ 10991da177e4SLinus Torvalds skb_reserve(n, headerlen); 11001da177e4SLinus Torvalds /* Set the tail pointer and length */ 11011da177e4SLinus Torvalds skb_put(n, skb->len); 11021da177e4SLinus Torvalds 11031da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 11041da177e4SLinus Torvalds BUG(); 11051da177e4SLinus Torvalds 11061da177e4SLinus Torvalds copy_skb_header(n, skb); 11071da177e4SLinus Torvalds return n; 11081da177e4SLinus Torvalds } 1109b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 11101da177e4SLinus Torvalds 11111da177e4SLinus Torvalds /** 1112bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 11131da177e4SLinus Torvalds * @skb: buffer to copy 1114117632e6SEric Dumazet * @headroom: headroom of new skb 11151da177e4SLinus Torvalds * @gfp_mask: allocation priority 1116bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1117bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1118bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 11191da177e4SLinus Torvalds * 11201da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 11211da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 11221da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 11231da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 11241da177e4SLinus Torvalds * or the pointer to the buffer on success. 11251da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 11261da177e4SLinus Torvalds */ 11271da177e4SLinus Torvalds 1128bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1129bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 11301da177e4SLinus Torvalds { 1131117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1132bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1133bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 11346602cebbSEric Dumazet 11351da177e4SLinus Torvalds if (!n) 11361da177e4SLinus Torvalds goto out; 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds /* Set the data pointer */ 1139117632e6SEric Dumazet skb_reserve(n, headroom); 11401da177e4SLinus Torvalds /* Set the tail pointer and length */ 11411da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 11421da177e4SLinus Torvalds /* Copy the bytes */ 1143d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 11441da177e4SLinus Torvalds 114525f484a6SHerbert Xu n->truesize += skb->data_len; 11461da177e4SLinus Torvalds n->data_len = skb->data_len; 11471da177e4SLinus Torvalds n->len = skb->len; 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 11501da177e4SLinus Torvalds int i; 11511da177e4SLinus Torvalds 115270008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) { 11531511022cSDan Carpenter kfree_skb(n); 11541511022cSDan Carpenter n = NULL; 1155a6686f2fSShirley Ma goto out; 1156a6686f2fSShirley Ma } 11571da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 11581da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1159ea2ab693SIan Campbell skb_frag_ref(skb, i); 11601da177e4SLinus Torvalds } 11611da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 11621da177e4SLinus Torvalds } 11631da177e4SLinus Torvalds 116421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 11651da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 11661da177e4SLinus Torvalds skb_clone_fraglist(n); 11671da177e4SLinus Torvalds } 11681da177e4SLinus Torvalds 11691da177e4SLinus Torvalds copy_skb_header(n, skb); 11701da177e4SLinus Torvalds out: 11711da177e4SLinus Torvalds return n; 11721da177e4SLinus Torvalds } 1173bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 11741da177e4SLinus Torvalds 11751da177e4SLinus Torvalds /** 11761da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 11771da177e4SLinus Torvalds * @skb: buffer to reallocate 11781da177e4SLinus Torvalds * @nhead: room to add at head 11791da177e4SLinus Torvalds * @ntail: room to add at tail 11801da177e4SLinus Torvalds * @gfp_mask: allocation priority 11811da177e4SLinus Torvalds * 1182bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1183bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 11841da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 11851da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 11861da177e4SLinus Torvalds * 11871da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 11881da177e4SLinus Torvalds * reloaded after call to this function. 11891da177e4SLinus Torvalds */ 11901da177e4SLinus Torvalds 119186a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1192dd0fc66fSAl Viro gfp_t gfp_mask) 11931da177e4SLinus Torvalds { 1194158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1195158f323bSEric Dumazet int size = osize + nhead + ntail; 11961da177e4SLinus Torvalds long off; 1197158f323bSEric Dumazet u8 *data; 11981da177e4SLinus Torvalds 11994edd87adSHerbert Xu BUG_ON(nhead < 0); 12004edd87adSHerbert Xu 12011da177e4SLinus Torvalds if (skb_shared(skb)) 12021da177e4SLinus Torvalds BUG(); 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 12051da177e4SLinus Torvalds 1206c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1207c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1208c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1209c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 12101da177e4SLinus Torvalds if (!data) 12111da177e4SLinus Torvalds goto nodata; 121287151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 12131da177e4SLinus Torvalds 12141da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 12156602cebbSEric Dumazet * optimized for the cases when header is void. 12166602cebbSEric Dumazet */ 12176602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 12186602cebbSEric Dumazet 12196602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 12206602cebbSEric Dumazet skb_shinfo(skb), 1221fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 12221da177e4SLinus Torvalds 12233e24591aSAlexander Duyck /* 12243e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 12253e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 12263e24591aSAlexander Duyck * be since all we did is relocate the values 12273e24591aSAlexander Duyck */ 12283e24591aSAlexander Duyck if (skb_cloned(skb)) { 1229a6686f2fSShirley Ma /* copy this zero copy skb frags */ 123070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1231a6686f2fSShirley Ma goto nofrags; 12321da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1233ea2ab693SIan Campbell skb_frag_ref(skb, i); 12341da177e4SLinus Torvalds 123521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 12361da177e4SLinus Torvalds skb_clone_fraglist(skb); 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds skb_release_data(skb); 12393e24591aSAlexander Duyck } else { 12403e24591aSAlexander Duyck skb_free_head(skb); 12411fd63041SEric Dumazet } 12421da177e4SLinus Torvalds off = (data + nhead) - skb->head; 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds skb->head = data; 1245d3836f21SEric Dumazet skb->head_frag = 0; 12461da177e4SLinus Torvalds skb->data += off; 12474305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 12484305b541SArnaldo Carvalho de Melo skb->end = size; 124956eb8882SPatrick McHardy off = nhead; 12504305b541SArnaldo Carvalho de Melo #else 12514305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 125256eb8882SPatrick McHardy #endif 125327a884dcSArnaldo Carvalho de Melo skb->tail += off; 1254b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 12551da177e4SLinus Torvalds skb->cloned = 0; 1256334a8132SPatrick McHardy skb->hdr_len = 0; 12571da177e4SLinus Torvalds skb->nohdr = 0; 12581da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1259158f323bSEric Dumazet 1260158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1261158f323bSEric Dumazet * For the moment, we really care of rx path, or 1262158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1263158f323bSEric Dumazet */ 1264158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1265158f323bSEric Dumazet skb->truesize += size - osize; 1266158f323bSEric Dumazet 12671da177e4SLinus Torvalds return 0; 12681da177e4SLinus Torvalds 1269a6686f2fSShirley Ma nofrags: 1270a6686f2fSShirley Ma kfree(data); 12711da177e4SLinus Torvalds nodata: 12721da177e4SLinus Torvalds return -ENOMEM; 12731da177e4SLinus Torvalds } 1274b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 12771da177e4SLinus Torvalds 12781da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 12791da177e4SLinus Torvalds { 12801da177e4SLinus Torvalds struct sk_buff *skb2; 12811da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds if (delta <= 0) 12841da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 12851da177e4SLinus Torvalds else { 12861da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 12871da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 12881da177e4SLinus Torvalds GFP_ATOMIC)) { 12891da177e4SLinus Torvalds kfree_skb(skb2); 12901da177e4SLinus Torvalds skb2 = NULL; 12911da177e4SLinus Torvalds } 12921da177e4SLinus Torvalds } 12931da177e4SLinus Torvalds return skb2; 12941da177e4SLinus Torvalds } 1295b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds /** 12981da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 12991da177e4SLinus Torvalds * @skb: buffer to copy 13001da177e4SLinus Torvalds * @newheadroom: new free bytes at head 13011da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 13021da177e4SLinus Torvalds * @gfp_mask: allocation priority 13031da177e4SLinus Torvalds * 13041da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 13051da177e4SLinus Torvalds * allocate additional space. 13061da177e4SLinus Torvalds * 13071da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 13081da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 13091da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 13101da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 13111da177e4SLinus Torvalds * 13121da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 13131da177e4SLinus Torvalds * is called from an interrupt. 13141da177e4SLinus Torvalds */ 13151da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 131686a76cafSVictor Fusco int newheadroom, int newtailroom, 1317dd0fc66fSAl Viro gfp_t gfp_mask) 13181da177e4SLinus Torvalds { 13191da177e4SLinus Torvalds /* 13201da177e4SLinus Torvalds * Allocate the copy buffer 13211da177e4SLinus Torvalds */ 1322c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1323c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1324c93bdd0eSMel Gorman NUMA_NO_NODE); 1325efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 13261da177e4SLinus Torvalds int head_copy_len, head_copy_off; 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds if (!n) 13291da177e4SLinus Torvalds return NULL; 13301da177e4SLinus Torvalds 13311da177e4SLinus Torvalds skb_reserve(n, newheadroom); 13321da177e4SLinus Torvalds 13331da177e4SLinus Torvalds /* Set the tail pointer and length */ 13341da177e4SLinus Torvalds skb_put(n, skb->len); 13351da177e4SLinus Torvalds 1336efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 13371da177e4SLinus Torvalds head_copy_off = 0; 13381da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 13391da177e4SLinus Torvalds head_copy_len = newheadroom; 13401da177e4SLinus Torvalds else 13411da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 13421da177e4SLinus Torvalds 13431da177e4SLinus Torvalds /* Copy the linear header and data. */ 13441da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 13451da177e4SLinus Torvalds skb->len + head_copy_len)) 13461da177e4SLinus Torvalds BUG(); 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds copy_skb_header(n, skb); 13491da177e4SLinus Torvalds 1350030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1351efd1e8d5SPatrick McHardy 13521da177e4SLinus Torvalds return n; 13531da177e4SLinus Torvalds } 1354b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 13551da177e4SLinus Torvalds 13561da177e4SLinus Torvalds /** 13571da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 13581da177e4SLinus Torvalds * @skb: buffer to pad 13591da177e4SLinus Torvalds * @pad: space to pad 13601da177e4SLinus Torvalds * 13611da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 13621da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 13631da177e4SLinus Torvalds * beyond the buffer end onto the wire. 13641da177e4SLinus Torvalds * 13655b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 13661da177e4SLinus Torvalds */ 13671da177e4SLinus Torvalds 13685b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 13691da177e4SLinus Torvalds { 13705b057c6bSHerbert Xu int err; 13715b057c6bSHerbert Xu int ntail; 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 13745b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 13751da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 13765b057c6bSHerbert Xu return 0; 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds 13794305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 13805b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 13815b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 13825b057c6bSHerbert Xu if (unlikely(err)) 13835b057c6bSHerbert Xu goto free_skb; 13845b057c6bSHerbert Xu } 13855b057c6bSHerbert Xu 13865b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 13875b057c6bSHerbert Xu * to be audited. 13885b057c6bSHerbert Xu */ 13895b057c6bSHerbert Xu err = skb_linearize(skb); 13905b057c6bSHerbert Xu if (unlikely(err)) 13915b057c6bSHerbert Xu goto free_skb; 13925b057c6bSHerbert Xu 13935b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 13945b057c6bSHerbert Xu return 0; 13955b057c6bSHerbert Xu 13965b057c6bSHerbert Xu free_skb: 13971da177e4SLinus Torvalds kfree_skb(skb); 13985b057c6bSHerbert Xu return err; 13991da177e4SLinus Torvalds } 1400b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 14011da177e4SLinus Torvalds 14020dde3e16SIlpo Järvinen /** 14030c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 14040c7ddf36SMathias Krause * @skb: start of the buffer to use 14050c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 14060c7ddf36SMathias Krause * @len: amount of data to add 14070c7ddf36SMathias Krause * 14080c7ddf36SMathias Krause * This function extends the used data area of the potentially 14090c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 14100c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 14110c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 14120c7ddf36SMathias Krause * returned. 14130c7ddf36SMathias Krause */ 14140c7ddf36SMathias Krause 14150c7ddf36SMathias Krause unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 14160c7ddf36SMathias Krause { 14170c7ddf36SMathias Krause if (tail != skb) { 14180c7ddf36SMathias Krause skb->data_len += len; 14190c7ddf36SMathias Krause skb->len += len; 14200c7ddf36SMathias Krause } 14210c7ddf36SMathias Krause return skb_put(tail, len); 14220c7ddf36SMathias Krause } 14230c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 14240c7ddf36SMathias Krause 14250c7ddf36SMathias Krause /** 14260dde3e16SIlpo Järvinen * skb_put - add data to a buffer 14270dde3e16SIlpo Järvinen * @skb: buffer to use 14280dde3e16SIlpo Järvinen * @len: amount of data to add 14290dde3e16SIlpo Järvinen * 14300dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 14310dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 14320dde3e16SIlpo Järvinen * first byte of the extra data is returned. 14330dde3e16SIlpo Järvinen */ 14340dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 14350dde3e16SIlpo Järvinen { 14360dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 14370dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 14380dde3e16SIlpo Järvinen skb->tail += len; 14390dde3e16SIlpo Järvinen skb->len += len; 14400dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 14410dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 14420dde3e16SIlpo Järvinen return tmp; 14430dde3e16SIlpo Järvinen } 14440dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 14450dde3e16SIlpo Järvinen 14466be8ac2fSIlpo Järvinen /** 1447c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1448c2aa270aSIlpo Järvinen * @skb: buffer to use 1449c2aa270aSIlpo Järvinen * @len: amount of data to add 1450c2aa270aSIlpo Järvinen * 1451c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1452c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1453c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1454c2aa270aSIlpo Järvinen */ 1455c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1456c2aa270aSIlpo Järvinen { 1457c2aa270aSIlpo Järvinen skb->data -= len; 1458c2aa270aSIlpo Järvinen skb->len += len; 1459c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1460c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1461c2aa270aSIlpo Järvinen return skb->data; 1462c2aa270aSIlpo Järvinen } 1463c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1464c2aa270aSIlpo Järvinen 1465c2aa270aSIlpo Järvinen /** 14666be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 14676be8ac2fSIlpo Järvinen * @skb: buffer to use 14686be8ac2fSIlpo Järvinen * @len: amount of data to remove 14696be8ac2fSIlpo Järvinen * 14706be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 14716be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 14726be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 14736be8ac2fSIlpo Järvinen * the old data. 14746be8ac2fSIlpo Järvinen */ 14756be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 14766be8ac2fSIlpo Järvinen { 147747d29646SDavid S. Miller return skb_pull_inline(skb, len); 14786be8ac2fSIlpo Järvinen } 14796be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 14806be8ac2fSIlpo Järvinen 1481419ae74eSIlpo Järvinen /** 1482419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1483419ae74eSIlpo Järvinen * @skb: buffer to alter 1484419ae74eSIlpo Järvinen * @len: new length 1485419ae74eSIlpo Järvinen * 1486419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1487419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1488419ae74eSIlpo Järvinen * The skb must be linear. 1489419ae74eSIlpo Järvinen */ 1490419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1491419ae74eSIlpo Järvinen { 1492419ae74eSIlpo Järvinen if (skb->len > len) 1493419ae74eSIlpo Järvinen __skb_trim(skb, len); 1494419ae74eSIlpo Järvinen } 1495419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1496419ae74eSIlpo Järvinen 14973cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 14981da177e4SLinus Torvalds */ 14991da177e4SLinus Torvalds 15003cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 15011da177e4SLinus Torvalds { 150227b437c8SHerbert Xu struct sk_buff **fragp; 150327b437c8SHerbert Xu struct sk_buff *frag; 15041da177e4SLinus Torvalds int offset = skb_headlen(skb); 15051da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 15061da177e4SLinus Torvalds int i; 150727b437c8SHerbert Xu int err; 150827b437c8SHerbert Xu 150927b437c8SHerbert Xu if (skb_cloned(skb) && 151027b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 151127b437c8SHerbert Xu return err; 15121da177e4SLinus Torvalds 1513f4d26fb3SHerbert Xu i = 0; 1514f4d26fb3SHerbert Xu if (offset >= len) 1515f4d26fb3SHerbert Xu goto drop_pages; 1516f4d26fb3SHerbert Xu 1517f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 15189e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 151927b437c8SHerbert Xu 152027b437c8SHerbert Xu if (end < len) { 15211da177e4SLinus Torvalds offset = end; 152227b437c8SHerbert Xu continue; 15231da177e4SLinus Torvalds } 15241da177e4SLinus Torvalds 15259e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 152627b437c8SHerbert Xu 1527f4d26fb3SHerbert Xu drop_pages: 152827b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 152927b437c8SHerbert Xu 153027b437c8SHerbert Xu for (; i < nfrags; i++) 1531ea2ab693SIan Campbell skb_frag_unref(skb, i); 153227b437c8SHerbert Xu 153321dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 153427b437c8SHerbert Xu skb_drop_fraglist(skb); 1535f4d26fb3SHerbert Xu goto done; 153627b437c8SHerbert Xu } 153727b437c8SHerbert Xu 153827b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 153927b437c8SHerbert Xu fragp = &frag->next) { 154027b437c8SHerbert Xu int end = offset + frag->len; 154127b437c8SHerbert Xu 154227b437c8SHerbert Xu if (skb_shared(frag)) { 154327b437c8SHerbert Xu struct sk_buff *nfrag; 154427b437c8SHerbert Xu 154527b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 154627b437c8SHerbert Xu if (unlikely(!nfrag)) 154727b437c8SHerbert Xu return -ENOMEM; 154827b437c8SHerbert Xu 154927b437c8SHerbert Xu nfrag->next = frag->next; 155085bb2a60SEric Dumazet consume_skb(frag); 155127b437c8SHerbert Xu frag = nfrag; 155227b437c8SHerbert Xu *fragp = frag; 155327b437c8SHerbert Xu } 155427b437c8SHerbert Xu 155527b437c8SHerbert Xu if (end < len) { 155627b437c8SHerbert Xu offset = end; 155727b437c8SHerbert Xu continue; 155827b437c8SHerbert Xu } 155927b437c8SHerbert Xu 156027b437c8SHerbert Xu if (end > len && 156127b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 156227b437c8SHerbert Xu return err; 156327b437c8SHerbert Xu 156427b437c8SHerbert Xu if (frag->next) 156527b437c8SHerbert Xu skb_drop_list(&frag->next); 156627b437c8SHerbert Xu break; 156727b437c8SHerbert Xu } 156827b437c8SHerbert Xu 1569f4d26fb3SHerbert Xu done: 157027b437c8SHerbert Xu if (len > skb_headlen(skb)) { 15711da177e4SLinus Torvalds skb->data_len -= skb->len - len; 15721da177e4SLinus Torvalds skb->len = len; 15731da177e4SLinus Torvalds } else { 15741da177e4SLinus Torvalds skb->len = len; 15751da177e4SLinus Torvalds skb->data_len = 0; 157627a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 15771da177e4SLinus Torvalds } 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds return 0; 15801da177e4SLinus Torvalds } 1581b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 15821da177e4SLinus Torvalds 15831da177e4SLinus Torvalds /** 15841da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 15851da177e4SLinus Torvalds * @skb: buffer to reallocate 15861da177e4SLinus Torvalds * @delta: number of bytes to advance tail 15871da177e4SLinus Torvalds * 15881da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 15891da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 15901da177e4SLinus Torvalds * data from fragmented part. 15911da177e4SLinus Torvalds * 15921da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 15931da177e4SLinus Torvalds * 15941da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 15951da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 15961da177e4SLinus Torvalds * 15971da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 15981da177e4SLinus Torvalds * reloaded after call to this function. 15991da177e4SLinus Torvalds */ 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 16021da177e4SLinus Torvalds * when it is necessary. 16031da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 16041da177e4SLinus Torvalds * 2. It may change skb pointers. 16051da177e4SLinus Torvalds * 16061da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 16071da177e4SLinus Torvalds */ 16081da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 16091da177e4SLinus Torvalds { 16101da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 16111da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 16121da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 16131da177e4SLinus Torvalds */ 16144305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 16151da177e4SLinus Torvalds 16161da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 16171da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 16181da177e4SLinus Torvalds GFP_ATOMIC)) 16191da177e4SLinus Torvalds return NULL; 16201da177e4SLinus Torvalds } 16211da177e4SLinus Torvalds 162227a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 16231da177e4SLinus Torvalds BUG(); 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 16261da177e4SLinus Torvalds * size of pulled pages. Superb. 16271da177e4SLinus Torvalds */ 162821dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 16291da177e4SLinus Torvalds goto pull_pages; 16301da177e4SLinus Torvalds 16311da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 16321da177e4SLinus Torvalds eat = delta; 16331da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16349e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 16359e903e08SEric Dumazet 16369e903e08SEric Dumazet if (size >= eat) 16371da177e4SLinus Torvalds goto pull_pages; 16389e903e08SEric Dumazet eat -= size; 16391da177e4SLinus Torvalds } 16401da177e4SLinus Torvalds 16411da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 16421da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 16431da177e4SLinus Torvalds * but taking into account that pulling is expected to 16441da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 16451da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 16461da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 16471da177e4SLinus Torvalds */ 16481da177e4SLinus Torvalds if (eat) { 16491da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 16501da177e4SLinus Torvalds struct sk_buff *clone = NULL; 16511da177e4SLinus Torvalds struct sk_buff *insp = NULL; 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds do { 165409a62660SKris Katterjohn BUG_ON(!list); 16551da177e4SLinus Torvalds 16561da177e4SLinus Torvalds if (list->len <= eat) { 16571da177e4SLinus Torvalds /* Eaten as whole. */ 16581da177e4SLinus Torvalds eat -= list->len; 16591da177e4SLinus Torvalds list = list->next; 16601da177e4SLinus Torvalds insp = list; 16611da177e4SLinus Torvalds } else { 16621da177e4SLinus Torvalds /* Eaten partially. */ 16631da177e4SLinus Torvalds 16641da177e4SLinus Torvalds if (skb_shared(list)) { 16651da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 16661da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 16671da177e4SLinus Torvalds if (!clone) 16681da177e4SLinus Torvalds return NULL; 16691da177e4SLinus Torvalds insp = list->next; 16701da177e4SLinus Torvalds list = clone; 16711da177e4SLinus Torvalds } else { 16721da177e4SLinus Torvalds /* This may be pulled without 16731da177e4SLinus Torvalds * problems. */ 16741da177e4SLinus Torvalds insp = list; 16751da177e4SLinus Torvalds } 16761da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 16771da177e4SLinus Torvalds kfree_skb(clone); 16781da177e4SLinus Torvalds return NULL; 16791da177e4SLinus Torvalds } 16801da177e4SLinus Torvalds break; 16811da177e4SLinus Torvalds } 16821da177e4SLinus Torvalds } while (eat); 16831da177e4SLinus Torvalds 16841da177e4SLinus Torvalds /* Free pulled out fragments. */ 16851da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 16861da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 16871da177e4SLinus Torvalds kfree_skb(list); 16881da177e4SLinus Torvalds } 16891da177e4SLinus Torvalds /* And insert new clone at head. */ 16901da177e4SLinus Torvalds if (clone) { 16911da177e4SLinus Torvalds clone->next = list; 16921da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 16931da177e4SLinus Torvalds } 16941da177e4SLinus Torvalds } 16951da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds pull_pages: 16981da177e4SLinus Torvalds eat = delta; 16991da177e4SLinus Torvalds k = 0; 17001da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 17019e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 17029e903e08SEric Dumazet 17039e903e08SEric Dumazet if (size <= eat) { 1704ea2ab693SIan Campbell skb_frag_unref(skb, i); 17059e903e08SEric Dumazet eat -= size; 17061da177e4SLinus Torvalds } else { 17071da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 17081da177e4SLinus Torvalds if (eat) { 17091da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 17109e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 17111da177e4SLinus Torvalds eat = 0; 17121da177e4SLinus Torvalds } 17131da177e4SLinus Torvalds k++; 17141da177e4SLinus Torvalds } 17151da177e4SLinus Torvalds } 17161da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 17171da177e4SLinus Torvalds 17181da177e4SLinus Torvalds skb->tail += delta; 17191da177e4SLinus Torvalds skb->data_len -= delta; 17201da177e4SLinus Torvalds 172127a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 17221da177e4SLinus Torvalds } 1723b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 17241da177e4SLinus Torvalds 172522019b17SEric Dumazet /** 172622019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 172722019b17SEric Dumazet * @skb: source skb 172822019b17SEric Dumazet * @offset: offset in source 172922019b17SEric Dumazet * @to: destination buffer 173022019b17SEric Dumazet * @len: number of bytes to copy 173122019b17SEric Dumazet * 173222019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 173322019b17SEric Dumazet * destination buffer. 173422019b17SEric Dumazet * 173522019b17SEric Dumazet * CAUTION ! : 173622019b17SEric Dumazet * If its prototype is ever changed, 173722019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 173822019b17SEric Dumazet * since it is called from BPF assembly code. 173922019b17SEric Dumazet */ 17401da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 17411da177e4SLinus Torvalds { 17421a028e50SDavid S. Miller int start = skb_headlen(skb); 1743fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1744fbb398a8SDavid S. Miller int i, copy; 17451da177e4SLinus Torvalds 17461da177e4SLinus Torvalds if (offset > (int)skb->len - len) 17471da177e4SLinus Torvalds goto fault; 17481da177e4SLinus Torvalds 17491da177e4SLinus Torvalds /* Copy header. */ 17501a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 17511da177e4SLinus Torvalds if (copy > len) 17521da177e4SLinus Torvalds copy = len; 1753d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 17541da177e4SLinus Torvalds if ((len -= copy) == 0) 17551da177e4SLinus Torvalds return 0; 17561da177e4SLinus Torvalds offset += copy; 17571da177e4SLinus Torvalds to += copy; 17581da177e4SLinus Torvalds } 17591da177e4SLinus Torvalds 17601da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 17611a028e50SDavid S. Miller int end; 176251c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 17631da177e4SLinus Torvalds 1764547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17651a028e50SDavid S. Miller 176651c56b00SEric Dumazet end = start + skb_frag_size(f); 17671da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17681da177e4SLinus Torvalds u8 *vaddr; 17691da177e4SLinus Torvalds 17701da177e4SLinus Torvalds if (copy > len) 17711da177e4SLinus Torvalds copy = len; 17721da177e4SLinus Torvalds 177351c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 17741da177e4SLinus Torvalds memcpy(to, 177551c56b00SEric Dumazet vaddr + f->page_offset + offset - start, 177651c56b00SEric Dumazet copy); 177751c56b00SEric Dumazet kunmap_atomic(vaddr); 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds if ((len -= copy) == 0) 17801da177e4SLinus Torvalds return 0; 17811da177e4SLinus Torvalds offset += copy; 17821da177e4SLinus Torvalds to += copy; 17831da177e4SLinus Torvalds } 17841a028e50SDavid S. Miller start = end; 17851da177e4SLinus Torvalds } 17861da177e4SLinus Torvalds 1787fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 17881a028e50SDavid S. Miller int end; 17891da177e4SLinus Torvalds 1790547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17911a028e50SDavid S. Miller 1792fbb398a8SDavid S. Miller end = start + frag_iter->len; 17931da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17941da177e4SLinus Torvalds if (copy > len) 17951da177e4SLinus Torvalds copy = len; 1796fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 17971da177e4SLinus Torvalds goto fault; 17981da177e4SLinus Torvalds if ((len -= copy) == 0) 17991da177e4SLinus Torvalds return 0; 18001da177e4SLinus Torvalds offset += copy; 18011da177e4SLinus Torvalds to += copy; 18021da177e4SLinus Torvalds } 18031a028e50SDavid S. Miller start = end; 18041da177e4SLinus Torvalds } 1805a6686f2fSShirley Ma 18061da177e4SLinus Torvalds if (!len) 18071da177e4SLinus Torvalds return 0; 18081da177e4SLinus Torvalds 18091da177e4SLinus Torvalds fault: 18101da177e4SLinus Torvalds return -EFAULT; 18111da177e4SLinus Torvalds } 1812b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 18131da177e4SLinus Torvalds 18149c55e01cSJens Axboe /* 18159c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 18169c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 18179c55e01cSJens Axboe */ 18189c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 18199c55e01cSJens Axboe { 18208b9d3728SJarek Poplawski put_page(spd->pages[i]); 18218b9d3728SJarek Poplawski } 18229c55e01cSJens Axboe 1823a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 18244fb66994SJarek Poplawski unsigned int *offset, 182518aafc62SEric Dumazet struct sock *sk) 18268b9d3728SJarek Poplawski { 18275640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 18288b9d3728SJarek Poplawski 18295640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 18308b9d3728SJarek Poplawski return NULL; 18314fb66994SJarek Poplawski 18325640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 18334fb66994SJarek Poplawski 18345640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 18355640f768SEric Dumazet page_address(page) + *offset, *len); 18365640f768SEric Dumazet *offset = pfrag->offset; 18375640f768SEric Dumazet pfrag->offset += *len; 18384fb66994SJarek Poplawski 18395640f768SEric Dumazet return pfrag->page; 18409c55e01cSJens Axboe } 18419c55e01cSJens Axboe 184241c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 184341c73a0dSEric Dumazet struct page *page, 184441c73a0dSEric Dumazet unsigned int offset) 184541c73a0dSEric Dumazet { 184641c73a0dSEric Dumazet return spd->nr_pages && 184741c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 184841c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 184941c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 185041c73a0dSEric Dumazet } 185141c73a0dSEric Dumazet 18529c55e01cSJens Axboe /* 18539c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 18549c55e01cSJens Axboe */ 1855a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 185635f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 18574fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 185818aafc62SEric Dumazet bool linear, 18597a67e56fSJarek Poplawski struct sock *sk) 18609c55e01cSJens Axboe { 186141c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1862a108d5f3SDavid S. Miller return true; 18639c55e01cSJens Axboe 18648b9d3728SJarek Poplawski if (linear) { 186518aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 18668b9d3728SJarek Poplawski if (!page) 1867a108d5f3SDavid S. Miller return true; 186841c73a0dSEric Dumazet } 186941c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 187041c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 1871a108d5f3SDavid S. Miller return false; 187241c73a0dSEric Dumazet } 18738b9d3728SJarek Poplawski get_page(page); 18749c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 18754fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 18769c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 18779c55e01cSJens Axboe spd->nr_pages++; 18788b9d3728SJarek Poplawski 1879a108d5f3SDavid S. Miller return false; 18809c55e01cSJens Axboe } 18819c55e01cSJens Axboe 1882a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 18832870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 188418aafc62SEric Dumazet unsigned int *len, 1885d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 188635f3d14dSJens Axboe struct sock *sk, 188735f3d14dSJens Axboe struct pipe_inode_info *pipe) 18889c55e01cSJens Axboe { 18892870c43dSOctavian Purdila if (!*len) 1890a108d5f3SDavid S. Miller return true; 18919c55e01cSJens Axboe 18922870c43dSOctavian Purdila /* skip this segment if already processed */ 18932870c43dSOctavian Purdila if (*off >= plen) { 18942870c43dSOctavian Purdila *off -= plen; 1895a108d5f3SDavid S. Miller return false; 18962870c43dSOctavian Purdila } 18972870c43dSOctavian Purdila 18982870c43dSOctavian Purdila /* ignore any bits we already processed */ 18999ca1b22dSEric Dumazet poff += *off; 19009ca1b22dSEric Dumazet plen -= *off; 19012870c43dSOctavian Purdila *off = 0; 19022870c43dSOctavian Purdila 190318aafc62SEric Dumazet do { 190418aafc62SEric Dumazet unsigned int flen = min(*len, plen); 19052870c43dSOctavian Purdila 190618aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 190718aafc62SEric Dumazet linear, sk)) 1908a108d5f3SDavid S. Miller return true; 190918aafc62SEric Dumazet poff += flen; 191018aafc62SEric Dumazet plen -= flen; 19112870c43dSOctavian Purdila *len -= flen; 191218aafc62SEric Dumazet } while (*len && plen); 19132870c43dSOctavian Purdila 1914a108d5f3SDavid S. Miller return false; 1915db43a282SOctavian Purdila } 19169c55e01cSJens Axboe 19179c55e01cSJens Axboe /* 1918a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 19192870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 19209c55e01cSJens Axboe */ 1921a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 192235f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 192335f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 19242870c43dSOctavian Purdila { 19252870c43dSOctavian Purdila int seg; 1926fa9835e5STom Herbert struct sk_buff *iter; 19279c55e01cSJens Axboe 19281d0c0b32SEric Dumazet /* map the linear part : 19292996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 19302996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 19312996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 19329c55e01cSJens Axboe */ 19332870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 19342870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 19352870c43dSOctavian Purdila skb_headlen(skb), 193618aafc62SEric Dumazet offset, len, spd, 19373a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 19381d0c0b32SEric Dumazet sk, pipe)) 1939a108d5f3SDavid S. Miller return true; 19409c55e01cSJens Axboe 19419c55e01cSJens Axboe /* 19429c55e01cSJens Axboe * then map the fragments 19439c55e01cSJens Axboe */ 19449c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 19459c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 19469c55e01cSJens Axboe 1947ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 19489e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 194918aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 1950a108d5f3SDavid S. Miller return true; 19519c55e01cSJens Axboe } 19529c55e01cSJens Axboe 1953fa9835e5STom Herbert skb_walk_frags(skb, iter) { 1954fa9835e5STom Herbert if (*offset >= iter->len) { 1955fa9835e5STom Herbert *offset -= iter->len; 1956fa9835e5STom Herbert continue; 1957fa9835e5STom Herbert } 1958fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 1959fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 1960fa9835e5STom Herbert * case. 1961fa9835e5STom Herbert */ 1962fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 1963fa9835e5STom Herbert return true; 1964fa9835e5STom Herbert } 1965fa9835e5STom Herbert 1966a108d5f3SDavid S. Miller return false; 19679c55e01cSJens Axboe } 19689c55e01cSJens Axboe 19699c55e01cSJens Axboe /* 19709c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 1971fa9835e5STom Herbert * the fragments, and the frag list. 19729c55e01cSJens Axboe */ 1973a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 19749c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 197525869262SAl Viro unsigned int flags) 19769c55e01cSJens Axboe { 197741c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 197841c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 19799c55e01cSJens Axboe struct splice_pipe_desc spd = { 19809c55e01cSJens Axboe .pages = pages, 19819c55e01cSJens Axboe .partial = partial, 1982047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 19839c55e01cSJens Axboe .flags = flags, 198428a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 19859c55e01cSJens Axboe .spd_release = sock_spd_release, 19869c55e01cSJens Axboe }; 198735f3d14dSJens Axboe int ret = 0; 198835f3d14dSJens Axboe 1989fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 19909c55e01cSJens Axboe 1991a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 199225869262SAl Viro ret = splice_to_pipe(pipe, &spd); 19939c55e01cSJens Axboe 199435f3d14dSJens Axboe return ret; 19959c55e01cSJens Axboe } 19962b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 19979c55e01cSJens Axboe 1998357b40a1SHerbert Xu /** 1999357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2000357b40a1SHerbert Xu * @skb: destination buffer 2001357b40a1SHerbert Xu * @offset: offset in destination 2002357b40a1SHerbert Xu * @from: source buffer 2003357b40a1SHerbert Xu * @len: number of bytes to copy 2004357b40a1SHerbert Xu * 2005357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2006357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2007357b40a1SHerbert Xu * traversing fragment lists and such. 2008357b40a1SHerbert Xu */ 2009357b40a1SHerbert Xu 20100c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2011357b40a1SHerbert Xu { 20121a028e50SDavid S. Miller int start = skb_headlen(skb); 2013fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2014fbb398a8SDavid S. Miller int i, copy; 2015357b40a1SHerbert Xu 2016357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2017357b40a1SHerbert Xu goto fault; 2018357b40a1SHerbert Xu 20191a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2020357b40a1SHerbert Xu if (copy > len) 2021357b40a1SHerbert Xu copy = len; 202227d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2023357b40a1SHerbert Xu if ((len -= copy) == 0) 2024357b40a1SHerbert Xu return 0; 2025357b40a1SHerbert Xu offset += copy; 2026357b40a1SHerbert Xu from += copy; 2027357b40a1SHerbert Xu } 2028357b40a1SHerbert Xu 2029357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2030357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 20311a028e50SDavid S. Miller int end; 2032357b40a1SHerbert Xu 2033547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20341a028e50SDavid S. Miller 20359e903e08SEric Dumazet end = start + skb_frag_size(frag); 2036357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2037357b40a1SHerbert Xu u8 *vaddr; 2038357b40a1SHerbert Xu 2039357b40a1SHerbert Xu if (copy > len) 2040357b40a1SHerbert Xu copy = len; 2041357b40a1SHerbert Xu 204251c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 20431a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 20441a028e50SDavid S. Miller from, copy); 204551c56b00SEric Dumazet kunmap_atomic(vaddr); 2046357b40a1SHerbert Xu 2047357b40a1SHerbert Xu if ((len -= copy) == 0) 2048357b40a1SHerbert Xu return 0; 2049357b40a1SHerbert Xu offset += copy; 2050357b40a1SHerbert Xu from += copy; 2051357b40a1SHerbert Xu } 20521a028e50SDavid S. Miller start = end; 2053357b40a1SHerbert Xu } 2054357b40a1SHerbert Xu 2055fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 20561a028e50SDavid S. Miller int end; 2057357b40a1SHerbert Xu 2058547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20591a028e50SDavid S. Miller 2060fbb398a8SDavid S. Miller end = start + frag_iter->len; 2061357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2062357b40a1SHerbert Xu if (copy > len) 2063357b40a1SHerbert Xu copy = len; 2064fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 20651a028e50SDavid S. Miller from, copy)) 2066357b40a1SHerbert Xu goto fault; 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 if (!len) 2075357b40a1SHerbert Xu return 0; 2076357b40a1SHerbert Xu 2077357b40a1SHerbert Xu fault: 2078357b40a1SHerbert Xu return -EFAULT; 2079357b40a1SHerbert Xu } 2080357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2081357b40a1SHerbert Xu 20821da177e4SLinus Torvalds /* Checksum skb data. */ 20832817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 20842817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 20851da177e4SLinus Torvalds { 20861a028e50SDavid S. Miller int start = skb_headlen(skb); 20871a028e50SDavid S. Miller int i, copy = start - offset; 2088fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 20891da177e4SLinus Torvalds int pos = 0; 20901da177e4SLinus Torvalds 20911da177e4SLinus Torvalds /* Checksum header. */ 20921da177e4SLinus Torvalds if (copy > 0) { 20931da177e4SLinus Torvalds if (copy > len) 20941da177e4SLinus Torvalds copy = len; 20952817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 20961da177e4SLinus Torvalds if ((len -= copy) == 0) 20971da177e4SLinus Torvalds return csum; 20981da177e4SLinus Torvalds offset += copy; 20991da177e4SLinus Torvalds pos = copy; 21001da177e4SLinus Torvalds } 21011da177e4SLinus Torvalds 21021da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21031a028e50SDavid S. Miller int end; 210451c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 21051da177e4SLinus Torvalds 2106547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21071a028e50SDavid S. Miller 210851c56b00SEric Dumazet end = start + skb_frag_size(frag); 21091da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 211044bb9363SAl Viro __wsum csum2; 21111da177e4SLinus Torvalds u8 *vaddr; 21121da177e4SLinus Torvalds 21131da177e4SLinus Torvalds if (copy > len) 21141da177e4SLinus Torvalds copy = len; 211551c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 21162817a336SDaniel Borkmann csum2 = ops->update(vaddr + frag->page_offset + 21171a028e50SDavid S. Miller offset - start, copy, 0); 211851c56b00SEric Dumazet kunmap_atomic(vaddr); 21192817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 21201da177e4SLinus Torvalds if (!(len -= copy)) 21211da177e4SLinus Torvalds return csum; 21221da177e4SLinus Torvalds offset += copy; 21231da177e4SLinus Torvalds pos += copy; 21241da177e4SLinus Torvalds } 21251a028e50SDavid S. Miller start = end; 21261da177e4SLinus Torvalds } 21271da177e4SLinus Torvalds 2128fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 21291a028e50SDavid S. Miller int end; 21301da177e4SLinus Torvalds 2131547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21321a028e50SDavid S. Miller 2133fbb398a8SDavid S. Miller end = start + frag_iter->len; 21341da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21355f92a738SAl Viro __wsum csum2; 21361da177e4SLinus Torvalds if (copy > len) 21371da177e4SLinus Torvalds copy = len; 21382817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 21392817a336SDaniel Borkmann copy, 0, ops); 21402817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 21411da177e4SLinus Torvalds if ((len -= copy) == 0) 21421da177e4SLinus Torvalds return csum; 21431da177e4SLinus Torvalds offset += copy; 21441da177e4SLinus Torvalds pos += copy; 21451da177e4SLinus Torvalds } 21461a028e50SDavid S. Miller start = end; 21471da177e4SLinus Torvalds } 214809a62660SKris Katterjohn BUG_ON(len); 21491da177e4SLinus Torvalds 21501da177e4SLinus Torvalds return csum; 21511da177e4SLinus Torvalds } 21522817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 21532817a336SDaniel Borkmann 21542817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 21552817a336SDaniel Borkmann int len, __wsum csum) 21562817a336SDaniel Borkmann { 21572817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2158cea80ea8SDaniel Borkmann .update = csum_partial_ext, 21592817a336SDaniel Borkmann .combine = csum_block_add_ext, 21602817a336SDaniel Borkmann }; 21612817a336SDaniel Borkmann 21622817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 21632817a336SDaniel Borkmann } 2164b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 21651da177e4SLinus Torvalds 21661da177e4SLinus Torvalds /* Both of above in one bottle. */ 21671da177e4SLinus Torvalds 216881d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 216981d77662SAl Viro u8 *to, int len, __wsum csum) 21701da177e4SLinus Torvalds { 21711a028e50SDavid S. Miller int start = skb_headlen(skb); 21721a028e50SDavid S. Miller int i, copy = start - offset; 2173fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 21741da177e4SLinus Torvalds int pos = 0; 21751da177e4SLinus Torvalds 21761da177e4SLinus Torvalds /* Copy header. */ 21771da177e4SLinus Torvalds if (copy > 0) { 21781da177e4SLinus Torvalds if (copy > len) 21791da177e4SLinus Torvalds copy = len; 21801da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 21811da177e4SLinus Torvalds copy, csum); 21821da177e4SLinus Torvalds if ((len -= copy) == 0) 21831da177e4SLinus Torvalds return csum; 21841da177e4SLinus Torvalds offset += copy; 21851da177e4SLinus Torvalds to += copy; 21861da177e4SLinus Torvalds pos = copy; 21871da177e4SLinus Torvalds } 21881da177e4SLinus Torvalds 21891da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21901a028e50SDavid S. Miller int end; 21911da177e4SLinus Torvalds 2192547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21931a028e50SDavid S. Miller 21949e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 21951da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21965084205fSAl Viro __wsum csum2; 21971da177e4SLinus Torvalds u8 *vaddr; 21981da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 21991da177e4SLinus Torvalds 22001da177e4SLinus Torvalds if (copy > len) 22011da177e4SLinus Torvalds copy = len; 220251c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 22031da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 22041a028e50SDavid S. Miller frag->page_offset + 22051a028e50SDavid S. Miller offset - start, to, 22061a028e50SDavid S. Miller copy, 0); 220751c56b00SEric Dumazet kunmap_atomic(vaddr); 22081da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 22091da177e4SLinus Torvalds if (!(len -= copy)) 22101da177e4SLinus Torvalds return csum; 22111da177e4SLinus Torvalds offset += copy; 22121da177e4SLinus Torvalds to += copy; 22131da177e4SLinus Torvalds pos += copy; 22141da177e4SLinus Torvalds } 22151a028e50SDavid S. Miller start = end; 22161da177e4SLinus Torvalds } 22171da177e4SLinus Torvalds 2218fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 221981d77662SAl Viro __wsum csum2; 22201a028e50SDavid S. Miller int end; 22211da177e4SLinus Torvalds 2222547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22231a028e50SDavid S. Miller 2224fbb398a8SDavid S. Miller end = start + frag_iter->len; 22251da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 22261da177e4SLinus Torvalds if (copy > len) 22271da177e4SLinus Torvalds copy = len; 2228fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 22291a028e50SDavid S. Miller offset - start, 22301da177e4SLinus Torvalds to, copy, 0); 22311da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 22321da177e4SLinus Torvalds if ((len -= copy) == 0) 22331da177e4SLinus Torvalds return csum; 22341da177e4SLinus Torvalds offset += copy; 22351da177e4SLinus Torvalds to += copy; 22361da177e4SLinus Torvalds pos += copy; 22371da177e4SLinus Torvalds } 22381a028e50SDavid S. Miller start = end; 22391da177e4SLinus Torvalds } 224009a62660SKris Katterjohn BUG_ON(len); 22411da177e4SLinus Torvalds return csum; 22421da177e4SLinus Torvalds } 2243b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 22441da177e4SLinus Torvalds 2245af2806f8SThomas Graf /** 2246af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2247af2806f8SThomas Graf * @from: source buffer 2248af2806f8SThomas Graf * 2249af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2250af2806f8SThomas Graf * into skb_zerocopy(). 2251af2806f8SThomas Graf */ 2252af2806f8SThomas Graf unsigned int 2253af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2254af2806f8SThomas Graf { 2255af2806f8SThomas Graf unsigned int hlen = 0; 2256af2806f8SThomas Graf 2257af2806f8SThomas Graf if (!from->head_frag || 2258af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2259af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2260af2806f8SThomas Graf hlen = skb_headlen(from); 2261af2806f8SThomas Graf 2262af2806f8SThomas Graf if (skb_has_frag_list(from)) 2263af2806f8SThomas Graf hlen = from->len; 2264af2806f8SThomas Graf 2265af2806f8SThomas Graf return hlen; 2266af2806f8SThomas Graf } 2267af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2268af2806f8SThomas Graf 2269af2806f8SThomas Graf /** 2270af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2271af2806f8SThomas Graf * @to: destination buffer 22727fceb4deSMasanari Iida * @from: source buffer 2273af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2274af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2275af2806f8SThomas Graf * 2276af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2277af2806f8SThomas Graf * to the frags in the source buffer. 2278af2806f8SThomas Graf * 2279af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2280af2806f8SThomas Graf * headroom in the `to` buffer. 228136d5fe6aSZoltan Kiss * 228236d5fe6aSZoltan Kiss * Return value: 228336d5fe6aSZoltan Kiss * 0: everything is OK 228436d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 228536d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2286af2806f8SThomas Graf */ 228736d5fe6aSZoltan Kiss int 228836d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2289af2806f8SThomas Graf { 2290af2806f8SThomas Graf int i, j = 0; 2291af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 229236d5fe6aSZoltan Kiss int ret; 2293af2806f8SThomas Graf struct page *page; 2294af2806f8SThomas Graf unsigned int offset; 2295af2806f8SThomas Graf 2296af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2297af2806f8SThomas Graf 2298af2806f8SThomas Graf /* dont bother with small payloads */ 229936d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 230036d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2301af2806f8SThomas Graf 2302af2806f8SThomas Graf if (hlen) { 230336d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 230436d5fe6aSZoltan Kiss if (unlikely(ret)) 230536d5fe6aSZoltan Kiss return ret; 2306af2806f8SThomas Graf len -= hlen; 2307af2806f8SThomas Graf } else { 2308af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2309af2806f8SThomas Graf if (plen) { 2310af2806f8SThomas Graf page = virt_to_head_page(from->head); 2311af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2312af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2313af2806f8SThomas Graf get_page(page); 2314af2806f8SThomas Graf j = 1; 2315af2806f8SThomas Graf len -= plen; 2316af2806f8SThomas Graf } 2317af2806f8SThomas Graf } 2318af2806f8SThomas Graf 2319af2806f8SThomas Graf to->truesize += len + plen; 2320af2806f8SThomas Graf to->len += len + plen; 2321af2806f8SThomas Graf to->data_len += len + plen; 2322af2806f8SThomas Graf 232336d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 232436d5fe6aSZoltan Kiss skb_tx_error(from); 232536d5fe6aSZoltan Kiss return -ENOMEM; 232636d5fe6aSZoltan Kiss } 232736d5fe6aSZoltan Kiss 2328af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2329af2806f8SThomas Graf if (!len) 2330af2806f8SThomas Graf break; 2331af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2332af2806f8SThomas Graf skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2333af2806f8SThomas Graf len -= skb_shinfo(to)->frags[j].size; 2334af2806f8SThomas Graf skb_frag_ref(to, j); 2335af2806f8SThomas Graf j++; 2336af2806f8SThomas Graf } 2337af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 233836d5fe6aSZoltan Kiss 233936d5fe6aSZoltan Kiss return 0; 2340af2806f8SThomas Graf } 2341af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2342af2806f8SThomas Graf 23431da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 23441da177e4SLinus Torvalds { 2345d3bc23e7SAl Viro __wsum csum; 23461da177e4SLinus Torvalds long csstart; 23471da177e4SLinus Torvalds 234884fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 234955508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 23501da177e4SLinus Torvalds else 23511da177e4SLinus Torvalds csstart = skb_headlen(skb); 23521da177e4SLinus Torvalds 235309a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 23541da177e4SLinus Torvalds 2355d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 23561da177e4SLinus Torvalds 23571da177e4SLinus Torvalds csum = 0; 23581da177e4SLinus Torvalds if (csstart != skb->len) 23591da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 23601da177e4SLinus Torvalds skb->len - csstart, 0); 23611da177e4SLinus Torvalds 236284fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2363ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 23641da177e4SLinus Torvalds 2365d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 23661da177e4SLinus Torvalds } 23671da177e4SLinus Torvalds } 2368b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 23691da177e4SLinus Torvalds 23701da177e4SLinus Torvalds /** 23711da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 23721da177e4SLinus Torvalds * @list: list to dequeue from 23731da177e4SLinus Torvalds * 23741da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 23751da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 23761da177e4SLinus Torvalds * returned or %NULL if the list is empty. 23771da177e4SLinus Torvalds */ 23781da177e4SLinus Torvalds 23791da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 23801da177e4SLinus Torvalds { 23811da177e4SLinus Torvalds unsigned long flags; 23821da177e4SLinus Torvalds struct sk_buff *result; 23831da177e4SLinus Torvalds 23841da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 23851da177e4SLinus Torvalds result = __skb_dequeue(list); 23861da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 23871da177e4SLinus Torvalds return result; 23881da177e4SLinus Torvalds } 2389b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 23901da177e4SLinus Torvalds 23911da177e4SLinus Torvalds /** 23921da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 23931da177e4SLinus Torvalds * @list: list to dequeue from 23941da177e4SLinus Torvalds * 23951da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 23961da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 23971da177e4SLinus Torvalds * returned or %NULL if the list is empty. 23981da177e4SLinus Torvalds */ 23991da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(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_tail(list); 24061da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24071da177e4SLinus Torvalds return result; 24081da177e4SLinus Torvalds } 2409b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 24101da177e4SLinus Torvalds 24111da177e4SLinus Torvalds /** 24121da177e4SLinus Torvalds * skb_queue_purge - empty a list 24131da177e4SLinus Torvalds * @list: list to empty 24141da177e4SLinus Torvalds * 24151da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 24161da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 24171da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 24181da177e4SLinus Torvalds */ 24191da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 24201da177e4SLinus Torvalds { 24211da177e4SLinus Torvalds struct sk_buff *skb; 24221da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 24231da177e4SLinus Torvalds kfree_skb(skb); 24241da177e4SLinus Torvalds } 2425b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 24261da177e4SLinus Torvalds 24271da177e4SLinus Torvalds /** 24289f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 24299f5afeaeSYaogong Wang * @root: root of the rbtree to empty 24309f5afeaeSYaogong Wang * 24319f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 24329f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 24339f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 24349f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 24359f5afeaeSYaogong Wang */ 24369f5afeaeSYaogong Wang void skb_rbtree_purge(struct rb_root *root) 24379f5afeaeSYaogong Wang { 24389f5afeaeSYaogong Wang struct sk_buff *skb, *next; 24399f5afeaeSYaogong Wang 24409f5afeaeSYaogong Wang rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode) 24419f5afeaeSYaogong Wang kfree_skb(skb); 24429f5afeaeSYaogong Wang 24439f5afeaeSYaogong Wang *root = RB_ROOT; 24449f5afeaeSYaogong Wang } 24459f5afeaeSYaogong Wang 24469f5afeaeSYaogong Wang /** 24471da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 24481da177e4SLinus Torvalds * @list: list to use 24491da177e4SLinus Torvalds * @newsk: buffer to queue 24501da177e4SLinus Torvalds * 24511da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 24521da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 24531da177e4SLinus Torvalds * safely. 24541da177e4SLinus Torvalds * 24551da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24561da177e4SLinus Torvalds */ 24571da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 24581da177e4SLinus Torvalds { 24591da177e4SLinus Torvalds unsigned long flags; 24601da177e4SLinus Torvalds 24611da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24621da177e4SLinus Torvalds __skb_queue_head(list, newsk); 24631da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24641da177e4SLinus Torvalds } 2465b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 24661da177e4SLinus Torvalds 24671da177e4SLinus Torvalds /** 24681da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 24691da177e4SLinus Torvalds * @list: list to use 24701da177e4SLinus Torvalds * @newsk: buffer to queue 24711da177e4SLinus Torvalds * 24721da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 24731da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 24741da177e4SLinus Torvalds * safely. 24751da177e4SLinus Torvalds * 24761da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 24771da177e4SLinus Torvalds */ 24781da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 24791da177e4SLinus Torvalds { 24801da177e4SLinus Torvalds unsigned long flags; 24811da177e4SLinus Torvalds 24821da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 24831da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 24841da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 24851da177e4SLinus Torvalds } 2486b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 24878728b834SDavid S. Miller 24881da177e4SLinus Torvalds /** 24891da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 24901da177e4SLinus Torvalds * @skb: buffer to remove 24918728b834SDavid S. Miller * @list: list to use 24921da177e4SLinus Torvalds * 24938728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 24948728b834SDavid S. Miller * function is atomic with respect to other list locked calls 24951da177e4SLinus Torvalds * 24968728b834SDavid S. Miller * You must know what list the SKB is on. 24971da177e4SLinus Torvalds */ 24988728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 24991da177e4SLinus Torvalds { 25001da177e4SLinus Torvalds unsigned long flags; 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 25038728b834SDavid S. Miller __skb_unlink(skb, list); 25041da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 25051da177e4SLinus Torvalds } 2506b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 25071da177e4SLinus Torvalds 25081da177e4SLinus Torvalds /** 25091da177e4SLinus Torvalds * skb_append - append a buffer 25101da177e4SLinus Torvalds * @old: buffer to insert after 25111da177e4SLinus Torvalds * @newsk: buffer to insert 25128728b834SDavid S. Miller * @list: list to use 25131da177e4SLinus Torvalds * 25141da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 25151da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 25161da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 25171da177e4SLinus Torvalds */ 25188728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 25191da177e4SLinus Torvalds { 25201da177e4SLinus Torvalds unsigned long flags; 25211da177e4SLinus Torvalds 25228728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 25237de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 25248728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 25251da177e4SLinus Torvalds } 2526b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 25271da177e4SLinus Torvalds 25281da177e4SLinus Torvalds /** 25291da177e4SLinus Torvalds * skb_insert - insert a buffer 25301da177e4SLinus Torvalds * @old: buffer to insert before 25311da177e4SLinus Torvalds * @newsk: buffer to insert 25328728b834SDavid S. Miller * @list: list to use 25331da177e4SLinus Torvalds * 25348728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 25358728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 25368728b834SDavid S. Miller * calls. 25378728b834SDavid S. Miller * 25381da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 25391da177e4SLinus Torvalds */ 25408728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 25411da177e4SLinus Torvalds { 25421da177e4SLinus Torvalds unsigned long flags; 25431da177e4SLinus Torvalds 25448728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 25458728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 25468728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 25471da177e4SLinus Torvalds } 2548b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 25491da177e4SLinus Torvalds 25501da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 25511da177e4SLinus Torvalds struct sk_buff* skb1, 25521da177e4SLinus Torvalds const u32 len, const int pos) 25531da177e4SLinus Torvalds { 25541da177e4SLinus Torvalds int i; 25551da177e4SLinus Torvalds 2556d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2557d626f62bSArnaldo Carvalho de Melo pos - len); 25581da177e4SLinus Torvalds /* And move data appendix as is. */ 25591da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 25601da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 25631da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 25641da177e4SLinus Torvalds skb1->data_len = skb->data_len; 25651da177e4SLinus Torvalds skb1->len += skb1->data_len; 25661da177e4SLinus Torvalds skb->data_len = 0; 25671da177e4SLinus Torvalds skb->len = len; 256827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 25691da177e4SLinus Torvalds } 25701da177e4SLinus Torvalds 25711da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 25721da177e4SLinus Torvalds struct sk_buff* skb1, 25731da177e4SLinus Torvalds const u32 len, int pos) 25741da177e4SLinus Torvalds { 25751da177e4SLinus Torvalds int i, k = 0; 25761da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 25771da177e4SLinus Torvalds 25781da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 25791da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 25801da177e4SLinus Torvalds skb->len = len; 25811da177e4SLinus Torvalds skb->data_len = len - pos; 25821da177e4SLinus Torvalds 25831da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 25849e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 25851da177e4SLinus Torvalds 25861da177e4SLinus Torvalds if (pos + size > len) { 25871da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 25881da177e4SLinus Torvalds 25891da177e4SLinus Torvalds if (pos < len) { 25901da177e4SLinus Torvalds /* Split frag. 25911da177e4SLinus Torvalds * We have two variants in this case: 25921da177e4SLinus Torvalds * 1. Move all the frag to the second 25931da177e4SLinus Torvalds * part, if it is possible. F.e. 25941da177e4SLinus Torvalds * this approach is mandatory for TUX, 25951da177e4SLinus Torvalds * where splitting is expensive. 25961da177e4SLinus Torvalds * 2. Split is accurately. We make this. 25971da177e4SLinus Torvalds */ 2598ea2ab693SIan Campbell skb_frag_ref(skb, i); 25991da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 26009e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 26019e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 26021da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 26031da177e4SLinus Torvalds } 26041da177e4SLinus Torvalds k++; 26051da177e4SLinus Torvalds } else 26061da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 26071da177e4SLinus Torvalds pos += size; 26081da177e4SLinus Torvalds } 26091da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 26101da177e4SLinus Torvalds } 26111da177e4SLinus Torvalds 26121da177e4SLinus Torvalds /** 26131da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 26141da177e4SLinus Torvalds * @skb: the buffer to split 26151da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 26161da177e4SLinus Torvalds * @len: new length for skb 26171da177e4SLinus Torvalds */ 26181da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 26191da177e4SLinus Torvalds { 26201da177e4SLinus Torvalds int pos = skb_headlen(skb); 26211da177e4SLinus Torvalds 262268534c68SAmerigo Wang skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 26231da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 26241da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 26251da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 26261da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 26271da177e4SLinus Torvalds } 2628b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 26291da177e4SLinus Torvalds 26309f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 26319f782db3SIlpo Järvinen * 26329f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 26339f782db3SIlpo Järvinen */ 2634832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2635832d11c5SIlpo Järvinen { 26360ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2637832d11c5SIlpo Järvinen } 2638832d11c5SIlpo Järvinen 2639832d11c5SIlpo Järvinen /** 2640832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2641832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2642832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2643832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2644832d11c5SIlpo Järvinen * 2645832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 264620e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 2647832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2648832d11c5SIlpo Järvinen * 2649832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2650832d11c5SIlpo Järvinen * 2651832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2652832d11c5SIlpo Järvinen * to have non-paged data as well. 2653832d11c5SIlpo Järvinen * 2654832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2655832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2656832d11c5SIlpo Järvinen */ 2657832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2658832d11c5SIlpo Järvinen { 2659832d11c5SIlpo Järvinen int from, to, merge, todo; 2660832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2661832d11c5SIlpo Järvinen 2662832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2663f8071cdeSEric Dumazet 2664f8071cdeSEric Dumazet if (skb_headlen(skb)) 2665f8071cdeSEric Dumazet return 0; 2666832d11c5SIlpo Järvinen 2667832d11c5SIlpo Järvinen todo = shiftlen; 2668832d11c5SIlpo Järvinen from = 0; 2669832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2670832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2671832d11c5SIlpo Järvinen 2672832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2673832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2674832d11c5SIlpo Järvinen */ 2675832d11c5SIlpo Järvinen if (!to || 2676ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2677ea2ab693SIan Campbell fragfrom->page_offset)) { 2678832d11c5SIlpo Järvinen merge = -1; 2679832d11c5SIlpo Järvinen } else { 2680832d11c5SIlpo Järvinen merge = to - 1; 2681832d11c5SIlpo Järvinen 26829e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2683832d11c5SIlpo Järvinen if (todo < 0) { 2684832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2685832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2686832d11c5SIlpo Järvinen return 0; 2687832d11c5SIlpo Järvinen 26889f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 26899f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2690832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2691832d11c5SIlpo Järvinen 26929e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 26939e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 2694832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2695832d11c5SIlpo Järvinen 2696832d11c5SIlpo Järvinen goto onlymerged; 2697832d11c5SIlpo Järvinen } 2698832d11c5SIlpo Järvinen 2699832d11c5SIlpo Järvinen from++; 2700832d11c5SIlpo Järvinen } 2701832d11c5SIlpo Järvinen 2702832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2703832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2704832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2705832d11c5SIlpo Järvinen return 0; 2706832d11c5SIlpo Järvinen 2707832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2708832d11c5SIlpo Järvinen return 0; 2709832d11c5SIlpo Järvinen 2710832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2711832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2712832d11c5SIlpo Järvinen return 0; 2713832d11c5SIlpo Järvinen 2714832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2715832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2716832d11c5SIlpo Järvinen 27179e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 2718832d11c5SIlpo Järvinen *fragto = *fragfrom; 27199e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2720832d11c5SIlpo Järvinen from++; 2721832d11c5SIlpo Järvinen to++; 2722832d11c5SIlpo Järvinen 2723832d11c5SIlpo Järvinen } else { 2724ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2725832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2726832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 27279e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 2728832d11c5SIlpo Järvinen 2729832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 27309e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 2731832d11c5SIlpo Järvinen todo = 0; 2732832d11c5SIlpo Järvinen 2733832d11c5SIlpo Järvinen to++; 2734832d11c5SIlpo Järvinen break; 2735832d11c5SIlpo Järvinen } 2736832d11c5SIlpo Järvinen } 2737832d11c5SIlpo Järvinen 2738832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2739832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2740832d11c5SIlpo Järvinen 2741832d11c5SIlpo Järvinen if (merge >= 0) { 2742832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2743832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2744832d11c5SIlpo Järvinen 27459e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2746ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2747832d11c5SIlpo Järvinen } 2748832d11c5SIlpo Järvinen 2749832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2750832d11c5SIlpo Järvinen to = 0; 2751832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2752832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2753832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2754832d11c5SIlpo Järvinen 2755832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2756832d11c5SIlpo Järvinen 2757832d11c5SIlpo Järvinen onlymerged: 2758832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2759832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2760832d11c5SIlpo Järvinen */ 2761832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2762832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2763832d11c5SIlpo Järvinen 2764832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2765832d11c5SIlpo Järvinen skb->len -= shiftlen; 2766832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2767832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2768832d11c5SIlpo Järvinen tgt->len += shiftlen; 2769832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2770832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2771832d11c5SIlpo Järvinen 2772832d11c5SIlpo Järvinen return shiftlen; 2773832d11c5SIlpo Järvinen } 2774832d11c5SIlpo Järvinen 2775677e90edSThomas Graf /** 2776677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2777677e90edSThomas Graf * @skb: the buffer to read 2778677e90edSThomas Graf * @from: lower offset of data to be read 2779677e90edSThomas Graf * @to: upper offset of data to be read 2780677e90edSThomas Graf * @st: state variable 2781677e90edSThomas Graf * 2782677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2783677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2784677e90edSThomas Graf */ 2785677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2786677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2787677e90edSThomas Graf { 2788677e90edSThomas Graf st->lower_offset = from; 2789677e90edSThomas Graf st->upper_offset = to; 2790677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2791677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2792677e90edSThomas Graf st->frag_data = NULL; 2793677e90edSThomas Graf } 2794b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2795677e90edSThomas Graf 2796677e90edSThomas Graf /** 2797677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2798677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2799677e90edSThomas Graf * @data: destination pointer for data to be returned 2800677e90edSThomas Graf * @st: state variable 2801677e90edSThomas Graf * 2802bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 2803677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2804bc32383cSMathias Krause * the head of the data block to @data and returns the length 2805677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2806677e90edSThomas Graf * offset has been reached. 2807677e90edSThomas Graf * 2808677e90edSThomas Graf * The caller is not required to consume all of the data 2809bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 2810677e90edSThomas Graf * of bytes already consumed and the next call to 2811677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2812677e90edSThomas Graf * 281325985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2814e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 2815677e90edSThomas Graf * reads of potentially non linear data. 2816677e90edSThomas Graf * 2817bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2818677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2819677e90edSThomas Graf * a stack for this purpose. 2820677e90edSThomas Graf */ 2821677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2822677e90edSThomas Graf struct skb_seq_state *st) 2823677e90edSThomas Graf { 2824677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2825677e90edSThomas Graf skb_frag_t *frag; 2826677e90edSThomas Graf 2827aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 2828aeb193eaSWedson Almeida Filho if (st->frag_data) { 2829aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 2830aeb193eaSWedson Almeida Filho st->frag_data = NULL; 2831aeb193eaSWedson Almeida Filho } 2832677e90edSThomas Graf return 0; 2833aeb193eaSWedson Almeida Filho } 2834677e90edSThomas Graf 2835677e90edSThomas Graf next_skb: 283695e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2837677e90edSThomas Graf 2838995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 283995e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2840677e90edSThomas Graf return block_limit - abs_offset; 2841677e90edSThomas Graf } 2842677e90edSThomas Graf 2843677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2844677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2845677e90edSThomas Graf 2846677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2847677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 28489e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 2849677e90edSThomas Graf 2850677e90edSThomas Graf if (abs_offset < block_limit) { 2851677e90edSThomas Graf if (!st->frag_data) 285251c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 2853677e90edSThomas Graf 2854677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2855677e90edSThomas Graf (abs_offset - st->stepped_offset); 2856677e90edSThomas Graf 2857677e90edSThomas Graf return block_limit - abs_offset; 2858677e90edSThomas Graf } 2859677e90edSThomas Graf 2860677e90edSThomas Graf if (st->frag_data) { 286151c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2862677e90edSThomas Graf st->frag_data = NULL; 2863677e90edSThomas Graf } 2864677e90edSThomas Graf 2865677e90edSThomas Graf st->frag_idx++; 28669e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 2867677e90edSThomas Graf } 2868677e90edSThomas Graf 28695b5a60daSOlaf Kirch if (st->frag_data) { 287051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 28715b5a60daSOlaf Kirch st->frag_data = NULL; 28725b5a60daSOlaf Kirch } 28735b5a60daSOlaf Kirch 287421dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2875677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 287695e3b24cSHerbert Xu st->frag_idx = 0; 2877677e90edSThomas Graf goto next_skb; 287871b3346dSShyam Iyer } else if (st->cur_skb->next) { 287971b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 288071b3346dSShyam Iyer st->frag_idx = 0; 2881677e90edSThomas Graf goto next_skb; 2882677e90edSThomas Graf } 2883677e90edSThomas Graf 2884677e90edSThomas Graf return 0; 2885677e90edSThomas Graf } 2886b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2887677e90edSThomas Graf 2888677e90edSThomas Graf /** 2889677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2890677e90edSThomas Graf * @st: state variable 2891677e90edSThomas Graf * 2892677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2893677e90edSThomas Graf * returned 0. 2894677e90edSThomas Graf */ 2895677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2896677e90edSThomas Graf { 2897677e90edSThomas Graf if (st->frag_data) 289851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2899677e90edSThomas Graf } 2900b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2901677e90edSThomas Graf 29023fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 29033fc7e8a6SThomas Graf 29043fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 29053fc7e8a6SThomas Graf struct ts_config *conf, 29063fc7e8a6SThomas Graf struct ts_state *state) 29073fc7e8a6SThomas Graf { 29083fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 29093fc7e8a6SThomas Graf } 29103fc7e8a6SThomas Graf 29113fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 29123fc7e8a6SThomas Graf { 29133fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 29143fc7e8a6SThomas Graf } 29153fc7e8a6SThomas Graf 29163fc7e8a6SThomas Graf /** 29173fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 29183fc7e8a6SThomas Graf * @skb: the buffer to look in 29193fc7e8a6SThomas Graf * @from: search offset 29203fc7e8a6SThomas Graf * @to: search limit 29213fc7e8a6SThomas Graf * @config: textsearch configuration 29223fc7e8a6SThomas Graf * 29233fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 29243fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 29253fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 29263fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 29273fc7e8a6SThomas Graf */ 29283fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 2929059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 29303fc7e8a6SThomas Graf { 2931059a2440SBojan Prtvar struct ts_state state; 2932f72b948dSPhil Oester unsigned int ret; 2933f72b948dSPhil Oester 29343fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 29353fc7e8a6SThomas Graf config->finish = skb_ts_finish; 29363fc7e8a6SThomas Graf 2937059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 29383fc7e8a6SThomas Graf 2939059a2440SBojan Prtvar ret = textsearch_find(config, &state); 2940f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 29413fc7e8a6SThomas Graf } 2942b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 29433fc7e8a6SThomas Graf 2944e89e9cf5SAnanda Raju /** 29452c53040fSBen Hutchings * skb_append_datato_frags - append the user data to a skb 2946e89e9cf5SAnanda Raju * @sk: sock structure 2947e793c0f7SMasanari Iida * @skb: skb structure to be appended with user data. 2948e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2949e89e9cf5SAnanda Raju * @from: pointer to user message iov 2950e89e9cf5SAnanda Raju * @length: length of the iov message 2951e89e9cf5SAnanda Raju * 2952e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2953e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2954e89e9cf5SAnanda Raju */ 2955e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2956dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2957e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2958e89e9cf5SAnanda Raju void *from, int length) 2959e89e9cf5SAnanda Raju { 2960b2111724SEric Dumazet int frg_cnt = skb_shinfo(skb)->nr_frags; 2961b2111724SEric Dumazet int copy; 2962e89e9cf5SAnanda Raju int offset = 0; 2963e89e9cf5SAnanda Raju int ret; 2964b2111724SEric Dumazet struct page_frag *pfrag = ¤t->task_frag; 2965e89e9cf5SAnanda Raju 2966e89e9cf5SAnanda Raju do { 2967e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2968e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2969b2111724SEric Dumazet return -EMSGSIZE; 2970e89e9cf5SAnanda Raju 2971b2111724SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 2972e89e9cf5SAnanda Raju return -ENOMEM; 2973e89e9cf5SAnanda Raju 2974e89e9cf5SAnanda Raju /* copy the user data to page */ 2975b2111724SEric Dumazet copy = min_t(int, length, pfrag->size - pfrag->offset); 2976e89e9cf5SAnanda Raju 2977b2111724SEric Dumazet ret = getfrag(from, page_address(pfrag->page) + pfrag->offset, 2978e89e9cf5SAnanda Raju offset, copy, 0, skb); 2979e89e9cf5SAnanda Raju if (ret < 0) 2980e89e9cf5SAnanda Raju return -EFAULT; 2981e89e9cf5SAnanda Raju 2982e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2983b2111724SEric Dumazet skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset, 2984b2111724SEric Dumazet copy); 2985b2111724SEric Dumazet frg_cnt++; 2986b2111724SEric Dumazet pfrag->offset += copy; 2987b2111724SEric Dumazet get_page(pfrag->page); 2988b2111724SEric Dumazet 2989b2111724SEric Dumazet skb->truesize += copy; 2990b2111724SEric Dumazet atomic_add(copy, &sk->sk_wmem_alloc); 2991e89e9cf5SAnanda Raju skb->len += copy; 2992e89e9cf5SAnanda Raju skb->data_len += copy; 2993e89e9cf5SAnanda Raju offset += copy; 2994e89e9cf5SAnanda Raju length -= copy; 2995e89e9cf5SAnanda Raju 2996e89e9cf5SAnanda Raju } while (length > 0); 2997e89e9cf5SAnanda Raju 2998e89e9cf5SAnanda Raju return 0; 2999e89e9cf5SAnanda Raju } 3000b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 3001e89e9cf5SAnanda Raju 3002be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3003be12a1feSHannes Frederic Sowa int offset, size_t size) 3004be12a1feSHannes Frederic Sowa { 3005be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3006be12a1feSHannes Frederic Sowa 3007be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3008be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3009be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3010be12a1feSHannes Frederic Sowa get_page(page); 3011be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3012be12a1feSHannes Frederic Sowa } else { 3013be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3014be12a1feSHannes Frederic Sowa } 3015be12a1feSHannes Frederic Sowa 3016be12a1feSHannes Frederic Sowa return 0; 3017be12a1feSHannes Frederic Sowa } 3018be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3019be12a1feSHannes Frederic Sowa 3020cbb042f9SHerbert Xu /** 3021cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3022cbb042f9SHerbert Xu * @skb: buffer to update 3023cbb042f9SHerbert Xu * @len: length of data pulled 3024cbb042f9SHerbert Xu * 3025cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3026fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 302784fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 302884fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 302984fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3030cbb042f9SHerbert Xu */ 3031cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3032cbb042f9SHerbert Xu { 303331b33dfbSPravin B Shelar unsigned char *data = skb->data; 303431b33dfbSPravin B Shelar 3035cbb042f9SHerbert Xu BUG_ON(len > skb->len); 303631b33dfbSPravin B Shelar __skb_pull(skb, len); 303731b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 303831b33dfbSPravin B Shelar return skb->data; 3039cbb042f9SHerbert Xu } 3040f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3041f94691acSArnaldo Carvalho de Melo 3042f4c50d99SHerbert Xu /** 3043f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3044df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3045576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3046f4c50d99SHerbert Xu * 3047f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 30484c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 30494c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3050f4c50d99SHerbert Xu */ 3051df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3052df5771ffSMichael S. Tsirkin netdev_features_t features) 3053f4c50d99SHerbert Xu { 3054f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3055f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 30561a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3057df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3058df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3059df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 30601fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3061f4c50d99SHerbert Xu unsigned int offset = doffset; 3062df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3063802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3064f4c50d99SHerbert Xu unsigned int headroom; 3065802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3066ec5f0615SPravin B Shelar __be16 proto; 306736c98382SAlexander Duyck bool csum, sg; 3068df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3069f4c50d99SHerbert Xu int err = -ENOMEM; 3070f4c50d99SHerbert Xu int i = 0; 3071f4c50d99SHerbert Xu int pos; 307253d6471cSVlad Yasevich int dummy; 3073f4c50d99SHerbert Xu 30745882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 307553d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 3076ec5f0615SPravin B Shelar if (unlikely(!proto)) 3077ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3078ec5f0615SPravin B Shelar 307936c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3080f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 30817e2b10c1STom Herbert 308207b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 308307b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 308407b26c94SSteffen Klassert struct sk_buff *iter; 308507b26c94SSteffen Klassert 308607b26c94SSteffen Klassert if (!list_skb || 308707b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 308807b26c94SSteffen Klassert goto normal; 308907b26c94SSteffen Klassert 309007b26c94SSteffen Klassert /* Split the buffer at the frag_list pointer. 309107b26c94SSteffen Klassert * This is based on the assumption that all 309207b26c94SSteffen Klassert * buffers in the chain excluding the last 309307b26c94SSteffen Klassert * containing the same amount of data. 309407b26c94SSteffen Klassert */ 309507b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 309607b26c94SSteffen Klassert if (skb_headlen(iter)) 309707b26c94SSteffen Klassert goto normal; 309807b26c94SSteffen Klassert 309907b26c94SSteffen Klassert len -= iter->len; 310007b26c94SSteffen Klassert } 310107b26c94SSteffen Klassert } 310207b26c94SSteffen Klassert 3103802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 3104802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 3105802ab55aSAlexander Duyck * now. 3106802ab55aSAlexander Duyck */ 3107802ab55aSAlexander Duyck partial_segs = len / mss; 3108d7fb5a80SAlexander Duyck if (partial_segs > 1) 3109802ab55aSAlexander Duyck mss *= partial_segs; 3110d7fb5a80SAlexander Duyck else 3111d7fb5a80SAlexander Duyck partial_segs = 0; 3112802ab55aSAlexander Duyck } 3113802ab55aSAlexander Duyck 311407b26c94SSteffen Klassert normal: 3115df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3116df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3117f4c50d99SHerbert Xu 3118f4c50d99SHerbert Xu do { 3119f4c50d99SHerbert Xu struct sk_buff *nskb; 31208cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3121c8884eddSHerbert Xu int hsize; 3122f4c50d99SHerbert Xu int size; 3123f4c50d99SHerbert Xu 31243953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 31253953c46cSMarcelo Ricardo Leitner len = list_skb->len; 31263953c46cSMarcelo Ricardo Leitner } else { 3127df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3128f4c50d99SHerbert Xu if (len > mss) 3129f4c50d99SHerbert Xu len = mss; 31303953c46cSMarcelo Ricardo Leitner } 3131f4c50d99SHerbert Xu 3132df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3133f4c50d99SHerbert Xu if (hsize < 0) 3134f4c50d99SHerbert Xu hsize = 0; 3135c8884eddSHerbert Xu if (hsize > len || !sg) 3136c8884eddSHerbert Xu hsize = len; 3137f4c50d99SHerbert Xu 31381a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 31391a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 31401a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 314189319d38SHerbert Xu 31429d8506ccSHerbert Xu i = 0; 31431a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 31441a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 31451fd819ecSMichael S. Tsirkin frag_skb = list_skb; 31461a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 31479d8506ccSHerbert Xu 31489d8506ccSHerbert Xu while (pos < offset + len) { 31499d8506ccSHerbert Xu BUG_ON(i >= nfrags); 31509d8506ccSHerbert Xu 31514e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 31529d8506ccSHerbert Xu if (pos + size > offset + len) 31539d8506ccSHerbert Xu break; 31549d8506ccSHerbert Xu 31559d8506ccSHerbert Xu i++; 31569d8506ccSHerbert Xu pos += size; 31574e1beba1SMichael S. Tsirkin frag++; 31589d8506ccSHerbert Xu } 31599d8506ccSHerbert Xu 31601a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 31611a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 316289319d38SHerbert Xu 3163f4c50d99SHerbert Xu if (unlikely(!nskb)) 3164f4c50d99SHerbert Xu goto err; 3165f4c50d99SHerbert Xu 31669d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 31679d8506ccSHerbert Xu kfree_skb(nskb); 31689d8506ccSHerbert Xu goto err; 31699d8506ccSHerbert Xu } 31709d8506ccSHerbert Xu 3171ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 317289319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 317389319d38SHerbert Xu kfree_skb(nskb); 317489319d38SHerbert Xu goto err; 317589319d38SHerbert Xu } 317689319d38SHerbert Xu 3177ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 317889319d38SHerbert Xu skb_release_head_state(nskb); 317989319d38SHerbert Xu __skb_push(nskb, doffset); 318089319d38SHerbert Xu } else { 3181c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3182df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3183c93bdd0eSMel Gorman NUMA_NO_NODE); 318489319d38SHerbert Xu 318589319d38SHerbert Xu if (unlikely(!nskb)) 318689319d38SHerbert Xu goto err; 318789319d38SHerbert Xu 318889319d38SHerbert Xu skb_reserve(nskb, headroom); 318989319d38SHerbert Xu __skb_put(nskb, doffset); 319089319d38SHerbert Xu } 319189319d38SHerbert Xu 3192f4c50d99SHerbert Xu if (segs) 3193f4c50d99SHerbert Xu tail->next = nskb; 3194f4c50d99SHerbert Xu else 3195f4c50d99SHerbert Xu segs = nskb; 3196f4c50d99SHerbert Xu tail = nskb; 3197f4c50d99SHerbert Xu 3198df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3199f4c50d99SHerbert Xu 3200030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3201fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 320268c33163SPravin B Shelar 3203df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 320468c33163SPravin B Shelar nskb->data - tnl_hlen, 320568c33163SPravin B Shelar doffset + tnl_hlen); 320689319d38SHerbert Xu 32079d8506ccSHerbert Xu if (nskb->len == len + doffset) 32081cdbcb79SSimon Horman goto perform_csum_check; 320989319d38SHerbert Xu 32107fbeffedSAlexander Duyck if (!sg) { 32117fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 32126f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 321376443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 321476443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 3215f4c50d99SHerbert Xu skb_put(nskb, len), 3216f4c50d99SHerbert Xu len, 0); 32177e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3218de843723STom Herbert skb_headroom(nskb) + doffset; 3219f4c50d99SHerbert Xu continue; 3220f4c50d99SHerbert Xu } 3221f4c50d99SHerbert Xu 32228cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3223f4c50d99SHerbert Xu 3224df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3225d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3226f4c50d99SHerbert Xu 3227df5771ffSMichael S. Tsirkin skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags & 3228df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3229cef401deSEric Dumazet 32309d8506ccSHerbert Xu while (pos < offset + len) { 32319d8506ccSHerbert Xu if (i >= nfrags) { 32321a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb)); 32339d8506ccSHerbert Xu 32349d8506ccSHerbert Xu i = 0; 32351a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 32361a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 32371fd819ecSMichael S. Tsirkin frag_skb = list_skb; 32389d8506ccSHerbert Xu 32399d8506ccSHerbert Xu BUG_ON(!nfrags); 32409d8506ccSHerbert Xu 32411a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 32429d8506ccSHerbert Xu } 32439d8506ccSHerbert Xu 32449d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 32459d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 32469d8506ccSHerbert Xu net_warn_ratelimited( 32479d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 32489d8506ccSHerbert Xu pos, mss); 32499d8506ccSHerbert Xu goto err; 32509d8506ccSHerbert Xu } 32519d8506ccSHerbert Xu 32521fd819ecSMichael S. Tsirkin if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC))) 32531fd819ecSMichael S. Tsirkin goto err; 32541fd819ecSMichael S. Tsirkin 32554e1beba1SMichael S. Tsirkin *nskb_frag = *frag; 32568cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 32578cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3258f4c50d99SHerbert Xu 3259f4c50d99SHerbert Xu if (pos < offset) { 32608cb19905SMichael S. Tsirkin nskb_frag->page_offset += offset - pos; 32618cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3262f4c50d99SHerbert Xu } 3263f4c50d99SHerbert Xu 326489319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3265f4c50d99SHerbert Xu 3266f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3267f4c50d99SHerbert Xu i++; 32684e1beba1SMichael S. Tsirkin frag++; 3269f4c50d99SHerbert Xu pos += size; 3270f4c50d99SHerbert Xu } else { 32718cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 327289319d38SHerbert Xu goto skip_fraglist; 3273f4c50d99SHerbert Xu } 3274f4c50d99SHerbert Xu 32758cb19905SMichael S. Tsirkin nskb_frag++; 3276f4c50d99SHerbert Xu } 3277f4c50d99SHerbert Xu 327889319d38SHerbert Xu skip_fraglist: 3279f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3280f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3281f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3282ec5f0615SPravin B Shelar 32831cdbcb79SSimon Horman perform_csum_check: 32847fbeffedSAlexander Duyck if (!csum) { 3285ddff00d4SAlexander Duyck if (skb_has_shared_frag(nskb)) { 3286ddff00d4SAlexander Duyck err = __skb_linearize(nskb); 3287ddff00d4SAlexander Duyck if (err) 3288ddff00d4SAlexander Duyck goto err; 3289ddff00d4SAlexander Duyck } 32907fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 3291ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 329276443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 329376443456SAlexander Duyck skb_checksum(nskb, doffset, 329476443456SAlexander Duyck nskb->len - doffset, 0); 32957e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 32967e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3297ec5f0615SPravin B Shelar } 3298df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3299f4c50d99SHerbert Xu 3300bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3301bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3302bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3303bec3cfdcSEric Dumazet */ 3304bec3cfdcSEric Dumazet segs->prev = tail; 3305432c856fSToshiaki Makita 3306802ab55aSAlexander Duyck if (partial_segs) { 330707b26c94SSteffen Klassert struct sk_buff *iter; 3308802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 330907b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 3310802ab55aSAlexander Duyck 3311802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 331207b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 3313802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 3314802ab55aSAlexander Duyck 3315802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 3316802ab55aSAlexander Duyck * our way back down the stack of protocols. 3317802ab55aSAlexander Duyck */ 331807b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 331907b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 332007b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 332107b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 332207b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 332307b26c94SSteffen Klassert } 332407b26c94SSteffen Klassert 332507b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 332607b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 332707b26c94SSteffen Klassert else if (tail != segs) 332807b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 3329802ab55aSAlexander Duyck } 3330802ab55aSAlexander Duyck 3331432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 3332432c856fSToshiaki Makita * using skb_set_owner_w(). 3333432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 3334432c856fSToshiaki Makita */ 3335432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 3336432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 3337432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 3338432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 3339432c856fSToshiaki Makita } 3340f4c50d99SHerbert Xu return segs; 3341f4c50d99SHerbert Xu 3342f4c50d99SHerbert Xu err: 3343289dccbeSEric Dumazet kfree_skb_list(segs); 3344f4c50d99SHerbert Xu return ERR_PTR(err); 3345f4c50d99SHerbert Xu } 3346f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3347f4c50d99SHerbert Xu 334871d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 334971d93b39SHerbert Xu { 33508a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 335167147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 335267147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 33538a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 335458025e46SEric Dumazet struct sk_buff *lp, *p = *head; 3355715dc1f3SEric Dumazet unsigned int delta_truesize; 335671d93b39SHerbert Xu 33578a29111cSEric Dumazet if (unlikely(p->len + len >= 65536)) 335871d93b39SHerbert Xu return -E2BIG; 335971d93b39SHerbert Xu 336029e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 33618a29111cSEric Dumazet pinfo = skb_shinfo(lp); 33628a29111cSEric Dumazet 33638a29111cSEric Dumazet if (headlen <= offset) { 336442da6994SHerbert Xu skb_frag_t *frag; 336566e92fcfSHerbert Xu skb_frag_t *frag2; 33669aaa156cSHerbert Xu int i = skbinfo->nr_frags; 33679aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 336842da6994SHerbert Xu 336966e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 33708a29111cSEric Dumazet goto merge; 337181705ad1SHerbert Xu 33728a29111cSEric Dumazet offset -= headlen; 33739aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 33749aaa156cSHerbert Xu skbinfo->nr_frags = 0; 3375f5572068SHerbert Xu 33769aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 33779aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 337866e92fcfSHerbert Xu do { 337966e92fcfSHerbert Xu *--frag = *--frag2; 338066e92fcfSHerbert Xu } while (--i); 338166e92fcfSHerbert Xu 338266e92fcfSHerbert Xu frag->page_offset += offset; 33839e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 338466e92fcfSHerbert Xu 3385715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3386ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3387ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3388715dc1f3SEric Dumazet 3389f5572068SHerbert Xu skb->truesize -= skb->data_len; 3390f5572068SHerbert Xu skb->len -= skb->data_len; 3391f5572068SHerbert Xu skb->data_len = 0; 3392f5572068SHerbert Xu 3393715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 33945d38a079SHerbert Xu goto done; 3395d7e8883cSEric Dumazet } else if (skb->head_frag) { 3396d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3397d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3398d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3399d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3400d7e8883cSEric Dumazet unsigned int first_offset; 3401d7e8883cSEric Dumazet 3402d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 34038a29111cSEric Dumazet goto merge; 3404d7e8883cSEric Dumazet 3405d7e8883cSEric Dumazet first_offset = skb->data - 3406d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3407d7e8883cSEric Dumazet offset; 3408d7e8883cSEric Dumazet 3409d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3410d7e8883cSEric Dumazet 3411d7e8883cSEric Dumazet frag->page.p = page; 3412d7e8883cSEric Dumazet frag->page_offset = first_offset; 3413d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3414d7e8883cSEric Dumazet 3415d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3416d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3417d7e8883cSEric Dumazet 3418715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3419d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3420d7e8883cSEric Dumazet goto done; 34218a29111cSEric Dumazet } 342271d93b39SHerbert Xu 342371d93b39SHerbert Xu merge: 3424715dc1f3SEric Dumazet delta_truesize = skb->truesize; 342567147ba9SHerbert Xu if (offset > headlen) { 3426d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3427d1dc7abfSMichal Schmidt 3428d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 34299e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3430d1dc7abfSMichal Schmidt skb->data_len -= eat; 3431d1dc7abfSMichal Schmidt skb->len -= eat; 343267147ba9SHerbert Xu offset = headlen; 343356035022SHerbert Xu } 343456035022SHerbert Xu 343567147ba9SHerbert Xu __skb_pull(skb, offset); 343656035022SHerbert Xu 343729e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 34388a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 34398a29111cSEric Dumazet else 3440c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3441c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 3442f4a775d1SEric Dumazet __skb_header_release(skb); 34438a29111cSEric Dumazet lp = p; 344471d93b39SHerbert Xu 34455d38a079SHerbert Xu done: 34465d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 344737fe4732SHerbert Xu p->data_len += len; 3448715dc1f3SEric Dumazet p->truesize += delta_truesize; 344937fe4732SHerbert Xu p->len += len; 34508a29111cSEric Dumazet if (lp != p) { 34518a29111cSEric Dumazet lp->data_len += len; 34528a29111cSEric Dumazet lp->truesize += delta_truesize; 34538a29111cSEric Dumazet lp->len += len; 34548a29111cSEric Dumazet } 345571d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 345671d93b39SHerbert Xu return 0; 345771d93b39SHerbert Xu } 345857c05650SMarcelo Ricardo Leitner EXPORT_SYMBOL_GPL(skb_gro_receive); 345971d93b39SHerbert Xu 34601da177e4SLinus Torvalds void __init skb_init(void) 34611da177e4SLinus Torvalds { 34621da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 34631da177e4SLinus Torvalds sizeof(struct sk_buff), 34641da177e4SLinus Torvalds 0, 3465e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 346620c2df83SPaul Mundt NULL); 3467d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3468d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 3469d179cd12SDavid S. Miller 0, 3470e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 347120c2df83SPaul Mundt NULL); 34721da177e4SLinus Torvalds } 34731da177e4SLinus Torvalds 3474716ea3a7SDavid Howells /** 3475716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3476716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 3477716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 3478716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 3479716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 3480716ea3a7SDavid Howells * 3481716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 3482716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 3483716ea3a7SDavid Howells */ 348451c739d1SDavid S. Miller static int 348551c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3486716ea3a7SDavid Howells { 34871a028e50SDavid S. Miller int start = skb_headlen(skb); 34881a028e50SDavid S. Miller int i, copy = start - offset; 3489fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3490716ea3a7SDavid Howells int elt = 0; 3491716ea3a7SDavid Howells 3492716ea3a7SDavid Howells if (copy > 0) { 3493716ea3a7SDavid Howells if (copy > len) 3494716ea3a7SDavid Howells copy = len; 3495642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 3496716ea3a7SDavid Howells elt++; 3497716ea3a7SDavid Howells if ((len -= copy) == 0) 3498716ea3a7SDavid Howells return elt; 3499716ea3a7SDavid Howells offset += copy; 3500716ea3a7SDavid Howells } 3501716ea3a7SDavid Howells 3502716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 35031a028e50SDavid S. Miller int end; 3504716ea3a7SDavid Howells 3505547b792cSIlpo Järvinen WARN_ON(start > offset + len); 35061a028e50SDavid S. Miller 35079e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3508716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3509716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3510716ea3a7SDavid Howells 3511716ea3a7SDavid Howells if (copy > len) 3512716ea3a7SDavid Howells copy = len; 3513ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3514642f1490SJens Axboe frag->page_offset+offset-start); 3515716ea3a7SDavid Howells elt++; 3516716ea3a7SDavid Howells if (!(len -= copy)) 3517716ea3a7SDavid Howells return elt; 3518716ea3a7SDavid Howells offset += copy; 3519716ea3a7SDavid Howells } 35201a028e50SDavid S. Miller start = end; 3521716ea3a7SDavid Howells } 3522716ea3a7SDavid Howells 3523fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 35241a028e50SDavid S. Miller int end; 3525716ea3a7SDavid Howells 3526547b792cSIlpo Järvinen WARN_ON(start > offset + len); 35271a028e50SDavid S. Miller 3528fbb398a8SDavid S. Miller end = start + frag_iter->len; 3529716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3530716ea3a7SDavid Howells if (copy > len) 3531716ea3a7SDavid Howells copy = len; 3532fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 353351c739d1SDavid S. Miller copy); 3534716ea3a7SDavid Howells if ((len -= copy) == 0) 3535716ea3a7SDavid Howells return elt; 3536716ea3a7SDavid Howells offset += copy; 3537716ea3a7SDavid Howells } 35381a028e50SDavid S. Miller start = end; 3539716ea3a7SDavid Howells } 3540716ea3a7SDavid Howells BUG_ON(len); 3541716ea3a7SDavid Howells return elt; 3542716ea3a7SDavid Howells } 3543716ea3a7SDavid Howells 354425a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 354525a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 354625a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 354725a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 354825a91d8dSFan Du * 354925a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 355025a91d8dSFan Du * 1. sg_init_table 355125a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 355225a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 355325a91d8dSFan Du * 355425a91d8dSFan Du * This is equivalent to: 355525a91d8dSFan Du * 1. sg_init_table 355625a91d8dSFan Du * 2. skb_to_sgvec(payload1) 355725a91d8dSFan Du * 3. sg_unmark_end 355825a91d8dSFan Du * 4. skb_to_sgvec(payload2) 355925a91d8dSFan Du * 356025a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 356125a91d8dSFan Du * is more preferable. 356225a91d8dSFan Du */ 356325a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 356425a91d8dSFan Du int offset, int len) 356525a91d8dSFan Du { 356625a91d8dSFan Du return __skb_to_sgvec(skb, sg, offset, len); 356725a91d8dSFan Du } 356825a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 356925a91d8dSFan Du 357051c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 357151c739d1SDavid S. Miller { 357251c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 357351c739d1SDavid S. Miller 3574c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 357551c739d1SDavid S. Miller 357651c739d1SDavid S. Miller return nsg; 357751c739d1SDavid S. Miller } 3578b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 357951c739d1SDavid S. Miller 3580716ea3a7SDavid Howells /** 3581716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 3582716ea3a7SDavid Howells * @skb: The socket buffer to check. 3583716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 3584716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 3585716ea3a7SDavid Howells * 3586716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 3587716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 3588716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 3589716ea3a7SDavid Howells * 3590716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 3591716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 3592716ea3a7SDavid Howells * set to point to the skb in which this space begins. 3593716ea3a7SDavid Howells * 3594716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3595716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3596716ea3a7SDavid Howells */ 3597716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3598716ea3a7SDavid Howells { 3599716ea3a7SDavid Howells int copyflag; 3600716ea3a7SDavid Howells int elt; 3601716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3602716ea3a7SDavid Howells 3603716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3604716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3605716ea3a7SDavid Howells * at the moment even if they are anonymous). 3606716ea3a7SDavid Howells */ 3607716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3608716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3609716ea3a7SDavid Howells return -ENOMEM; 3610716ea3a7SDavid Howells 3611716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 361221dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3613716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3614716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3615716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3616716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3617716ea3a7SDavid Howells 3618716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3619716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3620716ea3a7SDavid Howells return -ENOMEM; 3621716ea3a7SDavid Howells 3622716ea3a7SDavid Howells /* Voila! */ 3623716ea3a7SDavid Howells *trailer = skb; 3624716ea3a7SDavid Howells return 1; 3625716ea3a7SDavid Howells } 3626716ea3a7SDavid Howells 3627716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3628716ea3a7SDavid Howells 3629716ea3a7SDavid Howells elt = 1; 3630716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3631716ea3a7SDavid Howells copyflag = 0; 3632716ea3a7SDavid Howells 3633716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3634716ea3a7SDavid Howells int ntail = 0; 3635716ea3a7SDavid Howells 3636716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3637716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3638716ea3a7SDavid Howells * after it. */ 3639716ea3a7SDavid Howells 3640716ea3a7SDavid Howells if (skb_shared(skb1)) 3641716ea3a7SDavid Howells copyflag = 1; 3642716ea3a7SDavid Howells 3643716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3644716ea3a7SDavid Howells 3645716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3646716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 364721dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3648716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3649716ea3a7SDavid Howells ntail = tailbits + 128; 3650716ea3a7SDavid Howells } 3651716ea3a7SDavid Howells 3652716ea3a7SDavid Howells if (copyflag || 3653716ea3a7SDavid Howells skb_cloned(skb1) || 3654716ea3a7SDavid Howells ntail || 3655716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 365621dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3657716ea3a7SDavid Howells struct sk_buff *skb2; 3658716ea3a7SDavid Howells 3659716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3660716ea3a7SDavid Howells if (ntail == 0) 3661716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3662716ea3a7SDavid Howells else 3663716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3664716ea3a7SDavid Howells skb_headroom(skb1), 3665716ea3a7SDavid Howells ntail, 3666716ea3a7SDavid Howells GFP_ATOMIC); 3667716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3668716ea3a7SDavid Howells return -ENOMEM; 3669716ea3a7SDavid Howells 3670716ea3a7SDavid Howells if (skb1->sk) 3671716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3672716ea3a7SDavid Howells 3673716ea3a7SDavid Howells /* Looking around. Are we still alive? 3674716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3675716ea3a7SDavid Howells 3676716ea3a7SDavid Howells skb2->next = skb1->next; 3677716ea3a7SDavid Howells *skb_p = skb2; 3678716ea3a7SDavid Howells kfree_skb(skb1); 3679716ea3a7SDavid Howells skb1 = skb2; 3680716ea3a7SDavid Howells } 3681716ea3a7SDavid Howells elt++; 3682716ea3a7SDavid Howells *trailer = skb1; 3683716ea3a7SDavid Howells skb_p = &skb1->next; 3684716ea3a7SDavid Howells } 3685716ea3a7SDavid Howells 3686716ea3a7SDavid Howells return elt; 3687716ea3a7SDavid Howells } 3688b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3689716ea3a7SDavid Howells 3690b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3691b1faf566SEric Dumazet { 3692b1faf566SEric Dumazet struct sock *sk = skb->sk; 3693b1faf566SEric Dumazet 3694b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3695b1faf566SEric Dumazet } 3696b1faf566SEric Dumazet 36978605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 36988605330aSSoheil Hassas Yeganeh { 36998605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 37008605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 37018605330aSSoheil Hassas Yeganeh */ 37028605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 37038605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 37048605330aSSoheil Hassas Yeganeh } 37058605330aSSoheil Hassas Yeganeh 3706b1faf566SEric Dumazet /* 3707b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3708b1faf566SEric Dumazet */ 3709b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3710b1faf566SEric Dumazet { 3711b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 371295c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 3713b1faf566SEric Dumazet return -ENOMEM; 3714b1faf566SEric Dumazet 3715b1faf566SEric Dumazet skb_orphan(skb); 3716b1faf566SEric Dumazet skb->sk = sk; 3717b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3718b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 37198605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 3720b1faf566SEric Dumazet 3721abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3722abb57ea4SEric Dumazet skb_dst_force(skb); 3723abb57ea4SEric Dumazet 3724b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3725b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3726676d2369SDavid S. Miller sk->sk_data_ready(sk); 3727b1faf566SEric Dumazet return 0; 3728b1faf566SEric Dumazet } 3729b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3730b1faf566SEric Dumazet 373183a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 373283a1a1a7SSoheil Hassas Yeganeh { 373383a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 373483a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 373583a1a1a7SSoheil Hassas Yeganeh } 373683a1a1a7SSoheil Hassas Yeganeh 3737364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 3738364a9e93SWillem de Bruijn { 3739364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 374083a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 374183a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 3742997d5c3fSEric Dumazet unsigned long flags; 3743364a9e93SWillem de Bruijn 3744997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 3745364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 3746364a9e93SWillem de Bruijn if (skb && (skb_next = skb_peek(q))) 374783a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 3748997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 3749364a9e93SWillem de Bruijn 375083a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 375183a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 375283a1a1a7SSoheil Hassas Yeganeh 375383a1a1a7SSoheil Hassas Yeganeh if (skb_next) 3754364a9e93SWillem de Bruijn sk->sk_error_report(sk); 3755364a9e93SWillem de Bruijn 3756364a9e93SWillem de Bruijn return skb; 3757364a9e93SWillem de Bruijn } 3758364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 3759364a9e93SWillem de Bruijn 3760cab41c47SAlexander Duyck /** 3761cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 3762cab41c47SAlexander Duyck * @skb: the skb to clone 3763cab41c47SAlexander Duyck * 3764cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 3765cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 3766cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 3767cab41c47SAlexander Duyck * 3768cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 3769cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 3770cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 3771cab41c47SAlexander Duyck * the sk_error_queue. 3772cab41c47SAlexander Duyck */ 377362bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 377462bccb8cSAlexander Duyck { 377562bccb8cSAlexander Duyck struct sock *sk = skb->sk; 377662bccb8cSAlexander Duyck struct sk_buff *clone; 377762bccb8cSAlexander Duyck 377862bccb8cSAlexander Duyck if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt)) 377962bccb8cSAlexander Duyck return NULL; 378062bccb8cSAlexander Duyck 378162bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 378262bccb8cSAlexander Duyck if (!clone) { 378362bccb8cSAlexander Duyck sock_put(sk); 378462bccb8cSAlexander Duyck return NULL; 378562bccb8cSAlexander Duyck } 378662bccb8cSAlexander Duyck 378762bccb8cSAlexander Duyck clone->sk = sk; 378862bccb8cSAlexander Duyck clone->destructor = sock_efree; 378962bccb8cSAlexander Duyck 379062bccb8cSAlexander Duyck return clone; 379162bccb8cSAlexander Duyck } 379262bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 379362bccb8cSAlexander Duyck 379437846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 379537846ef0SAlexander Duyck struct sock *sk, 379637846ef0SAlexander Duyck int tstype) 3797ac45f602SPatrick Ohly { 3798ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3799ac45f602SPatrick Ohly int err; 3800ac45f602SPatrick Ohly 3801ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3802ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3803ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3804ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 3805e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 38064ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 380709c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 3808ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 3809ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 38104ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 38114ed2d765SWillem de Bruijn } 381229030374SEric Dumazet 3813ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 381429030374SEric Dumazet 3815ac45f602SPatrick Ohly if (err) 3816ac45f602SPatrick Ohly kfree_skb(skb); 3817ac45f602SPatrick Ohly } 381837846ef0SAlexander Duyck 3819b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 3820b245be1fSWillem de Bruijn { 3821b245be1fSWillem de Bruijn bool ret; 3822b245be1fSWillem de Bruijn 3823b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 3824b245be1fSWillem de Bruijn return true; 3825b245be1fSWillem de Bruijn 3826b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 3827b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 3828b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 3829b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 3830b245be1fSWillem de Bruijn return ret; 3831b245be1fSWillem de Bruijn } 3832b245be1fSWillem de Bruijn 383337846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 383437846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 383537846ef0SAlexander Duyck { 383637846ef0SAlexander Duyck struct sock *sk = skb->sk; 383737846ef0SAlexander Duyck 3838b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 3839b245be1fSWillem de Bruijn return; 3840b245be1fSWillem de Bruijn 38419ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 38429ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 38439ac25fc0SEric Dumazet */ 38449ac25fc0SEric Dumazet if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) { 384537846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 384637846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND); 384737846ef0SAlexander Duyck sock_put(sk); 384837846ef0SAlexander Duyck } 38499ac25fc0SEric Dumazet } 385037846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 385137846ef0SAlexander Duyck 385237846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 385337846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 385437846ef0SAlexander Duyck struct sock *sk, int tstype) 385537846ef0SAlexander Duyck { 385637846ef0SAlexander Duyck struct sk_buff *skb; 38573a8dd971SWillem de Bruijn bool tsonly; 385837846ef0SAlexander Duyck 38593a8dd971SWillem de Bruijn if (!sk) 38603a8dd971SWillem de Bruijn return; 38613a8dd971SWillem de Bruijn 38623a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 38633a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 386437846ef0SAlexander Duyck return; 386537846ef0SAlexander Duyck 38661c885808SFrancis Yan if (tsonly) { 38671c885808SFrancis Yan #ifdef CONFIG_INET 38681c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 38691c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 38701c885808SFrancis Yan sk->sk_type == SOCK_STREAM) 38711c885808SFrancis Yan skb = tcp_get_timestamping_opt_stats(sk); 387237846ef0SAlexander Duyck else 38731c885808SFrancis Yan #endif 38741c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 38751c885808SFrancis Yan } else { 387637846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 38771c885808SFrancis Yan } 387837846ef0SAlexander Duyck if (!skb) 387937846ef0SAlexander Duyck return; 388037846ef0SAlexander Duyck 388149ca0d8bSWillem de Bruijn if (tsonly) { 388249ca0d8bSWillem de Bruijn skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags; 388349ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 388449ca0d8bSWillem de Bruijn } 388549ca0d8bSWillem de Bruijn 388649ca0d8bSWillem de Bruijn if (hwtstamps) 388749ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 388849ca0d8bSWillem de Bruijn else 388949ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 389049ca0d8bSWillem de Bruijn 389137846ef0SAlexander Duyck __skb_complete_tx_timestamp(skb, sk, tstype); 389237846ef0SAlexander Duyck } 3893e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 3894e7fd2885SWillem de Bruijn 3895e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 3896e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 3897e7fd2885SWillem de Bruijn { 3898e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 3899e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 3900e7fd2885SWillem de Bruijn } 3901ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3902ac45f602SPatrick Ohly 39036e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 39046e3e939fSJohannes Berg { 39056e3e939fSJohannes Berg struct sock *sk = skb->sk; 39066e3e939fSJohannes Berg struct sock_exterr_skb *serr; 3907dd4f1072SEric Dumazet int err = 1; 39086e3e939fSJohannes Berg 39096e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 39106e3e939fSJohannes Berg skb->wifi_acked = acked; 39116e3e939fSJohannes Berg 39126e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 39136e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 39146e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 39156e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 39166e3e939fSJohannes Berg 3917dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 3918dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 3919dd4f1072SEric Dumazet */ 3920dd4f1072SEric Dumazet if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) { 39216e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 3922dd4f1072SEric Dumazet sock_put(sk); 3923dd4f1072SEric Dumazet } 39246e3e939fSJohannes Berg if (err) 39256e3e939fSJohannes Berg kfree_skb(skb); 39266e3e939fSJohannes Berg } 39276e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 39286e3e939fSJohannes Berg 3929f35d9d8aSRusty Russell /** 3930f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3931f35d9d8aSRusty Russell * @skb: the skb to set 3932f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3933f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3934f35d9d8aSRusty Russell * 3935f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3936f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3937f35d9d8aSRusty Russell * 3938f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3939f35d9d8aSRusty Russell * returns false you should drop the packet. 3940f35d9d8aSRusty Russell */ 3941f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3942f35d9d8aSRusty Russell { 39435ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 39445ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3945e87cc472SJoe Perches net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 39465ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3947f35d9d8aSRusty Russell return false; 3948f35d9d8aSRusty Russell } 3949f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3950f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3951f35d9d8aSRusty Russell skb->csum_offset = off; 3952e5d5decaSJason Wang skb_set_transport_header(skb, start); 3953f35d9d8aSRusty Russell return true; 3954f35d9d8aSRusty Russell } 3955b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3956f35d9d8aSRusty Russell 3957ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 3958ed1f50c3SPaul Durrant unsigned int max) 3959ed1f50c3SPaul Durrant { 3960ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 3961ed1f50c3SPaul Durrant return 0; 3962ed1f50c3SPaul Durrant 3963ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 3964ed1f50c3SPaul Durrant * won't need to do it again. 3965ed1f50c3SPaul Durrant */ 3966ed1f50c3SPaul Durrant if (max > skb->len) 3967ed1f50c3SPaul Durrant max = skb->len; 3968ed1f50c3SPaul Durrant 3969ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 3970ed1f50c3SPaul Durrant return -ENOMEM; 3971ed1f50c3SPaul Durrant 3972ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 3973ed1f50c3SPaul Durrant return -EPROTO; 3974ed1f50c3SPaul Durrant 3975ed1f50c3SPaul Durrant return 0; 3976ed1f50c3SPaul Durrant } 3977ed1f50c3SPaul Durrant 3978f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 3979f9708b43SJan Beulich 3980f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 3981f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 3982f9708b43SJan Beulich unsigned int off) 3983f9708b43SJan Beulich { 3984f9708b43SJan Beulich switch (proto) { 3985f9708b43SJan Beulich int err; 3986f9708b43SJan Beulich 3987f9708b43SJan Beulich case IPPROTO_TCP: 3988f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 3989f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 3990f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 3991f9708b43SJan Beulich offsetof(struct tcphdr, 3992f9708b43SJan Beulich check))) 3993f9708b43SJan Beulich err = -EPROTO; 3994f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 3995f9708b43SJan Beulich 3996f9708b43SJan Beulich case IPPROTO_UDP: 3997f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 3998f9708b43SJan Beulich off + sizeof(struct udphdr)); 3999f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4000f9708b43SJan Beulich offsetof(struct udphdr, 4001f9708b43SJan Beulich check))) 4002f9708b43SJan Beulich err = -EPROTO; 4003f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4004f9708b43SJan Beulich } 4005f9708b43SJan Beulich 4006f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4007f9708b43SJan Beulich } 4008f9708b43SJan Beulich 4009ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4010ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4011ed1f50c3SPaul Durrant */ 4012ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4013ed1f50c3SPaul Durrant 4014f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4015ed1f50c3SPaul Durrant { 4016ed1f50c3SPaul Durrant unsigned int off; 4017ed1f50c3SPaul Durrant bool fragment; 4018f9708b43SJan Beulich __sum16 *csum; 4019ed1f50c3SPaul Durrant int err; 4020ed1f50c3SPaul Durrant 4021ed1f50c3SPaul Durrant fragment = false; 4022ed1f50c3SPaul Durrant 4023ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4024ed1f50c3SPaul Durrant sizeof(struct iphdr), 4025ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4026ed1f50c3SPaul Durrant if (err < 0) 4027ed1f50c3SPaul Durrant goto out; 4028ed1f50c3SPaul Durrant 4029ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 4030ed1f50c3SPaul Durrant fragment = true; 4031ed1f50c3SPaul Durrant 4032ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 4033ed1f50c3SPaul Durrant 4034ed1f50c3SPaul Durrant err = -EPROTO; 4035ed1f50c3SPaul Durrant 4036ed1f50c3SPaul Durrant if (fragment) 4037ed1f50c3SPaul Durrant goto out; 4038ed1f50c3SPaul Durrant 4039f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4040f9708b43SJan Beulich if (IS_ERR(csum)) 4041f9708b43SJan Beulich return PTR_ERR(csum); 4042ed1f50c3SPaul Durrant 4043ed1f50c3SPaul Durrant if (recalculate) 4044f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4045ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4046ed1f50c3SPaul Durrant skb->len - off, 4047f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4048ed1f50c3SPaul Durrant err = 0; 4049ed1f50c3SPaul Durrant 4050ed1f50c3SPaul Durrant out: 4051ed1f50c3SPaul Durrant return err; 4052ed1f50c3SPaul Durrant } 4053ed1f50c3SPaul Durrant 4054ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4055ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4056ed1f50c3SPaul Durrant */ 4057ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4058ed1f50c3SPaul Durrant 4059ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4060ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4061ed1f50c3SPaul Durrant 4062ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4063ed1f50c3SPaul Durrant { 4064ed1f50c3SPaul Durrant int err; 4065ed1f50c3SPaul Durrant u8 nexthdr; 4066ed1f50c3SPaul Durrant unsigned int off; 4067ed1f50c3SPaul Durrant unsigned int len; 4068ed1f50c3SPaul Durrant bool fragment; 4069ed1f50c3SPaul Durrant bool done; 4070f9708b43SJan Beulich __sum16 *csum; 4071ed1f50c3SPaul Durrant 4072ed1f50c3SPaul Durrant fragment = false; 4073ed1f50c3SPaul Durrant done = false; 4074ed1f50c3SPaul Durrant 4075ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4076ed1f50c3SPaul Durrant 4077ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4078ed1f50c3SPaul Durrant if (err < 0) 4079ed1f50c3SPaul Durrant goto out; 4080ed1f50c3SPaul Durrant 4081ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4082ed1f50c3SPaul Durrant 4083ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4084ed1f50c3SPaul Durrant while (off <= len && !done) { 4085ed1f50c3SPaul Durrant switch (nexthdr) { 4086ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4087ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4088ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4089ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 4090ed1f50c3SPaul Durrant 4091ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4092ed1f50c3SPaul Durrant off + 4093ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 4094ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4095ed1f50c3SPaul Durrant if (err < 0) 4096ed1f50c3SPaul Durrant goto out; 4097ed1f50c3SPaul Durrant 4098ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4099ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4100ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 4101ed1f50c3SPaul Durrant break; 4102ed1f50c3SPaul Durrant } 4103ed1f50c3SPaul Durrant case IPPROTO_AH: { 4104ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 4105ed1f50c3SPaul Durrant 4106ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4107ed1f50c3SPaul Durrant off + 4108ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 4109ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4110ed1f50c3SPaul Durrant if (err < 0) 4111ed1f50c3SPaul Durrant goto out; 4112ed1f50c3SPaul Durrant 4113ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4114ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4115ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 4116ed1f50c3SPaul Durrant break; 4117ed1f50c3SPaul Durrant } 4118ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 4119ed1f50c3SPaul Durrant struct frag_hdr *hp; 4120ed1f50c3SPaul Durrant 4121ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4122ed1f50c3SPaul Durrant off + 4123ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 4124ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4125ed1f50c3SPaul Durrant if (err < 0) 4126ed1f50c3SPaul Durrant goto out; 4127ed1f50c3SPaul Durrant 4128ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 4129ed1f50c3SPaul Durrant 4130ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 4131ed1f50c3SPaul Durrant fragment = true; 4132ed1f50c3SPaul Durrant 4133ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4134ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 4135ed1f50c3SPaul Durrant break; 4136ed1f50c3SPaul Durrant } 4137ed1f50c3SPaul Durrant default: 4138ed1f50c3SPaul Durrant done = true; 4139ed1f50c3SPaul Durrant break; 4140ed1f50c3SPaul Durrant } 4141ed1f50c3SPaul Durrant } 4142ed1f50c3SPaul Durrant 4143ed1f50c3SPaul Durrant err = -EPROTO; 4144ed1f50c3SPaul Durrant 4145ed1f50c3SPaul Durrant if (!done || fragment) 4146ed1f50c3SPaul Durrant goto out; 4147ed1f50c3SPaul Durrant 4148f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 4149f9708b43SJan Beulich if (IS_ERR(csum)) 4150f9708b43SJan Beulich return PTR_ERR(csum); 4151ed1f50c3SPaul Durrant 4152ed1f50c3SPaul Durrant if (recalculate) 4153f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 4154ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 4155f9708b43SJan Beulich skb->len - off, nexthdr, 0); 4156ed1f50c3SPaul Durrant err = 0; 4157ed1f50c3SPaul Durrant 4158ed1f50c3SPaul Durrant out: 4159ed1f50c3SPaul Durrant return err; 4160ed1f50c3SPaul Durrant } 4161ed1f50c3SPaul Durrant 4162ed1f50c3SPaul Durrant /** 4163ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 4164ed1f50c3SPaul Durrant * @skb: the skb to set up 4165ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 4166ed1f50c3SPaul Durrant */ 4167ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 4168ed1f50c3SPaul Durrant { 4169ed1f50c3SPaul Durrant int err; 4170ed1f50c3SPaul Durrant 4171ed1f50c3SPaul Durrant switch (skb->protocol) { 4172ed1f50c3SPaul Durrant case htons(ETH_P_IP): 4173f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 4174ed1f50c3SPaul Durrant break; 4175ed1f50c3SPaul Durrant 4176ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 4177ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 4178ed1f50c3SPaul Durrant break; 4179ed1f50c3SPaul Durrant 4180ed1f50c3SPaul Durrant default: 4181ed1f50c3SPaul Durrant err = -EPROTO; 4182ed1f50c3SPaul Durrant break; 4183ed1f50c3SPaul Durrant } 4184ed1f50c3SPaul Durrant 4185ed1f50c3SPaul Durrant return err; 4186ed1f50c3SPaul Durrant } 4187ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 4188ed1f50c3SPaul Durrant 41899afd85c9SLinus Lüssing /** 41909afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 41919afd85c9SLinus Lüssing * @skb: the skb to check 41929afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 41939afd85c9SLinus Lüssing * 41949afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 41959afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 41969afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 41979afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 41989afd85c9SLinus Lüssing * 4199a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4200a516993fSLinus Lüssing * differs from the provided skb. 42019afd85c9SLinus Lüssing */ 42029afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 42039afd85c9SLinus Lüssing unsigned int transport_len) 42049afd85c9SLinus Lüssing { 42059afd85c9SLinus Lüssing struct sk_buff *skb_chk; 42069afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 42079afd85c9SLinus Lüssing int ret; 42089afd85c9SLinus Lüssing 4209a516993fSLinus Lüssing if (skb->len < len) 42109afd85c9SLinus Lüssing return NULL; 4211a516993fSLinus Lüssing else if (skb->len == len) 42129afd85c9SLinus Lüssing return skb; 42139afd85c9SLinus Lüssing 42149afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 42159afd85c9SLinus Lüssing if (!skb_chk) 42169afd85c9SLinus Lüssing return NULL; 42179afd85c9SLinus Lüssing 42189afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 42199afd85c9SLinus Lüssing if (ret) { 42209afd85c9SLinus Lüssing kfree_skb(skb_chk); 42219afd85c9SLinus Lüssing return NULL; 42229afd85c9SLinus Lüssing } 42239afd85c9SLinus Lüssing 42249afd85c9SLinus Lüssing return skb_chk; 42259afd85c9SLinus Lüssing } 42269afd85c9SLinus Lüssing 42279afd85c9SLinus Lüssing /** 42289afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 42299afd85c9SLinus Lüssing * @skb: the skb to check 42309afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 42319afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 42329afd85c9SLinus Lüssing * 42339afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 42349afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 42359afd85c9SLinus Lüssing * 42369afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 42379afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 42389afd85c9SLinus Lüssing * 4239a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4240a516993fSLinus Lüssing * differs from the provided skb. 42419afd85c9SLinus Lüssing */ 42429afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 42439afd85c9SLinus Lüssing unsigned int transport_len, 42449afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 42459afd85c9SLinus Lüssing { 42469afd85c9SLinus Lüssing struct sk_buff *skb_chk; 42479afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 4248fcba67c9SLinus Lüssing __sum16 ret; 42499afd85c9SLinus Lüssing 42509afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 42519afd85c9SLinus Lüssing if (!skb_chk) 4252a516993fSLinus Lüssing goto err; 42539afd85c9SLinus Lüssing 4254a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 4255a516993fSLinus Lüssing goto err; 42569afd85c9SLinus Lüssing 42579b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 42589afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 42599b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 42609afd85c9SLinus Lüssing 4261a516993fSLinus Lüssing if (ret) 4262a516993fSLinus Lüssing goto err; 42639afd85c9SLinus Lüssing 42649afd85c9SLinus Lüssing return skb_chk; 4265a516993fSLinus Lüssing 4266a516993fSLinus Lüssing err: 4267a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 4268a516993fSLinus Lüssing kfree_skb(skb_chk); 4269a516993fSLinus Lüssing 4270a516993fSLinus Lüssing return NULL; 4271a516993fSLinus Lüssing 42729afd85c9SLinus Lüssing } 42739afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 42749afd85c9SLinus Lüssing 42754497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 42764497b076SBen Hutchings { 4277e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 4278e87cc472SJoe Perches skb->dev->name); 42794497b076SBen Hutchings } 42804497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 4281bad43ca8SEric Dumazet 4282bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 4283bad43ca8SEric Dumazet { 42843d861f66SEric Dumazet if (head_stolen) { 42853d861f66SEric Dumazet skb_release_head_state(skb); 4286bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 42873d861f66SEric Dumazet } else { 4288bad43ca8SEric Dumazet __kfree_skb(skb); 4289bad43ca8SEric Dumazet } 42903d861f66SEric Dumazet } 4291bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 4292bad43ca8SEric Dumazet 4293bad43ca8SEric Dumazet /** 4294bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 4295bad43ca8SEric Dumazet * @to: prior buffer 4296bad43ca8SEric Dumazet * @from: buffer to add 4297bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 4298c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 4299bad43ca8SEric Dumazet */ 4300bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 4301bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 4302bad43ca8SEric Dumazet { 4303bad43ca8SEric Dumazet int i, delta, len = from->len; 4304bad43ca8SEric Dumazet 4305bad43ca8SEric Dumazet *fragstolen = false; 4306bad43ca8SEric Dumazet 4307bad43ca8SEric Dumazet if (skb_cloned(to)) 4308bad43ca8SEric Dumazet return false; 4309bad43ca8SEric Dumazet 4310bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 4311e93a0435SEric Dumazet if (len) 4312bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 4313bad43ca8SEric Dumazet *delta_truesize = 0; 4314bad43ca8SEric Dumazet return true; 4315bad43ca8SEric Dumazet } 4316bad43ca8SEric Dumazet 4317bad43ca8SEric Dumazet if (skb_has_frag_list(to) || skb_has_frag_list(from)) 4318bad43ca8SEric Dumazet return false; 4319bad43ca8SEric Dumazet 4320bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 4321bad43ca8SEric Dumazet struct page *page; 4322bad43ca8SEric Dumazet unsigned int offset; 4323bad43ca8SEric Dumazet 4324bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4325bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 4326bad43ca8SEric Dumazet return false; 4327bad43ca8SEric Dumazet 4328bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 4329bad43ca8SEric Dumazet return false; 4330bad43ca8SEric Dumazet 4331bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4332bad43ca8SEric Dumazet 4333bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 4334bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 4335bad43ca8SEric Dumazet 4336bad43ca8SEric Dumazet skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 4337bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 4338bad43ca8SEric Dumazet *fragstolen = true; 4339bad43ca8SEric Dumazet } else { 4340bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 4341bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 4342bad43ca8SEric Dumazet return false; 4343bad43ca8SEric Dumazet 4344f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 4345bad43ca8SEric Dumazet } 4346bad43ca8SEric Dumazet 4347bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 4348bad43ca8SEric Dumazet 4349bad43ca8SEric Dumazet memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 4350bad43ca8SEric Dumazet skb_shinfo(from)->frags, 4351bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 4352bad43ca8SEric Dumazet skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 4353bad43ca8SEric Dumazet 4354bad43ca8SEric Dumazet if (!skb_cloned(from)) 4355bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags = 0; 4356bad43ca8SEric Dumazet 43578ea853fdSLi RongQing /* if the skb is not cloned this does nothing 43588ea853fdSLi RongQing * since we set nr_frags to 0. 43598ea853fdSLi RongQing */ 4360bad43ca8SEric Dumazet for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 4361bad43ca8SEric Dumazet skb_frag_ref(from, i); 4362bad43ca8SEric Dumazet 4363bad43ca8SEric Dumazet to->truesize += delta; 4364bad43ca8SEric Dumazet to->len += len; 4365bad43ca8SEric Dumazet to->data_len += len; 4366bad43ca8SEric Dumazet 4367bad43ca8SEric Dumazet *delta_truesize = delta; 4368bad43ca8SEric Dumazet return true; 4369bad43ca8SEric Dumazet } 4370bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 4371621e84d6SNicolas Dichtel 4372621e84d6SNicolas Dichtel /** 43738b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 4374621e84d6SNicolas Dichtel * 4375621e84d6SNicolas Dichtel * @skb: buffer to clean 43768b27f277SNicolas Dichtel * @xnet: packet is crossing netns 4377621e84d6SNicolas Dichtel * 43788b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 43798b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 43808b27f277SNicolas Dichtel * operations. 43818b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 43828b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 43838b27f277SNicolas Dichtel * skb that could impact namespace isolation. 4384621e84d6SNicolas Dichtel */ 43858b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4386621e84d6SNicolas Dichtel { 43872456e855SThomas Gleixner skb->tstamp = 0; 4388621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 4389621e84d6SNicolas Dichtel skb->skb_iif = 0; 439060ff7467SWANG Cong skb->ignore_df = 0; 4391621e84d6SNicolas Dichtel skb_dst_drop(skb); 4392621e84d6SNicolas Dichtel secpath_reset(skb); 4393621e84d6SNicolas Dichtel nf_reset(skb); 4394621e84d6SNicolas Dichtel nf_reset_trace(skb); 4395213dd74aSHerbert Xu 4396213dd74aSHerbert Xu if (!xnet) 4397213dd74aSHerbert Xu return; 4398213dd74aSHerbert Xu 4399213dd74aSHerbert Xu skb_orphan(skb); 4400213dd74aSHerbert Xu skb->mark = 0; 4401621e84d6SNicolas Dichtel } 4402621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 4403de960aa9SFlorian Westphal 4404de960aa9SFlorian Westphal /** 4405de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4406de960aa9SFlorian Westphal * 4407de960aa9SFlorian Westphal * @skb: GSO skb 4408de960aa9SFlorian Westphal * 4409de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 4410de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 4411de960aa9SFlorian Westphal * 4412de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4413de960aa9SFlorian Westphal */ 4414de960aa9SFlorian Westphal unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4415de960aa9SFlorian Westphal { 4416de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 4417f993bc25SFlorian Westphal unsigned int thlen = 0; 4418f993bc25SFlorian Westphal 4419f993bc25SFlorian Westphal if (skb->encapsulation) { 4420f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 4421f993bc25SFlorian Westphal skb_transport_header(skb); 4422de960aa9SFlorian Westphal 4423de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 4424f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 4425f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 4426f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 442790017accSMarcelo Ricardo Leitner } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) { 442890017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 4429f993bc25SFlorian Westphal } 44306d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 44316d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 44326d39d589SFlorian Westphal * accounted for. 44336d39d589SFlorian Westphal */ 4434f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 4435de960aa9SFlorian Westphal } 4436de960aa9SFlorian Westphal EXPORT_SYMBOL_GPL(skb_gso_transport_seglen); 44370d5501c1SVlad Yasevich 4438ae7ef81eSMarcelo Ricardo Leitner /** 4439ae7ef81eSMarcelo Ricardo Leitner * skb_gso_validate_mtu - Return in case such skb fits a given MTU 4440ae7ef81eSMarcelo Ricardo Leitner * 4441ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 444276f21b99SDavid S. Miller * @mtu: MTU to validate against 4443ae7ef81eSMarcelo Ricardo Leitner * 4444ae7ef81eSMarcelo Ricardo Leitner * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU 4445ae7ef81eSMarcelo Ricardo Leitner * once split. 4446ae7ef81eSMarcelo Ricardo Leitner */ 4447ae7ef81eSMarcelo Ricardo Leitner bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu) 4448ae7ef81eSMarcelo Ricardo Leitner { 4449ae7ef81eSMarcelo Ricardo Leitner const struct skb_shared_info *shinfo = skb_shinfo(skb); 4450ae7ef81eSMarcelo Ricardo Leitner const struct sk_buff *iter; 4451ae7ef81eSMarcelo Ricardo Leitner unsigned int hlen; 4452ae7ef81eSMarcelo Ricardo Leitner 4453ae7ef81eSMarcelo Ricardo Leitner hlen = skb_gso_network_seglen(skb); 4454ae7ef81eSMarcelo Ricardo Leitner 4455ae7ef81eSMarcelo Ricardo Leitner if (shinfo->gso_size != GSO_BY_FRAGS) 4456ae7ef81eSMarcelo Ricardo Leitner return hlen <= mtu; 4457ae7ef81eSMarcelo Ricardo Leitner 4458ae7ef81eSMarcelo Ricardo Leitner /* Undo this so we can re-use header sizes */ 4459ae7ef81eSMarcelo Ricardo Leitner hlen -= GSO_BY_FRAGS; 4460ae7ef81eSMarcelo Ricardo Leitner 4461ae7ef81eSMarcelo Ricardo Leitner skb_walk_frags(skb, iter) { 4462ae7ef81eSMarcelo Ricardo Leitner if (hlen + skb_headlen(iter) > mtu) 4463ae7ef81eSMarcelo Ricardo Leitner return false; 4464ae7ef81eSMarcelo Ricardo Leitner } 4465ae7ef81eSMarcelo Ricardo Leitner 4466ae7ef81eSMarcelo Ricardo Leitner return true; 4467ae7ef81eSMarcelo Ricardo Leitner } 4468ae7ef81eSMarcelo Ricardo Leitner EXPORT_SYMBOL_GPL(skb_gso_validate_mtu); 4469ae7ef81eSMarcelo Ricardo Leitner 44700d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 44710d5501c1SVlad Yasevich { 44720d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 44730d5501c1SVlad Yasevich kfree_skb(skb); 44740d5501c1SVlad Yasevich return NULL; 44750d5501c1SVlad Yasevich } 44760d5501c1SVlad Yasevich 4477f6548615SVlad Yasevich memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN, 4478a6e18ff1SVlad Yasevich 2 * ETH_ALEN); 44790d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 44800d5501c1SVlad Yasevich return skb; 44810d5501c1SVlad Yasevich } 44820d5501c1SVlad Yasevich 44830d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 44840d5501c1SVlad Yasevich { 44850d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 44860d5501c1SVlad Yasevich u16 vlan_tci; 44870d5501c1SVlad Yasevich 4488df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 44890d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 44900d5501c1SVlad Yasevich return skb; 44910d5501c1SVlad Yasevich } 44920d5501c1SVlad Yasevich 44930d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 44940d5501c1SVlad Yasevich if (unlikely(!skb)) 44950d5501c1SVlad Yasevich goto err_free; 44960d5501c1SVlad Yasevich 44970d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 44980d5501c1SVlad Yasevich goto err_free; 44990d5501c1SVlad Yasevich 45000d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 45010d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 45020d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 45030d5501c1SVlad Yasevich 45040d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 45050d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 45060d5501c1SVlad Yasevich 45070d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 45080d5501c1SVlad Yasevich if (unlikely(!skb)) 45090d5501c1SVlad Yasevich goto err_free; 45100d5501c1SVlad Yasevich 45110d5501c1SVlad Yasevich skb_reset_network_header(skb); 45120d5501c1SVlad Yasevich skb_reset_transport_header(skb); 45130d5501c1SVlad Yasevich skb_reset_mac_len(skb); 45140d5501c1SVlad Yasevich 45150d5501c1SVlad Yasevich return skb; 45160d5501c1SVlad Yasevich 45170d5501c1SVlad Yasevich err_free: 45180d5501c1SVlad Yasevich kfree_skb(skb); 45190d5501c1SVlad Yasevich return NULL; 45200d5501c1SVlad Yasevich } 45210d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 45222e4e4410SEric Dumazet 4523e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 4524e2195121SJiri Pirko { 4525e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 4526e2195121SJiri Pirko return -ENOMEM; 4527e2195121SJiri Pirko 4528e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 4529e2195121SJiri Pirko return 0; 4530e2195121SJiri Pirko 4531e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 4532e2195121SJiri Pirko } 4533e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 4534e2195121SJiri Pirko 4535bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 4536bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 4537bfca4c52SShmulik Ladkani */ 4538bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 453993515d53SJiri Pirko { 454093515d53SJiri Pirko struct vlan_hdr *vhdr; 4541b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 454293515d53SJiri Pirko int err; 454393515d53SJiri Pirko 4544b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 4545b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 4546b6a79208SShmulik Ladkani offset)) { 4547b6a79208SShmulik Ladkani return -EINVAL; 4548b6a79208SShmulik Ladkani } 4549b6a79208SShmulik Ladkani 455093515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 455193515d53SJiri Pirko if (unlikely(err)) 4552b6a79208SShmulik Ladkani return err; 455393515d53SJiri Pirko 455493515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 455593515d53SJiri Pirko 455693515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 455793515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 455893515d53SJiri Pirko 455993515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 456093515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 456193515d53SJiri Pirko 456293515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 456393515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 456493515d53SJiri Pirko 456593515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 456693515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 456793515d53SJiri Pirko 456893515d53SJiri Pirko skb_reset_mac_len(skb); 456993515d53SJiri Pirko 457093515d53SJiri Pirko return err; 457193515d53SJiri Pirko } 4572bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 457393515d53SJiri Pirko 4574b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 4575b6a79208SShmulik Ladkani * Expects skb->data at mac header. 4576b6a79208SShmulik Ladkani */ 457793515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 457893515d53SJiri Pirko { 457993515d53SJiri Pirko u16 vlan_tci; 458093515d53SJiri Pirko __be16 vlan_proto; 458193515d53SJiri Pirko int err; 458293515d53SJiri Pirko 4583df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 458493515d53SJiri Pirko skb->vlan_tci = 0; 458593515d53SJiri Pirko } else { 4586ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 458793515d53SJiri Pirko return 0; 458893515d53SJiri Pirko 458993515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 459093515d53SJiri Pirko if (err) 459193515d53SJiri Pirko return err; 459293515d53SJiri Pirko } 459393515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 4594ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 459593515d53SJiri Pirko return 0; 459693515d53SJiri Pirko 459793515d53SJiri Pirko vlan_proto = skb->protocol; 459893515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 459993515d53SJiri Pirko if (unlikely(err)) 460093515d53SJiri Pirko return err; 460193515d53SJiri Pirko 460293515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 460393515d53SJiri Pirko return 0; 460493515d53SJiri Pirko } 460593515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 460693515d53SJiri Pirko 4607b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 4608b6a79208SShmulik Ladkani * Expects skb->data at mac header. 4609b6a79208SShmulik Ladkani */ 461093515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 461193515d53SJiri Pirko { 4612df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 4613b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 461493515d53SJiri Pirko int err; 461593515d53SJiri Pirko 4616b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 4617b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 4618b6a79208SShmulik Ladkani offset)) { 4619b6a79208SShmulik Ladkani return -EINVAL; 4620b6a79208SShmulik Ladkani } 4621b6a79208SShmulik Ladkani 462293515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 4623df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 4624b6a79208SShmulik Ladkani if (err) 462593515d53SJiri Pirko return err; 46269241e2dfSDaniel Borkmann 462793515d53SJiri Pirko skb->protocol = skb->vlan_proto; 462893515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 462993515d53SJiri Pirko 46306b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 463193515d53SJiri Pirko } 463293515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 463393515d53SJiri Pirko return 0; 463493515d53SJiri Pirko } 463593515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 463693515d53SJiri Pirko 46372e4e4410SEric Dumazet /** 46382e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 46392e4e4410SEric Dumazet * 4640de3f0d0eSMasanari Iida * @header_len: size of linear part 4641de3f0d0eSMasanari Iida * @data_len: needed length in frags 4642de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 4643de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 4644de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 46452e4e4410SEric Dumazet * 46462e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 46472e4e4410SEric Dumazet */ 46482e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 46492e4e4410SEric Dumazet unsigned long data_len, 46502e4e4410SEric Dumazet int max_page_order, 46512e4e4410SEric Dumazet int *errcode, 46522e4e4410SEric Dumazet gfp_t gfp_mask) 46532e4e4410SEric Dumazet { 46542e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 46552e4e4410SEric Dumazet unsigned long chunk; 46562e4e4410SEric Dumazet struct sk_buff *skb; 46572e4e4410SEric Dumazet struct page *page; 46582e4e4410SEric Dumazet gfp_t gfp_head; 46592e4e4410SEric Dumazet int i; 46602e4e4410SEric Dumazet 46612e4e4410SEric Dumazet *errcode = -EMSGSIZE; 46622e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 46632e4e4410SEric Dumazet * high order pages... 46642e4e4410SEric Dumazet */ 46652e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 46662e4e4410SEric Dumazet return NULL; 46672e4e4410SEric Dumazet 46682e4e4410SEric Dumazet gfp_head = gfp_mask; 4669d0164adcSMel Gorman if (gfp_head & __GFP_DIRECT_RECLAIM) 46702e4e4410SEric Dumazet gfp_head |= __GFP_REPEAT; 46712e4e4410SEric Dumazet 46722e4e4410SEric Dumazet *errcode = -ENOBUFS; 46732e4e4410SEric Dumazet skb = alloc_skb(header_len, gfp_head); 46742e4e4410SEric Dumazet if (!skb) 46752e4e4410SEric Dumazet return NULL; 46762e4e4410SEric Dumazet 46772e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 46782e4e4410SEric Dumazet 46792e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 46802e4e4410SEric Dumazet int order = max_page_order; 46812e4e4410SEric Dumazet 46822e4e4410SEric Dumazet while (order) { 46832e4e4410SEric Dumazet if (npages >= 1 << order) { 4684d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 46852e4e4410SEric Dumazet __GFP_COMP | 46862e4e4410SEric Dumazet __GFP_NOWARN | 46872e4e4410SEric Dumazet __GFP_NORETRY, 46882e4e4410SEric Dumazet order); 46892e4e4410SEric Dumazet if (page) 46902e4e4410SEric Dumazet goto fill_page; 46912e4e4410SEric Dumazet /* Do not retry other high order allocations */ 46922e4e4410SEric Dumazet order = 1; 46932e4e4410SEric Dumazet max_page_order = 0; 46942e4e4410SEric Dumazet } 46952e4e4410SEric Dumazet order--; 46962e4e4410SEric Dumazet } 46972e4e4410SEric Dumazet page = alloc_page(gfp_mask); 46982e4e4410SEric Dumazet if (!page) 46992e4e4410SEric Dumazet goto failure; 47002e4e4410SEric Dumazet fill_page: 47012e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 47022e4e4410SEric Dumazet PAGE_SIZE << order); 47032e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 47042e4e4410SEric Dumazet data_len -= chunk; 47052e4e4410SEric Dumazet npages -= 1 << order; 47062e4e4410SEric Dumazet } 47072e4e4410SEric Dumazet return skb; 47082e4e4410SEric Dumazet 47092e4e4410SEric Dumazet failure: 47102e4e4410SEric Dumazet kfree_skb(skb); 47112e4e4410SEric Dumazet return NULL; 47122e4e4410SEric Dumazet } 47132e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 47146fa01ccdSSowmini Varadhan 47156fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 47166fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 47176fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 47186fa01ccdSSowmini Varadhan { 47196fa01ccdSSowmini Varadhan int i; 47206fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 47216fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 47226fa01ccdSSowmini Varadhan u8 *data; 47236fa01ccdSSowmini Varadhan 47246fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 47256fa01ccdSSowmini Varadhan 47266fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 47276fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 47286fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 47296fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 47306fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 47316fa01ccdSSowmini Varadhan if (!data) 47326fa01ccdSSowmini Varadhan return -ENOMEM; 47336fa01ccdSSowmini Varadhan 47346fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 47356fa01ccdSSowmini Varadhan 47366fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 47376fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 47386fa01ccdSSowmini Varadhan skb->len -= off; 47396fa01ccdSSowmini Varadhan 47406fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 47416fa01ccdSSowmini Varadhan skb_shinfo(skb), 47426fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 47436fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 47446fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 47456fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 47466fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 47476fa01ccdSSowmini Varadhan kfree(data); 47486fa01ccdSSowmini Varadhan return -ENOMEM; 47496fa01ccdSSowmini Varadhan } 47506fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 47516fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 47526fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 47536fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 47546fa01ccdSSowmini Varadhan skb_release_data(skb); 47556fa01ccdSSowmini Varadhan } else { 47566fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 47576fa01ccdSSowmini Varadhan * relocate values 47586fa01ccdSSowmini Varadhan */ 47596fa01ccdSSowmini Varadhan skb_free_head(skb); 47606fa01ccdSSowmini Varadhan } 47616fa01ccdSSowmini Varadhan 47626fa01ccdSSowmini Varadhan skb->head = data; 47636fa01ccdSSowmini Varadhan skb->data = data; 47646fa01ccdSSowmini Varadhan skb->head_frag = 0; 47656fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 47666fa01ccdSSowmini Varadhan skb->end = size; 47676fa01ccdSSowmini Varadhan #else 47686fa01ccdSSowmini Varadhan skb->end = skb->head + size; 47696fa01ccdSSowmini Varadhan #endif 47706fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 47716fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 47726fa01ccdSSowmini Varadhan skb->cloned = 0; 47736fa01ccdSSowmini Varadhan skb->hdr_len = 0; 47746fa01ccdSSowmini Varadhan skb->nohdr = 0; 47756fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 47766fa01ccdSSowmini Varadhan 47776fa01ccdSSowmini Varadhan return 0; 47786fa01ccdSSowmini Varadhan } 47796fa01ccdSSowmini Varadhan 47806fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 47816fa01ccdSSowmini Varadhan 47826fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 47836fa01ccdSSowmini Varadhan * pskb_carve() 47846fa01ccdSSowmini Varadhan */ 47856fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 47866fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 47876fa01ccdSSowmini Varadhan gfp_t gfp_mask) 47886fa01ccdSSowmini Varadhan { 47896fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 47906fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 47916fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 47926fa01ccdSSowmini Varadhan 47936fa01ccdSSowmini Varadhan do { 47946fa01ccdSSowmini Varadhan if (!list) { 47956fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 47966fa01ccdSSowmini Varadhan return -EFAULT; 47976fa01ccdSSowmini Varadhan } 47986fa01ccdSSowmini Varadhan if (list->len <= eat) { 47996fa01ccdSSowmini Varadhan /* Eaten as whole. */ 48006fa01ccdSSowmini Varadhan eat -= list->len; 48016fa01ccdSSowmini Varadhan list = list->next; 48026fa01ccdSSowmini Varadhan insp = list; 48036fa01ccdSSowmini Varadhan } else { 48046fa01ccdSSowmini Varadhan /* Eaten partially. */ 48056fa01ccdSSowmini Varadhan if (skb_shared(list)) { 48066fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 48076fa01ccdSSowmini Varadhan if (!clone) 48086fa01ccdSSowmini Varadhan return -ENOMEM; 48096fa01ccdSSowmini Varadhan insp = list->next; 48106fa01ccdSSowmini Varadhan list = clone; 48116fa01ccdSSowmini Varadhan } else { 48126fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 48136fa01ccdSSowmini Varadhan insp = list; 48146fa01ccdSSowmini Varadhan } 48156fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 48166fa01ccdSSowmini Varadhan kfree_skb(clone); 48176fa01ccdSSowmini Varadhan return -ENOMEM; 48186fa01ccdSSowmini Varadhan } 48196fa01ccdSSowmini Varadhan break; 48206fa01ccdSSowmini Varadhan } 48216fa01ccdSSowmini Varadhan } while (eat); 48226fa01ccdSSowmini Varadhan 48236fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 48246fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 48256fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 48266fa01ccdSSowmini Varadhan kfree_skb(list); 48276fa01ccdSSowmini Varadhan } 48286fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 48296fa01ccdSSowmini Varadhan if (clone) { 48306fa01ccdSSowmini Varadhan clone->next = list; 48316fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 48326fa01ccdSSowmini Varadhan } 48336fa01ccdSSowmini Varadhan return 0; 48346fa01ccdSSowmini Varadhan } 48356fa01ccdSSowmini Varadhan 48366fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 48376fa01ccdSSowmini Varadhan * non-linear part of skb 48386fa01ccdSSowmini Varadhan */ 48396fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 48406fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 48416fa01ccdSSowmini Varadhan { 48426fa01ccdSSowmini Varadhan int i, k = 0; 48436fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 48446fa01ccdSSowmini Varadhan u8 *data; 48456fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 48466fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 48476fa01ccdSSowmini Varadhan 48486fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 48496fa01ccdSSowmini Varadhan 48506fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 48516fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 48526fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 48536fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 48546fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 48556fa01ccdSSowmini Varadhan if (!data) 48566fa01ccdSSowmini Varadhan return -ENOMEM; 48576fa01ccdSSowmini Varadhan 48586fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 48596fa01ccdSSowmini Varadhan 48606fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 48616fa01ccdSSowmini Varadhan skb_shinfo(skb), offsetof(struct skb_shared_info, 48626fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 48636fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 48646fa01ccdSSowmini Varadhan kfree(data); 48656fa01ccdSSowmini Varadhan return -ENOMEM; 48666fa01ccdSSowmini Varadhan } 48676fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 48686fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 48696fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 48706fa01ccdSSowmini Varadhan 48716fa01ccdSSowmini Varadhan if (pos + fsize > off) { 48726fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 48736fa01ccdSSowmini Varadhan 48746fa01ccdSSowmini Varadhan if (pos < off) { 48756fa01ccdSSowmini Varadhan /* Split frag. 48766fa01ccdSSowmini Varadhan * We have two variants in this case: 48776fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 48786fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 48796fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 48806fa01ccdSSowmini Varadhan * where splitting is expensive. 48816fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 48826fa01ccdSSowmini Varadhan */ 48836fa01ccdSSowmini Varadhan shinfo->frags[0].page_offset += off - pos; 48846fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 48856fa01ccdSSowmini Varadhan } 48866fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 48876fa01ccdSSowmini Varadhan k++; 48886fa01ccdSSowmini Varadhan } 48896fa01ccdSSowmini Varadhan pos += fsize; 48906fa01ccdSSowmini Varadhan } 48916fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 48926fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 48936fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 48946fa01ccdSSowmini Varadhan 48956fa01ccdSSowmini Varadhan if (k == 0) { 48966fa01ccdSSowmini Varadhan /* split line is in frag list */ 48976fa01ccdSSowmini Varadhan pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask); 48986fa01ccdSSowmini Varadhan } 48996fa01ccdSSowmini Varadhan skb_release_data(skb); 49006fa01ccdSSowmini Varadhan 49016fa01ccdSSowmini Varadhan skb->head = data; 49026fa01ccdSSowmini Varadhan skb->head_frag = 0; 49036fa01ccdSSowmini Varadhan skb->data = data; 49046fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 49056fa01ccdSSowmini Varadhan skb->end = size; 49066fa01ccdSSowmini Varadhan #else 49076fa01ccdSSowmini Varadhan skb->end = skb->head + size; 49086fa01ccdSSowmini Varadhan #endif 49096fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 49106fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 49116fa01ccdSSowmini Varadhan skb->cloned = 0; 49126fa01ccdSSowmini Varadhan skb->hdr_len = 0; 49136fa01ccdSSowmini Varadhan skb->nohdr = 0; 49146fa01ccdSSowmini Varadhan skb->len -= off; 49156fa01ccdSSowmini Varadhan skb->data_len = skb->len; 49166fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 49176fa01ccdSSowmini Varadhan return 0; 49186fa01ccdSSowmini Varadhan } 49196fa01ccdSSowmini Varadhan 49206fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 49216fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 49226fa01ccdSSowmini Varadhan { 49236fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 49246fa01ccdSSowmini Varadhan 49256fa01ccdSSowmini Varadhan if (len < headlen) 49266fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 49276fa01ccdSSowmini Varadhan else 49286fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 49296fa01ccdSSowmini Varadhan } 49306fa01ccdSSowmini Varadhan 49316fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 49326fa01ccdSSowmini Varadhan * a new skb 49336fa01ccdSSowmini Varadhan */ 49346fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 49356fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 49366fa01ccdSSowmini Varadhan { 49376fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 49386fa01ccdSSowmini Varadhan 49396fa01ccdSSowmini Varadhan if (!clone) 49406fa01ccdSSowmini Varadhan return NULL; 49416fa01ccdSSowmini Varadhan 49426fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 49436fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 49446fa01ccdSSowmini Varadhan kfree_skb(clone); 49456fa01ccdSSowmini Varadhan return NULL; 49466fa01ccdSSowmini Varadhan } 49476fa01ccdSSowmini Varadhan return clone; 49486fa01ccdSSowmini Varadhan } 49496fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 4950c8c8b127SEric Dumazet 4951c8c8b127SEric Dumazet /** 4952c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 4953c8c8b127SEric Dumazet * @skb: buffer 4954c8c8b127SEric Dumazet * 4955c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 4956c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 4957c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 4958c8c8b127SEric Dumazet * truesize. 4959c8c8b127SEric Dumazet * Notes: 4960c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 4961c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 4962c8c8b127SEric Dumazet */ 4963c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 4964c8c8b127SEric Dumazet { 49653174fed9SEric Dumazet if (skb->data_len) { 49663174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 4967c8c8b127SEric Dumazet skb_cloned(skb)) 4968c8c8b127SEric Dumazet return; 4969c8c8b127SEric Dumazet 4970c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 4971c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 49723174fed9SEric Dumazet } 49733174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 49743174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 49753174fed9SEric Dumazet * their truesize. 49763174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 49773174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 49783174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 4979c8c8b127SEric Dumazet */ 4980c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 4981c8c8b127SEric Dumazet } 4982