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> 441da177e4SLinus Torvalds #include <linux/mm.h> 451da177e4SLinus Torvalds #include <linux/interrupt.h> 461da177e4SLinus Torvalds #include <linux/in.h> 471da177e4SLinus Torvalds #include <linux/inet.h> 481da177e4SLinus Torvalds #include <linux/slab.h> 49de960aa9SFlorian Westphal #include <linux/tcp.h> 50de960aa9SFlorian Westphal #include <linux/udp.h> 5190017accSMarcelo Ricardo Leitner #include <linux/sctp.h> 521da177e4SLinus Torvalds #include <linux/netdevice.h> 531da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 541da177e4SLinus Torvalds #include <net/pkt_sched.h> 551da177e4SLinus Torvalds #endif 561da177e4SLinus Torvalds #include <linux/string.h> 571da177e4SLinus Torvalds #include <linux/skbuff.h> 589c55e01cSJens Axboe #include <linux/splice.h> 591da177e4SLinus Torvalds #include <linux/cache.h> 601da177e4SLinus Torvalds #include <linux/rtnetlink.h> 611da177e4SLinus Torvalds #include <linux/init.h> 62716ea3a7SDavid Howells #include <linux/scatterlist.h> 63ac45f602SPatrick Ohly #include <linux/errqueue.h> 64268bb0ceSLinus Torvalds #include <linux/prefetch.h> 650d5501c1SVlad Yasevich #include <linux/if_vlan.h> 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds #include <net/protocol.h> 681da177e4SLinus Torvalds #include <net/dst.h> 691da177e4SLinus Torvalds #include <net/sock.h> 701da177e4SLinus Torvalds #include <net/checksum.h> 71ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 721da177e4SLinus Torvalds #include <net/xfrm.h> 731da177e4SLinus Torvalds 747c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 75ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7651c56b00SEric Dumazet #include <linux/highmem.h> 77b245be1fSWillem de Bruijn #include <linux/capability.h> 78b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 79a1f8e7f7SAl Viro 807b7ed885SBart Van Assche #include "datagram.h" 817b7ed885SBart Van Assche 8208009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8308009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 84df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 85df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 86df5042f4SFlorian Westphal #endif 875f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 885f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds /** 91f05de73bSJean Sacren * skb_panic - private function for out-of-line support 921da177e4SLinus Torvalds * @skb: buffer 931da177e4SLinus Torvalds * @sz: size 94f05de73bSJean Sacren * @addr: address 9599d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 961da177e4SLinus Torvalds * 97f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 98f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 99f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 100f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1011da177e4SLinus Torvalds */ 102f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 10399d5851eSJames Hogan const char msg[]) 1041da177e4SLinus Torvalds { 105e005d193SJoe Perches pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 10699d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1074305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 10826095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1091da177e4SLinus Torvalds BUG(); 1101da177e4SLinus Torvalds } 1111da177e4SLinus Torvalds 112f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1131da177e4SLinus Torvalds { 114f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds 117f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 118f05de73bSJean Sacren { 119f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 120f05de73bSJean Sacren } 121c93bdd0eSMel Gorman 122c93bdd0eSMel Gorman /* 123c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 124c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 125c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 126c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 127c93bdd0eSMel Gorman * memory is free 128c93bdd0eSMel Gorman */ 129c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 130c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 13161c5e88aSstephen hemminger 13261c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 13361c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 134c93bdd0eSMel Gorman { 135c93bdd0eSMel Gorman void *obj; 136c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 137c93bdd0eSMel Gorman 138c93bdd0eSMel Gorman /* 139c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 140c93bdd0eSMel Gorman * to the reserves, fail. 141c93bdd0eSMel Gorman */ 142c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 143c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 144c93bdd0eSMel Gorman node); 145c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 146c93bdd0eSMel Gorman goto out; 147c93bdd0eSMel Gorman 148c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 149c93bdd0eSMel Gorman ret_pfmemalloc = true; 150c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 151c93bdd0eSMel Gorman 152c93bdd0eSMel Gorman out: 153c93bdd0eSMel Gorman if (pfmemalloc) 154c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 155c93bdd0eSMel Gorman 156c93bdd0eSMel Gorman return obj; 157c93bdd0eSMel Gorman } 158c93bdd0eSMel Gorman 1591da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1601da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1611da177e4SLinus Torvalds * [BEEP] leaks. 1621da177e4SLinus Torvalds * 1631da177e4SLinus Torvalds */ 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds /** 166d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1671da177e4SLinus Torvalds * @size: size to allocate 1681da177e4SLinus Torvalds * @gfp_mask: allocation mask 169c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 170c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 171c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 172c93bdd0eSMel Gorman * allocations in case the data is required for writeback 173b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1741da177e4SLinus Torvalds * 1751da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 17694b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 17794b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 1781da177e4SLinus Torvalds * 1791da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1801da177e4SLinus Torvalds * %GFP_ATOMIC. 1811da177e4SLinus Torvalds */ 182dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 183c93bdd0eSMel Gorman int flags, int node) 1841da177e4SLinus Torvalds { 185e18b890bSChristoph Lameter struct kmem_cache *cache; 1864947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1871da177e4SLinus Torvalds struct sk_buff *skb; 1881da177e4SLinus Torvalds u8 *data; 189c93bdd0eSMel Gorman bool pfmemalloc; 1901da177e4SLinus Torvalds 191c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 192c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 193c93bdd0eSMel Gorman 194c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 195c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1968798b3fbSHerbert Xu 1971da177e4SLinus Torvalds /* Get the HEAD */ 198b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1991da177e4SLinus Torvalds if (!skb) 2001da177e4SLinus Torvalds goto out; 201ec7d2f2cSEric Dumazet prefetchw(skb); 2021da177e4SLinus Torvalds 20387fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 20487fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 20587fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 20687fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 20787fb4b7bSEric Dumazet */ 208bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 20987fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 210c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2111da177e4SLinus Torvalds if (!data) 2121da177e4SLinus Torvalds goto nodata; 21387fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 21487fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 21587fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 21687fb4b7bSEric Dumazet */ 21787fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 218ec7d2f2cSEric Dumazet prefetchw(data + size); 2191da177e4SLinus Torvalds 220ca0605a7SArnaldo Carvalho de Melo /* 221c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 222c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 223c8005785SJohannes Berg * the tail pointer in struct sk_buff! 224ca0605a7SArnaldo Carvalho de Melo */ 225ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 22687fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 22787fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 228c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 22963354797SReshetova, Elena refcount_set(&skb->users, 1); 2301da177e4SLinus Torvalds skb->head = data; 2311da177e4SLinus Torvalds skb->data = data; 23227a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2334305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 23435d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 23535d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 23619633e12SStephen Hemminger 2374947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2384947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 239ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2404947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 2414947d3efSBenjamin LaHaise 242c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 243d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 2441da177e4SLinus Torvalds 245d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 246d0bf4a9eSEric Dumazet 247d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 2482638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 1); 249d179cd12SDavid S. Miller 2506ffe75ebSEric Dumazet fclones->skb2.fclone = SKB_FCLONE_CLONE; 251d179cd12SDavid S. Miller } 2521da177e4SLinus Torvalds out: 2531da177e4SLinus Torvalds return skb; 2541da177e4SLinus Torvalds nodata: 2558798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2561da177e4SLinus Torvalds skb = NULL; 2571da177e4SLinus Torvalds goto out; 2581da177e4SLinus Torvalds } 259b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2601da177e4SLinus Torvalds 261ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 262ba0509b6SJesper Dangaard Brouer static struct sk_buff *__build_skb_around(struct sk_buff *skb, 263ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 264ba0509b6SJesper Dangaard Brouer { 265ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 266ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 267ba0509b6SJesper Dangaard Brouer 268ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 269ba0509b6SJesper Dangaard Brouer 270ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 271ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 272ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 273ba0509b6SJesper Dangaard Brouer skb->head = data; 274ba0509b6SJesper Dangaard Brouer skb->data = data; 275ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 276ba0509b6SJesper Dangaard Brouer skb->end = skb->tail + size; 277ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 278ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 279ba0509b6SJesper Dangaard Brouer 280ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 281ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 282ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 283ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 284ba0509b6SJesper Dangaard Brouer 285ba0509b6SJesper Dangaard Brouer return skb; 286ba0509b6SJesper Dangaard Brouer } 287ba0509b6SJesper Dangaard Brouer 2881da177e4SLinus Torvalds /** 2892ea2f62cSEric Dumazet * __build_skb - build a network buffer 290b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2912ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 292b2b5ce9dSEric Dumazet * 293b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 294deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2952ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2962ea2f62cSEric Dumazet * or vmalloc() 297b2b5ce9dSEric Dumazet * The return is the new skb buffer. 298b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 299b2b5ce9dSEric Dumazet * Notes : 300b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 301b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 302b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 303b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 304b2b5ce9dSEric Dumazet * before giving packet to stack. 305b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 306b2b5ce9dSEric Dumazet */ 3072ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 308b2b5ce9dSEric Dumazet { 309b2b5ce9dSEric Dumazet struct sk_buff *skb; 310b2b5ce9dSEric Dumazet 311b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 312ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 313b2b5ce9dSEric Dumazet return NULL; 314b2b5ce9dSEric Dumazet 315b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 316b2b5ce9dSEric Dumazet 317ba0509b6SJesper Dangaard Brouer return __build_skb_around(skb, data, frag_size); 318b2b5ce9dSEric Dumazet } 3192ea2f62cSEric Dumazet 3202ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3212ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3222ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 3232ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 3242ea2f62cSEric Dumazet */ 3252ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3262ea2f62cSEric Dumazet { 3272ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3282ea2f62cSEric Dumazet 3292ea2f62cSEric Dumazet if (skb && frag_size) { 3302ea2f62cSEric Dumazet skb->head_frag = 1; 3312f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 3322ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3332ea2f62cSEric Dumazet } 3342ea2f62cSEric Dumazet return skb; 3352ea2f62cSEric Dumazet } 336b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 337b2b5ce9dSEric Dumazet 338ba0509b6SJesper Dangaard Brouer /** 339ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 340ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 341ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 342ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 343ba0509b6SJesper Dangaard Brouer */ 344ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 345ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 346ba0509b6SJesper Dangaard Brouer { 347ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 348ba0509b6SJesper Dangaard Brouer return NULL; 349ba0509b6SJesper Dangaard Brouer 350ba0509b6SJesper Dangaard Brouer skb = __build_skb_around(skb, data, frag_size); 351ba0509b6SJesper Dangaard Brouer 352ba0509b6SJesper Dangaard Brouer if (skb && frag_size) { 353ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 354ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 355ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 356ba0509b6SJesper Dangaard Brouer } 357ba0509b6SJesper Dangaard Brouer return skb; 358ba0509b6SJesper Dangaard Brouer } 359ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 360ba0509b6SJesper Dangaard Brouer 361795bb1c0SJesper Dangaard Brouer #define NAPI_SKB_CACHE_SIZE 64 362795bb1c0SJesper Dangaard Brouer 363795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache { 364795bb1c0SJesper Dangaard Brouer struct page_frag_cache page; 365e0d7924aSAlexey Dobriyan unsigned int skb_count; 366795bb1c0SJesper Dangaard Brouer void *skb_cache[NAPI_SKB_CACHE_SIZE]; 367795bb1c0SJesper Dangaard Brouer }; 368795bb1c0SJesper Dangaard Brouer 369b63ae8caSAlexander Duyck static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 370795bb1c0SJesper Dangaard Brouer static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 371ffde7328SAlexander Duyck 372ffde7328SAlexander Duyck static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 373ffde7328SAlexander Duyck { 374b63ae8caSAlexander Duyck struct page_frag_cache *nc; 375ffde7328SAlexander Duyck unsigned long flags; 376ffde7328SAlexander Duyck void *data; 377ffde7328SAlexander Duyck 378ffde7328SAlexander Duyck local_irq_save(flags); 3799451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 3808c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, fragsz, gfp_mask); 3816f532612SEric Dumazet local_irq_restore(flags); 3826f532612SEric Dumazet return data; 3836f532612SEric Dumazet } 384c93bdd0eSMel Gorman 385c93bdd0eSMel Gorman /** 386c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 387c93bdd0eSMel Gorman * @fragsz: fragment size 388c93bdd0eSMel Gorman * 389c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 390c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 391c93bdd0eSMel Gorman */ 392c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 393c93bdd0eSMel Gorman { 3943bed3cc4SAlexander Duyck fragsz = SKB_DATA_ALIGN(fragsz); 3953bed3cc4SAlexander Duyck 396453f85d4SMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC); 397c93bdd0eSMel Gorman } 3986f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 3996f532612SEric Dumazet 400ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 401ffde7328SAlexander Duyck { 402795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 4039451980aSAlexander Duyck 4048c2dd3e4SAlexander Duyck return page_frag_alloc(&nc->page, fragsz, gfp_mask); 405ffde7328SAlexander Duyck } 406ffde7328SAlexander Duyck 407ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 408ffde7328SAlexander Duyck { 4093bed3cc4SAlexander Duyck fragsz = SKB_DATA_ALIGN(fragsz); 4103bed3cc4SAlexander Duyck 411453f85d4SMel Gorman return __napi_alloc_frag(fragsz, GFP_ATOMIC); 412ffde7328SAlexander Duyck } 413ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 414ffde7328SAlexander Duyck 4156f532612SEric Dumazet /** 416fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 417fd11a83dSAlexander Duyck * @dev: network device to receive on 418d7499160SMasanari Iida * @len: length to allocate 419fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 420fd11a83dSAlexander Duyck * 421fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 422fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 423fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 424fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 425fd11a83dSAlexander Duyck * 426fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 427fd11a83dSAlexander Duyck */ 4289451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4299451980aSAlexander Duyck gfp_t gfp_mask) 430fd11a83dSAlexander Duyck { 431b63ae8caSAlexander Duyck struct page_frag_cache *nc; 4329451980aSAlexander Duyck unsigned long flags; 433fd11a83dSAlexander Duyck struct sk_buff *skb; 4349451980aSAlexander Duyck bool pfmemalloc; 4359451980aSAlexander Duyck void *data; 436fd11a83dSAlexander Duyck 4379451980aSAlexander Duyck len += NET_SKB_PAD; 438fd11a83dSAlexander Duyck 4399451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 440d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 441a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 442a080e7bdSAlexander Duyck if (!skb) 443a080e7bdSAlexander Duyck goto skb_fail; 444a080e7bdSAlexander Duyck goto skb_success; 445a080e7bdSAlexander Duyck } 4469451980aSAlexander Duyck 4479451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4489451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 4499451980aSAlexander Duyck 4509451980aSAlexander Duyck if (sk_memalloc_socks()) 4519451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 4529451980aSAlexander Duyck 4539451980aSAlexander Duyck local_irq_save(flags); 4549451980aSAlexander Duyck 4559451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 4568c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 4579451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 4589451980aSAlexander Duyck 4599451980aSAlexander Duyck local_irq_restore(flags); 4609451980aSAlexander Duyck 4619451980aSAlexander Duyck if (unlikely(!data)) 4629451980aSAlexander Duyck return NULL; 4639451980aSAlexander Duyck 4649451980aSAlexander Duyck skb = __build_skb(data, len); 4659451980aSAlexander Duyck if (unlikely(!skb)) { 466181edb2bSAlexander Duyck skb_free_frag(data); 4679451980aSAlexander Duyck return NULL; 4689451980aSAlexander Duyck } 4699451980aSAlexander Duyck 4709451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 4719451980aSAlexander Duyck if (pfmemalloc) 4729451980aSAlexander Duyck skb->pfmemalloc = 1; 4739451980aSAlexander Duyck skb->head_frag = 1; 4749451980aSAlexander Duyck 475a080e7bdSAlexander Duyck skb_success: 4768af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4777b2e497aSChristoph Hellwig skb->dev = dev; 478fd11a83dSAlexander Duyck 479a080e7bdSAlexander Duyck skb_fail: 4808af27456SChristoph Hellwig return skb; 4818af27456SChristoph Hellwig } 482b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4831da177e4SLinus Torvalds 484fd11a83dSAlexander Duyck /** 485fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 486fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 487d7499160SMasanari Iida * @len: length to allocate 488fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 489fd11a83dSAlexander Duyck * 490fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 491fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 492fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 493fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 494fd11a83dSAlexander Duyck * 495fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 496fd11a83dSAlexander Duyck */ 4979451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 4989451980aSAlexander Duyck gfp_t gfp_mask) 499fd11a83dSAlexander Duyck { 500795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 501fd11a83dSAlexander Duyck struct sk_buff *skb; 5029451980aSAlexander Duyck void *data; 503fd11a83dSAlexander Duyck 5049451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 505fd11a83dSAlexander Duyck 5069451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 507d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 508a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 509a080e7bdSAlexander Duyck if (!skb) 510a080e7bdSAlexander Duyck goto skb_fail; 511a080e7bdSAlexander Duyck goto skb_success; 512a080e7bdSAlexander Duyck } 5139451980aSAlexander Duyck 5149451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5159451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5169451980aSAlexander Duyck 5179451980aSAlexander Duyck if (sk_memalloc_socks()) 5189451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5199451980aSAlexander Duyck 5208c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5219451980aSAlexander Duyck if (unlikely(!data)) 5229451980aSAlexander Duyck return NULL; 5239451980aSAlexander Duyck 5249451980aSAlexander Duyck skb = __build_skb(data, len); 5259451980aSAlexander Duyck if (unlikely(!skb)) { 526181edb2bSAlexander Duyck skb_free_frag(data); 5279451980aSAlexander Duyck return NULL; 5289451980aSAlexander Duyck } 5299451980aSAlexander Duyck 5309451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 531795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5329451980aSAlexander Duyck skb->pfmemalloc = 1; 5339451980aSAlexander Duyck skb->head_frag = 1; 5349451980aSAlexander Duyck 535a080e7bdSAlexander Duyck skb_success: 536fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 537fd11a83dSAlexander Duyck skb->dev = napi->dev; 538fd11a83dSAlexander Duyck 539a080e7bdSAlexander Duyck skb_fail: 540fd11a83dSAlexander Duyck return skb; 541fd11a83dSAlexander Duyck } 542fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 543fd11a83dSAlexander Duyck 544654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 54550269e19SEric Dumazet int size, unsigned int truesize) 546654bed16SPeter Zijlstra { 547654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 548654bed16SPeter Zijlstra skb->len += size; 549654bed16SPeter Zijlstra skb->data_len += size; 55050269e19SEric Dumazet skb->truesize += truesize; 551654bed16SPeter Zijlstra } 552654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 553654bed16SPeter Zijlstra 554f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 555f8e617e1SJason Wang unsigned int truesize) 556f8e617e1SJason Wang { 557f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 558f8e617e1SJason Wang 559f8e617e1SJason Wang skb_frag_size_add(frag, size); 560f8e617e1SJason Wang skb->len += size; 561f8e617e1SJason Wang skb->data_len += size; 562f8e617e1SJason Wang skb->truesize += truesize; 563f8e617e1SJason Wang } 564f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 565f8e617e1SJason Wang 56627b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5671da177e4SLinus Torvalds { 568bd8a7036SEric Dumazet kfree_skb_list(*listp); 56927b437c8SHerbert Xu *listp = NULL; 5701da177e4SLinus Torvalds } 5711da177e4SLinus Torvalds 57227b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 57327b437c8SHerbert Xu { 57427b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 57527b437c8SHerbert Xu } 57627b437c8SHerbert Xu 5771da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5781da177e4SLinus Torvalds { 5791da177e4SLinus Torvalds struct sk_buff *list; 5801da177e4SLinus Torvalds 581fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5821da177e4SLinus Torvalds skb_get(list); 5831da177e4SLinus Torvalds } 5841da177e4SLinus Torvalds 585d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 586d3836f21SEric Dumazet { 587181edb2bSAlexander Duyck unsigned char *head = skb->head; 588181edb2bSAlexander Duyck 589d3836f21SEric Dumazet if (skb->head_frag) 590181edb2bSAlexander Duyck skb_free_frag(head); 591d3836f21SEric Dumazet else 592181edb2bSAlexander Duyck kfree(head); 593d3836f21SEric Dumazet } 594d3836f21SEric Dumazet 5955bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5961da177e4SLinus Torvalds { 597ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 5981da177e4SLinus Torvalds int i; 599ff04a771SEric Dumazet 600ff04a771SEric Dumazet if (skb->cloned && 601ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 602ff04a771SEric Dumazet &shinfo->dataref)) 603ff04a771SEric Dumazet return; 604ff04a771SEric Dumazet 605ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 606ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 6071da177e4SLinus Torvalds 608ff04a771SEric Dumazet if (shinfo->frag_list) 609ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6101da177e4SLinus Torvalds 6111f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 612d3836f21SEric Dumazet skb_free_head(skb); 6131da177e4SLinus Torvalds } 6141da177e4SLinus Torvalds 6151da177e4SLinus Torvalds /* 6161da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6171da177e4SLinus Torvalds */ 6182d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6191da177e4SLinus Torvalds { 620d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 621d179cd12SDavid S. Miller 622d179cd12SDavid S. Miller switch (skb->fclone) { 623d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6241da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6256ffe75ebSEric Dumazet return; 626d179cd12SDavid S. Miller 627d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 628d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6296ffe75ebSEric Dumazet 6306ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6316ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6326ffe75ebSEric Dumazet * while here, branch prediction will be good. 6336ffe75ebSEric Dumazet */ 6342638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 6356ffe75ebSEric Dumazet goto fastpath; 636d179cd12SDavid S. Miller break; 637d179cd12SDavid S. Miller 6386ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 639d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 640d179cd12SDavid S. Miller break; 6413ff50b79SStephen Hemminger } 6422638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 6436ffe75ebSEric Dumazet return; 6446ffe75ebSEric Dumazet fastpath: 6456ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6461da177e4SLinus Torvalds } 6471da177e4SLinus Torvalds 6480a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 6491da177e4SLinus Torvalds { 650adf30907SEric Dumazet skb_dst_drop(skb); 6511da177e4SLinus Torvalds if (skb->destructor) { 6529c2b3328SStephen Hemminger WARN_ON(in_irq()); 6531da177e4SLinus Torvalds skb->destructor(skb); 6541da177e4SLinus Torvalds } 655a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 656cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 6572fc72c7bSKOVACS Krisztian #endif 658df5042f4SFlorian Westphal skb_ext_put(skb); 65904a4bb55SLennert Buytenhek } 66004a4bb55SLennert Buytenhek 66104a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 66204a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 66304a4bb55SLennert Buytenhek { 66404a4bb55SLennert Buytenhek skb_release_head_state(skb); 665a28b1b90SFlorian Westphal if (likely(skb->head)) 6662d4baff8SHerbert Xu skb_release_data(skb); 6672d4baff8SHerbert Xu } 6681da177e4SLinus Torvalds 6692d4baff8SHerbert Xu /** 6702d4baff8SHerbert Xu * __kfree_skb - private function 6712d4baff8SHerbert Xu * @skb: buffer 6722d4baff8SHerbert Xu * 6732d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6742d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6752d4baff8SHerbert Xu * always call kfree_skb 6762d4baff8SHerbert Xu */ 6772d4baff8SHerbert Xu 6782d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6792d4baff8SHerbert Xu { 6802d4baff8SHerbert Xu skb_release_all(skb); 6811da177e4SLinus Torvalds kfree_skbmem(skb); 6821da177e4SLinus Torvalds } 683b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds /** 686231d06aeSJörn Engel * kfree_skb - free an sk_buff 687231d06aeSJörn Engel * @skb: buffer to free 688231d06aeSJörn Engel * 689231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 690231d06aeSJörn Engel * hit zero. 691231d06aeSJörn Engel */ 692231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 693231d06aeSJörn Engel { 6943889a803SPaolo Abeni if (!skb_unref(skb)) 695231d06aeSJörn Engel return; 6963889a803SPaolo Abeni 697ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 698231d06aeSJörn Engel __kfree_skb(skb); 699231d06aeSJörn Engel } 700b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 701231d06aeSJörn Engel 702bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 703bd8a7036SEric Dumazet { 704bd8a7036SEric Dumazet while (segs) { 705bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 706bd8a7036SEric Dumazet 707bd8a7036SEric Dumazet kfree_skb(segs); 708bd8a7036SEric Dumazet segs = next; 709bd8a7036SEric Dumazet } 710bd8a7036SEric Dumazet } 711bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 712bd8a7036SEric Dumazet 713d1a203eaSStephen Hemminger /** 71425121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 71525121173SMichael S. Tsirkin * @skb: buffer that triggered an error 71625121173SMichael S. Tsirkin * 71725121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 71825121173SMichael S. Tsirkin * skb must be freed afterwards. 71925121173SMichael S. Tsirkin */ 72025121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 72125121173SMichael S. Tsirkin { 7221f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 72325121173SMichael S. Tsirkin } 72425121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 72525121173SMichael S. Tsirkin 72625121173SMichael S. Tsirkin /** 727ead2ceb0SNeil Horman * consume_skb - free an skbuff 728ead2ceb0SNeil Horman * @skb: buffer to free 729ead2ceb0SNeil Horman * 730ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 731ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 732ead2ceb0SNeil Horman * is being dropped after a failure and notes that 733ead2ceb0SNeil Horman */ 734ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 735ead2ceb0SNeil Horman { 7363889a803SPaolo Abeni if (!skb_unref(skb)) 737ead2ceb0SNeil Horman return; 7383889a803SPaolo Abeni 73907dc22e7SKoki Sanagi trace_consume_skb(skb); 740ead2ceb0SNeil Horman __kfree_skb(skb); 741ead2ceb0SNeil Horman } 742ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 743ead2ceb0SNeil Horman 7440a463c78SPaolo Abeni /** 7450a463c78SPaolo Abeni * consume_stateless_skb - free an skbuff, assuming it is stateless 7460a463c78SPaolo Abeni * @skb: buffer to free 7470a463c78SPaolo Abeni * 748ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 749ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 7500a463c78SPaolo Abeni */ 751ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 7520a463c78SPaolo Abeni { 7530a463c78SPaolo Abeni trace_consume_skb(skb); 7540a463c78SPaolo Abeni skb_release_data(skb); 7550a463c78SPaolo Abeni kfree_skbmem(skb); 7560a463c78SPaolo Abeni } 7570a463c78SPaolo Abeni 758795bb1c0SJesper Dangaard Brouer void __kfree_skb_flush(void) 759795bb1c0SJesper Dangaard Brouer { 760795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 761795bb1c0SJesper Dangaard Brouer 762795bb1c0SJesper Dangaard Brouer /* flush skb_cache if containing objects */ 763795bb1c0SJesper Dangaard Brouer if (nc->skb_count) { 764795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 765795bb1c0SJesper Dangaard Brouer nc->skb_cache); 766795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 767795bb1c0SJesper Dangaard Brouer } 768795bb1c0SJesper Dangaard Brouer } 769795bb1c0SJesper Dangaard Brouer 77015fad714SJesper Dangaard Brouer static inline void _kfree_skb_defer(struct sk_buff *skb) 771795bb1c0SJesper Dangaard Brouer { 772795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 773795bb1c0SJesper Dangaard Brouer 774795bb1c0SJesper Dangaard Brouer /* drop skb->head and call any destructors for packet */ 775795bb1c0SJesper Dangaard Brouer skb_release_all(skb); 776795bb1c0SJesper Dangaard Brouer 777795bb1c0SJesper Dangaard Brouer /* record skb to CPU local list */ 778795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 779795bb1c0SJesper Dangaard Brouer 780795bb1c0SJesper Dangaard Brouer #ifdef CONFIG_SLUB 781795bb1c0SJesper Dangaard Brouer /* SLUB writes into objects when freeing */ 782795bb1c0SJesper Dangaard Brouer prefetchw(skb); 783795bb1c0SJesper Dangaard Brouer #endif 784795bb1c0SJesper Dangaard Brouer 785795bb1c0SJesper Dangaard Brouer /* flush skb_cache if it is filled */ 786795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 787795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 788795bb1c0SJesper Dangaard Brouer nc->skb_cache); 789795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 790795bb1c0SJesper Dangaard Brouer } 791795bb1c0SJesper Dangaard Brouer } 79215fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 79315fad714SJesper Dangaard Brouer { 79415fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 79515fad714SJesper Dangaard Brouer } 796795bb1c0SJesper Dangaard Brouer 797795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 798795bb1c0SJesper Dangaard Brouer { 799795bb1c0SJesper Dangaard Brouer if (unlikely(!skb)) 800795bb1c0SJesper Dangaard Brouer return; 801795bb1c0SJesper Dangaard Brouer 802885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 803795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 804885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 805795bb1c0SJesper Dangaard Brouer return; 806795bb1c0SJesper Dangaard Brouer } 807795bb1c0SJesper Dangaard Brouer 8087608894eSPaolo Abeni if (!skb_unref(skb)) 809795bb1c0SJesper Dangaard Brouer return; 8107608894eSPaolo Abeni 811795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 812795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 813795bb1c0SJesper Dangaard Brouer 814795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 815abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 816795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 817795bb1c0SJesper Dangaard Brouer return; 818795bb1c0SJesper Dangaard Brouer } 819795bb1c0SJesper Dangaard Brouer 82015fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 821795bb1c0SJesper Dangaard Brouer } 822795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 823795bb1c0SJesper Dangaard Brouer 824b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 825b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 826b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 827b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 828b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 829b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 830b1937227SEric Dumazet 831dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 832dec18810SHerbert Xu { 833dec18810SHerbert Xu new->tstamp = old->tstamp; 834b1937227SEric Dumazet /* We do not copy old->sk */ 835dec18810SHerbert Xu new->dev = old->dev; 836b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 8377fee226aSEric Dumazet skb_dst_copy(new, old); 838df5042f4SFlorian Westphal __skb_ext_copy(new, old); 839b1937227SEric Dumazet __nf_copy(new, old, false); 8406aa895b0SPatrick McHardy 841b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 842b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 843b1937227SEric Dumazet */ 844b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 84506021292SEliezer Tamir 846b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 847b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 848b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 849b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 850b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 851b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 852b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 853b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 854b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 855b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 856b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 857b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 858b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 859b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 860b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 861b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 862b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 863b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 864b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 865b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 866b1937227SEric Dumazet #endif 867e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 868b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 86906021292SEliezer Tamir #endif 8702bd82484SEric Dumazet #ifdef CONFIG_XPS 8712bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 8722bd82484SEric Dumazet #endif 873b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 874b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 875b1937227SEric Dumazet #endif 876b1937227SEric Dumazet 877dec18810SHerbert Xu } 878dec18810SHerbert Xu 87982c49a35SHerbert Xu /* 88082c49a35SHerbert Xu * You should not add any new code to this function. Add it to 88182c49a35SHerbert Xu * __copy_skb_header above instead. 88282c49a35SHerbert Xu */ 883e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 8841da177e4SLinus Torvalds { 8851da177e4SLinus Torvalds #define C(x) n->x = skb->x 8861da177e4SLinus Torvalds 8871da177e4SLinus Torvalds n->next = n->prev = NULL; 8881da177e4SLinus Torvalds n->sk = NULL; 889dec18810SHerbert Xu __copy_skb_header(n, skb); 890dec18810SHerbert Xu 8911da177e4SLinus Torvalds C(len); 8921da177e4SLinus Torvalds C(data_len); 8933e6b3b2eSAlexey Dobriyan C(mac_len); 894334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 89502f1c89dSPaul Moore n->cloned = 1; 8961da177e4SLinus Torvalds n->nohdr = 0; 897b13dda9fSEric Dumazet n->peeked = 0; 898e78bfb07SStefano Brivio C(pfmemalloc); 8991da177e4SLinus Torvalds n->destructor = NULL; 9001da177e4SLinus Torvalds C(tail); 9011da177e4SLinus Torvalds C(end); 90202f1c89dSPaul Moore C(head); 903d3836f21SEric Dumazet C(head_frag); 90402f1c89dSPaul Moore C(data); 90502f1c89dSPaul Moore C(truesize); 90663354797SReshetova, Elena refcount_set(&n->users, 1); 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 9091da177e4SLinus Torvalds skb->cloned = 1; 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds return n; 912e0053ec0SHerbert Xu #undef C 913e0053ec0SHerbert Xu } 914e0053ec0SHerbert Xu 915e0053ec0SHerbert Xu /** 916e0053ec0SHerbert Xu * skb_morph - morph one skb into another 917e0053ec0SHerbert Xu * @dst: the skb to receive the contents 918e0053ec0SHerbert Xu * @src: the skb to supply the contents 919e0053ec0SHerbert Xu * 920e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 921e0053ec0SHerbert Xu * supplied by the user. 922e0053ec0SHerbert Xu * 923e0053ec0SHerbert Xu * The target skb is returned upon exit. 924e0053ec0SHerbert Xu */ 925e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 926e0053ec0SHerbert Xu { 9272d4baff8SHerbert Xu skb_release_all(dst); 928e0053ec0SHerbert Xu return __skb_clone(dst, src); 929e0053ec0SHerbert Xu } 930e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 931e0053ec0SHerbert Xu 9326f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 933a91dbff5SWillem de Bruijn { 934a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 935a91dbff5SWillem de Bruijn struct user_struct *user; 936a91dbff5SWillem de Bruijn 937a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 938a91dbff5SWillem de Bruijn return 0; 939a91dbff5SWillem de Bruijn 940a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 941a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 942a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 943a91dbff5SWillem de Bruijn 944a91dbff5SWillem de Bruijn do { 945a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 946a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 947a91dbff5SWillem de Bruijn if (new_pg > max_pg) 948a91dbff5SWillem de Bruijn return -ENOBUFS; 949a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 950a91dbff5SWillem de Bruijn old_pg); 951a91dbff5SWillem de Bruijn 952a91dbff5SWillem de Bruijn if (!mmp->user) { 953a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 954a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 955a91dbff5SWillem de Bruijn } else { 956a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 957a91dbff5SWillem de Bruijn } 958a91dbff5SWillem de Bruijn 959a91dbff5SWillem de Bruijn return 0; 960a91dbff5SWillem de Bruijn } 9616f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 962a91dbff5SWillem de Bruijn 9636f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 964a91dbff5SWillem de Bruijn { 965a91dbff5SWillem de Bruijn if (mmp->user) { 966a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 967a91dbff5SWillem de Bruijn free_uid(mmp->user); 968a91dbff5SWillem de Bruijn } 969a91dbff5SWillem de Bruijn } 9706f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 971a91dbff5SWillem de Bruijn 97252267790SWillem de Bruijn struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size) 97352267790SWillem de Bruijn { 97452267790SWillem de Bruijn struct ubuf_info *uarg; 97552267790SWillem de Bruijn struct sk_buff *skb; 97652267790SWillem de Bruijn 97752267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 97852267790SWillem de Bruijn 97952267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 98052267790SWillem de Bruijn if (!skb) 98152267790SWillem de Bruijn return NULL; 98252267790SWillem de Bruijn 98352267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 98452267790SWillem de Bruijn uarg = (void *)skb->cb; 985a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 986a91dbff5SWillem de Bruijn 987a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 988a91dbff5SWillem de Bruijn kfree_skb(skb); 989a91dbff5SWillem de Bruijn return NULL; 990a91dbff5SWillem de Bruijn } 99152267790SWillem de Bruijn 99252267790SWillem de Bruijn uarg->callback = sock_zerocopy_callback; 9934ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 9944ab6c99dSWillem de Bruijn uarg->len = 1; 9954ab6c99dSWillem de Bruijn uarg->bytelen = size; 99652267790SWillem de Bruijn uarg->zerocopy = 1; 997c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 99852267790SWillem de Bruijn sock_hold(sk); 99952267790SWillem de Bruijn 100052267790SWillem de Bruijn return uarg; 100152267790SWillem de Bruijn } 100252267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_alloc); 100352267790SWillem de Bruijn 100452267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 100552267790SWillem de Bruijn { 100652267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 100752267790SWillem de Bruijn } 100852267790SWillem de Bruijn 10094ab6c99dSWillem de Bruijn struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 10104ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 10114ab6c99dSWillem de Bruijn { 10124ab6c99dSWillem de Bruijn if (uarg) { 10134ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 10144ab6c99dSWillem de Bruijn u32 bytelen, next; 10154ab6c99dSWillem de Bruijn 10164ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 10174ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 10184ab6c99dSWillem de Bruijn */ 10194ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 10204ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 10214ab6c99dSWillem de Bruijn return NULL; 10224ab6c99dSWillem de Bruijn } 10234ab6c99dSWillem de Bruijn 10244ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 10254ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 10264ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 10274ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 10284ab6c99dSWillem de Bruijn goto new_alloc; 10294ab6c99dSWillem de Bruijn return NULL; 10304ab6c99dSWillem de Bruijn } 10314ab6c99dSWillem de Bruijn 10324ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 10334ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1034a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1035a91dbff5SWillem de Bruijn return NULL; 10364ab6c99dSWillem de Bruijn uarg->len++; 10374ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 10384ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1039100f6d8eSWillem de Bruijn 1040100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1041100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 1042db5bce32SEric Dumazet sock_zerocopy_get(uarg); 1043100f6d8eSWillem de Bruijn 10444ab6c99dSWillem de Bruijn return uarg; 10454ab6c99dSWillem de Bruijn } 10464ab6c99dSWillem de Bruijn } 10474ab6c99dSWillem de Bruijn 10484ab6c99dSWillem de Bruijn new_alloc: 10494ab6c99dSWillem de Bruijn return sock_zerocopy_alloc(sk, size); 10504ab6c99dSWillem de Bruijn } 10514ab6c99dSWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_realloc); 10524ab6c99dSWillem de Bruijn 10534ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 10544ab6c99dSWillem de Bruijn { 10554ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 10564ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 10574ab6c99dSWillem de Bruijn u64 sum_len; 10584ab6c99dSWillem de Bruijn 10594ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 10604ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 10614ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 10624ab6c99dSWillem de Bruijn 10634ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 10644ab6c99dSWillem de Bruijn return false; 10654ab6c99dSWillem de Bruijn 10664ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 10674ab6c99dSWillem de Bruijn return false; 10684ab6c99dSWillem de Bruijn 10694ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 10704ab6c99dSWillem de Bruijn return true; 10714ab6c99dSWillem de Bruijn } 10724ab6c99dSWillem de Bruijn 107352267790SWillem de Bruijn void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) 107452267790SWillem de Bruijn { 10754ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 107652267790SWillem de Bruijn struct sock_exterr_skb *serr; 107752267790SWillem de Bruijn struct sock *sk = skb->sk; 10784ab6c99dSWillem de Bruijn struct sk_buff_head *q; 10794ab6c99dSWillem de Bruijn unsigned long flags; 10804ab6c99dSWillem de Bruijn u32 lo, hi; 10814ab6c99dSWillem de Bruijn u16 len; 108252267790SWillem de Bruijn 1083ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1084ccaffff1SWillem de Bruijn 10854ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 10864ab6c99dSWillem de Bruijn * so do not queue a completion notification 10874ab6c99dSWillem de Bruijn */ 10884ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 108952267790SWillem de Bruijn goto release; 109052267790SWillem de Bruijn 10914ab6c99dSWillem de Bruijn len = uarg->len; 10924ab6c99dSWillem de Bruijn lo = uarg->id; 10934ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 10944ab6c99dSWillem de Bruijn 109552267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 109652267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 109752267790SWillem de Bruijn serr->ee.ee_errno = 0; 109852267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 10994ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 11004ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 110152267790SWillem de Bruijn if (!success) 110252267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 110352267790SWillem de Bruijn 11044ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 11054ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 11064ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 11074ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 11084ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 11094ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 111052267790SWillem de Bruijn skb = NULL; 11114ab6c99dSWillem de Bruijn } 11124ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 111352267790SWillem de Bruijn 111452267790SWillem de Bruijn sk->sk_error_report(sk); 111552267790SWillem de Bruijn 111652267790SWillem de Bruijn release: 111752267790SWillem de Bruijn consume_skb(skb); 111852267790SWillem de Bruijn sock_put(sk); 111952267790SWillem de Bruijn } 112052267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_callback); 112152267790SWillem de Bruijn 112252267790SWillem de Bruijn void sock_zerocopy_put(struct ubuf_info *uarg) 112352267790SWillem de Bruijn { 1124c1d1b437SEric Dumazet if (uarg && refcount_dec_and_test(&uarg->refcnt)) { 112552267790SWillem de Bruijn if (uarg->callback) 112652267790SWillem de Bruijn uarg->callback(uarg, uarg->zerocopy); 112752267790SWillem de Bruijn else 112852267790SWillem de Bruijn consume_skb(skb_from_uarg(uarg)); 112952267790SWillem de Bruijn } 113052267790SWillem de Bruijn } 113152267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put); 113252267790SWillem de Bruijn 113352900d22SWillem de Bruijn void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 113452267790SWillem de Bruijn { 113552267790SWillem de Bruijn if (uarg) { 113652267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 113752267790SWillem de Bruijn 113852267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 11394ab6c99dSWillem de Bruijn uarg->len--; 114052267790SWillem de Bruijn 114152900d22SWillem de Bruijn if (have_uref) 114252267790SWillem de Bruijn sock_zerocopy_put(uarg); 114352267790SWillem de Bruijn } 114452267790SWillem de Bruijn } 114552267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); 114652267790SWillem de Bruijn 1147b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1148b5947e5dSWillem de Bruijn { 1149b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1150b5947e5dSWillem de Bruijn } 1151b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1152b5947e5dSWillem de Bruijn 115352267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 115452267790SWillem de Bruijn struct msghdr *msg, int len, 115552267790SWillem de Bruijn struct ubuf_info *uarg) 115652267790SWillem de Bruijn { 11574ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 115852267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 115952267790SWillem de Bruijn int err, orig_len = skb->len; 116052267790SWillem de Bruijn 11614ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 11624ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 11634ab6c99dSWillem de Bruijn */ 11644ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 11654ab6c99dSWillem de Bruijn return -EEXIST; 11664ab6c99dSWillem de Bruijn 116752267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 116852267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 116954d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 117054d43117SWillem de Bruijn 117152267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 117252267790SWillem de Bruijn msg->msg_iter = orig_iter; 117354d43117SWillem de Bruijn skb->sk = sk; 117452267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 117554d43117SWillem de Bruijn skb->sk = save_sk; 117652267790SWillem de Bruijn return err; 117752267790SWillem de Bruijn } 117852267790SWillem de Bruijn 117952900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 118052267790SWillem de Bruijn return skb->len - orig_len; 118152267790SWillem de Bruijn } 118252267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 118352267790SWillem de Bruijn 11841f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 118552267790SWillem de Bruijn gfp_t gfp_mask) 118652267790SWillem de Bruijn { 118752267790SWillem de Bruijn if (skb_zcopy(orig)) { 118852267790SWillem de Bruijn if (skb_zcopy(nskb)) { 118952267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 119052267790SWillem de Bruijn if (!gfp_mask) { 119152267790SWillem de Bruijn WARN_ON_ONCE(1); 119252267790SWillem de Bruijn return -ENOMEM; 119352267790SWillem de Bruijn } 119452267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 119552267790SWillem de Bruijn return 0; 119652267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 119752267790SWillem de Bruijn return -EIO; 119852267790SWillem de Bruijn } 119952900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 120052267790SWillem de Bruijn } 120152267790SWillem de Bruijn return 0; 120252267790SWillem de Bruijn } 120352267790SWillem de Bruijn 12042c53040fSBen Hutchings /** 12052c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 120648c83012SMichael S. Tsirkin * @skb: the skb to modify 120748c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 120848c83012SMichael S. Tsirkin * 120948c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 121048c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 121148c83012SMichael S. Tsirkin * to userspace pages. 121248c83012SMichael S. Tsirkin * 121348c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 121448c83012SMichael S. Tsirkin * %GFP_ATOMIC. 121548c83012SMichael S. Tsirkin * 121648c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 121748c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 121848c83012SMichael S. Tsirkin */ 121948c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1220a6686f2fSShirley Ma { 1221a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1222a6686f2fSShirley Ma struct page *page, *head = NULL; 12233ece7826SWillem de Bruijn int i, new_frags; 12243ece7826SWillem de Bruijn u32 d_off; 1225a6686f2fSShirley Ma 12263ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 12273ece7826SWillem de Bruijn return -EINVAL; 12283ece7826SWillem de Bruijn 1229f72c4ac6SWillem de Bruijn if (!num_frags) 1230f72c4ac6SWillem de Bruijn goto release; 1231f72c4ac6SWillem de Bruijn 12323ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 12333ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 123402756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1235a6686f2fSShirley Ma if (!page) { 1236a6686f2fSShirley Ma while (head) { 123740dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1238a6686f2fSShirley Ma put_page(head); 1239a6686f2fSShirley Ma head = next; 1240a6686f2fSShirley Ma } 1241a6686f2fSShirley Ma return -ENOMEM; 1242a6686f2fSShirley Ma } 12433ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 12443ece7826SWillem de Bruijn head = page; 12453ece7826SWillem de Bruijn } 12463ece7826SWillem de Bruijn 12473ece7826SWillem de Bruijn page = head; 12483ece7826SWillem de Bruijn d_off = 0; 12493ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 12503ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 12513ece7826SWillem de Bruijn u32 p_off, p_len, copied; 12523ece7826SWillem de Bruijn struct page *p; 12533ece7826SWillem de Bruijn u8 *vaddr; 1254c613c209SWillem de Bruijn 1255c613c209SWillem de Bruijn skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f), 1256c613c209SWillem de Bruijn p, p_off, p_len, copied) { 12573ece7826SWillem de Bruijn u32 copy, done = 0; 1258c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 12593ece7826SWillem de Bruijn 12603ece7826SWillem de Bruijn while (done < p_len) { 12613ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 12623ece7826SWillem de Bruijn d_off = 0; 12633ece7826SWillem de Bruijn page = (struct page *)page_private(page); 12643ece7826SWillem de Bruijn } 12653ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 12663ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 12673ece7826SWillem de Bruijn vaddr + p_off + done, copy); 12683ece7826SWillem de Bruijn done += copy; 12693ece7826SWillem de Bruijn d_off += copy; 12703ece7826SWillem de Bruijn } 127151c56b00SEric Dumazet kunmap_atomic(vaddr); 1272c613c209SWillem de Bruijn } 1273a6686f2fSShirley Ma } 1274a6686f2fSShirley Ma 1275a6686f2fSShirley Ma /* skb frags release userspace buffers */ 127602756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1277a8605c60SIan Campbell skb_frag_unref(skb, i); 1278a6686f2fSShirley Ma 1279a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 12803ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 12813ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 128240dadff2SSunghan Suh head = (struct page *)page_private(head); 1283a6686f2fSShirley Ma } 12843ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 12853ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 128648c83012SMichael S. Tsirkin 1287b90ddd56SWillem de Bruijn release: 12881f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1289a6686f2fSShirley Ma return 0; 1290a6686f2fSShirley Ma } 1291dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1292a6686f2fSShirley Ma 1293e0053ec0SHerbert Xu /** 1294e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1295e0053ec0SHerbert Xu * @skb: buffer to clone 1296e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1297e0053ec0SHerbert Xu * 1298e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1299e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1300e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1301e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1302e0053ec0SHerbert Xu * 1303e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1304e0053ec0SHerbert Xu * %GFP_ATOMIC. 1305e0053ec0SHerbert Xu */ 1306e0053ec0SHerbert Xu 1307e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1308e0053ec0SHerbert Xu { 1309d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1310d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1311d0bf4a9eSEric Dumazet skb1); 13126ffe75ebSEric Dumazet struct sk_buff *n; 1313e0053ec0SHerbert Xu 131470008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1315a6686f2fSShirley Ma return NULL; 1316a6686f2fSShirley Ma 1317e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 13182638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 13196ffe75ebSEric Dumazet n = &fclones->skb2; 13202638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1321e0053ec0SHerbert Xu } else { 1322c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1323c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1324c93bdd0eSMel Gorman 1325e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1326e0053ec0SHerbert Xu if (!n) 1327e0053ec0SHerbert Xu return NULL; 1328fe55f6d5SVegard Nossum 1329e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1330e0053ec0SHerbert Xu } 1331e0053ec0SHerbert Xu 1332e0053ec0SHerbert Xu return __skb_clone(n, skb); 13331da177e4SLinus Torvalds } 1334b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 13351da177e4SLinus Torvalds 1336b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1337f5b17294SPravin B Shelar { 1338030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1339030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1340030737bcSEric Dumazet skb->csum_start += off; 1341f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1342f5b17294SPravin B Shelar skb->transport_header += off; 1343f5b17294SPravin B Shelar skb->network_header += off; 1344f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1345f5b17294SPravin B Shelar skb->mac_header += off; 1346f5b17294SPravin B Shelar skb->inner_transport_header += off; 1347f5b17294SPravin B Shelar skb->inner_network_header += off; 1348aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1349f5b17294SPravin B Shelar } 1350b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1351f5b17294SPravin B Shelar 135208303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 13531da177e4SLinus Torvalds { 1354dec18810SHerbert Xu __copy_skb_header(new, old); 1355dec18810SHerbert Xu 13567967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 13577967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 13587967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 13591da177e4SLinus Torvalds } 136008303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 13611da177e4SLinus Torvalds 1362c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1363c93bdd0eSMel Gorman { 1364c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1365c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1366c93bdd0eSMel Gorman return 0; 1367c93bdd0eSMel Gorman } 1368c93bdd0eSMel Gorman 13691da177e4SLinus Torvalds /** 13701da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 13711da177e4SLinus Torvalds * @skb: buffer to copy 13721da177e4SLinus Torvalds * @gfp_mask: allocation priority 13731da177e4SLinus Torvalds * 13741da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 13751da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 13761da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 13771da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 13781da177e4SLinus Torvalds * 13791da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 13801da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 13811da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 13821da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 13831da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 13841da177e4SLinus Torvalds */ 13851da177e4SLinus Torvalds 1386dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 13871da177e4SLinus Torvalds { 13886602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1389ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1390c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1391c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 13926602cebbSEric Dumazet 13931da177e4SLinus Torvalds if (!n) 13941da177e4SLinus Torvalds return NULL; 13951da177e4SLinus Torvalds 13961da177e4SLinus Torvalds /* Set the data pointer */ 13971da177e4SLinus Torvalds skb_reserve(n, headerlen); 13981da177e4SLinus Torvalds /* Set the tail pointer and length */ 13991da177e4SLinus Torvalds skb_put(n, skb->len); 14001da177e4SLinus Torvalds 14019f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 14021da177e4SLinus Torvalds 140308303c18SIlya Lesokhin skb_copy_header(n, skb); 14041da177e4SLinus Torvalds return n; 14051da177e4SLinus Torvalds } 1406b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 14071da177e4SLinus Torvalds 14081da177e4SLinus Torvalds /** 1409bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 14101da177e4SLinus Torvalds * @skb: buffer to copy 1411117632e6SEric Dumazet * @headroom: headroom of new skb 14121da177e4SLinus Torvalds * @gfp_mask: allocation priority 1413bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1414bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1415bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 14161da177e4SLinus Torvalds * 14171da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 14181da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 14191da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 14201da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 14211da177e4SLinus Torvalds * or the pointer to the buffer on success. 14221da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 14231da177e4SLinus Torvalds */ 14241da177e4SLinus Torvalds 1425bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1426bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 14271da177e4SLinus Torvalds { 1428117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1429bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1430bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 14316602cebbSEric Dumazet 14321da177e4SLinus Torvalds if (!n) 14331da177e4SLinus Torvalds goto out; 14341da177e4SLinus Torvalds 14351da177e4SLinus Torvalds /* Set the data pointer */ 1436117632e6SEric Dumazet skb_reserve(n, headroom); 14371da177e4SLinus Torvalds /* Set the tail pointer and length */ 14381da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 14391da177e4SLinus Torvalds /* Copy the bytes */ 1440d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 14411da177e4SLinus Torvalds 144225f484a6SHerbert Xu n->truesize += skb->data_len; 14431da177e4SLinus Torvalds n->data_len = skb->data_len; 14441da177e4SLinus Torvalds n->len = skb->len; 14451da177e4SLinus Torvalds 14461da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 14471da177e4SLinus Torvalds int i; 14481da177e4SLinus Torvalds 14491f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 14501f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 14511511022cSDan Carpenter kfree_skb(n); 14521511022cSDan Carpenter n = NULL; 1453a6686f2fSShirley Ma goto out; 1454a6686f2fSShirley Ma } 14551da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 14561da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1457ea2ab693SIan Campbell skb_frag_ref(skb, i); 14581da177e4SLinus Torvalds } 14591da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 14601da177e4SLinus Torvalds } 14611da177e4SLinus Torvalds 146221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 14631da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 14641da177e4SLinus Torvalds skb_clone_fraglist(n); 14651da177e4SLinus Torvalds } 14661da177e4SLinus Torvalds 146708303c18SIlya Lesokhin skb_copy_header(n, skb); 14681da177e4SLinus Torvalds out: 14691da177e4SLinus Torvalds return n; 14701da177e4SLinus Torvalds } 1471bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds /** 14741da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 14751da177e4SLinus Torvalds * @skb: buffer to reallocate 14761da177e4SLinus Torvalds * @nhead: room to add at head 14771da177e4SLinus Torvalds * @ntail: room to add at tail 14781da177e4SLinus Torvalds * @gfp_mask: allocation priority 14791da177e4SLinus Torvalds * 1480bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1481bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 14821da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 14831da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 14841da177e4SLinus Torvalds * 14851da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 14861da177e4SLinus Torvalds * reloaded after call to this function. 14871da177e4SLinus Torvalds */ 14881da177e4SLinus Torvalds 148986a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1490dd0fc66fSAl Viro gfp_t gfp_mask) 14911da177e4SLinus Torvalds { 1492158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1493158f323bSEric Dumazet int size = osize + nhead + ntail; 14941da177e4SLinus Torvalds long off; 1495158f323bSEric Dumazet u8 *data; 14961da177e4SLinus Torvalds 14974edd87adSHerbert Xu BUG_ON(nhead < 0); 14984edd87adSHerbert Xu 14999f77fad3STim Hansen BUG_ON(skb_shared(skb)); 15001da177e4SLinus Torvalds 15011da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 15021da177e4SLinus Torvalds 1503c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1504c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1505c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1506c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 15071da177e4SLinus Torvalds if (!data) 15081da177e4SLinus Torvalds goto nodata; 150987151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 15101da177e4SLinus Torvalds 15111da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 15126602cebbSEric Dumazet * optimized for the cases when header is void. 15136602cebbSEric Dumazet */ 15146602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 15156602cebbSEric Dumazet 15166602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 15176602cebbSEric Dumazet skb_shinfo(skb), 1518fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 15191da177e4SLinus Torvalds 15203e24591aSAlexander Duyck /* 15213e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 15223e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 15233e24591aSAlexander Duyck * be since all we did is relocate the values 15243e24591aSAlexander Duyck */ 15253e24591aSAlexander Duyck if (skb_cloned(skb)) { 152670008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1527a6686f2fSShirley Ma goto nofrags; 15281f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1529c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 15301da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1531ea2ab693SIan Campbell skb_frag_ref(skb, i); 15321da177e4SLinus Torvalds 153321dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 15341da177e4SLinus Torvalds skb_clone_fraglist(skb); 15351da177e4SLinus Torvalds 15361da177e4SLinus Torvalds skb_release_data(skb); 15373e24591aSAlexander Duyck } else { 15383e24591aSAlexander Duyck skb_free_head(skb); 15391fd63041SEric Dumazet } 15401da177e4SLinus Torvalds off = (data + nhead) - skb->head; 15411da177e4SLinus Torvalds 15421da177e4SLinus Torvalds skb->head = data; 1543d3836f21SEric Dumazet skb->head_frag = 0; 15441da177e4SLinus Torvalds skb->data += off; 15454305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 15464305b541SArnaldo Carvalho de Melo skb->end = size; 154756eb8882SPatrick McHardy off = nhead; 15484305b541SArnaldo Carvalho de Melo #else 15494305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 155056eb8882SPatrick McHardy #endif 155127a884dcSArnaldo Carvalho de Melo skb->tail += off; 1552b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 15531da177e4SLinus Torvalds skb->cloned = 0; 1554334a8132SPatrick McHardy skb->hdr_len = 0; 15551da177e4SLinus Torvalds skb->nohdr = 0; 15561da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1557158f323bSEric Dumazet 1558de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1559de8f3a83SDaniel Borkmann 1560158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1561158f323bSEric Dumazet * For the moment, we really care of rx path, or 1562158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1563158f323bSEric Dumazet */ 1564158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1565158f323bSEric Dumazet skb->truesize += size - osize; 1566158f323bSEric Dumazet 15671da177e4SLinus Torvalds return 0; 15681da177e4SLinus Torvalds 1569a6686f2fSShirley Ma nofrags: 1570a6686f2fSShirley Ma kfree(data); 15711da177e4SLinus Torvalds nodata: 15721da177e4SLinus Torvalds return -ENOMEM; 15731da177e4SLinus Torvalds } 1574b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 15751da177e4SLinus Torvalds 15761da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 15771da177e4SLinus Torvalds 15781da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 15791da177e4SLinus Torvalds { 15801da177e4SLinus Torvalds struct sk_buff *skb2; 15811da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 15821da177e4SLinus Torvalds 15831da177e4SLinus Torvalds if (delta <= 0) 15841da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 15851da177e4SLinus Torvalds else { 15861da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 15871da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 15881da177e4SLinus Torvalds GFP_ATOMIC)) { 15891da177e4SLinus Torvalds kfree_skb(skb2); 15901da177e4SLinus Torvalds skb2 = NULL; 15911da177e4SLinus Torvalds } 15921da177e4SLinus Torvalds } 15931da177e4SLinus Torvalds return skb2; 15941da177e4SLinus Torvalds } 1595b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 15961da177e4SLinus Torvalds 15971da177e4SLinus Torvalds /** 15981da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 15991da177e4SLinus Torvalds * @skb: buffer to copy 16001da177e4SLinus Torvalds * @newheadroom: new free bytes at head 16011da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 16021da177e4SLinus Torvalds * @gfp_mask: allocation priority 16031da177e4SLinus Torvalds * 16041da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 16051da177e4SLinus Torvalds * allocate additional space. 16061da177e4SLinus Torvalds * 16071da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 16081da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 16091da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 16101da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 16111da177e4SLinus Torvalds * 16121da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 16131da177e4SLinus Torvalds * is called from an interrupt. 16141da177e4SLinus Torvalds */ 16151da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 161686a76cafSVictor Fusco int newheadroom, int newtailroom, 1617dd0fc66fSAl Viro gfp_t gfp_mask) 16181da177e4SLinus Torvalds { 16191da177e4SLinus Torvalds /* 16201da177e4SLinus Torvalds * Allocate the copy buffer 16211da177e4SLinus Torvalds */ 1622c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1623c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1624c93bdd0eSMel Gorman NUMA_NO_NODE); 1625efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 16261da177e4SLinus Torvalds int head_copy_len, head_copy_off; 16271da177e4SLinus Torvalds 16281da177e4SLinus Torvalds if (!n) 16291da177e4SLinus Torvalds return NULL; 16301da177e4SLinus Torvalds 16311da177e4SLinus Torvalds skb_reserve(n, newheadroom); 16321da177e4SLinus Torvalds 16331da177e4SLinus Torvalds /* Set the tail pointer and length */ 16341da177e4SLinus Torvalds skb_put(n, skb->len); 16351da177e4SLinus Torvalds 1636efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 16371da177e4SLinus Torvalds head_copy_off = 0; 16381da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 16391da177e4SLinus Torvalds head_copy_len = newheadroom; 16401da177e4SLinus Torvalds else 16411da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 16421da177e4SLinus Torvalds 16431da177e4SLinus Torvalds /* Copy the linear header and data. */ 16449f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 16459f77fad3STim Hansen skb->len + head_copy_len)); 16461da177e4SLinus Torvalds 164708303c18SIlya Lesokhin skb_copy_header(n, skb); 16481da177e4SLinus Torvalds 1649030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1650efd1e8d5SPatrick McHardy 16511da177e4SLinus Torvalds return n; 16521da177e4SLinus Torvalds } 1653b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds /** 1656cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 16571da177e4SLinus Torvalds * @skb: buffer to pad 16581da177e4SLinus Torvalds * @pad: space to pad 1659cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 16601da177e4SLinus Torvalds * 16611da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 16621da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 16631da177e4SLinus Torvalds * beyond the buffer end onto the wire. 16641da177e4SLinus Torvalds * 1665cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1666cd0a137aSFlorian Fainelli * if @free_on_error is true. 16671da177e4SLinus Torvalds */ 16681da177e4SLinus Torvalds 1669cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 16701da177e4SLinus Torvalds { 16715b057c6bSHerbert Xu int err; 16725b057c6bSHerbert Xu int ntail; 16731da177e4SLinus Torvalds 16741da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 16755b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 16761da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 16775b057c6bSHerbert Xu return 0; 16781da177e4SLinus Torvalds } 16791da177e4SLinus Torvalds 16804305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 16815b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 16825b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 16835b057c6bSHerbert Xu if (unlikely(err)) 16845b057c6bSHerbert Xu goto free_skb; 16855b057c6bSHerbert Xu } 16865b057c6bSHerbert Xu 16875b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 16885b057c6bSHerbert Xu * to be audited. 16895b057c6bSHerbert Xu */ 16905b057c6bSHerbert Xu err = skb_linearize(skb); 16915b057c6bSHerbert Xu if (unlikely(err)) 16925b057c6bSHerbert Xu goto free_skb; 16935b057c6bSHerbert Xu 16945b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 16955b057c6bSHerbert Xu return 0; 16965b057c6bSHerbert Xu 16975b057c6bSHerbert Xu free_skb: 1698cd0a137aSFlorian Fainelli if (free_on_error) 16991da177e4SLinus Torvalds kfree_skb(skb); 17005b057c6bSHerbert Xu return err; 17011da177e4SLinus Torvalds } 1702cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 17031da177e4SLinus Torvalds 17040dde3e16SIlpo Järvinen /** 17050c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 17060c7ddf36SMathias Krause * @skb: start of the buffer to use 17070c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 17080c7ddf36SMathias Krause * @len: amount of data to add 17090c7ddf36SMathias Krause * 17100c7ddf36SMathias Krause * This function extends the used data area of the potentially 17110c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 17120c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 17130c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 17140c7ddf36SMathias Krause * returned. 17150c7ddf36SMathias Krause */ 17160c7ddf36SMathias Krause 17174df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 17180c7ddf36SMathias Krause { 17190c7ddf36SMathias Krause if (tail != skb) { 17200c7ddf36SMathias Krause skb->data_len += len; 17210c7ddf36SMathias Krause skb->len += len; 17220c7ddf36SMathias Krause } 17230c7ddf36SMathias Krause return skb_put(tail, len); 17240c7ddf36SMathias Krause } 17250c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 17260c7ddf36SMathias Krause 17270c7ddf36SMathias Krause /** 17280dde3e16SIlpo Järvinen * skb_put - add data to a buffer 17290dde3e16SIlpo Järvinen * @skb: buffer to use 17300dde3e16SIlpo Järvinen * @len: amount of data to add 17310dde3e16SIlpo Järvinen * 17320dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 17330dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 17340dde3e16SIlpo Järvinen * first byte of the extra data is returned. 17350dde3e16SIlpo Järvinen */ 17364df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 17370dde3e16SIlpo Järvinen { 17384df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 17390dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 17400dde3e16SIlpo Järvinen skb->tail += len; 17410dde3e16SIlpo Järvinen skb->len += len; 17420dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 17430dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 17440dde3e16SIlpo Järvinen return tmp; 17450dde3e16SIlpo Järvinen } 17460dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 17470dde3e16SIlpo Järvinen 17486be8ac2fSIlpo Järvinen /** 1749c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1750c2aa270aSIlpo Järvinen * @skb: buffer to use 1751c2aa270aSIlpo Järvinen * @len: amount of data to add 1752c2aa270aSIlpo Järvinen * 1753c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1754c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1755c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1756c2aa270aSIlpo Järvinen */ 1757d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 1758c2aa270aSIlpo Järvinen { 1759c2aa270aSIlpo Järvinen skb->data -= len; 1760c2aa270aSIlpo Järvinen skb->len += len; 1761c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 1762c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1763c2aa270aSIlpo Järvinen return skb->data; 1764c2aa270aSIlpo Järvinen } 1765c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1766c2aa270aSIlpo Järvinen 1767c2aa270aSIlpo Järvinen /** 17686be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 17696be8ac2fSIlpo Järvinen * @skb: buffer to use 17706be8ac2fSIlpo Järvinen * @len: amount of data to remove 17716be8ac2fSIlpo Järvinen * 17726be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 17736be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 17746be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 17756be8ac2fSIlpo Järvinen * the old data. 17766be8ac2fSIlpo Järvinen */ 1777af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 17786be8ac2fSIlpo Järvinen { 177947d29646SDavid S. Miller return skb_pull_inline(skb, len); 17806be8ac2fSIlpo Järvinen } 17816be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 17826be8ac2fSIlpo Järvinen 1783419ae74eSIlpo Järvinen /** 1784419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1785419ae74eSIlpo Järvinen * @skb: buffer to alter 1786419ae74eSIlpo Järvinen * @len: new length 1787419ae74eSIlpo Järvinen * 1788419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1789419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1790419ae74eSIlpo Järvinen * The skb must be linear. 1791419ae74eSIlpo Järvinen */ 1792419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1793419ae74eSIlpo Järvinen { 1794419ae74eSIlpo Järvinen if (skb->len > len) 1795419ae74eSIlpo Järvinen __skb_trim(skb, len); 1796419ae74eSIlpo Järvinen } 1797419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1798419ae74eSIlpo Järvinen 17993cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 18001da177e4SLinus Torvalds */ 18011da177e4SLinus Torvalds 18023cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 18031da177e4SLinus Torvalds { 180427b437c8SHerbert Xu struct sk_buff **fragp; 180527b437c8SHerbert Xu struct sk_buff *frag; 18061da177e4SLinus Torvalds int offset = skb_headlen(skb); 18071da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 18081da177e4SLinus Torvalds int i; 180927b437c8SHerbert Xu int err; 181027b437c8SHerbert Xu 181127b437c8SHerbert Xu if (skb_cloned(skb) && 181227b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 181327b437c8SHerbert Xu return err; 18141da177e4SLinus Torvalds 1815f4d26fb3SHerbert Xu i = 0; 1816f4d26fb3SHerbert Xu if (offset >= len) 1817f4d26fb3SHerbert Xu goto drop_pages; 1818f4d26fb3SHerbert Xu 1819f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 18209e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 182127b437c8SHerbert Xu 182227b437c8SHerbert Xu if (end < len) { 18231da177e4SLinus Torvalds offset = end; 182427b437c8SHerbert Xu continue; 18251da177e4SLinus Torvalds } 18261da177e4SLinus Torvalds 18279e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 182827b437c8SHerbert Xu 1829f4d26fb3SHerbert Xu drop_pages: 183027b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 183127b437c8SHerbert Xu 183227b437c8SHerbert Xu for (; i < nfrags; i++) 1833ea2ab693SIan Campbell skb_frag_unref(skb, i); 183427b437c8SHerbert Xu 183521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 183627b437c8SHerbert Xu skb_drop_fraglist(skb); 1837f4d26fb3SHerbert Xu goto done; 183827b437c8SHerbert Xu } 183927b437c8SHerbert Xu 184027b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 184127b437c8SHerbert Xu fragp = &frag->next) { 184227b437c8SHerbert Xu int end = offset + frag->len; 184327b437c8SHerbert Xu 184427b437c8SHerbert Xu if (skb_shared(frag)) { 184527b437c8SHerbert Xu struct sk_buff *nfrag; 184627b437c8SHerbert Xu 184727b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 184827b437c8SHerbert Xu if (unlikely(!nfrag)) 184927b437c8SHerbert Xu return -ENOMEM; 185027b437c8SHerbert Xu 185127b437c8SHerbert Xu nfrag->next = frag->next; 185285bb2a60SEric Dumazet consume_skb(frag); 185327b437c8SHerbert Xu frag = nfrag; 185427b437c8SHerbert Xu *fragp = frag; 185527b437c8SHerbert Xu } 185627b437c8SHerbert Xu 185727b437c8SHerbert Xu if (end < len) { 185827b437c8SHerbert Xu offset = end; 185927b437c8SHerbert Xu continue; 186027b437c8SHerbert Xu } 186127b437c8SHerbert Xu 186227b437c8SHerbert Xu if (end > len && 186327b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 186427b437c8SHerbert Xu return err; 186527b437c8SHerbert Xu 186627b437c8SHerbert Xu if (frag->next) 186727b437c8SHerbert Xu skb_drop_list(&frag->next); 186827b437c8SHerbert Xu break; 186927b437c8SHerbert Xu } 187027b437c8SHerbert Xu 1871f4d26fb3SHerbert Xu done: 187227b437c8SHerbert Xu if (len > skb_headlen(skb)) { 18731da177e4SLinus Torvalds skb->data_len -= skb->len - len; 18741da177e4SLinus Torvalds skb->len = len; 18751da177e4SLinus Torvalds } else { 18761da177e4SLinus Torvalds skb->len = len; 18771da177e4SLinus Torvalds skb->data_len = 0; 187827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 18791da177e4SLinus Torvalds } 18801da177e4SLinus Torvalds 1881c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1882c21b48ccSEric Dumazet skb_condense(skb); 18831da177e4SLinus Torvalds return 0; 18841da177e4SLinus Torvalds } 1885b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 18861da177e4SLinus Torvalds 188788078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 188888078d98SEric Dumazet */ 188988078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 189088078d98SEric Dumazet { 189188078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 189288078d98SEric Dumazet int delta = skb->len - len; 189388078d98SEric Dumazet 1894d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 1895d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 1896d55bef50SDimitris Michailidis len); 189788078d98SEric Dumazet } 189888078d98SEric Dumazet return __pskb_trim(skb, len); 189988078d98SEric Dumazet } 190088078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 190188078d98SEric Dumazet 19021da177e4SLinus Torvalds /** 19031da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 19041da177e4SLinus Torvalds * @skb: buffer to reallocate 19051da177e4SLinus Torvalds * @delta: number of bytes to advance tail 19061da177e4SLinus Torvalds * 19071da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 19081da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 19091da177e4SLinus Torvalds * data from fragmented part. 19101da177e4SLinus Torvalds * 19111da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 19121da177e4SLinus Torvalds * 19131da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 19141da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 19151da177e4SLinus Torvalds * 19161da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 19171da177e4SLinus Torvalds * reloaded after call to this function. 19181da177e4SLinus Torvalds */ 19191da177e4SLinus Torvalds 19201da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 19211da177e4SLinus Torvalds * when it is necessary. 19221da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 19231da177e4SLinus Torvalds * 2. It may change skb pointers. 19241da177e4SLinus Torvalds * 19251da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 19261da177e4SLinus Torvalds */ 1927af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 19281da177e4SLinus Torvalds { 19291da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 19301da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 19311da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 19321da177e4SLinus Torvalds */ 19334305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 19341da177e4SLinus Torvalds 19351da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 19361da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 19371da177e4SLinus Torvalds GFP_ATOMIC)) 19381da177e4SLinus Torvalds return NULL; 19391da177e4SLinus Torvalds } 19401da177e4SLinus Torvalds 19419f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 19429f77fad3STim Hansen skb_tail_pointer(skb), delta)); 19431da177e4SLinus Torvalds 19441da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 19451da177e4SLinus Torvalds * size of pulled pages. Superb. 19461da177e4SLinus Torvalds */ 194721dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 19481da177e4SLinus Torvalds goto pull_pages; 19491da177e4SLinus Torvalds 19501da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 19511da177e4SLinus Torvalds eat = delta; 19521da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 19539e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 19549e903e08SEric Dumazet 19559e903e08SEric Dumazet if (size >= eat) 19561da177e4SLinus Torvalds goto pull_pages; 19579e903e08SEric Dumazet eat -= size; 19581da177e4SLinus Torvalds } 19591da177e4SLinus Torvalds 19601da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 196109001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 19621da177e4SLinus Torvalds * but taking into account that pulling is expected to 19631da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 19641da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 19651da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 19661da177e4SLinus Torvalds */ 19671da177e4SLinus Torvalds if (eat) { 19681da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 19691da177e4SLinus Torvalds struct sk_buff *clone = NULL; 19701da177e4SLinus Torvalds struct sk_buff *insp = NULL; 19711da177e4SLinus Torvalds 19721da177e4SLinus Torvalds do { 19731da177e4SLinus Torvalds if (list->len <= eat) { 19741da177e4SLinus Torvalds /* Eaten as whole. */ 19751da177e4SLinus Torvalds eat -= list->len; 19761da177e4SLinus Torvalds list = list->next; 19771da177e4SLinus Torvalds insp = list; 19781da177e4SLinus Torvalds } else { 19791da177e4SLinus Torvalds /* Eaten partially. */ 19801da177e4SLinus Torvalds 19811da177e4SLinus Torvalds if (skb_shared(list)) { 19821da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 19831da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 19841da177e4SLinus Torvalds if (!clone) 19851da177e4SLinus Torvalds return NULL; 19861da177e4SLinus Torvalds insp = list->next; 19871da177e4SLinus Torvalds list = clone; 19881da177e4SLinus Torvalds } else { 19891da177e4SLinus Torvalds /* This may be pulled without 19901da177e4SLinus Torvalds * problems. */ 19911da177e4SLinus Torvalds insp = list; 19921da177e4SLinus Torvalds } 19931da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 19941da177e4SLinus Torvalds kfree_skb(clone); 19951da177e4SLinus Torvalds return NULL; 19961da177e4SLinus Torvalds } 19971da177e4SLinus Torvalds break; 19981da177e4SLinus Torvalds } 19991da177e4SLinus Torvalds } while (eat); 20001da177e4SLinus Torvalds 20011da177e4SLinus Torvalds /* Free pulled out fragments. */ 20021da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 20031da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 20041da177e4SLinus Torvalds kfree_skb(list); 20051da177e4SLinus Torvalds } 20061da177e4SLinus Torvalds /* And insert new clone at head. */ 20071da177e4SLinus Torvalds if (clone) { 20081da177e4SLinus Torvalds clone->next = list; 20091da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 20101da177e4SLinus Torvalds } 20111da177e4SLinus Torvalds } 20121da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 20131da177e4SLinus Torvalds 20141da177e4SLinus Torvalds pull_pages: 20151da177e4SLinus Torvalds eat = delta; 20161da177e4SLinus Torvalds k = 0; 20171da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20189e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 20199e903e08SEric Dumazet 20209e903e08SEric Dumazet if (size <= eat) { 2021ea2ab693SIan Campbell skb_frag_unref(skb, i); 20229e903e08SEric Dumazet eat -= size; 20231da177e4SLinus Torvalds } else { 20241da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 20251da177e4SLinus Torvalds if (eat) { 20261da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 20279e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 20283ccc6c6fSlinzhang if (!i) 20293ccc6c6fSlinzhang goto end; 20301da177e4SLinus Torvalds eat = 0; 20311da177e4SLinus Torvalds } 20321da177e4SLinus Torvalds k++; 20331da177e4SLinus Torvalds } 20341da177e4SLinus Torvalds } 20351da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 20361da177e4SLinus Torvalds 20373ccc6c6fSlinzhang end: 20381da177e4SLinus Torvalds skb->tail += delta; 20391da177e4SLinus Torvalds skb->data_len -= delta; 20401da177e4SLinus Torvalds 20411f8b977aSWillem de Bruijn if (!skb->data_len) 20421f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 20431f8b977aSWillem de Bruijn 204427a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 20451da177e4SLinus Torvalds } 2046b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 20471da177e4SLinus Torvalds 204822019b17SEric Dumazet /** 204922019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 205022019b17SEric Dumazet * @skb: source skb 205122019b17SEric Dumazet * @offset: offset in source 205222019b17SEric Dumazet * @to: destination buffer 205322019b17SEric Dumazet * @len: number of bytes to copy 205422019b17SEric Dumazet * 205522019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 205622019b17SEric Dumazet * destination buffer. 205722019b17SEric Dumazet * 205822019b17SEric Dumazet * CAUTION ! : 205922019b17SEric Dumazet * If its prototype is ever changed, 206022019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 206122019b17SEric Dumazet * since it is called from BPF assembly code. 206222019b17SEric Dumazet */ 20631da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 20641da177e4SLinus Torvalds { 20651a028e50SDavid S. Miller int start = skb_headlen(skb); 2066fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2067fbb398a8SDavid S. Miller int i, copy; 20681da177e4SLinus Torvalds 20691da177e4SLinus Torvalds if (offset > (int)skb->len - len) 20701da177e4SLinus Torvalds goto fault; 20711da177e4SLinus Torvalds 20721da177e4SLinus Torvalds /* Copy header. */ 20731a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 20741da177e4SLinus Torvalds if (copy > len) 20751da177e4SLinus Torvalds copy = len; 2076d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 20771da177e4SLinus Torvalds if ((len -= copy) == 0) 20781da177e4SLinus Torvalds return 0; 20791da177e4SLinus Torvalds offset += copy; 20801da177e4SLinus Torvalds to += copy; 20811da177e4SLinus Torvalds } 20821da177e4SLinus Torvalds 20831da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20841a028e50SDavid S. Miller int end; 208551c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 20861da177e4SLinus Torvalds 2087547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20881a028e50SDavid S. Miller 208951c56b00SEric Dumazet end = start + skb_frag_size(f); 20901da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2091c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2092c613c209SWillem de Bruijn struct page *p; 20931da177e4SLinus Torvalds u8 *vaddr; 20941da177e4SLinus Torvalds 20951da177e4SLinus Torvalds if (copy > len) 20961da177e4SLinus Torvalds copy = len; 20971da177e4SLinus Torvalds 2098c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2099c613c209SWillem de Bruijn f->page_offset + offset - start, 2100c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2101c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2102c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 210351c56b00SEric Dumazet kunmap_atomic(vaddr); 2104c613c209SWillem de Bruijn } 21051da177e4SLinus Torvalds 21061da177e4SLinus Torvalds if ((len -= copy) == 0) 21071da177e4SLinus Torvalds return 0; 21081da177e4SLinus Torvalds offset += copy; 21091da177e4SLinus Torvalds to += copy; 21101da177e4SLinus Torvalds } 21111a028e50SDavid S. Miller start = end; 21121da177e4SLinus Torvalds } 21131da177e4SLinus Torvalds 2114fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 21151a028e50SDavid S. Miller int end; 21161da177e4SLinus Torvalds 2117547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21181a028e50SDavid S. Miller 2119fbb398a8SDavid S. Miller end = start + frag_iter->len; 21201da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21211da177e4SLinus Torvalds if (copy > len) 21221da177e4SLinus Torvalds copy = len; 2123fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 21241da177e4SLinus Torvalds goto fault; 21251da177e4SLinus Torvalds if ((len -= copy) == 0) 21261da177e4SLinus Torvalds return 0; 21271da177e4SLinus Torvalds offset += copy; 21281da177e4SLinus Torvalds to += copy; 21291da177e4SLinus Torvalds } 21301a028e50SDavid S. Miller start = end; 21311da177e4SLinus Torvalds } 2132a6686f2fSShirley Ma 21331da177e4SLinus Torvalds if (!len) 21341da177e4SLinus Torvalds return 0; 21351da177e4SLinus Torvalds 21361da177e4SLinus Torvalds fault: 21371da177e4SLinus Torvalds return -EFAULT; 21381da177e4SLinus Torvalds } 2139b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 21401da177e4SLinus Torvalds 21419c55e01cSJens Axboe /* 21429c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 21439c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 21449c55e01cSJens Axboe */ 21459c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 21469c55e01cSJens Axboe { 21478b9d3728SJarek Poplawski put_page(spd->pages[i]); 21488b9d3728SJarek Poplawski } 21499c55e01cSJens Axboe 2150a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 21514fb66994SJarek Poplawski unsigned int *offset, 215218aafc62SEric Dumazet struct sock *sk) 21538b9d3728SJarek Poplawski { 21545640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 21558b9d3728SJarek Poplawski 21565640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 21578b9d3728SJarek Poplawski return NULL; 21584fb66994SJarek Poplawski 21595640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 21604fb66994SJarek Poplawski 21615640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 21625640f768SEric Dumazet page_address(page) + *offset, *len); 21635640f768SEric Dumazet *offset = pfrag->offset; 21645640f768SEric Dumazet pfrag->offset += *len; 21654fb66994SJarek Poplawski 21665640f768SEric Dumazet return pfrag->page; 21679c55e01cSJens Axboe } 21689c55e01cSJens Axboe 216941c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 217041c73a0dSEric Dumazet struct page *page, 217141c73a0dSEric Dumazet unsigned int offset) 217241c73a0dSEric Dumazet { 217341c73a0dSEric Dumazet return spd->nr_pages && 217441c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 217541c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 217641c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 217741c73a0dSEric Dumazet } 217841c73a0dSEric Dumazet 21799c55e01cSJens Axboe /* 21809c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 21819c55e01cSJens Axboe */ 2182a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 218335f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 21844fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 218518aafc62SEric Dumazet bool linear, 21867a67e56fSJarek Poplawski struct sock *sk) 21879c55e01cSJens Axboe { 218841c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2189a108d5f3SDavid S. Miller return true; 21909c55e01cSJens Axboe 21918b9d3728SJarek Poplawski if (linear) { 219218aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 21938b9d3728SJarek Poplawski if (!page) 2194a108d5f3SDavid S. Miller return true; 219541c73a0dSEric Dumazet } 219641c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 219741c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2198a108d5f3SDavid S. Miller return false; 219941c73a0dSEric Dumazet } 22008b9d3728SJarek Poplawski get_page(page); 22019c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 22024fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 22039c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 22049c55e01cSJens Axboe spd->nr_pages++; 22058b9d3728SJarek Poplawski 2206a108d5f3SDavid S. Miller return false; 22079c55e01cSJens Axboe } 22089c55e01cSJens Axboe 2209a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 22102870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 221118aafc62SEric Dumazet unsigned int *len, 2212d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 221335f3d14dSJens Axboe struct sock *sk, 221435f3d14dSJens Axboe struct pipe_inode_info *pipe) 22159c55e01cSJens Axboe { 22162870c43dSOctavian Purdila if (!*len) 2217a108d5f3SDavid S. Miller return true; 22189c55e01cSJens Axboe 22192870c43dSOctavian Purdila /* skip this segment if already processed */ 22202870c43dSOctavian Purdila if (*off >= plen) { 22212870c43dSOctavian Purdila *off -= plen; 2222a108d5f3SDavid S. Miller return false; 22232870c43dSOctavian Purdila } 22242870c43dSOctavian Purdila 22252870c43dSOctavian Purdila /* ignore any bits we already processed */ 22269ca1b22dSEric Dumazet poff += *off; 22279ca1b22dSEric Dumazet plen -= *off; 22282870c43dSOctavian Purdila *off = 0; 22292870c43dSOctavian Purdila 223018aafc62SEric Dumazet do { 223118aafc62SEric Dumazet unsigned int flen = min(*len, plen); 22322870c43dSOctavian Purdila 223318aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 223418aafc62SEric Dumazet linear, sk)) 2235a108d5f3SDavid S. Miller return true; 223618aafc62SEric Dumazet poff += flen; 223718aafc62SEric Dumazet plen -= flen; 22382870c43dSOctavian Purdila *len -= flen; 223918aafc62SEric Dumazet } while (*len && plen); 22402870c43dSOctavian Purdila 2241a108d5f3SDavid S. Miller return false; 2242db43a282SOctavian Purdila } 22439c55e01cSJens Axboe 22449c55e01cSJens Axboe /* 2245a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 22462870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 22479c55e01cSJens Axboe */ 2248a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 224935f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 225035f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 22512870c43dSOctavian Purdila { 22522870c43dSOctavian Purdila int seg; 2253fa9835e5STom Herbert struct sk_buff *iter; 22549c55e01cSJens Axboe 22551d0c0b32SEric Dumazet /* map the linear part : 22562996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 22572996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 22582996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 22599c55e01cSJens Axboe */ 22602870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 22612870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 22622870c43dSOctavian Purdila skb_headlen(skb), 226318aafc62SEric Dumazet offset, len, spd, 22643a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 22651d0c0b32SEric Dumazet sk, pipe)) 2266a108d5f3SDavid S. Miller return true; 22679c55e01cSJens Axboe 22689c55e01cSJens Axboe /* 22699c55e01cSJens Axboe * then map the fragments 22709c55e01cSJens Axboe */ 22719c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 22729c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 22739c55e01cSJens Axboe 2274ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 22759e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 227618aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2277a108d5f3SDavid S. Miller return true; 22789c55e01cSJens Axboe } 22799c55e01cSJens Axboe 2280fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2281fa9835e5STom Herbert if (*offset >= iter->len) { 2282fa9835e5STom Herbert *offset -= iter->len; 2283fa9835e5STom Herbert continue; 2284fa9835e5STom Herbert } 2285fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2286fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2287fa9835e5STom Herbert * case. 2288fa9835e5STom Herbert */ 2289fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2290fa9835e5STom Herbert return true; 2291fa9835e5STom Herbert } 2292fa9835e5STom Herbert 2293a108d5f3SDavid S. Miller return false; 22949c55e01cSJens Axboe } 22959c55e01cSJens Axboe 22969c55e01cSJens Axboe /* 22979c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2298fa9835e5STom Herbert * the fragments, and the frag list. 22999c55e01cSJens Axboe */ 2300a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 23019c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 230225869262SAl Viro unsigned int flags) 23039c55e01cSJens Axboe { 230441c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 230541c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 23069c55e01cSJens Axboe struct splice_pipe_desc spd = { 23079c55e01cSJens Axboe .pages = pages, 23089c55e01cSJens Axboe .partial = partial, 2309047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 231028a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 23119c55e01cSJens Axboe .spd_release = sock_spd_release, 23129c55e01cSJens Axboe }; 231335f3d14dSJens Axboe int ret = 0; 231435f3d14dSJens Axboe 2315fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 23169c55e01cSJens Axboe 2317a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 231825869262SAl Viro ret = splice_to_pipe(pipe, &spd); 23199c55e01cSJens Axboe 232035f3d14dSJens Axboe return ret; 23219c55e01cSJens Axboe } 23222b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 23239c55e01cSJens Axboe 232420bf50deSTom Herbert /* Send skb data on a socket. Socket must be locked. */ 232520bf50deSTom Herbert int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 232620bf50deSTom Herbert int len) 232720bf50deSTom Herbert { 232820bf50deSTom Herbert unsigned int orig_len = len; 232920bf50deSTom Herbert struct sk_buff *head = skb; 233020bf50deSTom Herbert unsigned short fragidx; 233120bf50deSTom Herbert int slen, ret; 233220bf50deSTom Herbert 233320bf50deSTom Herbert do_frag_list: 233420bf50deSTom Herbert 233520bf50deSTom Herbert /* Deal with head data */ 233620bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 233720bf50deSTom Herbert struct kvec kv; 233820bf50deSTom Herbert struct msghdr msg; 233920bf50deSTom Herbert 234020bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 234120bf50deSTom Herbert kv.iov_base = skb->data + offset; 2342db5980d8SJohn Fastabend kv.iov_len = slen; 234320bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 234420bf50deSTom Herbert 234520bf50deSTom Herbert ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen); 234620bf50deSTom Herbert if (ret <= 0) 234720bf50deSTom Herbert goto error; 234820bf50deSTom Herbert 234920bf50deSTom Herbert offset += ret; 235020bf50deSTom Herbert len -= ret; 235120bf50deSTom Herbert } 235220bf50deSTom Herbert 235320bf50deSTom Herbert /* All the data was skb head? */ 235420bf50deSTom Herbert if (!len) 235520bf50deSTom Herbert goto out; 235620bf50deSTom Herbert 235720bf50deSTom Herbert /* Make offset relative to start of frags */ 235820bf50deSTom Herbert offset -= skb_headlen(skb); 235920bf50deSTom Herbert 236020bf50deSTom Herbert /* Find where we are in frag list */ 236120bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 236220bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 236320bf50deSTom Herbert 236420bf50deSTom Herbert if (offset < frag->size) 236520bf50deSTom Herbert break; 236620bf50deSTom Herbert 236720bf50deSTom Herbert offset -= frag->size; 236820bf50deSTom Herbert } 236920bf50deSTom Herbert 237020bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 237120bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 237220bf50deSTom Herbert 237320bf50deSTom Herbert slen = min_t(size_t, len, frag->size - offset); 237420bf50deSTom Herbert 237520bf50deSTom Herbert while (slen) { 237620bf50deSTom Herbert ret = kernel_sendpage_locked(sk, frag->page.p, 237720bf50deSTom Herbert frag->page_offset + offset, 237820bf50deSTom Herbert slen, MSG_DONTWAIT); 237920bf50deSTom Herbert if (ret <= 0) 238020bf50deSTom Herbert goto error; 238120bf50deSTom Herbert 238220bf50deSTom Herbert len -= ret; 238320bf50deSTom Herbert offset += ret; 238420bf50deSTom Herbert slen -= ret; 238520bf50deSTom Herbert } 238620bf50deSTom Herbert 238720bf50deSTom Herbert offset = 0; 238820bf50deSTom Herbert } 238920bf50deSTom Herbert 239020bf50deSTom Herbert if (len) { 239120bf50deSTom Herbert /* Process any frag lists */ 239220bf50deSTom Herbert 239320bf50deSTom Herbert if (skb == head) { 239420bf50deSTom Herbert if (skb_has_frag_list(skb)) { 239520bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 239620bf50deSTom Herbert goto do_frag_list; 239720bf50deSTom Herbert } 239820bf50deSTom Herbert } else if (skb->next) { 239920bf50deSTom Herbert skb = skb->next; 240020bf50deSTom Herbert goto do_frag_list; 240120bf50deSTom Herbert } 240220bf50deSTom Herbert } 240320bf50deSTom Herbert 240420bf50deSTom Herbert out: 240520bf50deSTom Herbert return orig_len - len; 240620bf50deSTom Herbert 240720bf50deSTom Herbert error: 240820bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 240920bf50deSTom Herbert } 241020bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 241120bf50deSTom Herbert 2412357b40a1SHerbert Xu /** 2413357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2414357b40a1SHerbert Xu * @skb: destination buffer 2415357b40a1SHerbert Xu * @offset: offset in destination 2416357b40a1SHerbert Xu * @from: source buffer 2417357b40a1SHerbert Xu * @len: number of bytes to copy 2418357b40a1SHerbert Xu * 2419357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2420357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2421357b40a1SHerbert Xu * traversing fragment lists and such. 2422357b40a1SHerbert Xu */ 2423357b40a1SHerbert Xu 24240c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2425357b40a1SHerbert Xu { 24261a028e50SDavid S. Miller int start = skb_headlen(skb); 2427fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2428fbb398a8SDavid S. Miller int i, copy; 2429357b40a1SHerbert Xu 2430357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2431357b40a1SHerbert Xu goto fault; 2432357b40a1SHerbert Xu 24331a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2434357b40a1SHerbert Xu if (copy > len) 2435357b40a1SHerbert Xu copy = len; 243627d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2437357b40a1SHerbert Xu if ((len -= copy) == 0) 2438357b40a1SHerbert Xu return 0; 2439357b40a1SHerbert Xu offset += copy; 2440357b40a1SHerbert Xu from += copy; 2441357b40a1SHerbert Xu } 2442357b40a1SHerbert Xu 2443357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2444357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 24451a028e50SDavid S. Miller int end; 2446357b40a1SHerbert Xu 2447547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24481a028e50SDavid S. Miller 24499e903e08SEric Dumazet end = start + skb_frag_size(frag); 2450357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2451c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2452c613c209SWillem de Bruijn struct page *p; 2453357b40a1SHerbert Xu u8 *vaddr; 2454357b40a1SHerbert Xu 2455357b40a1SHerbert Xu if (copy > len) 2456357b40a1SHerbert Xu copy = len; 2457357b40a1SHerbert Xu 2458c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2459c613c209SWillem de Bruijn frag->page_offset + offset - start, 2460c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2461c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2462c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 246351c56b00SEric Dumazet kunmap_atomic(vaddr); 2464c613c209SWillem de Bruijn } 2465357b40a1SHerbert Xu 2466357b40a1SHerbert Xu if ((len -= copy) == 0) 2467357b40a1SHerbert Xu return 0; 2468357b40a1SHerbert Xu offset += copy; 2469357b40a1SHerbert Xu from += copy; 2470357b40a1SHerbert Xu } 24711a028e50SDavid S. Miller start = end; 2472357b40a1SHerbert Xu } 2473357b40a1SHerbert Xu 2474fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 24751a028e50SDavid S. Miller int end; 2476357b40a1SHerbert Xu 2477547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24781a028e50SDavid S. Miller 2479fbb398a8SDavid S. Miller end = start + frag_iter->len; 2480357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2481357b40a1SHerbert Xu if (copy > len) 2482357b40a1SHerbert Xu copy = len; 2483fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 24841a028e50SDavid S. Miller from, copy)) 2485357b40a1SHerbert Xu goto fault; 2486357b40a1SHerbert Xu if ((len -= copy) == 0) 2487357b40a1SHerbert Xu return 0; 2488357b40a1SHerbert Xu offset += copy; 2489357b40a1SHerbert Xu from += copy; 2490357b40a1SHerbert Xu } 24911a028e50SDavid S. Miller start = end; 2492357b40a1SHerbert Xu } 2493357b40a1SHerbert Xu if (!len) 2494357b40a1SHerbert Xu return 0; 2495357b40a1SHerbert Xu 2496357b40a1SHerbert Xu fault: 2497357b40a1SHerbert Xu return -EFAULT; 2498357b40a1SHerbert Xu } 2499357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2500357b40a1SHerbert Xu 25011da177e4SLinus Torvalds /* Checksum skb data. */ 25022817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 25032817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 25041da177e4SLinus Torvalds { 25051a028e50SDavid S. Miller int start = skb_headlen(skb); 25061a028e50SDavid S. Miller int i, copy = start - offset; 2507fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 25081da177e4SLinus Torvalds int pos = 0; 25091da177e4SLinus Torvalds 25101da177e4SLinus Torvalds /* Checksum header. */ 25111da177e4SLinus Torvalds if (copy > 0) { 25121da177e4SLinus Torvalds if (copy > len) 25131da177e4SLinus Torvalds copy = len; 25142817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 25151da177e4SLinus Torvalds if ((len -= copy) == 0) 25161da177e4SLinus Torvalds return csum; 25171da177e4SLinus Torvalds offset += copy; 25181da177e4SLinus Torvalds pos = copy; 25191da177e4SLinus Torvalds } 25201da177e4SLinus Torvalds 25211da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 25221a028e50SDavid S. Miller int end; 252351c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 25241da177e4SLinus Torvalds 2525547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25261a028e50SDavid S. Miller 252751c56b00SEric Dumazet end = start + skb_frag_size(frag); 25281da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2529c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2530c613c209SWillem de Bruijn struct page *p; 253144bb9363SAl Viro __wsum csum2; 25321da177e4SLinus Torvalds u8 *vaddr; 25331da177e4SLinus Torvalds 25341da177e4SLinus Torvalds if (copy > len) 25351da177e4SLinus Torvalds copy = len; 2536c613c209SWillem de Bruijn 2537c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2538c613c209SWillem de Bruijn frag->page_offset + offset - start, 2539c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2540c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2541c613c209SWillem de Bruijn csum2 = ops->update(vaddr + p_off, p_len, 0); 254251c56b00SEric Dumazet kunmap_atomic(vaddr); 2543c613c209SWillem de Bruijn csum = ops->combine(csum, csum2, pos, p_len); 2544c613c209SWillem de Bruijn pos += p_len; 2545c613c209SWillem de Bruijn } 2546c613c209SWillem de Bruijn 25471da177e4SLinus Torvalds if (!(len -= copy)) 25481da177e4SLinus Torvalds return csum; 25491da177e4SLinus Torvalds offset += copy; 25501da177e4SLinus Torvalds } 25511a028e50SDavid S. Miller start = end; 25521da177e4SLinus Torvalds } 25531da177e4SLinus Torvalds 2554fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 25551a028e50SDavid S. Miller int end; 25561da177e4SLinus Torvalds 2557547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25581a028e50SDavid S. Miller 2559fbb398a8SDavid S. Miller end = start + frag_iter->len; 25601da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 25615f92a738SAl Viro __wsum csum2; 25621da177e4SLinus Torvalds if (copy > len) 25631da177e4SLinus Torvalds copy = len; 25642817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 25652817a336SDaniel Borkmann copy, 0, ops); 25662817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 25671da177e4SLinus Torvalds if ((len -= copy) == 0) 25681da177e4SLinus Torvalds return csum; 25691da177e4SLinus Torvalds offset += copy; 25701da177e4SLinus Torvalds pos += copy; 25711da177e4SLinus Torvalds } 25721a028e50SDavid S. Miller start = end; 25731da177e4SLinus Torvalds } 257409a62660SKris Katterjohn BUG_ON(len); 25751da177e4SLinus Torvalds 25761da177e4SLinus Torvalds return csum; 25771da177e4SLinus Torvalds } 25782817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 25792817a336SDaniel Borkmann 25802817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 25812817a336SDaniel Borkmann int len, __wsum csum) 25822817a336SDaniel Borkmann { 25832817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2584cea80ea8SDaniel Borkmann .update = csum_partial_ext, 25852817a336SDaniel Borkmann .combine = csum_block_add_ext, 25862817a336SDaniel Borkmann }; 25872817a336SDaniel Borkmann 25882817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 25892817a336SDaniel Borkmann } 2590b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 25911da177e4SLinus Torvalds 25921da177e4SLinus Torvalds /* Both of above in one bottle. */ 25931da177e4SLinus Torvalds 259481d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 259581d77662SAl Viro u8 *to, int len, __wsum csum) 25961da177e4SLinus Torvalds { 25971a028e50SDavid S. Miller int start = skb_headlen(skb); 25981a028e50SDavid S. Miller int i, copy = start - offset; 2599fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 26001da177e4SLinus Torvalds int pos = 0; 26011da177e4SLinus Torvalds 26021da177e4SLinus Torvalds /* Copy header. */ 26031da177e4SLinus Torvalds if (copy > 0) { 26041da177e4SLinus Torvalds if (copy > len) 26051da177e4SLinus Torvalds copy = len; 26061da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 26071da177e4SLinus Torvalds copy, csum); 26081da177e4SLinus Torvalds if ((len -= copy) == 0) 26091da177e4SLinus Torvalds return csum; 26101da177e4SLinus Torvalds offset += copy; 26111da177e4SLinus Torvalds to += copy; 26121da177e4SLinus Torvalds pos = copy; 26131da177e4SLinus Torvalds } 26141da177e4SLinus Torvalds 26151da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 26161a028e50SDavid S. Miller int end; 26171da177e4SLinus Torvalds 2618547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26191a028e50SDavid S. Miller 26209e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 26211da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2622c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2623c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2624c613c209SWillem de Bruijn struct page *p; 26255084205fSAl Viro __wsum csum2; 26261da177e4SLinus Torvalds u8 *vaddr; 26271da177e4SLinus Torvalds 26281da177e4SLinus Torvalds if (copy > len) 26291da177e4SLinus Torvalds copy = len; 2630c613c209SWillem de Bruijn 2631c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2632c613c209SWillem de Bruijn frag->page_offset + offset - start, 2633c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2634c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2635c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2636c613c209SWillem de Bruijn to + copied, 2637c613c209SWillem de Bruijn p_len, 0); 263851c56b00SEric Dumazet kunmap_atomic(vaddr); 26391da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2640c613c209SWillem de Bruijn pos += p_len; 2641c613c209SWillem de Bruijn } 2642c613c209SWillem de Bruijn 26431da177e4SLinus Torvalds if (!(len -= copy)) 26441da177e4SLinus Torvalds return csum; 26451da177e4SLinus Torvalds offset += copy; 26461da177e4SLinus Torvalds to += copy; 26471da177e4SLinus Torvalds } 26481a028e50SDavid S. Miller start = end; 26491da177e4SLinus Torvalds } 26501da177e4SLinus Torvalds 2651fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 265281d77662SAl Viro __wsum csum2; 26531a028e50SDavid S. Miller int end; 26541da177e4SLinus Torvalds 2655547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26561a028e50SDavid S. Miller 2657fbb398a8SDavid S. Miller end = start + frag_iter->len; 26581da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 26591da177e4SLinus Torvalds if (copy > len) 26601da177e4SLinus Torvalds copy = len; 2661fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 26621a028e50SDavid S. Miller offset - start, 26631da177e4SLinus Torvalds to, copy, 0); 26641da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 26651da177e4SLinus Torvalds if ((len -= copy) == 0) 26661da177e4SLinus Torvalds return csum; 26671da177e4SLinus Torvalds offset += copy; 26681da177e4SLinus Torvalds to += copy; 26691da177e4SLinus Torvalds pos += copy; 26701da177e4SLinus Torvalds } 26711a028e50SDavid S. Miller start = end; 26721da177e4SLinus Torvalds } 267309a62660SKris Katterjohn BUG_ON(len); 26741da177e4SLinus Torvalds return csum; 26751da177e4SLinus Torvalds } 2676b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 26771da177e4SLinus Torvalds 267849f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 267949f8e832SCong Wang { 268049f8e832SCong Wang __sum16 sum; 268149f8e832SCong Wang 268249f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 268314641931SCong Wang /* See comments in __skb_checksum_complete(). */ 268449f8e832SCong Wang if (likely(!sum)) { 268549f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 268649f8e832SCong Wang !skb->csum_complete_sw) 26877fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 268849f8e832SCong Wang } 268949f8e832SCong Wang if (!skb_shared(skb)) 269049f8e832SCong Wang skb->csum_valid = !sum; 269149f8e832SCong Wang return sum; 269249f8e832SCong Wang } 269349f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 269449f8e832SCong Wang 269514641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 269614641931SCong Wang * which has been changed from the hardware checksum, for example, by 269714641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 269814641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 269914641931SCong Wang * 270014641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 270114641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 270214641931SCong Wang * shared. 270314641931SCong Wang */ 270449f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 270549f8e832SCong Wang { 270649f8e832SCong Wang __wsum csum; 270749f8e832SCong Wang __sum16 sum; 270849f8e832SCong Wang 270949f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 271049f8e832SCong Wang 271149f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 271214641931SCong Wang /* This check is inverted, because we already knew the hardware 271314641931SCong Wang * checksum is invalid before calling this function. So, if the 271414641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 271514641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 271614641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 271714641931SCong Wang * when moving skb->data around. 271814641931SCong Wang */ 271949f8e832SCong Wang if (likely(!sum)) { 272049f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 272149f8e832SCong Wang !skb->csum_complete_sw) 27227fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 272349f8e832SCong Wang } 272449f8e832SCong Wang 272549f8e832SCong Wang if (!skb_shared(skb)) { 272649f8e832SCong Wang /* Save full packet checksum */ 272749f8e832SCong Wang skb->csum = csum; 272849f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 272949f8e832SCong Wang skb->csum_complete_sw = 1; 273049f8e832SCong Wang skb->csum_valid = !sum; 273149f8e832SCong Wang } 273249f8e832SCong Wang 273349f8e832SCong Wang return sum; 273449f8e832SCong Wang } 273549f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 273649f8e832SCong Wang 27379617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 27389617813dSDavide Caratti { 27399617813dSDavide Caratti net_warn_ratelimited( 27409617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 27419617813dSDavide Caratti __func__); 27429617813dSDavide Caratti return 0; 27439617813dSDavide Caratti } 27449617813dSDavide Caratti 27459617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 27469617813dSDavide Caratti int offset, int len) 27479617813dSDavide Caratti { 27489617813dSDavide Caratti net_warn_ratelimited( 27499617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 27509617813dSDavide Caratti __func__); 27519617813dSDavide Caratti return 0; 27529617813dSDavide Caratti } 27539617813dSDavide Caratti 27549617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 27559617813dSDavide Caratti .update = warn_crc32c_csum_update, 27569617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 27579617813dSDavide Caratti }; 27589617813dSDavide Caratti 27599617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 27609617813dSDavide Caratti &default_crc32c_ops; 27619617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 27629617813dSDavide Caratti 2763af2806f8SThomas Graf /** 2764af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2765af2806f8SThomas Graf * @from: source buffer 2766af2806f8SThomas Graf * 2767af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2768af2806f8SThomas Graf * into skb_zerocopy(). 2769af2806f8SThomas Graf */ 2770af2806f8SThomas Graf unsigned int 2771af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2772af2806f8SThomas Graf { 2773af2806f8SThomas Graf unsigned int hlen = 0; 2774af2806f8SThomas Graf 2775af2806f8SThomas Graf if (!from->head_frag || 2776af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2777af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2778af2806f8SThomas Graf hlen = skb_headlen(from); 2779af2806f8SThomas Graf 2780af2806f8SThomas Graf if (skb_has_frag_list(from)) 2781af2806f8SThomas Graf hlen = from->len; 2782af2806f8SThomas Graf 2783af2806f8SThomas Graf return hlen; 2784af2806f8SThomas Graf } 2785af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2786af2806f8SThomas Graf 2787af2806f8SThomas Graf /** 2788af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2789af2806f8SThomas Graf * @to: destination buffer 27907fceb4deSMasanari Iida * @from: source buffer 2791af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2792af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2793af2806f8SThomas Graf * 2794af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2795af2806f8SThomas Graf * to the frags in the source buffer. 2796af2806f8SThomas Graf * 2797af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2798af2806f8SThomas Graf * headroom in the `to` buffer. 279936d5fe6aSZoltan Kiss * 280036d5fe6aSZoltan Kiss * Return value: 280136d5fe6aSZoltan Kiss * 0: everything is OK 280236d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 280336d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2804af2806f8SThomas Graf */ 280536d5fe6aSZoltan Kiss int 280636d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2807af2806f8SThomas Graf { 2808af2806f8SThomas Graf int i, j = 0; 2809af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 281036d5fe6aSZoltan Kiss int ret; 2811af2806f8SThomas Graf struct page *page; 2812af2806f8SThomas Graf unsigned int offset; 2813af2806f8SThomas Graf 2814af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2815af2806f8SThomas Graf 2816af2806f8SThomas Graf /* dont bother with small payloads */ 281736d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 281836d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2819af2806f8SThomas Graf 2820af2806f8SThomas Graf if (hlen) { 282136d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 282236d5fe6aSZoltan Kiss if (unlikely(ret)) 282336d5fe6aSZoltan Kiss return ret; 2824af2806f8SThomas Graf len -= hlen; 2825af2806f8SThomas Graf } else { 2826af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2827af2806f8SThomas Graf if (plen) { 2828af2806f8SThomas Graf page = virt_to_head_page(from->head); 2829af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2830af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2831af2806f8SThomas Graf get_page(page); 2832af2806f8SThomas Graf j = 1; 2833af2806f8SThomas Graf len -= plen; 2834af2806f8SThomas Graf } 2835af2806f8SThomas Graf } 2836af2806f8SThomas Graf 2837af2806f8SThomas Graf to->truesize += len + plen; 2838af2806f8SThomas Graf to->len += len + plen; 2839af2806f8SThomas Graf to->data_len += len + plen; 2840af2806f8SThomas Graf 284136d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 284236d5fe6aSZoltan Kiss skb_tx_error(from); 284336d5fe6aSZoltan Kiss return -ENOMEM; 284436d5fe6aSZoltan Kiss } 28451f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 284636d5fe6aSZoltan Kiss 2847af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2848af2806f8SThomas Graf if (!len) 2849af2806f8SThomas Graf break; 2850af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2851af2806f8SThomas Graf skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2852af2806f8SThomas Graf len -= skb_shinfo(to)->frags[j].size; 2853af2806f8SThomas Graf skb_frag_ref(to, j); 2854af2806f8SThomas Graf j++; 2855af2806f8SThomas Graf } 2856af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 285736d5fe6aSZoltan Kiss 285836d5fe6aSZoltan Kiss return 0; 2859af2806f8SThomas Graf } 2860af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2861af2806f8SThomas Graf 28621da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 28631da177e4SLinus Torvalds { 2864d3bc23e7SAl Viro __wsum csum; 28651da177e4SLinus Torvalds long csstart; 28661da177e4SLinus Torvalds 286784fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 286855508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 28691da177e4SLinus Torvalds else 28701da177e4SLinus Torvalds csstart = skb_headlen(skb); 28711da177e4SLinus Torvalds 287209a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 28731da177e4SLinus Torvalds 2874d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 28751da177e4SLinus Torvalds 28761da177e4SLinus Torvalds csum = 0; 28771da177e4SLinus Torvalds if (csstart != skb->len) 28781da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 28791da177e4SLinus Torvalds skb->len - csstart, 0); 28801da177e4SLinus Torvalds 288184fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2882ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 28831da177e4SLinus Torvalds 2884d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 28851da177e4SLinus Torvalds } 28861da177e4SLinus Torvalds } 2887b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 28881da177e4SLinus Torvalds 28891da177e4SLinus Torvalds /** 28901da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 28911da177e4SLinus Torvalds * @list: list to dequeue from 28921da177e4SLinus Torvalds * 28931da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 28941da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 28951da177e4SLinus Torvalds * returned or %NULL if the list is empty. 28961da177e4SLinus Torvalds */ 28971da177e4SLinus Torvalds 28981da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 28991da177e4SLinus Torvalds { 29001da177e4SLinus Torvalds unsigned long flags; 29011da177e4SLinus Torvalds struct sk_buff *result; 29021da177e4SLinus Torvalds 29031da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 29041da177e4SLinus Torvalds result = __skb_dequeue(list); 29051da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 29061da177e4SLinus Torvalds return result; 29071da177e4SLinus Torvalds } 2908b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 29091da177e4SLinus Torvalds 29101da177e4SLinus Torvalds /** 29111da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 29121da177e4SLinus Torvalds * @list: list to dequeue from 29131da177e4SLinus Torvalds * 29141da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 29151da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 29161da177e4SLinus Torvalds * returned or %NULL if the list is empty. 29171da177e4SLinus Torvalds */ 29181da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 29191da177e4SLinus Torvalds { 29201da177e4SLinus Torvalds unsigned long flags; 29211da177e4SLinus Torvalds struct sk_buff *result; 29221da177e4SLinus Torvalds 29231da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 29241da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 29251da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 29261da177e4SLinus Torvalds return result; 29271da177e4SLinus Torvalds } 2928b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 29291da177e4SLinus Torvalds 29301da177e4SLinus Torvalds /** 29311da177e4SLinus Torvalds * skb_queue_purge - empty a list 29321da177e4SLinus Torvalds * @list: list to empty 29331da177e4SLinus Torvalds * 29341da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 29351da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 29361da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 29371da177e4SLinus Torvalds */ 29381da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 29391da177e4SLinus Torvalds { 29401da177e4SLinus Torvalds struct sk_buff *skb; 29411da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 29421da177e4SLinus Torvalds kfree_skb(skb); 29431da177e4SLinus Torvalds } 2944b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 29451da177e4SLinus Torvalds 29461da177e4SLinus Torvalds /** 29479f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 29489f5afeaeSYaogong Wang * @root: root of the rbtree to empty 2949385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 29509f5afeaeSYaogong Wang * 29519f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 29529f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 29539f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 29549f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 29559f5afeaeSYaogong Wang */ 2956385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 29579f5afeaeSYaogong Wang { 29587c90584cSEric Dumazet struct rb_node *p = rb_first(root); 2959385114deSPeter Oskolkov unsigned int sum = 0; 29609f5afeaeSYaogong Wang 29617c90584cSEric Dumazet while (p) { 29627c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 29637c90584cSEric Dumazet 29647c90584cSEric Dumazet p = rb_next(p); 29657c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 2966385114deSPeter Oskolkov sum += skb->truesize; 29679f5afeaeSYaogong Wang kfree_skb(skb); 29687c90584cSEric Dumazet } 2969385114deSPeter Oskolkov return sum; 29709f5afeaeSYaogong Wang } 29719f5afeaeSYaogong Wang 29729f5afeaeSYaogong Wang /** 29731da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 29741da177e4SLinus Torvalds * @list: list to use 29751da177e4SLinus Torvalds * @newsk: buffer to queue 29761da177e4SLinus Torvalds * 29771da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 29781da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 29791da177e4SLinus Torvalds * safely. 29801da177e4SLinus Torvalds * 29811da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 29821da177e4SLinus Torvalds */ 29831da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 29841da177e4SLinus Torvalds { 29851da177e4SLinus Torvalds unsigned long flags; 29861da177e4SLinus Torvalds 29871da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 29881da177e4SLinus Torvalds __skb_queue_head(list, newsk); 29891da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 29901da177e4SLinus Torvalds } 2991b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 29921da177e4SLinus Torvalds 29931da177e4SLinus Torvalds /** 29941da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 29951da177e4SLinus Torvalds * @list: list to use 29961da177e4SLinus Torvalds * @newsk: buffer to queue 29971da177e4SLinus Torvalds * 29981da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 29991da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 30001da177e4SLinus Torvalds * safely. 30011da177e4SLinus Torvalds * 30021da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 30031da177e4SLinus Torvalds */ 30041da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 30051da177e4SLinus Torvalds { 30061da177e4SLinus Torvalds unsigned long flags; 30071da177e4SLinus Torvalds 30081da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30091da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 30101da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30111da177e4SLinus Torvalds } 3012b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 30138728b834SDavid S. Miller 30141da177e4SLinus Torvalds /** 30151da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 30161da177e4SLinus Torvalds * @skb: buffer to remove 30178728b834SDavid S. Miller * @list: list to use 30181da177e4SLinus Torvalds * 30198728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 30208728b834SDavid S. Miller * function is atomic with respect to other list locked calls 30211da177e4SLinus Torvalds * 30228728b834SDavid S. Miller * You must know what list the SKB is on. 30231da177e4SLinus Torvalds */ 30248728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 30251da177e4SLinus Torvalds { 30261da177e4SLinus Torvalds unsigned long flags; 30271da177e4SLinus Torvalds 30281da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30298728b834SDavid S. Miller __skb_unlink(skb, list); 30301da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30311da177e4SLinus Torvalds } 3032b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 30331da177e4SLinus Torvalds 30341da177e4SLinus Torvalds /** 30351da177e4SLinus Torvalds * skb_append - append a buffer 30361da177e4SLinus Torvalds * @old: buffer to insert after 30371da177e4SLinus Torvalds * @newsk: buffer to insert 30388728b834SDavid S. Miller * @list: list to use 30391da177e4SLinus Torvalds * 30401da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 30411da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 30421da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 30431da177e4SLinus Torvalds */ 30448728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 30451da177e4SLinus Torvalds { 30461da177e4SLinus Torvalds unsigned long flags; 30471da177e4SLinus Torvalds 30488728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 30497de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 30508728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 30511da177e4SLinus Torvalds } 3052b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 30531da177e4SLinus Torvalds 30541da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 30551da177e4SLinus Torvalds struct sk_buff* skb1, 30561da177e4SLinus Torvalds const u32 len, const int pos) 30571da177e4SLinus Torvalds { 30581da177e4SLinus Torvalds int i; 30591da177e4SLinus Torvalds 3060d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3061d626f62bSArnaldo Carvalho de Melo pos - len); 30621da177e4SLinus Torvalds /* And move data appendix as is. */ 30631da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 30641da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 30651da177e4SLinus Torvalds 30661da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 30671da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 30681da177e4SLinus Torvalds skb1->data_len = skb->data_len; 30691da177e4SLinus Torvalds skb1->len += skb1->data_len; 30701da177e4SLinus Torvalds skb->data_len = 0; 30711da177e4SLinus Torvalds skb->len = len; 307227a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 30731da177e4SLinus Torvalds } 30741da177e4SLinus Torvalds 30751da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 30761da177e4SLinus Torvalds struct sk_buff* skb1, 30771da177e4SLinus Torvalds const u32 len, int pos) 30781da177e4SLinus Torvalds { 30791da177e4SLinus Torvalds int i, k = 0; 30801da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 30811da177e4SLinus Torvalds 30821da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 30831da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 30841da177e4SLinus Torvalds skb->len = len; 30851da177e4SLinus Torvalds skb->data_len = len - pos; 30861da177e4SLinus Torvalds 30871da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 30889e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 30891da177e4SLinus Torvalds 30901da177e4SLinus Torvalds if (pos + size > len) { 30911da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 30921da177e4SLinus Torvalds 30931da177e4SLinus Torvalds if (pos < len) { 30941da177e4SLinus Torvalds /* Split frag. 30951da177e4SLinus Torvalds * We have two variants in this case: 30961da177e4SLinus Torvalds * 1. Move all the frag to the second 30971da177e4SLinus Torvalds * part, if it is possible. F.e. 30981da177e4SLinus Torvalds * this approach is mandatory for TUX, 30991da177e4SLinus Torvalds * where splitting is expensive. 31001da177e4SLinus Torvalds * 2. Split is accurately. We make this. 31011da177e4SLinus Torvalds */ 3102ea2ab693SIan Campbell skb_frag_ref(skb, i); 31031da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 31049e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 31059e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 31061da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 31071da177e4SLinus Torvalds } 31081da177e4SLinus Torvalds k++; 31091da177e4SLinus Torvalds } else 31101da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 31111da177e4SLinus Torvalds pos += size; 31121da177e4SLinus Torvalds } 31131da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 31141da177e4SLinus Torvalds } 31151da177e4SLinus Torvalds 31161da177e4SLinus Torvalds /** 31171da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 31181da177e4SLinus Torvalds * @skb: the buffer to split 31191da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 31201da177e4SLinus Torvalds * @len: new length for skb 31211da177e4SLinus Torvalds */ 31221da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 31231da177e4SLinus Torvalds { 31241da177e4SLinus Torvalds int pos = skb_headlen(skb); 31251da177e4SLinus Torvalds 3126fff88030SWillem de Bruijn skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags & 3127fff88030SWillem de Bruijn SKBTX_SHARED_FRAG; 31281f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 31291da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 31301da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 31311da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 31321da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 31331da177e4SLinus Torvalds } 3134b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 31351da177e4SLinus Torvalds 31369f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 31379f782db3SIlpo Järvinen * 31389f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 31399f782db3SIlpo Järvinen */ 3140832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3141832d11c5SIlpo Järvinen { 31420ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3143832d11c5SIlpo Järvinen } 3144832d11c5SIlpo Järvinen 3145832d11c5SIlpo Järvinen /** 3146832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3147832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3148832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3149832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3150832d11c5SIlpo Järvinen * 3151832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 315220e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3153832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3154832d11c5SIlpo Järvinen * 3155832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3156832d11c5SIlpo Järvinen * 3157832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3158832d11c5SIlpo Järvinen * to have non-paged data as well. 3159832d11c5SIlpo Järvinen * 3160832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3161832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3162832d11c5SIlpo Järvinen */ 3163832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3164832d11c5SIlpo Järvinen { 3165832d11c5SIlpo Järvinen int from, to, merge, todo; 3166832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 3167832d11c5SIlpo Järvinen 3168832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3169f8071cdeSEric Dumazet 3170f8071cdeSEric Dumazet if (skb_headlen(skb)) 3171f8071cdeSEric Dumazet return 0; 31721f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 31731f8b977aSWillem de Bruijn return 0; 3174832d11c5SIlpo Järvinen 3175832d11c5SIlpo Järvinen todo = shiftlen; 3176832d11c5SIlpo Järvinen from = 0; 3177832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3178832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3179832d11c5SIlpo Järvinen 3180832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3181832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3182832d11c5SIlpo Järvinen */ 3183832d11c5SIlpo Järvinen if (!to || 3184ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3185ea2ab693SIan Campbell fragfrom->page_offset)) { 3186832d11c5SIlpo Järvinen merge = -1; 3187832d11c5SIlpo Järvinen } else { 3188832d11c5SIlpo Järvinen merge = to - 1; 3189832d11c5SIlpo Järvinen 31909e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3191832d11c5SIlpo Järvinen if (todo < 0) { 3192832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3193832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3194832d11c5SIlpo Järvinen return 0; 3195832d11c5SIlpo Järvinen 31969f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 31979f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3198832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3199832d11c5SIlpo Järvinen 32009e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 32019e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3202832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 3203832d11c5SIlpo Järvinen 3204832d11c5SIlpo Järvinen goto onlymerged; 3205832d11c5SIlpo Järvinen } 3206832d11c5SIlpo Järvinen 3207832d11c5SIlpo Järvinen from++; 3208832d11c5SIlpo Järvinen } 3209832d11c5SIlpo Järvinen 3210832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3211832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3212832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3213832d11c5SIlpo Järvinen return 0; 3214832d11c5SIlpo Järvinen 3215832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3216832d11c5SIlpo Järvinen return 0; 3217832d11c5SIlpo Järvinen 3218832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3219832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3220832d11c5SIlpo Järvinen return 0; 3221832d11c5SIlpo Järvinen 3222832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3223832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3224832d11c5SIlpo Järvinen 32259e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3226832d11c5SIlpo Järvinen *fragto = *fragfrom; 32279e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3228832d11c5SIlpo Järvinen from++; 3229832d11c5SIlpo Järvinen to++; 3230832d11c5SIlpo Järvinen 3231832d11c5SIlpo Järvinen } else { 3232ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3233832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 3234832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 32359e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3236832d11c5SIlpo Järvinen 3237832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 32389e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3239832d11c5SIlpo Järvinen todo = 0; 3240832d11c5SIlpo Järvinen 3241832d11c5SIlpo Järvinen to++; 3242832d11c5SIlpo Järvinen break; 3243832d11c5SIlpo Järvinen } 3244832d11c5SIlpo Järvinen } 3245832d11c5SIlpo Järvinen 3246832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3247832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3248832d11c5SIlpo Järvinen 3249832d11c5SIlpo Järvinen if (merge >= 0) { 3250832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3251832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3252832d11c5SIlpo Järvinen 32539e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 3254ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 3255832d11c5SIlpo Järvinen } 3256832d11c5SIlpo Järvinen 3257832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3258832d11c5SIlpo Järvinen to = 0; 3259832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3260832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3261832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3262832d11c5SIlpo Järvinen 3263832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3264832d11c5SIlpo Järvinen 3265832d11c5SIlpo Järvinen onlymerged: 3266832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3267832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3268832d11c5SIlpo Järvinen */ 3269832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3270832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3271832d11c5SIlpo Järvinen 3272832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3273832d11c5SIlpo Järvinen skb->len -= shiftlen; 3274832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3275832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3276832d11c5SIlpo Järvinen tgt->len += shiftlen; 3277832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3278832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3279832d11c5SIlpo Järvinen 3280832d11c5SIlpo Järvinen return shiftlen; 3281832d11c5SIlpo Järvinen } 3282832d11c5SIlpo Järvinen 3283677e90edSThomas Graf /** 3284677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3285677e90edSThomas Graf * @skb: the buffer to read 3286677e90edSThomas Graf * @from: lower offset of data to be read 3287677e90edSThomas Graf * @to: upper offset of data to be read 3288677e90edSThomas Graf * @st: state variable 3289677e90edSThomas Graf * 3290677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3291677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3292677e90edSThomas Graf */ 3293677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3294677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3295677e90edSThomas Graf { 3296677e90edSThomas Graf st->lower_offset = from; 3297677e90edSThomas Graf st->upper_offset = to; 3298677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3299677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3300677e90edSThomas Graf st->frag_data = NULL; 3301677e90edSThomas Graf } 3302b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3303677e90edSThomas Graf 3304677e90edSThomas Graf /** 3305677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3306677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3307677e90edSThomas Graf * @data: destination pointer for data to be returned 3308677e90edSThomas Graf * @st: state variable 3309677e90edSThomas Graf * 3310bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3311677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3312bc32383cSMathias Krause * the head of the data block to @data and returns the length 3313677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3314677e90edSThomas Graf * offset has been reached. 3315677e90edSThomas Graf * 3316677e90edSThomas Graf * The caller is not required to consume all of the data 3317bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3318677e90edSThomas Graf * of bytes already consumed and the next call to 3319677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3320677e90edSThomas Graf * 332125985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3322e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3323677e90edSThomas Graf * reads of potentially non linear data. 3324677e90edSThomas Graf * 3325bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3326677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3327677e90edSThomas Graf * a stack for this purpose. 3328677e90edSThomas Graf */ 3329677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3330677e90edSThomas Graf struct skb_seq_state *st) 3331677e90edSThomas Graf { 3332677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3333677e90edSThomas Graf skb_frag_t *frag; 3334677e90edSThomas Graf 3335aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3336aeb193eaSWedson Almeida Filho if (st->frag_data) { 3337aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3338aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3339aeb193eaSWedson Almeida Filho } 3340677e90edSThomas Graf return 0; 3341aeb193eaSWedson Almeida Filho } 3342677e90edSThomas Graf 3343677e90edSThomas Graf next_skb: 334495e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3345677e90edSThomas Graf 3346995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 334795e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3348677e90edSThomas Graf return block_limit - abs_offset; 3349677e90edSThomas Graf } 3350677e90edSThomas Graf 3351677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3352677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3353677e90edSThomas Graf 3354677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 3355677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 33569e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 3357677e90edSThomas Graf 3358677e90edSThomas Graf if (abs_offset < block_limit) { 3359677e90edSThomas Graf if (!st->frag_data) 336051c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 3361677e90edSThomas Graf 3362677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 3363677e90edSThomas Graf (abs_offset - st->stepped_offset); 3364677e90edSThomas Graf 3365677e90edSThomas Graf return block_limit - abs_offset; 3366677e90edSThomas Graf } 3367677e90edSThomas Graf 3368677e90edSThomas Graf if (st->frag_data) { 336951c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3370677e90edSThomas Graf st->frag_data = NULL; 3371677e90edSThomas Graf } 3372677e90edSThomas Graf 3373677e90edSThomas Graf st->frag_idx++; 33749e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 3375677e90edSThomas Graf } 3376677e90edSThomas Graf 33775b5a60daSOlaf Kirch if (st->frag_data) { 337851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 33795b5a60daSOlaf Kirch st->frag_data = NULL; 33805b5a60daSOlaf Kirch } 33815b5a60daSOlaf Kirch 338221dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3383677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 338495e3b24cSHerbert Xu st->frag_idx = 0; 3385677e90edSThomas Graf goto next_skb; 338671b3346dSShyam Iyer } else if (st->cur_skb->next) { 338771b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 338871b3346dSShyam Iyer st->frag_idx = 0; 3389677e90edSThomas Graf goto next_skb; 3390677e90edSThomas Graf } 3391677e90edSThomas Graf 3392677e90edSThomas Graf return 0; 3393677e90edSThomas Graf } 3394b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3395677e90edSThomas Graf 3396677e90edSThomas Graf /** 3397677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3398677e90edSThomas Graf * @st: state variable 3399677e90edSThomas Graf * 3400677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3401677e90edSThomas Graf * returned 0. 3402677e90edSThomas Graf */ 3403677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3404677e90edSThomas Graf { 3405677e90edSThomas Graf if (st->frag_data) 340651c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3407677e90edSThomas Graf } 3408b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3409677e90edSThomas Graf 34103fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 34113fc7e8a6SThomas Graf 34123fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 34133fc7e8a6SThomas Graf struct ts_config *conf, 34143fc7e8a6SThomas Graf struct ts_state *state) 34153fc7e8a6SThomas Graf { 34163fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 34173fc7e8a6SThomas Graf } 34183fc7e8a6SThomas Graf 34193fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 34203fc7e8a6SThomas Graf { 34213fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 34223fc7e8a6SThomas Graf } 34233fc7e8a6SThomas Graf 34243fc7e8a6SThomas Graf /** 34253fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 34263fc7e8a6SThomas Graf * @skb: the buffer to look in 34273fc7e8a6SThomas Graf * @from: search offset 34283fc7e8a6SThomas Graf * @to: search limit 34293fc7e8a6SThomas Graf * @config: textsearch configuration 34303fc7e8a6SThomas Graf * 34313fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 34323fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 34333fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 34343fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 34353fc7e8a6SThomas Graf */ 34363fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3437059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 34383fc7e8a6SThomas Graf { 3439059a2440SBojan Prtvar struct ts_state state; 3440f72b948dSPhil Oester unsigned int ret; 3441f72b948dSPhil Oester 34423fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 34433fc7e8a6SThomas Graf config->finish = skb_ts_finish; 34443fc7e8a6SThomas Graf 3445059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 34463fc7e8a6SThomas Graf 3447059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3448f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 34493fc7e8a6SThomas Graf } 3450b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 34513fc7e8a6SThomas Graf 3452be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3453be12a1feSHannes Frederic Sowa int offset, size_t size) 3454be12a1feSHannes Frederic Sowa { 3455be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3456be12a1feSHannes Frederic Sowa 3457be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3458be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3459be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3460be12a1feSHannes Frederic Sowa get_page(page); 3461be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3462be12a1feSHannes Frederic Sowa } else { 3463be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3464be12a1feSHannes Frederic Sowa } 3465be12a1feSHannes Frederic Sowa 3466be12a1feSHannes Frederic Sowa return 0; 3467be12a1feSHannes Frederic Sowa } 3468be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3469be12a1feSHannes Frederic Sowa 3470cbb042f9SHerbert Xu /** 3471cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3472cbb042f9SHerbert Xu * @skb: buffer to update 3473cbb042f9SHerbert Xu * @len: length of data pulled 3474cbb042f9SHerbert Xu * 3475cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3476fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 347784fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 347884fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 347984fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3480cbb042f9SHerbert Xu */ 3481af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3482cbb042f9SHerbert Xu { 348331b33dfbSPravin B Shelar unsigned char *data = skb->data; 348431b33dfbSPravin B Shelar 3485cbb042f9SHerbert Xu BUG_ON(len > skb->len); 348631b33dfbSPravin B Shelar __skb_pull(skb, len); 348731b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 348831b33dfbSPravin B Shelar return skb->data; 3489cbb042f9SHerbert Xu } 3490f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3491f94691acSArnaldo Carvalho de Melo 349213acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 349313acc94eSYonghong Song { 349413acc94eSYonghong Song skb_frag_t head_frag; 349513acc94eSYonghong Song struct page *page; 349613acc94eSYonghong Song 349713acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 349813acc94eSYonghong Song head_frag.page.p = page; 349913acc94eSYonghong Song head_frag.page_offset = frag_skb->data - 350013acc94eSYonghong Song (unsigned char *)page_address(page); 350113acc94eSYonghong Song head_frag.size = skb_headlen(frag_skb); 350213acc94eSYonghong Song return head_frag; 350313acc94eSYonghong Song } 350413acc94eSYonghong Song 3505f4c50d99SHerbert Xu /** 3506f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3507df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3508576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3509f4c50d99SHerbert Xu * 3510f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 35114c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 35124c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3513f4c50d99SHerbert Xu */ 3514df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3515df5771ffSMichael S. Tsirkin netdev_features_t features) 3516f4c50d99SHerbert Xu { 3517f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3518f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 35191a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3520df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3521df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3522df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 35231fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3524f4c50d99SHerbert Xu unsigned int offset = doffset; 3525df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3526802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3527f4c50d99SHerbert Xu unsigned int headroom; 3528802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3529ec5f0615SPravin B Shelar __be16 proto; 353036c98382SAlexander Duyck bool csum, sg; 3531df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3532f4c50d99SHerbert Xu int err = -ENOMEM; 3533f4c50d99SHerbert Xu int i = 0; 3534f4c50d99SHerbert Xu int pos; 353553d6471cSVlad Yasevich int dummy; 3536f4c50d99SHerbert Xu 35375882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 353853d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 3539ec5f0615SPravin B Shelar if (unlikely(!proto)) 3540ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3541ec5f0615SPravin B Shelar 354236c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3543f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 35447e2b10c1STom Herbert 354507b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 354607b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 354707b26c94SSteffen Klassert struct sk_buff *iter; 354843170c4eSIlan Tayari unsigned int frag_len; 354907b26c94SSteffen Klassert 355007b26c94SSteffen Klassert if (!list_skb || 355107b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 355207b26c94SSteffen Klassert goto normal; 355307b26c94SSteffen Klassert 355443170c4eSIlan Tayari /* If we get here then all the required 355543170c4eSIlan Tayari * GSO features except frag_list are supported. 355643170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 355743170c4eSIlan Tayari * with no frag_list. 355843170c4eSIlan Tayari * Currently we can do that only when the buffers don't 355943170c4eSIlan Tayari * have a linear part and all the buffers except 356043170c4eSIlan Tayari * the last are of the same length. 356107b26c94SSteffen Klassert */ 356243170c4eSIlan Tayari frag_len = list_skb->len; 356307b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 356443170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 356543170c4eSIlan Tayari goto normal; 3566eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 356707b26c94SSteffen Klassert goto normal; 356807b26c94SSteffen Klassert 356907b26c94SSteffen Klassert len -= iter->len; 357007b26c94SSteffen Klassert } 357143170c4eSIlan Tayari 357243170c4eSIlan Tayari if (len != frag_len) 357343170c4eSIlan Tayari goto normal; 357407b26c94SSteffen Klassert } 357507b26c94SSteffen Klassert 3576802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 3577802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 3578802ab55aSAlexander Duyck * now. 3579802ab55aSAlexander Duyck */ 3580802ab55aSAlexander Duyck partial_segs = len / mss; 3581d7fb5a80SAlexander Duyck if (partial_segs > 1) 3582802ab55aSAlexander Duyck mss *= partial_segs; 3583d7fb5a80SAlexander Duyck else 3584d7fb5a80SAlexander Duyck partial_segs = 0; 3585802ab55aSAlexander Duyck } 3586802ab55aSAlexander Duyck 358707b26c94SSteffen Klassert normal: 3588df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3589df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3590f4c50d99SHerbert Xu 3591f4c50d99SHerbert Xu do { 3592f4c50d99SHerbert Xu struct sk_buff *nskb; 35938cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3594c8884eddSHerbert Xu int hsize; 3595f4c50d99SHerbert Xu int size; 3596f4c50d99SHerbert Xu 35973953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 35983953c46cSMarcelo Ricardo Leitner len = list_skb->len; 35993953c46cSMarcelo Ricardo Leitner } else { 3600df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3601f4c50d99SHerbert Xu if (len > mss) 3602f4c50d99SHerbert Xu len = mss; 36033953c46cSMarcelo Ricardo Leitner } 3604f4c50d99SHerbert Xu 3605df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3606f4c50d99SHerbert Xu if (hsize < 0) 3607f4c50d99SHerbert Xu hsize = 0; 3608c8884eddSHerbert Xu if (hsize > len || !sg) 3609c8884eddSHerbert Xu hsize = len; 3610f4c50d99SHerbert Xu 36111a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 36121a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 36131a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 361489319d38SHerbert Xu 36159d8506ccSHerbert Xu i = 0; 36161a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 36171a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 36181fd819ecSMichael S. Tsirkin frag_skb = list_skb; 36191a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 36209d8506ccSHerbert Xu 36219d8506ccSHerbert Xu while (pos < offset + len) { 36229d8506ccSHerbert Xu BUG_ON(i >= nfrags); 36239d8506ccSHerbert Xu 36244e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 36259d8506ccSHerbert Xu if (pos + size > offset + len) 36269d8506ccSHerbert Xu break; 36279d8506ccSHerbert Xu 36289d8506ccSHerbert Xu i++; 36299d8506ccSHerbert Xu pos += size; 36304e1beba1SMichael S. Tsirkin frag++; 36319d8506ccSHerbert Xu } 36329d8506ccSHerbert Xu 36331a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 36341a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 363589319d38SHerbert Xu 3636f4c50d99SHerbert Xu if (unlikely(!nskb)) 3637f4c50d99SHerbert Xu goto err; 3638f4c50d99SHerbert Xu 36399d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 36409d8506ccSHerbert Xu kfree_skb(nskb); 36419d8506ccSHerbert Xu goto err; 36429d8506ccSHerbert Xu } 36439d8506ccSHerbert Xu 3644ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 364589319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 364689319d38SHerbert Xu kfree_skb(nskb); 364789319d38SHerbert Xu goto err; 364889319d38SHerbert Xu } 364989319d38SHerbert Xu 3650ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 365189319d38SHerbert Xu skb_release_head_state(nskb); 365289319d38SHerbert Xu __skb_push(nskb, doffset); 365389319d38SHerbert Xu } else { 3654c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3655df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3656c93bdd0eSMel Gorman NUMA_NO_NODE); 365789319d38SHerbert Xu 365889319d38SHerbert Xu if (unlikely(!nskb)) 365989319d38SHerbert Xu goto err; 366089319d38SHerbert Xu 366189319d38SHerbert Xu skb_reserve(nskb, headroom); 366289319d38SHerbert Xu __skb_put(nskb, doffset); 366389319d38SHerbert Xu } 366489319d38SHerbert Xu 3665f4c50d99SHerbert Xu if (segs) 3666f4c50d99SHerbert Xu tail->next = nskb; 3667f4c50d99SHerbert Xu else 3668f4c50d99SHerbert Xu segs = nskb; 3669f4c50d99SHerbert Xu tail = nskb; 3670f4c50d99SHerbert Xu 3671df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3672f4c50d99SHerbert Xu 3673030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3674fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 367568c33163SPravin B Shelar 3676df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 367768c33163SPravin B Shelar nskb->data - tnl_hlen, 367868c33163SPravin B Shelar doffset + tnl_hlen); 367989319d38SHerbert Xu 36809d8506ccSHerbert Xu if (nskb->len == len + doffset) 36811cdbcb79SSimon Horman goto perform_csum_check; 368289319d38SHerbert Xu 36837fbeffedSAlexander Duyck if (!sg) { 36847fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 36856f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 368676443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 368776443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 3688f4c50d99SHerbert Xu skb_put(nskb, len), 3689f4c50d99SHerbert Xu len, 0); 36907e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3691de843723STom Herbert skb_headroom(nskb) + doffset; 3692f4c50d99SHerbert Xu continue; 3693f4c50d99SHerbert Xu } 3694f4c50d99SHerbert Xu 36958cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3696f4c50d99SHerbert Xu 3697df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3698d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3699f4c50d99SHerbert Xu 3700fff88030SWillem de Bruijn skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags & 3701df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3702cef401deSEric Dumazet 3703bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 3704bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 3705bf5c25d6SWillem de Bruijn goto err; 3706bf5c25d6SWillem de Bruijn 37079d8506ccSHerbert Xu while (pos < offset + len) { 37089d8506ccSHerbert Xu if (i >= nfrags) { 37099d8506ccSHerbert Xu i = 0; 37101a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 37111a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 37121fd819ecSMichael S. Tsirkin frag_skb = list_skb; 371313acc94eSYonghong Song if (!skb_headlen(list_skb)) { 37149d8506ccSHerbert Xu BUG_ON(!nfrags); 371513acc94eSYonghong Song } else { 371613acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 37179d8506ccSHerbert Xu 371813acc94eSYonghong Song /* to make room for head_frag. */ 371913acc94eSYonghong Song i--; 372013acc94eSYonghong Song frag--; 372113acc94eSYonghong Song } 3722bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 3723bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 3724bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 3725bf5c25d6SWillem de Bruijn goto err; 3726bf5c25d6SWillem de Bruijn 37271a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 37289d8506ccSHerbert Xu } 37299d8506ccSHerbert Xu 37309d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 37319d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 37329d8506ccSHerbert Xu net_warn_ratelimited( 37339d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 37349d8506ccSHerbert Xu pos, mss); 3735ff907a11SEric Dumazet err = -EINVAL; 37369d8506ccSHerbert Xu goto err; 37379d8506ccSHerbert Xu } 37389d8506ccSHerbert Xu 373913acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 37408cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 37418cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3742f4c50d99SHerbert Xu 3743f4c50d99SHerbert Xu if (pos < offset) { 37448cb19905SMichael S. Tsirkin nskb_frag->page_offset += offset - pos; 37458cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3746f4c50d99SHerbert Xu } 3747f4c50d99SHerbert Xu 374889319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3749f4c50d99SHerbert Xu 3750f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3751f4c50d99SHerbert Xu i++; 37524e1beba1SMichael S. Tsirkin frag++; 3753f4c50d99SHerbert Xu pos += size; 3754f4c50d99SHerbert Xu } else { 37558cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 375689319d38SHerbert Xu goto skip_fraglist; 3757f4c50d99SHerbert Xu } 3758f4c50d99SHerbert Xu 37598cb19905SMichael S. Tsirkin nskb_frag++; 3760f4c50d99SHerbert Xu } 3761f4c50d99SHerbert Xu 376289319d38SHerbert Xu skip_fraglist: 3763f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3764f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3765f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3766ec5f0615SPravin B Shelar 37671cdbcb79SSimon Horman perform_csum_check: 37687fbeffedSAlexander Duyck if (!csum) { 3769ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 3770ff907a11SEric Dumazet __skb_linearize(nskb)) 3771ddff00d4SAlexander Duyck goto err; 3772ff907a11SEric Dumazet 37737fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 3774ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 377576443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 377676443456SAlexander Duyck skb_checksum(nskb, doffset, 377776443456SAlexander Duyck nskb->len - doffset, 0); 37787e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 37797e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3780ec5f0615SPravin B Shelar } 3781df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3782f4c50d99SHerbert Xu 3783bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3784bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3785bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3786bec3cfdcSEric Dumazet */ 3787bec3cfdcSEric Dumazet segs->prev = tail; 3788432c856fSToshiaki Makita 3789802ab55aSAlexander Duyck if (partial_segs) { 379007b26c94SSteffen Klassert struct sk_buff *iter; 3791802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 379207b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 3793802ab55aSAlexander Duyck 3794802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 379507b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 3796802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 3797802ab55aSAlexander Duyck 3798802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 3799802ab55aSAlexander Duyck * our way back down the stack of protocols. 3800802ab55aSAlexander Duyck */ 380107b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 380207b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 380307b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 380407b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 380507b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 380607b26c94SSteffen Klassert } 380707b26c94SSteffen Klassert 380807b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 380907b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 381007b26c94SSteffen Klassert else if (tail != segs) 381107b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 3812802ab55aSAlexander Duyck } 3813802ab55aSAlexander Duyck 3814432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 3815432c856fSToshiaki Makita * using skb_set_owner_w(). 3816432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 3817432c856fSToshiaki Makita */ 3818432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 3819432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 3820432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 3821432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 3822432c856fSToshiaki Makita } 3823f4c50d99SHerbert Xu return segs; 3824f4c50d99SHerbert Xu 3825f4c50d99SHerbert Xu err: 3826289dccbeSEric Dumazet kfree_skb_list(segs); 3827f4c50d99SHerbert Xu return ERR_PTR(err); 3828f4c50d99SHerbert Xu } 3829f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3830f4c50d99SHerbert Xu 3831d4546c25SDavid Miller int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb) 383271d93b39SHerbert Xu { 38338a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 383467147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 383567147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 38368a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 3837715dc1f3SEric Dumazet unsigned int delta_truesize; 3838d4546c25SDavid Miller struct sk_buff *lp; 383971d93b39SHerbert Xu 38400ab03f35SSteffen Klassert if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush)) 384171d93b39SHerbert Xu return -E2BIG; 384271d93b39SHerbert Xu 384329e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 38448a29111cSEric Dumazet pinfo = skb_shinfo(lp); 38458a29111cSEric Dumazet 38468a29111cSEric Dumazet if (headlen <= offset) { 384742da6994SHerbert Xu skb_frag_t *frag; 384866e92fcfSHerbert Xu skb_frag_t *frag2; 38499aaa156cSHerbert Xu int i = skbinfo->nr_frags; 38509aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 385142da6994SHerbert Xu 385266e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 38538a29111cSEric Dumazet goto merge; 385481705ad1SHerbert Xu 38558a29111cSEric Dumazet offset -= headlen; 38569aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 38579aaa156cSHerbert Xu skbinfo->nr_frags = 0; 3858f5572068SHerbert Xu 38599aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 38609aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 386166e92fcfSHerbert Xu do { 386266e92fcfSHerbert Xu *--frag = *--frag2; 386366e92fcfSHerbert Xu } while (--i); 386466e92fcfSHerbert Xu 386566e92fcfSHerbert Xu frag->page_offset += offset; 38669e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 386766e92fcfSHerbert Xu 3868715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3869ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3870ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3871715dc1f3SEric Dumazet 3872f5572068SHerbert Xu skb->truesize -= skb->data_len; 3873f5572068SHerbert Xu skb->len -= skb->data_len; 3874f5572068SHerbert Xu skb->data_len = 0; 3875f5572068SHerbert Xu 3876715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 38775d38a079SHerbert Xu goto done; 3878d7e8883cSEric Dumazet } else if (skb->head_frag) { 3879d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3880d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3881d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3882d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3883d7e8883cSEric Dumazet unsigned int first_offset; 3884d7e8883cSEric Dumazet 3885d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 38868a29111cSEric Dumazet goto merge; 3887d7e8883cSEric Dumazet 3888d7e8883cSEric Dumazet first_offset = skb->data - 3889d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3890d7e8883cSEric Dumazet offset; 3891d7e8883cSEric Dumazet 3892d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3893d7e8883cSEric Dumazet 3894d7e8883cSEric Dumazet frag->page.p = page; 3895d7e8883cSEric Dumazet frag->page_offset = first_offset; 3896d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3897d7e8883cSEric Dumazet 3898d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3899d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3900d7e8883cSEric Dumazet 3901715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3902d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3903d7e8883cSEric Dumazet goto done; 39048a29111cSEric Dumazet } 390571d93b39SHerbert Xu 390671d93b39SHerbert Xu merge: 3907715dc1f3SEric Dumazet delta_truesize = skb->truesize; 390867147ba9SHerbert Xu if (offset > headlen) { 3909d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3910d1dc7abfSMichal Schmidt 3911d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 39129e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3913d1dc7abfSMichal Schmidt skb->data_len -= eat; 3914d1dc7abfSMichal Schmidt skb->len -= eat; 391567147ba9SHerbert Xu offset = headlen; 391656035022SHerbert Xu } 391756035022SHerbert Xu 391867147ba9SHerbert Xu __skb_pull(skb, offset); 391956035022SHerbert Xu 392029e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 39218a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 39228a29111cSEric Dumazet else 3923c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3924c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 3925f4a775d1SEric Dumazet __skb_header_release(skb); 39268a29111cSEric Dumazet lp = p; 392771d93b39SHerbert Xu 39285d38a079SHerbert Xu done: 39295d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 393037fe4732SHerbert Xu p->data_len += len; 3931715dc1f3SEric Dumazet p->truesize += delta_truesize; 393237fe4732SHerbert Xu p->len += len; 39338a29111cSEric Dumazet if (lp != p) { 39348a29111cSEric Dumazet lp->data_len += len; 39358a29111cSEric Dumazet lp->truesize += delta_truesize; 39368a29111cSEric Dumazet lp->len += len; 39378a29111cSEric Dumazet } 393871d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 393971d93b39SHerbert Xu return 0; 394071d93b39SHerbert Xu } 394157c05650SMarcelo Ricardo Leitner EXPORT_SYMBOL_GPL(skb_gro_receive); 394271d93b39SHerbert Xu 3943df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 3944df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 3945df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 3946df5042f4SFlorian Westphal 3947df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 3948df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3949df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 3950df5042f4SFlorian Westphal #endif 39514165079bSFlorian Westphal #ifdef CONFIG_XFRM 39524165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 39534165079bSFlorian Westphal #endif 3954df5042f4SFlorian Westphal }; 3955df5042f4SFlorian Westphal 3956df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 3957df5042f4SFlorian Westphal { 3958df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 3959df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3960df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 3961df5042f4SFlorian Westphal #endif 39624165079bSFlorian Westphal #ifdef CONFIG_XFRM 39634165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 39644165079bSFlorian Westphal #endif 3965df5042f4SFlorian Westphal 0; 3966df5042f4SFlorian Westphal } 3967df5042f4SFlorian Westphal 3968df5042f4SFlorian Westphal static void skb_extensions_init(void) 3969df5042f4SFlorian Westphal { 3970df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 3971df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 3972df5042f4SFlorian Westphal 3973df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 3974df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 3975df5042f4SFlorian Westphal 0, 3976df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 3977df5042f4SFlorian Westphal NULL); 3978df5042f4SFlorian Westphal } 3979df5042f4SFlorian Westphal #else 3980df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 3981df5042f4SFlorian Westphal #endif 3982df5042f4SFlorian Westphal 39831da177e4SLinus Torvalds void __init skb_init(void) 39841da177e4SLinus Torvalds { 398579a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 39861da177e4SLinus Torvalds sizeof(struct sk_buff), 39871da177e4SLinus Torvalds 0, 3988e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 398979a8a642SKees Cook offsetof(struct sk_buff, cb), 399079a8a642SKees Cook sizeof_field(struct sk_buff, cb), 399120c2df83SPaul Mundt NULL); 3992d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3993d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 3994d179cd12SDavid S. Miller 0, 3995e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 399620c2df83SPaul Mundt NULL); 3997df5042f4SFlorian Westphal skb_extensions_init(); 39981da177e4SLinus Torvalds } 39991da177e4SLinus Torvalds 400051c739d1SDavid S. Miller static int 400148a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 400248a1df65SJason A. Donenfeld unsigned int recursion_level) 4003716ea3a7SDavid Howells { 40041a028e50SDavid S. Miller int start = skb_headlen(skb); 40051a028e50SDavid S. Miller int i, copy = start - offset; 4006fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4007716ea3a7SDavid Howells int elt = 0; 4008716ea3a7SDavid Howells 400948a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 401048a1df65SJason A. Donenfeld return -EMSGSIZE; 401148a1df65SJason A. Donenfeld 4012716ea3a7SDavid Howells if (copy > 0) { 4013716ea3a7SDavid Howells if (copy > len) 4014716ea3a7SDavid Howells copy = len; 4015642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4016716ea3a7SDavid Howells elt++; 4017716ea3a7SDavid Howells if ((len -= copy) == 0) 4018716ea3a7SDavid Howells return elt; 4019716ea3a7SDavid Howells offset += copy; 4020716ea3a7SDavid Howells } 4021716ea3a7SDavid Howells 4022716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 40231a028e50SDavid S. Miller int end; 4024716ea3a7SDavid Howells 4025547b792cSIlpo Järvinen WARN_ON(start > offset + len); 40261a028e50SDavid S. Miller 40279e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4028716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4029716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 403048a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 403148a1df65SJason A. Donenfeld return -EMSGSIZE; 4032716ea3a7SDavid Howells 4033716ea3a7SDavid Howells if (copy > len) 4034716ea3a7SDavid Howells copy = len; 4035ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4036642f1490SJens Axboe frag->page_offset+offset-start); 4037716ea3a7SDavid Howells elt++; 4038716ea3a7SDavid Howells if (!(len -= copy)) 4039716ea3a7SDavid Howells return elt; 4040716ea3a7SDavid Howells offset += copy; 4041716ea3a7SDavid Howells } 40421a028e50SDavid S. Miller start = end; 4043716ea3a7SDavid Howells } 4044716ea3a7SDavid Howells 4045fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 404648a1df65SJason A. Donenfeld int end, ret; 4047716ea3a7SDavid Howells 4048547b792cSIlpo Järvinen WARN_ON(start > offset + len); 40491a028e50SDavid S. Miller 4050fbb398a8SDavid S. Miller end = start + frag_iter->len; 4051716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 405248a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 405348a1df65SJason A. Donenfeld return -EMSGSIZE; 405448a1df65SJason A. Donenfeld 4055716ea3a7SDavid Howells if (copy > len) 4056716ea3a7SDavid Howells copy = len; 405748a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 405848a1df65SJason A. Donenfeld copy, recursion_level + 1); 405948a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 406048a1df65SJason A. Donenfeld return ret; 406148a1df65SJason A. Donenfeld elt += ret; 4062716ea3a7SDavid Howells if ((len -= copy) == 0) 4063716ea3a7SDavid Howells return elt; 4064716ea3a7SDavid Howells offset += copy; 4065716ea3a7SDavid Howells } 40661a028e50SDavid S. Miller start = end; 4067716ea3a7SDavid Howells } 4068716ea3a7SDavid Howells BUG_ON(len); 4069716ea3a7SDavid Howells return elt; 4070716ea3a7SDavid Howells } 4071716ea3a7SDavid Howells 407248a1df65SJason A. Donenfeld /** 407348a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 407448a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 407548a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 407648a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 407748a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 407848a1df65SJason A. Donenfeld * 407948a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 408048a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 408148a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 408248a1df65SJason A. Donenfeld * could not fit. 408348a1df65SJason A. Donenfeld */ 408448a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 408548a1df65SJason A. Donenfeld { 408648a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 408748a1df65SJason A. Donenfeld 408848a1df65SJason A. Donenfeld if (nsg <= 0) 408948a1df65SJason A. Donenfeld return nsg; 409048a1df65SJason A. Donenfeld 409148a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 409248a1df65SJason A. Donenfeld 409348a1df65SJason A. Donenfeld return nsg; 409448a1df65SJason A. Donenfeld } 409548a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 409648a1df65SJason A. Donenfeld 409725a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 409825a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 409925a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 410025a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 410125a91d8dSFan Du * 410225a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 410325a91d8dSFan Du * 1. sg_init_table 410425a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 410525a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 410625a91d8dSFan Du * 410725a91d8dSFan Du * This is equivalent to: 410825a91d8dSFan Du * 1. sg_init_table 410925a91d8dSFan Du * 2. skb_to_sgvec(payload1) 411025a91d8dSFan Du * 3. sg_unmark_end 411125a91d8dSFan Du * 4. skb_to_sgvec(payload2) 411225a91d8dSFan Du * 411325a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 411425a91d8dSFan Du * is more preferable. 411525a91d8dSFan Du */ 411625a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 411725a91d8dSFan Du int offset, int len) 411825a91d8dSFan Du { 411948a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 412025a91d8dSFan Du } 412125a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 412225a91d8dSFan Du 412351c739d1SDavid S. Miller 412451c739d1SDavid S. Miller 4125716ea3a7SDavid Howells /** 4126716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4127716ea3a7SDavid Howells * @skb: The socket buffer to check. 4128716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4129716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4130716ea3a7SDavid Howells * 4131716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4132716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4133716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4134716ea3a7SDavid Howells * 4135716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4136716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4137716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4138716ea3a7SDavid Howells * 4139716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4140716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4141716ea3a7SDavid Howells */ 4142716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4143716ea3a7SDavid Howells { 4144716ea3a7SDavid Howells int copyflag; 4145716ea3a7SDavid Howells int elt; 4146716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4147716ea3a7SDavid Howells 4148716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4149716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4150716ea3a7SDavid Howells * at the moment even if they are anonymous). 4151716ea3a7SDavid Howells */ 4152716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4153716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 4154716ea3a7SDavid Howells return -ENOMEM; 4155716ea3a7SDavid Howells 4156716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 415721dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4158716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4159716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4160716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4161716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4162716ea3a7SDavid Howells 4163716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4164716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4165716ea3a7SDavid Howells return -ENOMEM; 4166716ea3a7SDavid Howells 4167716ea3a7SDavid Howells /* Voila! */ 4168716ea3a7SDavid Howells *trailer = skb; 4169716ea3a7SDavid Howells return 1; 4170716ea3a7SDavid Howells } 4171716ea3a7SDavid Howells 4172716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4173716ea3a7SDavid Howells 4174716ea3a7SDavid Howells elt = 1; 4175716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4176716ea3a7SDavid Howells copyflag = 0; 4177716ea3a7SDavid Howells 4178716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4179716ea3a7SDavid Howells int ntail = 0; 4180716ea3a7SDavid Howells 4181716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4182716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4183716ea3a7SDavid Howells * after it. */ 4184716ea3a7SDavid Howells 4185716ea3a7SDavid Howells if (skb_shared(skb1)) 4186716ea3a7SDavid Howells copyflag = 1; 4187716ea3a7SDavid Howells 4188716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4189716ea3a7SDavid Howells 4190716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4191716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 419221dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4193716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4194716ea3a7SDavid Howells ntail = tailbits + 128; 4195716ea3a7SDavid Howells } 4196716ea3a7SDavid Howells 4197716ea3a7SDavid Howells if (copyflag || 4198716ea3a7SDavid Howells skb_cloned(skb1) || 4199716ea3a7SDavid Howells ntail || 4200716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 420121dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4202716ea3a7SDavid Howells struct sk_buff *skb2; 4203716ea3a7SDavid Howells 4204716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4205716ea3a7SDavid Howells if (ntail == 0) 4206716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4207716ea3a7SDavid Howells else 4208716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4209716ea3a7SDavid Howells skb_headroom(skb1), 4210716ea3a7SDavid Howells ntail, 4211716ea3a7SDavid Howells GFP_ATOMIC); 4212716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4213716ea3a7SDavid Howells return -ENOMEM; 4214716ea3a7SDavid Howells 4215716ea3a7SDavid Howells if (skb1->sk) 4216716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4217716ea3a7SDavid Howells 4218716ea3a7SDavid Howells /* Looking around. Are we still alive? 4219716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4220716ea3a7SDavid Howells 4221716ea3a7SDavid Howells skb2->next = skb1->next; 4222716ea3a7SDavid Howells *skb_p = skb2; 4223716ea3a7SDavid Howells kfree_skb(skb1); 4224716ea3a7SDavid Howells skb1 = skb2; 4225716ea3a7SDavid Howells } 4226716ea3a7SDavid Howells elt++; 4227716ea3a7SDavid Howells *trailer = skb1; 4228716ea3a7SDavid Howells skb_p = &skb1->next; 4229716ea3a7SDavid Howells } 4230716ea3a7SDavid Howells 4231716ea3a7SDavid Howells return elt; 4232716ea3a7SDavid Howells } 4233b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4234716ea3a7SDavid Howells 4235b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4236b1faf566SEric Dumazet { 4237b1faf566SEric Dumazet struct sock *sk = skb->sk; 4238b1faf566SEric Dumazet 4239b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4240b1faf566SEric Dumazet } 4241b1faf566SEric Dumazet 42428605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 42438605330aSSoheil Hassas Yeganeh { 42448605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 42458605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 42468605330aSSoheil Hassas Yeganeh */ 42478605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 42488605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 42498605330aSSoheil Hassas Yeganeh } 42508605330aSSoheil Hassas Yeganeh 4251b1faf566SEric Dumazet /* 4252b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4253b1faf566SEric Dumazet */ 4254b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4255b1faf566SEric Dumazet { 4256b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 425795c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 4258b1faf566SEric Dumazet return -ENOMEM; 4259b1faf566SEric Dumazet 4260b1faf566SEric Dumazet skb_orphan(skb); 4261b1faf566SEric Dumazet skb->sk = sk; 4262b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4263b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 42648605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4265b1faf566SEric Dumazet 4266abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4267abb57ea4SEric Dumazet skb_dst_force(skb); 4268abb57ea4SEric Dumazet 4269b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4270b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 42716e5d58fdSVinicius Costa Gomes sk->sk_error_report(sk); 4272b1faf566SEric Dumazet return 0; 4273b1faf566SEric Dumazet } 4274b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4275b1faf566SEric Dumazet 427683a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 427783a1a1a7SSoheil Hassas Yeganeh { 427883a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 427983a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 428083a1a1a7SSoheil Hassas Yeganeh } 428183a1a1a7SSoheil Hassas Yeganeh 4282364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4283364a9e93SWillem de Bruijn { 4284364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 428583a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 428683a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4287997d5c3fSEric Dumazet unsigned long flags; 4288364a9e93SWillem de Bruijn 4289997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4290364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 429138b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 429283a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 429338b25793SSoheil Hassas Yeganeh if (icmp_next) 429438b25793SSoheil Hassas Yeganeh sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin; 429538b25793SSoheil Hassas Yeganeh } 4296997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4297364a9e93SWillem de Bruijn 429883a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 429983a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 430083a1a1a7SSoheil Hassas Yeganeh 430183a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4302364a9e93SWillem de Bruijn sk->sk_error_report(sk); 4303364a9e93SWillem de Bruijn 4304364a9e93SWillem de Bruijn return skb; 4305364a9e93SWillem de Bruijn } 4306364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4307364a9e93SWillem de Bruijn 4308cab41c47SAlexander Duyck /** 4309cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4310cab41c47SAlexander Duyck * @skb: the skb to clone 4311cab41c47SAlexander Duyck * 4312cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4313cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4314cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4315cab41c47SAlexander Duyck * 4316cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4317cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4318cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4319cab41c47SAlexander Duyck * the sk_error_queue. 4320cab41c47SAlexander Duyck */ 432162bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 432262bccb8cSAlexander Duyck { 432362bccb8cSAlexander Duyck struct sock *sk = skb->sk; 432462bccb8cSAlexander Duyck struct sk_buff *clone; 432562bccb8cSAlexander Duyck 432641c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 432762bccb8cSAlexander Duyck return NULL; 432862bccb8cSAlexander Duyck 432962bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 433062bccb8cSAlexander Duyck if (!clone) { 433162bccb8cSAlexander Duyck sock_put(sk); 433262bccb8cSAlexander Duyck return NULL; 433362bccb8cSAlexander Duyck } 433462bccb8cSAlexander Duyck 433562bccb8cSAlexander Duyck clone->sk = sk; 433662bccb8cSAlexander Duyck clone->destructor = sock_efree; 433762bccb8cSAlexander Duyck 433862bccb8cSAlexander Duyck return clone; 433962bccb8cSAlexander Duyck } 434062bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 434162bccb8cSAlexander Duyck 434237846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 434337846ef0SAlexander Duyck struct sock *sk, 43444ef1b286SSoheil Hassas Yeganeh int tstype, 43454ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4346ac45f602SPatrick Ohly { 4347ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4348ac45f602SPatrick Ohly int err; 4349ac45f602SPatrick Ohly 43504ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 43514ef1b286SSoheil Hassas Yeganeh 4352ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4353ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4354ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4355ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4356e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 43574ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 43581862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 43594ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 436009c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 4361ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 4362ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 43634ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 43644ed2d765SWillem de Bruijn } 436529030374SEric Dumazet 4366ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 436729030374SEric Dumazet 4368ac45f602SPatrick Ohly if (err) 4369ac45f602SPatrick Ohly kfree_skb(skb); 4370ac45f602SPatrick Ohly } 437137846ef0SAlexander Duyck 4372b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4373b245be1fSWillem de Bruijn { 4374b245be1fSWillem de Bruijn bool ret; 4375b245be1fSWillem de Bruijn 4376b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4377b245be1fSWillem de Bruijn return true; 4378b245be1fSWillem de Bruijn 4379b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4380b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4381b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4382b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4383b245be1fSWillem de Bruijn return ret; 4384b245be1fSWillem de Bruijn } 4385b245be1fSWillem de Bruijn 438637846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 438737846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 438837846ef0SAlexander Duyck { 438937846ef0SAlexander Duyck struct sock *sk = skb->sk; 439037846ef0SAlexander Duyck 4391b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 439235b99dffSWillem de Bruijn goto err; 4393b245be1fSWillem de Bruijn 43949ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 43959ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 43969ac25fc0SEric Dumazet */ 439741c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 439837846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 43994ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 440037846ef0SAlexander Duyck sock_put(sk); 440135b99dffSWillem de Bruijn return; 440237846ef0SAlexander Duyck } 440335b99dffSWillem de Bruijn 440435b99dffSWillem de Bruijn err: 440535b99dffSWillem de Bruijn kfree_skb(skb); 44069ac25fc0SEric Dumazet } 440737846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 440837846ef0SAlexander Duyck 440937846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 441037846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 441137846ef0SAlexander Duyck struct sock *sk, int tstype) 441237846ef0SAlexander Duyck { 441337846ef0SAlexander Duyck struct sk_buff *skb; 44144ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 441537846ef0SAlexander Duyck 44163a8dd971SWillem de Bruijn if (!sk) 44173a8dd971SWillem de Bruijn return; 44183a8dd971SWillem de Bruijn 4419b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4420b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4421b50a5c70SMiroslav Lichvar return; 4422b50a5c70SMiroslav Lichvar 44233a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 44243a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 442537846ef0SAlexander Duyck return; 442637846ef0SAlexander Duyck 44271c885808SFrancis Yan if (tsonly) { 44281c885808SFrancis Yan #ifdef CONFIG_INET 44291c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 44301c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 44314ef1b286SSoheil Hassas Yeganeh sk->sk_type == SOCK_STREAM) { 44321c885808SFrancis Yan skb = tcp_get_timestamping_opt_stats(sk); 44334ef1b286SSoheil Hassas Yeganeh opt_stats = true; 44344ef1b286SSoheil Hassas Yeganeh } else 44351c885808SFrancis Yan #endif 44361c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 44371c885808SFrancis Yan } else { 443837846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 44391c885808SFrancis Yan } 444037846ef0SAlexander Duyck if (!skb) 444137846ef0SAlexander Duyck return; 444237846ef0SAlexander Duyck 444349ca0d8bSWillem de Bruijn if (tsonly) { 4444fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4445fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 444649ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 444749ca0d8bSWillem de Bruijn } 444849ca0d8bSWillem de Bruijn 444949ca0d8bSWillem de Bruijn if (hwtstamps) 445049ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 445149ca0d8bSWillem de Bruijn else 445249ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 445349ca0d8bSWillem de Bruijn 44544ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 445537846ef0SAlexander Duyck } 4456e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4457e7fd2885SWillem de Bruijn 4458e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4459e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4460e7fd2885SWillem de Bruijn { 4461e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 4462e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4463e7fd2885SWillem de Bruijn } 4464ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4465ac45f602SPatrick Ohly 44666e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 44676e3e939fSJohannes Berg { 44686e3e939fSJohannes Berg struct sock *sk = skb->sk; 44696e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4470dd4f1072SEric Dumazet int err = 1; 44716e3e939fSJohannes Berg 44726e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 44736e3e939fSJohannes Berg skb->wifi_acked = acked; 44746e3e939fSJohannes Berg 44756e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 44766e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 44776e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 44786e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 44796e3e939fSJohannes Berg 4480dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4481dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4482dd4f1072SEric Dumazet */ 448341c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 44846e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4485dd4f1072SEric Dumazet sock_put(sk); 4486dd4f1072SEric Dumazet } 44876e3e939fSJohannes Berg if (err) 44886e3e939fSJohannes Berg kfree_skb(skb); 44896e3e939fSJohannes Berg } 44906e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 44916e3e939fSJohannes Berg 4492f35d9d8aSRusty Russell /** 4493f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4494f35d9d8aSRusty Russell * @skb: the skb to set 4495f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4496f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4497f35d9d8aSRusty Russell * 4498f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4499f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4500f35d9d8aSRusty Russell * 4501f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4502f35d9d8aSRusty Russell * returns false you should drop the packet. 4503f35d9d8aSRusty Russell */ 4504f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4505f35d9d8aSRusty Russell { 450652b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 450752b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 450852b5d6f5SEric Dumazet 450952b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 451052b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 451152b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4512f35d9d8aSRusty Russell return false; 4513f35d9d8aSRusty Russell } 4514f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 451552b5d6f5SEric Dumazet skb->csum_start = csum_start; 4516f35d9d8aSRusty Russell skb->csum_offset = off; 4517e5d5decaSJason Wang skb_set_transport_header(skb, start); 4518f35d9d8aSRusty Russell return true; 4519f35d9d8aSRusty Russell } 4520b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4521f35d9d8aSRusty Russell 4522ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4523ed1f50c3SPaul Durrant unsigned int max) 4524ed1f50c3SPaul Durrant { 4525ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4526ed1f50c3SPaul Durrant return 0; 4527ed1f50c3SPaul Durrant 4528ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4529ed1f50c3SPaul Durrant * won't need to do it again. 4530ed1f50c3SPaul Durrant */ 4531ed1f50c3SPaul Durrant if (max > skb->len) 4532ed1f50c3SPaul Durrant max = skb->len; 4533ed1f50c3SPaul Durrant 4534ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4535ed1f50c3SPaul Durrant return -ENOMEM; 4536ed1f50c3SPaul Durrant 4537ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4538ed1f50c3SPaul Durrant return -EPROTO; 4539ed1f50c3SPaul Durrant 4540ed1f50c3SPaul Durrant return 0; 4541ed1f50c3SPaul Durrant } 4542ed1f50c3SPaul Durrant 4543f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4544f9708b43SJan Beulich 4545f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4546f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4547f9708b43SJan Beulich unsigned int off) 4548f9708b43SJan Beulich { 4549f9708b43SJan Beulich switch (proto) { 4550f9708b43SJan Beulich int err; 4551f9708b43SJan Beulich 4552f9708b43SJan Beulich case IPPROTO_TCP: 4553f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4554f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4555f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4556f9708b43SJan Beulich offsetof(struct tcphdr, 4557f9708b43SJan Beulich check))) 4558f9708b43SJan Beulich err = -EPROTO; 4559f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4560f9708b43SJan Beulich 4561f9708b43SJan Beulich case IPPROTO_UDP: 4562f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4563f9708b43SJan Beulich off + sizeof(struct udphdr)); 4564f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4565f9708b43SJan Beulich offsetof(struct udphdr, 4566f9708b43SJan Beulich check))) 4567f9708b43SJan Beulich err = -EPROTO; 4568f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4569f9708b43SJan Beulich } 4570f9708b43SJan Beulich 4571f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4572f9708b43SJan Beulich } 4573f9708b43SJan Beulich 4574ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4575ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4576ed1f50c3SPaul Durrant */ 4577ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4578ed1f50c3SPaul Durrant 4579f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4580ed1f50c3SPaul Durrant { 4581ed1f50c3SPaul Durrant unsigned int off; 4582ed1f50c3SPaul Durrant bool fragment; 4583f9708b43SJan Beulich __sum16 *csum; 4584ed1f50c3SPaul Durrant int err; 4585ed1f50c3SPaul Durrant 4586ed1f50c3SPaul Durrant fragment = false; 4587ed1f50c3SPaul Durrant 4588ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4589ed1f50c3SPaul Durrant sizeof(struct iphdr), 4590ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4591ed1f50c3SPaul Durrant if (err < 0) 4592ed1f50c3SPaul Durrant goto out; 4593ed1f50c3SPaul Durrant 4594ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 4595ed1f50c3SPaul Durrant fragment = true; 4596ed1f50c3SPaul Durrant 4597ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 4598ed1f50c3SPaul Durrant 4599ed1f50c3SPaul Durrant err = -EPROTO; 4600ed1f50c3SPaul Durrant 4601ed1f50c3SPaul Durrant if (fragment) 4602ed1f50c3SPaul Durrant goto out; 4603ed1f50c3SPaul Durrant 4604f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4605f9708b43SJan Beulich if (IS_ERR(csum)) 4606f9708b43SJan Beulich return PTR_ERR(csum); 4607ed1f50c3SPaul Durrant 4608ed1f50c3SPaul Durrant if (recalculate) 4609f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4610ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4611ed1f50c3SPaul Durrant skb->len - off, 4612f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4613ed1f50c3SPaul Durrant err = 0; 4614ed1f50c3SPaul Durrant 4615ed1f50c3SPaul Durrant out: 4616ed1f50c3SPaul Durrant return err; 4617ed1f50c3SPaul Durrant } 4618ed1f50c3SPaul Durrant 4619ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4620ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4621ed1f50c3SPaul Durrant */ 4622ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4623ed1f50c3SPaul Durrant 4624ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4625ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4626ed1f50c3SPaul Durrant 4627ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4628ed1f50c3SPaul Durrant { 4629ed1f50c3SPaul Durrant int err; 4630ed1f50c3SPaul Durrant u8 nexthdr; 4631ed1f50c3SPaul Durrant unsigned int off; 4632ed1f50c3SPaul Durrant unsigned int len; 4633ed1f50c3SPaul Durrant bool fragment; 4634ed1f50c3SPaul Durrant bool done; 4635f9708b43SJan Beulich __sum16 *csum; 4636ed1f50c3SPaul Durrant 4637ed1f50c3SPaul Durrant fragment = false; 4638ed1f50c3SPaul Durrant done = false; 4639ed1f50c3SPaul Durrant 4640ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4641ed1f50c3SPaul Durrant 4642ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4643ed1f50c3SPaul Durrant if (err < 0) 4644ed1f50c3SPaul Durrant goto out; 4645ed1f50c3SPaul Durrant 4646ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4647ed1f50c3SPaul Durrant 4648ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4649ed1f50c3SPaul Durrant while (off <= len && !done) { 4650ed1f50c3SPaul Durrant switch (nexthdr) { 4651ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4652ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4653ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4654ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 4655ed1f50c3SPaul Durrant 4656ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4657ed1f50c3SPaul Durrant off + 4658ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 4659ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4660ed1f50c3SPaul Durrant if (err < 0) 4661ed1f50c3SPaul Durrant goto out; 4662ed1f50c3SPaul Durrant 4663ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4664ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4665ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 4666ed1f50c3SPaul Durrant break; 4667ed1f50c3SPaul Durrant } 4668ed1f50c3SPaul Durrant case IPPROTO_AH: { 4669ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 4670ed1f50c3SPaul Durrant 4671ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4672ed1f50c3SPaul Durrant off + 4673ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 4674ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4675ed1f50c3SPaul Durrant if (err < 0) 4676ed1f50c3SPaul Durrant goto out; 4677ed1f50c3SPaul Durrant 4678ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4679ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4680ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 4681ed1f50c3SPaul Durrant break; 4682ed1f50c3SPaul Durrant } 4683ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 4684ed1f50c3SPaul Durrant struct frag_hdr *hp; 4685ed1f50c3SPaul Durrant 4686ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4687ed1f50c3SPaul Durrant off + 4688ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 4689ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4690ed1f50c3SPaul Durrant if (err < 0) 4691ed1f50c3SPaul Durrant goto out; 4692ed1f50c3SPaul Durrant 4693ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 4694ed1f50c3SPaul Durrant 4695ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 4696ed1f50c3SPaul Durrant fragment = true; 4697ed1f50c3SPaul Durrant 4698ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4699ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 4700ed1f50c3SPaul Durrant break; 4701ed1f50c3SPaul Durrant } 4702ed1f50c3SPaul Durrant default: 4703ed1f50c3SPaul Durrant done = true; 4704ed1f50c3SPaul Durrant break; 4705ed1f50c3SPaul Durrant } 4706ed1f50c3SPaul Durrant } 4707ed1f50c3SPaul Durrant 4708ed1f50c3SPaul Durrant err = -EPROTO; 4709ed1f50c3SPaul Durrant 4710ed1f50c3SPaul Durrant if (!done || fragment) 4711ed1f50c3SPaul Durrant goto out; 4712ed1f50c3SPaul Durrant 4713f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 4714f9708b43SJan Beulich if (IS_ERR(csum)) 4715f9708b43SJan Beulich return PTR_ERR(csum); 4716ed1f50c3SPaul Durrant 4717ed1f50c3SPaul Durrant if (recalculate) 4718f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 4719ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 4720f9708b43SJan Beulich skb->len - off, nexthdr, 0); 4721ed1f50c3SPaul Durrant err = 0; 4722ed1f50c3SPaul Durrant 4723ed1f50c3SPaul Durrant out: 4724ed1f50c3SPaul Durrant return err; 4725ed1f50c3SPaul Durrant } 4726ed1f50c3SPaul Durrant 4727ed1f50c3SPaul Durrant /** 4728ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 4729ed1f50c3SPaul Durrant * @skb: the skb to set up 4730ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 4731ed1f50c3SPaul Durrant */ 4732ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 4733ed1f50c3SPaul Durrant { 4734ed1f50c3SPaul Durrant int err; 4735ed1f50c3SPaul Durrant 4736ed1f50c3SPaul Durrant switch (skb->protocol) { 4737ed1f50c3SPaul Durrant case htons(ETH_P_IP): 4738f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 4739ed1f50c3SPaul Durrant break; 4740ed1f50c3SPaul Durrant 4741ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 4742ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 4743ed1f50c3SPaul Durrant break; 4744ed1f50c3SPaul Durrant 4745ed1f50c3SPaul Durrant default: 4746ed1f50c3SPaul Durrant err = -EPROTO; 4747ed1f50c3SPaul Durrant break; 4748ed1f50c3SPaul Durrant } 4749ed1f50c3SPaul Durrant 4750ed1f50c3SPaul Durrant return err; 4751ed1f50c3SPaul Durrant } 4752ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 4753ed1f50c3SPaul Durrant 47549afd85c9SLinus Lüssing /** 47559afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 47569afd85c9SLinus Lüssing * @skb: the skb to check 47579afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 47589afd85c9SLinus Lüssing * 47599afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 47609afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 47619afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 47629afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 47639afd85c9SLinus Lüssing * 4764a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4765a516993fSLinus Lüssing * differs from the provided skb. 47669afd85c9SLinus Lüssing */ 47679afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 47689afd85c9SLinus Lüssing unsigned int transport_len) 47699afd85c9SLinus Lüssing { 47709afd85c9SLinus Lüssing struct sk_buff *skb_chk; 47719afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 47729afd85c9SLinus Lüssing int ret; 47739afd85c9SLinus Lüssing 4774a516993fSLinus Lüssing if (skb->len < len) 47759afd85c9SLinus Lüssing return NULL; 4776a516993fSLinus Lüssing else if (skb->len == len) 47779afd85c9SLinus Lüssing return skb; 47789afd85c9SLinus Lüssing 47799afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 47809afd85c9SLinus Lüssing if (!skb_chk) 47819afd85c9SLinus Lüssing return NULL; 47829afd85c9SLinus Lüssing 47839afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 47849afd85c9SLinus Lüssing if (ret) { 47859afd85c9SLinus Lüssing kfree_skb(skb_chk); 47869afd85c9SLinus Lüssing return NULL; 47879afd85c9SLinus Lüssing } 47889afd85c9SLinus Lüssing 47899afd85c9SLinus Lüssing return skb_chk; 47909afd85c9SLinus Lüssing } 47919afd85c9SLinus Lüssing 47929afd85c9SLinus Lüssing /** 47939afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 47949afd85c9SLinus Lüssing * @skb: the skb to check 47959afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 47969afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 47979afd85c9SLinus Lüssing * 47989afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 47999afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 48009afd85c9SLinus Lüssing * 48019afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 48029afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 48039afd85c9SLinus Lüssing * 4804a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4805a516993fSLinus Lüssing * differs from the provided skb. 48069afd85c9SLinus Lüssing */ 48079afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 48089afd85c9SLinus Lüssing unsigned int transport_len, 48099afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 48109afd85c9SLinus Lüssing { 48119afd85c9SLinus Lüssing struct sk_buff *skb_chk; 48129afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 4813fcba67c9SLinus Lüssing __sum16 ret; 48149afd85c9SLinus Lüssing 48159afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 48169afd85c9SLinus Lüssing if (!skb_chk) 4817a516993fSLinus Lüssing goto err; 48189afd85c9SLinus Lüssing 4819a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 4820a516993fSLinus Lüssing goto err; 48219afd85c9SLinus Lüssing 48229b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 48239afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 48249b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 48259afd85c9SLinus Lüssing 4826a516993fSLinus Lüssing if (ret) 4827a516993fSLinus Lüssing goto err; 48289afd85c9SLinus Lüssing 48299afd85c9SLinus Lüssing return skb_chk; 4830a516993fSLinus Lüssing 4831a516993fSLinus Lüssing err: 4832a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 4833a516993fSLinus Lüssing kfree_skb(skb_chk); 4834a516993fSLinus Lüssing 4835a516993fSLinus Lüssing return NULL; 4836a516993fSLinus Lüssing 48379afd85c9SLinus Lüssing } 48389afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 48399afd85c9SLinus Lüssing 48404497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 48414497b076SBen Hutchings { 4842e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 4843e87cc472SJoe Perches skb->dev->name); 48444497b076SBen Hutchings } 48454497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 4846bad43ca8SEric Dumazet 4847bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 4848bad43ca8SEric Dumazet { 48493d861f66SEric Dumazet if (head_stolen) { 48503d861f66SEric Dumazet skb_release_head_state(skb); 4851bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 48523d861f66SEric Dumazet } else { 4853bad43ca8SEric Dumazet __kfree_skb(skb); 4854bad43ca8SEric Dumazet } 48553d861f66SEric Dumazet } 4856bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 4857bad43ca8SEric Dumazet 4858bad43ca8SEric Dumazet /** 4859bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 4860bad43ca8SEric Dumazet * @to: prior buffer 4861bad43ca8SEric Dumazet * @from: buffer to add 4862bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 4863c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 4864bad43ca8SEric Dumazet */ 4865bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 4866bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 4867bad43ca8SEric Dumazet { 4868c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 4869bad43ca8SEric Dumazet int i, delta, len = from->len; 4870bad43ca8SEric Dumazet 4871bad43ca8SEric Dumazet *fragstolen = false; 4872bad43ca8SEric Dumazet 4873bad43ca8SEric Dumazet if (skb_cloned(to)) 4874bad43ca8SEric Dumazet return false; 4875bad43ca8SEric Dumazet 4876bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 4877e93a0435SEric Dumazet if (len) 4878bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 4879bad43ca8SEric Dumazet *delta_truesize = 0; 4880bad43ca8SEric Dumazet return true; 4881bad43ca8SEric Dumazet } 4882bad43ca8SEric Dumazet 4883c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 4884c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 4885c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 4886bad43ca8SEric Dumazet return false; 48871f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 48881f8b977aSWillem de Bruijn return false; 4889bad43ca8SEric Dumazet 4890bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 4891bad43ca8SEric Dumazet struct page *page; 4892bad43ca8SEric Dumazet unsigned int offset; 4893bad43ca8SEric Dumazet 4894c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 4895c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 4896bad43ca8SEric Dumazet return false; 4897bad43ca8SEric Dumazet 4898bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 4899bad43ca8SEric Dumazet return false; 4900bad43ca8SEric Dumazet 4901bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4902bad43ca8SEric Dumazet 4903bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 4904bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 4905bad43ca8SEric Dumazet 4906c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 4907bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 4908bad43ca8SEric Dumazet *fragstolen = true; 4909bad43ca8SEric Dumazet } else { 4910c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 4911c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 4912bad43ca8SEric Dumazet return false; 4913bad43ca8SEric Dumazet 4914f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 4915bad43ca8SEric Dumazet } 4916bad43ca8SEric Dumazet 4917bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 4918bad43ca8SEric Dumazet 4919c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 4920c818fa9eSEric Dumazet from_shinfo->frags, 4921c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 4922c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 4923bad43ca8SEric Dumazet 4924bad43ca8SEric Dumazet if (!skb_cloned(from)) 4925c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 4926bad43ca8SEric Dumazet 49278ea853fdSLi RongQing /* if the skb is not cloned this does nothing 49288ea853fdSLi RongQing * since we set nr_frags to 0. 49298ea853fdSLi RongQing */ 4930c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 4931c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 4932bad43ca8SEric Dumazet 4933bad43ca8SEric Dumazet to->truesize += delta; 4934bad43ca8SEric Dumazet to->len += len; 4935bad43ca8SEric Dumazet to->data_len += len; 4936bad43ca8SEric Dumazet 4937bad43ca8SEric Dumazet *delta_truesize = delta; 4938bad43ca8SEric Dumazet return true; 4939bad43ca8SEric Dumazet } 4940bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 4941621e84d6SNicolas Dichtel 4942621e84d6SNicolas Dichtel /** 49438b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 4944621e84d6SNicolas Dichtel * 4945621e84d6SNicolas Dichtel * @skb: buffer to clean 49468b27f277SNicolas Dichtel * @xnet: packet is crossing netns 4947621e84d6SNicolas Dichtel * 49488b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 49498b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 49508b27f277SNicolas Dichtel * operations. 49518b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 49528b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 49538b27f277SNicolas Dichtel * skb that could impact namespace isolation. 4954621e84d6SNicolas Dichtel */ 49558b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4956621e84d6SNicolas Dichtel { 4957621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 4958621e84d6SNicolas Dichtel skb->skb_iif = 0; 495960ff7467SWANG Cong skb->ignore_df = 0; 4960621e84d6SNicolas Dichtel skb_dst_drop(skb); 4961621e84d6SNicolas Dichtel secpath_reset(skb); 4962621e84d6SNicolas Dichtel nf_reset(skb); 4963621e84d6SNicolas Dichtel nf_reset_trace(skb); 4964213dd74aSHerbert Xu 49656f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 49666f9a5069SPetr Machata skb->offload_fwd_mark = 0; 4967875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 49686f9a5069SPetr Machata #endif 49696f9a5069SPetr Machata 4970213dd74aSHerbert Xu if (!xnet) 4971213dd74aSHerbert Xu return; 4972213dd74aSHerbert Xu 49732b5ec1a5SYe Yin ipvs_reset(skb); 4974213dd74aSHerbert Xu skb->mark = 0; 4975c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 4976621e84d6SNicolas Dichtel } 4977621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 4978de960aa9SFlorian Westphal 4979de960aa9SFlorian Westphal /** 4980de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4981de960aa9SFlorian Westphal * 4982de960aa9SFlorian Westphal * @skb: GSO skb 4983de960aa9SFlorian Westphal * 4984de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 4985de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 4986de960aa9SFlorian Westphal * 4987de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4988de960aa9SFlorian Westphal */ 4989a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4990de960aa9SFlorian Westphal { 4991de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 4992f993bc25SFlorian Westphal unsigned int thlen = 0; 4993f993bc25SFlorian Westphal 4994f993bc25SFlorian Westphal if (skb->encapsulation) { 4995f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 4996f993bc25SFlorian Westphal skb_transport_header(skb); 4997de960aa9SFlorian Westphal 4998de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 4999f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5000f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5001f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 50021dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 500390017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5004ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5005ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5006f993bc25SFlorian Westphal } 50076d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 50086d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 50096d39d589SFlorian Westphal * accounted for. 50106d39d589SFlorian Westphal */ 5011f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5012de960aa9SFlorian Westphal } 5013a4a77718SDaniel Axtens 5014a4a77718SDaniel Axtens /** 5015a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5016a4a77718SDaniel Axtens * 5017a4a77718SDaniel Axtens * @skb: GSO skb 5018a4a77718SDaniel Axtens * 5019a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5020a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5021a4a77718SDaniel Axtens * 5022a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5023a4a77718SDaniel Axtens */ 5024a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5025a4a77718SDaniel Axtens { 5026a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5027a4a77718SDaniel Axtens skb_network_header(skb); 5028a4a77718SDaniel Axtens 5029a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5030a4a77718SDaniel Axtens } 5031a4a77718SDaniel Axtens 5032a4a77718SDaniel Axtens /** 5033a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5034a4a77718SDaniel Axtens * 5035a4a77718SDaniel Axtens * @skb: GSO skb 5036a4a77718SDaniel Axtens * 5037a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5038a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5039a4a77718SDaniel Axtens * headers (TCP/UDP). 5040a4a77718SDaniel Axtens */ 5041a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5042a4a77718SDaniel Axtens { 5043a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5044a4a77718SDaniel Axtens 5045a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5046a4a77718SDaniel Axtens } 50470d5501c1SVlad Yasevich 5048ae7ef81eSMarcelo Ricardo Leitner /** 50492b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 50502b16f048SDaniel Axtens * 50512b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 50522b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 50532b16f048SDaniel Axtens * 50542b16f048SDaniel Axtens * We might want to check: 50552b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 50562b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 50572b16f048SDaniel Axtens * 50582b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 50592b16f048SDaniel Axtens * 506049682bfaSMathieu Malaterre * @skb: GSO skb 506149682bfaSMathieu Malaterre * 50622b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 50632b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 50642b16f048SDaniel Axtens * 50652b16f048SDaniel Axtens * @max_len: The maximum permissible length. 50662b16f048SDaniel Axtens * 50672b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 50682b16f048SDaniel Axtens */ 50692b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 50702b16f048SDaniel Axtens unsigned int seg_len, 50712b16f048SDaniel Axtens unsigned int max_len) { 50722b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 50732b16f048SDaniel Axtens const struct sk_buff *iter; 50742b16f048SDaniel Axtens 50752b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 50762b16f048SDaniel Axtens return seg_len <= max_len; 50772b16f048SDaniel Axtens 50782b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 50792b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 50802b16f048SDaniel Axtens 50812b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 50822b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 50832b16f048SDaniel Axtens return false; 50842b16f048SDaniel Axtens } 50852b16f048SDaniel Axtens 50862b16f048SDaniel Axtens return true; 50872b16f048SDaniel Axtens } 50882b16f048SDaniel Axtens 50892b16f048SDaniel Axtens /** 5090779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5091ae7ef81eSMarcelo Ricardo Leitner * 5092ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 509376f21b99SDavid S. Miller * @mtu: MTU to validate against 5094ae7ef81eSMarcelo Ricardo Leitner * 5095779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5096779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5097779b7931SDaniel Axtens * payload. 5098ae7ef81eSMarcelo Ricardo Leitner */ 5099779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5100ae7ef81eSMarcelo Ricardo Leitner { 51012b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5102ae7ef81eSMarcelo Ricardo Leitner } 5103779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5104ae7ef81eSMarcelo Ricardo Leitner 51052b16f048SDaniel Axtens /** 51062b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 51072b16f048SDaniel Axtens * 51082b16f048SDaniel Axtens * @skb: GSO skb 51092b16f048SDaniel Axtens * @len: length to validate against 51102b16f048SDaniel Axtens * 51112b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 51122b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 51132b16f048SDaniel Axtens */ 51142b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 51152b16f048SDaniel Axtens { 51162b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 51172b16f048SDaniel Axtens } 51182b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 51192b16f048SDaniel Axtens 51200d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 51210d5501c1SVlad Yasevich { 5122d85e8be2SYuya Kusakabe int mac_len, meta_len; 5123d85e8be2SYuya Kusakabe void *meta; 51244bbb3e0eSToshiaki Makita 51250d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 51260d5501c1SVlad Yasevich kfree_skb(skb); 51270d5501c1SVlad Yasevich return NULL; 51280d5501c1SVlad Yasevich } 51290d5501c1SVlad Yasevich 51304bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5131ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 51324bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 51334bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5134ae474573SToshiaki Makita } 5135d85e8be2SYuya Kusakabe 5136d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5137d85e8be2SYuya Kusakabe if (meta_len) { 5138d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5139d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5140d85e8be2SYuya Kusakabe } 5141d85e8be2SYuya Kusakabe 51420d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 51430d5501c1SVlad Yasevich return skb; 51440d5501c1SVlad Yasevich } 51450d5501c1SVlad Yasevich 51460d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 51470d5501c1SVlad Yasevich { 51480d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 51490d5501c1SVlad Yasevich u16 vlan_tci; 51500d5501c1SVlad Yasevich 5151df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 51520d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 51530d5501c1SVlad Yasevich return skb; 51540d5501c1SVlad Yasevich } 51550d5501c1SVlad Yasevich 51560d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 51570d5501c1SVlad Yasevich if (unlikely(!skb)) 51580d5501c1SVlad Yasevich goto err_free; 51590d5501c1SVlad Yasevich 51600d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 51610d5501c1SVlad Yasevich goto err_free; 51620d5501c1SVlad Yasevich 51630d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 51640d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 51650d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 51660d5501c1SVlad Yasevich 51670d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 51680d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 51690d5501c1SVlad Yasevich 51700d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 51710d5501c1SVlad Yasevich if (unlikely(!skb)) 51720d5501c1SVlad Yasevich goto err_free; 51730d5501c1SVlad Yasevich 51740d5501c1SVlad Yasevich skb_reset_network_header(skb); 51750d5501c1SVlad Yasevich skb_reset_transport_header(skb); 51760d5501c1SVlad Yasevich skb_reset_mac_len(skb); 51770d5501c1SVlad Yasevich 51780d5501c1SVlad Yasevich return skb; 51790d5501c1SVlad Yasevich 51800d5501c1SVlad Yasevich err_free: 51810d5501c1SVlad Yasevich kfree_skb(skb); 51820d5501c1SVlad Yasevich return NULL; 51830d5501c1SVlad Yasevich } 51840d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 51852e4e4410SEric Dumazet 5186e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5187e2195121SJiri Pirko { 5188e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5189e2195121SJiri Pirko return -ENOMEM; 5190e2195121SJiri Pirko 5191e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5192e2195121SJiri Pirko return 0; 5193e2195121SJiri Pirko 5194e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5195e2195121SJiri Pirko } 5196e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5197e2195121SJiri Pirko 5198bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5199bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5200bfca4c52SShmulik Ladkani */ 5201bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 520293515d53SJiri Pirko { 520393515d53SJiri Pirko struct vlan_hdr *vhdr; 5204b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 520593515d53SJiri Pirko int err; 520693515d53SJiri Pirko 5207b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5208b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5209b6a79208SShmulik Ladkani offset)) { 5210b6a79208SShmulik Ladkani return -EINVAL; 5211b6a79208SShmulik Ladkani } 5212b6a79208SShmulik Ladkani 521393515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 521493515d53SJiri Pirko if (unlikely(err)) 5215b6a79208SShmulik Ladkani return err; 521693515d53SJiri Pirko 521793515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 521893515d53SJiri Pirko 521993515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 522093515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 522193515d53SJiri Pirko 522293515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 522393515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 522493515d53SJiri Pirko 522593515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 522693515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 522793515d53SJiri Pirko 522893515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 522993515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 523093515d53SJiri Pirko 523193515d53SJiri Pirko skb_reset_mac_len(skb); 523293515d53SJiri Pirko 523393515d53SJiri Pirko return err; 523493515d53SJiri Pirko } 5235bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 523693515d53SJiri Pirko 5237b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5238b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5239b6a79208SShmulik Ladkani */ 524093515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 524193515d53SJiri Pirko { 524293515d53SJiri Pirko u16 vlan_tci; 524393515d53SJiri Pirko __be16 vlan_proto; 524493515d53SJiri Pirko int err; 524593515d53SJiri Pirko 5246df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5247b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 524893515d53SJiri Pirko } else { 5249ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 525093515d53SJiri Pirko return 0; 525193515d53SJiri Pirko 525293515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 525393515d53SJiri Pirko if (err) 525493515d53SJiri Pirko return err; 525593515d53SJiri Pirko } 525693515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5257ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 525893515d53SJiri Pirko return 0; 525993515d53SJiri Pirko 526093515d53SJiri Pirko vlan_proto = skb->protocol; 526193515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 526293515d53SJiri Pirko if (unlikely(err)) 526393515d53SJiri Pirko return err; 526493515d53SJiri Pirko 526593515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 526693515d53SJiri Pirko return 0; 526793515d53SJiri Pirko } 526893515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 526993515d53SJiri Pirko 5270b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5271b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5272b6a79208SShmulik Ladkani */ 527393515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 527493515d53SJiri Pirko { 5275df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5276b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 527793515d53SJiri Pirko int err; 527893515d53SJiri Pirko 5279b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5280b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5281b6a79208SShmulik Ladkani offset)) { 5282b6a79208SShmulik Ladkani return -EINVAL; 5283b6a79208SShmulik Ladkani } 5284b6a79208SShmulik Ladkani 528593515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5286df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5287b6a79208SShmulik Ladkani if (err) 528893515d53SJiri Pirko return err; 52899241e2dfSDaniel Borkmann 529093515d53SJiri Pirko skb->protocol = skb->vlan_proto; 529193515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 529293515d53SJiri Pirko 52936b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 529493515d53SJiri Pirko } 529593515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 529693515d53SJiri Pirko return 0; 529793515d53SJiri Pirko } 529893515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 529993515d53SJiri Pirko 53002e4e4410SEric Dumazet /** 53012e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 53022e4e4410SEric Dumazet * 5303de3f0d0eSMasanari Iida * @header_len: size of linear part 5304de3f0d0eSMasanari Iida * @data_len: needed length in frags 5305de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5306de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5307de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 53082e4e4410SEric Dumazet * 53092e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 53102e4e4410SEric Dumazet */ 53112e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 53122e4e4410SEric Dumazet unsigned long data_len, 53132e4e4410SEric Dumazet int max_page_order, 53142e4e4410SEric Dumazet int *errcode, 53152e4e4410SEric Dumazet gfp_t gfp_mask) 53162e4e4410SEric Dumazet { 53172e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 53182e4e4410SEric Dumazet unsigned long chunk; 53192e4e4410SEric Dumazet struct sk_buff *skb; 53202e4e4410SEric Dumazet struct page *page; 53212e4e4410SEric Dumazet int i; 53222e4e4410SEric Dumazet 53232e4e4410SEric Dumazet *errcode = -EMSGSIZE; 53242e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 53252e4e4410SEric Dumazet * high order pages... 53262e4e4410SEric Dumazet */ 53272e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 53282e4e4410SEric Dumazet return NULL; 53292e4e4410SEric Dumazet 53302e4e4410SEric Dumazet *errcode = -ENOBUFS; 5331f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 53322e4e4410SEric Dumazet if (!skb) 53332e4e4410SEric Dumazet return NULL; 53342e4e4410SEric Dumazet 53352e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 53362e4e4410SEric Dumazet 53372e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 53382e4e4410SEric Dumazet int order = max_page_order; 53392e4e4410SEric Dumazet 53402e4e4410SEric Dumazet while (order) { 53412e4e4410SEric Dumazet if (npages >= 1 << order) { 5342d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 53432e4e4410SEric Dumazet __GFP_COMP | 5344d14b56f5SMichal Hocko __GFP_NOWARN, 53452e4e4410SEric Dumazet order); 53462e4e4410SEric Dumazet if (page) 53472e4e4410SEric Dumazet goto fill_page; 53482e4e4410SEric Dumazet /* Do not retry other high order allocations */ 53492e4e4410SEric Dumazet order = 1; 53502e4e4410SEric Dumazet max_page_order = 0; 53512e4e4410SEric Dumazet } 53522e4e4410SEric Dumazet order--; 53532e4e4410SEric Dumazet } 53542e4e4410SEric Dumazet page = alloc_page(gfp_mask); 53552e4e4410SEric Dumazet if (!page) 53562e4e4410SEric Dumazet goto failure; 53572e4e4410SEric Dumazet fill_page: 53582e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 53592e4e4410SEric Dumazet PAGE_SIZE << order); 53602e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 53612e4e4410SEric Dumazet data_len -= chunk; 53622e4e4410SEric Dumazet npages -= 1 << order; 53632e4e4410SEric Dumazet } 53642e4e4410SEric Dumazet return skb; 53652e4e4410SEric Dumazet 53662e4e4410SEric Dumazet failure: 53672e4e4410SEric Dumazet kfree_skb(skb); 53682e4e4410SEric Dumazet return NULL; 53692e4e4410SEric Dumazet } 53702e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 53716fa01ccdSSowmini Varadhan 53726fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 53736fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 53746fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 53756fa01ccdSSowmini Varadhan { 53766fa01ccdSSowmini Varadhan int i; 53776fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 53786fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 53796fa01ccdSSowmini Varadhan u8 *data; 53806fa01ccdSSowmini Varadhan 53816fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 53826fa01ccdSSowmini Varadhan 53836fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 53846fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 53856fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 53866fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 53876fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 53886fa01ccdSSowmini Varadhan if (!data) 53896fa01ccdSSowmini Varadhan return -ENOMEM; 53906fa01ccdSSowmini Varadhan 53916fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 53926fa01ccdSSowmini Varadhan 53936fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 53946fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 53956fa01ccdSSowmini Varadhan skb->len -= off; 53966fa01ccdSSowmini Varadhan 53976fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 53986fa01ccdSSowmini Varadhan skb_shinfo(skb), 53996fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 54006fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 54016fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 54026fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 54036fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 54046fa01ccdSSowmini Varadhan kfree(data); 54056fa01ccdSSowmini Varadhan return -ENOMEM; 54066fa01ccdSSowmini Varadhan } 54076fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 54086fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 54096fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 54106fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 54116fa01ccdSSowmini Varadhan skb_release_data(skb); 54126fa01ccdSSowmini Varadhan } else { 54136fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 54146fa01ccdSSowmini Varadhan * relocate values 54156fa01ccdSSowmini Varadhan */ 54166fa01ccdSSowmini Varadhan skb_free_head(skb); 54176fa01ccdSSowmini Varadhan } 54186fa01ccdSSowmini Varadhan 54196fa01ccdSSowmini Varadhan skb->head = data; 54206fa01ccdSSowmini Varadhan skb->data = data; 54216fa01ccdSSowmini Varadhan skb->head_frag = 0; 54226fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 54236fa01ccdSSowmini Varadhan skb->end = size; 54246fa01ccdSSowmini Varadhan #else 54256fa01ccdSSowmini Varadhan skb->end = skb->head + size; 54266fa01ccdSSowmini Varadhan #endif 54276fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 54286fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 54296fa01ccdSSowmini Varadhan skb->cloned = 0; 54306fa01ccdSSowmini Varadhan skb->hdr_len = 0; 54316fa01ccdSSowmini Varadhan skb->nohdr = 0; 54326fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 54336fa01ccdSSowmini Varadhan 54346fa01ccdSSowmini Varadhan return 0; 54356fa01ccdSSowmini Varadhan } 54366fa01ccdSSowmini Varadhan 54376fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 54386fa01ccdSSowmini Varadhan 54396fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 54406fa01ccdSSowmini Varadhan * pskb_carve() 54416fa01ccdSSowmini Varadhan */ 54426fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 54436fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 54446fa01ccdSSowmini Varadhan gfp_t gfp_mask) 54456fa01ccdSSowmini Varadhan { 54466fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 54476fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 54486fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 54496fa01ccdSSowmini Varadhan 54506fa01ccdSSowmini Varadhan do { 54516fa01ccdSSowmini Varadhan if (!list) { 54526fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 54536fa01ccdSSowmini Varadhan return -EFAULT; 54546fa01ccdSSowmini Varadhan } 54556fa01ccdSSowmini Varadhan if (list->len <= eat) { 54566fa01ccdSSowmini Varadhan /* Eaten as whole. */ 54576fa01ccdSSowmini Varadhan eat -= list->len; 54586fa01ccdSSowmini Varadhan list = list->next; 54596fa01ccdSSowmini Varadhan insp = list; 54606fa01ccdSSowmini Varadhan } else { 54616fa01ccdSSowmini Varadhan /* Eaten partially. */ 54626fa01ccdSSowmini Varadhan if (skb_shared(list)) { 54636fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 54646fa01ccdSSowmini Varadhan if (!clone) 54656fa01ccdSSowmini Varadhan return -ENOMEM; 54666fa01ccdSSowmini Varadhan insp = list->next; 54676fa01ccdSSowmini Varadhan list = clone; 54686fa01ccdSSowmini Varadhan } else { 54696fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 54706fa01ccdSSowmini Varadhan insp = list; 54716fa01ccdSSowmini Varadhan } 54726fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 54736fa01ccdSSowmini Varadhan kfree_skb(clone); 54746fa01ccdSSowmini Varadhan return -ENOMEM; 54756fa01ccdSSowmini Varadhan } 54766fa01ccdSSowmini Varadhan break; 54776fa01ccdSSowmini Varadhan } 54786fa01ccdSSowmini Varadhan } while (eat); 54796fa01ccdSSowmini Varadhan 54806fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 54816fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 54826fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 54836fa01ccdSSowmini Varadhan kfree_skb(list); 54846fa01ccdSSowmini Varadhan } 54856fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 54866fa01ccdSSowmini Varadhan if (clone) { 54876fa01ccdSSowmini Varadhan clone->next = list; 54886fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 54896fa01ccdSSowmini Varadhan } 54906fa01ccdSSowmini Varadhan return 0; 54916fa01ccdSSowmini Varadhan } 54926fa01ccdSSowmini Varadhan 54936fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 54946fa01ccdSSowmini Varadhan * non-linear part of skb 54956fa01ccdSSowmini Varadhan */ 54966fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 54976fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 54986fa01ccdSSowmini Varadhan { 54996fa01ccdSSowmini Varadhan int i, k = 0; 55006fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 55016fa01ccdSSowmini Varadhan u8 *data; 55026fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 55036fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 55046fa01ccdSSowmini Varadhan 55056fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 55066fa01ccdSSowmini Varadhan 55076fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 55086fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 55096fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 55106fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 55116fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 55126fa01ccdSSowmini Varadhan if (!data) 55136fa01ccdSSowmini Varadhan return -ENOMEM; 55146fa01ccdSSowmini Varadhan 55156fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 55166fa01ccdSSowmini Varadhan 55176fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 55186fa01ccdSSowmini Varadhan skb_shinfo(skb), offsetof(struct skb_shared_info, 55196fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 55206fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 55216fa01ccdSSowmini Varadhan kfree(data); 55226fa01ccdSSowmini Varadhan return -ENOMEM; 55236fa01ccdSSowmini Varadhan } 55246fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 55256fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 55266fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 55276fa01ccdSSowmini Varadhan 55286fa01ccdSSowmini Varadhan if (pos + fsize > off) { 55296fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 55306fa01ccdSSowmini Varadhan 55316fa01ccdSSowmini Varadhan if (pos < off) { 55326fa01ccdSSowmini Varadhan /* Split frag. 55336fa01ccdSSowmini Varadhan * We have two variants in this case: 55346fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 55356fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 55366fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 55376fa01ccdSSowmini Varadhan * where splitting is expensive. 55386fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 55396fa01ccdSSowmini Varadhan */ 55406fa01ccdSSowmini Varadhan shinfo->frags[0].page_offset += off - pos; 55416fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 55426fa01ccdSSowmini Varadhan } 55436fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 55446fa01ccdSSowmini Varadhan k++; 55456fa01ccdSSowmini Varadhan } 55466fa01ccdSSowmini Varadhan pos += fsize; 55476fa01ccdSSowmini Varadhan } 55486fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 55496fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 55506fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 55516fa01ccdSSowmini Varadhan 55526fa01ccdSSowmini Varadhan if (k == 0) { 55536fa01ccdSSowmini Varadhan /* split line is in frag list */ 55546fa01ccdSSowmini Varadhan pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask); 55556fa01ccdSSowmini Varadhan } 55566fa01ccdSSowmini Varadhan skb_release_data(skb); 55576fa01ccdSSowmini Varadhan 55586fa01ccdSSowmini Varadhan skb->head = data; 55596fa01ccdSSowmini Varadhan skb->head_frag = 0; 55606fa01ccdSSowmini Varadhan skb->data = data; 55616fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 55626fa01ccdSSowmini Varadhan skb->end = size; 55636fa01ccdSSowmini Varadhan #else 55646fa01ccdSSowmini Varadhan skb->end = skb->head + size; 55656fa01ccdSSowmini Varadhan #endif 55666fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 55676fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 55686fa01ccdSSowmini Varadhan skb->cloned = 0; 55696fa01ccdSSowmini Varadhan skb->hdr_len = 0; 55706fa01ccdSSowmini Varadhan skb->nohdr = 0; 55716fa01ccdSSowmini Varadhan skb->len -= off; 55726fa01ccdSSowmini Varadhan skb->data_len = skb->len; 55736fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 55746fa01ccdSSowmini Varadhan return 0; 55756fa01ccdSSowmini Varadhan } 55766fa01ccdSSowmini Varadhan 55776fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 55786fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 55796fa01ccdSSowmini Varadhan { 55806fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 55816fa01ccdSSowmini Varadhan 55826fa01ccdSSowmini Varadhan if (len < headlen) 55836fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 55846fa01ccdSSowmini Varadhan else 55856fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 55866fa01ccdSSowmini Varadhan } 55876fa01ccdSSowmini Varadhan 55886fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 55896fa01ccdSSowmini Varadhan * a new skb 55906fa01ccdSSowmini Varadhan */ 55916fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 55926fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 55936fa01ccdSSowmini Varadhan { 55946fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 55956fa01ccdSSowmini Varadhan 55966fa01ccdSSowmini Varadhan if (!clone) 55976fa01ccdSSowmini Varadhan return NULL; 55986fa01ccdSSowmini Varadhan 55996fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 56006fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 56016fa01ccdSSowmini Varadhan kfree_skb(clone); 56026fa01ccdSSowmini Varadhan return NULL; 56036fa01ccdSSowmini Varadhan } 56046fa01ccdSSowmini Varadhan return clone; 56056fa01ccdSSowmini Varadhan } 56066fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 5607c8c8b127SEric Dumazet 5608c8c8b127SEric Dumazet /** 5609c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 5610c8c8b127SEric Dumazet * @skb: buffer 5611c8c8b127SEric Dumazet * 5612c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 5613c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 5614c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 5615c8c8b127SEric Dumazet * truesize. 5616c8c8b127SEric Dumazet * Notes: 5617c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 5618c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 5619c8c8b127SEric Dumazet */ 5620c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 5621c8c8b127SEric Dumazet { 56223174fed9SEric Dumazet if (skb->data_len) { 56233174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 5624c8c8b127SEric Dumazet skb_cloned(skb)) 5625c8c8b127SEric Dumazet return; 5626c8c8b127SEric Dumazet 5627c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 5628c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 56293174fed9SEric Dumazet } 56303174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 56313174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 56323174fed9SEric Dumazet * their truesize. 56333174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 56343174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 56353174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 5636c8c8b127SEric Dumazet */ 5637c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 5638c8c8b127SEric Dumazet } 5639df5042f4SFlorian Westphal 5640df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 5641df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 5642df5042f4SFlorian Westphal { 5643df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 5644df5042f4SFlorian Westphal } 5645df5042f4SFlorian Westphal 5646df5042f4SFlorian Westphal static struct skb_ext *skb_ext_alloc(void) 5647df5042f4SFlorian Westphal { 5648df5042f4SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 5649df5042f4SFlorian Westphal 5650df5042f4SFlorian Westphal if (new) { 5651df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 5652df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 5653df5042f4SFlorian Westphal } 5654df5042f4SFlorian Westphal 5655df5042f4SFlorian Westphal return new; 5656df5042f4SFlorian Westphal } 5657df5042f4SFlorian Westphal 56584165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 56594165079bSFlorian Westphal unsigned int old_active) 5660df5042f4SFlorian Westphal { 5661df5042f4SFlorian Westphal struct skb_ext *new; 5662df5042f4SFlorian Westphal 5663df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 5664df5042f4SFlorian Westphal return old; 5665df5042f4SFlorian Westphal 5666df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 5667df5042f4SFlorian Westphal if (!new) 5668df5042f4SFlorian Westphal return NULL; 5669df5042f4SFlorian Westphal 5670df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 5671df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 5672df5042f4SFlorian Westphal 56734165079bSFlorian Westphal #ifdef CONFIG_XFRM 56744165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 56754165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 56764165079bSFlorian Westphal unsigned int i; 56774165079bSFlorian Westphal 56784165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 56794165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 56804165079bSFlorian Westphal } 56814165079bSFlorian Westphal #endif 5682df5042f4SFlorian Westphal __skb_ext_put(old); 5683df5042f4SFlorian Westphal return new; 5684df5042f4SFlorian Westphal } 5685df5042f4SFlorian Westphal 5686df5042f4SFlorian Westphal /** 5687df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 5688df5042f4SFlorian Westphal * @skb: buffer 5689df5042f4SFlorian Westphal * @id: extension to allocate space for 5690df5042f4SFlorian Westphal * 5691df5042f4SFlorian Westphal * Allocates enough space for the given extension. 5692df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 5693df5042f4SFlorian Westphal * is returned. 5694df5042f4SFlorian Westphal * 5695df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 5696df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 5697df5042f4SFlorian Westphal * 5698df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 5699df5042f4SFlorian Westphal */ 5700df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 5701df5042f4SFlorian Westphal { 5702df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 5703df5042f4SFlorian Westphal unsigned int newlen, newoff; 5704df5042f4SFlorian Westphal 5705df5042f4SFlorian Westphal if (skb->active_extensions) { 5706df5042f4SFlorian Westphal old = skb->extensions; 5707df5042f4SFlorian Westphal 57084165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 5709df5042f4SFlorian Westphal if (!new) 5710df5042f4SFlorian Westphal return NULL; 5711df5042f4SFlorian Westphal 5712682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 5713df5042f4SFlorian Westphal goto set_active; 5714df5042f4SFlorian Westphal 5715e94e50bdSPaolo Abeni newoff = new->chunks; 5716df5042f4SFlorian Westphal } else { 5717df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 5718df5042f4SFlorian Westphal 5719df5042f4SFlorian Westphal new = skb_ext_alloc(); 5720df5042f4SFlorian Westphal if (!new) 5721df5042f4SFlorian Westphal return NULL; 5722df5042f4SFlorian Westphal } 5723df5042f4SFlorian Westphal 5724df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 5725df5042f4SFlorian Westphal new->chunks = newlen; 5726df5042f4SFlorian Westphal new->offset[id] = newoff; 5727df5042f4SFlorian Westphal set_active: 5728682ec859SPaolo Abeni skb->extensions = new; 5729df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 5730df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 5731df5042f4SFlorian Westphal } 5732df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 5733df5042f4SFlorian Westphal 57344165079bSFlorian Westphal #ifdef CONFIG_XFRM 57354165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 57364165079bSFlorian Westphal { 57374165079bSFlorian Westphal unsigned int i; 57384165079bSFlorian Westphal 57394165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 57404165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 57414165079bSFlorian Westphal } 57424165079bSFlorian Westphal #endif 57434165079bSFlorian Westphal 5744df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 5745df5042f4SFlorian Westphal { 5746df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 5747df5042f4SFlorian Westphal 5748df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 5749df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 5750df5042f4SFlorian Westphal skb->extensions = NULL; 5751df5042f4SFlorian Westphal __skb_ext_put(ext); 57524165079bSFlorian Westphal #ifdef CONFIG_XFRM 57534165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 57544165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 57554165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 57564165079bSFlorian Westphal 57574165079bSFlorian Westphal skb_ext_put_sp(sp); 57584165079bSFlorian Westphal sp->len = 0; 57594165079bSFlorian Westphal #endif 5760df5042f4SFlorian Westphal } 5761df5042f4SFlorian Westphal } 5762df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 5763df5042f4SFlorian Westphal 5764df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 5765df5042f4SFlorian Westphal { 5766df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 5767df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 5768df5042f4SFlorian Westphal */ 5769df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 5770df5042f4SFlorian Westphal goto free_now; 5771df5042f4SFlorian Westphal 5772df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 5773df5042f4SFlorian Westphal return; 5774df5042f4SFlorian Westphal free_now: 57754165079bSFlorian Westphal #ifdef CONFIG_XFRM 57764165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 57774165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 57784165079bSFlorian Westphal #endif 57794165079bSFlorian Westphal 5780df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 5781df5042f4SFlorian Westphal } 5782df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 5783df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 5784