12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 41da177e4SLinus Torvalds * 5113aa838SAlan Cox * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 61da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Fixes: 91da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 101da177e4SLinus Torvalds * balancer bugs. 111da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 121da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 131da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 141da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 151da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 161da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 171da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 181da177e4SLinus Torvalds * only put in the headers 191da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 201da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 211da177e4SLinus Torvalds * Andi Kleen : slabified it. 221da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 231da177e4SLinus Torvalds * 241da177e4SLinus Torvalds * NOTE: 251da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 261da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 271da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 281da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 331da177e4SLinus Torvalds */ 341da177e4SLinus Torvalds 35e005d193SJoe Perches #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 36e005d193SJoe Perches 371da177e4SLinus Torvalds #include <linux/module.h> 381da177e4SLinus Torvalds #include <linux/types.h> 391da177e4SLinus Torvalds #include <linux/kernel.h> 401da177e4SLinus Torvalds #include <linux/mm.h> 411da177e4SLinus Torvalds #include <linux/interrupt.h> 421da177e4SLinus Torvalds #include <linux/in.h> 431da177e4SLinus Torvalds #include <linux/inet.h> 441da177e4SLinus Torvalds #include <linux/slab.h> 45de960aa9SFlorian Westphal #include <linux/tcp.h> 46de960aa9SFlorian Westphal #include <linux/udp.h> 4790017accSMarcelo Ricardo Leitner #include <linux/sctp.h> 481da177e4SLinus Torvalds #include <linux/netdevice.h> 491da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 501da177e4SLinus Torvalds #include <net/pkt_sched.h> 511da177e4SLinus Torvalds #endif 521da177e4SLinus Torvalds #include <linux/string.h> 531da177e4SLinus Torvalds #include <linux/skbuff.h> 549c55e01cSJens Axboe #include <linux/splice.h> 551da177e4SLinus Torvalds #include <linux/cache.h> 561da177e4SLinus Torvalds #include <linux/rtnetlink.h> 571da177e4SLinus Torvalds #include <linux/init.h> 58716ea3a7SDavid Howells #include <linux/scatterlist.h> 59ac45f602SPatrick Ohly #include <linux/errqueue.h> 60268bb0ceSLinus Torvalds #include <linux/prefetch.h> 610d5501c1SVlad Yasevich #include <linux/if_vlan.h> 622a2ea508SJohn Hurley #include <linux/mpls.h> 631da177e4SLinus Torvalds 641da177e4SLinus Torvalds #include <net/protocol.h> 651da177e4SLinus Torvalds #include <net/dst.h> 661da177e4SLinus Torvalds #include <net/sock.h> 671da177e4SLinus Torvalds #include <net/checksum.h> 68ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 691da177e4SLinus Torvalds #include <net/xfrm.h> 708822e270SJohn Hurley #include <net/mpls.h> 711da177e4SLinus Torvalds 727c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 73ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7451c56b00SEric Dumazet #include <linux/highmem.h> 75b245be1fSWillem de Bruijn #include <linux/capability.h> 76b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 772544af03SMatteo Croce #include <linux/indirect_call_wrapper.h> 78a1f8e7f7SAl Viro 797b7ed885SBart Van Assche #include "datagram.h" 807b7ed885SBart Van Assche 8108009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8208009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 83df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 84df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 85df5042f4SFlorian Westphal #endif 865f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 875f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 881da177e4SLinus Torvalds 891da177e4SLinus Torvalds /** 90f05de73bSJean Sacren * skb_panic - private function for out-of-line support 911da177e4SLinus Torvalds * @skb: buffer 921da177e4SLinus Torvalds * @sz: size 93f05de73bSJean Sacren * @addr: address 9499d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 951da177e4SLinus Torvalds * 96f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 97f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 98f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 99f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1001da177e4SLinus Torvalds */ 101f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 10299d5851eSJames Hogan const char msg[]) 1031da177e4SLinus Torvalds { 104e005d193SJoe Perches pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 10599d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1064305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 10726095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1081da177e4SLinus Torvalds BUG(); 1091da177e4SLinus Torvalds } 1101da177e4SLinus Torvalds 111f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1121da177e4SLinus Torvalds { 113f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1141da177e4SLinus Torvalds } 1151da177e4SLinus Torvalds 116f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 117f05de73bSJean Sacren { 118f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 119f05de73bSJean Sacren } 120c93bdd0eSMel Gorman 121c93bdd0eSMel Gorman /* 122c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 123c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 124c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 125c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 126c93bdd0eSMel Gorman * memory is free 127c93bdd0eSMel Gorman */ 128c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 129c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 13061c5e88aSstephen hemminger 13161c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 13261c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 133c93bdd0eSMel Gorman { 134c93bdd0eSMel Gorman void *obj; 135c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 136c93bdd0eSMel Gorman 137c93bdd0eSMel Gorman /* 138c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 139c93bdd0eSMel Gorman * to the reserves, fail. 140c93bdd0eSMel Gorman */ 141c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 142c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 143c93bdd0eSMel Gorman node); 144c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 145c93bdd0eSMel Gorman goto out; 146c93bdd0eSMel Gorman 147c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 148c93bdd0eSMel Gorman ret_pfmemalloc = true; 149c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 150c93bdd0eSMel Gorman 151c93bdd0eSMel Gorman out: 152c93bdd0eSMel Gorman if (pfmemalloc) 153c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 154c93bdd0eSMel Gorman 155c93bdd0eSMel Gorman return obj; 156c93bdd0eSMel Gorman } 157c93bdd0eSMel Gorman 1581da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1591da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1601da177e4SLinus Torvalds * [BEEP] leaks. 1611da177e4SLinus Torvalds * 1621da177e4SLinus Torvalds */ 1631da177e4SLinus Torvalds 1641da177e4SLinus Torvalds /** 165d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1661da177e4SLinus Torvalds * @size: size to allocate 1671da177e4SLinus Torvalds * @gfp_mask: allocation mask 168c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 169c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 170c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 171c93bdd0eSMel Gorman * allocations in case the data is required for writeback 172b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1731da177e4SLinus Torvalds * 1741da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 17594b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 17694b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 1771da177e4SLinus Torvalds * 1781da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1791da177e4SLinus Torvalds * %GFP_ATOMIC. 1801da177e4SLinus Torvalds */ 181dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 182c93bdd0eSMel Gorman int flags, int node) 1831da177e4SLinus Torvalds { 184e18b890bSChristoph Lameter struct kmem_cache *cache; 1854947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1861da177e4SLinus Torvalds struct sk_buff *skb; 1871da177e4SLinus Torvalds u8 *data; 188c93bdd0eSMel Gorman bool pfmemalloc; 1891da177e4SLinus Torvalds 190c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 191c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 192c93bdd0eSMel Gorman 193c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 194c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1958798b3fbSHerbert Xu 1961da177e4SLinus Torvalds /* Get the HEAD */ 197b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1981da177e4SLinus Torvalds if (!skb) 1991da177e4SLinus Torvalds goto out; 200ec7d2f2cSEric Dumazet prefetchw(skb); 2011da177e4SLinus Torvalds 20287fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 20387fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 20487fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 20587fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 20687fb4b7bSEric Dumazet */ 207bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 20887fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 209c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2101da177e4SLinus Torvalds if (!data) 2111da177e4SLinus Torvalds goto nodata; 21287fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 21387fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 21487fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 21587fb4b7bSEric Dumazet */ 21687fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 217ec7d2f2cSEric Dumazet prefetchw(data + size); 2181da177e4SLinus Torvalds 219ca0605a7SArnaldo Carvalho de Melo /* 220c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 221c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 222c8005785SJohannes Berg * the tail pointer in struct sk_buff! 223ca0605a7SArnaldo Carvalho de Melo */ 224ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 22587fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 22687fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 227c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 22863354797SReshetova, Elena refcount_set(&skb->users, 1); 2291da177e4SLinus Torvalds skb->head = data; 2301da177e4SLinus Torvalds skb->data = data; 23127a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2324305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 23335d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 23435d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 23519633e12SStephen Hemminger 2364947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2374947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 238ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2394947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 2404947d3efSBenjamin LaHaise 241c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 242d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 2431da177e4SLinus Torvalds 244d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 245d0bf4a9eSEric Dumazet 246d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 2472638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 1); 248d179cd12SDavid S. Miller 2496ffe75ebSEric Dumazet fclones->skb2.fclone = SKB_FCLONE_CLONE; 250d179cd12SDavid S. Miller } 2511da177e4SLinus Torvalds out: 2521da177e4SLinus Torvalds return skb; 2531da177e4SLinus Torvalds nodata: 2548798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2551da177e4SLinus Torvalds skb = NULL; 2561da177e4SLinus Torvalds goto out; 2571da177e4SLinus Torvalds } 258b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2591da177e4SLinus Torvalds 260ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 261ba0509b6SJesper Dangaard Brouer static struct sk_buff *__build_skb_around(struct sk_buff *skb, 262ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 263ba0509b6SJesper Dangaard Brouer { 264ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 265ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 266ba0509b6SJesper Dangaard Brouer 267ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 268ba0509b6SJesper Dangaard Brouer 269ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 270ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 271ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 272ba0509b6SJesper Dangaard Brouer skb->head = data; 273ba0509b6SJesper Dangaard Brouer skb->data = data; 274ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 275ba0509b6SJesper Dangaard Brouer skb->end = skb->tail + size; 276ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 277ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 278ba0509b6SJesper Dangaard Brouer 279ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 280ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 281ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 282ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 283ba0509b6SJesper Dangaard Brouer 284ba0509b6SJesper Dangaard Brouer return skb; 285ba0509b6SJesper Dangaard Brouer } 286ba0509b6SJesper Dangaard Brouer 2871da177e4SLinus Torvalds /** 2882ea2f62cSEric Dumazet * __build_skb - build a network buffer 289b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2902ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 291b2b5ce9dSEric Dumazet * 292b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 293deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2942ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2952ea2f62cSEric Dumazet * or vmalloc() 296b2b5ce9dSEric Dumazet * The return is the new skb buffer. 297b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 298b2b5ce9dSEric Dumazet * Notes : 299b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 300b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 301b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 302b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 303b2b5ce9dSEric Dumazet * before giving packet to stack. 304b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 305b2b5ce9dSEric Dumazet */ 3062ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 307b2b5ce9dSEric Dumazet { 308b2b5ce9dSEric Dumazet struct sk_buff *skb; 309b2b5ce9dSEric Dumazet 310b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 311ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 312b2b5ce9dSEric Dumazet return NULL; 313b2b5ce9dSEric Dumazet 314b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 315b2b5ce9dSEric Dumazet 316ba0509b6SJesper Dangaard Brouer return __build_skb_around(skb, data, frag_size); 317b2b5ce9dSEric Dumazet } 3182ea2f62cSEric Dumazet 3192ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3202ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3212ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 3222ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 3232ea2f62cSEric Dumazet */ 3242ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3252ea2f62cSEric Dumazet { 3262ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3272ea2f62cSEric Dumazet 3282ea2f62cSEric Dumazet if (skb && frag_size) { 3292ea2f62cSEric Dumazet skb->head_frag = 1; 3302f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 3312ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3322ea2f62cSEric Dumazet } 3332ea2f62cSEric Dumazet return skb; 3342ea2f62cSEric Dumazet } 335b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 336b2b5ce9dSEric Dumazet 337ba0509b6SJesper Dangaard Brouer /** 338ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 339ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 340ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 341ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 342ba0509b6SJesper Dangaard Brouer */ 343ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 344ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 345ba0509b6SJesper Dangaard Brouer { 346ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 347ba0509b6SJesper Dangaard Brouer return NULL; 348ba0509b6SJesper Dangaard Brouer 349ba0509b6SJesper Dangaard Brouer skb = __build_skb_around(skb, data, frag_size); 350ba0509b6SJesper Dangaard Brouer 351ba0509b6SJesper Dangaard Brouer if (skb && frag_size) { 352ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 353ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 354ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 355ba0509b6SJesper Dangaard Brouer } 356ba0509b6SJesper Dangaard Brouer return skb; 357ba0509b6SJesper Dangaard Brouer } 358ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 359ba0509b6SJesper Dangaard Brouer 360795bb1c0SJesper Dangaard Brouer #define NAPI_SKB_CACHE_SIZE 64 361795bb1c0SJesper Dangaard Brouer 362795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache { 363795bb1c0SJesper Dangaard Brouer struct page_frag_cache page; 364e0d7924aSAlexey Dobriyan unsigned int skb_count; 365795bb1c0SJesper Dangaard Brouer void *skb_cache[NAPI_SKB_CACHE_SIZE]; 366795bb1c0SJesper Dangaard Brouer }; 367795bb1c0SJesper Dangaard Brouer 368b63ae8caSAlexander Duyck static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 369795bb1c0SJesper Dangaard Brouer static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 370ffde7328SAlexander Duyck 371ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 372ffde7328SAlexander Duyck { 373795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 3749451980aSAlexander Duyck 3758c2dd3e4SAlexander Duyck return page_frag_alloc(&nc->page, fragsz, gfp_mask); 376ffde7328SAlexander Duyck } 377ffde7328SAlexander Duyck 378ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 379ffde7328SAlexander Duyck { 3803bed3cc4SAlexander Duyck fragsz = SKB_DATA_ALIGN(fragsz); 3813bed3cc4SAlexander Duyck 382453f85d4SMel Gorman return __napi_alloc_frag(fragsz, GFP_ATOMIC); 383ffde7328SAlexander Duyck } 384ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 385ffde7328SAlexander Duyck 3866f532612SEric Dumazet /** 3877ba7aeabSSebastian Andrzej Siewior * netdev_alloc_frag - allocate a page fragment 3887ba7aeabSSebastian Andrzej Siewior * @fragsz: fragment size 3897ba7aeabSSebastian Andrzej Siewior * 3907ba7aeabSSebastian Andrzej Siewior * Allocates a frag from a page for receive buffer. 3917ba7aeabSSebastian Andrzej Siewior * Uses GFP_ATOMIC allocations. 3927ba7aeabSSebastian Andrzej Siewior */ 3937ba7aeabSSebastian Andrzej Siewior void *netdev_alloc_frag(unsigned int fragsz) 3947ba7aeabSSebastian Andrzej Siewior { 3957ba7aeabSSebastian Andrzej Siewior struct page_frag_cache *nc; 3967ba7aeabSSebastian Andrzej Siewior void *data; 3977ba7aeabSSebastian Andrzej Siewior 3987ba7aeabSSebastian Andrzej Siewior fragsz = SKB_DATA_ALIGN(fragsz); 3997ba7aeabSSebastian Andrzej Siewior if (in_irq() || irqs_disabled()) { 4007ba7aeabSSebastian Andrzej Siewior nc = this_cpu_ptr(&netdev_alloc_cache); 4017ba7aeabSSebastian Andrzej Siewior data = page_frag_alloc(nc, fragsz, GFP_ATOMIC); 4027ba7aeabSSebastian Andrzej Siewior } else { 4037ba7aeabSSebastian Andrzej Siewior local_bh_disable(); 4047ba7aeabSSebastian Andrzej Siewior data = __napi_alloc_frag(fragsz, GFP_ATOMIC); 4057ba7aeabSSebastian Andrzej Siewior local_bh_enable(); 4067ba7aeabSSebastian Andrzej Siewior } 4077ba7aeabSSebastian Andrzej Siewior return data; 4087ba7aeabSSebastian Andrzej Siewior } 4097ba7aeabSSebastian Andrzej Siewior EXPORT_SYMBOL(netdev_alloc_frag); 4107ba7aeabSSebastian Andrzej Siewior 4117ba7aeabSSebastian Andrzej Siewior /** 412fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 413fd11a83dSAlexander Duyck * @dev: network device to receive on 414d7499160SMasanari Iida * @len: length to allocate 415fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 416fd11a83dSAlexander Duyck * 417fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 418fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 419fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 420fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 421fd11a83dSAlexander Duyck * 422fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 423fd11a83dSAlexander Duyck */ 4249451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4259451980aSAlexander Duyck gfp_t gfp_mask) 426fd11a83dSAlexander Duyck { 427b63ae8caSAlexander Duyck struct page_frag_cache *nc; 428fd11a83dSAlexander Duyck struct sk_buff *skb; 4299451980aSAlexander Duyck bool pfmemalloc; 4309451980aSAlexander Duyck void *data; 431fd11a83dSAlexander Duyck 4329451980aSAlexander Duyck len += NET_SKB_PAD; 433fd11a83dSAlexander Duyck 4349451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 435d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 436a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 437a080e7bdSAlexander Duyck if (!skb) 438a080e7bdSAlexander Duyck goto skb_fail; 439a080e7bdSAlexander Duyck goto skb_success; 440a080e7bdSAlexander Duyck } 4419451980aSAlexander Duyck 4429451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4439451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 4449451980aSAlexander Duyck 4459451980aSAlexander Duyck if (sk_memalloc_socks()) 4469451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 4479451980aSAlexander Duyck 44892dcabd7SSebastian Andrzej Siewior if (in_irq() || irqs_disabled()) { 4499451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 4508c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 4519451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 45292dcabd7SSebastian Andrzej Siewior } else { 45392dcabd7SSebastian Andrzej Siewior local_bh_disable(); 45492dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 45592dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 45692dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 45792dcabd7SSebastian Andrzej Siewior local_bh_enable(); 45892dcabd7SSebastian Andrzej Siewior } 4599451980aSAlexander Duyck 4609451980aSAlexander Duyck if (unlikely(!data)) 4619451980aSAlexander Duyck return NULL; 4629451980aSAlexander Duyck 4639451980aSAlexander Duyck skb = __build_skb(data, len); 4649451980aSAlexander Duyck if (unlikely(!skb)) { 465181edb2bSAlexander Duyck skb_free_frag(data); 4669451980aSAlexander Duyck return NULL; 4679451980aSAlexander Duyck } 4689451980aSAlexander Duyck 4699451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 4709451980aSAlexander Duyck if (pfmemalloc) 4719451980aSAlexander Duyck skb->pfmemalloc = 1; 4729451980aSAlexander Duyck skb->head_frag = 1; 4739451980aSAlexander Duyck 474a080e7bdSAlexander Duyck skb_success: 4758af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4767b2e497aSChristoph Hellwig skb->dev = dev; 477fd11a83dSAlexander Duyck 478a080e7bdSAlexander Duyck skb_fail: 4798af27456SChristoph Hellwig return skb; 4808af27456SChristoph Hellwig } 481b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4821da177e4SLinus Torvalds 483fd11a83dSAlexander Duyck /** 484fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 485fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 486d7499160SMasanari Iida * @len: length to allocate 487fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 488fd11a83dSAlexander Duyck * 489fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 490fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 491fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 492fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 493fd11a83dSAlexander Duyck * 494fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 495fd11a83dSAlexander Duyck */ 4969451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 4979451980aSAlexander Duyck gfp_t gfp_mask) 498fd11a83dSAlexander Duyck { 499795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 500fd11a83dSAlexander Duyck struct sk_buff *skb; 5019451980aSAlexander Duyck void *data; 502fd11a83dSAlexander Duyck 5039451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 504fd11a83dSAlexander Duyck 5059451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 506d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 507a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 508a080e7bdSAlexander Duyck if (!skb) 509a080e7bdSAlexander Duyck goto skb_fail; 510a080e7bdSAlexander Duyck goto skb_success; 511a080e7bdSAlexander Duyck } 5129451980aSAlexander Duyck 5139451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5149451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5159451980aSAlexander Duyck 5169451980aSAlexander Duyck if (sk_memalloc_socks()) 5179451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5189451980aSAlexander Duyck 5198c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5209451980aSAlexander Duyck if (unlikely(!data)) 5219451980aSAlexander Duyck return NULL; 5229451980aSAlexander Duyck 5239451980aSAlexander Duyck skb = __build_skb(data, len); 5249451980aSAlexander Duyck if (unlikely(!skb)) { 525181edb2bSAlexander Duyck skb_free_frag(data); 5269451980aSAlexander Duyck return NULL; 5279451980aSAlexander Duyck } 5289451980aSAlexander Duyck 5299451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 530795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5319451980aSAlexander Duyck skb->pfmemalloc = 1; 5329451980aSAlexander Duyck skb->head_frag = 1; 5339451980aSAlexander Duyck 534a080e7bdSAlexander Duyck skb_success: 535fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 536fd11a83dSAlexander Duyck skb->dev = napi->dev; 537fd11a83dSAlexander Duyck 538a080e7bdSAlexander Duyck skb_fail: 539fd11a83dSAlexander Duyck return skb; 540fd11a83dSAlexander Duyck } 541fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 542fd11a83dSAlexander Duyck 543654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 54450269e19SEric Dumazet int size, unsigned int truesize) 545654bed16SPeter Zijlstra { 546654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 547654bed16SPeter Zijlstra skb->len += size; 548654bed16SPeter Zijlstra skb->data_len += size; 54950269e19SEric Dumazet skb->truesize += truesize; 550654bed16SPeter Zijlstra } 551654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 552654bed16SPeter Zijlstra 553f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 554f8e617e1SJason Wang unsigned int truesize) 555f8e617e1SJason Wang { 556f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 557f8e617e1SJason Wang 558f8e617e1SJason Wang skb_frag_size_add(frag, size); 559f8e617e1SJason Wang skb->len += size; 560f8e617e1SJason Wang skb->data_len += size; 561f8e617e1SJason Wang skb->truesize += truesize; 562f8e617e1SJason Wang } 563f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 564f8e617e1SJason Wang 56527b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5661da177e4SLinus Torvalds { 567bd8a7036SEric Dumazet kfree_skb_list(*listp); 56827b437c8SHerbert Xu *listp = NULL; 5691da177e4SLinus Torvalds } 5701da177e4SLinus Torvalds 57127b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 57227b437c8SHerbert Xu { 57327b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 57427b437c8SHerbert Xu } 57527b437c8SHerbert Xu 5761da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5771da177e4SLinus Torvalds { 5781da177e4SLinus Torvalds struct sk_buff *list; 5791da177e4SLinus Torvalds 580fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5811da177e4SLinus Torvalds skb_get(list); 5821da177e4SLinus Torvalds } 5831da177e4SLinus Torvalds 584d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 585d3836f21SEric Dumazet { 586181edb2bSAlexander Duyck unsigned char *head = skb->head; 587181edb2bSAlexander Duyck 588d3836f21SEric Dumazet if (skb->head_frag) 589181edb2bSAlexander Duyck skb_free_frag(head); 590d3836f21SEric Dumazet else 591181edb2bSAlexander Duyck kfree(head); 592d3836f21SEric Dumazet } 593d3836f21SEric Dumazet 5945bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5951da177e4SLinus Torvalds { 596ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 5971da177e4SLinus Torvalds int i; 598ff04a771SEric Dumazet 599ff04a771SEric Dumazet if (skb->cloned && 600ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 601ff04a771SEric Dumazet &shinfo->dataref)) 602ff04a771SEric Dumazet return; 603ff04a771SEric Dumazet 604ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 605ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 6061da177e4SLinus Torvalds 607ff04a771SEric Dumazet if (shinfo->frag_list) 608ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6091da177e4SLinus Torvalds 6101f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 611d3836f21SEric Dumazet skb_free_head(skb); 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds 6141da177e4SLinus Torvalds /* 6151da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6161da177e4SLinus Torvalds */ 6172d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6181da177e4SLinus Torvalds { 619d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 620d179cd12SDavid S. Miller 621d179cd12SDavid S. Miller switch (skb->fclone) { 622d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6231da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6246ffe75ebSEric Dumazet return; 625d179cd12SDavid S. Miller 626d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 627d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6286ffe75ebSEric Dumazet 6296ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6306ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6316ffe75ebSEric Dumazet * while here, branch prediction will be good. 6326ffe75ebSEric Dumazet */ 6332638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 6346ffe75ebSEric Dumazet goto fastpath; 635d179cd12SDavid S. Miller break; 636d179cd12SDavid S. Miller 6376ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 638d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 639d179cd12SDavid S. Miller break; 6403ff50b79SStephen Hemminger } 6412638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 6426ffe75ebSEric Dumazet return; 6436ffe75ebSEric Dumazet fastpath: 6446ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6451da177e4SLinus Torvalds } 6461da177e4SLinus Torvalds 6470a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 6481da177e4SLinus Torvalds { 649adf30907SEric Dumazet skb_dst_drop(skb); 6501da177e4SLinus Torvalds if (skb->destructor) { 6519c2b3328SStephen Hemminger WARN_ON(in_irq()); 6521da177e4SLinus Torvalds skb->destructor(skb); 6531da177e4SLinus Torvalds } 654a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 655cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 6562fc72c7bSKOVACS Krisztian #endif 657df5042f4SFlorian Westphal skb_ext_put(skb); 65804a4bb55SLennert Buytenhek } 65904a4bb55SLennert Buytenhek 66004a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 66104a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 66204a4bb55SLennert Buytenhek { 66304a4bb55SLennert Buytenhek skb_release_head_state(skb); 664a28b1b90SFlorian Westphal if (likely(skb->head)) 6652d4baff8SHerbert Xu skb_release_data(skb); 6662d4baff8SHerbert Xu } 6671da177e4SLinus Torvalds 6682d4baff8SHerbert Xu /** 6692d4baff8SHerbert Xu * __kfree_skb - private function 6702d4baff8SHerbert Xu * @skb: buffer 6712d4baff8SHerbert Xu * 6722d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6732d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6742d4baff8SHerbert Xu * always call kfree_skb 6752d4baff8SHerbert Xu */ 6762d4baff8SHerbert Xu 6772d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6782d4baff8SHerbert Xu { 6792d4baff8SHerbert Xu skb_release_all(skb); 6801da177e4SLinus Torvalds kfree_skbmem(skb); 6811da177e4SLinus Torvalds } 682b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6831da177e4SLinus Torvalds 6841da177e4SLinus Torvalds /** 685231d06aeSJörn Engel * kfree_skb - free an sk_buff 686231d06aeSJörn Engel * @skb: buffer to free 687231d06aeSJörn Engel * 688231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 689231d06aeSJörn Engel * hit zero. 690231d06aeSJörn Engel */ 691231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 692231d06aeSJörn Engel { 6933889a803SPaolo Abeni if (!skb_unref(skb)) 694231d06aeSJörn Engel return; 6953889a803SPaolo Abeni 696ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 697231d06aeSJörn Engel __kfree_skb(skb); 698231d06aeSJörn Engel } 699b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 700231d06aeSJörn Engel 701bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 702bd8a7036SEric Dumazet { 703bd8a7036SEric Dumazet while (segs) { 704bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 705bd8a7036SEric Dumazet 706bd8a7036SEric Dumazet kfree_skb(segs); 707bd8a7036SEric Dumazet segs = next; 708bd8a7036SEric Dumazet } 709bd8a7036SEric Dumazet } 710bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 711bd8a7036SEric Dumazet 7126413139dSWillem de Bruijn /* Dump skb information and contents. 7136413139dSWillem de Bruijn * 7146413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 7156413139dSWillem de Bruijn * 7166413139dSWillem de Bruijn * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise. 7176413139dSWillem de Bruijn */ 7186413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 7196413139dSWillem de Bruijn { 7206413139dSWillem de Bruijn static atomic_t can_dump_full = ATOMIC_INIT(5); 7216413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 7226413139dSWillem de Bruijn struct net_device *dev = skb->dev; 7236413139dSWillem de Bruijn struct sock *sk = skb->sk; 7246413139dSWillem de Bruijn struct sk_buff *list_skb; 7256413139dSWillem de Bruijn bool has_mac, has_trans; 7266413139dSWillem de Bruijn int headroom, tailroom; 7276413139dSWillem de Bruijn int i, len, seg_len; 7286413139dSWillem de Bruijn 7296413139dSWillem de Bruijn if (full_pkt) 7306413139dSWillem de Bruijn full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0; 7316413139dSWillem de Bruijn 7326413139dSWillem de Bruijn if (full_pkt) 7336413139dSWillem de Bruijn len = skb->len; 7346413139dSWillem de Bruijn else 7356413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 7366413139dSWillem de Bruijn 7376413139dSWillem de Bruijn headroom = skb_headroom(skb); 7386413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 7396413139dSWillem de Bruijn 7406413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 7416413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 7426413139dSWillem de Bruijn 7436413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 7446413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 7456413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 7466413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 7476413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 7486413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 7496413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 7506413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 7516413139dSWillem de Bruijn skb->network_header, 7526413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 7536413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 7546413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 7556413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 7566413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 7576413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 7586413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 7596413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 7606413139dSWillem de Bruijn 7616413139dSWillem de Bruijn if (dev) 7626413139dSWillem de Bruijn printk("%sdev name=%s feat=0x%pNF\n", 7636413139dSWillem de Bruijn level, dev->name, &dev->features); 7646413139dSWillem de Bruijn if (sk) 765db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 7666413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 7676413139dSWillem de Bruijn 7686413139dSWillem de Bruijn if (full_pkt && headroom) 7696413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 7706413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 7716413139dSWillem de Bruijn 7726413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 7736413139dSWillem de Bruijn if (seg_len) 7746413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 7756413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 7766413139dSWillem de Bruijn len -= seg_len; 7776413139dSWillem de Bruijn 7786413139dSWillem de Bruijn if (full_pkt && tailroom) 7796413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 7806413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 7816413139dSWillem de Bruijn 7826413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 7836413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 7846413139dSWillem de Bruijn u32 p_off, p_len, copied; 7856413139dSWillem de Bruijn struct page *p; 7866413139dSWillem de Bruijn u8 *vaddr; 7876413139dSWillem de Bruijn 788b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 7896413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 7906413139dSWillem de Bruijn copied) { 7916413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 7926413139dSWillem de Bruijn vaddr = kmap_atomic(p); 7936413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 7946413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 7956413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 7966413139dSWillem de Bruijn kunmap_atomic(vaddr); 7976413139dSWillem de Bruijn len -= seg_len; 7986413139dSWillem de Bruijn if (!len) 7996413139dSWillem de Bruijn break; 8006413139dSWillem de Bruijn } 8016413139dSWillem de Bruijn } 8026413139dSWillem de Bruijn 8036413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8046413139dSWillem de Bruijn printk("skb fraglist:\n"); 8056413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8066413139dSWillem de Bruijn skb_dump(level, list_skb, true); 8076413139dSWillem de Bruijn } 8086413139dSWillem de Bruijn } 8096413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 8106413139dSWillem de Bruijn 811d1a203eaSStephen Hemminger /** 81225121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 81325121173SMichael S. Tsirkin * @skb: buffer that triggered an error 81425121173SMichael S. Tsirkin * 81525121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 81625121173SMichael S. Tsirkin * skb must be freed afterwards. 81725121173SMichael S. Tsirkin */ 81825121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 81925121173SMichael S. Tsirkin { 8201f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 82125121173SMichael S. Tsirkin } 82225121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 82325121173SMichael S. Tsirkin 82425121173SMichael S. Tsirkin /** 825ead2ceb0SNeil Horman * consume_skb - free an skbuff 826ead2ceb0SNeil Horman * @skb: buffer to free 827ead2ceb0SNeil Horman * 828ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 829ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 830ead2ceb0SNeil Horman * is being dropped after a failure and notes that 831ead2ceb0SNeil Horman */ 832ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 833ead2ceb0SNeil Horman { 8343889a803SPaolo Abeni if (!skb_unref(skb)) 835ead2ceb0SNeil Horman return; 8363889a803SPaolo Abeni 83707dc22e7SKoki Sanagi trace_consume_skb(skb); 838ead2ceb0SNeil Horman __kfree_skb(skb); 839ead2ceb0SNeil Horman } 840ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 841ead2ceb0SNeil Horman 8420a463c78SPaolo Abeni /** 8430a463c78SPaolo Abeni * consume_stateless_skb - free an skbuff, assuming it is stateless 8440a463c78SPaolo Abeni * @skb: buffer to free 8450a463c78SPaolo Abeni * 846ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 847ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 8480a463c78SPaolo Abeni */ 849ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 8500a463c78SPaolo Abeni { 8510a463c78SPaolo Abeni trace_consume_skb(skb); 8520a463c78SPaolo Abeni skb_release_data(skb); 8530a463c78SPaolo Abeni kfree_skbmem(skb); 8540a463c78SPaolo Abeni } 8550a463c78SPaolo Abeni 856795bb1c0SJesper Dangaard Brouer void __kfree_skb_flush(void) 857795bb1c0SJesper Dangaard Brouer { 858795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 859795bb1c0SJesper Dangaard Brouer 860795bb1c0SJesper Dangaard Brouer /* flush skb_cache if containing objects */ 861795bb1c0SJesper Dangaard Brouer if (nc->skb_count) { 862795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 863795bb1c0SJesper Dangaard Brouer nc->skb_cache); 864795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 865795bb1c0SJesper Dangaard Brouer } 866795bb1c0SJesper Dangaard Brouer } 867795bb1c0SJesper Dangaard Brouer 86815fad714SJesper Dangaard Brouer static inline void _kfree_skb_defer(struct sk_buff *skb) 869795bb1c0SJesper Dangaard Brouer { 870795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 871795bb1c0SJesper Dangaard Brouer 872795bb1c0SJesper Dangaard Brouer /* drop skb->head and call any destructors for packet */ 873795bb1c0SJesper Dangaard Brouer skb_release_all(skb); 874795bb1c0SJesper Dangaard Brouer 875795bb1c0SJesper Dangaard Brouer /* record skb to CPU local list */ 876795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 877795bb1c0SJesper Dangaard Brouer 878795bb1c0SJesper Dangaard Brouer #ifdef CONFIG_SLUB 879795bb1c0SJesper Dangaard Brouer /* SLUB writes into objects when freeing */ 880795bb1c0SJesper Dangaard Brouer prefetchw(skb); 881795bb1c0SJesper Dangaard Brouer #endif 882795bb1c0SJesper Dangaard Brouer 883795bb1c0SJesper Dangaard Brouer /* flush skb_cache if it is filled */ 884795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 885795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 886795bb1c0SJesper Dangaard Brouer nc->skb_cache); 887795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 888795bb1c0SJesper Dangaard Brouer } 889795bb1c0SJesper Dangaard Brouer } 89015fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 89115fad714SJesper Dangaard Brouer { 89215fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 89315fad714SJesper Dangaard Brouer } 894795bb1c0SJesper Dangaard Brouer 895795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 896795bb1c0SJesper Dangaard Brouer { 897795bb1c0SJesper Dangaard Brouer if (unlikely(!skb)) 898795bb1c0SJesper Dangaard Brouer return; 899795bb1c0SJesper Dangaard Brouer 900885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 901795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 902885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 903795bb1c0SJesper Dangaard Brouer return; 904795bb1c0SJesper Dangaard Brouer } 905795bb1c0SJesper Dangaard Brouer 9067608894eSPaolo Abeni if (!skb_unref(skb)) 907795bb1c0SJesper Dangaard Brouer return; 9087608894eSPaolo Abeni 909795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 910795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 911795bb1c0SJesper Dangaard Brouer 912795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 913abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 914795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 915795bb1c0SJesper Dangaard Brouer return; 916795bb1c0SJesper Dangaard Brouer } 917795bb1c0SJesper Dangaard Brouer 91815fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 919795bb1c0SJesper Dangaard Brouer } 920795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 921795bb1c0SJesper Dangaard Brouer 922b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 923b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 924b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 925b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 926b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 927b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 928b1937227SEric Dumazet 929dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 930dec18810SHerbert Xu { 931dec18810SHerbert Xu new->tstamp = old->tstamp; 932b1937227SEric Dumazet /* We do not copy old->sk */ 933dec18810SHerbert Xu new->dev = old->dev; 934b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 9357fee226aSEric Dumazet skb_dst_copy(new, old); 936df5042f4SFlorian Westphal __skb_ext_copy(new, old); 937b1937227SEric Dumazet __nf_copy(new, old, false); 9386aa895b0SPatrick McHardy 939b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 940b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 941b1937227SEric Dumazet */ 942b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 94306021292SEliezer Tamir 944b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 945b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 946b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 947b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 948b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 949b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 950b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 951b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 952b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 953b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 954b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 955b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 956b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 957b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 958b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 959b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 960b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 961b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 962b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 963b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 964b1937227SEric Dumazet #endif 965e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 966b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 96706021292SEliezer Tamir #endif 9682bd82484SEric Dumazet #ifdef CONFIG_XPS 9692bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 9702bd82484SEric Dumazet #endif 971b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 972b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 973b1937227SEric Dumazet #endif 974b1937227SEric Dumazet 975dec18810SHerbert Xu } 976dec18810SHerbert Xu 97782c49a35SHerbert Xu /* 97882c49a35SHerbert Xu * You should not add any new code to this function. Add it to 97982c49a35SHerbert Xu * __copy_skb_header above instead. 98082c49a35SHerbert Xu */ 981e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 9821da177e4SLinus Torvalds { 9831da177e4SLinus Torvalds #define C(x) n->x = skb->x 9841da177e4SLinus Torvalds 9851da177e4SLinus Torvalds n->next = n->prev = NULL; 9861da177e4SLinus Torvalds n->sk = NULL; 987dec18810SHerbert Xu __copy_skb_header(n, skb); 988dec18810SHerbert Xu 9891da177e4SLinus Torvalds C(len); 9901da177e4SLinus Torvalds C(data_len); 9913e6b3b2eSAlexey Dobriyan C(mac_len); 992334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 99302f1c89dSPaul Moore n->cloned = 1; 9941da177e4SLinus Torvalds n->nohdr = 0; 995b13dda9fSEric Dumazet n->peeked = 0; 996e78bfb07SStefano Brivio C(pfmemalloc); 9971da177e4SLinus Torvalds n->destructor = NULL; 9981da177e4SLinus Torvalds C(tail); 9991da177e4SLinus Torvalds C(end); 100002f1c89dSPaul Moore C(head); 1001d3836f21SEric Dumazet C(head_frag); 100202f1c89dSPaul Moore C(data); 100302f1c89dSPaul Moore C(truesize); 100463354797SReshetova, Elena refcount_set(&n->users, 1); 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 10071da177e4SLinus Torvalds skb->cloned = 1; 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds return n; 1010e0053ec0SHerbert Xu #undef C 1011e0053ec0SHerbert Xu } 1012e0053ec0SHerbert Xu 1013e0053ec0SHerbert Xu /** 1014da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1015da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1016da29e4b4SJakub Kicinski */ 1017da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1018da29e4b4SJakub Kicinski { 1019da29e4b4SJakub Kicinski struct sk_buff *n; 1020da29e4b4SJakub Kicinski 1021da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1022da29e4b4SJakub Kicinski if (!n) 1023da29e4b4SJakub Kicinski return NULL; 1024da29e4b4SJakub Kicinski 1025da29e4b4SJakub Kicinski n->len = first->len; 1026da29e4b4SJakub Kicinski n->data_len = first->len; 1027da29e4b4SJakub Kicinski n->truesize = first->truesize; 1028da29e4b4SJakub Kicinski 1029da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1030da29e4b4SJakub Kicinski 1031da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1032da29e4b4SJakub Kicinski n->destructor = NULL; 1033da29e4b4SJakub Kicinski 1034da29e4b4SJakub Kicinski return n; 1035da29e4b4SJakub Kicinski } 1036da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1037da29e4b4SJakub Kicinski 1038da29e4b4SJakub Kicinski /** 1039e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1040e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1041e0053ec0SHerbert Xu * @src: the skb to supply the contents 1042e0053ec0SHerbert Xu * 1043e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1044e0053ec0SHerbert Xu * supplied by the user. 1045e0053ec0SHerbert Xu * 1046e0053ec0SHerbert Xu * The target skb is returned upon exit. 1047e0053ec0SHerbert Xu */ 1048e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1049e0053ec0SHerbert Xu { 10502d4baff8SHerbert Xu skb_release_all(dst); 1051e0053ec0SHerbert Xu return __skb_clone(dst, src); 1052e0053ec0SHerbert Xu } 1053e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1054e0053ec0SHerbert Xu 10556f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1056a91dbff5SWillem de Bruijn { 1057a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1058a91dbff5SWillem de Bruijn struct user_struct *user; 1059a91dbff5SWillem de Bruijn 1060a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1061a91dbff5SWillem de Bruijn return 0; 1062a91dbff5SWillem de Bruijn 1063a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1064a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1065a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1066a91dbff5SWillem de Bruijn 1067a91dbff5SWillem de Bruijn do { 1068a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1069a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1070a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1071a91dbff5SWillem de Bruijn return -ENOBUFS; 1072a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1073a91dbff5SWillem de Bruijn old_pg); 1074a91dbff5SWillem de Bruijn 1075a91dbff5SWillem de Bruijn if (!mmp->user) { 1076a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1077a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1078a91dbff5SWillem de Bruijn } else { 1079a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1080a91dbff5SWillem de Bruijn } 1081a91dbff5SWillem de Bruijn 1082a91dbff5SWillem de Bruijn return 0; 1083a91dbff5SWillem de Bruijn } 10846f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1085a91dbff5SWillem de Bruijn 10866f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1087a91dbff5SWillem de Bruijn { 1088a91dbff5SWillem de Bruijn if (mmp->user) { 1089a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1090a91dbff5SWillem de Bruijn free_uid(mmp->user); 1091a91dbff5SWillem de Bruijn } 1092a91dbff5SWillem de Bruijn } 10936f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1094a91dbff5SWillem de Bruijn 109552267790SWillem de Bruijn struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size) 109652267790SWillem de Bruijn { 109752267790SWillem de Bruijn struct ubuf_info *uarg; 109852267790SWillem de Bruijn struct sk_buff *skb; 109952267790SWillem de Bruijn 110052267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 110152267790SWillem de Bruijn 110252267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 110352267790SWillem de Bruijn if (!skb) 110452267790SWillem de Bruijn return NULL; 110552267790SWillem de Bruijn 110652267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 110752267790SWillem de Bruijn uarg = (void *)skb->cb; 1108a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1109a91dbff5SWillem de Bruijn 1110a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1111a91dbff5SWillem de Bruijn kfree_skb(skb); 1112a91dbff5SWillem de Bruijn return NULL; 1113a91dbff5SWillem de Bruijn } 111452267790SWillem de Bruijn 111552267790SWillem de Bruijn uarg->callback = sock_zerocopy_callback; 11164ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 11174ab6c99dSWillem de Bruijn uarg->len = 1; 11184ab6c99dSWillem de Bruijn uarg->bytelen = size; 111952267790SWillem de Bruijn uarg->zerocopy = 1; 1120c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 112152267790SWillem de Bruijn sock_hold(sk); 112252267790SWillem de Bruijn 112352267790SWillem de Bruijn return uarg; 112452267790SWillem de Bruijn } 112552267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_alloc); 112652267790SWillem de Bruijn 112752267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 112852267790SWillem de Bruijn { 112952267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 113052267790SWillem de Bruijn } 113152267790SWillem de Bruijn 11324ab6c99dSWillem de Bruijn struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 11334ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 11344ab6c99dSWillem de Bruijn { 11354ab6c99dSWillem de Bruijn if (uarg) { 11364ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 11374ab6c99dSWillem de Bruijn u32 bytelen, next; 11384ab6c99dSWillem de Bruijn 11394ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 11404ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 11414ab6c99dSWillem de Bruijn */ 11424ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 11434ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 11444ab6c99dSWillem de Bruijn return NULL; 11454ab6c99dSWillem de Bruijn } 11464ab6c99dSWillem de Bruijn 11474ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 11484ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 11494ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 11504ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 11514ab6c99dSWillem de Bruijn goto new_alloc; 11524ab6c99dSWillem de Bruijn return NULL; 11534ab6c99dSWillem de Bruijn } 11544ab6c99dSWillem de Bruijn 11554ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 11564ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1157a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1158a91dbff5SWillem de Bruijn return NULL; 11594ab6c99dSWillem de Bruijn uarg->len++; 11604ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 11614ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1162100f6d8eSWillem de Bruijn 1163100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1164100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 1165db5bce32SEric Dumazet sock_zerocopy_get(uarg); 1166100f6d8eSWillem de Bruijn 11674ab6c99dSWillem de Bruijn return uarg; 11684ab6c99dSWillem de Bruijn } 11694ab6c99dSWillem de Bruijn } 11704ab6c99dSWillem de Bruijn 11714ab6c99dSWillem de Bruijn new_alloc: 11724ab6c99dSWillem de Bruijn return sock_zerocopy_alloc(sk, size); 11734ab6c99dSWillem de Bruijn } 11744ab6c99dSWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_realloc); 11754ab6c99dSWillem de Bruijn 11764ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 11774ab6c99dSWillem de Bruijn { 11784ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 11794ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 11804ab6c99dSWillem de Bruijn u64 sum_len; 11814ab6c99dSWillem de Bruijn 11824ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 11834ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 11844ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 11854ab6c99dSWillem de Bruijn 11864ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 11874ab6c99dSWillem de Bruijn return false; 11884ab6c99dSWillem de Bruijn 11894ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 11904ab6c99dSWillem de Bruijn return false; 11914ab6c99dSWillem de Bruijn 11924ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 11934ab6c99dSWillem de Bruijn return true; 11944ab6c99dSWillem de Bruijn } 11954ab6c99dSWillem de Bruijn 119652267790SWillem de Bruijn void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) 119752267790SWillem de Bruijn { 11984ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 119952267790SWillem de Bruijn struct sock_exterr_skb *serr; 120052267790SWillem de Bruijn struct sock *sk = skb->sk; 12014ab6c99dSWillem de Bruijn struct sk_buff_head *q; 12024ab6c99dSWillem de Bruijn unsigned long flags; 12034ab6c99dSWillem de Bruijn u32 lo, hi; 12044ab6c99dSWillem de Bruijn u16 len; 120552267790SWillem de Bruijn 1206ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1207ccaffff1SWillem de Bruijn 12084ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 12094ab6c99dSWillem de Bruijn * so do not queue a completion notification 12104ab6c99dSWillem de Bruijn */ 12114ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 121252267790SWillem de Bruijn goto release; 121352267790SWillem de Bruijn 12144ab6c99dSWillem de Bruijn len = uarg->len; 12154ab6c99dSWillem de Bruijn lo = uarg->id; 12164ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 12174ab6c99dSWillem de Bruijn 121852267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 121952267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 122052267790SWillem de Bruijn serr->ee.ee_errno = 0; 122152267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 12224ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 12234ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 122452267790SWillem de Bruijn if (!success) 122552267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 122652267790SWillem de Bruijn 12274ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 12284ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 12294ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 12304ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 12314ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 12324ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 123352267790SWillem de Bruijn skb = NULL; 12344ab6c99dSWillem de Bruijn } 12354ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 123652267790SWillem de Bruijn 123752267790SWillem de Bruijn sk->sk_error_report(sk); 123852267790SWillem de Bruijn 123952267790SWillem de Bruijn release: 124052267790SWillem de Bruijn consume_skb(skb); 124152267790SWillem de Bruijn sock_put(sk); 124252267790SWillem de Bruijn } 124352267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_callback); 124452267790SWillem de Bruijn 124552267790SWillem de Bruijn void sock_zerocopy_put(struct ubuf_info *uarg) 124652267790SWillem de Bruijn { 1247c1d1b437SEric Dumazet if (uarg && refcount_dec_and_test(&uarg->refcnt)) { 124852267790SWillem de Bruijn if (uarg->callback) 124952267790SWillem de Bruijn uarg->callback(uarg, uarg->zerocopy); 125052267790SWillem de Bruijn else 125152267790SWillem de Bruijn consume_skb(skb_from_uarg(uarg)); 125252267790SWillem de Bruijn } 125352267790SWillem de Bruijn } 125452267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put); 125552267790SWillem de Bruijn 125652900d22SWillem de Bruijn void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 125752267790SWillem de Bruijn { 125852267790SWillem de Bruijn if (uarg) { 125952267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 126052267790SWillem de Bruijn 126152267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 12624ab6c99dSWillem de Bruijn uarg->len--; 126352267790SWillem de Bruijn 126452900d22SWillem de Bruijn if (have_uref) 126552267790SWillem de Bruijn sock_zerocopy_put(uarg); 126652267790SWillem de Bruijn } 126752267790SWillem de Bruijn } 126852267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); 126952267790SWillem de Bruijn 1270b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1271b5947e5dSWillem de Bruijn { 1272b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1273b5947e5dSWillem de Bruijn } 1274b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1275b5947e5dSWillem de Bruijn 127652267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 127752267790SWillem de Bruijn struct msghdr *msg, int len, 127852267790SWillem de Bruijn struct ubuf_info *uarg) 127952267790SWillem de Bruijn { 12804ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 128152267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 128252267790SWillem de Bruijn int err, orig_len = skb->len; 128352267790SWillem de Bruijn 12844ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 12854ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 12864ab6c99dSWillem de Bruijn */ 12874ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 12884ab6c99dSWillem de Bruijn return -EEXIST; 12894ab6c99dSWillem de Bruijn 129052267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 129152267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 129254d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 129354d43117SWillem de Bruijn 129452267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 129552267790SWillem de Bruijn msg->msg_iter = orig_iter; 129654d43117SWillem de Bruijn skb->sk = sk; 129752267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 129854d43117SWillem de Bruijn skb->sk = save_sk; 129952267790SWillem de Bruijn return err; 130052267790SWillem de Bruijn } 130152267790SWillem de Bruijn 130252900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 130352267790SWillem de Bruijn return skb->len - orig_len; 130452267790SWillem de Bruijn } 130552267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 130652267790SWillem de Bruijn 13071f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 130852267790SWillem de Bruijn gfp_t gfp_mask) 130952267790SWillem de Bruijn { 131052267790SWillem de Bruijn if (skb_zcopy(orig)) { 131152267790SWillem de Bruijn if (skb_zcopy(nskb)) { 131252267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 131352267790SWillem de Bruijn if (!gfp_mask) { 131452267790SWillem de Bruijn WARN_ON_ONCE(1); 131552267790SWillem de Bruijn return -ENOMEM; 131652267790SWillem de Bruijn } 131752267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 131852267790SWillem de Bruijn return 0; 131952267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 132052267790SWillem de Bruijn return -EIO; 132152267790SWillem de Bruijn } 132252900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 132352267790SWillem de Bruijn } 132452267790SWillem de Bruijn return 0; 132552267790SWillem de Bruijn } 132652267790SWillem de Bruijn 13272c53040fSBen Hutchings /** 13282c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 132948c83012SMichael S. Tsirkin * @skb: the skb to modify 133048c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 133148c83012SMichael S. Tsirkin * 133248c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 133348c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 133448c83012SMichael S. Tsirkin * to userspace pages. 133548c83012SMichael S. Tsirkin * 133648c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 133748c83012SMichael S. Tsirkin * %GFP_ATOMIC. 133848c83012SMichael S. Tsirkin * 133948c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 134048c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 134148c83012SMichael S. Tsirkin */ 134248c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1343a6686f2fSShirley Ma { 1344a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1345a6686f2fSShirley Ma struct page *page, *head = NULL; 13463ece7826SWillem de Bruijn int i, new_frags; 13473ece7826SWillem de Bruijn u32 d_off; 1348a6686f2fSShirley Ma 13493ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 13503ece7826SWillem de Bruijn return -EINVAL; 13513ece7826SWillem de Bruijn 1352f72c4ac6SWillem de Bruijn if (!num_frags) 1353f72c4ac6SWillem de Bruijn goto release; 1354f72c4ac6SWillem de Bruijn 13553ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 13563ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 135702756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1358a6686f2fSShirley Ma if (!page) { 1359a6686f2fSShirley Ma while (head) { 136040dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1361a6686f2fSShirley Ma put_page(head); 1362a6686f2fSShirley Ma head = next; 1363a6686f2fSShirley Ma } 1364a6686f2fSShirley Ma return -ENOMEM; 1365a6686f2fSShirley Ma } 13663ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 13673ece7826SWillem de Bruijn head = page; 13683ece7826SWillem de Bruijn } 13693ece7826SWillem de Bruijn 13703ece7826SWillem de Bruijn page = head; 13713ece7826SWillem de Bruijn d_off = 0; 13723ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 13733ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 13743ece7826SWillem de Bruijn u32 p_off, p_len, copied; 13753ece7826SWillem de Bruijn struct page *p; 13763ece7826SWillem de Bruijn u8 *vaddr; 1377c613c209SWillem de Bruijn 1378b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1379c613c209SWillem de Bruijn p, p_off, p_len, copied) { 13803ece7826SWillem de Bruijn u32 copy, done = 0; 1381c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 13823ece7826SWillem de Bruijn 13833ece7826SWillem de Bruijn while (done < p_len) { 13843ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 13853ece7826SWillem de Bruijn d_off = 0; 13863ece7826SWillem de Bruijn page = (struct page *)page_private(page); 13873ece7826SWillem de Bruijn } 13883ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 13893ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 13903ece7826SWillem de Bruijn vaddr + p_off + done, copy); 13913ece7826SWillem de Bruijn done += copy; 13923ece7826SWillem de Bruijn d_off += copy; 13933ece7826SWillem de Bruijn } 139451c56b00SEric Dumazet kunmap_atomic(vaddr); 1395c613c209SWillem de Bruijn } 1396a6686f2fSShirley Ma } 1397a6686f2fSShirley Ma 1398a6686f2fSShirley Ma /* skb frags release userspace buffers */ 139902756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1400a8605c60SIan Campbell skb_frag_unref(skb, i); 1401a6686f2fSShirley Ma 1402a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 14033ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 14043ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 140540dadff2SSunghan Suh head = (struct page *)page_private(head); 1406a6686f2fSShirley Ma } 14073ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 14083ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 140948c83012SMichael S. Tsirkin 1410b90ddd56SWillem de Bruijn release: 14111f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1412a6686f2fSShirley Ma return 0; 1413a6686f2fSShirley Ma } 1414dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1415a6686f2fSShirley Ma 1416e0053ec0SHerbert Xu /** 1417e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1418e0053ec0SHerbert Xu * @skb: buffer to clone 1419e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1420e0053ec0SHerbert Xu * 1421e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1422e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1423e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1424e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1425e0053ec0SHerbert Xu * 1426e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1427e0053ec0SHerbert Xu * %GFP_ATOMIC. 1428e0053ec0SHerbert Xu */ 1429e0053ec0SHerbert Xu 1430e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1431e0053ec0SHerbert Xu { 1432d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1433d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1434d0bf4a9eSEric Dumazet skb1); 14356ffe75ebSEric Dumazet struct sk_buff *n; 1436e0053ec0SHerbert Xu 143770008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1438a6686f2fSShirley Ma return NULL; 1439a6686f2fSShirley Ma 1440e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 14412638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 14426ffe75ebSEric Dumazet n = &fclones->skb2; 14432638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1444e0053ec0SHerbert Xu } else { 1445c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1446c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1447c93bdd0eSMel Gorman 1448e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1449e0053ec0SHerbert Xu if (!n) 1450e0053ec0SHerbert Xu return NULL; 1451fe55f6d5SVegard Nossum 1452e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1453e0053ec0SHerbert Xu } 1454e0053ec0SHerbert Xu 1455e0053ec0SHerbert Xu return __skb_clone(n, skb); 14561da177e4SLinus Torvalds } 1457b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 14581da177e4SLinus Torvalds 1459b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1460f5b17294SPravin B Shelar { 1461030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1462030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1463030737bcSEric Dumazet skb->csum_start += off; 1464f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1465f5b17294SPravin B Shelar skb->transport_header += off; 1466f5b17294SPravin B Shelar skb->network_header += off; 1467f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1468f5b17294SPravin B Shelar skb->mac_header += off; 1469f5b17294SPravin B Shelar skb->inner_transport_header += off; 1470f5b17294SPravin B Shelar skb->inner_network_header += off; 1471aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1472f5b17294SPravin B Shelar } 1473b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1474f5b17294SPravin B Shelar 147508303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 14761da177e4SLinus Torvalds { 1477dec18810SHerbert Xu __copy_skb_header(new, old); 1478dec18810SHerbert Xu 14797967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 14807967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 14817967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 14821da177e4SLinus Torvalds } 148308303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 14841da177e4SLinus Torvalds 1485c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1486c93bdd0eSMel Gorman { 1487c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1488c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1489c93bdd0eSMel Gorman return 0; 1490c93bdd0eSMel Gorman } 1491c93bdd0eSMel Gorman 14921da177e4SLinus Torvalds /** 14931da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 14941da177e4SLinus Torvalds * @skb: buffer to copy 14951da177e4SLinus Torvalds * @gfp_mask: allocation priority 14961da177e4SLinus Torvalds * 14971da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 14981da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 14991da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 15001da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 15011da177e4SLinus Torvalds * 15021da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 15031da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 15041da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 15051da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 15061da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 15071da177e4SLinus Torvalds */ 15081da177e4SLinus Torvalds 1509dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 15101da177e4SLinus Torvalds { 15116602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1512ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1513c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1514c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 15156602cebbSEric Dumazet 15161da177e4SLinus Torvalds if (!n) 15171da177e4SLinus Torvalds return NULL; 15181da177e4SLinus Torvalds 15191da177e4SLinus Torvalds /* Set the data pointer */ 15201da177e4SLinus Torvalds skb_reserve(n, headerlen); 15211da177e4SLinus Torvalds /* Set the tail pointer and length */ 15221da177e4SLinus Torvalds skb_put(n, skb->len); 15231da177e4SLinus Torvalds 15249f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 15251da177e4SLinus Torvalds 152608303c18SIlya Lesokhin skb_copy_header(n, skb); 15271da177e4SLinus Torvalds return n; 15281da177e4SLinus Torvalds } 1529b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 15301da177e4SLinus Torvalds 15311da177e4SLinus Torvalds /** 1532bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 15331da177e4SLinus Torvalds * @skb: buffer to copy 1534117632e6SEric Dumazet * @headroom: headroom of new skb 15351da177e4SLinus Torvalds * @gfp_mask: allocation priority 1536bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1537bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1538bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 15391da177e4SLinus Torvalds * 15401da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 15411da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 15421da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 15431da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 15441da177e4SLinus Torvalds * or the pointer to the buffer on success. 15451da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 15461da177e4SLinus Torvalds */ 15471da177e4SLinus Torvalds 1548bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1549bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 15501da177e4SLinus Torvalds { 1551117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1552bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1553bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 15546602cebbSEric Dumazet 15551da177e4SLinus Torvalds if (!n) 15561da177e4SLinus Torvalds goto out; 15571da177e4SLinus Torvalds 15581da177e4SLinus Torvalds /* Set the data pointer */ 1559117632e6SEric Dumazet skb_reserve(n, headroom); 15601da177e4SLinus Torvalds /* Set the tail pointer and length */ 15611da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 15621da177e4SLinus Torvalds /* Copy the bytes */ 1563d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 15641da177e4SLinus Torvalds 156525f484a6SHerbert Xu n->truesize += skb->data_len; 15661da177e4SLinus Torvalds n->data_len = skb->data_len; 15671da177e4SLinus Torvalds n->len = skb->len; 15681da177e4SLinus Torvalds 15691da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 15701da177e4SLinus Torvalds int i; 15711da177e4SLinus Torvalds 15721f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 15731f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 15741511022cSDan Carpenter kfree_skb(n); 15751511022cSDan Carpenter n = NULL; 1576a6686f2fSShirley Ma goto out; 1577a6686f2fSShirley Ma } 15781da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 15791da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1580ea2ab693SIan Campbell skb_frag_ref(skb, i); 15811da177e4SLinus Torvalds } 15821da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 15831da177e4SLinus Torvalds } 15841da177e4SLinus Torvalds 158521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 15861da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 15871da177e4SLinus Torvalds skb_clone_fraglist(n); 15881da177e4SLinus Torvalds } 15891da177e4SLinus Torvalds 159008303c18SIlya Lesokhin skb_copy_header(n, skb); 15911da177e4SLinus Torvalds out: 15921da177e4SLinus Torvalds return n; 15931da177e4SLinus Torvalds } 1594bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 15951da177e4SLinus Torvalds 15961da177e4SLinus Torvalds /** 15971da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 15981da177e4SLinus Torvalds * @skb: buffer to reallocate 15991da177e4SLinus Torvalds * @nhead: room to add at head 16001da177e4SLinus Torvalds * @ntail: room to add at tail 16011da177e4SLinus Torvalds * @gfp_mask: allocation priority 16021da177e4SLinus Torvalds * 1603bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1604bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 16051da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 16061da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 16071da177e4SLinus Torvalds * 16081da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 16091da177e4SLinus Torvalds * reloaded after call to this function. 16101da177e4SLinus Torvalds */ 16111da177e4SLinus Torvalds 161286a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1613dd0fc66fSAl Viro gfp_t gfp_mask) 16141da177e4SLinus Torvalds { 1615158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1616158f323bSEric Dumazet int size = osize + nhead + ntail; 16171da177e4SLinus Torvalds long off; 1618158f323bSEric Dumazet u8 *data; 16191da177e4SLinus Torvalds 16204edd87adSHerbert Xu BUG_ON(nhead < 0); 16214edd87adSHerbert Xu 16229f77fad3STim Hansen BUG_ON(skb_shared(skb)); 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 16251da177e4SLinus Torvalds 1626c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1627c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1628c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1629c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 16301da177e4SLinus Torvalds if (!data) 16311da177e4SLinus Torvalds goto nodata; 163287151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 16331da177e4SLinus Torvalds 16341da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 16356602cebbSEric Dumazet * optimized for the cases when header is void. 16366602cebbSEric Dumazet */ 16376602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 16386602cebbSEric Dumazet 16396602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 16406602cebbSEric Dumazet skb_shinfo(skb), 1641fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 16421da177e4SLinus Torvalds 16433e24591aSAlexander Duyck /* 16443e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 16453e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 16463e24591aSAlexander Duyck * be since all we did is relocate the values 16473e24591aSAlexander Duyck */ 16483e24591aSAlexander Duyck if (skb_cloned(skb)) { 164970008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1650a6686f2fSShirley Ma goto nofrags; 16511f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1652c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 16531da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1654ea2ab693SIan Campbell skb_frag_ref(skb, i); 16551da177e4SLinus Torvalds 165621dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 16571da177e4SLinus Torvalds skb_clone_fraglist(skb); 16581da177e4SLinus Torvalds 16591da177e4SLinus Torvalds skb_release_data(skb); 16603e24591aSAlexander Duyck } else { 16613e24591aSAlexander Duyck skb_free_head(skb); 16621fd63041SEric Dumazet } 16631da177e4SLinus Torvalds off = (data + nhead) - skb->head; 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds skb->head = data; 1666d3836f21SEric Dumazet skb->head_frag = 0; 16671da177e4SLinus Torvalds skb->data += off; 16684305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 16694305b541SArnaldo Carvalho de Melo skb->end = size; 167056eb8882SPatrick McHardy off = nhead; 16714305b541SArnaldo Carvalho de Melo #else 16724305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 167356eb8882SPatrick McHardy #endif 167427a884dcSArnaldo Carvalho de Melo skb->tail += off; 1675b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 16761da177e4SLinus Torvalds skb->cloned = 0; 1677334a8132SPatrick McHardy skb->hdr_len = 0; 16781da177e4SLinus Torvalds skb->nohdr = 0; 16791da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1680158f323bSEric Dumazet 1681de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1682de8f3a83SDaniel Borkmann 1683158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1684158f323bSEric Dumazet * For the moment, we really care of rx path, or 1685158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1686158f323bSEric Dumazet */ 1687158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1688158f323bSEric Dumazet skb->truesize += size - osize; 1689158f323bSEric Dumazet 16901da177e4SLinus Torvalds return 0; 16911da177e4SLinus Torvalds 1692a6686f2fSShirley Ma nofrags: 1693a6686f2fSShirley Ma kfree(data); 16941da177e4SLinus Torvalds nodata: 16951da177e4SLinus Torvalds return -ENOMEM; 16961da177e4SLinus Torvalds } 1697b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 16981da177e4SLinus Torvalds 16991da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 17001da177e4SLinus Torvalds 17011da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 17021da177e4SLinus Torvalds { 17031da177e4SLinus Torvalds struct sk_buff *skb2; 17041da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 17051da177e4SLinus Torvalds 17061da177e4SLinus Torvalds if (delta <= 0) 17071da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 17081da177e4SLinus Torvalds else { 17091da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 17101da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 17111da177e4SLinus Torvalds GFP_ATOMIC)) { 17121da177e4SLinus Torvalds kfree_skb(skb2); 17131da177e4SLinus Torvalds skb2 = NULL; 17141da177e4SLinus Torvalds } 17151da177e4SLinus Torvalds } 17161da177e4SLinus Torvalds return skb2; 17171da177e4SLinus Torvalds } 1718b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 17191da177e4SLinus Torvalds 17201da177e4SLinus Torvalds /** 17211da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 17221da177e4SLinus Torvalds * @skb: buffer to copy 17231da177e4SLinus Torvalds * @newheadroom: new free bytes at head 17241da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 17251da177e4SLinus Torvalds * @gfp_mask: allocation priority 17261da177e4SLinus Torvalds * 17271da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 17281da177e4SLinus Torvalds * allocate additional space. 17291da177e4SLinus Torvalds * 17301da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 17311da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 17321da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 17331da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 17341da177e4SLinus Torvalds * 17351da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 17361da177e4SLinus Torvalds * is called from an interrupt. 17371da177e4SLinus Torvalds */ 17381da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 173986a76cafSVictor Fusco int newheadroom, int newtailroom, 1740dd0fc66fSAl Viro gfp_t gfp_mask) 17411da177e4SLinus Torvalds { 17421da177e4SLinus Torvalds /* 17431da177e4SLinus Torvalds * Allocate the copy buffer 17441da177e4SLinus Torvalds */ 1745c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1746c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1747c93bdd0eSMel Gorman NUMA_NO_NODE); 1748efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 17491da177e4SLinus Torvalds int head_copy_len, head_copy_off; 17501da177e4SLinus Torvalds 17511da177e4SLinus Torvalds if (!n) 17521da177e4SLinus Torvalds return NULL; 17531da177e4SLinus Torvalds 17541da177e4SLinus Torvalds skb_reserve(n, newheadroom); 17551da177e4SLinus Torvalds 17561da177e4SLinus Torvalds /* Set the tail pointer and length */ 17571da177e4SLinus Torvalds skb_put(n, skb->len); 17581da177e4SLinus Torvalds 1759efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 17601da177e4SLinus Torvalds head_copy_off = 0; 17611da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 17621da177e4SLinus Torvalds head_copy_len = newheadroom; 17631da177e4SLinus Torvalds else 17641da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds /* Copy the linear header and data. */ 17679f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 17689f77fad3STim Hansen skb->len + head_copy_len)); 17691da177e4SLinus Torvalds 177008303c18SIlya Lesokhin skb_copy_header(n, skb); 17711da177e4SLinus Torvalds 1772030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1773efd1e8d5SPatrick McHardy 17741da177e4SLinus Torvalds return n; 17751da177e4SLinus Torvalds } 1776b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 17771da177e4SLinus Torvalds 17781da177e4SLinus Torvalds /** 1779cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 17801da177e4SLinus Torvalds * @skb: buffer to pad 17811da177e4SLinus Torvalds * @pad: space to pad 1782cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 17831da177e4SLinus Torvalds * 17841da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 17851da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 17861da177e4SLinus Torvalds * beyond the buffer end onto the wire. 17871da177e4SLinus Torvalds * 1788cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1789cd0a137aSFlorian Fainelli * if @free_on_error is true. 17901da177e4SLinus Torvalds */ 17911da177e4SLinus Torvalds 1792cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 17931da177e4SLinus Torvalds { 17945b057c6bSHerbert Xu int err; 17955b057c6bSHerbert Xu int ntail; 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 17985b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 17991da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 18005b057c6bSHerbert Xu return 0; 18011da177e4SLinus Torvalds } 18021da177e4SLinus Torvalds 18034305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 18045b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 18055b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 18065b057c6bSHerbert Xu if (unlikely(err)) 18075b057c6bSHerbert Xu goto free_skb; 18085b057c6bSHerbert Xu } 18095b057c6bSHerbert Xu 18105b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 18115b057c6bSHerbert Xu * to be audited. 18125b057c6bSHerbert Xu */ 18135b057c6bSHerbert Xu err = skb_linearize(skb); 18145b057c6bSHerbert Xu if (unlikely(err)) 18155b057c6bSHerbert Xu goto free_skb; 18165b057c6bSHerbert Xu 18175b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 18185b057c6bSHerbert Xu return 0; 18195b057c6bSHerbert Xu 18205b057c6bSHerbert Xu free_skb: 1821cd0a137aSFlorian Fainelli if (free_on_error) 18221da177e4SLinus Torvalds kfree_skb(skb); 18235b057c6bSHerbert Xu return err; 18241da177e4SLinus Torvalds } 1825cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 18261da177e4SLinus Torvalds 18270dde3e16SIlpo Järvinen /** 18280c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 18290c7ddf36SMathias Krause * @skb: start of the buffer to use 18300c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 18310c7ddf36SMathias Krause * @len: amount of data to add 18320c7ddf36SMathias Krause * 18330c7ddf36SMathias Krause * This function extends the used data area of the potentially 18340c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 18350c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 18360c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 18370c7ddf36SMathias Krause * returned. 18380c7ddf36SMathias Krause */ 18390c7ddf36SMathias Krause 18404df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 18410c7ddf36SMathias Krause { 18420c7ddf36SMathias Krause if (tail != skb) { 18430c7ddf36SMathias Krause skb->data_len += len; 18440c7ddf36SMathias Krause skb->len += len; 18450c7ddf36SMathias Krause } 18460c7ddf36SMathias Krause return skb_put(tail, len); 18470c7ddf36SMathias Krause } 18480c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 18490c7ddf36SMathias Krause 18500c7ddf36SMathias Krause /** 18510dde3e16SIlpo Järvinen * skb_put - add data to a buffer 18520dde3e16SIlpo Järvinen * @skb: buffer to use 18530dde3e16SIlpo Järvinen * @len: amount of data to add 18540dde3e16SIlpo Järvinen * 18550dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 18560dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 18570dde3e16SIlpo Järvinen * first byte of the extra data is returned. 18580dde3e16SIlpo Järvinen */ 18594df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 18600dde3e16SIlpo Järvinen { 18614df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 18620dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 18630dde3e16SIlpo Järvinen skb->tail += len; 18640dde3e16SIlpo Järvinen skb->len += len; 18650dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 18660dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 18670dde3e16SIlpo Järvinen return tmp; 18680dde3e16SIlpo Järvinen } 18690dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 18700dde3e16SIlpo Järvinen 18716be8ac2fSIlpo Järvinen /** 1872c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1873c2aa270aSIlpo Järvinen * @skb: buffer to use 1874c2aa270aSIlpo Järvinen * @len: amount of data to add 1875c2aa270aSIlpo Järvinen * 1876c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1877c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1878c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1879c2aa270aSIlpo Järvinen */ 1880d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 1881c2aa270aSIlpo Järvinen { 1882c2aa270aSIlpo Järvinen skb->data -= len; 1883c2aa270aSIlpo Järvinen skb->len += len; 1884c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 1885c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1886c2aa270aSIlpo Järvinen return skb->data; 1887c2aa270aSIlpo Järvinen } 1888c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1889c2aa270aSIlpo Järvinen 1890c2aa270aSIlpo Järvinen /** 18916be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 18926be8ac2fSIlpo Järvinen * @skb: buffer to use 18936be8ac2fSIlpo Järvinen * @len: amount of data to remove 18946be8ac2fSIlpo Järvinen * 18956be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 18966be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 18976be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 18986be8ac2fSIlpo Järvinen * the old data. 18996be8ac2fSIlpo Järvinen */ 1900af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 19016be8ac2fSIlpo Järvinen { 190247d29646SDavid S. Miller return skb_pull_inline(skb, len); 19036be8ac2fSIlpo Järvinen } 19046be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 19056be8ac2fSIlpo Järvinen 1906419ae74eSIlpo Järvinen /** 1907419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1908419ae74eSIlpo Järvinen * @skb: buffer to alter 1909419ae74eSIlpo Järvinen * @len: new length 1910419ae74eSIlpo Järvinen * 1911419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1912419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1913419ae74eSIlpo Järvinen * The skb must be linear. 1914419ae74eSIlpo Järvinen */ 1915419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1916419ae74eSIlpo Järvinen { 1917419ae74eSIlpo Järvinen if (skb->len > len) 1918419ae74eSIlpo Järvinen __skb_trim(skb, len); 1919419ae74eSIlpo Järvinen } 1920419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1921419ae74eSIlpo Järvinen 19223cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 19231da177e4SLinus Torvalds */ 19241da177e4SLinus Torvalds 19253cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 19261da177e4SLinus Torvalds { 192727b437c8SHerbert Xu struct sk_buff **fragp; 192827b437c8SHerbert Xu struct sk_buff *frag; 19291da177e4SLinus Torvalds int offset = skb_headlen(skb); 19301da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 19311da177e4SLinus Torvalds int i; 193227b437c8SHerbert Xu int err; 193327b437c8SHerbert Xu 193427b437c8SHerbert Xu if (skb_cloned(skb) && 193527b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 193627b437c8SHerbert Xu return err; 19371da177e4SLinus Torvalds 1938f4d26fb3SHerbert Xu i = 0; 1939f4d26fb3SHerbert Xu if (offset >= len) 1940f4d26fb3SHerbert Xu goto drop_pages; 1941f4d26fb3SHerbert Xu 1942f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 19439e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 194427b437c8SHerbert Xu 194527b437c8SHerbert Xu if (end < len) { 19461da177e4SLinus Torvalds offset = end; 194727b437c8SHerbert Xu continue; 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds 19509e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 195127b437c8SHerbert Xu 1952f4d26fb3SHerbert Xu drop_pages: 195327b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 195427b437c8SHerbert Xu 195527b437c8SHerbert Xu for (; i < nfrags; i++) 1956ea2ab693SIan Campbell skb_frag_unref(skb, i); 195727b437c8SHerbert Xu 195821dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 195927b437c8SHerbert Xu skb_drop_fraglist(skb); 1960f4d26fb3SHerbert Xu goto done; 196127b437c8SHerbert Xu } 196227b437c8SHerbert Xu 196327b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 196427b437c8SHerbert Xu fragp = &frag->next) { 196527b437c8SHerbert Xu int end = offset + frag->len; 196627b437c8SHerbert Xu 196727b437c8SHerbert Xu if (skb_shared(frag)) { 196827b437c8SHerbert Xu struct sk_buff *nfrag; 196927b437c8SHerbert Xu 197027b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 197127b437c8SHerbert Xu if (unlikely(!nfrag)) 197227b437c8SHerbert Xu return -ENOMEM; 197327b437c8SHerbert Xu 197427b437c8SHerbert Xu nfrag->next = frag->next; 197585bb2a60SEric Dumazet consume_skb(frag); 197627b437c8SHerbert Xu frag = nfrag; 197727b437c8SHerbert Xu *fragp = frag; 197827b437c8SHerbert Xu } 197927b437c8SHerbert Xu 198027b437c8SHerbert Xu if (end < len) { 198127b437c8SHerbert Xu offset = end; 198227b437c8SHerbert Xu continue; 198327b437c8SHerbert Xu } 198427b437c8SHerbert Xu 198527b437c8SHerbert Xu if (end > len && 198627b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 198727b437c8SHerbert Xu return err; 198827b437c8SHerbert Xu 198927b437c8SHerbert Xu if (frag->next) 199027b437c8SHerbert Xu skb_drop_list(&frag->next); 199127b437c8SHerbert Xu break; 199227b437c8SHerbert Xu } 199327b437c8SHerbert Xu 1994f4d26fb3SHerbert Xu done: 199527b437c8SHerbert Xu if (len > skb_headlen(skb)) { 19961da177e4SLinus Torvalds skb->data_len -= skb->len - len; 19971da177e4SLinus Torvalds skb->len = len; 19981da177e4SLinus Torvalds } else { 19991da177e4SLinus Torvalds skb->len = len; 20001da177e4SLinus Torvalds skb->data_len = 0; 200127a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 20021da177e4SLinus Torvalds } 20031da177e4SLinus Torvalds 2004c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2005c21b48ccSEric Dumazet skb_condense(skb); 20061da177e4SLinus Torvalds return 0; 20071da177e4SLinus Torvalds } 2008b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 20091da177e4SLinus Torvalds 201088078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 201188078d98SEric Dumazet */ 201288078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 201388078d98SEric Dumazet { 201488078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 201588078d98SEric Dumazet int delta = skb->len - len; 201688078d98SEric Dumazet 2017d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2018d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2019d55bef50SDimitris Michailidis len); 202088078d98SEric Dumazet } 202188078d98SEric Dumazet return __pskb_trim(skb, len); 202288078d98SEric Dumazet } 202388078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 202488078d98SEric Dumazet 20251da177e4SLinus Torvalds /** 20261da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 20271da177e4SLinus Torvalds * @skb: buffer to reallocate 20281da177e4SLinus Torvalds * @delta: number of bytes to advance tail 20291da177e4SLinus Torvalds * 20301da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 20311da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 20321da177e4SLinus Torvalds * data from fragmented part. 20331da177e4SLinus Torvalds * 20341da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 20351da177e4SLinus Torvalds * 20361da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 20371da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 20381da177e4SLinus Torvalds * 20391da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 20401da177e4SLinus Torvalds * reloaded after call to this function. 20411da177e4SLinus Torvalds */ 20421da177e4SLinus Torvalds 20431da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 20441da177e4SLinus Torvalds * when it is necessary. 20451da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 20461da177e4SLinus Torvalds * 2. It may change skb pointers. 20471da177e4SLinus Torvalds * 20481da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 20491da177e4SLinus Torvalds */ 2050af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 20511da177e4SLinus Torvalds { 20521da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 20531da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 20541da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 20551da177e4SLinus Torvalds */ 20564305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 20571da177e4SLinus Torvalds 20581da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 20591da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 20601da177e4SLinus Torvalds GFP_ATOMIC)) 20611da177e4SLinus Torvalds return NULL; 20621da177e4SLinus Torvalds } 20631da177e4SLinus Torvalds 20649f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 20659f77fad3STim Hansen skb_tail_pointer(skb), delta)); 20661da177e4SLinus Torvalds 20671da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 20681da177e4SLinus Torvalds * size of pulled pages. Superb. 20691da177e4SLinus Torvalds */ 207021dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 20711da177e4SLinus Torvalds goto pull_pages; 20721da177e4SLinus Torvalds 20731da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 20741da177e4SLinus Torvalds eat = delta; 20751da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20769e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 20779e903e08SEric Dumazet 20789e903e08SEric Dumazet if (size >= eat) 20791da177e4SLinus Torvalds goto pull_pages; 20809e903e08SEric Dumazet eat -= size; 20811da177e4SLinus Torvalds } 20821da177e4SLinus Torvalds 20831da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 208409001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 20851da177e4SLinus Torvalds * but taking into account that pulling is expected to 20861da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 20871da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 20881da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 20891da177e4SLinus Torvalds */ 20901da177e4SLinus Torvalds if (eat) { 20911da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 20921da177e4SLinus Torvalds struct sk_buff *clone = NULL; 20931da177e4SLinus Torvalds struct sk_buff *insp = NULL; 20941da177e4SLinus Torvalds 20951da177e4SLinus Torvalds do { 20961da177e4SLinus Torvalds if (list->len <= eat) { 20971da177e4SLinus Torvalds /* Eaten as whole. */ 20981da177e4SLinus Torvalds eat -= list->len; 20991da177e4SLinus Torvalds list = list->next; 21001da177e4SLinus Torvalds insp = list; 21011da177e4SLinus Torvalds } else { 21021da177e4SLinus Torvalds /* Eaten partially. */ 21031da177e4SLinus Torvalds 21041da177e4SLinus Torvalds if (skb_shared(list)) { 21051da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 21061da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 21071da177e4SLinus Torvalds if (!clone) 21081da177e4SLinus Torvalds return NULL; 21091da177e4SLinus Torvalds insp = list->next; 21101da177e4SLinus Torvalds list = clone; 21111da177e4SLinus Torvalds } else { 21121da177e4SLinus Torvalds /* This may be pulled without 21131da177e4SLinus Torvalds * problems. */ 21141da177e4SLinus Torvalds insp = list; 21151da177e4SLinus Torvalds } 21161da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 21171da177e4SLinus Torvalds kfree_skb(clone); 21181da177e4SLinus Torvalds return NULL; 21191da177e4SLinus Torvalds } 21201da177e4SLinus Torvalds break; 21211da177e4SLinus Torvalds } 21221da177e4SLinus Torvalds } while (eat); 21231da177e4SLinus Torvalds 21241da177e4SLinus Torvalds /* Free pulled out fragments. */ 21251da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 21261da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 21271da177e4SLinus Torvalds kfree_skb(list); 21281da177e4SLinus Torvalds } 21291da177e4SLinus Torvalds /* And insert new clone at head. */ 21301da177e4SLinus Torvalds if (clone) { 21311da177e4SLinus Torvalds clone->next = list; 21321da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 21331da177e4SLinus Torvalds } 21341da177e4SLinus Torvalds } 21351da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 21361da177e4SLinus Torvalds 21371da177e4SLinus Torvalds pull_pages: 21381da177e4SLinus Torvalds eat = delta; 21391da177e4SLinus Torvalds k = 0; 21401da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21419e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 21429e903e08SEric Dumazet 21439e903e08SEric Dumazet if (size <= eat) { 2144ea2ab693SIan Campbell skb_frag_unref(skb, i); 21459e903e08SEric Dumazet eat -= size; 21461da177e4SLinus Torvalds } else { 2147b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2148b54c9d5bSJonathan Lemon 2149b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 21501da177e4SLinus Torvalds if (eat) { 2151b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2152b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 21533ccc6c6fSlinzhang if (!i) 21543ccc6c6fSlinzhang goto end; 21551da177e4SLinus Torvalds eat = 0; 21561da177e4SLinus Torvalds } 21571da177e4SLinus Torvalds k++; 21581da177e4SLinus Torvalds } 21591da177e4SLinus Torvalds } 21601da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 21611da177e4SLinus Torvalds 21623ccc6c6fSlinzhang end: 21631da177e4SLinus Torvalds skb->tail += delta; 21641da177e4SLinus Torvalds skb->data_len -= delta; 21651da177e4SLinus Torvalds 21661f8b977aSWillem de Bruijn if (!skb->data_len) 21671f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 21681f8b977aSWillem de Bruijn 216927a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 21701da177e4SLinus Torvalds } 2171b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 21721da177e4SLinus Torvalds 217322019b17SEric Dumazet /** 217422019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 217522019b17SEric Dumazet * @skb: source skb 217622019b17SEric Dumazet * @offset: offset in source 217722019b17SEric Dumazet * @to: destination buffer 217822019b17SEric Dumazet * @len: number of bytes to copy 217922019b17SEric Dumazet * 218022019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 218122019b17SEric Dumazet * destination buffer. 218222019b17SEric Dumazet * 218322019b17SEric Dumazet * CAUTION ! : 218422019b17SEric Dumazet * If its prototype is ever changed, 218522019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 218622019b17SEric Dumazet * since it is called from BPF assembly code. 218722019b17SEric Dumazet */ 21881da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 21891da177e4SLinus Torvalds { 21901a028e50SDavid S. Miller int start = skb_headlen(skb); 2191fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2192fbb398a8SDavid S. Miller int i, copy; 21931da177e4SLinus Torvalds 21941da177e4SLinus Torvalds if (offset > (int)skb->len - len) 21951da177e4SLinus Torvalds goto fault; 21961da177e4SLinus Torvalds 21971da177e4SLinus Torvalds /* Copy header. */ 21981a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 21991da177e4SLinus Torvalds if (copy > len) 22001da177e4SLinus Torvalds copy = len; 2201d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 22021da177e4SLinus Torvalds if ((len -= copy) == 0) 22031da177e4SLinus Torvalds return 0; 22041da177e4SLinus Torvalds offset += copy; 22051da177e4SLinus Torvalds to += copy; 22061da177e4SLinus Torvalds } 22071da177e4SLinus Torvalds 22081da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22091a028e50SDavid S. Miller int end; 221051c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 22111da177e4SLinus Torvalds 2212547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22131a028e50SDavid S. Miller 221451c56b00SEric Dumazet end = start + skb_frag_size(f); 22151da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2216c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2217c613c209SWillem de Bruijn struct page *p; 22181da177e4SLinus Torvalds u8 *vaddr; 22191da177e4SLinus Torvalds 22201da177e4SLinus Torvalds if (copy > len) 22211da177e4SLinus Torvalds copy = len; 22221da177e4SLinus Torvalds 2223c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2224b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2225c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2226c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2227c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 222851c56b00SEric Dumazet kunmap_atomic(vaddr); 2229c613c209SWillem de Bruijn } 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds if ((len -= copy) == 0) 22321da177e4SLinus Torvalds return 0; 22331da177e4SLinus Torvalds offset += copy; 22341da177e4SLinus Torvalds to += copy; 22351da177e4SLinus Torvalds } 22361a028e50SDavid S. Miller start = end; 22371da177e4SLinus Torvalds } 22381da177e4SLinus Torvalds 2239fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 22401a028e50SDavid S. Miller int end; 22411da177e4SLinus Torvalds 2242547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22431a028e50SDavid S. Miller 2244fbb398a8SDavid S. Miller end = start + frag_iter->len; 22451da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 22461da177e4SLinus Torvalds if (copy > len) 22471da177e4SLinus Torvalds copy = len; 2248fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 22491da177e4SLinus Torvalds goto fault; 22501da177e4SLinus Torvalds if ((len -= copy) == 0) 22511da177e4SLinus Torvalds return 0; 22521da177e4SLinus Torvalds offset += copy; 22531da177e4SLinus Torvalds to += copy; 22541da177e4SLinus Torvalds } 22551a028e50SDavid S. Miller start = end; 22561da177e4SLinus Torvalds } 2257a6686f2fSShirley Ma 22581da177e4SLinus Torvalds if (!len) 22591da177e4SLinus Torvalds return 0; 22601da177e4SLinus Torvalds 22611da177e4SLinus Torvalds fault: 22621da177e4SLinus Torvalds return -EFAULT; 22631da177e4SLinus Torvalds } 2264b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 22651da177e4SLinus Torvalds 22669c55e01cSJens Axboe /* 22679c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 22689c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 22699c55e01cSJens Axboe */ 22709c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 22719c55e01cSJens Axboe { 22728b9d3728SJarek Poplawski put_page(spd->pages[i]); 22738b9d3728SJarek Poplawski } 22749c55e01cSJens Axboe 2275a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 22764fb66994SJarek Poplawski unsigned int *offset, 227718aafc62SEric Dumazet struct sock *sk) 22788b9d3728SJarek Poplawski { 22795640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 22808b9d3728SJarek Poplawski 22815640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 22828b9d3728SJarek Poplawski return NULL; 22834fb66994SJarek Poplawski 22845640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 22854fb66994SJarek Poplawski 22865640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 22875640f768SEric Dumazet page_address(page) + *offset, *len); 22885640f768SEric Dumazet *offset = pfrag->offset; 22895640f768SEric Dumazet pfrag->offset += *len; 22904fb66994SJarek Poplawski 22915640f768SEric Dumazet return pfrag->page; 22929c55e01cSJens Axboe } 22939c55e01cSJens Axboe 229441c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 229541c73a0dSEric Dumazet struct page *page, 229641c73a0dSEric Dumazet unsigned int offset) 229741c73a0dSEric Dumazet { 229841c73a0dSEric Dumazet return spd->nr_pages && 229941c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 230041c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 230141c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 230241c73a0dSEric Dumazet } 230341c73a0dSEric Dumazet 23049c55e01cSJens Axboe /* 23059c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 23069c55e01cSJens Axboe */ 2307a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 230835f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 23094fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 231018aafc62SEric Dumazet bool linear, 23117a67e56fSJarek Poplawski struct sock *sk) 23129c55e01cSJens Axboe { 231341c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2314a108d5f3SDavid S. Miller return true; 23159c55e01cSJens Axboe 23168b9d3728SJarek Poplawski if (linear) { 231718aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 23188b9d3728SJarek Poplawski if (!page) 2319a108d5f3SDavid S. Miller return true; 232041c73a0dSEric Dumazet } 232141c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 232241c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2323a108d5f3SDavid S. Miller return false; 232441c73a0dSEric Dumazet } 23258b9d3728SJarek Poplawski get_page(page); 23269c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 23274fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 23289c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 23299c55e01cSJens Axboe spd->nr_pages++; 23308b9d3728SJarek Poplawski 2331a108d5f3SDavid S. Miller return false; 23329c55e01cSJens Axboe } 23339c55e01cSJens Axboe 2334a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 23352870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 233618aafc62SEric Dumazet unsigned int *len, 2337d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 233835f3d14dSJens Axboe struct sock *sk, 233935f3d14dSJens Axboe struct pipe_inode_info *pipe) 23409c55e01cSJens Axboe { 23412870c43dSOctavian Purdila if (!*len) 2342a108d5f3SDavid S. Miller return true; 23439c55e01cSJens Axboe 23442870c43dSOctavian Purdila /* skip this segment if already processed */ 23452870c43dSOctavian Purdila if (*off >= plen) { 23462870c43dSOctavian Purdila *off -= plen; 2347a108d5f3SDavid S. Miller return false; 23482870c43dSOctavian Purdila } 23492870c43dSOctavian Purdila 23502870c43dSOctavian Purdila /* ignore any bits we already processed */ 23519ca1b22dSEric Dumazet poff += *off; 23529ca1b22dSEric Dumazet plen -= *off; 23532870c43dSOctavian Purdila *off = 0; 23542870c43dSOctavian Purdila 235518aafc62SEric Dumazet do { 235618aafc62SEric Dumazet unsigned int flen = min(*len, plen); 23572870c43dSOctavian Purdila 235818aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 235918aafc62SEric Dumazet linear, sk)) 2360a108d5f3SDavid S. Miller return true; 236118aafc62SEric Dumazet poff += flen; 236218aafc62SEric Dumazet plen -= flen; 23632870c43dSOctavian Purdila *len -= flen; 236418aafc62SEric Dumazet } while (*len && plen); 23652870c43dSOctavian Purdila 2366a108d5f3SDavid S. Miller return false; 2367db43a282SOctavian Purdila } 23689c55e01cSJens Axboe 23699c55e01cSJens Axboe /* 2370a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 23712870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 23729c55e01cSJens Axboe */ 2373a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 237435f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 237535f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 23762870c43dSOctavian Purdila { 23772870c43dSOctavian Purdila int seg; 2378fa9835e5STom Herbert struct sk_buff *iter; 23799c55e01cSJens Axboe 23801d0c0b32SEric Dumazet /* map the linear part : 23812996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 23822996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 23832996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 23849c55e01cSJens Axboe */ 23852870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 23862870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 23872870c43dSOctavian Purdila skb_headlen(skb), 238818aafc62SEric Dumazet offset, len, spd, 23893a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 23901d0c0b32SEric Dumazet sk, pipe)) 2391a108d5f3SDavid S. Miller return true; 23929c55e01cSJens Axboe 23939c55e01cSJens Axboe /* 23949c55e01cSJens Axboe * then map the fragments 23959c55e01cSJens Axboe */ 23969c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 23979c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 23989c55e01cSJens Axboe 2399ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2400b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 240118aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2402a108d5f3SDavid S. Miller return true; 24039c55e01cSJens Axboe } 24049c55e01cSJens Axboe 2405fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2406fa9835e5STom Herbert if (*offset >= iter->len) { 2407fa9835e5STom Herbert *offset -= iter->len; 2408fa9835e5STom Herbert continue; 2409fa9835e5STom Herbert } 2410fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2411fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2412fa9835e5STom Herbert * case. 2413fa9835e5STom Herbert */ 2414fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2415fa9835e5STom Herbert return true; 2416fa9835e5STom Herbert } 2417fa9835e5STom Herbert 2418a108d5f3SDavid S. Miller return false; 24199c55e01cSJens Axboe } 24209c55e01cSJens Axboe 24219c55e01cSJens Axboe /* 24229c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2423fa9835e5STom Herbert * the fragments, and the frag list. 24249c55e01cSJens Axboe */ 2425a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 24269c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 242725869262SAl Viro unsigned int flags) 24289c55e01cSJens Axboe { 242941c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 243041c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 24319c55e01cSJens Axboe struct splice_pipe_desc spd = { 24329c55e01cSJens Axboe .pages = pages, 24339c55e01cSJens Axboe .partial = partial, 2434047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 243528a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 24369c55e01cSJens Axboe .spd_release = sock_spd_release, 24379c55e01cSJens Axboe }; 243835f3d14dSJens Axboe int ret = 0; 243935f3d14dSJens Axboe 2440fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 24419c55e01cSJens Axboe 2442a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 244325869262SAl Viro ret = splice_to_pipe(pipe, &spd); 24449c55e01cSJens Axboe 244535f3d14dSJens Axboe return ret; 24469c55e01cSJens Axboe } 24472b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 24489c55e01cSJens Axboe 244920bf50deSTom Herbert /* Send skb data on a socket. Socket must be locked. */ 245020bf50deSTom Herbert int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 245120bf50deSTom Herbert int len) 245220bf50deSTom Herbert { 245320bf50deSTom Herbert unsigned int orig_len = len; 245420bf50deSTom Herbert struct sk_buff *head = skb; 245520bf50deSTom Herbert unsigned short fragidx; 245620bf50deSTom Herbert int slen, ret; 245720bf50deSTom Herbert 245820bf50deSTom Herbert do_frag_list: 245920bf50deSTom Herbert 246020bf50deSTom Herbert /* Deal with head data */ 246120bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 246220bf50deSTom Herbert struct kvec kv; 246320bf50deSTom Herbert struct msghdr msg; 246420bf50deSTom Herbert 246520bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 246620bf50deSTom Herbert kv.iov_base = skb->data + offset; 2467db5980d8SJohn Fastabend kv.iov_len = slen; 246820bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2469bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 247020bf50deSTom Herbert 247120bf50deSTom Herbert ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen); 247220bf50deSTom Herbert if (ret <= 0) 247320bf50deSTom Herbert goto error; 247420bf50deSTom Herbert 247520bf50deSTom Herbert offset += ret; 247620bf50deSTom Herbert len -= ret; 247720bf50deSTom Herbert } 247820bf50deSTom Herbert 247920bf50deSTom Herbert /* All the data was skb head? */ 248020bf50deSTom Herbert if (!len) 248120bf50deSTom Herbert goto out; 248220bf50deSTom Herbert 248320bf50deSTom Herbert /* Make offset relative to start of frags */ 248420bf50deSTom Herbert offset -= skb_headlen(skb); 248520bf50deSTom Herbert 248620bf50deSTom Herbert /* Find where we are in frag list */ 248720bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 248820bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 248920bf50deSTom Herbert 2490d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 249120bf50deSTom Herbert break; 249220bf50deSTom Herbert 2493d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 249420bf50deSTom Herbert } 249520bf50deSTom Herbert 249620bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 249720bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 249820bf50deSTom Herbert 2499d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 250020bf50deSTom Herbert 250120bf50deSTom Herbert while (slen) { 2502d8e18a51SMatthew Wilcox (Oracle) ret = kernel_sendpage_locked(sk, skb_frag_page(frag), 2503b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 250420bf50deSTom Herbert slen, MSG_DONTWAIT); 250520bf50deSTom Herbert if (ret <= 0) 250620bf50deSTom Herbert goto error; 250720bf50deSTom Herbert 250820bf50deSTom Herbert len -= ret; 250920bf50deSTom Herbert offset += ret; 251020bf50deSTom Herbert slen -= ret; 251120bf50deSTom Herbert } 251220bf50deSTom Herbert 251320bf50deSTom Herbert offset = 0; 251420bf50deSTom Herbert } 251520bf50deSTom Herbert 251620bf50deSTom Herbert if (len) { 251720bf50deSTom Herbert /* Process any frag lists */ 251820bf50deSTom Herbert 251920bf50deSTom Herbert if (skb == head) { 252020bf50deSTom Herbert if (skb_has_frag_list(skb)) { 252120bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 252220bf50deSTom Herbert goto do_frag_list; 252320bf50deSTom Herbert } 252420bf50deSTom Herbert } else if (skb->next) { 252520bf50deSTom Herbert skb = skb->next; 252620bf50deSTom Herbert goto do_frag_list; 252720bf50deSTom Herbert } 252820bf50deSTom Herbert } 252920bf50deSTom Herbert 253020bf50deSTom Herbert out: 253120bf50deSTom Herbert return orig_len - len; 253220bf50deSTom Herbert 253320bf50deSTom Herbert error: 253420bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 253520bf50deSTom Herbert } 253620bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 253720bf50deSTom Herbert 2538357b40a1SHerbert Xu /** 2539357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2540357b40a1SHerbert Xu * @skb: destination buffer 2541357b40a1SHerbert Xu * @offset: offset in destination 2542357b40a1SHerbert Xu * @from: source buffer 2543357b40a1SHerbert Xu * @len: number of bytes to copy 2544357b40a1SHerbert Xu * 2545357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2546357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2547357b40a1SHerbert Xu * traversing fragment lists and such. 2548357b40a1SHerbert Xu */ 2549357b40a1SHerbert Xu 25500c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2551357b40a1SHerbert Xu { 25521a028e50SDavid S. Miller int start = skb_headlen(skb); 2553fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2554fbb398a8SDavid S. Miller int i, copy; 2555357b40a1SHerbert Xu 2556357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2557357b40a1SHerbert Xu goto fault; 2558357b40a1SHerbert Xu 25591a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2560357b40a1SHerbert Xu if (copy > len) 2561357b40a1SHerbert Xu copy = len; 256227d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2563357b40a1SHerbert Xu if ((len -= copy) == 0) 2564357b40a1SHerbert Xu return 0; 2565357b40a1SHerbert Xu offset += copy; 2566357b40a1SHerbert Xu from += copy; 2567357b40a1SHerbert Xu } 2568357b40a1SHerbert Xu 2569357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2570357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 25711a028e50SDavid S. Miller int end; 2572357b40a1SHerbert Xu 2573547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25741a028e50SDavid S. Miller 25759e903e08SEric Dumazet end = start + skb_frag_size(frag); 2576357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2577c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2578c613c209SWillem de Bruijn struct page *p; 2579357b40a1SHerbert Xu u8 *vaddr; 2580357b40a1SHerbert Xu 2581357b40a1SHerbert Xu if (copy > len) 2582357b40a1SHerbert Xu copy = len; 2583357b40a1SHerbert Xu 2584c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2585b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2586c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2587c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2588c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 258951c56b00SEric Dumazet kunmap_atomic(vaddr); 2590c613c209SWillem de Bruijn } 2591357b40a1SHerbert Xu 2592357b40a1SHerbert Xu if ((len -= copy) == 0) 2593357b40a1SHerbert Xu return 0; 2594357b40a1SHerbert Xu offset += copy; 2595357b40a1SHerbert Xu from += copy; 2596357b40a1SHerbert Xu } 25971a028e50SDavid S. Miller start = end; 2598357b40a1SHerbert Xu } 2599357b40a1SHerbert Xu 2600fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 26011a028e50SDavid S. Miller int end; 2602357b40a1SHerbert Xu 2603547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26041a028e50SDavid S. Miller 2605fbb398a8SDavid S. Miller end = start + frag_iter->len; 2606357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2607357b40a1SHerbert Xu if (copy > len) 2608357b40a1SHerbert Xu copy = len; 2609fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 26101a028e50SDavid S. Miller from, copy)) 2611357b40a1SHerbert Xu goto fault; 2612357b40a1SHerbert Xu if ((len -= copy) == 0) 2613357b40a1SHerbert Xu return 0; 2614357b40a1SHerbert Xu offset += copy; 2615357b40a1SHerbert Xu from += copy; 2616357b40a1SHerbert Xu } 26171a028e50SDavid S. Miller start = end; 2618357b40a1SHerbert Xu } 2619357b40a1SHerbert Xu if (!len) 2620357b40a1SHerbert Xu return 0; 2621357b40a1SHerbert Xu 2622357b40a1SHerbert Xu fault: 2623357b40a1SHerbert Xu return -EFAULT; 2624357b40a1SHerbert Xu } 2625357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2626357b40a1SHerbert Xu 26271da177e4SLinus Torvalds /* Checksum skb data. */ 26282817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 26292817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 26301da177e4SLinus Torvalds { 26311a028e50SDavid S. Miller int start = skb_headlen(skb); 26321a028e50SDavid S. Miller int i, copy = start - offset; 2633fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 26341da177e4SLinus Torvalds int pos = 0; 26351da177e4SLinus Torvalds 26361da177e4SLinus Torvalds /* Checksum header. */ 26371da177e4SLinus Torvalds if (copy > 0) { 26381da177e4SLinus Torvalds if (copy > len) 26391da177e4SLinus Torvalds copy = len; 26402544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 26412544af03SMatteo Croce skb->data + offset, copy, csum); 26421da177e4SLinus Torvalds if ((len -= copy) == 0) 26431da177e4SLinus Torvalds return csum; 26441da177e4SLinus Torvalds offset += copy; 26451da177e4SLinus Torvalds pos = copy; 26461da177e4SLinus Torvalds } 26471da177e4SLinus Torvalds 26481da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 26491a028e50SDavid S. Miller int end; 265051c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 26511da177e4SLinus Torvalds 2652547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26531a028e50SDavid S. Miller 265451c56b00SEric Dumazet end = start + skb_frag_size(frag); 26551da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2656c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2657c613c209SWillem de Bruijn struct page *p; 265844bb9363SAl Viro __wsum csum2; 26591da177e4SLinus Torvalds u8 *vaddr; 26601da177e4SLinus Torvalds 26611da177e4SLinus Torvalds if (copy > len) 26621da177e4SLinus Torvalds copy = len; 2663c613c209SWillem de Bruijn 2664c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2665b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2666c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2667c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 26682544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 26692544af03SMatteo Croce csum_partial_ext, 26702544af03SMatteo Croce vaddr + p_off, p_len, 0); 267151c56b00SEric Dumazet kunmap_atomic(vaddr); 26722544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 26732544af03SMatteo Croce csum_block_add_ext, csum, 26742544af03SMatteo Croce csum2, pos, p_len); 2675c613c209SWillem de Bruijn pos += p_len; 2676c613c209SWillem de Bruijn } 2677c613c209SWillem de Bruijn 26781da177e4SLinus Torvalds if (!(len -= copy)) 26791da177e4SLinus Torvalds return csum; 26801da177e4SLinus Torvalds offset += copy; 26811da177e4SLinus Torvalds } 26821a028e50SDavid S. Miller start = end; 26831da177e4SLinus Torvalds } 26841da177e4SLinus Torvalds 2685fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 26861a028e50SDavid S. Miller int end; 26871da177e4SLinus Torvalds 2688547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26891a028e50SDavid S. Miller 2690fbb398a8SDavid S. Miller end = start + frag_iter->len; 26911da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 26925f92a738SAl Viro __wsum csum2; 26931da177e4SLinus Torvalds if (copy > len) 26941da177e4SLinus Torvalds copy = len; 26952817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 26962817a336SDaniel Borkmann copy, 0, ops); 26972544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 26982544af03SMatteo Croce csum, csum2, pos, copy); 26991da177e4SLinus Torvalds if ((len -= copy) == 0) 27001da177e4SLinus Torvalds return csum; 27011da177e4SLinus Torvalds offset += copy; 27021da177e4SLinus Torvalds pos += copy; 27031da177e4SLinus Torvalds } 27041a028e50SDavid S. Miller start = end; 27051da177e4SLinus Torvalds } 270609a62660SKris Katterjohn BUG_ON(len); 27071da177e4SLinus Torvalds 27081da177e4SLinus Torvalds return csum; 27091da177e4SLinus Torvalds } 27102817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 27112817a336SDaniel Borkmann 27122817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 27132817a336SDaniel Borkmann int len, __wsum csum) 27142817a336SDaniel Borkmann { 27152817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2716cea80ea8SDaniel Borkmann .update = csum_partial_ext, 27172817a336SDaniel Borkmann .combine = csum_block_add_ext, 27182817a336SDaniel Borkmann }; 27192817a336SDaniel Borkmann 27202817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 27212817a336SDaniel Borkmann } 2722b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 27231da177e4SLinus Torvalds 27241da177e4SLinus Torvalds /* Both of above in one bottle. */ 27251da177e4SLinus Torvalds 272681d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 272781d77662SAl Viro u8 *to, int len, __wsum csum) 27281da177e4SLinus Torvalds { 27291a028e50SDavid S. Miller int start = skb_headlen(skb); 27301a028e50SDavid S. Miller int i, copy = start - offset; 2731fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 27321da177e4SLinus Torvalds int pos = 0; 27331da177e4SLinus Torvalds 27341da177e4SLinus Torvalds /* Copy header. */ 27351da177e4SLinus Torvalds if (copy > 0) { 27361da177e4SLinus Torvalds if (copy > len) 27371da177e4SLinus Torvalds copy = len; 27381da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 27391da177e4SLinus Torvalds copy, csum); 27401da177e4SLinus Torvalds if ((len -= copy) == 0) 27411da177e4SLinus Torvalds return csum; 27421da177e4SLinus Torvalds offset += copy; 27431da177e4SLinus Torvalds to += copy; 27441da177e4SLinus Torvalds pos = copy; 27451da177e4SLinus Torvalds } 27461da177e4SLinus Torvalds 27471da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 27481a028e50SDavid S. Miller int end; 27491da177e4SLinus Torvalds 2750547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27511a028e50SDavid S. Miller 27529e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 27531da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2754c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2755c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2756c613c209SWillem de Bruijn struct page *p; 27575084205fSAl Viro __wsum csum2; 27581da177e4SLinus Torvalds u8 *vaddr; 27591da177e4SLinus Torvalds 27601da177e4SLinus Torvalds if (copy > len) 27611da177e4SLinus Torvalds copy = len; 2762c613c209SWillem de Bruijn 2763c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2764b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2765c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2766c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2767c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2768c613c209SWillem de Bruijn to + copied, 2769c613c209SWillem de Bruijn p_len, 0); 277051c56b00SEric Dumazet kunmap_atomic(vaddr); 27711da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2772c613c209SWillem de Bruijn pos += p_len; 2773c613c209SWillem de Bruijn } 2774c613c209SWillem de Bruijn 27751da177e4SLinus Torvalds if (!(len -= copy)) 27761da177e4SLinus Torvalds return csum; 27771da177e4SLinus Torvalds offset += copy; 27781da177e4SLinus Torvalds to += copy; 27791da177e4SLinus Torvalds } 27801a028e50SDavid S. Miller start = end; 27811da177e4SLinus Torvalds } 27821da177e4SLinus Torvalds 2783fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 278481d77662SAl Viro __wsum csum2; 27851a028e50SDavid S. Miller int end; 27861da177e4SLinus Torvalds 2787547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27881a028e50SDavid S. Miller 2789fbb398a8SDavid S. Miller end = start + frag_iter->len; 27901da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 27911da177e4SLinus Torvalds if (copy > len) 27921da177e4SLinus Torvalds copy = len; 2793fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 27941a028e50SDavid S. Miller offset - start, 27951da177e4SLinus Torvalds to, copy, 0); 27961da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 27971da177e4SLinus Torvalds if ((len -= copy) == 0) 27981da177e4SLinus Torvalds return csum; 27991da177e4SLinus Torvalds offset += copy; 28001da177e4SLinus Torvalds to += copy; 28011da177e4SLinus Torvalds pos += copy; 28021da177e4SLinus Torvalds } 28031a028e50SDavid S. Miller start = end; 28041da177e4SLinus Torvalds } 280509a62660SKris Katterjohn BUG_ON(len); 28061da177e4SLinus Torvalds return csum; 28071da177e4SLinus Torvalds } 2808b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 28091da177e4SLinus Torvalds 281049f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 281149f8e832SCong Wang { 281249f8e832SCong Wang __sum16 sum; 281349f8e832SCong Wang 281449f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 281514641931SCong Wang /* See comments in __skb_checksum_complete(). */ 281649f8e832SCong Wang if (likely(!sum)) { 281749f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 281849f8e832SCong Wang !skb->csum_complete_sw) 28197fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 282049f8e832SCong Wang } 282149f8e832SCong Wang if (!skb_shared(skb)) 282249f8e832SCong Wang skb->csum_valid = !sum; 282349f8e832SCong Wang return sum; 282449f8e832SCong Wang } 282549f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 282649f8e832SCong Wang 282714641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 282814641931SCong Wang * which has been changed from the hardware checksum, for example, by 282914641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 283014641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 283114641931SCong Wang * 283214641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 283314641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 283414641931SCong Wang * shared. 283514641931SCong Wang */ 283649f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 283749f8e832SCong Wang { 283849f8e832SCong Wang __wsum csum; 283949f8e832SCong Wang __sum16 sum; 284049f8e832SCong Wang 284149f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 284249f8e832SCong Wang 284349f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 284414641931SCong Wang /* This check is inverted, because we already knew the hardware 284514641931SCong Wang * checksum is invalid before calling this function. So, if the 284614641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 284714641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 284814641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 284914641931SCong Wang * when moving skb->data around. 285014641931SCong Wang */ 285149f8e832SCong Wang if (likely(!sum)) { 285249f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 285349f8e832SCong Wang !skb->csum_complete_sw) 28547fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 285549f8e832SCong Wang } 285649f8e832SCong Wang 285749f8e832SCong Wang if (!skb_shared(skb)) { 285849f8e832SCong Wang /* Save full packet checksum */ 285949f8e832SCong Wang skb->csum = csum; 286049f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 286149f8e832SCong Wang skb->csum_complete_sw = 1; 286249f8e832SCong Wang skb->csum_valid = !sum; 286349f8e832SCong Wang } 286449f8e832SCong Wang 286549f8e832SCong Wang return sum; 286649f8e832SCong Wang } 286749f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 286849f8e832SCong Wang 28699617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 28709617813dSDavide Caratti { 28719617813dSDavide Caratti net_warn_ratelimited( 28729617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 28739617813dSDavide Caratti __func__); 28749617813dSDavide Caratti return 0; 28759617813dSDavide Caratti } 28769617813dSDavide Caratti 28779617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 28789617813dSDavide Caratti int offset, int len) 28799617813dSDavide Caratti { 28809617813dSDavide Caratti net_warn_ratelimited( 28819617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 28829617813dSDavide Caratti __func__); 28839617813dSDavide Caratti return 0; 28849617813dSDavide Caratti } 28859617813dSDavide Caratti 28869617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 28879617813dSDavide Caratti .update = warn_crc32c_csum_update, 28889617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 28899617813dSDavide Caratti }; 28909617813dSDavide Caratti 28919617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 28929617813dSDavide Caratti &default_crc32c_ops; 28939617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 28949617813dSDavide Caratti 2895af2806f8SThomas Graf /** 2896af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2897af2806f8SThomas Graf * @from: source buffer 2898af2806f8SThomas Graf * 2899af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2900af2806f8SThomas Graf * into skb_zerocopy(). 2901af2806f8SThomas Graf */ 2902af2806f8SThomas Graf unsigned int 2903af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2904af2806f8SThomas Graf { 2905af2806f8SThomas Graf unsigned int hlen = 0; 2906af2806f8SThomas Graf 2907af2806f8SThomas Graf if (!from->head_frag || 2908af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2909af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2910af2806f8SThomas Graf hlen = skb_headlen(from); 2911af2806f8SThomas Graf 2912af2806f8SThomas Graf if (skb_has_frag_list(from)) 2913af2806f8SThomas Graf hlen = from->len; 2914af2806f8SThomas Graf 2915af2806f8SThomas Graf return hlen; 2916af2806f8SThomas Graf } 2917af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2918af2806f8SThomas Graf 2919af2806f8SThomas Graf /** 2920af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2921af2806f8SThomas Graf * @to: destination buffer 29227fceb4deSMasanari Iida * @from: source buffer 2923af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2924af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2925af2806f8SThomas Graf * 2926af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2927af2806f8SThomas Graf * to the frags in the source buffer. 2928af2806f8SThomas Graf * 2929af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2930af2806f8SThomas Graf * headroom in the `to` buffer. 293136d5fe6aSZoltan Kiss * 293236d5fe6aSZoltan Kiss * Return value: 293336d5fe6aSZoltan Kiss * 0: everything is OK 293436d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 293536d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2936af2806f8SThomas Graf */ 293736d5fe6aSZoltan Kiss int 293836d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2939af2806f8SThomas Graf { 2940af2806f8SThomas Graf int i, j = 0; 2941af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 294236d5fe6aSZoltan Kiss int ret; 2943af2806f8SThomas Graf struct page *page; 2944af2806f8SThomas Graf unsigned int offset; 2945af2806f8SThomas Graf 2946af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2947af2806f8SThomas Graf 2948af2806f8SThomas Graf /* dont bother with small payloads */ 294936d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 295036d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2951af2806f8SThomas Graf 2952af2806f8SThomas Graf if (hlen) { 295336d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 295436d5fe6aSZoltan Kiss if (unlikely(ret)) 295536d5fe6aSZoltan Kiss return ret; 2956af2806f8SThomas Graf len -= hlen; 2957af2806f8SThomas Graf } else { 2958af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2959af2806f8SThomas Graf if (plen) { 2960af2806f8SThomas Graf page = virt_to_head_page(from->head); 2961af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2962af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2963af2806f8SThomas Graf get_page(page); 2964af2806f8SThomas Graf j = 1; 2965af2806f8SThomas Graf len -= plen; 2966af2806f8SThomas Graf } 2967af2806f8SThomas Graf } 2968af2806f8SThomas Graf 2969af2806f8SThomas Graf to->truesize += len + plen; 2970af2806f8SThomas Graf to->len += len + plen; 2971af2806f8SThomas Graf to->data_len += len + plen; 2972af2806f8SThomas Graf 297336d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 297436d5fe6aSZoltan Kiss skb_tx_error(from); 297536d5fe6aSZoltan Kiss return -ENOMEM; 297636d5fe6aSZoltan Kiss } 29771f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 297836d5fe6aSZoltan Kiss 2979af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2980d8e18a51SMatthew Wilcox (Oracle) int size; 2981d8e18a51SMatthew Wilcox (Oracle) 2982af2806f8SThomas Graf if (!len) 2983af2806f8SThomas Graf break; 2984af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2985d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 2986d8e18a51SMatthew Wilcox (Oracle) len); 2987d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 2988d8e18a51SMatthew Wilcox (Oracle) len -= size; 2989af2806f8SThomas Graf skb_frag_ref(to, j); 2990af2806f8SThomas Graf j++; 2991af2806f8SThomas Graf } 2992af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 299336d5fe6aSZoltan Kiss 299436d5fe6aSZoltan Kiss return 0; 2995af2806f8SThomas Graf } 2996af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2997af2806f8SThomas Graf 29981da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 29991da177e4SLinus Torvalds { 3000d3bc23e7SAl Viro __wsum csum; 30011da177e4SLinus Torvalds long csstart; 30021da177e4SLinus Torvalds 300384fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 300455508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 30051da177e4SLinus Torvalds else 30061da177e4SLinus Torvalds csstart = skb_headlen(skb); 30071da177e4SLinus Torvalds 300809a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 30091da177e4SLinus Torvalds 3010d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 30111da177e4SLinus Torvalds 30121da177e4SLinus Torvalds csum = 0; 30131da177e4SLinus Torvalds if (csstart != skb->len) 30141da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 30151da177e4SLinus Torvalds skb->len - csstart, 0); 30161da177e4SLinus Torvalds 301784fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3018ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 30191da177e4SLinus Torvalds 3020d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 30211da177e4SLinus Torvalds } 30221da177e4SLinus Torvalds } 3023b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 30241da177e4SLinus Torvalds 30251da177e4SLinus Torvalds /** 30261da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 30271da177e4SLinus Torvalds * @list: list to dequeue from 30281da177e4SLinus Torvalds * 30291da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 30301da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 30311da177e4SLinus Torvalds * returned or %NULL if the list is empty. 30321da177e4SLinus Torvalds */ 30331da177e4SLinus Torvalds 30341da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 30351da177e4SLinus Torvalds { 30361da177e4SLinus Torvalds unsigned long flags; 30371da177e4SLinus Torvalds struct sk_buff *result; 30381da177e4SLinus Torvalds 30391da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30401da177e4SLinus Torvalds result = __skb_dequeue(list); 30411da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30421da177e4SLinus Torvalds return result; 30431da177e4SLinus Torvalds } 3044b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 30451da177e4SLinus Torvalds 30461da177e4SLinus Torvalds /** 30471da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 30481da177e4SLinus Torvalds * @list: list to dequeue from 30491da177e4SLinus Torvalds * 30501da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 30511da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 30521da177e4SLinus Torvalds * returned or %NULL if the list is empty. 30531da177e4SLinus Torvalds */ 30541da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 30551da177e4SLinus Torvalds { 30561da177e4SLinus Torvalds unsigned long flags; 30571da177e4SLinus Torvalds struct sk_buff *result; 30581da177e4SLinus Torvalds 30591da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30601da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 30611da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30621da177e4SLinus Torvalds return result; 30631da177e4SLinus Torvalds } 3064b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 30651da177e4SLinus Torvalds 30661da177e4SLinus Torvalds /** 30671da177e4SLinus Torvalds * skb_queue_purge - empty a list 30681da177e4SLinus Torvalds * @list: list to empty 30691da177e4SLinus Torvalds * 30701da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 30711da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 30721da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 30731da177e4SLinus Torvalds */ 30741da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 30751da177e4SLinus Torvalds { 30761da177e4SLinus Torvalds struct sk_buff *skb; 30771da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 30781da177e4SLinus Torvalds kfree_skb(skb); 30791da177e4SLinus Torvalds } 3080b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 30811da177e4SLinus Torvalds 30821da177e4SLinus Torvalds /** 30839f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 30849f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3085385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 30869f5afeaeSYaogong Wang * 30879f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 30889f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 30899f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 30909f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 30919f5afeaeSYaogong Wang */ 3092385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 30939f5afeaeSYaogong Wang { 30947c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3095385114deSPeter Oskolkov unsigned int sum = 0; 30969f5afeaeSYaogong Wang 30977c90584cSEric Dumazet while (p) { 30987c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 30997c90584cSEric Dumazet 31007c90584cSEric Dumazet p = rb_next(p); 31017c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3102385114deSPeter Oskolkov sum += skb->truesize; 31039f5afeaeSYaogong Wang kfree_skb(skb); 31047c90584cSEric Dumazet } 3105385114deSPeter Oskolkov return sum; 31069f5afeaeSYaogong Wang } 31079f5afeaeSYaogong Wang 31089f5afeaeSYaogong Wang /** 31091da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 31101da177e4SLinus Torvalds * @list: list to use 31111da177e4SLinus Torvalds * @newsk: buffer to queue 31121da177e4SLinus Torvalds * 31131da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 31141da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 31151da177e4SLinus Torvalds * safely. 31161da177e4SLinus Torvalds * 31171da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31181da177e4SLinus Torvalds */ 31191da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 31201da177e4SLinus Torvalds { 31211da177e4SLinus Torvalds unsigned long flags; 31221da177e4SLinus Torvalds 31231da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31241da177e4SLinus Torvalds __skb_queue_head(list, newsk); 31251da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31261da177e4SLinus Torvalds } 3127b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds /** 31301da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 31311da177e4SLinus Torvalds * @list: list to use 31321da177e4SLinus Torvalds * @newsk: buffer to queue 31331da177e4SLinus Torvalds * 31341da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 31351da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 31361da177e4SLinus Torvalds * safely. 31371da177e4SLinus Torvalds * 31381da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31391da177e4SLinus Torvalds */ 31401da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 31411da177e4SLinus Torvalds { 31421da177e4SLinus Torvalds unsigned long flags; 31431da177e4SLinus Torvalds 31441da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31451da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 31461da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31471da177e4SLinus Torvalds } 3148b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 31498728b834SDavid S. Miller 31501da177e4SLinus Torvalds /** 31511da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 31521da177e4SLinus Torvalds * @skb: buffer to remove 31538728b834SDavid S. Miller * @list: list to use 31541da177e4SLinus Torvalds * 31558728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 31568728b834SDavid S. Miller * function is atomic with respect to other list locked calls 31571da177e4SLinus Torvalds * 31588728b834SDavid S. Miller * You must know what list the SKB is on. 31591da177e4SLinus Torvalds */ 31608728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 31611da177e4SLinus Torvalds { 31621da177e4SLinus Torvalds unsigned long flags; 31631da177e4SLinus Torvalds 31641da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31658728b834SDavid S. Miller __skb_unlink(skb, list); 31661da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31671da177e4SLinus Torvalds } 3168b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 31691da177e4SLinus Torvalds 31701da177e4SLinus Torvalds /** 31711da177e4SLinus Torvalds * skb_append - append a buffer 31721da177e4SLinus Torvalds * @old: buffer to insert after 31731da177e4SLinus Torvalds * @newsk: buffer to insert 31748728b834SDavid S. Miller * @list: list to use 31751da177e4SLinus Torvalds * 31761da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 31771da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 31781da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31791da177e4SLinus Torvalds */ 31808728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 31811da177e4SLinus Torvalds { 31821da177e4SLinus Torvalds unsigned long flags; 31831da177e4SLinus Torvalds 31848728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 31857de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 31868728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 31871da177e4SLinus Torvalds } 3188b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 31891da177e4SLinus Torvalds 31901da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 31911da177e4SLinus Torvalds struct sk_buff* skb1, 31921da177e4SLinus Torvalds const u32 len, const int pos) 31931da177e4SLinus Torvalds { 31941da177e4SLinus Torvalds int i; 31951da177e4SLinus Torvalds 3196d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3197d626f62bSArnaldo Carvalho de Melo pos - len); 31981da177e4SLinus Torvalds /* And move data appendix as is. */ 31991da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 32001da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 32011da177e4SLinus Torvalds 32021da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 32031da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 32041da177e4SLinus Torvalds skb1->data_len = skb->data_len; 32051da177e4SLinus Torvalds skb1->len += skb1->data_len; 32061da177e4SLinus Torvalds skb->data_len = 0; 32071da177e4SLinus Torvalds skb->len = len; 320827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 32091da177e4SLinus Torvalds } 32101da177e4SLinus Torvalds 32111da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 32121da177e4SLinus Torvalds struct sk_buff* skb1, 32131da177e4SLinus Torvalds const u32 len, int pos) 32141da177e4SLinus Torvalds { 32151da177e4SLinus Torvalds int i, k = 0; 32161da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 32171da177e4SLinus Torvalds 32181da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 32191da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 32201da177e4SLinus Torvalds skb->len = len; 32211da177e4SLinus Torvalds skb->data_len = len - pos; 32221da177e4SLinus Torvalds 32231da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 32249e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 32251da177e4SLinus Torvalds 32261da177e4SLinus Torvalds if (pos + size > len) { 32271da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 32281da177e4SLinus Torvalds 32291da177e4SLinus Torvalds if (pos < len) { 32301da177e4SLinus Torvalds /* Split frag. 32311da177e4SLinus Torvalds * We have two variants in this case: 32321da177e4SLinus Torvalds * 1. Move all the frag to the second 32331da177e4SLinus Torvalds * part, if it is possible. F.e. 32341da177e4SLinus Torvalds * this approach is mandatory for TUX, 32351da177e4SLinus Torvalds * where splitting is expensive. 32361da177e4SLinus Torvalds * 2. Split is accurately. We make this. 32371da177e4SLinus Torvalds */ 3238ea2ab693SIan Campbell skb_frag_ref(skb, i); 3239b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 32409e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 32419e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 32421da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 32431da177e4SLinus Torvalds } 32441da177e4SLinus Torvalds k++; 32451da177e4SLinus Torvalds } else 32461da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 32471da177e4SLinus Torvalds pos += size; 32481da177e4SLinus Torvalds } 32491da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 32501da177e4SLinus Torvalds } 32511da177e4SLinus Torvalds 32521da177e4SLinus Torvalds /** 32531da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 32541da177e4SLinus Torvalds * @skb: the buffer to split 32551da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 32561da177e4SLinus Torvalds * @len: new length for skb 32571da177e4SLinus Torvalds */ 32581da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 32591da177e4SLinus Torvalds { 32601da177e4SLinus Torvalds int pos = skb_headlen(skb); 32611da177e4SLinus Torvalds 3262fff88030SWillem de Bruijn skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags & 3263fff88030SWillem de Bruijn SKBTX_SHARED_FRAG; 32641f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 32651da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 32661da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 32671da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 32681da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 32691da177e4SLinus Torvalds } 3270b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 32711da177e4SLinus Torvalds 32729f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 32739f782db3SIlpo Järvinen * 32749f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 32759f782db3SIlpo Järvinen */ 3276832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3277832d11c5SIlpo Järvinen { 32780ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3279832d11c5SIlpo Järvinen } 3280832d11c5SIlpo Järvinen 3281832d11c5SIlpo Järvinen /** 3282832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3283832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3284832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3285832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3286832d11c5SIlpo Järvinen * 3287832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 328820e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3289832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3290832d11c5SIlpo Järvinen * 3291832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3292832d11c5SIlpo Järvinen * 3293832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3294832d11c5SIlpo Järvinen * to have non-paged data as well. 3295832d11c5SIlpo Järvinen * 3296832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3297832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3298832d11c5SIlpo Järvinen */ 3299832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3300832d11c5SIlpo Järvinen { 3301832d11c5SIlpo Järvinen int from, to, merge, todo; 3302d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3303832d11c5SIlpo Järvinen 3304832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3305f8071cdeSEric Dumazet 3306f8071cdeSEric Dumazet if (skb_headlen(skb)) 3307f8071cdeSEric Dumazet return 0; 33081f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 33091f8b977aSWillem de Bruijn return 0; 3310832d11c5SIlpo Järvinen 3311832d11c5SIlpo Järvinen todo = shiftlen; 3312832d11c5SIlpo Järvinen from = 0; 3313832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3314832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3315832d11c5SIlpo Järvinen 3316832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3317832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3318832d11c5SIlpo Järvinen */ 3319832d11c5SIlpo Järvinen if (!to || 3320ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3321b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3322832d11c5SIlpo Järvinen merge = -1; 3323832d11c5SIlpo Järvinen } else { 3324832d11c5SIlpo Järvinen merge = to - 1; 3325832d11c5SIlpo Järvinen 33269e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3327832d11c5SIlpo Järvinen if (todo < 0) { 3328832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3329832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3330832d11c5SIlpo Järvinen return 0; 3331832d11c5SIlpo Järvinen 33329f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 33339f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3334832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3335832d11c5SIlpo Järvinen 33369e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 33379e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3338b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3339832d11c5SIlpo Järvinen 3340832d11c5SIlpo Järvinen goto onlymerged; 3341832d11c5SIlpo Järvinen } 3342832d11c5SIlpo Järvinen 3343832d11c5SIlpo Järvinen from++; 3344832d11c5SIlpo Järvinen } 3345832d11c5SIlpo Järvinen 3346832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3347832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3348832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3349832d11c5SIlpo Järvinen return 0; 3350832d11c5SIlpo Järvinen 3351832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3352832d11c5SIlpo Järvinen return 0; 3353832d11c5SIlpo Järvinen 3354832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3355832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3356832d11c5SIlpo Järvinen return 0; 3357832d11c5SIlpo Järvinen 3358832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3359832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3360832d11c5SIlpo Järvinen 33619e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3362832d11c5SIlpo Järvinen *fragto = *fragfrom; 33639e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3364832d11c5SIlpo Järvinen from++; 3365832d11c5SIlpo Järvinen to++; 3366832d11c5SIlpo Järvinen 3367832d11c5SIlpo Järvinen } else { 3368ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3369b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3370b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 33719e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3372832d11c5SIlpo Järvinen 3373b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 33749e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3375832d11c5SIlpo Järvinen todo = 0; 3376832d11c5SIlpo Järvinen 3377832d11c5SIlpo Järvinen to++; 3378832d11c5SIlpo Järvinen break; 3379832d11c5SIlpo Järvinen } 3380832d11c5SIlpo Järvinen } 3381832d11c5SIlpo Järvinen 3382832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3383832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3384832d11c5SIlpo Järvinen 3385832d11c5SIlpo Järvinen if (merge >= 0) { 3386832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3387832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3388832d11c5SIlpo Järvinen 33899e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 3390ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 3391832d11c5SIlpo Järvinen } 3392832d11c5SIlpo Järvinen 3393832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3394832d11c5SIlpo Järvinen to = 0; 3395832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3396832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3397832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3398832d11c5SIlpo Järvinen 3399832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3400832d11c5SIlpo Järvinen 3401832d11c5SIlpo Järvinen onlymerged: 3402832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3403832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3404832d11c5SIlpo Järvinen */ 3405832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3406832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3407832d11c5SIlpo Järvinen 3408832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3409832d11c5SIlpo Järvinen skb->len -= shiftlen; 3410832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3411832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3412832d11c5SIlpo Järvinen tgt->len += shiftlen; 3413832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3414832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3415832d11c5SIlpo Järvinen 3416832d11c5SIlpo Järvinen return shiftlen; 3417832d11c5SIlpo Järvinen } 3418832d11c5SIlpo Järvinen 3419677e90edSThomas Graf /** 3420677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3421677e90edSThomas Graf * @skb: the buffer to read 3422677e90edSThomas Graf * @from: lower offset of data to be read 3423677e90edSThomas Graf * @to: upper offset of data to be read 3424677e90edSThomas Graf * @st: state variable 3425677e90edSThomas Graf * 3426677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3427677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3428677e90edSThomas Graf */ 3429677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3430677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3431677e90edSThomas Graf { 3432677e90edSThomas Graf st->lower_offset = from; 3433677e90edSThomas Graf st->upper_offset = to; 3434677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3435677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3436677e90edSThomas Graf st->frag_data = NULL; 3437677e90edSThomas Graf } 3438b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3439677e90edSThomas Graf 3440677e90edSThomas Graf /** 3441677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3442677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3443677e90edSThomas Graf * @data: destination pointer for data to be returned 3444677e90edSThomas Graf * @st: state variable 3445677e90edSThomas Graf * 3446bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3447677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3448bc32383cSMathias Krause * the head of the data block to @data and returns the length 3449677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3450677e90edSThomas Graf * offset has been reached. 3451677e90edSThomas Graf * 3452677e90edSThomas Graf * The caller is not required to consume all of the data 3453bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3454677e90edSThomas Graf * of bytes already consumed and the next call to 3455677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3456677e90edSThomas Graf * 345725985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3458e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3459677e90edSThomas Graf * reads of potentially non linear data. 3460677e90edSThomas Graf * 3461bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3462677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3463677e90edSThomas Graf * a stack for this purpose. 3464677e90edSThomas Graf */ 3465677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3466677e90edSThomas Graf struct skb_seq_state *st) 3467677e90edSThomas Graf { 3468677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3469677e90edSThomas Graf skb_frag_t *frag; 3470677e90edSThomas Graf 3471aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3472aeb193eaSWedson Almeida Filho if (st->frag_data) { 3473aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3474aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3475aeb193eaSWedson Almeida Filho } 3476677e90edSThomas Graf return 0; 3477aeb193eaSWedson Almeida Filho } 3478677e90edSThomas Graf 3479677e90edSThomas Graf next_skb: 348095e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3481677e90edSThomas Graf 3482995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 348395e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3484677e90edSThomas Graf return block_limit - abs_offset; 3485677e90edSThomas Graf } 3486677e90edSThomas Graf 3487677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3488677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3489677e90edSThomas Graf 3490677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 3491677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 34929e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 3493677e90edSThomas Graf 3494677e90edSThomas Graf if (abs_offset < block_limit) { 3495677e90edSThomas Graf if (!st->frag_data) 349651c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 3497677e90edSThomas Graf 3498b54c9d5bSJonathan Lemon *data = (u8 *) st->frag_data + skb_frag_off(frag) + 3499677e90edSThomas Graf (abs_offset - st->stepped_offset); 3500677e90edSThomas Graf 3501677e90edSThomas Graf return block_limit - abs_offset; 3502677e90edSThomas Graf } 3503677e90edSThomas Graf 3504677e90edSThomas Graf if (st->frag_data) { 350551c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3506677e90edSThomas Graf st->frag_data = NULL; 3507677e90edSThomas Graf } 3508677e90edSThomas Graf 3509677e90edSThomas Graf st->frag_idx++; 35109e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 3511677e90edSThomas Graf } 3512677e90edSThomas Graf 35135b5a60daSOlaf Kirch if (st->frag_data) { 351451c56b00SEric Dumazet kunmap_atomic(st->frag_data); 35155b5a60daSOlaf Kirch st->frag_data = NULL; 35165b5a60daSOlaf Kirch } 35175b5a60daSOlaf Kirch 351821dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3519677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 352095e3b24cSHerbert Xu st->frag_idx = 0; 3521677e90edSThomas Graf goto next_skb; 352271b3346dSShyam Iyer } else if (st->cur_skb->next) { 352371b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 352471b3346dSShyam Iyer st->frag_idx = 0; 3525677e90edSThomas Graf goto next_skb; 3526677e90edSThomas Graf } 3527677e90edSThomas Graf 3528677e90edSThomas Graf return 0; 3529677e90edSThomas Graf } 3530b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3531677e90edSThomas Graf 3532677e90edSThomas Graf /** 3533677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3534677e90edSThomas Graf * @st: state variable 3535677e90edSThomas Graf * 3536677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3537677e90edSThomas Graf * returned 0. 3538677e90edSThomas Graf */ 3539677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3540677e90edSThomas Graf { 3541677e90edSThomas Graf if (st->frag_data) 354251c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3543677e90edSThomas Graf } 3544b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3545677e90edSThomas Graf 35463fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 35473fc7e8a6SThomas Graf 35483fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 35493fc7e8a6SThomas Graf struct ts_config *conf, 35503fc7e8a6SThomas Graf struct ts_state *state) 35513fc7e8a6SThomas Graf { 35523fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 35533fc7e8a6SThomas Graf } 35543fc7e8a6SThomas Graf 35553fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 35563fc7e8a6SThomas Graf { 35573fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 35583fc7e8a6SThomas Graf } 35593fc7e8a6SThomas Graf 35603fc7e8a6SThomas Graf /** 35613fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 35623fc7e8a6SThomas Graf * @skb: the buffer to look in 35633fc7e8a6SThomas Graf * @from: search offset 35643fc7e8a6SThomas Graf * @to: search limit 35653fc7e8a6SThomas Graf * @config: textsearch configuration 35663fc7e8a6SThomas Graf * 35673fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 35683fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 35693fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 35703fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 35713fc7e8a6SThomas Graf */ 35723fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3573059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 35743fc7e8a6SThomas Graf { 3575059a2440SBojan Prtvar struct ts_state state; 3576f72b948dSPhil Oester unsigned int ret; 3577f72b948dSPhil Oester 35783fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 35793fc7e8a6SThomas Graf config->finish = skb_ts_finish; 35803fc7e8a6SThomas Graf 3581059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 35823fc7e8a6SThomas Graf 3583059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3584f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 35853fc7e8a6SThomas Graf } 3586b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 35873fc7e8a6SThomas Graf 3588be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3589be12a1feSHannes Frederic Sowa int offset, size_t size) 3590be12a1feSHannes Frederic Sowa { 3591be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3592be12a1feSHannes Frederic Sowa 3593be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3594be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3595be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3596be12a1feSHannes Frederic Sowa get_page(page); 3597be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3598be12a1feSHannes Frederic Sowa } else { 3599be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3600be12a1feSHannes Frederic Sowa } 3601be12a1feSHannes Frederic Sowa 3602be12a1feSHannes Frederic Sowa return 0; 3603be12a1feSHannes Frederic Sowa } 3604be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3605be12a1feSHannes Frederic Sowa 3606cbb042f9SHerbert Xu /** 3607cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3608cbb042f9SHerbert Xu * @skb: buffer to update 3609cbb042f9SHerbert Xu * @len: length of data pulled 3610cbb042f9SHerbert Xu * 3611cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3612fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 361384fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 361484fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 361584fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3616cbb042f9SHerbert Xu */ 3617af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3618cbb042f9SHerbert Xu { 361931b33dfbSPravin B Shelar unsigned char *data = skb->data; 362031b33dfbSPravin B Shelar 3621cbb042f9SHerbert Xu BUG_ON(len > skb->len); 362231b33dfbSPravin B Shelar __skb_pull(skb, len); 362331b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 362431b33dfbSPravin B Shelar return skb->data; 3625cbb042f9SHerbert Xu } 3626f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3627f94691acSArnaldo Carvalho de Melo 362813acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 362913acc94eSYonghong Song { 363013acc94eSYonghong Song skb_frag_t head_frag; 363113acc94eSYonghong Song struct page *page; 363213acc94eSYonghong Song 363313acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3634d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3635b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3636b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3637d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 363813acc94eSYonghong Song return head_frag; 363913acc94eSYonghong Song } 364013acc94eSYonghong Song 3641f4c50d99SHerbert Xu /** 3642f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3643df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3644576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3645f4c50d99SHerbert Xu * 3646f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 36474c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 36484c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3649f4c50d99SHerbert Xu */ 3650df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3651df5771ffSMichael S. Tsirkin netdev_features_t features) 3652f4c50d99SHerbert Xu { 3653f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3654f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 36551a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3656df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3657df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3658df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 36591fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3660f4c50d99SHerbert Xu unsigned int offset = doffset; 3661df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3662802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3663f4c50d99SHerbert Xu unsigned int headroom; 3664802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3665ec5f0615SPravin B Shelar __be16 proto; 366636c98382SAlexander Duyck bool csum, sg; 3667df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3668f4c50d99SHerbert Xu int err = -ENOMEM; 3669f4c50d99SHerbert Xu int i = 0; 3670f4c50d99SHerbert Xu int pos; 367153d6471cSVlad Yasevich int dummy; 3672f4c50d99SHerbert Xu 36733dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 36743dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 36753dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 36763dcbdb13SShmulik Ladkani * non head_frag head. 36773dcbdb13SShmulik Ladkani * 36783dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 36793dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 36803dcbdb13SShmulik Ladkani * head, then the first one has too). 36813dcbdb13SShmulik Ladkani * 36823dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 36833dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 36843dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 36853dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 36863dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 36873dcbdb13SShmulik Ladkani */ 36883dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 36893dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 36903dcbdb13SShmulik Ladkani } 36913dcbdb13SShmulik Ladkani 36925882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 369353d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 3694ec5f0615SPravin B Shelar if (unlikely(!proto)) 3695ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3696ec5f0615SPravin B Shelar 369736c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3698f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 36997e2b10c1STom Herbert 370007b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 370107b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 370207b26c94SSteffen Klassert struct sk_buff *iter; 370343170c4eSIlan Tayari unsigned int frag_len; 370407b26c94SSteffen Klassert 370507b26c94SSteffen Klassert if (!list_skb || 370607b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 370707b26c94SSteffen Klassert goto normal; 370807b26c94SSteffen Klassert 370943170c4eSIlan Tayari /* If we get here then all the required 371043170c4eSIlan Tayari * GSO features except frag_list are supported. 371143170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 371243170c4eSIlan Tayari * with no frag_list. 371343170c4eSIlan Tayari * Currently we can do that only when the buffers don't 371443170c4eSIlan Tayari * have a linear part and all the buffers except 371543170c4eSIlan Tayari * the last are of the same length. 371607b26c94SSteffen Klassert */ 371743170c4eSIlan Tayari frag_len = list_skb->len; 371807b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 371943170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 372043170c4eSIlan Tayari goto normal; 3721eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 372207b26c94SSteffen Klassert goto normal; 372307b26c94SSteffen Klassert 372407b26c94SSteffen Klassert len -= iter->len; 372507b26c94SSteffen Klassert } 372643170c4eSIlan Tayari 372743170c4eSIlan Tayari if (len != frag_len) 372843170c4eSIlan Tayari goto normal; 372907b26c94SSteffen Klassert } 373007b26c94SSteffen Klassert 3731802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 3732802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 3733802ab55aSAlexander Duyck * now. 3734802ab55aSAlexander Duyck */ 3735802ab55aSAlexander Duyck partial_segs = len / mss; 3736d7fb5a80SAlexander Duyck if (partial_segs > 1) 3737802ab55aSAlexander Duyck mss *= partial_segs; 3738d7fb5a80SAlexander Duyck else 3739d7fb5a80SAlexander Duyck partial_segs = 0; 3740802ab55aSAlexander Duyck } 3741802ab55aSAlexander Duyck 374207b26c94SSteffen Klassert normal: 3743df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3744df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3745f4c50d99SHerbert Xu 3746f4c50d99SHerbert Xu do { 3747f4c50d99SHerbert Xu struct sk_buff *nskb; 37488cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3749c8884eddSHerbert Xu int hsize; 3750f4c50d99SHerbert Xu int size; 3751f4c50d99SHerbert Xu 37523953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 37533953c46cSMarcelo Ricardo Leitner len = list_skb->len; 37543953c46cSMarcelo Ricardo Leitner } else { 3755df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3756f4c50d99SHerbert Xu if (len > mss) 3757f4c50d99SHerbert Xu len = mss; 37583953c46cSMarcelo Ricardo Leitner } 3759f4c50d99SHerbert Xu 3760df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3761f4c50d99SHerbert Xu if (hsize < 0) 3762f4c50d99SHerbert Xu hsize = 0; 3763c8884eddSHerbert Xu if (hsize > len || !sg) 3764c8884eddSHerbert Xu hsize = len; 3765f4c50d99SHerbert Xu 37661a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 37671a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 37681a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 376989319d38SHerbert Xu 37709d8506ccSHerbert Xu i = 0; 37711a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 37721a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 37731fd819ecSMichael S. Tsirkin frag_skb = list_skb; 37741a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 37759d8506ccSHerbert Xu 37769d8506ccSHerbert Xu while (pos < offset + len) { 37779d8506ccSHerbert Xu BUG_ON(i >= nfrags); 37789d8506ccSHerbert Xu 37794e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 37809d8506ccSHerbert Xu if (pos + size > offset + len) 37819d8506ccSHerbert Xu break; 37829d8506ccSHerbert Xu 37839d8506ccSHerbert Xu i++; 37849d8506ccSHerbert Xu pos += size; 37854e1beba1SMichael S. Tsirkin frag++; 37869d8506ccSHerbert Xu } 37879d8506ccSHerbert Xu 37881a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 37891a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 379089319d38SHerbert Xu 3791f4c50d99SHerbert Xu if (unlikely(!nskb)) 3792f4c50d99SHerbert Xu goto err; 3793f4c50d99SHerbert Xu 37949d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 37959d8506ccSHerbert Xu kfree_skb(nskb); 37969d8506ccSHerbert Xu goto err; 37979d8506ccSHerbert Xu } 37989d8506ccSHerbert Xu 3799ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 380089319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 380189319d38SHerbert Xu kfree_skb(nskb); 380289319d38SHerbert Xu goto err; 380389319d38SHerbert Xu } 380489319d38SHerbert Xu 3805ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 380689319d38SHerbert Xu skb_release_head_state(nskb); 380789319d38SHerbert Xu __skb_push(nskb, doffset); 380889319d38SHerbert Xu } else { 3809c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3810df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3811c93bdd0eSMel Gorman NUMA_NO_NODE); 381289319d38SHerbert Xu 381389319d38SHerbert Xu if (unlikely(!nskb)) 381489319d38SHerbert Xu goto err; 381589319d38SHerbert Xu 381689319d38SHerbert Xu skb_reserve(nskb, headroom); 381789319d38SHerbert Xu __skb_put(nskb, doffset); 381889319d38SHerbert Xu } 381989319d38SHerbert Xu 3820f4c50d99SHerbert Xu if (segs) 3821f4c50d99SHerbert Xu tail->next = nskb; 3822f4c50d99SHerbert Xu else 3823f4c50d99SHerbert Xu segs = nskb; 3824f4c50d99SHerbert Xu tail = nskb; 3825f4c50d99SHerbert Xu 3826df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3827f4c50d99SHerbert Xu 3828030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3829fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 383068c33163SPravin B Shelar 3831df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 383268c33163SPravin B Shelar nskb->data - tnl_hlen, 383368c33163SPravin B Shelar doffset + tnl_hlen); 383489319d38SHerbert Xu 38359d8506ccSHerbert Xu if (nskb->len == len + doffset) 38361cdbcb79SSimon Horman goto perform_csum_check; 383789319d38SHerbert Xu 38387fbeffedSAlexander Duyck if (!sg) { 38397fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 38406f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 384176443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 384276443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 3843f4c50d99SHerbert Xu skb_put(nskb, len), 3844f4c50d99SHerbert Xu len, 0); 38457e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3846de843723STom Herbert skb_headroom(nskb) + doffset; 3847f4c50d99SHerbert Xu continue; 3848f4c50d99SHerbert Xu } 3849f4c50d99SHerbert Xu 38508cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3851f4c50d99SHerbert Xu 3852df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3853d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3854f4c50d99SHerbert Xu 3855fff88030SWillem de Bruijn skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags & 3856df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3857cef401deSEric Dumazet 3858bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 3859bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 3860bf5c25d6SWillem de Bruijn goto err; 3861bf5c25d6SWillem de Bruijn 38629d8506ccSHerbert Xu while (pos < offset + len) { 38639d8506ccSHerbert Xu if (i >= nfrags) { 38649d8506ccSHerbert Xu i = 0; 38651a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 38661a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 38671fd819ecSMichael S. Tsirkin frag_skb = list_skb; 386813acc94eSYonghong Song if (!skb_headlen(list_skb)) { 38699d8506ccSHerbert Xu BUG_ON(!nfrags); 387013acc94eSYonghong Song } else { 387113acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 38729d8506ccSHerbert Xu 387313acc94eSYonghong Song /* to make room for head_frag. */ 387413acc94eSYonghong Song i--; 387513acc94eSYonghong Song frag--; 387613acc94eSYonghong Song } 3877bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 3878bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 3879bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 3880bf5c25d6SWillem de Bruijn goto err; 3881bf5c25d6SWillem de Bruijn 38821a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 38839d8506ccSHerbert Xu } 38849d8506ccSHerbert Xu 38859d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 38869d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 38879d8506ccSHerbert Xu net_warn_ratelimited( 38889d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 38899d8506ccSHerbert Xu pos, mss); 3890ff907a11SEric Dumazet err = -EINVAL; 38919d8506ccSHerbert Xu goto err; 38929d8506ccSHerbert Xu } 38939d8506ccSHerbert Xu 389413acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 38958cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 38968cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3897f4c50d99SHerbert Xu 3898f4c50d99SHerbert Xu if (pos < offset) { 3899b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 39008cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3901f4c50d99SHerbert Xu } 3902f4c50d99SHerbert Xu 390389319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3904f4c50d99SHerbert Xu 3905f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3906f4c50d99SHerbert Xu i++; 39074e1beba1SMichael S. Tsirkin frag++; 3908f4c50d99SHerbert Xu pos += size; 3909f4c50d99SHerbert Xu } else { 39108cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 391189319d38SHerbert Xu goto skip_fraglist; 3912f4c50d99SHerbert Xu } 3913f4c50d99SHerbert Xu 39148cb19905SMichael S. Tsirkin nskb_frag++; 3915f4c50d99SHerbert Xu } 3916f4c50d99SHerbert Xu 391789319d38SHerbert Xu skip_fraglist: 3918f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3919f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3920f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3921ec5f0615SPravin B Shelar 39221cdbcb79SSimon Horman perform_csum_check: 39237fbeffedSAlexander Duyck if (!csum) { 3924ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 3925ff907a11SEric Dumazet __skb_linearize(nskb)) 3926ddff00d4SAlexander Duyck goto err; 3927ff907a11SEric Dumazet 39287fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 3929ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 393076443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 393176443456SAlexander Duyck skb_checksum(nskb, doffset, 393276443456SAlexander Duyck nskb->len - doffset, 0); 39337e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 39347e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3935ec5f0615SPravin B Shelar } 3936df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3937f4c50d99SHerbert Xu 3938bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3939bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3940bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3941bec3cfdcSEric Dumazet */ 3942bec3cfdcSEric Dumazet segs->prev = tail; 3943432c856fSToshiaki Makita 3944802ab55aSAlexander Duyck if (partial_segs) { 394507b26c94SSteffen Klassert struct sk_buff *iter; 3946802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 394707b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 3948802ab55aSAlexander Duyck 3949802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 395007b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 3951802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 3952802ab55aSAlexander Duyck 3953802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 3954802ab55aSAlexander Duyck * our way back down the stack of protocols. 3955802ab55aSAlexander Duyck */ 395607b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 395707b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 395807b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 395907b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 396007b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 396107b26c94SSteffen Klassert } 396207b26c94SSteffen Klassert 396307b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 396407b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 396507b26c94SSteffen Klassert else if (tail != segs) 396607b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 3967802ab55aSAlexander Duyck } 3968802ab55aSAlexander Duyck 3969432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 3970432c856fSToshiaki Makita * using skb_set_owner_w(). 3971432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 3972432c856fSToshiaki Makita */ 3973432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 3974432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 3975432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 3976432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 3977432c856fSToshiaki Makita } 3978f4c50d99SHerbert Xu return segs; 3979f4c50d99SHerbert Xu 3980f4c50d99SHerbert Xu err: 3981289dccbeSEric Dumazet kfree_skb_list(segs); 3982f4c50d99SHerbert Xu return ERR_PTR(err); 3983f4c50d99SHerbert Xu } 3984f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3985f4c50d99SHerbert Xu 3986d4546c25SDavid Miller int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb) 398771d93b39SHerbert Xu { 39888a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 398967147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 399067147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 39918a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 3992715dc1f3SEric Dumazet unsigned int delta_truesize; 3993d4546c25SDavid Miller struct sk_buff *lp; 399471d93b39SHerbert Xu 39950ab03f35SSteffen Klassert if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush)) 399671d93b39SHerbert Xu return -E2BIG; 399771d93b39SHerbert Xu 399829e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 39998a29111cSEric Dumazet pinfo = skb_shinfo(lp); 40008a29111cSEric Dumazet 40018a29111cSEric Dumazet if (headlen <= offset) { 400242da6994SHerbert Xu skb_frag_t *frag; 400366e92fcfSHerbert Xu skb_frag_t *frag2; 40049aaa156cSHerbert Xu int i = skbinfo->nr_frags; 40059aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 400642da6994SHerbert Xu 400766e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 40088a29111cSEric Dumazet goto merge; 400981705ad1SHerbert Xu 40108a29111cSEric Dumazet offset -= headlen; 40119aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 40129aaa156cSHerbert Xu skbinfo->nr_frags = 0; 4013f5572068SHerbert Xu 40149aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 40159aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 401666e92fcfSHerbert Xu do { 401766e92fcfSHerbert Xu *--frag = *--frag2; 401866e92fcfSHerbert Xu } while (--i); 401966e92fcfSHerbert Xu 4020b54c9d5bSJonathan Lemon skb_frag_off_add(frag, offset); 40219e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 402266e92fcfSHerbert Xu 4023715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 4024ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 4025ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 4026715dc1f3SEric Dumazet 4027f5572068SHerbert Xu skb->truesize -= skb->data_len; 4028f5572068SHerbert Xu skb->len -= skb->data_len; 4029f5572068SHerbert Xu skb->data_len = 0; 4030f5572068SHerbert Xu 4031715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 40325d38a079SHerbert Xu goto done; 4033d7e8883cSEric Dumazet } else if (skb->head_frag) { 4034d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 4035d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 4036d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 4037d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 4038d7e8883cSEric Dumazet unsigned int first_offset; 4039d7e8883cSEric Dumazet 4040d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 40418a29111cSEric Dumazet goto merge; 4042d7e8883cSEric Dumazet 4043d7e8883cSEric Dumazet first_offset = skb->data - 4044d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 4045d7e8883cSEric Dumazet offset; 4046d7e8883cSEric Dumazet 4047d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 4048d7e8883cSEric Dumazet 4049d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(frag, page); 4050b54c9d5bSJonathan Lemon skb_frag_off_set(frag, first_offset); 4051d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 4052d7e8883cSEric Dumazet 4053d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 4054d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 4055d7e8883cSEric Dumazet 4056715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4057d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 4058d7e8883cSEric Dumazet goto done; 40598a29111cSEric Dumazet } 406071d93b39SHerbert Xu 406171d93b39SHerbert Xu merge: 4062715dc1f3SEric Dumazet delta_truesize = skb->truesize; 406367147ba9SHerbert Xu if (offset > headlen) { 4064d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 4065d1dc7abfSMichal Schmidt 4066b54c9d5bSJonathan Lemon skb_frag_off_add(&skbinfo->frags[0], eat); 40679e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 4068d1dc7abfSMichal Schmidt skb->data_len -= eat; 4069d1dc7abfSMichal Schmidt skb->len -= eat; 407067147ba9SHerbert Xu offset = headlen; 407156035022SHerbert Xu } 407256035022SHerbert Xu 407367147ba9SHerbert Xu __skb_pull(skb, offset); 407456035022SHerbert Xu 407529e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 40768a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 40778a29111cSEric Dumazet else 4078c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 4079c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 4080f4a775d1SEric Dumazet __skb_header_release(skb); 40818a29111cSEric Dumazet lp = p; 408271d93b39SHerbert Xu 40835d38a079SHerbert Xu done: 40845d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 408537fe4732SHerbert Xu p->data_len += len; 4086715dc1f3SEric Dumazet p->truesize += delta_truesize; 408737fe4732SHerbert Xu p->len += len; 40888a29111cSEric Dumazet if (lp != p) { 40898a29111cSEric Dumazet lp->data_len += len; 40908a29111cSEric Dumazet lp->truesize += delta_truesize; 40918a29111cSEric Dumazet lp->len += len; 40928a29111cSEric Dumazet } 409371d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 409471d93b39SHerbert Xu return 0; 409571d93b39SHerbert Xu } 409657c05650SMarcelo Ricardo Leitner EXPORT_SYMBOL_GPL(skb_gro_receive); 409771d93b39SHerbert Xu 4098df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4099df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4100df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4101df5042f4SFlorian Westphal 4102df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4103df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4104df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4105df5042f4SFlorian Westphal #endif 41064165079bSFlorian Westphal #ifdef CONFIG_XFRM 41074165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 41084165079bSFlorian Westphal #endif 410995a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 411095a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 411195a7233cSPaul Blakey #endif 4112df5042f4SFlorian Westphal }; 4113df5042f4SFlorian Westphal 4114df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4115df5042f4SFlorian Westphal { 4116df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4117df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4118df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4119df5042f4SFlorian Westphal #endif 41204165079bSFlorian Westphal #ifdef CONFIG_XFRM 41214165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 41224165079bSFlorian Westphal #endif 412395a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 412495a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 412595a7233cSPaul Blakey #endif 4126df5042f4SFlorian Westphal 0; 4127df5042f4SFlorian Westphal } 4128df5042f4SFlorian Westphal 4129df5042f4SFlorian Westphal static void skb_extensions_init(void) 4130df5042f4SFlorian Westphal { 4131df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4132df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4133df5042f4SFlorian Westphal 4134df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4135df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4136df5042f4SFlorian Westphal 0, 4137df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4138df5042f4SFlorian Westphal NULL); 4139df5042f4SFlorian Westphal } 4140df5042f4SFlorian Westphal #else 4141df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4142df5042f4SFlorian Westphal #endif 4143df5042f4SFlorian Westphal 41441da177e4SLinus Torvalds void __init skb_init(void) 41451da177e4SLinus Torvalds { 414679a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 41471da177e4SLinus Torvalds sizeof(struct sk_buff), 41481da177e4SLinus Torvalds 0, 4149e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 415079a8a642SKees Cook offsetof(struct sk_buff, cb), 415179a8a642SKees Cook sizeof_field(struct sk_buff, cb), 415220c2df83SPaul Mundt NULL); 4153d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4154d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4155d179cd12SDavid S. Miller 0, 4156e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 415720c2df83SPaul Mundt NULL); 4158df5042f4SFlorian Westphal skb_extensions_init(); 41591da177e4SLinus Torvalds } 41601da177e4SLinus Torvalds 416151c739d1SDavid S. Miller static int 416248a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 416348a1df65SJason A. Donenfeld unsigned int recursion_level) 4164716ea3a7SDavid Howells { 41651a028e50SDavid S. Miller int start = skb_headlen(skb); 41661a028e50SDavid S. Miller int i, copy = start - offset; 4167fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4168716ea3a7SDavid Howells int elt = 0; 4169716ea3a7SDavid Howells 417048a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 417148a1df65SJason A. Donenfeld return -EMSGSIZE; 417248a1df65SJason A. Donenfeld 4173716ea3a7SDavid Howells if (copy > 0) { 4174716ea3a7SDavid Howells if (copy > len) 4175716ea3a7SDavid Howells copy = len; 4176642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4177716ea3a7SDavid Howells elt++; 4178716ea3a7SDavid Howells if ((len -= copy) == 0) 4179716ea3a7SDavid Howells return elt; 4180716ea3a7SDavid Howells offset += copy; 4181716ea3a7SDavid Howells } 4182716ea3a7SDavid Howells 4183716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 41841a028e50SDavid S. Miller int end; 4185716ea3a7SDavid Howells 4186547b792cSIlpo Järvinen WARN_ON(start > offset + len); 41871a028e50SDavid S. Miller 41889e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4189716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4190716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 419148a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 419248a1df65SJason A. Donenfeld return -EMSGSIZE; 4193716ea3a7SDavid Howells 4194716ea3a7SDavid Howells if (copy > len) 4195716ea3a7SDavid Howells copy = len; 4196ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4197b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4198716ea3a7SDavid Howells elt++; 4199716ea3a7SDavid Howells if (!(len -= copy)) 4200716ea3a7SDavid Howells return elt; 4201716ea3a7SDavid Howells offset += copy; 4202716ea3a7SDavid Howells } 42031a028e50SDavid S. Miller start = end; 4204716ea3a7SDavid Howells } 4205716ea3a7SDavid Howells 4206fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 420748a1df65SJason A. Donenfeld int end, ret; 4208716ea3a7SDavid Howells 4209547b792cSIlpo Järvinen WARN_ON(start > offset + len); 42101a028e50SDavid S. Miller 4211fbb398a8SDavid S. Miller end = start + frag_iter->len; 4212716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 421348a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 421448a1df65SJason A. Donenfeld return -EMSGSIZE; 421548a1df65SJason A. Donenfeld 4216716ea3a7SDavid Howells if (copy > len) 4217716ea3a7SDavid Howells copy = len; 421848a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 421948a1df65SJason A. Donenfeld copy, recursion_level + 1); 422048a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 422148a1df65SJason A. Donenfeld return ret; 422248a1df65SJason A. Donenfeld elt += ret; 4223716ea3a7SDavid Howells if ((len -= copy) == 0) 4224716ea3a7SDavid Howells return elt; 4225716ea3a7SDavid Howells offset += copy; 4226716ea3a7SDavid Howells } 42271a028e50SDavid S. Miller start = end; 4228716ea3a7SDavid Howells } 4229716ea3a7SDavid Howells BUG_ON(len); 4230716ea3a7SDavid Howells return elt; 4231716ea3a7SDavid Howells } 4232716ea3a7SDavid Howells 423348a1df65SJason A. Donenfeld /** 423448a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 423548a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 423648a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 423748a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 423848a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 423948a1df65SJason A. Donenfeld * 424048a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 424148a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 424248a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 424348a1df65SJason A. Donenfeld * could not fit. 424448a1df65SJason A. Donenfeld */ 424548a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 424648a1df65SJason A. Donenfeld { 424748a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 424848a1df65SJason A. Donenfeld 424948a1df65SJason A. Donenfeld if (nsg <= 0) 425048a1df65SJason A. Donenfeld return nsg; 425148a1df65SJason A. Donenfeld 425248a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 425348a1df65SJason A. Donenfeld 425448a1df65SJason A. Donenfeld return nsg; 425548a1df65SJason A. Donenfeld } 425648a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 425748a1df65SJason A. Donenfeld 425825a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 425925a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 426025a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 426125a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 426225a91d8dSFan Du * 426325a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 426425a91d8dSFan Du * 1. sg_init_table 426525a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 426625a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 426725a91d8dSFan Du * 426825a91d8dSFan Du * This is equivalent to: 426925a91d8dSFan Du * 1. sg_init_table 427025a91d8dSFan Du * 2. skb_to_sgvec(payload1) 427125a91d8dSFan Du * 3. sg_unmark_end 427225a91d8dSFan Du * 4. skb_to_sgvec(payload2) 427325a91d8dSFan Du * 427425a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 427525a91d8dSFan Du * is more preferable. 427625a91d8dSFan Du */ 427725a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 427825a91d8dSFan Du int offset, int len) 427925a91d8dSFan Du { 428048a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 428125a91d8dSFan Du } 428225a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 428325a91d8dSFan Du 428451c739d1SDavid S. Miller 428551c739d1SDavid S. Miller 4286716ea3a7SDavid Howells /** 4287716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4288716ea3a7SDavid Howells * @skb: The socket buffer to check. 4289716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4290716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4291716ea3a7SDavid Howells * 4292716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4293716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4294716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4295716ea3a7SDavid Howells * 4296716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4297716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4298716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4299716ea3a7SDavid Howells * 4300716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4301716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4302716ea3a7SDavid Howells */ 4303716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4304716ea3a7SDavid Howells { 4305716ea3a7SDavid Howells int copyflag; 4306716ea3a7SDavid Howells int elt; 4307716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4308716ea3a7SDavid Howells 4309716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4310716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4311716ea3a7SDavid Howells * at the moment even if they are anonymous). 4312716ea3a7SDavid Howells */ 4313716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4314716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 4315716ea3a7SDavid Howells return -ENOMEM; 4316716ea3a7SDavid Howells 4317716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 431821dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4319716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4320716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4321716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4322716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4323716ea3a7SDavid Howells 4324716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4325716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4326716ea3a7SDavid Howells return -ENOMEM; 4327716ea3a7SDavid Howells 4328716ea3a7SDavid Howells /* Voila! */ 4329716ea3a7SDavid Howells *trailer = skb; 4330716ea3a7SDavid Howells return 1; 4331716ea3a7SDavid Howells } 4332716ea3a7SDavid Howells 4333716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4334716ea3a7SDavid Howells 4335716ea3a7SDavid Howells elt = 1; 4336716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4337716ea3a7SDavid Howells copyflag = 0; 4338716ea3a7SDavid Howells 4339716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4340716ea3a7SDavid Howells int ntail = 0; 4341716ea3a7SDavid Howells 4342716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4343716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4344716ea3a7SDavid Howells * after it. */ 4345716ea3a7SDavid Howells 4346716ea3a7SDavid Howells if (skb_shared(skb1)) 4347716ea3a7SDavid Howells copyflag = 1; 4348716ea3a7SDavid Howells 4349716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4350716ea3a7SDavid Howells 4351716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4352716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 435321dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4354716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4355716ea3a7SDavid Howells ntail = tailbits + 128; 4356716ea3a7SDavid Howells } 4357716ea3a7SDavid Howells 4358716ea3a7SDavid Howells if (copyflag || 4359716ea3a7SDavid Howells skb_cloned(skb1) || 4360716ea3a7SDavid Howells ntail || 4361716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 436221dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4363716ea3a7SDavid Howells struct sk_buff *skb2; 4364716ea3a7SDavid Howells 4365716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4366716ea3a7SDavid Howells if (ntail == 0) 4367716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4368716ea3a7SDavid Howells else 4369716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4370716ea3a7SDavid Howells skb_headroom(skb1), 4371716ea3a7SDavid Howells ntail, 4372716ea3a7SDavid Howells GFP_ATOMIC); 4373716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4374716ea3a7SDavid Howells return -ENOMEM; 4375716ea3a7SDavid Howells 4376716ea3a7SDavid Howells if (skb1->sk) 4377716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4378716ea3a7SDavid Howells 4379716ea3a7SDavid Howells /* Looking around. Are we still alive? 4380716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4381716ea3a7SDavid Howells 4382716ea3a7SDavid Howells skb2->next = skb1->next; 4383716ea3a7SDavid Howells *skb_p = skb2; 4384716ea3a7SDavid Howells kfree_skb(skb1); 4385716ea3a7SDavid Howells skb1 = skb2; 4386716ea3a7SDavid Howells } 4387716ea3a7SDavid Howells elt++; 4388716ea3a7SDavid Howells *trailer = skb1; 4389716ea3a7SDavid Howells skb_p = &skb1->next; 4390716ea3a7SDavid Howells } 4391716ea3a7SDavid Howells 4392716ea3a7SDavid Howells return elt; 4393716ea3a7SDavid Howells } 4394b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4395716ea3a7SDavid Howells 4396b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4397b1faf566SEric Dumazet { 4398b1faf566SEric Dumazet struct sock *sk = skb->sk; 4399b1faf566SEric Dumazet 4400b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4401b1faf566SEric Dumazet } 4402b1faf566SEric Dumazet 44038605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 44048605330aSSoheil Hassas Yeganeh { 44058605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 44068605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 44078605330aSSoheil Hassas Yeganeh */ 44088605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 44098605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 44108605330aSSoheil Hassas Yeganeh } 44118605330aSSoheil Hassas Yeganeh 4412b1faf566SEric Dumazet /* 4413b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4414b1faf566SEric Dumazet */ 4415b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4416b1faf566SEric Dumazet { 4417b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4418ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4419b1faf566SEric Dumazet return -ENOMEM; 4420b1faf566SEric Dumazet 4421b1faf566SEric Dumazet skb_orphan(skb); 4422b1faf566SEric Dumazet skb->sk = sk; 4423b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4424b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 44258605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4426b1faf566SEric Dumazet 4427abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4428abb57ea4SEric Dumazet skb_dst_force(skb); 4429abb57ea4SEric Dumazet 4430b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4431b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 44326e5d58fdSVinicius Costa Gomes sk->sk_error_report(sk); 4433b1faf566SEric Dumazet return 0; 4434b1faf566SEric Dumazet } 4435b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4436b1faf566SEric Dumazet 443783a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 443883a1a1a7SSoheil Hassas Yeganeh { 443983a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 444083a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 444183a1a1a7SSoheil Hassas Yeganeh } 444283a1a1a7SSoheil Hassas Yeganeh 4443364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4444364a9e93SWillem de Bruijn { 4445364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 444683a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 444783a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4448997d5c3fSEric Dumazet unsigned long flags; 4449364a9e93SWillem de Bruijn 4450997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4451364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 445238b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 445383a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 445438b25793SSoheil Hassas Yeganeh if (icmp_next) 445538b25793SSoheil Hassas Yeganeh sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin; 445638b25793SSoheil Hassas Yeganeh } 4457997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4458364a9e93SWillem de Bruijn 445983a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 446083a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 446183a1a1a7SSoheil Hassas Yeganeh 446283a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4463364a9e93SWillem de Bruijn sk->sk_error_report(sk); 4464364a9e93SWillem de Bruijn 4465364a9e93SWillem de Bruijn return skb; 4466364a9e93SWillem de Bruijn } 4467364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4468364a9e93SWillem de Bruijn 4469cab41c47SAlexander Duyck /** 4470cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4471cab41c47SAlexander Duyck * @skb: the skb to clone 4472cab41c47SAlexander Duyck * 4473cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4474cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4475cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4476cab41c47SAlexander Duyck * 4477cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4478cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4479cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4480cab41c47SAlexander Duyck * the sk_error_queue. 4481cab41c47SAlexander Duyck */ 448262bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 448362bccb8cSAlexander Duyck { 448462bccb8cSAlexander Duyck struct sock *sk = skb->sk; 448562bccb8cSAlexander Duyck struct sk_buff *clone; 448662bccb8cSAlexander Duyck 448741c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 448862bccb8cSAlexander Duyck return NULL; 448962bccb8cSAlexander Duyck 449062bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 449162bccb8cSAlexander Duyck if (!clone) { 449262bccb8cSAlexander Duyck sock_put(sk); 449362bccb8cSAlexander Duyck return NULL; 449462bccb8cSAlexander Duyck } 449562bccb8cSAlexander Duyck 449662bccb8cSAlexander Duyck clone->sk = sk; 449762bccb8cSAlexander Duyck clone->destructor = sock_efree; 449862bccb8cSAlexander Duyck 449962bccb8cSAlexander Duyck return clone; 450062bccb8cSAlexander Duyck } 450162bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 450262bccb8cSAlexander Duyck 450337846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 450437846ef0SAlexander Duyck struct sock *sk, 45054ef1b286SSoheil Hassas Yeganeh int tstype, 45064ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4507ac45f602SPatrick Ohly { 4508ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4509ac45f602SPatrick Ohly int err; 4510ac45f602SPatrick Ohly 45114ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 45124ef1b286SSoheil Hassas Yeganeh 4513ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4514ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4515ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4516ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4517e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 45184ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 45191862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 45204ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 452109c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 4522ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 4523ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 45244ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 45254ed2d765SWillem de Bruijn } 452629030374SEric Dumazet 4527ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 452829030374SEric Dumazet 4529ac45f602SPatrick Ohly if (err) 4530ac45f602SPatrick Ohly kfree_skb(skb); 4531ac45f602SPatrick Ohly } 453237846ef0SAlexander Duyck 4533b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4534b245be1fSWillem de Bruijn { 4535b245be1fSWillem de Bruijn bool ret; 4536b245be1fSWillem de Bruijn 4537b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4538b245be1fSWillem de Bruijn return true; 4539b245be1fSWillem de Bruijn 4540b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4541b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4542b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4543b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4544b245be1fSWillem de Bruijn return ret; 4545b245be1fSWillem de Bruijn } 4546b245be1fSWillem de Bruijn 454737846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 454837846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 454937846ef0SAlexander Duyck { 455037846ef0SAlexander Duyck struct sock *sk = skb->sk; 455137846ef0SAlexander Duyck 4552b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 455335b99dffSWillem de Bruijn goto err; 4554b245be1fSWillem de Bruijn 45559ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 45569ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 45579ac25fc0SEric Dumazet */ 455841c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 455937846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 45604ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 456137846ef0SAlexander Duyck sock_put(sk); 456235b99dffSWillem de Bruijn return; 456337846ef0SAlexander Duyck } 456435b99dffSWillem de Bruijn 456535b99dffSWillem de Bruijn err: 456635b99dffSWillem de Bruijn kfree_skb(skb); 45679ac25fc0SEric Dumazet } 456837846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 456937846ef0SAlexander Duyck 457037846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 457137846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 457237846ef0SAlexander Duyck struct sock *sk, int tstype) 457337846ef0SAlexander Duyck { 457437846ef0SAlexander Duyck struct sk_buff *skb; 45754ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 457637846ef0SAlexander Duyck 45773a8dd971SWillem de Bruijn if (!sk) 45783a8dd971SWillem de Bruijn return; 45793a8dd971SWillem de Bruijn 4580b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4581b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4582b50a5c70SMiroslav Lichvar return; 4583b50a5c70SMiroslav Lichvar 45843a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 45853a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 458637846ef0SAlexander Duyck return; 458737846ef0SAlexander Duyck 45881c885808SFrancis Yan if (tsonly) { 45891c885808SFrancis Yan #ifdef CONFIG_INET 45901c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 45911c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 45924ef1b286SSoheil Hassas Yeganeh sk->sk_type == SOCK_STREAM) { 45931c885808SFrancis Yan skb = tcp_get_timestamping_opt_stats(sk); 45944ef1b286SSoheil Hassas Yeganeh opt_stats = true; 45954ef1b286SSoheil Hassas Yeganeh } else 45961c885808SFrancis Yan #endif 45971c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 45981c885808SFrancis Yan } else { 459937846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 46001c885808SFrancis Yan } 460137846ef0SAlexander Duyck if (!skb) 460237846ef0SAlexander Duyck return; 460337846ef0SAlexander Duyck 460449ca0d8bSWillem de Bruijn if (tsonly) { 4605fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4606fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 460749ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 460849ca0d8bSWillem de Bruijn } 460949ca0d8bSWillem de Bruijn 461049ca0d8bSWillem de Bruijn if (hwtstamps) 461149ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 461249ca0d8bSWillem de Bruijn else 461349ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 461449ca0d8bSWillem de Bruijn 46154ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 461637846ef0SAlexander Duyck } 4617e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4618e7fd2885SWillem de Bruijn 4619e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4620e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4621e7fd2885SWillem de Bruijn { 4622e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 4623e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4624e7fd2885SWillem de Bruijn } 4625ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4626ac45f602SPatrick Ohly 46276e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 46286e3e939fSJohannes Berg { 46296e3e939fSJohannes Berg struct sock *sk = skb->sk; 46306e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4631dd4f1072SEric Dumazet int err = 1; 46326e3e939fSJohannes Berg 46336e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 46346e3e939fSJohannes Berg skb->wifi_acked = acked; 46356e3e939fSJohannes Berg 46366e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 46376e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 46386e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 46396e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 46406e3e939fSJohannes Berg 4641dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4642dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4643dd4f1072SEric Dumazet */ 464441c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 46456e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4646dd4f1072SEric Dumazet sock_put(sk); 4647dd4f1072SEric Dumazet } 46486e3e939fSJohannes Berg if (err) 46496e3e939fSJohannes Berg kfree_skb(skb); 46506e3e939fSJohannes Berg } 46516e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 46526e3e939fSJohannes Berg 4653f35d9d8aSRusty Russell /** 4654f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4655f35d9d8aSRusty Russell * @skb: the skb to set 4656f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4657f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4658f35d9d8aSRusty Russell * 4659f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4660f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4661f35d9d8aSRusty Russell * 4662f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4663f35d9d8aSRusty Russell * returns false you should drop the packet. 4664f35d9d8aSRusty Russell */ 4665f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4666f35d9d8aSRusty Russell { 466752b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 466852b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 466952b5d6f5SEric Dumazet 467052b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 467152b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 467252b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4673f35d9d8aSRusty Russell return false; 4674f35d9d8aSRusty Russell } 4675f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 467652b5d6f5SEric Dumazet skb->csum_start = csum_start; 4677f35d9d8aSRusty Russell skb->csum_offset = off; 4678e5d5decaSJason Wang skb_set_transport_header(skb, start); 4679f35d9d8aSRusty Russell return true; 4680f35d9d8aSRusty Russell } 4681b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4682f35d9d8aSRusty Russell 4683ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4684ed1f50c3SPaul Durrant unsigned int max) 4685ed1f50c3SPaul Durrant { 4686ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4687ed1f50c3SPaul Durrant return 0; 4688ed1f50c3SPaul Durrant 4689ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4690ed1f50c3SPaul Durrant * won't need to do it again. 4691ed1f50c3SPaul Durrant */ 4692ed1f50c3SPaul Durrant if (max > skb->len) 4693ed1f50c3SPaul Durrant max = skb->len; 4694ed1f50c3SPaul Durrant 4695ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4696ed1f50c3SPaul Durrant return -ENOMEM; 4697ed1f50c3SPaul Durrant 4698ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4699ed1f50c3SPaul Durrant return -EPROTO; 4700ed1f50c3SPaul Durrant 4701ed1f50c3SPaul Durrant return 0; 4702ed1f50c3SPaul Durrant } 4703ed1f50c3SPaul Durrant 4704f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4705f9708b43SJan Beulich 4706f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4707f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4708f9708b43SJan Beulich unsigned int off) 4709f9708b43SJan Beulich { 4710f9708b43SJan Beulich switch (proto) { 4711f9708b43SJan Beulich int err; 4712f9708b43SJan Beulich 4713f9708b43SJan Beulich case IPPROTO_TCP: 4714f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4715f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4716f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4717f9708b43SJan Beulich offsetof(struct tcphdr, 4718f9708b43SJan Beulich check))) 4719f9708b43SJan Beulich err = -EPROTO; 4720f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4721f9708b43SJan Beulich 4722f9708b43SJan Beulich case IPPROTO_UDP: 4723f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4724f9708b43SJan Beulich off + sizeof(struct udphdr)); 4725f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4726f9708b43SJan Beulich offsetof(struct udphdr, 4727f9708b43SJan Beulich check))) 4728f9708b43SJan Beulich err = -EPROTO; 4729f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4730f9708b43SJan Beulich } 4731f9708b43SJan Beulich 4732f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4733f9708b43SJan Beulich } 4734f9708b43SJan Beulich 4735ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4736ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4737ed1f50c3SPaul Durrant */ 4738ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4739ed1f50c3SPaul Durrant 4740f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4741ed1f50c3SPaul Durrant { 4742ed1f50c3SPaul Durrant unsigned int off; 4743ed1f50c3SPaul Durrant bool fragment; 4744f9708b43SJan Beulich __sum16 *csum; 4745ed1f50c3SPaul Durrant int err; 4746ed1f50c3SPaul Durrant 4747ed1f50c3SPaul Durrant fragment = false; 4748ed1f50c3SPaul Durrant 4749ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4750ed1f50c3SPaul Durrant sizeof(struct iphdr), 4751ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4752ed1f50c3SPaul Durrant if (err < 0) 4753ed1f50c3SPaul Durrant goto out; 4754ed1f50c3SPaul Durrant 4755ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 4756ed1f50c3SPaul Durrant fragment = true; 4757ed1f50c3SPaul Durrant 4758ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 4759ed1f50c3SPaul Durrant 4760ed1f50c3SPaul Durrant err = -EPROTO; 4761ed1f50c3SPaul Durrant 4762ed1f50c3SPaul Durrant if (fragment) 4763ed1f50c3SPaul Durrant goto out; 4764ed1f50c3SPaul Durrant 4765f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4766f9708b43SJan Beulich if (IS_ERR(csum)) 4767f9708b43SJan Beulich return PTR_ERR(csum); 4768ed1f50c3SPaul Durrant 4769ed1f50c3SPaul Durrant if (recalculate) 4770f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4771ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4772ed1f50c3SPaul Durrant skb->len - off, 4773f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4774ed1f50c3SPaul Durrant err = 0; 4775ed1f50c3SPaul Durrant 4776ed1f50c3SPaul Durrant out: 4777ed1f50c3SPaul Durrant return err; 4778ed1f50c3SPaul Durrant } 4779ed1f50c3SPaul Durrant 4780ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4781ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4782ed1f50c3SPaul Durrant */ 4783ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4784ed1f50c3SPaul Durrant 4785ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4786ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4787ed1f50c3SPaul Durrant 4788ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4789ed1f50c3SPaul Durrant { 4790ed1f50c3SPaul Durrant int err; 4791ed1f50c3SPaul Durrant u8 nexthdr; 4792ed1f50c3SPaul Durrant unsigned int off; 4793ed1f50c3SPaul Durrant unsigned int len; 4794ed1f50c3SPaul Durrant bool fragment; 4795ed1f50c3SPaul Durrant bool done; 4796f9708b43SJan Beulich __sum16 *csum; 4797ed1f50c3SPaul Durrant 4798ed1f50c3SPaul Durrant fragment = false; 4799ed1f50c3SPaul Durrant done = false; 4800ed1f50c3SPaul Durrant 4801ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4802ed1f50c3SPaul Durrant 4803ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4804ed1f50c3SPaul Durrant if (err < 0) 4805ed1f50c3SPaul Durrant goto out; 4806ed1f50c3SPaul Durrant 4807ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4808ed1f50c3SPaul Durrant 4809ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4810ed1f50c3SPaul Durrant while (off <= len && !done) { 4811ed1f50c3SPaul Durrant switch (nexthdr) { 4812ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4813ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4814ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4815ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 4816ed1f50c3SPaul Durrant 4817ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4818ed1f50c3SPaul Durrant off + 4819ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 4820ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4821ed1f50c3SPaul Durrant if (err < 0) 4822ed1f50c3SPaul Durrant goto out; 4823ed1f50c3SPaul Durrant 4824ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4825ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4826ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 4827ed1f50c3SPaul Durrant break; 4828ed1f50c3SPaul Durrant } 4829ed1f50c3SPaul Durrant case IPPROTO_AH: { 4830ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 4831ed1f50c3SPaul Durrant 4832ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4833ed1f50c3SPaul Durrant off + 4834ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 4835ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4836ed1f50c3SPaul Durrant if (err < 0) 4837ed1f50c3SPaul Durrant goto out; 4838ed1f50c3SPaul Durrant 4839ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4840ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4841ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 4842ed1f50c3SPaul Durrant break; 4843ed1f50c3SPaul Durrant } 4844ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 4845ed1f50c3SPaul Durrant struct frag_hdr *hp; 4846ed1f50c3SPaul Durrant 4847ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4848ed1f50c3SPaul Durrant off + 4849ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 4850ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4851ed1f50c3SPaul Durrant if (err < 0) 4852ed1f50c3SPaul Durrant goto out; 4853ed1f50c3SPaul Durrant 4854ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 4855ed1f50c3SPaul Durrant 4856ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 4857ed1f50c3SPaul Durrant fragment = true; 4858ed1f50c3SPaul Durrant 4859ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4860ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 4861ed1f50c3SPaul Durrant break; 4862ed1f50c3SPaul Durrant } 4863ed1f50c3SPaul Durrant default: 4864ed1f50c3SPaul Durrant done = true; 4865ed1f50c3SPaul Durrant break; 4866ed1f50c3SPaul Durrant } 4867ed1f50c3SPaul Durrant } 4868ed1f50c3SPaul Durrant 4869ed1f50c3SPaul Durrant err = -EPROTO; 4870ed1f50c3SPaul Durrant 4871ed1f50c3SPaul Durrant if (!done || fragment) 4872ed1f50c3SPaul Durrant goto out; 4873ed1f50c3SPaul Durrant 4874f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 4875f9708b43SJan Beulich if (IS_ERR(csum)) 4876f9708b43SJan Beulich return PTR_ERR(csum); 4877ed1f50c3SPaul Durrant 4878ed1f50c3SPaul Durrant if (recalculate) 4879f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 4880ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 4881f9708b43SJan Beulich skb->len - off, nexthdr, 0); 4882ed1f50c3SPaul Durrant err = 0; 4883ed1f50c3SPaul Durrant 4884ed1f50c3SPaul Durrant out: 4885ed1f50c3SPaul Durrant return err; 4886ed1f50c3SPaul Durrant } 4887ed1f50c3SPaul Durrant 4888ed1f50c3SPaul Durrant /** 4889ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 4890ed1f50c3SPaul Durrant * @skb: the skb to set up 4891ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 4892ed1f50c3SPaul Durrant */ 4893ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 4894ed1f50c3SPaul Durrant { 4895ed1f50c3SPaul Durrant int err; 4896ed1f50c3SPaul Durrant 4897ed1f50c3SPaul Durrant switch (skb->protocol) { 4898ed1f50c3SPaul Durrant case htons(ETH_P_IP): 4899f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 4900ed1f50c3SPaul Durrant break; 4901ed1f50c3SPaul Durrant 4902ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 4903ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 4904ed1f50c3SPaul Durrant break; 4905ed1f50c3SPaul Durrant 4906ed1f50c3SPaul Durrant default: 4907ed1f50c3SPaul Durrant err = -EPROTO; 4908ed1f50c3SPaul Durrant break; 4909ed1f50c3SPaul Durrant } 4910ed1f50c3SPaul Durrant 4911ed1f50c3SPaul Durrant return err; 4912ed1f50c3SPaul Durrant } 4913ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 4914ed1f50c3SPaul Durrant 49159afd85c9SLinus Lüssing /** 49169afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 49179afd85c9SLinus Lüssing * @skb: the skb to check 49189afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 49199afd85c9SLinus Lüssing * 49209afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 49219afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 49229afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 49239afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 49249afd85c9SLinus Lüssing * 4925a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4926a516993fSLinus Lüssing * differs from the provided skb. 49279afd85c9SLinus Lüssing */ 49289afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 49299afd85c9SLinus Lüssing unsigned int transport_len) 49309afd85c9SLinus Lüssing { 49319afd85c9SLinus Lüssing struct sk_buff *skb_chk; 49329afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 49339afd85c9SLinus Lüssing int ret; 49349afd85c9SLinus Lüssing 4935a516993fSLinus Lüssing if (skb->len < len) 49369afd85c9SLinus Lüssing return NULL; 4937a516993fSLinus Lüssing else if (skb->len == len) 49389afd85c9SLinus Lüssing return skb; 49399afd85c9SLinus Lüssing 49409afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 49419afd85c9SLinus Lüssing if (!skb_chk) 49429afd85c9SLinus Lüssing return NULL; 49439afd85c9SLinus Lüssing 49449afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 49459afd85c9SLinus Lüssing if (ret) { 49469afd85c9SLinus Lüssing kfree_skb(skb_chk); 49479afd85c9SLinus Lüssing return NULL; 49489afd85c9SLinus Lüssing } 49499afd85c9SLinus Lüssing 49509afd85c9SLinus Lüssing return skb_chk; 49519afd85c9SLinus Lüssing } 49529afd85c9SLinus Lüssing 49539afd85c9SLinus Lüssing /** 49549afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 49559afd85c9SLinus Lüssing * @skb: the skb to check 49569afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 49579afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 49589afd85c9SLinus Lüssing * 49599afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 49609afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 49619afd85c9SLinus Lüssing * 49629afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 49639afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 49649afd85c9SLinus Lüssing * 4965a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4966a516993fSLinus Lüssing * differs from the provided skb. 49679afd85c9SLinus Lüssing */ 49689afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 49699afd85c9SLinus Lüssing unsigned int transport_len, 49709afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 49719afd85c9SLinus Lüssing { 49729afd85c9SLinus Lüssing struct sk_buff *skb_chk; 49739afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 4974fcba67c9SLinus Lüssing __sum16 ret; 49759afd85c9SLinus Lüssing 49769afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 49779afd85c9SLinus Lüssing if (!skb_chk) 4978a516993fSLinus Lüssing goto err; 49799afd85c9SLinus Lüssing 4980a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 4981a516993fSLinus Lüssing goto err; 49829afd85c9SLinus Lüssing 49839b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 49849afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 49859b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 49869afd85c9SLinus Lüssing 4987a516993fSLinus Lüssing if (ret) 4988a516993fSLinus Lüssing goto err; 49899afd85c9SLinus Lüssing 49909afd85c9SLinus Lüssing return skb_chk; 4991a516993fSLinus Lüssing 4992a516993fSLinus Lüssing err: 4993a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 4994a516993fSLinus Lüssing kfree_skb(skb_chk); 4995a516993fSLinus Lüssing 4996a516993fSLinus Lüssing return NULL; 4997a516993fSLinus Lüssing 49989afd85c9SLinus Lüssing } 49999afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 50009afd85c9SLinus Lüssing 50014497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 50024497b076SBen Hutchings { 5003e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5004e87cc472SJoe Perches skb->dev->name); 50054497b076SBen Hutchings } 50064497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5007bad43ca8SEric Dumazet 5008bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5009bad43ca8SEric Dumazet { 50103d861f66SEric Dumazet if (head_stolen) { 50113d861f66SEric Dumazet skb_release_head_state(skb); 5012bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 50133d861f66SEric Dumazet } else { 5014bad43ca8SEric Dumazet __kfree_skb(skb); 5015bad43ca8SEric Dumazet } 50163d861f66SEric Dumazet } 5017bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5018bad43ca8SEric Dumazet 5019bad43ca8SEric Dumazet /** 5020bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5021bad43ca8SEric Dumazet * @to: prior buffer 5022bad43ca8SEric Dumazet * @from: buffer to add 5023bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5024c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5025bad43ca8SEric Dumazet */ 5026bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5027bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5028bad43ca8SEric Dumazet { 5029c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5030bad43ca8SEric Dumazet int i, delta, len = from->len; 5031bad43ca8SEric Dumazet 5032bad43ca8SEric Dumazet *fragstolen = false; 5033bad43ca8SEric Dumazet 5034bad43ca8SEric Dumazet if (skb_cloned(to)) 5035bad43ca8SEric Dumazet return false; 5036bad43ca8SEric Dumazet 5037bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5038e93a0435SEric Dumazet if (len) 5039bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5040bad43ca8SEric Dumazet *delta_truesize = 0; 5041bad43ca8SEric Dumazet return true; 5042bad43ca8SEric Dumazet } 5043bad43ca8SEric Dumazet 5044c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5045c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5046c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5047bad43ca8SEric Dumazet return false; 50481f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 50491f8b977aSWillem de Bruijn return false; 5050bad43ca8SEric Dumazet 5051bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5052bad43ca8SEric Dumazet struct page *page; 5053bad43ca8SEric Dumazet unsigned int offset; 5054bad43ca8SEric Dumazet 5055c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5056c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5057bad43ca8SEric Dumazet return false; 5058bad43ca8SEric Dumazet 5059bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5060bad43ca8SEric Dumazet return false; 5061bad43ca8SEric Dumazet 5062bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5063bad43ca8SEric Dumazet 5064bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5065bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5066bad43ca8SEric Dumazet 5067c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5068bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5069bad43ca8SEric Dumazet *fragstolen = true; 5070bad43ca8SEric Dumazet } else { 5071c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5072c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5073bad43ca8SEric Dumazet return false; 5074bad43ca8SEric Dumazet 5075f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5076bad43ca8SEric Dumazet } 5077bad43ca8SEric Dumazet 5078bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5079bad43ca8SEric Dumazet 5080c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5081c818fa9eSEric Dumazet from_shinfo->frags, 5082c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5083c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5084bad43ca8SEric Dumazet 5085bad43ca8SEric Dumazet if (!skb_cloned(from)) 5086c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5087bad43ca8SEric Dumazet 50888ea853fdSLi RongQing /* if the skb is not cloned this does nothing 50898ea853fdSLi RongQing * since we set nr_frags to 0. 50908ea853fdSLi RongQing */ 5091c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5092c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5093bad43ca8SEric Dumazet 5094bad43ca8SEric Dumazet to->truesize += delta; 5095bad43ca8SEric Dumazet to->len += len; 5096bad43ca8SEric Dumazet to->data_len += len; 5097bad43ca8SEric Dumazet 5098bad43ca8SEric Dumazet *delta_truesize = delta; 5099bad43ca8SEric Dumazet return true; 5100bad43ca8SEric Dumazet } 5101bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5102621e84d6SNicolas Dichtel 5103621e84d6SNicolas Dichtel /** 51048b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5105621e84d6SNicolas Dichtel * 5106621e84d6SNicolas Dichtel * @skb: buffer to clean 51078b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5108621e84d6SNicolas Dichtel * 51098b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 51108b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 51118b27f277SNicolas Dichtel * operations. 51128b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 51138b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 51148b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5115621e84d6SNicolas Dichtel */ 51168b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5117621e84d6SNicolas Dichtel { 5118621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5119621e84d6SNicolas Dichtel skb->skb_iif = 0; 512060ff7467SWANG Cong skb->ignore_df = 0; 5121621e84d6SNicolas Dichtel skb_dst_drop(skb); 5122174e2381SFlorian Westphal skb_ext_reset(skb); 5123895b5c9fSFlorian Westphal nf_reset_ct(skb); 5124621e84d6SNicolas Dichtel nf_reset_trace(skb); 5125213dd74aSHerbert Xu 51266f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 51276f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5128875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 51296f9a5069SPetr Machata #endif 51306f9a5069SPetr Machata 5131213dd74aSHerbert Xu if (!xnet) 5132213dd74aSHerbert Xu return; 5133213dd74aSHerbert Xu 51342b5ec1a5SYe Yin ipvs_reset(skb); 5135213dd74aSHerbert Xu skb->mark = 0; 5136c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 5137621e84d6SNicolas Dichtel } 5138621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5139de960aa9SFlorian Westphal 5140de960aa9SFlorian Westphal /** 5141de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5142de960aa9SFlorian Westphal * 5143de960aa9SFlorian Westphal * @skb: GSO skb 5144de960aa9SFlorian Westphal * 5145de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5146de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5147de960aa9SFlorian Westphal * 5148de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5149de960aa9SFlorian Westphal */ 5150a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5151de960aa9SFlorian Westphal { 5152de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5153f993bc25SFlorian Westphal unsigned int thlen = 0; 5154f993bc25SFlorian Westphal 5155f993bc25SFlorian Westphal if (skb->encapsulation) { 5156f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5157f993bc25SFlorian Westphal skb_transport_header(skb); 5158de960aa9SFlorian Westphal 5159de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5160f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5161f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5162f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 51631dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 516490017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5165ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5166ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5167f993bc25SFlorian Westphal } 51686d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 51696d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 51706d39d589SFlorian Westphal * accounted for. 51716d39d589SFlorian Westphal */ 5172f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5173de960aa9SFlorian Westphal } 5174a4a77718SDaniel Axtens 5175a4a77718SDaniel Axtens /** 5176a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5177a4a77718SDaniel Axtens * 5178a4a77718SDaniel Axtens * @skb: GSO skb 5179a4a77718SDaniel Axtens * 5180a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5181a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5182a4a77718SDaniel Axtens * 5183a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5184a4a77718SDaniel Axtens */ 5185a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5186a4a77718SDaniel Axtens { 5187a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5188a4a77718SDaniel Axtens skb_network_header(skb); 5189a4a77718SDaniel Axtens 5190a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5191a4a77718SDaniel Axtens } 5192a4a77718SDaniel Axtens 5193a4a77718SDaniel Axtens /** 5194a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5195a4a77718SDaniel Axtens * 5196a4a77718SDaniel Axtens * @skb: GSO skb 5197a4a77718SDaniel Axtens * 5198a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5199a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5200a4a77718SDaniel Axtens * headers (TCP/UDP). 5201a4a77718SDaniel Axtens */ 5202a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5203a4a77718SDaniel Axtens { 5204a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5205a4a77718SDaniel Axtens 5206a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5207a4a77718SDaniel Axtens } 52080d5501c1SVlad Yasevich 5209ae7ef81eSMarcelo Ricardo Leitner /** 52102b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 52112b16f048SDaniel Axtens * 52122b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 52132b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 52142b16f048SDaniel Axtens * 52152b16f048SDaniel Axtens * We might want to check: 52162b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 52172b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 52182b16f048SDaniel Axtens * 52192b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 52202b16f048SDaniel Axtens * 522149682bfaSMathieu Malaterre * @skb: GSO skb 522249682bfaSMathieu Malaterre * 52232b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 52242b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 52252b16f048SDaniel Axtens * 52262b16f048SDaniel Axtens * @max_len: The maximum permissible length. 52272b16f048SDaniel Axtens * 52282b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 52292b16f048SDaniel Axtens */ 52302b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 52312b16f048SDaniel Axtens unsigned int seg_len, 52322b16f048SDaniel Axtens unsigned int max_len) { 52332b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 52342b16f048SDaniel Axtens const struct sk_buff *iter; 52352b16f048SDaniel Axtens 52362b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 52372b16f048SDaniel Axtens return seg_len <= max_len; 52382b16f048SDaniel Axtens 52392b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 52402b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 52412b16f048SDaniel Axtens 52422b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 52432b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 52442b16f048SDaniel Axtens return false; 52452b16f048SDaniel Axtens } 52462b16f048SDaniel Axtens 52472b16f048SDaniel Axtens return true; 52482b16f048SDaniel Axtens } 52492b16f048SDaniel Axtens 52502b16f048SDaniel Axtens /** 5251779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5252ae7ef81eSMarcelo Ricardo Leitner * 5253ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 525476f21b99SDavid S. Miller * @mtu: MTU to validate against 5255ae7ef81eSMarcelo Ricardo Leitner * 5256779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5257779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5258779b7931SDaniel Axtens * payload. 5259ae7ef81eSMarcelo Ricardo Leitner */ 5260779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5261ae7ef81eSMarcelo Ricardo Leitner { 52622b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5263ae7ef81eSMarcelo Ricardo Leitner } 5264779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5265ae7ef81eSMarcelo Ricardo Leitner 52662b16f048SDaniel Axtens /** 52672b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 52682b16f048SDaniel Axtens * 52692b16f048SDaniel Axtens * @skb: GSO skb 52702b16f048SDaniel Axtens * @len: length to validate against 52712b16f048SDaniel Axtens * 52722b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 52732b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 52742b16f048SDaniel Axtens */ 52752b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 52762b16f048SDaniel Axtens { 52772b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 52782b16f048SDaniel Axtens } 52792b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 52802b16f048SDaniel Axtens 52810d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 52820d5501c1SVlad Yasevich { 5283d85e8be2SYuya Kusakabe int mac_len, meta_len; 5284d85e8be2SYuya Kusakabe void *meta; 52854bbb3e0eSToshiaki Makita 52860d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 52870d5501c1SVlad Yasevich kfree_skb(skb); 52880d5501c1SVlad Yasevich return NULL; 52890d5501c1SVlad Yasevich } 52900d5501c1SVlad Yasevich 52914bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5292ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 52934bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 52944bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5295ae474573SToshiaki Makita } 5296d85e8be2SYuya Kusakabe 5297d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5298d85e8be2SYuya Kusakabe if (meta_len) { 5299d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5300d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5301d85e8be2SYuya Kusakabe } 5302d85e8be2SYuya Kusakabe 53030d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 53040d5501c1SVlad Yasevich return skb; 53050d5501c1SVlad Yasevich } 53060d5501c1SVlad Yasevich 53070d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 53080d5501c1SVlad Yasevich { 53090d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 53100d5501c1SVlad Yasevich u16 vlan_tci; 53110d5501c1SVlad Yasevich 5312df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 53130d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 53140d5501c1SVlad Yasevich return skb; 53150d5501c1SVlad Yasevich } 53160d5501c1SVlad Yasevich 53170d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 53180d5501c1SVlad Yasevich if (unlikely(!skb)) 53190d5501c1SVlad Yasevich goto err_free; 53200d5501c1SVlad Yasevich 53210d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 53220d5501c1SVlad Yasevich goto err_free; 53230d5501c1SVlad Yasevich 53240d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 53250d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 53260d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 53270d5501c1SVlad Yasevich 53280d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 53290d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 53300d5501c1SVlad Yasevich 53310d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 53320d5501c1SVlad Yasevich if (unlikely(!skb)) 53330d5501c1SVlad Yasevich goto err_free; 53340d5501c1SVlad Yasevich 53350d5501c1SVlad Yasevich skb_reset_network_header(skb); 53360d5501c1SVlad Yasevich skb_reset_transport_header(skb); 53370d5501c1SVlad Yasevich skb_reset_mac_len(skb); 53380d5501c1SVlad Yasevich 53390d5501c1SVlad Yasevich return skb; 53400d5501c1SVlad Yasevich 53410d5501c1SVlad Yasevich err_free: 53420d5501c1SVlad Yasevich kfree_skb(skb); 53430d5501c1SVlad Yasevich return NULL; 53440d5501c1SVlad Yasevich } 53450d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 53462e4e4410SEric Dumazet 5347e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5348e2195121SJiri Pirko { 5349e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5350e2195121SJiri Pirko return -ENOMEM; 5351e2195121SJiri Pirko 5352e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5353e2195121SJiri Pirko return 0; 5354e2195121SJiri Pirko 5355e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5356e2195121SJiri Pirko } 5357e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5358e2195121SJiri Pirko 5359bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5360bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5361bfca4c52SShmulik Ladkani */ 5362bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 536393515d53SJiri Pirko { 536493515d53SJiri Pirko struct vlan_hdr *vhdr; 5365b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 536693515d53SJiri Pirko int err; 536793515d53SJiri Pirko 5368b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5369b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5370b6a79208SShmulik Ladkani offset)) { 5371b6a79208SShmulik Ladkani return -EINVAL; 5372b6a79208SShmulik Ladkani } 5373b6a79208SShmulik Ladkani 537493515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 537593515d53SJiri Pirko if (unlikely(err)) 5376b6a79208SShmulik Ladkani return err; 537793515d53SJiri Pirko 537893515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 537993515d53SJiri Pirko 538093515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 538193515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 538293515d53SJiri Pirko 538393515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 538493515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 538593515d53SJiri Pirko 538693515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 538793515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 538893515d53SJiri Pirko 538993515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 539093515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 539193515d53SJiri Pirko 539293515d53SJiri Pirko skb_reset_mac_len(skb); 539393515d53SJiri Pirko 539493515d53SJiri Pirko return err; 539593515d53SJiri Pirko } 5396bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 539793515d53SJiri Pirko 5398b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5399b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5400b6a79208SShmulik Ladkani */ 540193515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 540293515d53SJiri Pirko { 540393515d53SJiri Pirko u16 vlan_tci; 540493515d53SJiri Pirko __be16 vlan_proto; 540593515d53SJiri Pirko int err; 540693515d53SJiri Pirko 5407df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5408b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 540993515d53SJiri Pirko } else { 5410ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 541193515d53SJiri Pirko return 0; 541293515d53SJiri Pirko 541393515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 541493515d53SJiri Pirko if (err) 541593515d53SJiri Pirko return err; 541693515d53SJiri Pirko } 541793515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5418ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 541993515d53SJiri Pirko return 0; 542093515d53SJiri Pirko 542193515d53SJiri Pirko vlan_proto = skb->protocol; 542293515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 542393515d53SJiri Pirko if (unlikely(err)) 542493515d53SJiri Pirko return err; 542593515d53SJiri Pirko 542693515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 542793515d53SJiri Pirko return 0; 542893515d53SJiri Pirko } 542993515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 543093515d53SJiri Pirko 5431b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5432b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5433b6a79208SShmulik Ladkani */ 543493515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 543593515d53SJiri Pirko { 5436df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5437b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 543893515d53SJiri Pirko int err; 543993515d53SJiri Pirko 5440b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5441b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5442b6a79208SShmulik Ladkani offset)) { 5443b6a79208SShmulik Ladkani return -EINVAL; 5444b6a79208SShmulik Ladkani } 5445b6a79208SShmulik Ladkani 544693515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5447df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5448b6a79208SShmulik Ladkani if (err) 544993515d53SJiri Pirko return err; 54509241e2dfSDaniel Borkmann 545193515d53SJiri Pirko skb->protocol = skb->vlan_proto; 545293515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 545393515d53SJiri Pirko 54546b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 545593515d53SJiri Pirko } 545693515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 545793515d53SJiri Pirko return 0; 545893515d53SJiri Pirko } 545993515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 546093515d53SJiri Pirko 54618822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 54628822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 54638822e270SJohn Hurley __be16 ethertype) 54648822e270SJohn Hurley { 54658822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 54668822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 54678822e270SJohn Hurley 54688822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 54698822e270SJohn Hurley } 54708822e270SJohn Hurley 54718822e270SJohn Hurley hdr->h_proto = ethertype; 54728822e270SJohn Hurley } 54738822e270SJohn Hurley 54748822e270SJohn Hurley /** 5475e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5476e7dbfed1SMartin Varghese * the packet 54778822e270SJohn Hurley * 54788822e270SJohn Hurley * @skb: buffer 54798822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 54808822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5481fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5482e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5483e7dbfed1SMartin Varghese * ethernet 54848822e270SJohn Hurley * 54858822e270SJohn Hurley * Expects skb->data at mac header. 54868822e270SJohn Hurley * 54878822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 54888822e270SJohn Hurley */ 5489fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5490d04ac224SMartin Varghese int mac_len, bool ethernet) 54918822e270SJohn Hurley { 54928822e270SJohn Hurley struct mpls_shim_hdr *lse; 54938822e270SJohn Hurley int err; 54948822e270SJohn Hurley 54958822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 54968822e270SJohn Hurley return -EINVAL; 54978822e270SJohn Hurley 54988822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 54998822e270SJohn Hurley if (skb->encapsulation) 55008822e270SJohn Hurley return -EINVAL; 55018822e270SJohn Hurley 55028822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 55038822e270SJohn Hurley if (unlikely(err)) 55048822e270SJohn Hurley return err; 55058822e270SJohn Hurley 55068822e270SJohn Hurley if (!skb->inner_protocol) { 5507e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 55088822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 55098822e270SJohn Hurley } 55108822e270SJohn Hurley 55118822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 55128822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5513fa4e0f88SDavide Caratti mac_len); 55148822e270SJohn Hurley skb_reset_mac_header(skb); 5515fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5516e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 55178822e270SJohn Hurley 55188822e270SJohn Hurley lse = mpls_hdr(skb); 55198822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 55208822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 55218822e270SJohn Hurley 5522d04ac224SMartin Varghese if (ethernet) 55238822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 55248822e270SJohn Hurley skb->protocol = mpls_proto; 55258822e270SJohn Hurley 55268822e270SJohn Hurley return 0; 55278822e270SJohn Hurley } 55288822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 55298822e270SJohn Hurley 55302e4e4410SEric Dumazet /** 5531ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5532ed246ceeSJohn Hurley * 5533ed246ceeSJohn Hurley * @skb: buffer 5534ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5535fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5536040b5cfbSMartin Varghese * @ethernet: flag to indicate if ethernet header is present in packet 5537ed246ceeSJohn Hurley * 5538ed246ceeSJohn Hurley * Expects skb->data at mac header. 5539ed246ceeSJohn Hurley * 5540ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5541ed246ceeSJohn Hurley */ 5542040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5543040b5cfbSMartin Varghese bool ethernet) 5544ed246ceeSJohn Hurley { 5545ed246ceeSJohn Hurley int err; 5546ed246ceeSJohn Hurley 5547ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5548dedc5a08SDavide Caratti return 0; 5549ed246ceeSJohn Hurley 5550fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5551ed246ceeSJohn Hurley if (unlikely(err)) 5552ed246ceeSJohn Hurley return err; 5553ed246ceeSJohn Hurley 5554ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5555ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5556fa4e0f88SDavide Caratti mac_len); 5557ed246ceeSJohn Hurley 5558ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5559ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5560fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5561ed246ceeSJohn Hurley 5562040b5cfbSMartin Varghese if (ethernet) { 5563ed246ceeSJohn Hurley struct ethhdr *hdr; 5564ed246ceeSJohn Hurley 5565ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5566ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5567ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5568ed246ceeSJohn Hurley } 5569ed246ceeSJohn Hurley skb->protocol = next_proto; 5570ed246ceeSJohn Hurley 5571ed246ceeSJohn Hurley return 0; 5572ed246ceeSJohn Hurley } 5573ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5574ed246ceeSJohn Hurley 5575ed246ceeSJohn Hurley /** 5576d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5577d27cf5c5SJohn Hurley * 5578d27cf5c5SJohn Hurley * @skb: buffer 5579d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5580d27cf5c5SJohn Hurley * 5581d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5582d27cf5c5SJohn Hurley * 5583d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5584d27cf5c5SJohn Hurley */ 5585d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5586d27cf5c5SJohn Hurley { 5587d27cf5c5SJohn Hurley int err; 5588d27cf5c5SJohn Hurley 5589d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5590d27cf5c5SJohn Hurley return -EINVAL; 5591d27cf5c5SJohn Hurley 5592d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5593d27cf5c5SJohn Hurley if (unlikely(err)) 5594d27cf5c5SJohn Hurley return err; 5595d27cf5c5SJohn Hurley 5596d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5597d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5598d27cf5c5SJohn Hurley 5599d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5600d27cf5c5SJohn Hurley } 5601d27cf5c5SJohn Hurley 5602d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 5603d27cf5c5SJohn Hurley 5604d27cf5c5SJohn Hurley return 0; 5605d27cf5c5SJohn Hurley } 5606d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 5607d27cf5c5SJohn Hurley 5608d27cf5c5SJohn Hurley /** 56092a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 56102a2ea508SJohn Hurley * 56112a2ea508SJohn Hurley * @skb: buffer 56122a2ea508SJohn Hurley * 56132a2ea508SJohn Hurley * Expects skb->data at mac header. 56142a2ea508SJohn Hurley * 56152a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 56162a2ea508SJohn Hurley */ 56172a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 56182a2ea508SJohn Hurley { 56192a2ea508SJohn Hurley u32 lse; 56202a2ea508SJohn Hurley u8 ttl; 56212a2ea508SJohn Hurley 56222a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 56232a2ea508SJohn Hurley return -EINVAL; 56242a2ea508SJohn Hurley 56252a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 56262a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 56272a2ea508SJohn Hurley if (!--ttl) 56282a2ea508SJohn Hurley return -EINVAL; 56292a2ea508SJohn Hurley 56302a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 56312a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 56322a2ea508SJohn Hurley 56332a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 56342a2ea508SJohn Hurley } 56352a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 56362a2ea508SJohn Hurley 56372a2ea508SJohn Hurley /** 56382e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 56392e4e4410SEric Dumazet * 5640de3f0d0eSMasanari Iida * @header_len: size of linear part 5641de3f0d0eSMasanari Iida * @data_len: needed length in frags 5642de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5643de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5644de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 56452e4e4410SEric Dumazet * 56462e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 56472e4e4410SEric Dumazet */ 56482e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 56492e4e4410SEric Dumazet unsigned long data_len, 56502e4e4410SEric Dumazet int max_page_order, 56512e4e4410SEric Dumazet int *errcode, 56522e4e4410SEric Dumazet gfp_t gfp_mask) 56532e4e4410SEric Dumazet { 56542e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 56552e4e4410SEric Dumazet unsigned long chunk; 56562e4e4410SEric Dumazet struct sk_buff *skb; 56572e4e4410SEric Dumazet struct page *page; 56582e4e4410SEric Dumazet int i; 56592e4e4410SEric Dumazet 56602e4e4410SEric Dumazet *errcode = -EMSGSIZE; 56612e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 56622e4e4410SEric Dumazet * high order pages... 56632e4e4410SEric Dumazet */ 56642e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 56652e4e4410SEric Dumazet return NULL; 56662e4e4410SEric Dumazet 56672e4e4410SEric Dumazet *errcode = -ENOBUFS; 5668f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 56692e4e4410SEric Dumazet if (!skb) 56702e4e4410SEric Dumazet return NULL; 56712e4e4410SEric Dumazet 56722e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 56732e4e4410SEric Dumazet 56742e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 56752e4e4410SEric Dumazet int order = max_page_order; 56762e4e4410SEric Dumazet 56772e4e4410SEric Dumazet while (order) { 56782e4e4410SEric Dumazet if (npages >= 1 << order) { 5679d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 56802e4e4410SEric Dumazet __GFP_COMP | 5681d14b56f5SMichal Hocko __GFP_NOWARN, 56822e4e4410SEric Dumazet order); 56832e4e4410SEric Dumazet if (page) 56842e4e4410SEric Dumazet goto fill_page; 56852e4e4410SEric Dumazet /* Do not retry other high order allocations */ 56862e4e4410SEric Dumazet order = 1; 56872e4e4410SEric Dumazet max_page_order = 0; 56882e4e4410SEric Dumazet } 56892e4e4410SEric Dumazet order--; 56902e4e4410SEric Dumazet } 56912e4e4410SEric Dumazet page = alloc_page(gfp_mask); 56922e4e4410SEric Dumazet if (!page) 56932e4e4410SEric Dumazet goto failure; 56942e4e4410SEric Dumazet fill_page: 56952e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 56962e4e4410SEric Dumazet PAGE_SIZE << order); 56972e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 56982e4e4410SEric Dumazet data_len -= chunk; 56992e4e4410SEric Dumazet npages -= 1 << order; 57002e4e4410SEric Dumazet } 57012e4e4410SEric Dumazet return skb; 57022e4e4410SEric Dumazet 57032e4e4410SEric Dumazet failure: 57042e4e4410SEric Dumazet kfree_skb(skb); 57052e4e4410SEric Dumazet return NULL; 57062e4e4410SEric Dumazet } 57072e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 57086fa01ccdSSowmini Varadhan 57096fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 57106fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 57116fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 57126fa01ccdSSowmini Varadhan { 57136fa01ccdSSowmini Varadhan int i; 57146fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 57156fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 57166fa01ccdSSowmini Varadhan u8 *data; 57176fa01ccdSSowmini Varadhan 57186fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 57196fa01ccdSSowmini Varadhan 57206fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 57216fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 57226fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 57236fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 57246fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 57256fa01ccdSSowmini Varadhan if (!data) 57266fa01ccdSSowmini Varadhan return -ENOMEM; 57276fa01ccdSSowmini Varadhan 57286fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 57296fa01ccdSSowmini Varadhan 57306fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 57316fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 57326fa01ccdSSowmini Varadhan skb->len -= off; 57336fa01ccdSSowmini Varadhan 57346fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 57356fa01ccdSSowmini Varadhan skb_shinfo(skb), 57366fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 57376fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 57386fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 57396fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 57406fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 57416fa01ccdSSowmini Varadhan kfree(data); 57426fa01ccdSSowmini Varadhan return -ENOMEM; 57436fa01ccdSSowmini Varadhan } 57446fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 57456fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 57466fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 57476fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 57486fa01ccdSSowmini Varadhan skb_release_data(skb); 57496fa01ccdSSowmini Varadhan } else { 57506fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 57516fa01ccdSSowmini Varadhan * relocate values 57526fa01ccdSSowmini Varadhan */ 57536fa01ccdSSowmini Varadhan skb_free_head(skb); 57546fa01ccdSSowmini Varadhan } 57556fa01ccdSSowmini Varadhan 57566fa01ccdSSowmini Varadhan skb->head = data; 57576fa01ccdSSowmini Varadhan skb->data = data; 57586fa01ccdSSowmini Varadhan skb->head_frag = 0; 57596fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 57606fa01ccdSSowmini Varadhan skb->end = size; 57616fa01ccdSSowmini Varadhan #else 57626fa01ccdSSowmini Varadhan skb->end = skb->head + size; 57636fa01ccdSSowmini Varadhan #endif 57646fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 57656fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 57666fa01ccdSSowmini Varadhan skb->cloned = 0; 57676fa01ccdSSowmini Varadhan skb->hdr_len = 0; 57686fa01ccdSSowmini Varadhan skb->nohdr = 0; 57696fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 57706fa01ccdSSowmini Varadhan 57716fa01ccdSSowmini Varadhan return 0; 57726fa01ccdSSowmini Varadhan } 57736fa01ccdSSowmini Varadhan 57746fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 57756fa01ccdSSowmini Varadhan 57766fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 57776fa01ccdSSowmini Varadhan * pskb_carve() 57786fa01ccdSSowmini Varadhan */ 57796fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 57806fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 57816fa01ccdSSowmini Varadhan gfp_t gfp_mask) 57826fa01ccdSSowmini Varadhan { 57836fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 57846fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 57856fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 57866fa01ccdSSowmini Varadhan 57876fa01ccdSSowmini Varadhan do { 57886fa01ccdSSowmini Varadhan if (!list) { 57896fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 57906fa01ccdSSowmini Varadhan return -EFAULT; 57916fa01ccdSSowmini Varadhan } 57926fa01ccdSSowmini Varadhan if (list->len <= eat) { 57936fa01ccdSSowmini Varadhan /* Eaten as whole. */ 57946fa01ccdSSowmini Varadhan eat -= list->len; 57956fa01ccdSSowmini Varadhan list = list->next; 57966fa01ccdSSowmini Varadhan insp = list; 57976fa01ccdSSowmini Varadhan } else { 57986fa01ccdSSowmini Varadhan /* Eaten partially. */ 57996fa01ccdSSowmini Varadhan if (skb_shared(list)) { 58006fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 58016fa01ccdSSowmini Varadhan if (!clone) 58026fa01ccdSSowmini Varadhan return -ENOMEM; 58036fa01ccdSSowmini Varadhan insp = list->next; 58046fa01ccdSSowmini Varadhan list = clone; 58056fa01ccdSSowmini Varadhan } else { 58066fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 58076fa01ccdSSowmini Varadhan insp = list; 58086fa01ccdSSowmini Varadhan } 58096fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 58106fa01ccdSSowmini Varadhan kfree_skb(clone); 58116fa01ccdSSowmini Varadhan return -ENOMEM; 58126fa01ccdSSowmini Varadhan } 58136fa01ccdSSowmini Varadhan break; 58146fa01ccdSSowmini Varadhan } 58156fa01ccdSSowmini Varadhan } while (eat); 58166fa01ccdSSowmini Varadhan 58176fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 58186fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 58196fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 58206fa01ccdSSowmini Varadhan kfree_skb(list); 58216fa01ccdSSowmini Varadhan } 58226fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 58236fa01ccdSSowmini Varadhan if (clone) { 58246fa01ccdSSowmini Varadhan clone->next = list; 58256fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 58266fa01ccdSSowmini Varadhan } 58276fa01ccdSSowmini Varadhan return 0; 58286fa01ccdSSowmini Varadhan } 58296fa01ccdSSowmini Varadhan 58306fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 58316fa01ccdSSowmini Varadhan * non-linear part of skb 58326fa01ccdSSowmini Varadhan */ 58336fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 58346fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 58356fa01ccdSSowmini Varadhan { 58366fa01ccdSSowmini Varadhan int i, k = 0; 58376fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 58386fa01ccdSSowmini Varadhan u8 *data; 58396fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 58406fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 58416fa01ccdSSowmini Varadhan 58426fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 58436fa01ccdSSowmini Varadhan 58446fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 58456fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 58466fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 58476fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 58486fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 58496fa01ccdSSowmini Varadhan if (!data) 58506fa01ccdSSowmini Varadhan return -ENOMEM; 58516fa01ccdSSowmini Varadhan 58526fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 58536fa01ccdSSowmini Varadhan 58546fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 58556fa01ccdSSowmini Varadhan skb_shinfo(skb), offsetof(struct skb_shared_info, 58566fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 58576fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 58586fa01ccdSSowmini Varadhan kfree(data); 58596fa01ccdSSowmini Varadhan return -ENOMEM; 58606fa01ccdSSowmini Varadhan } 58616fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 58626fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 58636fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 58646fa01ccdSSowmini Varadhan 58656fa01ccdSSowmini Varadhan if (pos + fsize > off) { 58666fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 58676fa01ccdSSowmini Varadhan 58686fa01ccdSSowmini Varadhan if (pos < off) { 58696fa01ccdSSowmini Varadhan /* Split frag. 58706fa01ccdSSowmini Varadhan * We have two variants in this case: 58716fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 58726fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 58736fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 58746fa01ccdSSowmini Varadhan * where splitting is expensive. 58756fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 58766fa01ccdSSowmini Varadhan */ 5877b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 58786fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 58796fa01ccdSSowmini Varadhan } 58806fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 58816fa01ccdSSowmini Varadhan k++; 58826fa01ccdSSowmini Varadhan } 58836fa01ccdSSowmini Varadhan pos += fsize; 58846fa01ccdSSowmini Varadhan } 58856fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 58866fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 58876fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 58886fa01ccdSSowmini Varadhan 58896fa01ccdSSowmini Varadhan if (k == 0) { 58906fa01ccdSSowmini Varadhan /* split line is in frag list */ 58916fa01ccdSSowmini Varadhan pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask); 58926fa01ccdSSowmini Varadhan } 58936fa01ccdSSowmini Varadhan skb_release_data(skb); 58946fa01ccdSSowmini Varadhan 58956fa01ccdSSowmini Varadhan skb->head = data; 58966fa01ccdSSowmini Varadhan skb->head_frag = 0; 58976fa01ccdSSowmini Varadhan skb->data = data; 58986fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 58996fa01ccdSSowmini Varadhan skb->end = size; 59006fa01ccdSSowmini Varadhan #else 59016fa01ccdSSowmini Varadhan skb->end = skb->head + size; 59026fa01ccdSSowmini Varadhan #endif 59036fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 59046fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 59056fa01ccdSSowmini Varadhan skb->cloned = 0; 59066fa01ccdSSowmini Varadhan skb->hdr_len = 0; 59076fa01ccdSSowmini Varadhan skb->nohdr = 0; 59086fa01ccdSSowmini Varadhan skb->len -= off; 59096fa01ccdSSowmini Varadhan skb->data_len = skb->len; 59106fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 59116fa01ccdSSowmini Varadhan return 0; 59126fa01ccdSSowmini Varadhan } 59136fa01ccdSSowmini Varadhan 59146fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 59156fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 59166fa01ccdSSowmini Varadhan { 59176fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 59186fa01ccdSSowmini Varadhan 59196fa01ccdSSowmini Varadhan if (len < headlen) 59206fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 59216fa01ccdSSowmini Varadhan else 59226fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 59236fa01ccdSSowmini Varadhan } 59246fa01ccdSSowmini Varadhan 59256fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 59266fa01ccdSSowmini Varadhan * a new skb 59276fa01ccdSSowmini Varadhan */ 59286fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 59296fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 59306fa01ccdSSowmini Varadhan { 59316fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 59326fa01ccdSSowmini Varadhan 59336fa01ccdSSowmini Varadhan if (!clone) 59346fa01ccdSSowmini Varadhan return NULL; 59356fa01ccdSSowmini Varadhan 59366fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 59376fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 59386fa01ccdSSowmini Varadhan kfree_skb(clone); 59396fa01ccdSSowmini Varadhan return NULL; 59406fa01ccdSSowmini Varadhan } 59416fa01ccdSSowmini Varadhan return clone; 59426fa01ccdSSowmini Varadhan } 59436fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 5944c8c8b127SEric Dumazet 5945c8c8b127SEric Dumazet /** 5946c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 5947c8c8b127SEric Dumazet * @skb: buffer 5948c8c8b127SEric Dumazet * 5949c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 5950c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 5951c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 5952c8c8b127SEric Dumazet * truesize. 5953c8c8b127SEric Dumazet * Notes: 5954c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 5955c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 5956c8c8b127SEric Dumazet */ 5957c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 5958c8c8b127SEric Dumazet { 59593174fed9SEric Dumazet if (skb->data_len) { 59603174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 5961c8c8b127SEric Dumazet skb_cloned(skb)) 5962c8c8b127SEric Dumazet return; 5963c8c8b127SEric Dumazet 5964c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 5965c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 59663174fed9SEric Dumazet } 59673174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 59683174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 59693174fed9SEric Dumazet * their truesize. 59703174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 59713174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 59723174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 5973c8c8b127SEric Dumazet */ 5974c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 5975c8c8b127SEric Dumazet } 5976df5042f4SFlorian Westphal 5977df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 5978df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 5979df5042f4SFlorian Westphal { 5980df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 5981df5042f4SFlorian Westphal } 5982df5042f4SFlorian Westphal 5983df5042f4SFlorian Westphal static struct skb_ext *skb_ext_alloc(void) 5984df5042f4SFlorian Westphal { 5985df5042f4SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 5986df5042f4SFlorian Westphal 5987df5042f4SFlorian Westphal if (new) { 5988df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 5989df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 5990df5042f4SFlorian Westphal } 5991df5042f4SFlorian Westphal 5992df5042f4SFlorian Westphal return new; 5993df5042f4SFlorian Westphal } 5994df5042f4SFlorian Westphal 59954165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 59964165079bSFlorian Westphal unsigned int old_active) 5997df5042f4SFlorian Westphal { 5998df5042f4SFlorian Westphal struct skb_ext *new; 5999df5042f4SFlorian Westphal 6000df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6001df5042f4SFlorian Westphal return old; 6002df5042f4SFlorian Westphal 6003df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6004df5042f4SFlorian Westphal if (!new) 6005df5042f4SFlorian Westphal return NULL; 6006df5042f4SFlorian Westphal 6007df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6008df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6009df5042f4SFlorian Westphal 60104165079bSFlorian Westphal #ifdef CONFIG_XFRM 60114165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 60124165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 60134165079bSFlorian Westphal unsigned int i; 60144165079bSFlorian Westphal 60154165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 60164165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 60174165079bSFlorian Westphal } 60184165079bSFlorian Westphal #endif 6019df5042f4SFlorian Westphal __skb_ext_put(old); 6020df5042f4SFlorian Westphal return new; 6021df5042f4SFlorian Westphal } 6022df5042f4SFlorian Westphal 6023df5042f4SFlorian Westphal /** 6024df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6025df5042f4SFlorian Westphal * @skb: buffer 6026df5042f4SFlorian Westphal * @id: extension to allocate space for 6027df5042f4SFlorian Westphal * 6028df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6029df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6030df5042f4SFlorian Westphal * is returned. 6031df5042f4SFlorian Westphal * 6032df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6033df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6034df5042f4SFlorian Westphal * 6035df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6036df5042f4SFlorian Westphal */ 6037df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6038df5042f4SFlorian Westphal { 6039df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6040df5042f4SFlorian Westphal unsigned int newlen, newoff; 6041df5042f4SFlorian Westphal 6042df5042f4SFlorian Westphal if (skb->active_extensions) { 6043df5042f4SFlorian Westphal old = skb->extensions; 6044df5042f4SFlorian Westphal 60454165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6046df5042f4SFlorian Westphal if (!new) 6047df5042f4SFlorian Westphal return NULL; 6048df5042f4SFlorian Westphal 6049682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6050df5042f4SFlorian Westphal goto set_active; 6051df5042f4SFlorian Westphal 6052e94e50bdSPaolo Abeni newoff = new->chunks; 6053df5042f4SFlorian Westphal } else { 6054df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6055df5042f4SFlorian Westphal 6056df5042f4SFlorian Westphal new = skb_ext_alloc(); 6057df5042f4SFlorian Westphal if (!new) 6058df5042f4SFlorian Westphal return NULL; 6059df5042f4SFlorian Westphal } 6060df5042f4SFlorian Westphal 6061df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6062df5042f4SFlorian Westphal new->chunks = newlen; 6063df5042f4SFlorian Westphal new->offset[id] = newoff; 6064df5042f4SFlorian Westphal set_active: 6065682ec859SPaolo Abeni skb->extensions = new; 6066df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6067df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6068df5042f4SFlorian Westphal } 6069df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6070df5042f4SFlorian Westphal 60714165079bSFlorian Westphal #ifdef CONFIG_XFRM 60724165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 60734165079bSFlorian Westphal { 60744165079bSFlorian Westphal unsigned int i; 60754165079bSFlorian Westphal 60764165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 60774165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 60784165079bSFlorian Westphal } 60794165079bSFlorian Westphal #endif 60804165079bSFlorian Westphal 6081df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6082df5042f4SFlorian Westphal { 6083df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6084df5042f4SFlorian Westphal 6085df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6086df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6087df5042f4SFlorian Westphal skb->extensions = NULL; 6088df5042f4SFlorian Westphal __skb_ext_put(ext); 60894165079bSFlorian Westphal #ifdef CONFIG_XFRM 60904165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 60914165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 60924165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 60934165079bSFlorian Westphal 60944165079bSFlorian Westphal skb_ext_put_sp(sp); 60954165079bSFlorian Westphal sp->len = 0; 60964165079bSFlorian Westphal #endif 6097df5042f4SFlorian Westphal } 6098df5042f4SFlorian Westphal } 6099df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6100df5042f4SFlorian Westphal 6101df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6102df5042f4SFlorian Westphal { 6103df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6104df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6105df5042f4SFlorian Westphal */ 6106df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6107df5042f4SFlorian Westphal goto free_now; 6108df5042f4SFlorian Westphal 6109df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6110df5042f4SFlorian Westphal return; 6111df5042f4SFlorian Westphal free_now: 61124165079bSFlorian Westphal #ifdef CONFIG_XFRM 61134165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 61144165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 61154165079bSFlorian Westphal #endif 61164165079bSFlorian Westphal 6117df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6118df5042f4SFlorian Westphal } 6119df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6120df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 6121