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> 713ee17bc7SMat Martineau #include <net/mptcp.h> 721da177e4SLinus Torvalds 737c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 74ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7551c56b00SEric Dumazet #include <linux/highmem.h> 76b245be1fSWillem de Bruijn #include <linux/capability.h> 77b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 782544af03SMatteo Croce #include <linux/indirect_call_wrapper.h> 79a1f8e7f7SAl Viro 807b7ed885SBart Van Assche #include "datagram.h" 817b7ed885SBart Van Assche 8208009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8308009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 84df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 85df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 86df5042f4SFlorian Westphal #endif 875f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 885f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds /** 91f05de73bSJean Sacren * skb_panic - private function for out-of-line support 921da177e4SLinus Torvalds * @skb: buffer 931da177e4SLinus Torvalds * @sz: size 94f05de73bSJean Sacren * @addr: address 9599d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 961da177e4SLinus Torvalds * 97f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 98f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 99f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 100f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1011da177e4SLinus Torvalds */ 102f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 10399d5851eSJames Hogan const char msg[]) 1041da177e4SLinus Torvalds { 10541a46913SJesper Dangaard Brouer pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", 10699d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1074305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 10826095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1091da177e4SLinus Torvalds BUG(); 1101da177e4SLinus Torvalds } 1111da177e4SLinus Torvalds 112f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1131da177e4SLinus Torvalds { 114f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds 117f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 118f05de73bSJean Sacren { 119f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 120f05de73bSJean Sacren } 121c93bdd0eSMel Gorman 122c93bdd0eSMel Gorman /* 123c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 124c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 125c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 126c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 127c93bdd0eSMel Gorman * memory is free 128c93bdd0eSMel Gorman */ 129c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 130c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 13161c5e88aSstephen hemminger 13261c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 13361c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 134c93bdd0eSMel Gorman { 135c93bdd0eSMel Gorman void *obj; 136c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 137c93bdd0eSMel Gorman 138c93bdd0eSMel Gorman /* 139c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 140c93bdd0eSMel Gorman * to the reserves, fail. 141c93bdd0eSMel Gorman */ 142c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 143c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 144c93bdd0eSMel Gorman node); 145c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 146c93bdd0eSMel Gorman goto out; 147c93bdd0eSMel Gorman 148c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 149c93bdd0eSMel Gorman ret_pfmemalloc = true; 150c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 151c93bdd0eSMel Gorman 152c93bdd0eSMel Gorman out: 153c93bdd0eSMel Gorman if (pfmemalloc) 154c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 155c93bdd0eSMel Gorman 156c93bdd0eSMel Gorman return obj; 157c93bdd0eSMel Gorman } 158c93bdd0eSMel Gorman 1591da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1601da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1611da177e4SLinus Torvalds * [BEEP] leaks. 1621da177e4SLinus Torvalds * 1631da177e4SLinus Torvalds */ 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds /** 166d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1671da177e4SLinus Torvalds * @size: size to allocate 1681da177e4SLinus Torvalds * @gfp_mask: allocation mask 169c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 170c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 171c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 172c93bdd0eSMel Gorman * allocations in case the data is required for writeback 173b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1741da177e4SLinus Torvalds * 1751da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 17694b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 17794b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 1781da177e4SLinus Torvalds * 1791da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1801da177e4SLinus Torvalds * %GFP_ATOMIC. 1811da177e4SLinus Torvalds */ 182dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 183c93bdd0eSMel Gorman int flags, int node) 1841da177e4SLinus Torvalds { 185e18b890bSChristoph Lameter struct kmem_cache *cache; 1864947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1871da177e4SLinus Torvalds struct sk_buff *skb; 1881da177e4SLinus Torvalds u8 *data; 189c93bdd0eSMel Gorman bool pfmemalloc; 1901da177e4SLinus Torvalds 191c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 192c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 193c93bdd0eSMel Gorman 194c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 195c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1968798b3fbSHerbert Xu 1971da177e4SLinus Torvalds /* Get the HEAD */ 198b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1991da177e4SLinus Torvalds if (!skb) 2001da177e4SLinus Torvalds goto out; 201ec7d2f2cSEric Dumazet prefetchw(skb); 2021da177e4SLinus Torvalds 20387fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 20487fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 20587fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 20687fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 20787fb4b7bSEric Dumazet */ 208bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 20987fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 210c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2111da177e4SLinus Torvalds if (!data) 2121da177e4SLinus Torvalds goto nodata; 21387fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 21487fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 21587fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 21687fb4b7bSEric Dumazet */ 21787fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 218ec7d2f2cSEric Dumazet prefetchw(data + size); 2191da177e4SLinus Torvalds 220ca0605a7SArnaldo Carvalho de Melo /* 221c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 222c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 223c8005785SJohannes Berg * the tail pointer in struct sk_buff! 224ca0605a7SArnaldo Carvalho de Melo */ 225ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 22687fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 22787fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 228c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 22963354797SReshetova, Elena refcount_set(&skb->users, 1); 2301da177e4SLinus Torvalds skb->head = data; 2311da177e4SLinus Torvalds skb->data = data; 23227a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2334305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 23435d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 23535d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 23619633e12SStephen Hemminger 2374947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2384947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 239ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2404947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 2414947d3efSBenjamin LaHaise 242c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 243d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 2441da177e4SLinus Torvalds 245d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 246d0bf4a9eSEric Dumazet 247d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 2482638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 1); 249d179cd12SDavid S. Miller 2506ffe75ebSEric Dumazet fclones->skb2.fclone = SKB_FCLONE_CLONE; 251d179cd12SDavid S. Miller } 2526370cc3bSAleksandr Nogikh 2536370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 2546370cc3bSAleksandr Nogikh 2551da177e4SLinus Torvalds out: 2561da177e4SLinus Torvalds return skb; 2571da177e4SLinus Torvalds nodata: 2588798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2591da177e4SLinus Torvalds skb = NULL; 2601da177e4SLinus Torvalds goto out; 2611da177e4SLinus Torvalds } 262b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2631da177e4SLinus Torvalds 264ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 265ba0509b6SJesper Dangaard Brouer static struct sk_buff *__build_skb_around(struct sk_buff *skb, 266ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 267ba0509b6SJesper Dangaard Brouer { 268ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 269ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 270ba0509b6SJesper Dangaard Brouer 271ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 272ba0509b6SJesper Dangaard Brouer 273ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 274ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 275ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 276ba0509b6SJesper Dangaard Brouer skb->head = data; 277ba0509b6SJesper Dangaard Brouer skb->data = data; 278ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 279ba0509b6SJesper Dangaard Brouer skb->end = skb->tail + size; 280ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 281ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 282ba0509b6SJesper Dangaard Brouer 283ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 284ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 285ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 286ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 287ba0509b6SJesper Dangaard Brouer 2886370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 2896370cc3bSAleksandr Nogikh 290ba0509b6SJesper Dangaard Brouer return skb; 291ba0509b6SJesper Dangaard Brouer } 292ba0509b6SJesper Dangaard Brouer 2931da177e4SLinus Torvalds /** 2942ea2f62cSEric Dumazet * __build_skb - build a network buffer 295b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2962ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 297b2b5ce9dSEric Dumazet * 298b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 299deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 3002ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 3012ea2f62cSEric Dumazet * or vmalloc() 302b2b5ce9dSEric Dumazet * The return is the new skb buffer. 303b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 304b2b5ce9dSEric Dumazet * Notes : 305b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 306b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 307b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 308b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 309b2b5ce9dSEric Dumazet * before giving packet to stack. 310b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 311b2b5ce9dSEric Dumazet */ 3122ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 313b2b5ce9dSEric Dumazet { 314b2b5ce9dSEric Dumazet struct sk_buff *skb; 315b2b5ce9dSEric Dumazet 316b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 317ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 318b2b5ce9dSEric Dumazet return NULL; 319b2b5ce9dSEric Dumazet 320b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 321b2b5ce9dSEric Dumazet 322ba0509b6SJesper Dangaard Brouer return __build_skb_around(skb, data, frag_size); 323b2b5ce9dSEric Dumazet } 3242ea2f62cSEric Dumazet 3252ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3262ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3272ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 3282ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 3292ea2f62cSEric Dumazet */ 3302ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3312ea2f62cSEric Dumazet { 3322ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3332ea2f62cSEric Dumazet 3342ea2f62cSEric Dumazet if (skb && frag_size) { 3352ea2f62cSEric Dumazet skb->head_frag = 1; 3362f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 3372ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3382ea2f62cSEric Dumazet } 3392ea2f62cSEric Dumazet return skb; 3402ea2f62cSEric Dumazet } 341b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 342b2b5ce9dSEric Dumazet 343ba0509b6SJesper Dangaard Brouer /** 344ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 345ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 346ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 347ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 348ba0509b6SJesper Dangaard Brouer */ 349ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 350ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 351ba0509b6SJesper Dangaard Brouer { 352ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 353ba0509b6SJesper Dangaard Brouer return NULL; 354ba0509b6SJesper Dangaard Brouer 355ba0509b6SJesper Dangaard Brouer skb = __build_skb_around(skb, data, frag_size); 356ba0509b6SJesper Dangaard Brouer 357ba0509b6SJesper Dangaard Brouer if (skb && frag_size) { 358ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 359ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 360ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 361ba0509b6SJesper Dangaard Brouer } 362ba0509b6SJesper Dangaard Brouer return skb; 363ba0509b6SJesper Dangaard Brouer } 364ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 365ba0509b6SJesper Dangaard Brouer 366795bb1c0SJesper Dangaard Brouer #define NAPI_SKB_CACHE_SIZE 64 367795bb1c0SJesper Dangaard Brouer 368795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache { 369795bb1c0SJesper Dangaard Brouer struct page_frag_cache page; 370e0d7924aSAlexey Dobriyan unsigned int skb_count; 371795bb1c0SJesper Dangaard Brouer void *skb_cache[NAPI_SKB_CACHE_SIZE]; 372795bb1c0SJesper Dangaard Brouer }; 373795bb1c0SJesper Dangaard Brouer 374b63ae8caSAlexander Duyck static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 375795bb1c0SJesper Dangaard Brouer static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 376ffde7328SAlexander Duyck 377ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 378ffde7328SAlexander Duyck { 379795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 3809451980aSAlexander Duyck 3818c2dd3e4SAlexander Duyck return page_frag_alloc(&nc->page, fragsz, gfp_mask); 382ffde7328SAlexander Duyck } 383ffde7328SAlexander Duyck 384ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 385ffde7328SAlexander Duyck { 3863bed3cc4SAlexander Duyck fragsz = SKB_DATA_ALIGN(fragsz); 3873bed3cc4SAlexander Duyck 388453f85d4SMel Gorman return __napi_alloc_frag(fragsz, GFP_ATOMIC); 389ffde7328SAlexander Duyck } 390ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 391ffde7328SAlexander Duyck 3926f532612SEric Dumazet /** 3937ba7aeabSSebastian Andrzej Siewior * netdev_alloc_frag - allocate a page fragment 3947ba7aeabSSebastian Andrzej Siewior * @fragsz: fragment size 3957ba7aeabSSebastian Andrzej Siewior * 3967ba7aeabSSebastian Andrzej Siewior * Allocates a frag from a page for receive buffer. 3977ba7aeabSSebastian Andrzej Siewior * Uses GFP_ATOMIC allocations. 3987ba7aeabSSebastian Andrzej Siewior */ 3997ba7aeabSSebastian Andrzej Siewior void *netdev_alloc_frag(unsigned int fragsz) 4007ba7aeabSSebastian Andrzej Siewior { 4017ba7aeabSSebastian Andrzej Siewior struct page_frag_cache *nc; 4027ba7aeabSSebastian Andrzej Siewior void *data; 4037ba7aeabSSebastian Andrzej Siewior 4047ba7aeabSSebastian Andrzej Siewior fragsz = SKB_DATA_ALIGN(fragsz); 4057ba7aeabSSebastian Andrzej Siewior if (in_irq() || irqs_disabled()) { 4067ba7aeabSSebastian Andrzej Siewior nc = this_cpu_ptr(&netdev_alloc_cache); 4077ba7aeabSSebastian Andrzej Siewior data = page_frag_alloc(nc, fragsz, GFP_ATOMIC); 4087ba7aeabSSebastian Andrzej Siewior } else { 4097ba7aeabSSebastian Andrzej Siewior local_bh_disable(); 4107ba7aeabSSebastian Andrzej Siewior data = __napi_alloc_frag(fragsz, GFP_ATOMIC); 4117ba7aeabSSebastian Andrzej Siewior local_bh_enable(); 4127ba7aeabSSebastian Andrzej Siewior } 4137ba7aeabSSebastian Andrzej Siewior return data; 4147ba7aeabSSebastian Andrzej Siewior } 4157ba7aeabSSebastian Andrzej Siewior EXPORT_SYMBOL(netdev_alloc_frag); 4167ba7aeabSSebastian Andrzej Siewior 4177ba7aeabSSebastian Andrzej Siewior /** 418fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 419fd11a83dSAlexander Duyck * @dev: network device to receive on 420d7499160SMasanari Iida * @len: length to allocate 421fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 422fd11a83dSAlexander Duyck * 423fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 424fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 425fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 426fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 427fd11a83dSAlexander Duyck * 428fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 429fd11a83dSAlexander Duyck */ 4309451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4319451980aSAlexander Duyck gfp_t gfp_mask) 432fd11a83dSAlexander Duyck { 433b63ae8caSAlexander Duyck struct page_frag_cache *nc; 434fd11a83dSAlexander Duyck struct sk_buff *skb; 4359451980aSAlexander Duyck bool pfmemalloc; 4369451980aSAlexander Duyck void *data; 437fd11a83dSAlexander Duyck 4389451980aSAlexander Duyck len += NET_SKB_PAD; 439fd11a83dSAlexander Duyck 4409451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 441d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 442a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 443a080e7bdSAlexander Duyck if (!skb) 444a080e7bdSAlexander Duyck goto skb_fail; 445a080e7bdSAlexander Duyck goto skb_success; 446a080e7bdSAlexander Duyck } 4479451980aSAlexander Duyck 4489451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4499451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 4509451980aSAlexander Duyck 4519451980aSAlexander Duyck if (sk_memalloc_socks()) 4529451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 4539451980aSAlexander Duyck 45492dcabd7SSebastian Andrzej Siewior if (in_irq() || irqs_disabled()) { 4559451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 4568c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 4579451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 45892dcabd7SSebastian Andrzej Siewior } else { 45992dcabd7SSebastian Andrzej Siewior local_bh_disable(); 46092dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 46192dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 46292dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 46392dcabd7SSebastian Andrzej Siewior local_bh_enable(); 46492dcabd7SSebastian Andrzej Siewior } 4659451980aSAlexander Duyck 4669451980aSAlexander Duyck if (unlikely(!data)) 4679451980aSAlexander Duyck return NULL; 4689451980aSAlexander Duyck 4699451980aSAlexander Duyck skb = __build_skb(data, len); 4709451980aSAlexander Duyck if (unlikely(!skb)) { 471181edb2bSAlexander Duyck skb_free_frag(data); 4729451980aSAlexander Duyck return NULL; 4739451980aSAlexander Duyck } 4749451980aSAlexander Duyck 4759451980aSAlexander Duyck if (pfmemalloc) 4769451980aSAlexander Duyck skb->pfmemalloc = 1; 4779451980aSAlexander Duyck skb->head_frag = 1; 4789451980aSAlexander Duyck 479a080e7bdSAlexander Duyck skb_success: 4808af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4817b2e497aSChristoph Hellwig skb->dev = dev; 482fd11a83dSAlexander Duyck 483a080e7bdSAlexander Duyck skb_fail: 4848af27456SChristoph Hellwig return skb; 4858af27456SChristoph Hellwig } 486b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4871da177e4SLinus Torvalds 488fd11a83dSAlexander Duyck /** 489fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 490fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 491d7499160SMasanari Iida * @len: length to allocate 492fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 493fd11a83dSAlexander Duyck * 494fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 495fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 496fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 497fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 498fd11a83dSAlexander Duyck * 499fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 500fd11a83dSAlexander Duyck */ 5019451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 5029451980aSAlexander Duyck gfp_t gfp_mask) 503fd11a83dSAlexander Duyck { 504795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 505fd11a83dSAlexander Duyck struct sk_buff *skb; 5069451980aSAlexander Duyck void *data; 507fd11a83dSAlexander Duyck 5089451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 509fd11a83dSAlexander Duyck 5109451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 511d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 512a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 513a080e7bdSAlexander Duyck if (!skb) 514a080e7bdSAlexander Duyck goto skb_fail; 515a080e7bdSAlexander Duyck goto skb_success; 516a080e7bdSAlexander Duyck } 5179451980aSAlexander Duyck 5189451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5199451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5209451980aSAlexander Duyck 5219451980aSAlexander Duyck if (sk_memalloc_socks()) 5229451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5239451980aSAlexander Duyck 5248c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5259451980aSAlexander Duyck if (unlikely(!data)) 5269451980aSAlexander Duyck return NULL; 5279451980aSAlexander Duyck 5289451980aSAlexander Duyck skb = __build_skb(data, len); 5299451980aSAlexander Duyck if (unlikely(!skb)) { 530181edb2bSAlexander Duyck skb_free_frag(data); 5319451980aSAlexander Duyck return NULL; 5329451980aSAlexander Duyck } 5339451980aSAlexander Duyck 534795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5359451980aSAlexander Duyck skb->pfmemalloc = 1; 5369451980aSAlexander Duyck skb->head_frag = 1; 5379451980aSAlexander Duyck 538a080e7bdSAlexander Duyck skb_success: 539fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 540fd11a83dSAlexander Duyck skb->dev = napi->dev; 541fd11a83dSAlexander Duyck 542a080e7bdSAlexander Duyck skb_fail: 543fd11a83dSAlexander Duyck return skb; 544fd11a83dSAlexander Duyck } 545fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 546fd11a83dSAlexander Duyck 547654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 54850269e19SEric Dumazet int size, unsigned int truesize) 549654bed16SPeter Zijlstra { 550654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 551654bed16SPeter Zijlstra skb->len += size; 552654bed16SPeter Zijlstra skb->data_len += size; 55350269e19SEric Dumazet skb->truesize += truesize; 554654bed16SPeter Zijlstra } 555654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 556654bed16SPeter Zijlstra 557f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 558f8e617e1SJason Wang unsigned int truesize) 559f8e617e1SJason Wang { 560f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 561f8e617e1SJason Wang 562f8e617e1SJason Wang skb_frag_size_add(frag, size); 563f8e617e1SJason Wang skb->len += size; 564f8e617e1SJason Wang skb->data_len += size; 565f8e617e1SJason Wang skb->truesize += truesize; 566f8e617e1SJason Wang } 567f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 568f8e617e1SJason Wang 56927b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5701da177e4SLinus Torvalds { 571bd8a7036SEric Dumazet kfree_skb_list(*listp); 57227b437c8SHerbert Xu *listp = NULL; 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds 57527b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 57627b437c8SHerbert Xu { 57727b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 57827b437c8SHerbert Xu } 57927b437c8SHerbert Xu 5801da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5811da177e4SLinus Torvalds { 5821da177e4SLinus Torvalds struct sk_buff *list; 5831da177e4SLinus Torvalds 584fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5851da177e4SLinus Torvalds skb_get(list); 5861da177e4SLinus Torvalds } 5871da177e4SLinus Torvalds 588d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 589d3836f21SEric Dumazet { 590181edb2bSAlexander Duyck unsigned char *head = skb->head; 591181edb2bSAlexander Duyck 592d3836f21SEric Dumazet if (skb->head_frag) 593181edb2bSAlexander Duyck skb_free_frag(head); 594d3836f21SEric Dumazet else 595181edb2bSAlexander Duyck kfree(head); 596d3836f21SEric Dumazet } 597d3836f21SEric Dumazet 5985bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5991da177e4SLinus Torvalds { 600ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6011da177e4SLinus Torvalds int i; 602ff04a771SEric Dumazet 603ff04a771SEric Dumazet if (skb->cloned && 604ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 605ff04a771SEric Dumazet &shinfo->dataref)) 606ff04a771SEric Dumazet return; 607ff04a771SEric Dumazet 608ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 609ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 6101da177e4SLinus Torvalds 611ff04a771SEric Dumazet if (shinfo->frag_list) 612ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6131da177e4SLinus Torvalds 6141f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 615d3836f21SEric Dumazet skb_free_head(skb); 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds 6181da177e4SLinus Torvalds /* 6191da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6201da177e4SLinus Torvalds */ 6212d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6221da177e4SLinus Torvalds { 623d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 624d179cd12SDavid S. Miller 625d179cd12SDavid S. Miller switch (skb->fclone) { 626d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6271da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6286ffe75ebSEric Dumazet return; 629d179cd12SDavid S. Miller 630d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 631d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6326ffe75ebSEric Dumazet 6336ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6346ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6356ffe75ebSEric Dumazet * while here, branch prediction will be good. 6366ffe75ebSEric Dumazet */ 6372638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 6386ffe75ebSEric Dumazet goto fastpath; 639d179cd12SDavid S. Miller break; 640d179cd12SDavid S. Miller 6416ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 642d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 643d179cd12SDavid S. Miller break; 6443ff50b79SStephen Hemminger } 6452638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 6466ffe75ebSEric Dumazet return; 6476ffe75ebSEric Dumazet fastpath: 6486ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6491da177e4SLinus Torvalds } 6501da177e4SLinus Torvalds 6510a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 6521da177e4SLinus Torvalds { 653adf30907SEric Dumazet skb_dst_drop(skb); 6541da177e4SLinus Torvalds if (skb->destructor) { 6559c2b3328SStephen Hemminger WARN_ON(in_irq()); 6561da177e4SLinus Torvalds skb->destructor(skb); 6571da177e4SLinus Torvalds } 658a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 659cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 6602fc72c7bSKOVACS Krisztian #endif 661df5042f4SFlorian Westphal skb_ext_put(skb); 66204a4bb55SLennert Buytenhek } 66304a4bb55SLennert Buytenhek 66404a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 66504a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 66604a4bb55SLennert Buytenhek { 66704a4bb55SLennert Buytenhek skb_release_head_state(skb); 668a28b1b90SFlorian Westphal if (likely(skb->head)) 6692d4baff8SHerbert Xu skb_release_data(skb); 6702d4baff8SHerbert Xu } 6711da177e4SLinus Torvalds 6722d4baff8SHerbert Xu /** 6732d4baff8SHerbert Xu * __kfree_skb - private function 6742d4baff8SHerbert Xu * @skb: buffer 6752d4baff8SHerbert Xu * 6762d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6772d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6782d4baff8SHerbert Xu * always call kfree_skb 6792d4baff8SHerbert Xu */ 6802d4baff8SHerbert Xu 6812d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6822d4baff8SHerbert Xu { 6832d4baff8SHerbert Xu skb_release_all(skb); 6841da177e4SLinus Torvalds kfree_skbmem(skb); 6851da177e4SLinus Torvalds } 686b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds /** 689231d06aeSJörn Engel * kfree_skb - free an sk_buff 690231d06aeSJörn Engel * @skb: buffer to free 691231d06aeSJörn Engel * 692231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 693231d06aeSJörn Engel * hit zero. 694231d06aeSJörn Engel */ 695231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 696231d06aeSJörn Engel { 6973889a803SPaolo Abeni if (!skb_unref(skb)) 698231d06aeSJörn Engel return; 6993889a803SPaolo Abeni 700ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 701231d06aeSJörn Engel __kfree_skb(skb); 702231d06aeSJörn Engel } 703b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 704231d06aeSJörn Engel 705bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 706bd8a7036SEric Dumazet { 707bd8a7036SEric Dumazet while (segs) { 708bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 709bd8a7036SEric Dumazet 710bd8a7036SEric Dumazet kfree_skb(segs); 711bd8a7036SEric Dumazet segs = next; 712bd8a7036SEric Dumazet } 713bd8a7036SEric Dumazet } 714bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 715bd8a7036SEric Dumazet 7166413139dSWillem de Bruijn /* Dump skb information and contents. 7176413139dSWillem de Bruijn * 7186413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 7196413139dSWillem de Bruijn * 720302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 7216413139dSWillem de Bruijn */ 7226413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 7236413139dSWillem de Bruijn { 7246413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 7256413139dSWillem de Bruijn struct net_device *dev = skb->dev; 7266413139dSWillem de Bruijn struct sock *sk = skb->sk; 7276413139dSWillem de Bruijn struct sk_buff *list_skb; 7286413139dSWillem de Bruijn bool has_mac, has_trans; 7296413139dSWillem de Bruijn int headroom, tailroom; 7306413139dSWillem de Bruijn int i, len, seg_len; 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 824be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 82525121173SMichael S. Tsirkin /** 826ead2ceb0SNeil Horman * consume_skb - free an skbuff 827ead2ceb0SNeil Horman * @skb: buffer to free 828ead2ceb0SNeil Horman * 829ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 830ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 831ead2ceb0SNeil Horman * is being dropped after a failure and notes that 832ead2ceb0SNeil Horman */ 833ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 834ead2ceb0SNeil Horman { 8353889a803SPaolo Abeni if (!skb_unref(skb)) 836ead2ceb0SNeil Horman return; 8373889a803SPaolo Abeni 83807dc22e7SKoki Sanagi trace_consume_skb(skb); 839ead2ceb0SNeil Horman __kfree_skb(skb); 840ead2ceb0SNeil Horman } 841ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 842be769db2SHerbert Xu #endif 843ead2ceb0SNeil Horman 8440a463c78SPaolo Abeni /** 845c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 8460a463c78SPaolo Abeni * @skb: buffer to free 8470a463c78SPaolo Abeni * 848ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 849ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 8500a463c78SPaolo Abeni */ 851ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 8520a463c78SPaolo Abeni { 8530a463c78SPaolo Abeni trace_consume_skb(skb); 8540a463c78SPaolo Abeni skb_release_data(skb); 8550a463c78SPaolo Abeni kfree_skbmem(skb); 8560a463c78SPaolo Abeni } 8570a463c78SPaolo Abeni 858795bb1c0SJesper Dangaard Brouer void __kfree_skb_flush(void) 859795bb1c0SJesper Dangaard Brouer { 860795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 861795bb1c0SJesper Dangaard Brouer 862795bb1c0SJesper Dangaard Brouer /* flush skb_cache if containing objects */ 863795bb1c0SJesper Dangaard Brouer if (nc->skb_count) { 864795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 865795bb1c0SJesper Dangaard Brouer nc->skb_cache); 866795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 867795bb1c0SJesper Dangaard Brouer } 868795bb1c0SJesper Dangaard Brouer } 869795bb1c0SJesper Dangaard Brouer 87015fad714SJesper Dangaard Brouer static inline void _kfree_skb_defer(struct sk_buff *skb) 871795bb1c0SJesper Dangaard Brouer { 872795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 873795bb1c0SJesper Dangaard Brouer 874795bb1c0SJesper Dangaard Brouer /* drop skb->head and call any destructors for packet */ 875795bb1c0SJesper Dangaard Brouer skb_release_all(skb); 876795bb1c0SJesper Dangaard Brouer 877795bb1c0SJesper Dangaard Brouer /* record skb to CPU local list */ 878795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 879795bb1c0SJesper Dangaard Brouer 880795bb1c0SJesper Dangaard Brouer #ifdef CONFIG_SLUB 881795bb1c0SJesper Dangaard Brouer /* SLUB writes into objects when freeing */ 882795bb1c0SJesper Dangaard Brouer prefetchw(skb); 883795bb1c0SJesper Dangaard Brouer #endif 884795bb1c0SJesper Dangaard Brouer 885795bb1c0SJesper Dangaard Brouer /* flush skb_cache if it is filled */ 886795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 887795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 888795bb1c0SJesper Dangaard Brouer nc->skb_cache); 889795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 890795bb1c0SJesper Dangaard Brouer } 891795bb1c0SJesper Dangaard Brouer } 89215fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 89315fad714SJesper Dangaard Brouer { 89415fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 89515fad714SJesper Dangaard Brouer } 896795bb1c0SJesper Dangaard Brouer 897795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 898795bb1c0SJesper Dangaard Brouer { 899885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 900795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 901885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 902795bb1c0SJesper Dangaard Brouer return; 903795bb1c0SJesper Dangaard Brouer } 904795bb1c0SJesper Dangaard Brouer 9056454eca8SYunsheng Lin lockdep_assert_in_softirq(); 9066454eca8SYunsheng Lin 9077608894eSPaolo Abeni if (!skb_unref(skb)) 908795bb1c0SJesper Dangaard Brouer return; 9097608894eSPaolo Abeni 910795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 911795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 912795bb1c0SJesper Dangaard Brouer 913795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 914abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 915795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 916795bb1c0SJesper Dangaard Brouer return; 917795bb1c0SJesper Dangaard Brouer } 918795bb1c0SJesper Dangaard Brouer 91915fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 920795bb1c0SJesper Dangaard Brouer } 921795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 922795bb1c0SJesper Dangaard Brouer 923b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 924b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 925b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 926b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 927b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 928b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 929b1937227SEric Dumazet 930dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 931dec18810SHerbert Xu { 932dec18810SHerbert Xu new->tstamp = old->tstamp; 933b1937227SEric Dumazet /* We do not copy old->sk */ 934dec18810SHerbert Xu new->dev = old->dev; 935b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 9367fee226aSEric Dumazet skb_dst_copy(new, old); 937df5042f4SFlorian Westphal __skb_ext_copy(new, old); 938b1937227SEric Dumazet __nf_copy(new, old, false); 9396aa895b0SPatrick McHardy 940b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 941b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 942b1937227SEric Dumazet */ 943b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 94406021292SEliezer Tamir 945b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 946b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 947b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 948b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 949b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 950b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 951b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 952b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 953b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 954b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 955b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 956b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 957b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 958b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 959b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 960b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 961b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 962b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 963b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 964b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 965b1937227SEric Dumazet #endif 966e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 967b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 96806021292SEliezer Tamir #endif 9692bd82484SEric Dumazet #ifdef CONFIG_XPS 9702bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 9712bd82484SEric Dumazet #endif 972b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 973b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 974b1937227SEric Dumazet #endif 975b1937227SEric Dumazet 976dec18810SHerbert Xu } 977dec18810SHerbert Xu 97882c49a35SHerbert Xu /* 97982c49a35SHerbert Xu * You should not add any new code to this function. Add it to 98082c49a35SHerbert Xu * __copy_skb_header above instead. 98182c49a35SHerbert Xu */ 982e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 9831da177e4SLinus Torvalds { 9841da177e4SLinus Torvalds #define C(x) n->x = skb->x 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds n->next = n->prev = NULL; 9871da177e4SLinus Torvalds n->sk = NULL; 988dec18810SHerbert Xu __copy_skb_header(n, skb); 989dec18810SHerbert Xu 9901da177e4SLinus Torvalds C(len); 9911da177e4SLinus Torvalds C(data_len); 9923e6b3b2eSAlexey Dobriyan C(mac_len); 993334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 99402f1c89dSPaul Moore n->cloned = 1; 9951da177e4SLinus Torvalds n->nohdr = 0; 996b13dda9fSEric Dumazet n->peeked = 0; 997e78bfb07SStefano Brivio C(pfmemalloc); 9981da177e4SLinus Torvalds n->destructor = NULL; 9991da177e4SLinus Torvalds C(tail); 10001da177e4SLinus Torvalds C(end); 100102f1c89dSPaul Moore C(head); 1002d3836f21SEric Dumazet C(head_frag); 100302f1c89dSPaul Moore C(data); 100402f1c89dSPaul Moore C(truesize); 100563354797SReshetova, Elena refcount_set(&n->users, 1); 10061da177e4SLinus Torvalds 10071da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 10081da177e4SLinus Torvalds skb->cloned = 1; 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds return n; 1011e0053ec0SHerbert Xu #undef C 1012e0053ec0SHerbert Xu } 1013e0053ec0SHerbert Xu 1014e0053ec0SHerbert Xu /** 1015da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1016da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1017da29e4b4SJakub Kicinski */ 1018da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1019da29e4b4SJakub Kicinski { 1020da29e4b4SJakub Kicinski struct sk_buff *n; 1021da29e4b4SJakub Kicinski 1022da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1023da29e4b4SJakub Kicinski if (!n) 1024da29e4b4SJakub Kicinski return NULL; 1025da29e4b4SJakub Kicinski 1026da29e4b4SJakub Kicinski n->len = first->len; 1027da29e4b4SJakub Kicinski n->data_len = first->len; 1028da29e4b4SJakub Kicinski n->truesize = first->truesize; 1029da29e4b4SJakub Kicinski 1030da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1031da29e4b4SJakub Kicinski 1032da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1033da29e4b4SJakub Kicinski n->destructor = NULL; 1034da29e4b4SJakub Kicinski 1035da29e4b4SJakub Kicinski return n; 1036da29e4b4SJakub Kicinski } 1037da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1038da29e4b4SJakub Kicinski 1039da29e4b4SJakub Kicinski /** 1040e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1041e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1042e0053ec0SHerbert Xu * @src: the skb to supply the contents 1043e0053ec0SHerbert Xu * 1044e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1045e0053ec0SHerbert Xu * supplied by the user. 1046e0053ec0SHerbert Xu * 1047e0053ec0SHerbert Xu * The target skb is returned upon exit. 1048e0053ec0SHerbert Xu */ 1049e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1050e0053ec0SHerbert Xu { 10512d4baff8SHerbert Xu skb_release_all(dst); 1052e0053ec0SHerbert Xu return __skb_clone(dst, src); 1053e0053ec0SHerbert Xu } 1054e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1055e0053ec0SHerbert Xu 10566f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1057a91dbff5SWillem de Bruijn { 1058a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1059a91dbff5SWillem de Bruijn struct user_struct *user; 1060a91dbff5SWillem de Bruijn 1061a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1062a91dbff5SWillem de Bruijn return 0; 1063a91dbff5SWillem de Bruijn 1064a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1065a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1066a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1067a91dbff5SWillem de Bruijn 1068a91dbff5SWillem de Bruijn do { 1069a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1070a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1071a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1072a91dbff5SWillem de Bruijn return -ENOBUFS; 1073a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1074a91dbff5SWillem de Bruijn old_pg); 1075a91dbff5SWillem de Bruijn 1076a91dbff5SWillem de Bruijn if (!mmp->user) { 1077a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1078a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1079a91dbff5SWillem de Bruijn } else { 1080a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1081a91dbff5SWillem de Bruijn } 1082a91dbff5SWillem de Bruijn 1083a91dbff5SWillem de Bruijn return 0; 1084a91dbff5SWillem de Bruijn } 10856f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1086a91dbff5SWillem de Bruijn 10876f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1088a91dbff5SWillem de Bruijn { 1089a91dbff5SWillem de Bruijn if (mmp->user) { 1090a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1091a91dbff5SWillem de Bruijn free_uid(mmp->user); 1092a91dbff5SWillem de Bruijn } 1093a91dbff5SWillem de Bruijn } 10946f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1095a91dbff5SWillem de Bruijn 109652267790SWillem de Bruijn struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size) 109752267790SWillem de Bruijn { 109852267790SWillem de Bruijn struct ubuf_info *uarg; 109952267790SWillem de Bruijn struct sk_buff *skb; 110052267790SWillem de Bruijn 110152267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 110252267790SWillem de Bruijn 110352267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 110452267790SWillem de Bruijn if (!skb) 110552267790SWillem de Bruijn return NULL; 110652267790SWillem de Bruijn 110752267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 110852267790SWillem de Bruijn uarg = (void *)skb->cb; 1109a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1110a91dbff5SWillem de Bruijn 1111a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1112a91dbff5SWillem de Bruijn kfree_skb(skb); 1113a91dbff5SWillem de Bruijn return NULL; 1114a91dbff5SWillem de Bruijn } 111552267790SWillem de Bruijn 111652267790SWillem de Bruijn uarg->callback = sock_zerocopy_callback; 11174ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 11184ab6c99dSWillem de Bruijn uarg->len = 1; 11194ab6c99dSWillem de Bruijn uarg->bytelen = size; 112052267790SWillem de Bruijn uarg->zerocopy = 1; 1121c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 112252267790SWillem de Bruijn sock_hold(sk); 112352267790SWillem de Bruijn 112452267790SWillem de Bruijn return uarg; 112552267790SWillem de Bruijn } 112652267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_alloc); 112752267790SWillem de Bruijn 112852267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 112952267790SWillem de Bruijn { 113052267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 113152267790SWillem de Bruijn } 113252267790SWillem de Bruijn 11334ab6c99dSWillem de Bruijn struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 11344ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 11354ab6c99dSWillem de Bruijn { 11364ab6c99dSWillem de Bruijn if (uarg) { 11374ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 11384ab6c99dSWillem de Bruijn u32 bytelen, next; 11394ab6c99dSWillem de Bruijn 11404ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 11414ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 11424ab6c99dSWillem de Bruijn */ 11434ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 11444ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 11454ab6c99dSWillem de Bruijn return NULL; 11464ab6c99dSWillem de Bruijn } 11474ab6c99dSWillem de Bruijn 11484ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 11494ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 11504ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 11514ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 11524ab6c99dSWillem de Bruijn goto new_alloc; 11534ab6c99dSWillem de Bruijn return NULL; 11544ab6c99dSWillem de Bruijn } 11554ab6c99dSWillem de Bruijn 11564ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 11574ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1158a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1159a91dbff5SWillem de Bruijn return NULL; 11604ab6c99dSWillem de Bruijn uarg->len++; 11614ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 11624ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1163100f6d8eSWillem de Bruijn 1164100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1165100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 1166db5bce32SEric Dumazet sock_zerocopy_get(uarg); 1167100f6d8eSWillem de Bruijn 11684ab6c99dSWillem de Bruijn return uarg; 11694ab6c99dSWillem de Bruijn } 11704ab6c99dSWillem de Bruijn } 11714ab6c99dSWillem de Bruijn 11724ab6c99dSWillem de Bruijn new_alloc: 11734ab6c99dSWillem de Bruijn return sock_zerocopy_alloc(sk, size); 11744ab6c99dSWillem de Bruijn } 11754ab6c99dSWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_realloc); 11764ab6c99dSWillem de Bruijn 11774ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 11784ab6c99dSWillem de Bruijn { 11794ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 11804ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 11814ab6c99dSWillem de Bruijn u64 sum_len; 11824ab6c99dSWillem de Bruijn 11834ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 11844ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 11854ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 11864ab6c99dSWillem de Bruijn 11874ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 11884ab6c99dSWillem de Bruijn return false; 11894ab6c99dSWillem de Bruijn 11904ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 11914ab6c99dSWillem de Bruijn return false; 11924ab6c99dSWillem de Bruijn 11934ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 11944ab6c99dSWillem de Bruijn return true; 11954ab6c99dSWillem de Bruijn } 11964ab6c99dSWillem de Bruijn 119752267790SWillem de Bruijn void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) 119852267790SWillem de Bruijn { 11994ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 120052267790SWillem de Bruijn struct sock_exterr_skb *serr; 120152267790SWillem de Bruijn struct sock *sk = skb->sk; 12024ab6c99dSWillem de Bruijn struct sk_buff_head *q; 12034ab6c99dSWillem de Bruijn unsigned long flags; 12044ab6c99dSWillem de Bruijn u32 lo, hi; 12054ab6c99dSWillem de Bruijn u16 len; 120652267790SWillem de Bruijn 1207ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1208ccaffff1SWillem de Bruijn 12094ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 12104ab6c99dSWillem de Bruijn * so do not queue a completion notification 12114ab6c99dSWillem de Bruijn */ 12124ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 121352267790SWillem de Bruijn goto release; 121452267790SWillem de Bruijn 12154ab6c99dSWillem de Bruijn len = uarg->len; 12164ab6c99dSWillem de Bruijn lo = uarg->id; 12174ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 12184ab6c99dSWillem de Bruijn 121952267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 122052267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 122152267790SWillem de Bruijn serr->ee.ee_errno = 0; 122252267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 12234ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 12244ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 122552267790SWillem de Bruijn if (!success) 122652267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 122752267790SWillem de Bruijn 12284ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 12294ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 12304ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 12314ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 12324ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 12334ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 123452267790SWillem de Bruijn skb = NULL; 12354ab6c99dSWillem de Bruijn } 12364ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 123752267790SWillem de Bruijn 123852267790SWillem de Bruijn sk->sk_error_report(sk); 123952267790SWillem de Bruijn 124052267790SWillem de Bruijn release: 124152267790SWillem de Bruijn consume_skb(skb); 124252267790SWillem de Bruijn sock_put(sk); 124352267790SWillem de Bruijn } 124452267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_callback); 124552267790SWillem de Bruijn 124652267790SWillem de Bruijn void sock_zerocopy_put(struct ubuf_info *uarg) 124752267790SWillem de Bruijn { 1248c1d1b437SEric Dumazet if (uarg && refcount_dec_and_test(&uarg->refcnt)) { 124952267790SWillem de Bruijn if (uarg->callback) 125052267790SWillem de Bruijn uarg->callback(uarg, uarg->zerocopy); 125152267790SWillem de Bruijn else 125252267790SWillem de Bruijn consume_skb(skb_from_uarg(uarg)); 125352267790SWillem de Bruijn } 125452267790SWillem de Bruijn } 125552267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put); 125652267790SWillem de Bruijn 125752900d22SWillem de Bruijn void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 125852267790SWillem de Bruijn { 125952267790SWillem de Bruijn if (uarg) { 126052267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 126152267790SWillem de Bruijn 126252267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 12634ab6c99dSWillem de Bruijn uarg->len--; 126452267790SWillem de Bruijn 126552900d22SWillem de Bruijn if (have_uref) 126652267790SWillem de Bruijn sock_zerocopy_put(uarg); 126752267790SWillem de Bruijn } 126852267790SWillem de Bruijn } 126952267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); 127052267790SWillem de Bruijn 1271b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1272b5947e5dSWillem de Bruijn { 1273b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1274b5947e5dSWillem de Bruijn } 1275b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1276b5947e5dSWillem de Bruijn 127752267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 127852267790SWillem de Bruijn struct msghdr *msg, int len, 127952267790SWillem de Bruijn struct ubuf_info *uarg) 128052267790SWillem de Bruijn { 12814ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 128252267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 128352267790SWillem de Bruijn int err, orig_len = skb->len; 128452267790SWillem de Bruijn 12854ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 12864ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 12874ab6c99dSWillem de Bruijn */ 12884ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 12894ab6c99dSWillem de Bruijn return -EEXIST; 12904ab6c99dSWillem de Bruijn 129152267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 129252267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 129354d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 129454d43117SWillem de Bruijn 129552267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 129652267790SWillem de Bruijn msg->msg_iter = orig_iter; 129754d43117SWillem de Bruijn skb->sk = sk; 129852267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 129954d43117SWillem de Bruijn skb->sk = save_sk; 130052267790SWillem de Bruijn return err; 130152267790SWillem de Bruijn } 130252267790SWillem de Bruijn 130352900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 130452267790SWillem de Bruijn return skb->len - orig_len; 130552267790SWillem de Bruijn } 130652267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 130752267790SWillem de Bruijn 13081f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 130952267790SWillem de Bruijn gfp_t gfp_mask) 131052267790SWillem de Bruijn { 131152267790SWillem de Bruijn if (skb_zcopy(orig)) { 131252267790SWillem de Bruijn if (skb_zcopy(nskb)) { 131352267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 131452267790SWillem de Bruijn if (!gfp_mask) { 131552267790SWillem de Bruijn WARN_ON_ONCE(1); 131652267790SWillem de Bruijn return -ENOMEM; 131752267790SWillem de Bruijn } 131852267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 131952267790SWillem de Bruijn return 0; 132052267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 132152267790SWillem de Bruijn return -EIO; 132252267790SWillem de Bruijn } 132352900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 132452267790SWillem de Bruijn } 132552267790SWillem de Bruijn return 0; 132652267790SWillem de Bruijn } 132752267790SWillem de Bruijn 13282c53040fSBen Hutchings /** 13292c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 133048c83012SMichael S. Tsirkin * @skb: the skb to modify 133148c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 133248c83012SMichael S. Tsirkin * 133348c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 133448c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 133548c83012SMichael S. Tsirkin * to userspace pages. 133648c83012SMichael S. Tsirkin * 133748c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 133848c83012SMichael S. Tsirkin * %GFP_ATOMIC. 133948c83012SMichael S. Tsirkin * 134048c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 134148c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 134248c83012SMichael S. Tsirkin */ 134348c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1344a6686f2fSShirley Ma { 1345a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1346a6686f2fSShirley Ma struct page *page, *head = NULL; 13473ece7826SWillem de Bruijn int i, new_frags; 13483ece7826SWillem de Bruijn u32 d_off; 1349a6686f2fSShirley Ma 13503ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 13513ece7826SWillem de Bruijn return -EINVAL; 13523ece7826SWillem de Bruijn 1353f72c4ac6SWillem de Bruijn if (!num_frags) 1354f72c4ac6SWillem de Bruijn goto release; 1355f72c4ac6SWillem de Bruijn 13563ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 13573ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 135802756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1359a6686f2fSShirley Ma if (!page) { 1360a6686f2fSShirley Ma while (head) { 136140dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1362a6686f2fSShirley Ma put_page(head); 1363a6686f2fSShirley Ma head = next; 1364a6686f2fSShirley Ma } 1365a6686f2fSShirley Ma return -ENOMEM; 1366a6686f2fSShirley Ma } 13673ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 13683ece7826SWillem de Bruijn head = page; 13693ece7826SWillem de Bruijn } 13703ece7826SWillem de Bruijn 13713ece7826SWillem de Bruijn page = head; 13723ece7826SWillem de Bruijn d_off = 0; 13733ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 13743ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 13753ece7826SWillem de Bruijn u32 p_off, p_len, copied; 13763ece7826SWillem de Bruijn struct page *p; 13773ece7826SWillem de Bruijn u8 *vaddr; 1378c613c209SWillem de Bruijn 1379b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1380c613c209SWillem de Bruijn p, p_off, p_len, copied) { 13813ece7826SWillem de Bruijn u32 copy, done = 0; 1382c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 13833ece7826SWillem de Bruijn 13843ece7826SWillem de Bruijn while (done < p_len) { 13853ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 13863ece7826SWillem de Bruijn d_off = 0; 13873ece7826SWillem de Bruijn page = (struct page *)page_private(page); 13883ece7826SWillem de Bruijn } 13893ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 13903ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 13913ece7826SWillem de Bruijn vaddr + p_off + done, copy); 13923ece7826SWillem de Bruijn done += copy; 13933ece7826SWillem de Bruijn d_off += copy; 13943ece7826SWillem de Bruijn } 139551c56b00SEric Dumazet kunmap_atomic(vaddr); 1396c613c209SWillem de Bruijn } 1397a6686f2fSShirley Ma } 1398a6686f2fSShirley Ma 1399a6686f2fSShirley Ma /* skb frags release userspace buffers */ 140002756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1401a8605c60SIan Campbell skb_frag_unref(skb, i); 1402a6686f2fSShirley Ma 1403a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 14043ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 14053ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 140640dadff2SSunghan Suh head = (struct page *)page_private(head); 1407a6686f2fSShirley Ma } 14083ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 14093ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 141048c83012SMichael S. Tsirkin 1411b90ddd56SWillem de Bruijn release: 14121f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1413a6686f2fSShirley Ma return 0; 1414a6686f2fSShirley Ma } 1415dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1416a6686f2fSShirley Ma 1417e0053ec0SHerbert Xu /** 1418e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1419e0053ec0SHerbert Xu * @skb: buffer to clone 1420e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1421e0053ec0SHerbert Xu * 1422e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1423e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1424e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1425e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1426e0053ec0SHerbert Xu * 1427e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1428e0053ec0SHerbert Xu * %GFP_ATOMIC. 1429e0053ec0SHerbert Xu */ 1430e0053ec0SHerbert Xu 1431e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1432e0053ec0SHerbert Xu { 1433d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1434d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1435d0bf4a9eSEric Dumazet skb1); 14366ffe75ebSEric Dumazet struct sk_buff *n; 1437e0053ec0SHerbert Xu 143870008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1439a6686f2fSShirley Ma return NULL; 1440a6686f2fSShirley Ma 1441e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 14422638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 14436ffe75ebSEric Dumazet n = &fclones->skb2; 14442638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1445e0053ec0SHerbert Xu } else { 1446c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1447c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1448c93bdd0eSMel Gorman 1449e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1450e0053ec0SHerbert Xu if (!n) 1451e0053ec0SHerbert Xu return NULL; 1452fe55f6d5SVegard Nossum 1453e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1454e0053ec0SHerbert Xu } 1455e0053ec0SHerbert Xu 1456e0053ec0SHerbert Xu return __skb_clone(n, skb); 14571da177e4SLinus Torvalds } 1458b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 14591da177e4SLinus Torvalds 1460b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1461f5b17294SPravin B Shelar { 1462030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1463030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1464030737bcSEric Dumazet skb->csum_start += off; 1465f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1466f5b17294SPravin B Shelar skb->transport_header += off; 1467f5b17294SPravin B Shelar skb->network_header += off; 1468f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1469f5b17294SPravin B Shelar skb->mac_header += off; 1470f5b17294SPravin B Shelar skb->inner_transport_header += off; 1471f5b17294SPravin B Shelar skb->inner_network_header += off; 1472aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1473f5b17294SPravin B Shelar } 1474b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1475f5b17294SPravin B Shelar 147608303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 14771da177e4SLinus Torvalds { 1478dec18810SHerbert Xu __copy_skb_header(new, old); 1479dec18810SHerbert Xu 14807967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 14817967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 14827967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 14831da177e4SLinus Torvalds } 148408303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 14851da177e4SLinus Torvalds 1486c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1487c93bdd0eSMel Gorman { 1488c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1489c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1490c93bdd0eSMel Gorman return 0; 1491c93bdd0eSMel Gorman } 1492c93bdd0eSMel Gorman 14931da177e4SLinus Torvalds /** 14941da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 14951da177e4SLinus Torvalds * @skb: buffer to copy 14961da177e4SLinus Torvalds * @gfp_mask: allocation priority 14971da177e4SLinus Torvalds * 14981da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 14991da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 15001da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 15011da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 15021da177e4SLinus Torvalds * 15031da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 15041da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 15051da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 15061da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 15071da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 15081da177e4SLinus Torvalds */ 15091da177e4SLinus Torvalds 1510dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 15111da177e4SLinus Torvalds { 15126602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1513ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1514c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1515c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 15166602cebbSEric Dumazet 15171da177e4SLinus Torvalds if (!n) 15181da177e4SLinus Torvalds return NULL; 15191da177e4SLinus Torvalds 15201da177e4SLinus Torvalds /* Set the data pointer */ 15211da177e4SLinus Torvalds skb_reserve(n, headerlen); 15221da177e4SLinus Torvalds /* Set the tail pointer and length */ 15231da177e4SLinus Torvalds skb_put(n, skb->len); 15241da177e4SLinus Torvalds 15259f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 15261da177e4SLinus Torvalds 152708303c18SIlya Lesokhin skb_copy_header(n, skb); 15281da177e4SLinus Torvalds return n; 15291da177e4SLinus Torvalds } 1530b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 15311da177e4SLinus Torvalds 15321da177e4SLinus Torvalds /** 1533bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 15341da177e4SLinus Torvalds * @skb: buffer to copy 1535117632e6SEric Dumazet * @headroom: headroom of new skb 15361da177e4SLinus Torvalds * @gfp_mask: allocation priority 1537bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1538bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1539bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 15401da177e4SLinus Torvalds * 15411da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 15421da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 15431da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 15441da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 15451da177e4SLinus Torvalds * or the pointer to the buffer on success. 15461da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 15471da177e4SLinus Torvalds */ 15481da177e4SLinus Torvalds 1549bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1550bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 15511da177e4SLinus Torvalds { 1552117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1553bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1554bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 15556602cebbSEric Dumazet 15561da177e4SLinus Torvalds if (!n) 15571da177e4SLinus Torvalds goto out; 15581da177e4SLinus Torvalds 15591da177e4SLinus Torvalds /* Set the data pointer */ 1560117632e6SEric Dumazet skb_reserve(n, headroom); 15611da177e4SLinus Torvalds /* Set the tail pointer and length */ 15621da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 15631da177e4SLinus Torvalds /* Copy the bytes */ 1564d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 15651da177e4SLinus Torvalds 156625f484a6SHerbert Xu n->truesize += skb->data_len; 15671da177e4SLinus Torvalds n->data_len = skb->data_len; 15681da177e4SLinus Torvalds n->len = skb->len; 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 15711da177e4SLinus Torvalds int i; 15721da177e4SLinus Torvalds 15731f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 15741f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 15751511022cSDan Carpenter kfree_skb(n); 15761511022cSDan Carpenter n = NULL; 1577a6686f2fSShirley Ma goto out; 1578a6686f2fSShirley Ma } 15791da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 15801da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1581ea2ab693SIan Campbell skb_frag_ref(skb, i); 15821da177e4SLinus Torvalds } 15831da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 15841da177e4SLinus Torvalds } 15851da177e4SLinus Torvalds 158621dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 15871da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 15881da177e4SLinus Torvalds skb_clone_fraglist(n); 15891da177e4SLinus Torvalds } 15901da177e4SLinus Torvalds 159108303c18SIlya Lesokhin skb_copy_header(n, skb); 15921da177e4SLinus Torvalds out: 15931da177e4SLinus Torvalds return n; 15941da177e4SLinus Torvalds } 1595bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 15961da177e4SLinus Torvalds 15971da177e4SLinus Torvalds /** 15981da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 15991da177e4SLinus Torvalds * @skb: buffer to reallocate 16001da177e4SLinus Torvalds * @nhead: room to add at head 16011da177e4SLinus Torvalds * @ntail: room to add at tail 16021da177e4SLinus Torvalds * @gfp_mask: allocation priority 16031da177e4SLinus Torvalds * 1604bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1605bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 16061da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 16071da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 16081da177e4SLinus Torvalds * 16091da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 16101da177e4SLinus Torvalds * reloaded after call to this function. 16111da177e4SLinus Torvalds */ 16121da177e4SLinus Torvalds 161386a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1614dd0fc66fSAl Viro gfp_t gfp_mask) 16151da177e4SLinus Torvalds { 1616158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1617158f323bSEric Dumazet int size = osize + nhead + ntail; 16181da177e4SLinus Torvalds long off; 1619158f323bSEric Dumazet u8 *data; 16201da177e4SLinus Torvalds 16214edd87adSHerbert Xu BUG_ON(nhead < 0); 16224edd87adSHerbert Xu 16239f77fad3STim Hansen BUG_ON(skb_shared(skb)); 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 16261da177e4SLinus Torvalds 1627c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1628c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1629c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1630c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 16311da177e4SLinus Torvalds if (!data) 16321da177e4SLinus Torvalds goto nodata; 163387151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 16366602cebbSEric Dumazet * optimized for the cases when header is void. 16376602cebbSEric Dumazet */ 16386602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 16396602cebbSEric Dumazet 16406602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 16416602cebbSEric Dumazet skb_shinfo(skb), 1642fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 16431da177e4SLinus Torvalds 16443e24591aSAlexander Duyck /* 16453e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 16463e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 16473e24591aSAlexander Duyck * be since all we did is relocate the values 16483e24591aSAlexander Duyck */ 16493e24591aSAlexander Duyck if (skb_cloned(skb)) { 165070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1651a6686f2fSShirley Ma goto nofrags; 16521f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1653c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 16541da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1655ea2ab693SIan Campbell skb_frag_ref(skb, i); 16561da177e4SLinus Torvalds 165721dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 16581da177e4SLinus Torvalds skb_clone_fraglist(skb); 16591da177e4SLinus Torvalds 16601da177e4SLinus Torvalds skb_release_data(skb); 16613e24591aSAlexander Duyck } else { 16623e24591aSAlexander Duyck skb_free_head(skb); 16631fd63041SEric Dumazet } 16641da177e4SLinus Torvalds off = (data + nhead) - skb->head; 16651da177e4SLinus Torvalds 16661da177e4SLinus Torvalds skb->head = data; 1667d3836f21SEric Dumazet skb->head_frag = 0; 16681da177e4SLinus Torvalds skb->data += off; 16694305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 16704305b541SArnaldo Carvalho de Melo skb->end = size; 167156eb8882SPatrick McHardy off = nhead; 16724305b541SArnaldo Carvalho de Melo #else 16734305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 167456eb8882SPatrick McHardy #endif 167527a884dcSArnaldo Carvalho de Melo skb->tail += off; 1676b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 16771da177e4SLinus Torvalds skb->cloned = 0; 1678334a8132SPatrick McHardy skb->hdr_len = 0; 16791da177e4SLinus Torvalds skb->nohdr = 0; 16801da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1681158f323bSEric Dumazet 1682de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1683de8f3a83SDaniel Borkmann 1684158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1685158f323bSEric Dumazet * For the moment, we really care of rx path, or 1686158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1687158f323bSEric Dumazet */ 1688158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1689158f323bSEric Dumazet skb->truesize += size - osize; 1690158f323bSEric Dumazet 16911da177e4SLinus Torvalds return 0; 16921da177e4SLinus Torvalds 1693a6686f2fSShirley Ma nofrags: 1694a6686f2fSShirley Ma kfree(data); 16951da177e4SLinus Torvalds nodata: 16961da177e4SLinus Torvalds return -ENOMEM; 16971da177e4SLinus Torvalds } 1698b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 16991da177e4SLinus Torvalds 17001da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 17011da177e4SLinus Torvalds 17021da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 17031da177e4SLinus Torvalds { 17041da177e4SLinus Torvalds struct sk_buff *skb2; 17051da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 17061da177e4SLinus Torvalds 17071da177e4SLinus Torvalds if (delta <= 0) 17081da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 17091da177e4SLinus Torvalds else { 17101da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 17111da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 17121da177e4SLinus Torvalds GFP_ATOMIC)) { 17131da177e4SLinus Torvalds kfree_skb(skb2); 17141da177e4SLinus Torvalds skb2 = NULL; 17151da177e4SLinus Torvalds } 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds return skb2; 17181da177e4SLinus Torvalds } 1719b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds /** 17221da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 17231da177e4SLinus Torvalds * @skb: buffer to copy 17241da177e4SLinus Torvalds * @newheadroom: new free bytes at head 17251da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 17261da177e4SLinus Torvalds * @gfp_mask: allocation priority 17271da177e4SLinus Torvalds * 17281da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 17291da177e4SLinus Torvalds * allocate additional space. 17301da177e4SLinus Torvalds * 17311da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 17321da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 17331da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 17341da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 17351da177e4SLinus Torvalds * 17361da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 17371da177e4SLinus Torvalds * is called from an interrupt. 17381da177e4SLinus Torvalds */ 17391da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 174086a76cafSVictor Fusco int newheadroom, int newtailroom, 1741dd0fc66fSAl Viro gfp_t gfp_mask) 17421da177e4SLinus Torvalds { 17431da177e4SLinus Torvalds /* 17441da177e4SLinus Torvalds * Allocate the copy buffer 17451da177e4SLinus Torvalds */ 1746c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1747c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1748c93bdd0eSMel Gorman NUMA_NO_NODE); 1749efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 17501da177e4SLinus Torvalds int head_copy_len, head_copy_off; 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds if (!n) 17531da177e4SLinus Torvalds return NULL; 17541da177e4SLinus Torvalds 17551da177e4SLinus Torvalds skb_reserve(n, newheadroom); 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds /* Set the tail pointer and length */ 17581da177e4SLinus Torvalds skb_put(n, skb->len); 17591da177e4SLinus Torvalds 1760efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 17611da177e4SLinus Torvalds head_copy_off = 0; 17621da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 17631da177e4SLinus Torvalds head_copy_len = newheadroom; 17641da177e4SLinus Torvalds else 17651da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds /* Copy the linear header and data. */ 17689f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 17699f77fad3STim Hansen skb->len + head_copy_len)); 17701da177e4SLinus Torvalds 177108303c18SIlya Lesokhin skb_copy_header(n, skb); 17721da177e4SLinus Torvalds 1773030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1774efd1e8d5SPatrick McHardy 17751da177e4SLinus Torvalds return n; 17761da177e4SLinus Torvalds } 1777b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds /** 1780cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 17811da177e4SLinus Torvalds * @skb: buffer to pad 17821da177e4SLinus Torvalds * @pad: space to pad 1783cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 17841da177e4SLinus Torvalds * 17851da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 17861da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 17871da177e4SLinus Torvalds * beyond the buffer end onto the wire. 17881da177e4SLinus Torvalds * 1789cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1790cd0a137aSFlorian Fainelli * if @free_on_error is true. 17911da177e4SLinus Torvalds */ 17921da177e4SLinus Torvalds 1793cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 17941da177e4SLinus Torvalds { 17955b057c6bSHerbert Xu int err; 17965b057c6bSHerbert Xu int ntail; 17971da177e4SLinus Torvalds 17981da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 17995b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 18001da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 18015b057c6bSHerbert Xu return 0; 18021da177e4SLinus Torvalds } 18031da177e4SLinus Torvalds 18044305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 18055b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 18065b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 18075b057c6bSHerbert Xu if (unlikely(err)) 18085b057c6bSHerbert Xu goto free_skb; 18095b057c6bSHerbert Xu } 18105b057c6bSHerbert Xu 18115b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 18125b057c6bSHerbert Xu * to be audited. 18135b057c6bSHerbert Xu */ 18145b057c6bSHerbert Xu err = skb_linearize(skb); 18155b057c6bSHerbert Xu if (unlikely(err)) 18165b057c6bSHerbert Xu goto free_skb; 18175b057c6bSHerbert Xu 18185b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 18195b057c6bSHerbert Xu return 0; 18205b057c6bSHerbert Xu 18215b057c6bSHerbert Xu free_skb: 1822cd0a137aSFlorian Fainelli if (free_on_error) 18231da177e4SLinus Torvalds kfree_skb(skb); 18245b057c6bSHerbert Xu return err; 18251da177e4SLinus Torvalds } 1826cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 18271da177e4SLinus Torvalds 18280dde3e16SIlpo Järvinen /** 18290c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 18300c7ddf36SMathias Krause * @skb: start of the buffer to use 18310c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 18320c7ddf36SMathias Krause * @len: amount of data to add 18330c7ddf36SMathias Krause * 18340c7ddf36SMathias Krause * This function extends the used data area of the potentially 18350c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 18360c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 18370c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 18380c7ddf36SMathias Krause * returned. 18390c7ddf36SMathias Krause */ 18400c7ddf36SMathias Krause 18414df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 18420c7ddf36SMathias Krause { 18430c7ddf36SMathias Krause if (tail != skb) { 18440c7ddf36SMathias Krause skb->data_len += len; 18450c7ddf36SMathias Krause skb->len += len; 18460c7ddf36SMathias Krause } 18470c7ddf36SMathias Krause return skb_put(tail, len); 18480c7ddf36SMathias Krause } 18490c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 18500c7ddf36SMathias Krause 18510c7ddf36SMathias Krause /** 18520dde3e16SIlpo Järvinen * skb_put - add data to a buffer 18530dde3e16SIlpo Järvinen * @skb: buffer to use 18540dde3e16SIlpo Järvinen * @len: amount of data to add 18550dde3e16SIlpo Järvinen * 18560dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 18570dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 18580dde3e16SIlpo Järvinen * first byte of the extra data is returned. 18590dde3e16SIlpo Järvinen */ 18604df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 18610dde3e16SIlpo Järvinen { 18624df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 18630dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 18640dde3e16SIlpo Järvinen skb->tail += len; 18650dde3e16SIlpo Järvinen skb->len += len; 18660dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 18670dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 18680dde3e16SIlpo Järvinen return tmp; 18690dde3e16SIlpo Järvinen } 18700dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 18710dde3e16SIlpo Järvinen 18726be8ac2fSIlpo Järvinen /** 1873c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1874c2aa270aSIlpo Järvinen * @skb: buffer to use 1875c2aa270aSIlpo Järvinen * @len: amount of data to add 1876c2aa270aSIlpo Järvinen * 1877c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1878c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1879c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1880c2aa270aSIlpo Järvinen */ 1881d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 1882c2aa270aSIlpo Järvinen { 1883c2aa270aSIlpo Järvinen skb->data -= len; 1884c2aa270aSIlpo Järvinen skb->len += len; 1885c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 1886c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1887c2aa270aSIlpo Järvinen return skb->data; 1888c2aa270aSIlpo Järvinen } 1889c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1890c2aa270aSIlpo Järvinen 1891c2aa270aSIlpo Järvinen /** 18926be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 18936be8ac2fSIlpo Järvinen * @skb: buffer to use 18946be8ac2fSIlpo Järvinen * @len: amount of data to remove 18956be8ac2fSIlpo Järvinen * 18966be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 18976be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 18986be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 18996be8ac2fSIlpo Järvinen * the old data. 19006be8ac2fSIlpo Järvinen */ 1901af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 19026be8ac2fSIlpo Järvinen { 190347d29646SDavid S. Miller return skb_pull_inline(skb, len); 19046be8ac2fSIlpo Järvinen } 19056be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 19066be8ac2fSIlpo Järvinen 1907419ae74eSIlpo Järvinen /** 1908419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1909419ae74eSIlpo Järvinen * @skb: buffer to alter 1910419ae74eSIlpo Järvinen * @len: new length 1911419ae74eSIlpo Järvinen * 1912419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1913419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1914419ae74eSIlpo Järvinen * The skb must be linear. 1915419ae74eSIlpo Järvinen */ 1916419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1917419ae74eSIlpo Järvinen { 1918419ae74eSIlpo Järvinen if (skb->len > len) 1919419ae74eSIlpo Järvinen __skb_trim(skb, len); 1920419ae74eSIlpo Järvinen } 1921419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1922419ae74eSIlpo Järvinen 19233cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 19241da177e4SLinus Torvalds */ 19251da177e4SLinus Torvalds 19263cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 19271da177e4SLinus Torvalds { 192827b437c8SHerbert Xu struct sk_buff **fragp; 192927b437c8SHerbert Xu struct sk_buff *frag; 19301da177e4SLinus Torvalds int offset = skb_headlen(skb); 19311da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 19321da177e4SLinus Torvalds int i; 193327b437c8SHerbert Xu int err; 193427b437c8SHerbert Xu 193527b437c8SHerbert Xu if (skb_cloned(skb) && 193627b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 193727b437c8SHerbert Xu return err; 19381da177e4SLinus Torvalds 1939f4d26fb3SHerbert Xu i = 0; 1940f4d26fb3SHerbert Xu if (offset >= len) 1941f4d26fb3SHerbert Xu goto drop_pages; 1942f4d26fb3SHerbert Xu 1943f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 19449e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 194527b437c8SHerbert Xu 194627b437c8SHerbert Xu if (end < len) { 19471da177e4SLinus Torvalds offset = end; 194827b437c8SHerbert Xu continue; 19491da177e4SLinus Torvalds } 19501da177e4SLinus Torvalds 19519e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 195227b437c8SHerbert Xu 1953f4d26fb3SHerbert Xu drop_pages: 195427b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 195527b437c8SHerbert Xu 195627b437c8SHerbert Xu for (; i < nfrags; i++) 1957ea2ab693SIan Campbell skb_frag_unref(skb, i); 195827b437c8SHerbert Xu 195921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 196027b437c8SHerbert Xu skb_drop_fraglist(skb); 1961f4d26fb3SHerbert Xu goto done; 196227b437c8SHerbert Xu } 196327b437c8SHerbert Xu 196427b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 196527b437c8SHerbert Xu fragp = &frag->next) { 196627b437c8SHerbert Xu int end = offset + frag->len; 196727b437c8SHerbert Xu 196827b437c8SHerbert Xu if (skb_shared(frag)) { 196927b437c8SHerbert Xu struct sk_buff *nfrag; 197027b437c8SHerbert Xu 197127b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 197227b437c8SHerbert Xu if (unlikely(!nfrag)) 197327b437c8SHerbert Xu return -ENOMEM; 197427b437c8SHerbert Xu 197527b437c8SHerbert Xu nfrag->next = frag->next; 197685bb2a60SEric Dumazet consume_skb(frag); 197727b437c8SHerbert Xu frag = nfrag; 197827b437c8SHerbert Xu *fragp = frag; 197927b437c8SHerbert Xu } 198027b437c8SHerbert Xu 198127b437c8SHerbert Xu if (end < len) { 198227b437c8SHerbert Xu offset = end; 198327b437c8SHerbert Xu continue; 198427b437c8SHerbert Xu } 198527b437c8SHerbert Xu 198627b437c8SHerbert Xu if (end > len && 198727b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 198827b437c8SHerbert Xu return err; 198927b437c8SHerbert Xu 199027b437c8SHerbert Xu if (frag->next) 199127b437c8SHerbert Xu skb_drop_list(&frag->next); 199227b437c8SHerbert Xu break; 199327b437c8SHerbert Xu } 199427b437c8SHerbert Xu 1995f4d26fb3SHerbert Xu done: 199627b437c8SHerbert Xu if (len > skb_headlen(skb)) { 19971da177e4SLinus Torvalds skb->data_len -= skb->len - len; 19981da177e4SLinus Torvalds skb->len = len; 19991da177e4SLinus Torvalds } else { 20001da177e4SLinus Torvalds skb->len = len; 20011da177e4SLinus Torvalds skb->data_len = 0; 200227a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 20031da177e4SLinus Torvalds } 20041da177e4SLinus Torvalds 2005c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2006c21b48ccSEric Dumazet skb_condense(skb); 20071da177e4SLinus Torvalds return 0; 20081da177e4SLinus Torvalds } 2009b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 20101da177e4SLinus Torvalds 201188078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 201288078d98SEric Dumazet */ 201388078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 201488078d98SEric Dumazet { 201588078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 201688078d98SEric Dumazet int delta = skb->len - len; 201788078d98SEric Dumazet 2018d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2019d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2020d55bef50SDimitris Michailidis len); 202154970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 202254970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 202354970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 202454970a2fSVasily Averin 202554970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 202654970a2fSVasily Averin return -EINVAL; 202788078d98SEric Dumazet } 202888078d98SEric Dumazet return __pskb_trim(skb, len); 202988078d98SEric Dumazet } 203088078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 203188078d98SEric Dumazet 20321da177e4SLinus Torvalds /** 20331da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 20341da177e4SLinus Torvalds * @skb: buffer to reallocate 20351da177e4SLinus Torvalds * @delta: number of bytes to advance tail 20361da177e4SLinus Torvalds * 20371da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 20381da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 20391da177e4SLinus Torvalds * data from fragmented part. 20401da177e4SLinus Torvalds * 20411da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 20421da177e4SLinus Torvalds * 20431da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 20441da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 20451da177e4SLinus Torvalds * 20461da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 20471da177e4SLinus Torvalds * reloaded after call to this function. 20481da177e4SLinus Torvalds */ 20491da177e4SLinus Torvalds 20501da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 20511da177e4SLinus Torvalds * when it is necessary. 20521da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 20531da177e4SLinus Torvalds * 2. It may change skb pointers. 20541da177e4SLinus Torvalds * 20551da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 20561da177e4SLinus Torvalds */ 2057af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 20581da177e4SLinus Torvalds { 20591da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 20601da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 20611da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 20621da177e4SLinus Torvalds */ 20634305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 20641da177e4SLinus Torvalds 20651da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 20661da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 20671da177e4SLinus Torvalds GFP_ATOMIC)) 20681da177e4SLinus Torvalds return NULL; 20691da177e4SLinus Torvalds } 20701da177e4SLinus Torvalds 20719f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 20729f77fad3STim Hansen skb_tail_pointer(skb), delta)); 20731da177e4SLinus Torvalds 20741da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 20751da177e4SLinus Torvalds * size of pulled pages. Superb. 20761da177e4SLinus Torvalds */ 207721dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 20781da177e4SLinus Torvalds goto pull_pages; 20791da177e4SLinus Torvalds 20801da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 20811da177e4SLinus Torvalds eat = delta; 20821da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20839e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 20849e903e08SEric Dumazet 20859e903e08SEric Dumazet if (size >= eat) 20861da177e4SLinus Torvalds goto pull_pages; 20879e903e08SEric Dumazet eat -= size; 20881da177e4SLinus Torvalds } 20891da177e4SLinus Torvalds 20901da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 209109001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 20921da177e4SLinus Torvalds * but taking into account that pulling is expected to 20931da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 20941da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 20951da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 20961da177e4SLinus Torvalds */ 20971da177e4SLinus Torvalds if (eat) { 20981da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 20991da177e4SLinus Torvalds struct sk_buff *clone = NULL; 21001da177e4SLinus Torvalds struct sk_buff *insp = NULL; 21011da177e4SLinus Torvalds 21021da177e4SLinus Torvalds do { 21031da177e4SLinus Torvalds if (list->len <= eat) { 21041da177e4SLinus Torvalds /* Eaten as whole. */ 21051da177e4SLinus Torvalds eat -= list->len; 21061da177e4SLinus Torvalds list = list->next; 21071da177e4SLinus Torvalds insp = list; 21081da177e4SLinus Torvalds } else { 21091da177e4SLinus Torvalds /* Eaten partially. */ 21101da177e4SLinus Torvalds 21111da177e4SLinus Torvalds if (skb_shared(list)) { 21121da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 21131da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 21141da177e4SLinus Torvalds if (!clone) 21151da177e4SLinus Torvalds return NULL; 21161da177e4SLinus Torvalds insp = list->next; 21171da177e4SLinus Torvalds list = clone; 21181da177e4SLinus Torvalds } else { 21191da177e4SLinus Torvalds /* This may be pulled without 21201da177e4SLinus Torvalds * problems. */ 21211da177e4SLinus Torvalds insp = list; 21221da177e4SLinus Torvalds } 21231da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 21241da177e4SLinus Torvalds kfree_skb(clone); 21251da177e4SLinus Torvalds return NULL; 21261da177e4SLinus Torvalds } 21271da177e4SLinus Torvalds break; 21281da177e4SLinus Torvalds } 21291da177e4SLinus Torvalds } while (eat); 21301da177e4SLinus Torvalds 21311da177e4SLinus Torvalds /* Free pulled out fragments. */ 21321da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 21331da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 21341da177e4SLinus Torvalds kfree_skb(list); 21351da177e4SLinus Torvalds } 21361da177e4SLinus Torvalds /* And insert new clone at head. */ 21371da177e4SLinus Torvalds if (clone) { 21381da177e4SLinus Torvalds clone->next = list; 21391da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 21401da177e4SLinus Torvalds } 21411da177e4SLinus Torvalds } 21421da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 21431da177e4SLinus Torvalds 21441da177e4SLinus Torvalds pull_pages: 21451da177e4SLinus Torvalds eat = delta; 21461da177e4SLinus Torvalds k = 0; 21471da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21489e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 21499e903e08SEric Dumazet 21509e903e08SEric Dumazet if (size <= eat) { 2151ea2ab693SIan Campbell skb_frag_unref(skb, i); 21529e903e08SEric Dumazet eat -= size; 21531da177e4SLinus Torvalds } else { 2154b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2155b54c9d5bSJonathan Lemon 2156b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 21571da177e4SLinus Torvalds if (eat) { 2158b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2159b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 21603ccc6c6fSlinzhang if (!i) 21613ccc6c6fSlinzhang goto end; 21621da177e4SLinus Torvalds eat = 0; 21631da177e4SLinus Torvalds } 21641da177e4SLinus Torvalds k++; 21651da177e4SLinus Torvalds } 21661da177e4SLinus Torvalds } 21671da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 21681da177e4SLinus Torvalds 21693ccc6c6fSlinzhang end: 21701da177e4SLinus Torvalds skb->tail += delta; 21711da177e4SLinus Torvalds skb->data_len -= delta; 21721da177e4SLinus Torvalds 21731f8b977aSWillem de Bruijn if (!skb->data_len) 21741f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 21751f8b977aSWillem de Bruijn 217627a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 21771da177e4SLinus Torvalds } 2178b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 21791da177e4SLinus Torvalds 218022019b17SEric Dumazet /** 218122019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 218222019b17SEric Dumazet * @skb: source skb 218322019b17SEric Dumazet * @offset: offset in source 218422019b17SEric Dumazet * @to: destination buffer 218522019b17SEric Dumazet * @len: number of bytes to copy 218622019b17SEric Dumazet * 218722019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 218822019b17SEric Dumazet * destination buffer. 218922019b17SEric Dumazet * 219022019b17SEric Dumazet * CAUTION ! : 219122019b17SEric Dumazet * If its prototype is ever changed, 219222019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 219322019b17SEric Dumazet * since it is called from BPF assembly code. 219422019b17SEric Dumazet */ 21951da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 21961da177e4SLinus Torvalds { 21971a028e50SDavid S. Miller int start = skb_headlen(skb); 2198fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2199fbb398a8SDavid S. Miller int i, copy; 22001da177e4SLinus Torvalds 22011da177e4SLinus Torvalds if (offset > (int)skb->len - len) 22021da177e4SLinus Torvalds goto fault; 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds /* Copy header. */ 22051a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 22061da177e4SLinus Torvalds if (copy > len) 22071da177e4SLinus Torvalds copy = len; 2208d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 22091da177e4SLinus Torvalds if ((len -= copy) == 0) 22101da177e4SLinus Torvalds return 0; 22111da177e4SLinus Torvalds offset += copy; 22121da177e4SLinus Torvalds to += copy; 22131da177e4SLinus Torvalds } 22141da177e4SLinus Torvalds 22151da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22161a028e50SDavid S. Miller int end; 221751c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 22181da177e4SLinus Torvalds 2219547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22201a028e50SDavid S. Miller 222151c56b00SEric Dumazet end = start + skb_frag_size(f); 22221da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2223c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2224c613c209SWillem de Bruijn struct page *p; 22251da177e4SLinus Torvalds u8 *vaddr; 22261da177e4SLinus Torvalds 22271da177e4SLinus Torvalds if (copy > len) 22281da177e4SLinus Torvalds copy = len; 22291da177e4SLinus Torvalds 2230c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2231b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2232c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2233c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2234c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 223551c56b00SEric Dumazet kunmap_atomic(vaddr); 2236c613c209SWillem de Bruijn } 22371da177e4SLinus Torvalds 22381da177e4SLinus Torvalds if ((len -= copy) == 0) 22391da177e4SLinus Torvalds return 0; 22401da177e4SLinus Torvalds offset += copy; 22411da177e4SLinus Torvalds to += copy; 22421da177e4SLinus Torvalds } 22431a028e50SDavid S. Miller start = end; 22441da177e4SLinus Torvalds } 22451da177e4SLinus Torvalds 2246fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 22471a028e50SDavid S. Miller int end; 22481da177e4SLinus Torvalds 2249547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22501a028e50SDavid S. Miller 2251fbb398a8SDavid S. Miller end = start + frag_iter->len; 22521da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 22531da177e4SLinus Torvalds if (copy > len) 22541da177e4SLinus Torvalds copy = len; 2255fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 22561da177e4SLinus Torvalds goto fault; 22571da177e4SLinus Torvalds if ((len -= copy) == 0) 22581da177e4SLinus Torvalds return 0; 22591da177e4SLinus Torvalds offset += copy; 22601da177e4SLinus Torvalds to += copy; 22611da177e4SLinus Torvalds } 22621a028e50SDavid S. Miller start = end; 22631da177e4SLinus Torvalds } 2264a6686f2fSShirley Ma 22651da177e4SLinus Torvalds if (!len) 22661da177e4SLinus Torvalds return 0; 22671da177e4SLinus Torvalds 22681da177e4SLinus Torvalds fault: 22691da177e4SLinus Torvalds return -EFAULT; 22701da177e4SLinus Torvalds } 2271b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 22721da177e4SLinus Torvalds 22739c55e01cSJens Axboe /* 22749c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 22759c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 22769c55e01cSJens Axboe */ 22779c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 22789c55e01cSJens Axboe { 22798b9d3728SJarek Poplawski put_page(spd->pages[i]); 22808b9d3728SJarek Poplawski } 22819c55e01cSJens Axboe 2282a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 22834fb66994SJarek Poplawski unsigned int *offset, 228418aafc62SEric Dumazet struct sock *sk) 22858b9d3728SJarek Poplawski { 22865640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 22878b9d3728SJarek Poplawski 22885640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 22898b9d3728SJarek Poplawski return NULL; 22904fb66994SJarek Poplawski 22915640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 22924fb66994SJarek Poplawski 22935640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 22945640f768SEric Dumazet page_address(page) + *offset, *len); 22955640f768SEric Dumazet *offset = pfrag->offset; 22965640f768SEric Dumazet pfrag->offset += *len; 22974fb66994SJarek Poplawski 22985640f768SEric Dumazet return pfrag->page; 22999c55e01cSJens Axboe } 23009c55e01cSJens Axboe 230141c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 230241c73a0dSEric Dumazet struct page *page, 230341c73a0dSEric Dumazet unsigned int offset) 230441c73a0dSEric Dumazet { 230541c73a0dSEric Dumazet return spd->nr_pages && 230641c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 230741c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 230841c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 230941c73a0dSEric Dumazet } 231041c73a0dSEric Dumazet 23119c55e01cSJens Axboe /* 23129c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 23139c55e01cSJens Axboe */ 2314a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 231535f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 23164fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 231718aafc62SEric Dumazet bool linear, 23187a67e56fSJarek Poplawski struct sock *sk) 23199c55e01cSJens Axboe { 232041c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2321a108d5f3SDavid S. Miller return true; 23229c55e01cSJens Axboe 23238b9d3728SJarek Poplawski if (linear) { 232418aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 23258b9d3728SJarek Poplawski if (!page) 2326a108d5f3SDavid S. Miller return true; 232741c73a0dSEric Dumazet } 232841c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 232941c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2330a108d5f3SDavid S. Miller return false; 233141c73a0dSEric Dumazet } 23328b9d3728SJarek Poplawski get_page(page); 23339c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 23344fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 23359c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 23369c55e01cSJens Axboe spd->nr_pages++; 23378b9d3728SJarek Poplawski 2338a108d5f3SDavid S. Miller return false; 23399c55e01cSJens Axboe } 23409c55e01cSJens Axboe 2341a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 23422870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 234318aafc62SEric Dumazet unsigned int *len, 2344d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 234535f3d14dSJens Axboe struct sock *sk, 234635f3d14dSJens Axboe struct pipe_inode_info *pipe) 23479c55e01cSJens Axboe { 23482870c43dSOctavian Purdila if (!*len) 2349a108d5f3SDavid S. Miller return true; 23509c55e01cSJens Axboe 23512870c43dSOctavian Purdila /* skip this segment if already processed */ 23522870c43dSOctavian Purdila if (*off >= plen) { 23532870c43dSOctavian Purdila *off -= plen; 2354a108d5f3SDavid S. Miller return false; 23552870c43dSOctavian Purdila } 23562870c43dSOctavian Purdila 23572870c43dSOctavian Purdila /* ignore any bits we already processed */ 23589ca1b22dSEric Dumazet poff += *off; 23599ca1b22dSEric Dumazet plen -= *off; 23602870c43dSOctavian Purdila *off = 0; 23612870c43dSOctavian Purdila 236218aafc62SEric Dumazet do { 236318aafc62SEric Dumazet unsigned int flen = min(*len, plen); 23642870c43dSOctavian Purdila 236518aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 236618aafc62SEric Dumazet linear, sk)) 2367a108d5f3SDavid S. Miller return true; 236818aafc62SEric Dumazet poff += flen; 236918aafc62SEric Dumazet plen -= flen; 23702870c43dSOctavian Purdila *len -= flen; 237118aafc62SEric Dumazet } while (*len && plen); 23722870c43dSOctavian Purdila 2373a108d5f3SDavid S. Miller return false; 2374db43a282SOctavian Purdila } 23759c55e01cSJens Axboe 23769c55e01cSJens Axboe /* 2377a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 23782870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 23799c55e01cSJens Axboe */ 2380a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 238135f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 238235f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 23832870c43dSOctavian Purdila { 23842870c43dSOctavian Purdila int seg; 2385fa9835e5STom Herbert struct sk_buff *iter; 23869c55e01cSJens Axboe 23871d0c0b32SEric Dumazet /* map the linear part : 23882996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 23892996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 23902996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 23919c55e01cSJens Axboe */ 23922870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 23932870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 23942870c43dSOctavian Purdila skb_headlen(skb), 239518aafc62SEric Dumazet offset, len, spd, 23963a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 23971d0c0b32SEric Dumazet sk, pipe)) 2398a108d5f3SDavid S. Miller return true; 23999c55e01cSJens Axboe 24009c55e01cSJens Axboe /* 24019c55e01cSJens Axboe * then map the fragments 24029c55e01cSJens Axboe */ 24039c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 24049c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 24059c55e01cSJens Axboe 2406ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2407b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 240818aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2409a108d5f3SDavid S. Miller return true; 24109c55e01cSJens Axboe } 24119c55e01cSJens Axboe 2412fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2413fa9835e5STom Herbert if (*offset >= iter->len) { 2414fa9835e5STom Herbert *offset -= iter->len; 2415fa9835e5STom Herbert continue; 2416fa9835e5STom Herbert } 2417fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2418fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2419fa9835e5STom Herbert * case. 2420fa9835e5STom Herbert */ 2421fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2422fa9835e5STom Herbert return true; 2423fa9835e5STom Herbert } 2424fa9835e5STom Herbert 2425a108d5f3SDavid S. Miller return false; 24269c55e01cSJens Axboe } 24279c55e01cSJens Axboe 24289c55e01cSJens Axboe /* 24299c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2430fa9835e5STom Herbert * the fragments, and the frag list. 24319c55e01cSJens Axboe */ 2432a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 24339c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 243425869262SAl Viro unsigned int flags) 24359c55e01cSJens Axboe { 243641c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 243741c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 24389c55e01cSJens Axboe struct splice_pipe_desc spd = { 24399c55e01cSJens Axboe .pages = pages, 24409c55e01cSJens Axboe .partial = partial, 2441047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 244228a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 24439c55e01cSJens Axboe .spd_release = sock_spd_release, 24449c55e01cSJens Axboe }; 244535f3d14dSJens Axboe int ret = 0; 244635f3d14dSJens Axboe 2447fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 24489c55e01cSJens Axboe 2449a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 245025869262SAl Viro ret = splice_to_pipe(pipe, &spd); 24519c55e01cSJens Axboe 245235f3d14dSJens Axboe return ret; 24539c55e01cSJens Axboe } 24542b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 24559c55e01cSJens Axboe 245620bf50deSTom Herbert /* Send skb data on a socket. Socket must be locked. */ 245720bf50deSTom Herbert int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 245820bf50deSTom Herbert int len) 245920bf50deSTom Herbert { 246020bf50deSTom Herbert unsigned int orig_len = len; 246120bf50deSTom Herbert struct sk_buff *head = skb; 246220bf50deSTom Herbert unsigned short fragidx; 246320bf50deSTom Herbert int slen, ret; 246420bf50deSTom Herbert 246520bf50deSTom Herbert do_frag_list: 246620bf50deSTom Herbert 246720bf50deSTom Herbert /* Deal with head data */ 246820bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 246920bf50deSTom Herbert struct kvec kv; 247020bf50deSTom Herbert struct msghdr msg; 247120bf50deSTom Herbert 247220bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 247320bf50deSTom Herbert kv.iov_base = skb->data + offset; 2474db5980d8SJohn Fastabend kv.iov_len = slen; 247520bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2476bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 247720bf50deSTom Herbert 247820bf50deSTom Herbert ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen); 247920bf50deSTom Herbert if (ret <= 0) 248020bf50deSTom Herbert goto error; 248120bf50deSTom Herbert 248220bf50deSTom Herbert offset += ret; 248320bf50deSTom Herbert len -= ret; 248420bf50deSTom Herbert } 248520bf50deSTom Herbert 248620bf50deSTom Herbert /* All the data was skb head? */ 248720bf50deSTom Herbert if (!len) 248820bf50deSTom Herbert goto out; 248920bf50deSTom Herbert 249020bf50deSTom Herbert /* Make offset relative to start of frags */ 249120bf50deSTom Herbert offset -= skb_headlen(skb); 249220bf50deSTom Herbert 249320bf50deSTom Herbert /* Find where we are in frag list */ 249420bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 249520bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 249620bf50deSTom Herbert 2497d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 249820bf50deSTom Herbert break; 249920bf50deSTom Herbert 2500d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 250120bf50deSTom Herbert } 250220bf50deSTom Herbert 250320bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 250420bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 250520bf50deSTom Herbert 2506d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 250720bf50deSTom Herbert 250820bf50deSTom Herbert while (slen) { 2509d8e18a51SMatthew Wilcox (Oracle) ret = kernel_sendpage_locked(sk, skb_frag_page(frag), 2510b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 251120bf50deSTom Herbert slen, MSG_DONTWAIT); 251220bf50deSTom Herbert if (ret <= 0) 251320bf50deSTom Herbert goto error; 251420bf50deSTom Herbert 251520bf50deSTom Herbert len -= ret; 251620bf50deSTom Herbert offset += ret; 251720bf50deSTom Herbert slen -= ret; 251820bf50deSTom Herbert } 251920bf50deSTom Herbert 252020bf50deSTom Herbert offset = 0; 252120bf50deSTom Herbert } 252220bf50deSTom Herbert 252320bf50deSTom Herbert if (len) { 252420bf50deSTom Herbert /* Process any frag lists */ 252520bf50deSTom Herbert 252620bf50deSTom Herbert if (skb == head) { 252720bf50deSTom Herbert if (skb_has_frag_list(skb)) { 252820bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 252920bf50deSTom Herbert goto do_frag_list; 253020bf50deSTom Herbert } 253120bf50deSTom Herbert } else if (skb->next) { 253220bf50deSTom Herbert skb = skb->next; 253320bf50deSTom Herbert goto do_frag_list; 253420bf50deSTom Herbert } 253520bf50deSTom Herbert } 253620bf50deSTom Herbert 253720bf50deSTom Herbert out: 253820bf50deSTom Herbert return orig_len - len; 253920bf50deSTom Herbert 254020bf50deSTom Herbert error: 254120bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 254220bf50deSTom Herbert } 254320bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 254420bf50deSTom Herbert 2545357b40a1SHerbert Xu /** 2546357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2547357b40a1SHerbert Xu * @skb: destination buffer 2548357b40a1SHerbert Xu * @offset: offset in destination 2549357b40a1SHerbert Xu * @from: source buffer 2550357b40a1SHerbert Xu * @len: number of bytes to copy 2551357b40a1SHerbert Xu * 2552357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2553357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2554357b40a1SHerbert Xu * traversing fragment lists and such. 2555357b40a1SHerbert Xu */ 2556357b40a1SHerbert Xu 25570c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2558357b40a1SHerbert Xu { 25591a028e50SDavid S. Miller int start = skb_headlen(skb); 2560fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2561fbb398a8SDavid S. Miller int i, copy; 2562357b40a1SHerbert Xu 2563357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2564357b40a1SHerbert Xu goto fault; 2565357b40a1SHerbert Xu 25661a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2567357b40a1SHerbert Xu if (copy > len) 2568357b40a1SHerbert Xu copy = len; 256927d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2570357b40a1SHerbert Xu if ((len -= copy) == 0) 2571357b40a1SHerbert Xu return 0; 2572357b40a1SHerbert Xu offset += copy; 2573357b40a1SHerbert Xu from += copy; 2574357b40a1SHerbert Xu } 2575357b40a1SHerbert Xu 2576357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2577357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 25781a028e50SDavid S. Miller int end; 2579357b40a1SHerbert Xu 2580547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25811a028e50SDavid S. Miller 25829e903e08SEric Dumazet end = start + skb_frag_size(frag); 2583357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2584c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2585c613c209SWillem de Bruijn struct page *p; 2586357b40a1SHerbert Xu u8 *vaddr; 2587357b40a1SHerbert Xu 2588357b40a1SHerbert Xu if (copy > len) 2589357b40a1SHerbert Xu copy = len; 2590357b40a1SHerbert Xu 2591c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2592b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2593c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2594c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2595c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 259651c56b00SEric Dumazet kunmap_atomic(vaddr); 2597c613c209SWillem de Bruijn } 2598357b40a1SHerbert Xu 2599357b40a1SHerbert Xu if ((len -= copy) == 0) 2600357b40a1SHerbert Xu return 0; 2601357b40a1SHerbert Xu offset += copy; 2602357b40a1SHerbert Xu from += copy; 2603357b40a1SHerbert Xu } 26041a028e50SDavid S. Miller start = end; 2605357b40a1SHerbert Xu } 2606357b40a1SHerbert Xu 2607fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 26081a028e50SDavid S. Miller int end; 2609357b40a1SHerbert Xu 2610547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26111a028e50SDavid S. Miller 2612fbb398a8SDavid S. Miller end = start + frag_iter->len; 2613357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2614357b40a1SHerbert Xu if (copy > len) 2615357b40a1SHerbert Xu copy = len; 2616fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 26171a028e50SDavid S. Miller from, copy)) 2618357b40a1SHerbert Xu goto fault; 2619357b40a1SHerbert Xu if ((len -= copy) == 0) 2620357b40a1SHerbert Xu return 0; 2621357b40a1SHerbert Xu offset += copy; 2622357b40a1SHerbert Xu from += copy; 2623357b40a1SHerbert Xu } 26241a028e50SDavid S. Miller start = end; 2625357b40a1SHerbert Xu } 2626357b40a1SHerbert Xu if (!len) 2627357b40a1SHerbert Xu return 0; 2628357b40a1SHerbert Xu 2629357b40a1SHerbert Xu fault: 2630357b40a1SHerbert Xu return -EFAULT; 2631357b40a1SHerbert Xu } 2632357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2633357b40a1SHerbert Xu 26341da177e4SLinus Torvalds /* Checksum skb data. */ 26352817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 26362817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 26371da177e4SLinus Torvalds { 26381a028e50SDavid S. Miller int start = skb_headlen(skb); 26391a028e50SDavid S. Miller int i, copy = start - offset; 2640fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 26411da177e4SLinus Torvalds int pos = 0; 26421da177e4SLinus Torvalds 26431da177e4SLinus Torvalds /* Checksum header. */ 26441da177e4SLinus Torvalds if (copy > 0) { 26451da177e4SLinus Torvalds if (copy > len) 26461da177e4SLinus Torvalds copy = len; 26472544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 26482544af03SMatteo Croce skb->data + offset, copy, csum); 26491da177e4SLinus Torvalds if ((len -= copy) == 0) 26501da177e4SLinus Torvalds return csum; 26511da177e4SLinus Torvalds offset += copy; 26521da177e4SLinus Torvalds pos = copy; 26531da177e4SLinus Torvalds } 26541da177e4SLinus Torvalds 26551da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 26561a028e50SDavid S. Miller int end; 265751c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 26581da177e4SLinus Torvalds 2659547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26601a028e50SDavid S. Miller 266151c56b00SEric Dumazet end = start + skb_frag_size(frag); 26621da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2663c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2664c613c209SWillem de Bruijn struct page *p; 266544bb9363SAl Viro __wsum csum2; 26661da177e4SLinus Torvalds u8 *vaddr; 26671da177e4SLinus Torvalds 26681da177e4SLinus Torvalds if (copy > len) 26691da177e4SLinus Torvalds copy = len; 2670c613c209SWillem de Bruijn 2671c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2672b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2673c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2674c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 26752544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 26762544af03SMatteo Croce csum_partial_ext, 26772544af03SMatteo Croce vaddr + p_off, p_len, 0); 267851c56b00SEric Dumazet kunmap_atomic(vaddr); 26792544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 26802544af03SMatteo Croce csum_block_add_ext, csum, 26812544af03SMatteo Croce csum2, pos, p_len); 2682c613c209SWillem de Bruijn pos += p_len; 2683c613c209SWillem de Bruijn } 2684c613c209SWillem de Bruijn 26851da177e4SLinus Torvalds if (!(len -= copy)) 26861da177e4SLinus Torvalds return csum; 26871da177e4SLinus Torvalds offset += copy; 26881da177e4SLinus Torvalds } 26891a028e50SDavid S. Miller start = end; 26901da177e4SLinus Torvalds } 26911da177e4SLinus Torvalds 2692fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 26931a028e50SDavid S. Miller int end; 26941da177e4SLinus Torvalds 2695547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26961a028e50SDavid S. Miller 2697fbb398a8SDavid S. Miller end = start + frag_iter->len; 26981da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 26995f92a738SAl Viro __wsum csum2; 27001da177e4SLinus Torvalds if (copy > len) 27011da177e4SLinus Torvalds copy = len; 27022817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 27032817a336SDaniel Borkmann copy, 0, ops); 27042544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 27052544af03SMatteo Croce csum, csum2, pos, copy); 27061da177e4SLinus Torvalds if ((len -= copy) == 0) 27071da177e4SLinus Torvalds return csum; 27081da177e4SLinus Torvalds offset += copy; 27091da177e4SLinus Torvalds pos += copy; 27101da177e4SLinus Torvalds } 27111a028e50SDavid S. Miller start = end; 27121da177e4SLinus Torvalds } 271309a62660SKris Katterjohn BUG_ON(len); 27141da177e4SLinus Torvalds 27151da177e4SLinus Torvalds return csum; 27161da177e4SLinus Torvalds } 27172817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 27182817a336SDaniel Borkmann 27192817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 27202817a336SDaniel Borkmann int len, __wsum csum) 27212817a336SDaniel Borkmann { 27222817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2723cea80ea8SDaniel Borkmann .update = csum_partial_ext, 27242817a336SDaniel Borkmann .combine = csum_block_add_ext, 27252817a336SDaniel Borkmann }; 27262817a336SDaniel Borkmann 27272817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 27282817a336SDaniel Borkmann } 2729b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 27301da177e4SLinus Torvalds 27311da177e4SLinus Torvalds /* Both of above in one bottle. */ 27321da177e4SLinus Torvalds 273381d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 27348d5930dfSAl Viro u8 *to, int len) 27351da177e4SLinus Torvalds { 27361a028e50SDavid S. Miller int start = skb_headlen(skb); 27371a028e50SDavid S. Miller int i, copy = start - offset; 2738fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 27391da177e4SLinus Torvalds int pos = 0; 27408d5930dfSAl Viro __wsum csum = 0; 27411da177e4SLinus Torvalds 27421da177e4SLinus Torvalds /* Copy header. */ 27431da177e4SLinus Torvalds if (copy > 0) { 27441da177e4SLinus Torvalds if (copy > len) 27451da177e4SLinus Torvalds copy = len; 27461da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2747cc44c17bSAl Viro copy); 27481da177e4SLinus Torvalds if ((len -= copy) == 0) 27491da177e4SLinus Torvalds return csum; 27501da177e4SLinus Torvalds offset += copy; 27511da177e4SLinus Torvalds to += copy; 27521da177e4SLinus Torvalds pos = copy; 27531da177e4SLinus Torvalds } 27541da177e4SLinus Torvalds 27551da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 27561a028e50SDavid S. Miller int end; 27571da177e4SLinus Torvalds 2758547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27591a028e50SDavid S. Miller 27609e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 27611da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2762c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2763c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2764c613c209SWillem de Bruijn struct page *p; 27655084205fSAl Viro __wsum csum2; 27661da177e4SLinus Torvalds u8 *vaddr; 27671da177e4SLinus Torvalds 27681da177e4SLinus Torvalds if (copy > len) 27691da177e4SLinus Torvalds copy = len; 2770c613c209SWillem de Bruijn 2771c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2772b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2773c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2774c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2775c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2776c613c209SWillem de Bruijn to + copied, 2777cc44c17bSAl Viro p_len); 277851c56b00SEric Dumazet kunmap_atomic(vaddr); 27791da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2780c613c209SWillem de Bruijn pos += p_len; 2781c613c209SWillem de Bruijn } 2782c613c209SWillem de Bruijn 27831da177e4SLinus Torvalds if (!(len -= copy)) 27841da177e4SLinus Torvalds return csum; 27851da177e4SLinus Torvalds offset += copy; 27861da177e4SLinus Torvalds to += copy; 27871da177e4SLinus Torvalds } 27881a028e50SDavid S. Miller start = end; 27891da177e4SLinus Torvalds } 27901da177e4SLinus Torvalds 2791fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 279281d77662SAl Viro __wsum csum2; 27931a028e50SDavid S. Miller int end; 27941da177e4SLinus Torvalds 2795547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27961a028e50SDavid S. Miller 2797fbb398a8SDavid S. Miller end = start + frag_iter->len; 27981da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 27991da177e4SLinus Torvalds if (copy > len) 28001da177e4SLinus Torvalds copy = len; 2801fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 28021a028e50SDavid S. Miller offset - start, 28038d5930dfSAl Viro to, copy); 28041da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 28051da177e4SLinus Torvalds if ((len -= copy) == 0) 28061da177e4SLinus Torvalds return csum; 28071da177e4SLinus Torvalds offset += copy; 28081da177e4SLinus Torvalds to += copy; 28091da177e4SLinus Torvalds pos += copy; 28101da177e4SLinus Torvalds } 28111a028e50SDavid S. Miller start = end; 28121da177e4SLinus Torvalds } 281309a62660SKris Katterjohn BUG_ON(len); 28141da177e4SLinus Torvalds return csum; 28151da177e4SLinus Torvalds } 2816b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 28171da177e4SLinus Torvalds 281849f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 281949f8e832SCong Wang { 282049f8e832SCong Wang __sum16 sum; 282149f8e832SCong Wang 282249f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 282314641931SCong Wang /* See comments in __skb_checksum_complete(). */ 282449f8e832SCong Wang if (likely(!sum)) { 282549f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 282649f8e832SCong Wang !skb->csum_complete_sw) 28277fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 282849f8e832SCong Wang } 282949f8e832SCong Wang if (!skb_shared(skb)) 283049f8e832SCong Wang skb->csum_valid = !sum; 283149f8e832SCong Wang return sum; 283249f8e832SCong Wang } 283349f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 283449f8e832SCong Wang 283514641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 283614641931SCong Wang * which has been changed from the hardware checksum, for example, by 283714641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 283814641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 283914641931SCong Wang * 284014641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 284114641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 284214641931SCong Wang * shared. 284314641931SCong Wang */ 284449f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 284549f8e832SCong Wang { 284649f8e832SCong Wang __wsum csum; 284749f8e832SCong Wang __sum16 sum; 284849f8e832SCong Wang 284949f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 285049f8e832SCong Wang 285149f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 285214641931SCong Wang /* This check is inverted, because we already knew the hardware 285314641931SCong Wang * checksum is invalid before calling this function. So, if the 285414641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 285514641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 285614641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 285714641931SCong Wang * when moving skb->data around. 285814641931SCong Wang */ 285949f8e832SCong Wang if (likely(!sum)) { 286049f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 286149f8e832SCong Wang !skb->csum_complete_sw) 28627fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 286349f8e832SCong Wang } 286449f8e832SCong Wang 286549f8e832SCong Wang if (!skb_shared(skb)) { 286649f8e832SCong Wang /* Save full packet checksum */ 286749f8e832SCong Wang skb->csum = csum; 286849f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 286949f8e832SCong Wang skb->csum_complete_sw = 1; 287049f8e832SCong Wang skb->csum_valid = !sum; 287149f8e832SCong Wang } 287249f8e832SCong Wang 287349f8e832SCong Wang return sum; 287449f8e832SCong Wang } 287549f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 287649f8e832SCong Wang 28779617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 28789617813dSDavide Caratti { 28799617813dSDavide Caratti net_warn_ratelimited( 28809617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 28819617813dSDavide Caratti __func__); 28829617813dSDavide Caratti return 0; 28839617813dSDavide Caratti } 28849617813dSDavide Caratti 28859617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 28869617813dSDavide Caratti int offset, int len) 28879617813dSDavide Caratti { 28889617813dSDavide Caratti net_warn_ratelimited( 28899617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 28909617813dSDavide Caratti __func__); 28919617813dSDavide Caratti return 0; 28929617813dSDavide Caratti } 28939617813dSDavide Caratti 28949617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 28959617813dSDavide Caratti .update = warn_crc32c_csum_update, 28969617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 28979617813dSDavide Caratti }; 28989617813dSDavide Caratti 28999617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 29009617813dSDavide Caratti &default_crc32c_ops; 29019617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 29029617813dSDavide Caratti 2903af2806f8SThomas Graf /** 2904af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2905af2806f8SThomas Graf * @from: source buffer 2906af2806f8SThomas Graf * 2907af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2908af2806f8SThomas Graf * into skb_zerocopy(). 2909af2806f8SThomas Graf */ 2910af2806f8SThomas Graf unsigned int 2911af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2912af2806f8SThomas Graf { 2913af2806f8SThomas Graf unsigned int hlen = 0; 2914af2806f8SThomas Graf 2915af2806f8SThomas Graf if (!from->head_frag || 2916af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2917af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2918af2806f8SThomas Graf hlen = skb_headlen(from); 2919af2806f8SThomas Graf 2920af2806f8SThomas Graf if (skb_has_frag_list(from)) 2921af2806f8SThomas Graf hlen = from->len; 2922af2806f8SThomas Graf 2923af2806f8SThomas Graf return hlen; 2924af2806f8SThomas Graf } 2925af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2926af2806f8SThomas Graf 2927af2806f8SThomas Graf /** 2928af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2929af2806f8SThomas Graf * @to: destination buffer 29307fceb4deSMasanari Iida * @from: source buffer 2931af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2932af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2933af2806f8SThomas Graf * 2934af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2935af2806f8SThomas Graf * to the frags in the source buffer. 2936af2806f8SThomas Graf * 2937af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2938af2806f8SThomas Graf * headroom in the `to` buffer. 293936d5fe6aSZoltan Kiss * 294036d5fe6aSZoltan Kiss * Return value: 294136d5fe6aSZoltan Kiss * 0: everything is OK 294236d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 294336d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2944af2806f8SThomas Graf */ 294536d5fe6aSZoltan Kiss int 294636d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2947af2806f8SThomas Graf { 2948af2806f8SThomas Graf int i, j = 0; 2949af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 295036d5fe6aSZoltan Kiss int ret; 2951af2806f8SThomas Graf struct page *page; 2952af2806f8SThomas Graf unsigned int offset; 2953af2806f8SThomas Graf 2954af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2955af2806f8SThomas Graf 2956af2806f8SThomas Graf /* dont bother with small payloads */ 295736d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 295836d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2959af2806f8SThomas Graf 2960af2806f8SThomas Graf if (hlen) { 296136d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 296236d5fe6aSZoltan Kiss if (unlikely(ret)) 296336d5fe6aSZoltan Kiss return ret; 2964af2806f8SThomas Graf len -= hlen; 2965af2806f8SThomas Graf } else { 2966af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2967af2806f8SThomas Graf if (plen) { 2968af2806f8SThomas Graf page = virt_to_head_page(from->head); 2969af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2970af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2971af2806f8SThomas Graf get_page(page); 2972af2806f8SThomas Graf j = 1; 2973af2806f8SThomas Graf len -= plen; 2974af2806f8SThomas Graf } 2975af2806f8SThomas Graf } 2976af2806f8SThomas Graf 2977af2806f8SThomas Graf to->truesize += len + plen; 2978af2806f8SThomas Graf to->len += len + plen; 2979af2806f8SThomas Graf to->data_len += len + plen; 2980af2806f8SThomas Graf 298136d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 298236d5fe6aSZoltan Kiss skb_tx_error(from); 298336d5fe6aSZoltan Kiss return -ENOMEM; 298436d5fe6aSZoltan Kiss } 29851f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 298636d5fe6aSZoltan Kiss 2987af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2988d8e18a51SMatthew Wilcox (Oracle) int size; 2989d8e18a51SMatthew Wilcox (Oracle) 2990af2806f8SThomas Graf if (!len) 2991af2806f8SThomas Graf break; 2992af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2993d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 2994d8e18a51SMatthew Wilcox (Oracle) len); 2995d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 2996d8e18a51SMatthew Wilcox (Oracle) len -= size; 2997af2806f8SThomas Graf skb_frag_ref(to, j); 2998af2806f8SThomas Graf j++; 2999af2806f8SThomas Graf } 3000af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 300136d5fe6aSZoltan Kiss 300236d5fe6aSZoltan Kiss return 0; 3003af2806f8SThomas Graf } 3004af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3005af2806f8SThomas Graf 30061da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 30071da177e4SLinus Torvalds { 3008d3bc23e7SAl Viro __wsum csum; 30091da177e4SLinus Torvalds long csstart; 30101da177e4SLinus Torvalds 301184fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 301255508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 30131da177e4SLinus Torvalds else 30141da177e4SLinus Torvalds csstart = skb_headlen(skb); 30151da177e4SLinus Torvalds 301609a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 30171da177e4SLinus Torvalds 3018d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 30191da177e4SLinus Torvalds 30201da177e4SLinus Torvalds csum = 0; 30211da177e4SLinus Torvalds if (csstart != skb->len) 30221da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 30238d5930dfSAl Viro skb->len - csstart); 30241da177e4SLinus Torvalds 302584fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3026ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 30271da177e4SLinus Torvalds 3028d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 30291da177e4SLinus Torvalds } 30301da177e4SLinus Torvalds } 3031b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 30321da177e4SLinus Torvalds 30331da177e4SLinus Torvalds /** 30341da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 30351da177e4SLinus Torvalds * @list: list to dequeue from 30361da177e4SLinus Torvalds * 30371da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 30381da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 30391da177e4SLinus Torvalds * returned or %NULL if the list is empty. 30401da177e4SLinus Torvalds */ 30411da177e4SLinus Torvalds 30421da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 30431da177e4SLinus Torvalds { 30441da177e4SLinus Torvalds unsigned long flags; 30451da177e4SLinus Torvalds struct sk_buff *result; 30461da177e4SLinus Torvalds 30471da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30481da177e4SLinus Torvalds result = __skb_dequeue(list); 30491da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30501da177e4SLinus Torvalds return result; 30511da177e4SLinus Torvalds } 3052b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 30531da177e4SLinus Torvalds 30541da177e4SLinus Torvalds /** 30551da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 30561da177e4SLinus Torvalds * @list: list to dequeue from 30571da177e4SLinus Torvalds * 30581da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 30591da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 30601da177e4SLinus Torvalds * returned or %NULL if the list is empty. 30611da177e4SLinus Torvalds */ 30621da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 30631da177e4SLinus Torvalds { 30641da177e4SLinus Torvalds unsigned long flags; 30651da177e4SLinus Torvalds struct sk_buff *result; 30661da177e4SLinus Torvalds 30671da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30681da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 30691da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30701da177e4SLinus Torvalds return result; 30711da177e4SLinus Torvalds } 3072b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 30731da177e4SLinus Torvalds 30741da177e4SLinus Torvalds /** 30751da177e4SLinus Torvalds * skb_queue_purge - empty a list 30761da177e4SLinus Torvalds * @list: list to empty 30771da177e4SLinus Torvalds * 30781da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 30791da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 30801da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 30811da177e4SLinus Torvalds */ 30821da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 30831da177e4SLinus Torvalds { 30841da177e4SLinus Torvalds struct sk_buff *skb; 30851da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 30861da177e4SLinus Torvalds kfree_skb(skb); 30871da177e4SLinus Torvalds } 3088b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 30891da177e4SLinus Torvalds 30901da177e4SLinus Torvalds /** 30919f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 30929f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3093385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 30949f5afeaeSYaogong Wang * 30959f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 30969f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 30979f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 30989f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 30999f5afeaeSYaogong Wang */ 3100385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 31019f5afeaeSYaogong Wang { 31027c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3103385114deSPeter Oskolkov unsigned int sum = 0; 31049f5afeaeSYaogong Wang 31057c90584cSEric Dumazet while (p) { 31067c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 31077c90584cSEric Dumazet 31087c90584cSEric Dumazet p = rb_next(p); 31097c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3110385114deSPeter Oskolkov sum += skb->truesize; 31119f5afeaeSYaogong Wang kfree_skb(skb); 31127c90584cSEric Dumazet } 3113385114deSPeter Oskolkov return sum; 31149f5afeaeSYaogong Wang } 31159f5afeaeSYaogong Wang 31169f5afeaeSYaogong Wang /** 31171da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 31181da177e4SLinus Torvalds * @list: list to use 31191da177e4SLinus Torvalds * @newsk: buffer to queue 31201da177e4SLinus Torvalds * 31211da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 31221da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 31231da177e4SLinus Torvalds * safely. 31241da177e4SLinus Torvalds * 31251da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31261da177e4SLinus Torvalds */ 31271da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 31281da177e4SLinus Torvalds { 31291da177e4SLinus Torvalds unsigned long flags; 31301da177e4SLinus Torvalds 31311da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31321da177e4SLinus Torvalds __skb_queue_head(list, newsk); 31331da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31341da177e4SLinus Torvalds } 3135b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 31361da177e4SLinus Torvalds 31371da177e4SLinus Torvalds /** 31381da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 31391da177e4SLinus Torvalds * @list: list to use 31401da177e4SLinus Torvalds * @newsk: buffer to queue 31411da177e4SLinus Torvalds * 31421da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 31431da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 31441da177e4SLinus Torvalds * safely. 31451da177e4SLinus Torvalds * 31461da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31471da177e4SLinus Torvalds */ 31481da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 31491da177e4SLinus Torvalds { 31501da177e4SLinus Torvalds unsigned long flags; 31511da177e4SLinus Torvalds 31521da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31531da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 31541da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31551da177e4SLinus Torvalds } 3156b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 31578728b834SDavid S. Miller 31581da177e4SLinus Torvalds /** 31591da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 31601da177e4SLinus Torvalds * @skb: buffer to remove 31618728b834SDavid S. Miller * @list: list to use 31621da177e4SLinus Torvalds * 31638728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 31648728b834SDavid S. Miller * function is atomic with respect to other list locked calls 31651da177e4SLinus Torvalds * 31668728b834SDavid S. Miller * You must know what list the SKB is on. 31671da177e4SLinus Torvalds */ 31688728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 31691da177e4SLinus Torvalds { 31701da177e4SLinus Torvalds unsigned long flags; 31711da177e4SLinus Torvalds 31721da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31738728b834SDavid S. Miller __skb_unlink(skb, list); 31741da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31751da177e4SLinus Torvalds } 3176b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 31771da177e4SLinus Torvalds 31781da177e4SLinus Torvalds /** 31791da177e4SLinus Torvalds * skb_append - append a buffer 31801da177e4SLinus Torvalds * @old: buffer to insert after 31811da177e4SLinus Torvalds * @newsk: buffer to insert 31828728b834SDavid S. Miller * @list: list to use 31831da177e4SLinus Torvalds * 31841da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 31851da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 31861da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31871da177e4SLinus Torvalds */ 31888728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 31891da177e4SLinus Torvalds { 31901da177e4SLinus Torvalds unsigned long flags; 31911da177e4SLinus Torvalds 31928728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 31937de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 31948728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 31951da177e4SLinus Torvalds } 3196b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 31971da177e4SLinus Torvalds 31981da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 31991da177e4SLinus Torvalds struct sk_buff* skb1, 32001da177e4SLinus Torvalds const u32 len, const int pos) 32011da177e4SLinus Torvalds { 32021da177e4SLinus Torvalds int i; 32031da177e4SLinus Torvalds 3204d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3205d626f62bSArnaldo Carvalho de Melo pos - len); 32061da177e4SLinus Torvalds /* And move data appendix as is. */ 32071da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 32081da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 32091da177e4SLinus Torvalds 32101da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 32111da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 32121da177e4SLinus Torvalds skb1->data_len = skb->data_len; 32131da177e4SLinus Torvalds skb1->len += skb1->data_len; 32141da177e4SLinus Torvalds skb->data_len = 0; 32151da177e4SLinus Torvalds skb->len = len; 321627a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 32171da177e4SLinus Torvalds } 32181da177e4SLinus Torvalds 32191da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 32201da177e4SLinus Torvalds struct sk_buff* skb1, 32211da177e4SLinus Torvalds const u32 len, int pos) 32221da177e4SLinus Torvalds { 32231da177e4SLinus Torvalds int i, k = 0; 32241da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 32251da177e4SLinus Torvalds 32261da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 32271da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 32281da177e4SLinus Torvalds skb->len = len; 32291da177e4SLinus Torvalds skb->data_len = len - pos; 32301da177e4SLinus Torvalds 32311da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 32329e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 32331da177e4SLinus Torvalds 32341da177e4SLinus Torvalds if (pos + size > len) { 32351da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 32361da177e4SLinus Torvalds 32371da177e4SLinus Torvalds if (pos < len) { 32381da177e4SLinus Torvalds /* Split frag. 32391da177e4SLinus Torvalds * We have two variants in this case: 32401da177e4SLinus Torvalds * 1. Move all the frag to the second 32411da177e4SLinus Torvalds * part, if it is possible. F.e. 32421da177e4SLinus Torvalds * this approach is mandatory for TUX, 32431da177e4SLinus Torvalds * where splitting is expensive. 32441da177e4SLinus Torvalds * 2. Split is accurately. We make this. 32451da177e4SLinus Torvalds */ 3246ea2ab693SIan Campbell skb_frag_ref(skb, i); 3247b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 32489e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 32499e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 32501da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 32511da177e4SLinus Torvalds } 32521da177e4SLinus Torvalds k++; 32531da177e4SLinus Torvalds } else 32541da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 32551da177e4SLinus Torvalds pos += size; 32561da177e4SLinus Torvalds } 32571da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 32581da177e4SLinus Torvalds } 32591da177e4SLinus Torvalds 32601da177e4SLinus Torvalds /** 32611da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 32621da177e4SLinus Torvalds * @skb: the buffer to split 32631da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 32641da177e4SLinus Torvalds * @len: new length for skb 32651da177e4SLinus Torvalds */ 32661da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 32671da177e4SLinus Torvalds { 32681da177e4SLinus Torvalds int pos = skb_headlen(skb); 32691da177e4SLinus Torvalds 3270fff88030SWillem de Bruijn skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags & 3271fff88030SWillem de Bruijn SKBTX_SHARED_FRAG; 32721f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 32731da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 32741da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 32751da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 32761da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 32771da177e4SLinus Torvalds } 3278b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 32791da177e4SLinus Torvalds 32809f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 32819f782db3SIlpo Järvinen * 32829f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 32839f782db3SIlpo Järvinen */ 3284832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3285832d11c5SIlpo Järvinen { 32860ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3287832d11c5SIlpo Järvinen } 3288832d11c5SIlpo Järvinen 3289832d11c5SIlpo Järvinen /** 3290832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3291832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3292832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3293832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3294832d11c5SIlpo Järvinen * 3295832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 329620e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3297832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3298832d11c5SIlpo Järvinen * 3299832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3300832d11c5SIlpo Järvinen * 3301832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3302832d11c5SIlpo Järvinen * to have non-paged data as well. 3303832d11c5SIlpo Järvinen * 3304832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3305832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3306832d11c5SIlpo Järvinen */ 3307832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3308832d11c5SIlpo Järvinen { 3309832d11c5SIlpo Järvinen int from, to, merge, todo; 3310d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3311832d11c5SIlpo Järvinen 3312832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3313f8071cdeSEric Dumazet 3314f8071cdeSEric Dumazet if (skb_headlen(skb)) 3315f8071cdeSEric Dumazet return 0; 33161f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 33171f8b977aSWillem de Bruijn return 0; 3318832d11c5SIlpo Järvinen 3319832d11c5SIlpo Järvinen todo = shiftlen; 3320832d11c5SIlpo Järvinen from = 0; 3321832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3322832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3323832d11c5SIlpo Järvinen 3324832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3325832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3326832d11c5SIlpo Järvinen */ 3327832d11c5SIlpo Järvinen if (!to || 3328ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3329b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3330832d11c5SIlpo Järvinen merge = -1; 3331832d11c5SIlpo Järvinen } else { 3332832d11c5SIlpo Järvinen merge = to - 1; 3333832d11c5SIlpo Järvinen 33349e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3335832d11c5SIlpo Järvinen if (todo < 0) { 3336832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3337832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3338832d11c5SIlpo Järvinen return 0; 3339832d11c5SIlpo Järvinen 33409f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 33419f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3342832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3343832d11c5SIlpo Järvinen 33449e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 33459e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3346b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3347832d11c5SIlpo Järvinen 3348832d11c5SIlpo Järvinen goto onlymerged; 3349832d11c5SIlpo Järvinen } 3350832d11c5SIlpo Järvinen 3351832d11c5SIlpo Järvinen from++; 3352832d11c5SIlpo Järvinen } 3353832d11c5SIlpo Järvinen 3354832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3355832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3356832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3357832d11c5SIlpo Järvinen return 0; 3358832d11c5SIlpo Järvinen 3359832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3360832d11c5SIlpo Järvinen return 0; 3361832d11c5SIlpo Järvinen 3362832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3363832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3364832d11c5SIlpo Järvinen return 0; 3365832d11c5SIlpo Järvinen 3366832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3367832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3368832d11c5SIlpo Järvinen 33699e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3370832d11c5SIlpo Järvinen *fragto = *fragfrom; 33719e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3372832d11c5SIlpo Järvinen from++; 3373832d11c5SIlpo Järvinen to++; 3374832d11c5SIlpo Järvinen 3375832d11c5SIlpo Järvinen } else { 3376ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3377b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3378b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 33799e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3380832d11c5SIlpo Järvinen 3381b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 33829e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3383832d11c5SIlpo Järvinen todo = 0; 3384832d11c5SIlpo Järvinen 3385832d11c5SIlpo Järvinen to++; 3386832d11c5SIlpo Järvinen break; 3387832d11c5SIlpo Järvinen } 3388832d11c5SIlpo Järvinen } 3389832d11c5SIlpo Järvinen 3390832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3391832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3392832d11c5SIlpo Järvinen 3393832d11c5SIlpo Järvinen if (merge >= 0) { 3394832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3395832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3396832d11c5SIlpo Järvinen 33979e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 3398ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 3399832d11c5SIlpo Järvinen } 3400832d11c5SIlpo Järvinen 3401832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3402832d11c5SIlpo Järvinen to = 0; 3403832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3404832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3405832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3406832d11c5SIlpo Järvinen 3407832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3408832d11c5SIlpo Järvinen 3409832d11c5SIlpo Järvinen onlymerged: 3410832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3411832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3412832d11c5SIlpo Järvinen */ 3413832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3414832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3415832d11c5SIlpo Järvinen 3416832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3417832d11c5SIlpo Järvinen skb->len -= shiftlen; 3418832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3419832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3420832d11c5SIlpo Järvinen tgt->len += shiftlen; 3421832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3422832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3423832d11c5SIlpo Järvinen 3424832d11c5SIlpo Järvinen return shiftlen; 3425832d11c5SIlpo Järvinen } 3426832d11c5SIlpo Järvinen 3427677e90edSThomas Graf /** 3428677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3429677e90edSThomas Graf * @skb: the buffer to read 3430677e90edSThomas Graf * @from: lower offset of data to be read 3431677e90edSThomas Graf * @to: upper offset of data to be read 3432677e90edSThomas Graf * @st: state variable 3433677e90edSThomas Graf * 3434677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3435677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3436677e90edSThomas Graf */ 3437677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3438677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3439677e90edSThomas Graf { 3440677e90edSThomas Graf st->lower_offset = from; 3441677e90edSThomas Graf st->upper_offset = to; 3442677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3443677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3444677e90edSThomas Graf st->frag_data = NULL; 3445*97550f6fSWillem de Bruijn st->frag_off = 0; 3446677e90edSThomas Graf } 3447b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3448677e90edSThomas Graf 3449677e90edSThomas Graf /** 3450677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3451677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3452677e90edSThomas Graf * @data: destination pointer for data to be returned 3453677e90edSThomas Graf * @st: state variable 3454677e90edSThomas Graf * 3455bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3456677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3457bc32383cSMathias Krause * the head of the data block to @data and returns the length 3458677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3459677e90edSThomas Graf * offset has been reached. 3460677e90edSThomas Graf * 3461677e90edSThomas Graf * The caller is not required to consume all of the data 3462bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3463677e90edSThomas Graf * of bytes already consumed and the next call to 3464677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3465677e90edSThomas Graf * 346625985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3467e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3468677e90edSThomas Graf * reads of potentially non linear data. 3469677e90edSThomas Graf * 3470bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3471677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3472677e90edSThomas Graf * a stack for this purpose. 3473677e90edSThomas Graf */ 3474677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3475677e90edSThomas Graf struct skb_seq_state *st) 3476677e90edSThomas Graf { 3477677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3478677e90edSThomas Graf skb_frag_t *frag; 3479677e90edSThomas Graf 3480aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3481aeb193eaSWedson Almeida Filho if (st->frag_data) { 3482aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3483aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3484aeb193eaSWedson Almeida Filho } 3485677e90edSThomas Graf return 0; 3486aeb193eaSWedson Almeida Filho } 3487677e90edSThomas Graf 3488677e90edSThomas Graf next_skb: 348995e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3490677e90edSThomas Graf 3491995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 349295e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3493677e90edSThomas Graf return block_limit - abs_offset; 3494677e90edSThomas Graf } 3495677e90edSThomas Graf 3496677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3497677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3498677e90edSThomas Graf 3499677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 3500*97550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3501677e90edSThomas Graf 3502*97550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 3503*97550f6fSWillem de Bruijn 3504*97550f6fSWillem de Bruijn pg_idx = 0; 3505*97550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 3506*97550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 3507*97550f6fSWillem de Bruijn 3508*97550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 3509*97550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 3510*97550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 3511*97550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 3512*97550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 3513*97550f6fSWillem de Bruijn } 3514*97550f6fSWillem de Bruijn 3515*97550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3516677e90edSThomas Graf if (abs_offset < block_limit) { 3517677e90edSThomas Graf if (!st->frag_data) 3518*97550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3519677e90edSThomas Graf 3520*97550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3521677e90edSThomas Graf (abs_offset - st->stepped_offset); 3522677e90edSThomas Graf 3523677e90edSThomas Graf return block_limit - abs_offset; 3524677e90edSThomas Graf } 3525677e90edSThomas Graf 3526677e90edSThomas Graf if (st->frag_data) { 352751c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3528677e90edSThomas Graf st->frag_data = NULL; 3529677e90edSThomas Graf } 3530677e90edSThomas Graf 3531*97550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 3532*97550f6fSWillem de Bruijn st->frag_off += pg_sz; 3533*97550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 3534*97550f6fSWillem de Bruijn st->frag_off = 0; 3535677e90edSThomas Graf st->frag_idx++; 3536*97550f6fSWillem de Bruijn } 3537677e90edSThomas Graf } 3538677e90edSThomas Graf 35395b5a60daSOlaf Kirch if (st->frag_data) { 354051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 35415b5a60daSOlaf Kirch st->frag_data = NULL; 35425b5a60daSOlaf Kirch } 35435b5a60daSOlaf Kirch 354421dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3545677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 354695e3b24cSHerbert Xu st->frag_idx = 0; 3547677e90edSThomas Graf goto next_skb; 354871b3346dSShyam Iyer } else if (st->cur_skb->next) { 354971b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 355071b3346dSShyam Iyer st->frag_idx = 0; 3551677e90edSThomas Graf goto next_skb; 3552677e90edSThomas Graf } 3553677e90edSThomas Graf 3554677e90edSThomas Graf return 0; 3555677e90edSThomas Graf } 3556b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3557677e90edSThomas Graf 3558677e90edSThomas Graf /** 3559677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3560677e90edSThomas Graf * @st: state variable 3561677e90edSThomas Graf * 3562677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3563677e90edSThomas Graf * returned 0. 3564677e90edSThomas Graf */ 3565677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3566677e90edSThomas Graf { 3567677e90edSThomas Graf if (st->frag_data) 356851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3569677e90edSThomas Graf } 3570b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3571677e90edSThomas Graf 35723fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 35733fc7e8a6SThomas Graf 35743fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 35753fc7e8a6SThomas Graf struct ts_config *conf, 35763fc7e8a6SThomas Graf struct ts_state *state) 35773fc7e8a6SThomas Graf { 35783fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 35793fc7e8a6SThomas Graf } 35803fc7e8a6SThomas Graf 35813fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 35823fc7e8a6SThomas Graf { 35833fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 35843fc7e8a6SThomas Graf } 35853fc7e8a6SThomas Graf 35863fc7e8a6SThomas Graf /** 35873fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 35883fc7e8a6SThomas Graf * @skb: the buffer to look in 35893fc7e8a6SThomas Graf * @from: search offset 35903fc7e8a6SThomas Graf * @to: search limit 35913fc7e8a6SThomas Graf * @config: textsearch configuration 35923fc7e8a6SThomas Graf * 35933fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 35943fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 35953fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 35963fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 35973fc7e8a6SThomas Graf */ 35983fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3599059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 36003fc7e8a6SThomas Graf { 3601059a2440SBojan Prtvar struct ts_state state; 3602f72b948dSPhil Oester unsigned int ret; 3603f72b948dSPhil Oester 36043fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 36053fc7e8a6SThomas Graf config->finish = skb_ts_finish; 36063fc7e8a6SThomas Graf 3607059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 36083fc7e8a6SThomas Graf 3609059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3610f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 36113fc7e8a6SThomas Graf } 3612b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 36133fc7e8a6SThomas Graf 3614be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3615be12a1feSHannes Frederic Sowa int offset, size_t size) 3616be12a1feSHannes Frederic Sowa { 3617be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3618be12a1feSHannes Frederic Sowa 3619be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3620be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3621be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3622be12a1feSHannes Frederic Sowa get_page(page); 3623be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3624be12a1feSHannes Frederic Sowa } else { 3625be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3626be12a1feSHannes Frederic Sowa } 3627be12a1feSHannes Frederic Sowa 3628be12a1feSHannes Frederic Sowa return 0; 3629be12a1feSHannes Frederic Sowa } 3630be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3631be12a1feSHannes Frederic Sowa 3632cbb042f9SHerbert Xu /** 3633cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3634cbb042f9SHerbert Xu * @skb: buffer to update 3635cbb042f9SHerbert Xu * @len: length of data pulled 3636cbb042f9SHerbert Xu * 3637cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3638fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 363984fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 364084fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 364184fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3642cbb042f9SHerbert Xu */ 3643af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3644cbb042f9SHerbert Xu { 364531b33dfbSPravin B Shelar unsigned char *data = skb->data; 364631b33dfbSPravin B Shelar 3647cbb042f9SHerbert Xu BUG_ON(len > skb->len); 364831b33dfbSPravin B Shelar __skb_pull(skb, len); 364931b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 365031b33dfbSPravin B Shelar return skb->data; 3651cbb042f9SHerbert Xu } 3652f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3653f94691acSArnaldo Carvalho de Melo 365413acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 365513acc94eSYonghong Song { 365613acc94eSYonghong Song skb_frag_t head_frag; 365713acc94eSYonghong Song struct page *page; 365813acc94eSYonghong Song 365913acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3660d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3661b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3662b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3663d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 366413acc94eSYonghong Song return head_frag; 366513acc94eSYonghong Song } 366613acc94eSYonghong Song 36673a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 36683a1296a3SSteffen Klassert netdev_features_t features, 36693a1296a3SSteffen Klassert unsigned int offset) 36703a1296a3SSteffen Klassert { 36713a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 36723a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 36733a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 36743a1296a3SSteffen Klassert unsigned int delta_len = 0; 36753a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 367653475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 367753475c5dSDongseok Yi int err; 36783a1296a3SSteffen Klassert 36793a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 36803a1296a3SSteffen Klassert 36813a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 36823a1296a3SSteffen Klassert 36833a1296a3SSteffen Klassert do { 36843a1296a3SSteffen Klassert nskb = list_skb; 36853a1296a3SSteffen Klassert list_skb = list_skb->next; 36863a1296a3SSteffen Klassert 368753475c5dSDongseok Yi err = 0; 368853475c5dSDongseok Yi if (skb_shared(nskb)) { 368953475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 369053475c5dSDongseok Yi if (tmp) { 369153475c5dSDongseok Yi consume_skb(nskb); 369253475c5dSDongseok Yi nskb = tmp; 369353475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 369453475c5dSDongseok Yi } else { 369553475c5dSDongseok Yi err = -ENOMEM; 369653475c5dSDongseok Yi } 369753475c5dSDongseok Yi } 369853475c5dSDongseok Yi 36993a1296a3SSteffen Klassert if (!tail) 37003a1296a3SSteffen Klassert skb->next = nskb; 37013a1296a3SSteffen Klassert else 37023a1296a3SSteffen Klassert tail->next = nskb; 37033a1296a3SSteffen Klassert 370453475c5dSDongseok Yi if (unlikely(err)) { 370553475c5dSDongseok Yi nskb->next = list_skb; 370653475c5dSDongseok Yi goto err_linearize; 370753475c5dSDongseok Yi } 370853475c5dSDongseok Yi 37093a1296a3SSteffen Klassert tail = nskb; 37103a1296a3SSteffen Klassert 37113a1296a3SSteffen Klassert delta_len += nskb->len; 37123a1296a3SSteffen Klassert delta_truesize += nskb->truesize; 37133a1296a3SSteffen Klassert 37143a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 37153a1296a3SSteffen Klassert 3716cf673ed0SFlorian Westphal skb_release_head_state(nskb); 37173a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 37183a1296a3SSteffen Klassert 37193a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 37203a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 37213a1296a3SSteffen Klassert nskb->data - tnl_hlen, 37223a1296a3SSteffen Klassert offset + tnl_hlen); 37233a1296a3SSteffen Klassert 37243a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 37253a1296a3SSteffen Klassert __skb_linearize(nskb)) 37263a1296a3SSteffen Klassert goto err_linearize; 37273a1296a3SSteffen Klassert 37283a1296a3SSteffen Klassert } while (list_skb); 37293a1296a3SSteffen Klassert 37303a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 37313a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 37323a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 37333a1296a3SSteffen Klassert 37343a1296a3SSteffen Klassert skb_gso_reset(skb); 37353a1296a3SSteffen Klassert 37363a1296a3SSteffen Klassert skb->prev = tail; 37373a1296a3SSteffen Klassert 37383a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 37393a1296a3SSteffen Klassert __skb_linearize(skb)) 37403a1296a3SSteffen Klassert goto err_linearize; 37413a1296a3SSteffen Klassert 37423a1296a3SSteffen Klassert skb_get(skb); 37433a1296a3SSteffen Klassert 37443a1296a3SSteffen Klassert return skb; 37453a1296a3SSteffen Klassert 37463a1296a3SSteffen Klassert err_linearize: 37473a1296a3SSteffen Klassert kfree_skb_list(skb->next); 37483a1296a3SSteffen Klassert skb->next = NULL; 37493a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 37503a1296a3SSteffen Klassert } 37513a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 37523a1296a3SSteffen Klassert 37533a1296a3SSteffen Klassert int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb) 37543a1296a3SSteffen Klassert { 37553a1296a3SSteffen Klassert if (unlikely(p->len + skb->len >= 65536)) 37563a1296a3SSteffen Klassert return -E2BIG; 37573a1296a3SSteffen Klassert 37583a1296a3SSteffen Klassert if (NAPI_GRO_CB(p)->last == p) 37593a1296a3SSteffen Klassert skb_shinfo(p)->frag_list = skb; 37603a1296a3SSteffen Klassert else 37613a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last->next = skb; 37623a1296a3SSteffen Klassert 37633a1296a3SSteffen Klassert skb_pull(skb, skb_gro_offset(skb)); 37643a1296a3SSteffen Klassert 37653a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last = skb; 37663a1296a3SSteffen Klassert NAPI_GRO_CB(p)->count++; 37673a1296a3SSteffen Klassert p->data_len += skb->len; 37683a1296a3SSteffen Klassert p->truesize += skb->truesize; 37693a1296a3SSteffen Klassert p->len += skb->len; 37703a1296a3SSteffen Klassert 37713a1296a3SSteffen Klassert NAPI_GRO_CB(skb)->same_flow = 1; 37723a1296a3SSteffen Klassert 37733a1296a3SSteffen Klassert return 0; 37743a1296a3SSteffen Klassert } 37753a1296a3SSteffen Klassert 3776f4c50d99SHerbert Xu /** 3777f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3778df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3779576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3780f4c50d99SHerbert Xu * 3781f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 37824c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 37834c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3784f4c50d99SHerbert Xu */ 3785df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3786df5771ffSMichael S. Tsirkin netdev_features_t features) 3787f4c50d99SHerbert Xu { 3788f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3789f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 37901a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3791df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3792df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3793df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 37941fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3795f4c50d99SHerbert Xu unsigned int offset = doffset; 3796df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3797802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3798f4c50d99SHerbert Xu unsigned int headroom; 3799802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3800ec5f0615SPravin B Shelar __be16 proto; 380136c98382SAlexander Duyck bool csum, sg; 3802df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3803f4c50d99SHerbert Xu int err = -ENOMEM; 3804f4c50d99SHerbert Xu int i = 0; 3805f4c50d99SHerbert Xu int pos; 3806f4c50d99SHerbert Xu 38073dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 38083dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 38093dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 38103dcbdb13SShmulik Ladkani * non head_frag head. 38113dcbdb13SShmulik Ladkani * 38123dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 38133dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 38143dcbdb13SShmulik Ladkani * head, then the first one has too). 38153dcbdb13SShmulik Ladkani * 38163dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 38173dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 38183dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 38193dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 38203dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 38213dcbdb13SShmulik Ladkani */ 38223dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 38233dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 38243dcbdb13SShmulik Ladkani } 38253dcbdb13SShmulik Ladkani 38265882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 38272f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 3828ec5f0615SPravin B Shelar if (unlikely(!proto)) 3829ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3830ec5f0615SPravin B Shelar 383136c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3832f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 38337e2b10c1STom Herbert 383407b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 383507b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 383607b26c94SSteffen Klassert struct sk_buff *iter; 383743170c4eSIlan Tayari unsigned int frag_len; 383807b26c94SSteffen Klassert 383907b26c94SSteffen Klassert if (!list_skb || 384007b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 384107b26c94SSteffen Klassert goto normal; 384207b26c94SSteffen Klassert 384343170c4eSIlan Tayari /* If we get here then all the required 384443170c4eSIlan Tayari * GSO features except frag_list are supported. 384543170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 384643170c4eSIlan Tayari * with no frag_list. 384743170c4eSIlan Tayari * Currently we can do that only when the buffers don't 384843170c4eSIlan Tayari * have a linear part and all the buffers except 384943170c4eSIlan Tayari * the last are of the same length. 385007b26c94SSteffen Klassert */ 385143170c4eSIlan Tayari frag_len = list_skb->len; 385207b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 385343170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 385443170c4eSIlan Tayari goto normal; 3855eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 385607b26c94SSteffen Klassert goto normal; 385707b26c94SSteffen Klassert 385807b26c94SSteffen Klassert len -= iter->len; 385907b26c94SSteffen Klassert } 386043170c4eSIlan Tayari 386143170c4eSIlan Tayari if (len != frag_len) 386243170c4eSIlan Tayari goto normal; 386307b26c94SSteffen Klassert } 386407b26c94SSteffen Klassert 3865802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 3866802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 3867802ab55aSAlexander Duyck * now. 3868802ab55aSAlexander Duyck */ 3869802ab55aSAlexander Duyck partial_segs = len / mss; 3870d7fb5a80SAlexander Duyck if (partial_segs > 1) 3871802ab55aSAlexander Duyck mss *= partial_segs; 3872d7fb5a80SAlexander Duyck else 3873d7fb5a80SAlexander Duyck partial_segs = 0; 3874802ab55aSAlexander Duyck } 3875802ab55aSAlexander Duyck 387607b26c94SSteffen Klassert normal: 3877df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3878df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3879f4c50d99SHerbert Xu 3880f4c50d99SHerbert Xu do { 3881f4c50d99SHerbert Xu struct sk_buff *nskb; 38828cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3883c8884eddSHerbert Xu int hsize; 3884f4c50d99SHerbert Xu int size; 3885f4c50d99SHerbert Xu 38863953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 38873953c46cSMarcelo Ricardo Leitner len = list_skb->len; 38883953c46cSMarcelo Ricardo Leitner } else { 3889df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3890f4c50d99SHerbert Xu if (len > mss) 3891f4c50d99SHerbert Xu len = mss; 38923953c46cSMarcelo Ricardo Leitner } 3893f4c50d99SHerbert Xu 3894df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3895f4c50d99SHerbert Xu if (hsize < 0) 3896f4c50d99SHerbert Xu hsize = 0; 3897c8884eddSHerbert Xu if (hsize > len || !sg) 3898c8884eddSHerbert Xu hsize = len; 3899f4c50d99SHerbert Xu 39001a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 39011a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 39021a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 390389319d38SHerbert Xu 39049d8506ccSHerbert Xu i = 0; 39051a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 39061a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 39071fd819ecSMichael S. Tsirkin frag_skb = list_skb; 39081a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 39099d8506ccSHerbert Xu 39109d8506ccSHerbert Xu while (pos < offset + len) { 39119d8506ccSHerbert Xu BUG_ON(i >= nfrags); 39129d8506ccSHerbert Xu 39134e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 39149d8506ccSHerbert Xu if (pos + size > offset + len) 39159d8506ccSHerbert Xu break; 39169d8506ccSHerbert Xu 39179d8506ccSHerbert Xu i++; 39189d8506ccSHerbert Xu pos += size; 39194e1beba1SMichael S. Tsirkin frag++; 39209d8506ccSHerbert Xu } 39219d8506ccSHerbert Xu 39221a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 39231a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 392489319d38SHerbert Xu 3925f4c50d99SHerbert Xu if (unlikely(!nskb)) 3926f4c50d99SHerbert Xu goto err; 3927f4c50d99SHerbert Xu 39289d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 39299d8506ccSHerbert Xu kfree_skb(nskb); 39309d8506ccSHerbert Xu goto err; 39319d8506ccSHerbert Xu } 39329d8506ccSHerbert Xu 3933ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 393489319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 393589319d38SHerbert Xu kfree_skb(nskb); 393689319d38SHerbert Xu goto err; 393789319d38SHerbert Xu } 393889319d38SHerbert Xu 3939ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 394089319d38SHerbert Xu skb_release_head_state(nskb); 394189319d38SHerbert Xu __skb_push(nskb, doffset); 394289319d38SHerbert Xu } else { 3943c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3944df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3945c93bdd0eSMel Gorman NUMA_NO_NODE); 394689319d38SHerbert Xu 394789319d38SHerbert Xu if (unlikely(!nskb)) 394889319d38SHerbert Xu goto err; 394989319d38SHerbert Xu 395089319d38SHerbert Xu skb_reserve(nskb, headroom); 395189319d38SHerbert Xu __skb_put(nskb, doffset); 395289319d38SHerbert Xu } 395389319d38SHerbert Xu 3954f4c50d99SHerbert Xu if (segs) 3955f4c50d99SHerbert Xu tail->next = nskb; 3956f4c50d99SHerbert Xu else 3957f4c50d99SHerbert Xu segs = nskb; 3958f4c50d99SHerbert Xu tail = nskb; 3959f4c50d99SHerbert Xu 3960df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3961f4c50d99SHerbert Xu 3962030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3963fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 396468c33163SPravin B Shelar 3965df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 396668c33163SPravin B Shelar nskb->data - tnl_hlen, 396768c33163SPravin B Shelar doffset + tnl_hlen); 396889319d38SHerbert Xu 39699d8506ccSHerbert Xu if (nskb->len == len + doffset) 39701cdbcb79SSimon Horman goto perform_csum_check; 397189319d38SHerbert Xu 39727fbeffedSAlexander Duyck if (!sg) { 39731454c9faSYadu Kishore if (!csum) { 39747fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 39756f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 397676443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 397776443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 39781454c9faSYadu Kishore skb_put(nskb, 39791454c9faSYadu Kishore len), 39808d5930dfSAl Viro len); 39817e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3982de843723STom Herbert skb_headroom(nskb) + doffset; 39831454c9faSYadu Kishore } else { 39841454c9faSYadu Kishore skb_copy_bits(head_skb, offset, 39851454c9faSYadu Kishore skb_put(nskb, len), 39861454c9faSYadu Kishore len); 39871454c9faSYadu Kishore } 3988f4c50d99SHerbert Xu continue; 3989f4c50d99SHerbert Xu } 3990f4c50d99SHerbert Xu 39918cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3992f4c50d99SHerbert Xu 3993df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3994d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3995f4c50d99SHerbert Xu 3996fff88030SWillem de Bruijn skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags & 3997df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3998cef401deSEric Dumazet 3999bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4000bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4001bf5c25d6SWillem de Bruijn goto err; 4002bf5c25d6SWillem de Bruijn 40039d8506ccSHerbert Xu while (pos < offset + len) { 40049d8506ccSHerbert Xu if (i >= nfrags) { 40059d8506ccSHerbert Xu i = 0; 40061a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 40071a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 40081fd819ecSMichael S. Tsirkin frag_skb = list_skb; 400913acc94eSYonghong Song if (!skb_headlen(list_skb)) { 40109d8506ccSHerbert Xu BUG_ON(!nfrags); 401113acc94eSYonghong Song } else { 401213acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 40139d8506ccSHerbert Xu 401413acc94eSYonghong Song /* to make room for head_frag. */ 401513acc94eSYonghong Song i--; 401613acc94eSYonghong Song frag--; 401713acc94eSYonghong Song } 4018bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4019bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4020bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4021bf5c25d6SWillem de Bruijn goto err; 4022bf5c25d6SWillem de Bruijn 40231a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 40249d8506ccSHerbert Xu } 40259d8506ccSHerbert Xu 40269d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 40279d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 40289d8506ccSHerbert Xu net_warn_ratelimited( 40299d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 40309d8506ccSHerbert Xu pos, mss); 4031ff907a11SEric Dumazet err = -EINVAL; 40329d8506ccSHerbert Xu goto err; 40339d8506ccSHerbert Xu } 40349d8506ccSHerbert Xu 403513acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 40368cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 40378cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4038f4c50d99SHerbert Xu 4039f4c50d99SHerbert Xu if (pos < offset) { 4040b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 40418cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4042f4c50d99SHerbert Xu } 4043f4c50d99SHerbert Xu 404489319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4045f4c50d99SHerbert Xu 4046f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4047f4c50d99SHerbert Xu i++; 40484e1beba1SMichael S. Tsirkin frag++; 4049f4c50d99SHerbert Xu pos += size; 4050f4c50d99SHerbert Xu } else { 40518cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 405289319d38SHerbert Xu goto skip_fraglist; 4053f4c50d99SHerbert Xu } 4054f4c50d99SHerbert Xu 40558cb19905SMichael S. Tsirkin nskb_frag++; 4056f4c50d99SHerbert Xu } 4057f4c50d99SHerbert Xu 405889319d38SHerbert Xu skip_fraglist: 4059f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4060f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4061f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4062ec5f0615SPravin B Shelar 40631cdbcb79SSimon Horman perform_csum_check: 40647fbeffedSAlexander Duyck if (!csum) { 4065ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4066ff907a11SEric Dumazet __skb_linearize(nskb)) 4067ddff00d4SAlexander Duyck goto err; 4068ff907a11SEric Dumazet 40697fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4070ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 407176443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 407276443456SAlexander Duyck skb_checksum(nskb, doffset, 407376443456SAlexander Duyck nskb->len - doffset, 0); 40747e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 40757e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4076ec5f0615SPravin B Shelar } 4077df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4078f4c50d99SHerbert Xu 4079bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4080bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4081bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4082bec3cfdcSEric Dumazet */ 4083bec3cfdcSEric Dumazet segs->prev = tail; 4084432c856fSToshiaki Makita 4085802ab55aSAlexander Duyck if (partial_segs) { 408607b26c94SSteffen Klassert struct sk_buff *iter; 4087802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 408807b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4089802ab55aSAlexander Duyck 4090802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 409107b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4092802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4093802ab55aSAlexander Duyck 4094802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4095802ab55aSAlexander Duyck * our way back down the stack of protocols. 4096802ab55aSAlexander Duyck */ 409707b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 409807b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 409907b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 410007b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 410107b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 410207b26c94SSteffen Klassert } 410307b26c94SSteffen Klassert 410407b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 410507b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 410607b26c94SSteffen Klassert else if (tail != segs) 410707b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4108802ab55aSAlexander Duyck } 4109802ab55aSAlexander Duyck 4110432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4111432c856fSToshiaki Makita * using skb_set_owner_w(). 4112432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4113432c856fSToshiaki Makita */ 4114432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4115432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4116432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4117432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4118432c856fSToshiaki Makita } 4119f4c50d99SHerbert Xu return segs; 4120f4c50d99SHerbert Xu 4121f4c50d99SHerbert Xu err: 4122289dccbeSEric Dumazet kfree_skb_list(segs); 4123f4c50d99SHerbert Xu return ERR_PTR(err); 4124f4c50d99SHerbert Xu } 4125f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4126f4c50d99SHerbert Xu 4127d4546c25SDavid Miller int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb) 412871d93b39SHerbert Xu { 41298a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 413067147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 413167147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 41328a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 4133715dc1f3SEric Dumazet unsigned int delta_truesize; 4134d4546c25SDavid Miller struct sk_buff *lp; 413571d93b39SHerbert Xu 41360ab03f35SSteffen Klassert if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush)) 413771d93b39SHerbert Xu return -E2BIG; 413871d93b39SHerbert Xu 413929e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 41408a29111cSEric Dumazet pinfo = skb_shinfo(lp); 41418a29111cSEric Dumazet 41428a29111cSEric Dumazet if (headlen <= offset) { 414342da6994SHerbert Xu skb_frag_t *frag; 414466e92fcfSHerbert Xu skb_frag_t *frag2; 41459aaa156cSHerbert Xu int i = skbinfo->nr_frags; 41469aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 414742da6994SHerbert Xu 414866e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 41498a29111cSEric Dumazet goto merge; 415081705ad1SHerbert Xu 41518a29111cSEric Dumazet offset -= headlen; 41529aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 41539aaa156cSHerbert Xu skbinfo->nr_frags = 0; 4154f5572068SHerbert Xu 41559aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 41569aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 415766e92fcfSHerbert Xu do { 415866e92fcfSHerbert Xu *--frag = *--frag2; 415966e92fcfSHerbert Xu } while (--i); 416066e92fcfSHerbert Xu 4161b54c9d5bSJonathan Lemon skb_frag_off_add(frag, offset); 41629e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 416366e92fcfSHerbert Xu 4164715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 4165ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 4166ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 4167715dc1f3SEric Dumazet 4168f5572068SHerbert Xu skb->truesize -= skb->data_len; 4169f5572068SHerbert Xu skb->len -= skb->data_len; 4170f5572068SHerbert Xu skb->data_len = 0; 4171f5572068SHerbert Xu 4172715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 41735d38a079SHerbert Xu goto done; 4174d7e8883cSEric Dumazet } else if (skb->head_frag) { 4175d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 4176d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 4177d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 4178d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 4179d7e8883cSEric Dumazet unsigned int first_offset; 4180d7e8883cSEric Dumazet 4181d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 41828a29111cSEric Dumazet goto merge; 4183d7e8883cSEric Dumazet 4184d7e8883cSEric Dumazet first_offset = skb->data - 4185d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 4186d7e8883cSEric Dumazet offset; 4187d7e8883cSEric Dumazet 4188d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 4189d7e8883cSEric Dumazet 4190d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(frag, page); 4191b54c9d5bSJonathan Lemon skb_frag_off_set(frag, first_offset); 4192d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 4193d7e8883cSEric Dumazet 4194d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 4195d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 4196d7e8883cSEric Dumazet 4197715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4198d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 4199d7e8883cSEric Dumazet goto done; 42008a29111cSEric Dumazet } 420171d93b39SHerbert Xu 420271d93b39SHerbert Xu merge: 4203715dc1f3SEric Dumazet delta_truesize = skb->truesize; 420467147ba9SHerbert Xu if (offset > headlen) { 4205d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 4206d1dc7abfSMichal Schmidt 4207b54c9d5bSJonathan Lemon skb_frag_off_add(&skbinfo->frags[0], eat); 42089e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 4209d1dc7abfSMichal Schmidt skb->data_len -= eat; 4210d1dc7abfSMichal Schmidt skb->len -= eat; 421167147ba9SHerbert Xu offset = headlen; 421256035022SHerbert Xu } 421356035022SHerbert Xu 421467147ba9SHerbert Xu __skb_pull(skb, offset); 421556035022SHerbert Xu 421629e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 42178a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 42188a29111cSEric Dumazet else 4219c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 4220c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 4221f4a775d1SEric Dumazet __skb_header_release(skb); 42228a29111cSEric Dumazet lp = p; 422371d93b39SHerbert Xu 42245d38a079SHerbert Xu done: 42255d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 422637fe4732SHerbert Xu p->data_len += len; 4227715dc1f3SEric Dumazet p->truesize += delta_truesize; 422837fe4732SHerbert Xu p->len += len; 42298a29111cSEric Dumazet if (lp != p) { 42308a29111cSEric Dumazet lp->data_len += len; 42318a29111cSEric Dumazet lp->truesize += delta_truesize; 42328a29111cSEric Dumazet lp->len += len; 42338a29111cSEric Dumazet } 423471d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 423571d93b39SHerbert Xu return 0; 423671d93b39SHerbert Xu } 423771d93b39SHerbert Xu 4238df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4239df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4240df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4241df5042f4SFlorian Westphal 4242df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4243df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4244df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4245df5042f4SFlorian Westphal #endif 42464165079bSFlorian Westphal #ifdef CONFIG_XFRM 42474165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 42484165079bSFlorian Westphal #endif 424995a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 425095a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 425195a7233cSPaul Blakey #endif 42523ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 42533ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 42543ee17bc7SMat Martineau #endif 4255df5042f4SFlorian Westphal }; 4256df5042f4SFlorian Westphal 4257df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4258df5042f4SFlorian Westphal { 4259df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4260df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4261df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4262df5042f4SFlorian Westphal #endif 42634165079bSFlorian Westphal #ifdef CONFIG_XFRM 42644165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 42654165079bSFlorian Westphal #endif 426695a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 426795a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 426895a7233cSPaul Blakey #endif 42693ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 42703ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 42713ee17bc7SMat Martineau #endif 4272df5042f4SFlorian Westphal 0; 4273df5042f4SFlorian Westphal } 4274df5042f4SFlorian Westphal 4275df5042f4SFlorian Westphal static void skb_extensions_init(void) 4276df5042f4SFlorian Westphal { 4277df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4278df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4279df5042f4SFlorian Westphal 4280df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4281df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4282df5042f4SFlorian Westphal 0, 4283df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4284df5042f4SFlorian Westphal NULL); 4285df5042f4SFlorian Westphal } 4286df5042f4SFlorian Westphal #else 4287df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4288df5042f4SFlorian Westphal #endif 4289df5042f4SFlorian Westphal 42901da177e4SLinus Torvalds void __init skb_init(void) 42911da177e4SLinus Torvalds { 429279a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 42931da177e4SLinus Torvalds sizeof(struct sk_buff), 42941da177e4SLinus Torvalds 0, 4295e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 429679a8a642SKees Cook offsetof(struct sk_buff, cb), 429779a8a642SKees Cook sizeof_field(struct sk_buff, cb), 429820c2df83SPaul Mundt NULL); 4299d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4300d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4301d179cd12SDavid S. Miller 0, 4302e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 430320c2df83SPaul Mundt NULL); 4304df5042f4SFlorian Westphal skb_extensions_init(); 43051da177e4SLinus Torvalds } 43061da177e4SLinus Torvalds 430751c739d1SDavid S. Miller static int 430848a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 430948a1df65SJason A. Donenfeld unsigned int recursion_level) 4310716ea3a7SDavid Howells { 43111a028e50SDavid S. Miller int start = skb_headlen(skb); 43121a028e50SDavid S. Miller int i, copy = start - offset; 4313fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4314716ea3a7SDavid Howells int elt = 0; 4315716ea3a7SDavid Howells 431648a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 431748a1df65SJason A. Donenfeld return -EMSGSIZE; 431848a1df65SJason A. Donenfeld 4319716ea3a7SDavid Howells if (copy > 0) { 4320716ea3a7SDavid Howells if (copy > len) 4321716ea3a7SDavid Howells copy = len; 4322642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4323716ea3a7SDavid Howells elt++; 4324716ea3a7SDavid Howells if ((len -= copy) == 0) 4325716ea3a7SDavid Howells return elt; 4326716ea3a7SDavid Howells offset += copy; 4327716ea3a7SDavid Howells } 4328716ea3a7SDavid Howells 4329716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 43301a028e50SDavid S. Miller int end; 4331716ea3a7SDavid Howells 4332547b792cSIlpo Järvinen WARN_ON(start > offset + len); 43331a028e50SDavid S. Miller 43349e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4335716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4336716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 433748a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 433848a1df65SJason A. Donenfeld return -EMSGSIZE; 4339716ea3a7SDavid Howells 4340716ea3a7SDavid Howells if (copy > len) 4341716ea3a7SDavid Howells copy = len; 4342ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4343b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4344716ea3a7SDavid Howells elt++; 4345716ea3a7SDavid Howells if (!(len -= copy)) 4346716ea3a7SDavid Howells return elt; 4347716ea3a7SDavid Howells offset += copy; 4348716ea3a7SDavid Howells } 43491a028e50SDavid S. Miller start = end; 4350716ea3a7SDavid Howells } 4351716ea3a7SDavid Howells 4352fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 435348a1df65SJason A. Donenfeld int end, ret; 4354716ea3a7SDavid Howells 4355547b792cSIlpo Järvinen WARN_ON(start > offset + len); 43561a028e50SDavid S. Miller 4357fbb398a8SDavid S. Miller end = start + frag_iter->len; 4358716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 435948a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 436048a1df65SJason A. Donenfeld return -EMSGSIZE; 436148a1df65SJason A. Donenfeld 4362716ea3a7SDavid Howells if (copy > len) 4363716ea3a7SDavid Howells copy = len; 436448a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 436548a1df65SJason A. Donenfeld copy, recursion_level + 1); 436648a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 436748a1df65SJason A. Donenfeld return ret; 436848a1df65SJason A. Donenfeld elt += ret; 4369716ea3a7SDavid Howells if ((len -= copy) == 0) 4370716ea3a7SDavid Howells return elt; 4371716ea3a7SDavid Howells offset += copy; 4372716ea3a7SDavid Howells } 43731a028e50SDavid S. Miller start = end; 4374716ea3a7SDavid Howells } 4375716ea3a7SDavid Howells BUG_ON(len); 4376716ea3a7SDavid Howells return elt; 4377716ea3a7SDavid Howells } 4378716ea3a7SDavid Howells 437948a1df65SJason A. Donenfeld /** 438048a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 438148a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 438248a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 438348a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 438448a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 438548a1df65SJason A. Donenfeld * 438648a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 438748a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 438848a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 438948a1df65SJason A. Donenfeld * could not fit. 439048a1df65SJason A. Donenfeld */ 439148a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 439248a1df65SJason A. Donenfeld { 439348a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 439448a1df65SJason A. Donenfeld 439548a1df65SJason A. Donenfeld if (nsg <= 0) 439648a1df65SJason A. Donenfeld return nsg; 439748a1df65SJason A. Donenfeld 439848a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 439948a1df65SJason A. Donenfeld 440048a1df65SJason A. Donenfeld return nsg; 440148a1df65SJason A. Donenfeld } 440248a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 440348a1df65SJason A. Donenfeld 440425a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 440525a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 440625a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 440725a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 440825a91d8dSFan Du * 440925a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 441025a91d8dSFan Du * 1. sg_init_table 441125a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 441225a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 441325a91d8dSFan Du * 441425a91d8dSFan Du * This is equivalent to: 441525a91d8dSFan Du * 1. sg_init_table 441625a91d8dSFan Du * 2. skb_to_sgvec(payload1) 441725a91d8dSFan Du * 3. sg_unmark_end 441825a91d8dSFan Du * 4. skb_to_sgvec(payload2) 441925a91d8dSFan Du * 442025a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 442125a91d8dSFan Du * is more preferable. 442225a91d8dSFan Du */ 442325a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 442425a91d8dSFan Du int offset, int len) 442525a91d8dSFan Du { 442648a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 442725a91d8dSFan Du } 442825a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 442925a91d8dSFan Du 443051c739d1SDavid S. Miller 443151c739d1SDavid S. Miller 4432716ea3a7SDavid Howells /** 4433716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4434716ea3a7SDavid Howells * @skb: The socket buffer to check. 4435716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4436716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4437716ea3a7SDavid Howells * 4438716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4439716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4440716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4441716ea3a7SDavid Howells * 4442716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4443716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4444716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4445716ea3a7SDavid Howells * 4446716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4447716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4448716ea3a7SDavid Howells */ 4449716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4450716ea3a7SDavid Howells { 4451716ea3a7SDavid Howells int copyflag; 4452716ea3a7SDavid Howells int elt; 4453716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4454716ea3a7SDavid Howells 4455716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4456716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4457716ea3a7SDavid Howells * at the moment even if they are anonymous). 4458716ea3a7SDavid Howells */ 4459716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4460c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4461716ea3a7SDavid Howells return -ENOMEM; 4462716ea3a7SDavid Howells 4463716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 446421dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4465716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4466716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4467716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4468716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4469716ea3a7SDavid Howells 4470716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4471716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4472716ea3a7SDavid Howells return -ENOMEM; 4473716ea3a7SDavid Howells 4474716ea3a7SDavid Howells /* Voila! */ 4475716ea3a7SDavid Howells *trailer = skb; 4476716ea3a7SDavid Howells return 1; 4477716ea3a7SDavid Howells } 4478716ea3a7SDavid Howells 4479716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4480716ea3a7SDavid Howells 4481716ea3a7SDavid Howells elt = 1; 4482716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4483716ea3a7SDavid Howells copyflag = 0; 4484716ea3a7SDavid Howells 4485716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4486716ea3a7SDavid Howells int ntail = 0; 4487716ea3a7SDavid Howells 4488716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4489716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4490716ea3a7SDavid Howells * after it. */ 4491716ea3a7SDavid Howells 4492716ea3a7SDavid Howells if (skb_shared(skb1)) 4493716ea3a7SDavid Howells copyflag = 1; 4494716ea3a7SDavid Howells 4495716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4496716ea3a7SDavid Howells 4497716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4498716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 449921dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4500716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4501716ea3a7SDavid Howells ntail = tailbits + 128; 4502716ea3a7SDavid Howells } 4503716ea3a7SDavid Howells 4504716ea3a7SDavid Howells if (copyflag || 4505716ea3a7SDavid Howells skb_cloned(skb1) || 4506716ea3a7SDavid Howells ntail || 4507716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 450821dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4509716ea3a7SDavid Howells struct sk_buff *skb2; 4510716ea3a7SDavid Howells 4511716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4512716ea3a7SDavid Howells if (ntail == 0) 4513716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4514716ea3a7SDavid Howells else 4515716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4516716ea3a7SDavid Howells skb_headroom(skb1), 4517716ea3a7SDavid Howells ntail, 4518716ea3a7SDavid Howells GFP_ATOMIC); 4519716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4520716ea3a7SDavid Howells return -ENOMEM; 4521716ea3a7SDavid Howells 4522716ea3a7SDavid Howells if (skb1->sk) 4523716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4524716ea3a7SDavid Howells 4525716ea3a7SDavid Howells /* Looking around. Are we still alive? 4526716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4527716ea3a7SDavid Howells 4528716ea3a7SDavid Howells skb2->next = skb1->next; 4529716ea3a7SDavid Howells *skb_p = skb2; 4530716ea3a7SDavid Howells kfree_skb(skb1); 4531716ea3a7SDavid Howells skb1 = skb2; 4532716ea3a7SDavid Howells } 4533716ea3a7SDavid Howells elt++; 4534716ea3a7SDavid Howells *trailer = skb1; 4535716ea3a7SDavid Howells skb_p = &skb1->next; 4536716ea3a7SDavid Howells } 4537716ea3a7SDavid Howells 4538716ea3a7SDavid Howells return elt; 4539716ea3a7SDavid Howells } 4540b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4541716ea3a7SDavid Howells 4542b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4543b1faf566SEric Dumazet { 4544b1faf566SEric Dumazet struct sock *sk = skb->sk; 4545b1faf566SEric Dumazet 4546b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4547b1faf566SEric Dumazet } 4548b1faf566SEric Dumazet 45498605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 45508605330aSSoheil Hassas Yeganeh { 45518605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 45528605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 45538605330aSSoheil Hassas Yeganeh */ 45548605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 45558605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 45568605330aSSoheil Hassas Yeganeh } 45578605330aSSoheil Hassas Yeganeh 4558b1faf566SEric Dumazet /* 4559b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4560b1faf566SEric Dumazet */ 4561b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4562b1faf566SEric Dumazet { 4563b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4564ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4565b1faf566SEric Dumazet return -ENOMEM; 4566b1faf566SEric Dumazet 4567b1faf566SEric Dumazet skb_orphan(skb); 4568b1faf566SEric Dumazet skb->sk = sk; 4569b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4570b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 45718605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4572b1faf566SEric Dumazet 4573abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4574abb57ea4SEric Dumazet skb_dst_force(skb); 4575abb57ea4SEric Dumazet 4576b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4577b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 45786e5d58fdSVinicius Costa Gomes sk->sk_error_report(sk); 4579b1faf566SEric Dumazet return 0; 4580b1faf566SEric Dumazet } 4581b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4582b1faf566SEric Dumazet 458383a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 458483a1a1a7SSoheil Hassas Yeganeh { 458583a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 458683a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 458783a1a1a7SSoheil Hassas Yeganeh } 458883a1a1a7SSoheil Hassas Yeganeh 4589364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4590364a9e93SWillem de Bruijn { 4591364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 459283a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 459383a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4594997d5c3fSEric Dumazet unsigned long flags; 4595364a9e93SWillem de Bruijn 4596997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4597364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 459838b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 459983a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 460038b25793SSoheil Hassas Yeganeh if (icmp_next) 4601985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 460238b25793SSoheil Hassas Yeganeh } 4603997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4604364a9e93SWillem de Bruijn 460583a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 460683a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 460783a1a1a7SSoheil Hassas Yeganeh 460883a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4609364a9e93SWillem de Bruijn sk->sk_error_report(sk); 4610364a9e93SWillem de Bruijn 4611364a9e93SWillem de Bruijn return skb; 4612364a9e93SWillem de Bruijn } 4613364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4614364a9e93SWillem de Bruijn 4615cab41c47SAlexander Duyck /** 4616cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4617cab41c47SAlexander Duyck * @skb: the skb to clone 4618cab41c47SAlexander Duyck * 4619cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4620cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4621cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4622cab41c47SAlexander Duyck * 4623cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4624cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4625cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4626cab41c47SAlexander Duyck * the sk_error_queue. 4627cab41c47SAlexander Duyck */ 462862bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 462962bccb8cSAlexander Duyck { 463062bccb8cSAlexander Duyck struct sock *sk = skb->sk; 463162bccb8cSAlexander Duyck struct sk_buff *clone; 463262bccb8cSAlexander Duyck 463341c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 463462bccb8cSAlexander Duyck return NULL; 463562bccb8cSAlexander Duyck 463662bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 463762bccb8cSAlexander Duyck if (!clone) { 463862bccb8cSAlexander Duyck sock_put(sk); 463962bccb8cSAlexander Duyck return NULL; 464062bccb8cSAlexander Duyck } 464162bccb8cSAlexander Duyck 464262bccb8cSAlexander Duyck clone->sk = sk; 464362bccb8cSAlexander Duyck clone->destructor = sock_efree; 464462bccb8cSAlexander Duyck 464562bccb8cSAlexander Duyck return clone; 464662bccb8cSAlexander Duyck } 464762bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 464862bccb8cSAlexander Duyck 464937846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 465037846ef0SAlexander Duyck struct sock *sk, 46514ef1b286SSoheil Hassas Yeganeh int tstype, 46524ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4653ac45f602SPatrick Ohly { 4654ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4655ac45f602SPatrick Ohly int err; 4656ac45f602SPatrick Ohly 46574ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 46584ef1b286SSoheil Hassas Yeganeh 4659ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4660ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4661ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4662ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4663e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 46644ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 46651862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 46664ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 466709c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 4668ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 4669ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 46704ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 46714ed2d765SWillem de Bruijn } 467229030374SEric Dumazet 4673ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 467429030374SEric Dumazet 4675ac45f602SPatrick Ohly if (err) 4676ac45f602SPatrick Ohly kfree_skb(skb); 4677ac45f602SPatrick Ohly } 467837846ef0SAlexander Duyck 4679b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4680b245be1fSWillem de Bruijn { 4681b245be1fSWillem de Bruijn bool ret; 4682b245be1fSWillem de Bruijn 4683b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4684b245be1fSWillem de Bruijn return true; 4685b245be1fSWillem de Bruijn 4686b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4687b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4688b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4689b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4690b245be1fSWillem de Bruijn return ret; 4691b245be1fSWillem de Bruijn } 4692b245be1fSWillem de Bruijn 469337846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 469437846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 469537846ef0SAlexander Duyck { 469637846ef0SAlexander Duyck struct sock *sk = skb->sk; 469737846ef0SAlexander Duyck 4698b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 469935b99dffSWillem de Bruijn goto err; 4700b245be1fSWillem de Bruijn 47019ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 47029ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 47039ac25fc0SEric Dumazet */ 470441c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 470537846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 47064ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 470737846ef0SAlexander Duyck sock_put(sk); 470835b99dffSWillem de Bruijn return; 470937846ef0SAlexander Duyck } 471035b99dffSWillem de Bruijn 471135b99dffSWillem de Bruijn err: 471235b99dffSWillem de Bruijn kfree_skb(skb); 47139ac25fc0SEric Dumazet } 471437846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 471537846ef0SAlexander Duyck 471637846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 471737846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 471837846ef0SAlexander Duyck struct sock *sk, int tstype) 471937846ef0SAlexander Duyck { 472037846ef0SAlexander Duyck struct sk_buff *skb; 47214ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 472237846ef0SAlexander Duyck 47233a8dd971SWillem de Bruijn if (!sk) 47243a8dd971SWillem de Bruijn return; 47253a8dd971SWillem de Bruijn 4726b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4727b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4728b50a5c70SMiroslav Lichvar return; 4729b50a5c70SMiroslav Lichvar 47303a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 47313a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 473237846ef0SAlexander Duyck return; 473337846ef0SAlexander Duyck 47341c885808SFrancis Yan if (tsonly) { 47351c885808SFrancis Yan #ifdef CONFIG_INET 47361c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 47371c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 47384ef1b286SSoheil Hassas Yeganeh sk->sk_type == SOCK_STREAM) { 473948040793SYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb); 47404ef1b286SSoheil Hassas Yeganeh opt_stats = true; 47414ef1b286SSoheil Hassas Yeganeh } else 47421c885808SFrancis Yan #endif 47431c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 47441c885808SFrancis Yan } else { 474537846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 47461c885808SFrancis Yan } 474737846ef0SAlexander Duyck if (!skb) 474837846ef0SAlexander Duyck return; 474937846ef0SAlexander Duyck 475049ca0d8bSWillem de Bruijn if (tsonly) { 4751fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4752fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 475349ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 475449ca0d8bSWillem de Bruijn } 475549ca0d8bSWillem de Bruijn 475649ca0d8bSWillem de Bruijn if (hwtstamps) 475749ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 475849ca0d8bSWillem de Bruijn else 475949ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 476049ca0d8bSWillem de Bruijn 47614ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 476237846ef0SAlexander Duyck } 4763e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4764e7fd2885SWillem de Bruijn 4765e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4766e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4767e7fd2885SWillem de Bruijn { 4768e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 4769e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4770e7fd2885SWillem de Bruijn } 4771ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4772ac45f602SPatrick Ohly 47736e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 47746e3e939fSJohannes Berg { 47756e3e939fSJohannes Berg struct sock *sk = skb->sk; 47766e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4777dd4f1072SEric Dumazet int err = 1; 47786e3e939fSJohannes Berg 47796e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 47806e3e939fSJohannes Berg skb->wifi_acked = acked; 47816e3e939fSJohannes Berg 47826e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 47836e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 47846e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 47856e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 47866e3e939fSJohannes Berg 4787dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4788dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4789dd4f1072SEric Dumazet */ 479041c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 47916e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4792dd4f1072SEric Dumazet sock_put(sk); 4793dd4f1072SEric Dumazet } 47946e3e939fSJohannes Berg if (err) 47956e3e939fSJohannes Berg kfree_skb(skb); 47966e3e939fSJohannes Berg } 47976e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 47986e3e939fSJohannes Berg 4799f35d9d8aSRusty Russell /** 4800f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4801f35d9d8aSRusty Russell * @skb: the skb to set 4802f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4803f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4804f35d9d8aSRusty Russell * 4805f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4806f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4807f35d9d8aSRusty Russell * 4808f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4809f35d9d8aSRusty Russell * returns false you should drop the packet. 4810f35d9d8aSRusty Russell */ 4811f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4812f35d9d8aSRusty Russell { 481352b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 481452b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 481552b5d6f5SEric Dumazet 481652b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 481752b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 481852b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4819f35d9d8aSRusty Russell return false; 4820f35d9d8aSRusty Russell } 4821f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 482252b5d6f5SEric Dumazet skb->csum_start = csum_start; 4823f35d9d8aSRusty Russell skb->csum_offset = off; 4824e5d5decaSJason Wang skb_set_transport_header(skb, start); 4825f35d9d8aSRusty Russell return true; 4826f35d9d8aSRusty Russell } 4827b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4828f35d9d8aSRusty Russell 4829ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4830ed1f50c3SPaul Durrant unsigned int max) 4831ed1f50c3SPaul Durrant { 4832ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4833ed1f50c3SPaul Durrant return 0; 4834ed1f50c3SPaul Durrant 4835ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4836ed1f50c3SPaul Durrant * won't need to do it again. 4837ed1f50c3SPaul Durrant */ 4838ed1f50c3SPaul Durrant if (max > skb->len) 4839ed1f50c3SPaul Durrant max = skb->len; 4840ed1f50c3SPaul Durrant 4841ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4842ed1f50c3SPaul Durrant return -ENOMEM; 4843ed1f50c3SPaul Durrant 4844ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4845ed1f50c3SPaul Durrant return -EPROTO; 4846ed1f50c3SPaul Durrant 4847ed1f50c3SPaul Durrant return 0; 4848ed1f50c3SPaul Durrant } 4849ed1f50c3SPaul Durrant 4850f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4851f9708b43SJan Beulich 4852f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4853f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4854f9708b43SJan Beulich unsigned int off) 4855f9708b43SJan Beulich { 4856f9708b43SJan Beulich int err; 4857f9708b43SJan Beulich 4858161d1792SKees Cook switch (proto) { 4859f9708b43SJan Beulich case IPPROTO_TCP: 4860f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4861f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4862f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4863f9708b43SJan Beulich offsetof(struct tcphdr, 4864f9708b43SJan Beulich check))) 4865f9708b43SJan Beulich err = -EPROTO; 4866f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4867f9708b43SJan Beulich 4868f9708b43SJan Beulich case IPPROTO_UDP: 4869f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4870f9708b43SJan Beulich off + sizeof(struct udphdr)); 4871f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4872f9708b43SJan Beulich offsetof(struct udphdr, 4873f9708b43SJan Beulich check))) 4874f9708b43SJan Beulich err = -EPROTO; 4875f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4876f9708b43SJan Beulich } 4877f9708b43SJan Beulich 4878f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4879f9708b43SJan Beulich } 4880f9708b43SJan Beulich 4881ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4882ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4883ed1f50c3SPaul Durrant */ 4884ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4885ed1f50c3SPaul Durrant 4886f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4887ed1f50c3SPaul Durrant { 4888ed1f50c3SPaul Durrant unsigned int off; 4889ed1f50c3SPaul Durrant bool fragment; 4890f9708b43SJan Beulich __sum16 *csum; 4891ed1f50c3SPaul Durrant int err; 4892ed1f50c3SPaul Durrant 4893ed1f50c3SPaul Durrant fragment = false; 4894ed1f50c3SPaul Durrant 4895ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4896ed1f50c3SPaul Durrant sizeof(struct iphdr), 4897ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4898ed1f50c3SPaul Durrant if (err < 0) 4899ed1f50c3SPaul Durrant goto out; 4900ed1f50c3SPaul Durrant 490111f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 4902ed1f50c3SPaul Durrant fragment = true; 4903ed1f50c3SPaul Durrant 4904ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 4905ed1f50c3SPaul Durrant 4906ed1f50c3SPaul Durrant err = -EPROTO; 4907ed1f50c3SPaul Durrant 4908ed1f50c3SPaul Durrant if (fragment) 4909ed1f50c3SPaul Durrant goto out; 4910ed1f50c3SPaul Durrant 4911f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4912f9708b43SJan Beulich if (IS_ERR(csum)) 4913f9708b43SJan Beulich return PTR_ERR(csum); 4914ed1f50c3SPaul Durrant 4915ed1f50c3SPaul Durrant if (recalculate) 4916f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4917ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4918ed1f50c3SPaul Durrant skb->len - off, 4919f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4920ed1f50c3SPaul Durrant err = 0; 4921ed1f50c3SPaul Durrant 4922ed1f50c3SPaul Durrant out: 4923ed1f50c3SPaul Durrant return err; 4924ed1f50c3SPaul Durrant } 4925ed1f50c3SPaul Durrant 4926ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4927ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4928ed1f50c3SPaul Durrant */ 4929ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4930ed1f50c3SPaul Durrant 4931ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4932ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4933ed1f50c3SPaul Durrant 4934ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4935ed1f50c3SPaul Durrant { 4936ed1f50c3SPaul Durrant int err; 4937ed1f50c3SPaul Durrant u8 nexthdr; 4938ed1f50c3SPaul Durrant unsigned int off; 4939ed1f50c3SPaul Durrant unsigned int len; 4940ed1f50c3SPaul Durrant bool fragment; 4941ed1f50c3SPaul Durrant bool done; 4942f9708b43SJan Beulich __sum16 *csum; 4943ed1f50c3SPaul Durrant 4944ed1f50c3SPaul Durrant fragment = false; 4945ed1f50c3SPaul Durrant done = false; 4946ed1f50c3SPaul Durrant 4947ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4948ed1f50c3SPaul Durrant 4949ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4950ed1f50c3SPaul Durrant if (err < 0) 4951ed1f50c3SPaul Durrant goto out; 4952ed1f50c3SPaul Durrant 4953ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4954ed1f50c3SPaul Durrant 4955ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4956ed1f50c3SPaul Durrant while (off <= len && !done) { 4957ed1f50c3SPaul Durrant switch (nexthdr) { 4958ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4959ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4960ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4961ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 4962ed1f50c3SPaul Durrant 4963ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4964ed1f50c3SPaul Durrant off + 4965ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 4966ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4967ed1f50c3SPaul Durrant if (err < 0) 4968ed1f50c3SPaul Durrant goto out; 4969ed1f50c3SPaul Durrant 4970ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4971ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4972ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 4973ed1f50c3SPaul Durrant break; 4974ed1f50c3SPaul Durrant } 4975ed1f50c3SPaul Durrant case IPPROTO_AH: { 4976ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 4977ed1f50c3SPaul Durrant 4978ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4979ed1f50c3SPaul Durrant off + 4980ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 4981ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4982ed1f50c3SPaul Durrant if (err < 0) 4983ed1f50c3SPaul Durrant goto out; 4984ed1f50c3SPaul Durrant 4985ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4986ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4987ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 4988ed1f50c3SPaul Durrant break; 4989ed1f50c3SPaul Durrant } 4990ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 4991ed1f50c3SPaul Durrant struct frag_hdr *hp; 4992ed1f50c3SPaul Durrant 4993ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4994ed1f50c3SPaul Durrant off + 4995ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 4996ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4997ed1f50c3SPaul Durrant if (err < 0) 4998ed1f50c3SPaul Durrant goto out; 4999ed1f50c3SPaul Durrant 5000ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5001ed1f50c3SPaul Durrant 5002ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5003ed1f50c3SPaul Durrant fragment = true; 5004ed1f50c3SPaul Durrant 5005ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5006ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5007ed1f50c3SPaul Durrant break; 5008ed1f50c3SPaul Durrant } 5009ed1f50c3SPaul Durrant default: 5010ed1f50c3SPaul Durrant done = true; 5011ed1f50c3SPaul Durrant break; 5012ed1f50c3SPaul Durrant } 5013ed1f50c3SPaul Durrant } 5014ed1f50c3SPaul Durrant 5015ed1f50c3SPaul Durrant err = -EPROTO; 5016ed1f50c3SPaul Durrant 5017ed1f50c3SPaul Durrant if (!done || fragment) 5018ed1f50c3SPaul Durrant goto out; 5019ed1f50c3SPaul Durrant 5020f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5021f9708b43SJan Beulich if (IS_ERR(csum)) 5022f9708b43SJan Beulich return PTR_ERR(csum); 5023ed1f50c3SPaul Durrant 5024ed1f50c3SPaul Durrant if (recalculate) 5025f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5026ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5027f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5028ed1f50c3SPaul Durrant err = 0; 5029ed1f50c3SPaul Durrant 5030ed1f50c3SPaul Durrant out: 5031ed1f50c3SPaul Durrant return err; 5032ed1f50c3SPaul Durrant } 5033ed1f50c3SPaul Durrant 5034ed1f50c3SPaul Durrant /** 5035ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5036ed1f50c3SPaul Durrant * @skb: the skb to set up 5037ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5038ed1f50c3SPaul Durrant */ 5039ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5040ed1f50c3SPaul Durrant { 5041ed1f50c3SPaul Durrant int err; 5042ed1f50c3SPaul Durrant 5043ed1f50c3SPaul Durrant switch (skb->protocol) { 5044ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5045f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5046ed1f50c3SPaul Durrant break; 5047ed1f50c3SPaul Durrant 5048ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5049ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5050ed1f50c3SPaul Durrant break; 5051ed1f50c3SPaul Durrant 5052ed1f50c3SPaul Durrant default: 5053ed1f50c3SPaul Durrant err = -EPROTO; 5054ed1f50c3SPaul Durrant break; 5055ed1f50c3SPaul Durrant } 5056ed1f50c3SPaul Durrant 5057ed1f50c3SPaul Durrant return err; 5058ed1f50c3SPaul Durrant } 5059ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5060ed1f50c3SPaul Durrant 50619afd85c9SLinus Lüssing /** 50629afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 50639afd85c9SLinus Lüssing * @skb: the skb to check 50649afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 50659afd85c9SLinus Lüssing * 50669afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 50679afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 50689afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 50699afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 50709afd85c9SLinus Lüssing * 5071a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5072a516993fSLinus Lüssing * differs from the provided skb. 50739afd85c9SLinus Lüssing */ 50749afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 50759afd85c9SLinus Lüssing unsigned int transport_len) 50769afd85c9SLinus Lüssing { 50779afd85c9SLinus Lüssing struct sk_buff *skb_chk; 50789afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 50799afd85c9SLinus Lüssing int ret; 50809afd85c9SLinus Lüssing 5081a516993fSLinus Lüssing if (skb->len < len) 50829afd85c9SLinus Lüssing return NULL; 5083a516993fSLinus Lüssing else if (skb->len == len) 50849afd85c9SLinus Lüssing return skb; 50859afd85c9SLinus Lüssing 50869afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 50879afd85c9SLinus Lüssing if (!skb_chk) 50889afd85c9SLinus Lüssing return NULL; 50899afd85c9SLinus Lüssing 50909afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 50919afd85c9SLinus Lüssing if (ret) { 50929afd85c9SLinus Lüssing kfree_skb(skb_chk); 50939afd85c9SLinus Lüssing return NULL; 50949afd85c9SLinus Lüssing } 50959afd85c9SLinus Lüssing 50969afd85c9SLinus Lüssing return skb_chk; 50979afd85c9SLinus Lüssing } 50989afd85c9SLinus Lüssing 50999afd85c9SLinus Lüssing /** 51009afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 51019afd85c9SLinus Lüssing * @skb: the skb to check 51029afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51039afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 51049afd85c9SLinus Lüssing * 51059afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 51069afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 51079afd85c9SLinus Lüssing * 51089afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 51099afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 51109afd85c9SLinus Lüssing * 5111a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5112a516993fSLinus Lüssing * differs from the provided skb. 51139afd85c9SLinus Lüssing */ 51149afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 51159afd85c9SLinus Lüssing unsigned int transport_len, 51169afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 51179afd85c9SLinus Lüssing { 51189afd85c9SLinus Lüssing struct sk_buff *skb_chk; 51199afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5120fcba67c9SLinus Lüssing __sum16 ret; 51219afd85c9SLinus Lüssing 51229afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 51239afd85c9SLinus Lüssing if (!skb_chk) 5124a516993fSLinus Lüssing goto err; 51259afd85c9SLinus Lüssing 5126a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5127a516993fSLinus Lüssing goto err; 51289afd85c9SLinus Lüssing 51299b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 51309afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 51319b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 51329afd85c9SLinus Lüssing 5133a516993fSLinus Lüssing if (ret) 5134a516993fSLinus Lüssing goto err; 51359afd85c9SLinus Lüssing 51369afd85c9SLinus Lüssing return skb_chk; 5137a516993fSLinus Lüssing 5138a516993fSLinus Lüssing err: 5139a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5140a516993fSLinus Lüssing kfree_skb(skb_chk); 5141a516993fSLinus Lüssing 5142a516993fSLinus Lüssing return NULL; 5143a516993fSLinus Lüssing 51449afd85c9SLinus Lüssing } 51459afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 51469afd85c9SLinus Lüssing 51474497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 51484497b076SBen Hutchings { 5149e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5150e87cc472SJoe Perches skb->dev->name); 51514497b076SBen Hutchings } 51524497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5153bad43ca8SEric Dumazet 5154bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5155bad43ca8SEric Dumazet { 51563d861f66SEric Dumazet if (head_stolen) { 51573d861f66SEric Dumazet skb_release_head_state(skb); 5158bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 51593d861f66SEric Dumazet } else { 5160bad43ca8SEric Dumazet __kfree_skb(skb); 5161bad43ca8SEric Dumazet } 51623d861f66SEric Dumazet } 5163bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5164bad43ca8SEric Dumazet 5165bad43ca8SEric Dumazet /** 5166bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5167bad43ca8SEric Dumazet * @to: prior buffer 5168bad43ca8SEric Dumazet * @from: buffer to add 5169bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5170c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5171bad43ca8SEric Dumazet */ 5172bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5173bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5174bad43ca8SEric Dumazet { 5175c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5176bad43ca8SEric Dumazet int i, delta, len = from->len; 5177bad43ca8SEric Dumazet 5178bad43ca8SEric Dumazet *fragstolen = false; 5179bad43ca8SEric Dumazet 5180bad43ca8SEric Dumazet if (skb_cloned(to)) 5181bad43ca8SEric Dumazet return false; 5182bad43ca8SEric Dumazet 5183bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5184e93a0435SEric Dumazet if (len) 5185bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5186bad43ca8SEric Dumazet *delta_truesize = 0; 5187bad43ca8SEric Dumazet return true; 5188bad43ca8SEric Dumazet } 5189bad43ca8SEric Dumazet 5190c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5191c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5192c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5193bad43ca8SEric Dumazet return false; 51941f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 51951f8b977aSWillem de Bruijn return false; 5196bad43ca8SEric Dumazet 5197bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5198bad43ca8SEric Dumazet struct page *page; 5199bad43ca8SEric Dumazet unsigned int offset; 5200bad43ca8SEric Dumazet 5201c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5202c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5203bad43ca8SEric Dumazet return false; 5204bad43ca8SEric Dumazet 5205bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5206bad43ca8SEric Dumazet return false; 5207bad43ca8SEric Dumazet 5208bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5209bad43ca8SEric Dumazet 5210bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5211bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5212bad43ca8SEric Dumazet 5213c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5214bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5215bad43ca8SEric Dumazet *fragstolen = true; 5216bad43ca8SEric Dumazet } else { 5217c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5218c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5219bad43ca8SEric Dumazet return false; 5220bad43ca8SEric Dumazet 5221f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5222bad43ca8SEric Dumazet } 5223bad43ca8SEric Dumazet 5224bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5225bad43ca8SEric Dumazet 5226c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5227c818fa9eSEric Dumazet from_shinfo->frags, 5228c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5229c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5230bad43ca8SEric Dumazet 5231bad43ca8SEric Dumazet if (!skb_cloned(from)) 5232c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5233bad43ca8SEric Dumazet 52348ea853fdSLi RongQing /* if the skb is not cloned this does nothing 52358ea853fdSLi RongQing * since we set nr_frags to 0. 52368ea853fdSLi RongQing */ 5237c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5238c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5239bad43ca8SEric Dumazet 5240bad43ca8SEric Dumazet to->truesize += delta; 5241bad43ca8SEric Dumazet to->len += len; 5242bad43ca8SEric Dumazet to->data_len += len; 5243bad43ca8SEric Dumazet 5244bad43ca8SEric Dumazet *delta_truesize = delta; 5245bad43ca8SEric Dumazet return true; 5246bad43ca8SEric Dumazet } 5247bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5248621e84d6SNicolas Dichtel 5249621e84d6SNicolas Dichtel /** 52508b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5251621e84d6SNicolas Dichtel * 5252621e84d6SNicolas Dichtel * @skb: buffer to clean 52538b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5254621e84d6SNicolas Dichtel * 52558b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 52568b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 52578b27f277SNicolas Dichtel * operations. 52588b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 52598b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 52608b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5261621e84d6SNicolas Dichtel */ 52628b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5263621e84d6SNicolas Dichtel { 5264621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5265621e84d6SNicolas Dichtel skb->skb_iif = 0; 526660ff7467SWANG Cong skb->ignore_df = 0; 5267621e84d6SNicolas Dichtel skb_dst_drop(skb); 5268174e2381SFlorian Westphal skb_ext_reset(skb); 5269895b5c9fSFlorian Westphal nf_reset_ct(skb); 5270621e84d6SNicolas Dichtel nf_reset_trace(skb); 5271213dd74aSHerbert Xu 52726f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 52736f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5274875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 52756f9a5069SPetr Machata #endif 52766f9a5069SPetr Machata 5277213dd74aSHerbert Xu if (!xnet) 5278213dd74aSHerbert Xu return; 5279213dd74aSHerbert Xu 52802b5ec1a5SYe Yin ipvs_reset(skb); 5281213dd74aSHerbert Xu skb->mark = 0; 5282c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 5283621e84d6SNicolas Dichtel } 5284621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5285de960aa9SFlorian Westphal 5286de960aa9SFlorian Westphal /** 5287de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5288de960aa9SFlorian Westphal * 5289de960aa9SFlorian Westphal * @skb: GSO skb 5290de960aa9SFlorian Westphal * 5291de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5292de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5293de960aa9SFlorian Westphal * 5294de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5295de960aa9SFlorian Westphal */ 5296a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5297de960aa9SFlorian Westphal { 5298de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5299f993bc25SFlorian Westphal unsigned int thlen = 0; 5300f993bc25SFlorian Westphal 5301f993bc25SFlorian Westphal if (skb->encapsulation) { 5302f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5303f993bc25SFlorian Westphal skb_transport_header(skb); 5304de960aa9SFlorian Westphal 5305de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5306f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5307f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5308f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 53091dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 531090017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5311ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5312ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5313f993bc25SFlorian Westphal } 53146d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 53156d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 53166d39d589SFlorian Westphal * accounted for. 53176d39d589SFlorian Westphal */ 5318f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5319de960aa9SFlorian Westphal } 5320a4a77718SDaniel Axtens 5321a4a77718SDaniel Axtens /** 5322a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5323a4a77718SDaniel Axtens * 5324a4a77718SDaniel Axtens * @skb: GSO skb 5325a4a77718SDaniel Axtens * 5326a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5327a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5328a4a77718SDaniel Axtens * 5329a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5330a4a77718SDaniel Axtens */ 5331a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5332a4a77718SDaniel Axtens { 5333a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5334a4a77718SDaniel Axtens skb_network_header(skb); 5335a4a77718SDaniel Axtens 5336a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5337a4a77718SDaniel Axtens } 5338a4a77718SDaniel Axtens 5339a4a77718SDaniel Axtens /** 5340a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5341a4a77718SDaniel Axtens * 5342a4a77718SDaniel Axtens * @skb: GSO skb 5343a4a77718SDaniel Axtens * 5344a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5345a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5346a4a77718SDaniel Axtens * headers (TCP/UDP). 5347a4a77718SDaniel Axtens */ 5348a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5349a4a77718SDaniel Axtens { 5350a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5351a4a77718SDaniel Axtens 5352a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5353a4a77718SDaniel Axtens } 53540d5501c1SVlad Yasevich 5355ae7ef81eSMarcelo Ricardo Leitner /** 53562b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 53572b16f048SDaniel Axtens * 53582b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 53592b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 53602b16f048SDaniel Axtens * 53612b16f048SDaniel Axtens * We might want to check: 53622b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 53632b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 53642b16f048SDaniel Axtens * 53652b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 53662b16f048SDaniel Axtens * 536749682bfaSMathieu Malaterre * @skb: GSO skb 536849682bfaSMathieu Malaterre * 53692b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 53702b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 53712b16f048SDaniel Axtens * 53722b16f048SDaniel Axtens * @max_len: The maximum permissible length. 53732b16f048SDaniel Axtens * 53742b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 53752b16f048SDaniel Axtens */ 53762b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 53772b16f048SDaniel Axtens unsigned int seg_len, 53782b16f048SDaniel Axtens unsigned int max_len) { 53792b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 53802b16f048SDaniel Axtens const struct sk_buff *iter; 53812b16f048SDaniel Axtens 53822b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 53832b16f048SDaniel Axtens return seg_len <= max_len; 53842b16f048SDaniel Axtens 53852b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 53862b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 53872b16f048SDaniel Axtens 53882b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 53892b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 53902b16f048SDaniel Axtens return false; 53912b16f048SDaniel Axtens } 53922b16f048SDaniel Axtens 53932b16f048SDaniel Axtens return true; 53942b16f048SDaniel Axtens } 53952b16f048SDaniel Axtens 53962b16f048SDaniel Axtens /** 5397779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5398ae7ef81eSMarcelo Ricardo Leitner * 5399ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 540076f21b99SDavid S. Miller * @mtu: MTU to validate against 5401ae7ef81eSMarcelo Ricardo Leitner * 5402779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5403779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5404779b7931SDaniel Axtens * payload. 5405ae7ef81eSMarcelo Ricardo Leitner */ 5406779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5407ae7ef81eSMarcelo Ricardo Leitner { 54082b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5409ae7ef81eSMarcelo Ricardo Leitner } 5410779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5411ae7ef81eSMarcelo Ricardo Leitner 54122b16f048SDaniel Axtens /** 54132b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 54142b16f048SDaniel Axtens * 54152b16f048SDaniel Axtens * @skb: GSO skb 54162b16f048SDaniel Axtens * @len: length to validate against 54172b16f048SDaniel Axtens * 54182b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 54192b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 54202b16f048SDaniel Axtens */ 54212b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 54222b16f048SDaniel Axtens { 54232b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 54242b16f048SDaniel Axtens } 54252b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 54262b16f048SDaniel Axtens 54270d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 54280d5501c1SVlad Yasevich { 5429d85e8be2SYuya Kusakabe int mac_len, meta_len; 5430d85e8be2SYuya Kusakabe void *meta; 54314bbb3e0eSToshiaki Makita 54320d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 54330d5501c1SVlad Yasevich kfree_skb(skb); 54340d5501c1SVlad Yasevich return NULL; 54350d5501c1SVlad Yasevich } 54360d5501c1SVlad Yasevich 54374bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5438ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 54394bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 54404bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5441ae474573SToshiaki Makita } 5442d85e8be2SYuya Kusakabe 5443d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5444d85e8be2SYuya Kusakabe if (meta_len) { 5445d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5446d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5447d85e8be2SYuya Kusakabe } 5448d85e8be2SYuya Kusakabe 54490d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 54500d5501c1SVlad Yasevich return skb; 54510d5501c1SVlad Yasevich } 54520d5501c1SVlad Yasevich 54530d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 54540d5501c1SVlad Yasevich { 54550d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 54560d5501c1SVlad Yasevich u16 vlan_tci; 54570d5501c1SVlad Yasevich 5458df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 54590d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 54600d5501c1SVlad Yasevich return skb; 54610d5501c1SVlad Yasevich } 54620d5501c1SVlad Yasevich 54630d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 54640d5501c1SVlad Yasevich if (unlikely(!skb)) 54650d5501c1SVlad Yasevich goto err_free; 546655eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 546755eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 54680d5501c1SVlad Yasevich goto err_free; 54690d5501c1SVlad Yasevich 54700d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 54710d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 54720d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 54730d5501c1SVlad Yasevich 54740d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 54750d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 54760d5501c1SVlad Yasevich 54770d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 54780d5501c1SVlad Yasevich if (unlikely(!skb)) 54790d5501c1SVlad Yasevich goto err_free; 54800d5501c1SVlad Yasevich 54810d5501c1SVlad Yasevich skb_reset_network_header(skb); 54828be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 54830d5501c1SVlad Yasevich skb_reset_transport_header(skb); 54840d5501c1SVlad Yasevich skb_reset_mac_len(skb); 54850d5501c1SVlad Yasevich 54860d5501c1SVlad Yasevich return skb; 54870d5501c1SVlad Yasevich 54880d5501c1SVlad Yasevich err_free: 54890d5501c1SVlad Yasevich kfree_skb(skb); 54900d5501c1SVlad Yasevich return NULL; 54910d5501c1SVlad Yasevich } 54920d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 54932e4e4410SEric Dumazet 5494e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5495e2195121SJiri Pirko { 5496e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5497e2195121SJiri Pirko return -ENOMEM; 5498e2195121SJiri Pirko 5499e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5500e2195121SJiri Pirko return 0; 5501e2195121SJiri Pirko 5502e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5503e2195121SJiri Pirko } 5504e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5505e2195121SJiri Pirko 5506bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5507bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5508bfca4c52SShmulik Ladkani */ 5509bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 551093515d53SJiri Pirko { 551193515d53SJiri Pirko struct vlan_hdr *vhdr; 5512b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 551393515d53SJiri Pirko int err; 551493515d53SJiri Pirko 5515b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5516b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5517b6a79208SShmulik Ladkani offset)) { 5518b6a79208SShmulik Ladkani return -EINVAL; 5519b6a79208SShmulik Ladkani } 5520b6a79208SShmulik Ladkani 552193515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 552293515d53SJiri Pirko if (unlikely(err)) 5523b6a79208SShmulik Ladkani return err; 552493515d53SJiri Pirko 552593515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 552693515d53SJiri Pirko 552793515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 552893515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 552993515d53SJiri Pirko 553093515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 553193515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 553293515d53SJiri Pirko 553393515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 553493515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 553593515d53SJiri Pirko 553693515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 553793515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 553893515d53SJiri Pirko 553993515d53SJiri Pirko skb_reset_mac_len(skb); 554093515d53SJiri Pirko 554193515d53SJiri Pirko return err; 554293515d53SJiri Pirko } 5543bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 554493515d53SJiri Pirko 5545b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5546b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5547b6a79208SShmulik Ladkani */ 554893515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 554993515d53SJiri Pirko { 555093515d53SJiri Pirko u16 vlan_tci; 555193515d53SJiri Pirko __be16 vlan_proto; 555293515d53SJiri Pirko int err; 555393515d53SJiri Pirko 5554df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5555b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 555693515d53SJiri Pirko } else { 5557ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 555893515d53SJiri Pirko return 0; 555993515d53SJiri Pirko 556093515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 556193515d53SJiri Pirko if (err) 556293515d53SJiri Pirko return err; 556393515d53SJiri Pirko } 556493515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5565ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 556693515d53SJiri Pirko return 0; 556793515d53SJiri Pirko 556893515d53SJiri Pirko vlan_proto = skb->protocol; 556993515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 557093515d53SJiri Pirko if (unlikely(err)) 557193515d53SJiri Pirko return err; 557293515d53SJiri Pirko 557393515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 557493515d53SJiri Pirko return 0; 557593515d53SJiri Pirko } 557693515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 557793515d53SJiri Pirko 5578b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5579b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5580b6a79208SShmulik Ladkani */ 558193515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 558293515d53SJiri Pirko { 5583df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5584b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 558593515d53SJiri Pirko int err; 558693515d53SJiri Pirko 5587b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5588b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5589b6a79208SShmulik Ladkani offset)) { 5590b6a79208SShmulik Ladkani return -EINVAL; 5591b6a79208SShmulik Ladkani } 5592b6a79208SShmulik Ladkani 559393515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5594df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5595b6a79208SShmulik Ladkani if (err) 559693515d53SJiri Pirko return err; 55979241e2dfSDaniel Borkmann 559893515d53SJiri Pirko skb->protocol = skb->vlan_proto; 559993515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 560093515d53SJiri Pirko 56016b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 560293515d53SJiri Pirko } 560393515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 560493515d53SJiri Pirko return 0; 560593515d53SJiri Pirko } 560693515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 560793515d53SJiri Pirko 560819fbcb36SGuillaume Nault /** 560919fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 561019fbcb36SGuillaume Nault * 561119fbcb36SGuillaume Nault * @skb: Socket buffer to modify 561219fbcb36SGuillaume Nault * 561319fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 561419fbcb36SGuillaume Nault * 561519fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 561619fbcb36SGuillaume Nault * present. 561719fbcb36SGuillaume Nault * 561819fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 561919fbcb36SGuillaume Nault */ 562019fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 562119fbcb36SGuillaume Nault { 562219fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 562319fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 562419fbcb36SGuillaume Nault return -EPROTO; 562519fbcb36SGuillaume Nault 562619fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 562719fbcb36SGuillaume Nault skb_reset_mac_header(skb); 562819fbcb36SGuillaume Nault skb_reset_mac_len(skb); 562919fbcb36SGuillaume Nault 563019fbcb36SGuillaume Nault return 0; 563119fbcb36SGuillaume Nault } 563219fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 563319fbcb36SGuillaume Nault 563419fbcb36SGuillaume Nault /** 563519fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 563619fbcb36SGuillaume Nault * 563719fbcb36SGuillaume Nault * @skb: Socket buffer to modify 563819fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 563919fbcb36SGuillaume Nault * @src: Source MAC address of the new header 564019fbcb36SGuillaume Nault * 564119fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 564219fbcb36SGuillaume Nault * 564319fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 564419fbcb36SGuillaume Nault * 564519fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 564619fbcb36SGuillaume Nault */ 564719fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 564819fbcb36SGuillaume Nault const unsigned char *src) 564919fbcb36SGuillaume Nault { 565019fbcb36SGuillaume Nault struct ethhdr *eth; 565119fbcb36SGuillaume Nault int err; 565219fbcb36SGuillaume Nault 565319fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 565419fbcb36SGuillaume Nault return -EPROTO; 565519fbcb36SGuillaume Nault 565619fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 565719fbcb36SGuillaume Nault if (err < 0) 565819fbcb36SGuillaume Nault return err; 565919fbcb36SGuillaume Nault 566019fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 566119fbcb36SGuillaume Nault skb_reset_mac_header(skb); 566219fbcb36SGuillaume Nault skb_reset_mac_len(skb); 566319fbcb36SGuillaume Nault 566419fbcb36SGuillaume Nault eth = eth_hdr(skb); 566519fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 566619fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 566719fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 566819fbcb36SGuillaume Nault 566919fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 567019fbcb36SGuillaume Nault 567119fbcb36SGuillaume Nault return 0; 567219fbcb36SGuillaume Nault } 567319fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 567419fbcb36SGuillaume Nault 56758822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 56768822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 56778822e270SJohn Hurley __be16 ethertype) 56788822e270SJohn Hurley { 56798822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 56808822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 56818822e270SJohn Hurley 56828822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 56838822e270SJohn Hurley } 56848822e270SJohn Hurley 56858822e270SJohn Hurley hdr->h_proto = ethertype; 56868822e270SJohn Hurley } 56878822e270SJohn Hurley 56888822e270SJohn Hurley /** 5689e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5690e7dbfed1SMartin Varghese * the packet 56918822e270SJohn Hurley * 56928822e270SJohn Hurley * @skb: buffer 56938822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 56948822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5695fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5696e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5697e7dbfed1SMartin Varghese * ethernet 56988822e270SJohn Hurley * 56998822e270SJohn Hurley * Expects skb->data at mac header. 57008822e270SJohn Hurley * 57018822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 57028822e270SJohn Hurley */ 5703fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5704d04ac224SMartin Varghese int mac_len, bool ethernet) 57058822e270SJohn Hurley { 57068822e270SJohn Hurley struct mpls_shim_hdr *lse; 57078822e270SJohn Hurley int err; 57088822e270SJohn Hurley 57098822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 57108822e270SJohn Hurley return -EINVAL; 57118822e270SJohn Hurley 57128822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 57138822e270SJohn Hurley if (skb->encapsulation) 57148822e270SJohn Hurley return -EINVAL; 57158822e270SJohn Hurley 57168822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 57178822e270SJohn Hurley if (unlikely(err)) 57188822e270SJohn Hurley return err; 57198822e270SJohn Hurley 57208822e270SJohn Hurley if (!skb->inner_protocol) { 5721e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 57228822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 57238822e270SJohn Hurley } 57248822e270SJohn Hurley 57258822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 57268822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5727fa4e0f88SDavide Caratti mac_len); 57288822e270SJohn Hurley skb_reset_mac_header(skb); 5729fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5730e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 57318822e270SJohn Hurley 57328822e270SJohn Hurley lse = mpls_hdr(skb); 57338822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 57348822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 57358822e270SJohn Hurley 57364296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 57378822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 57388822e270SJohn Hurley skb->protocol = mpls_proto; 57398822e270SJohn Hurley 57408822e270SJohn Hurley return 0; 57418822e270SJohn Hurley } 57428822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 57438822e270SJohn Hurley 57442e4e4410SEric Dumazet /** 5745ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5746ed246ceeSJohn Hurley * 5747ed246ceeSJohn Hurley * @skb: buffer 5748ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5749fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 575076f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5751ed246ceeSJohn Hurley * 5752ed246ceeSJohn Hurley * Expects skb->data at mac header. 5753ed246ceeSJohn Hurley * 5754ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5755ed246ceeSJohn Hurley */ 5756040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5757040b5cfbSMartin Varghese bool ethernet) 5758ed246ceeSJohn Hurley { 5759ed246ceeSJohn Hurley int err; 5760ed246ceeSJohn Hurley 5761ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5762dedc5a08SDavide Caratti return 0; 5763ed246ceeSJohn Hurley 5764fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5765ed246ceeSJohn Hurley if (unlikely(err)) 5766ed246ceeSJohn Hurley return err; 5767ed246ceeSJohn Hurley 5768ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5769ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5770fa4e0f88SDavide Caratti mac_len); 5771ed246ceeSJohn Hurley 5772ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5773ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5774fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5775ed246ceeSJohn Hurley 57764296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5777ed246ceeSJohn Hurley struct ethhdr *hdr; 5778ed246ceeSJohn Hurley 5779ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5780ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5781ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5782ed246ceeSJohn Hurley } 5783ed246ceeSJohn Hurley skb->protocol = next_proto; 5784ed246ceeSJohn Hurley 5785ed246ceeSJohn Hurley return 0; 5786ed246ceeSJohn Hurley } 5787ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5788ed246ceeSJohn Hurley 5789ed246ceeSJohn Hurley /** 5790d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5791d27cf5c5SJohn Hurley * 5792d27cf5c5SJohn Hurley * @skb: buffer 5793d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5794d27cf5c5SJohn Hurley * 5795d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5796d27cf5c5SJohn Hurley * 5797d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5798d27cf5c5SJohn Hurley */ 5799d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5800d27cf5c5SJohn Hurley { 5801d27cf5c5SJohn Hurley int err; 5802d27cf5c5SJohn Hurley 5803d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5804d27cf5c5SJohn Hurley return -EINVAL; 5805d27cf5c5SJohn Hurley 5806d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5807d27cf5c5SJohn Hurley if (unlikely(err)) 5808d27cf5c5SJohn Hurley return err; 5809d27cf5c5SJohn Hurley 5810d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5811d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5812d27cf5c5SJohn Hurley 5813d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5814d27cf5c5SJohn Hurley } 5815d27cf5c5SJohn Hurley 5816d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 5817d27cf5c5SJohn Hurley 5818d27cf5c5SJohn Hurley return 0; 5819d27cf5c5SJohn Hurley } 5820d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 5821d27cf5c5SJohn Hurley 5822d27cf5c5SJohn Hurley /** 58232a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 58242a2ea508SJohn Hurley * 58252a2ea508SJohn Hurley * @skb: buffer 58262a2ea508SJohn Hurley * 58272a2ea508SJohn Hurley * Expects skb->data at mac header. 58282a2ea508SJohn Hurley * 58292a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 58302a2ea508SJohn Hurley */ 58312a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 58322a2ea508SJohn Hurley { 58332a2ea508SJohn Hurley u32 lse; 58342a2ea508SJohn Hurley u8 ttl; 58352a2ea508SJohn Hurley 58362a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 58372a2ea508SJohn Hurley return -EINVAL; 58382a2ea508SJohn Hurley 583913de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 584013de4ed9SDavide Caratti return -ENOMEM; 584113de4ed9SDavide Caratti 58422a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 58432a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 58442a2ea508SJohn Hurley if (!--ttl) 58452a2ea508SJohn Hurley return -EINVAL; 58462a2ea508SJohn Hurley 58472a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 58482a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 58492a2ea508SJohn Hurley 58502a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 58512a2ea508SJohn Hurley } 58522a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 58532a2ea508SJohn Hurley 58542a2ea508SJohn Hurley /** 58552e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 58562e4e4410SEric Dumazet * 5857de3f0d0eSMasanari Iida * @header_len: size of linear part 5858de3f0d0eSMasanari Iida * @data_len: needed length in frags 5859de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5860de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5861de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 58622e4e4410SEric Dumazet * 58632e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 58642e4e4410SEric Dumazet */ 58652e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 58662e4e4410SEric Dumazet unsigned long data_len, 58672e4e4410SEric Dumazet int max_page_order, 58682e4e4410SEric Dumazet int *errcode, 58692e4e4410SEric Dumazet gfp_t gfp_mask) 58702e4e4410SEric Dumazet { 58712e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 58722e4e4410SEric Dumazet unsigned long chunk; 58732e4e4410SEric Dumazet struct sk_buff *skb; 58742e4e4410SEric Dumazet struct page *page; 58752e4e4410SEric Dumazet int i; 58762e4e4410SEric Dumazet 58772e4e4410SEric Dumazet *errcode = -EMSGSIZE; 58782e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 58792e4e4410SEric Dumazet * high order pages... 58802e4e4410SEric Dumazet */ 58812e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 58822e4e4410SEric Dumazet return NULL; 58832e4e4410SEric Dumazet 58842e4e4410SEric Dumazet *errcode = -ENOBUFS; 5885f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 58862e4e4410SEric Dumazet if (!skb) 58872e4e4410SEric Dumazet return NULL; 58882e4e4410SEric Dumazet 58892e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 58902e4e4410SEric Dumazet 58912e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 58922e4e4410SEric Dumazet int order = max_page_order; 58932e4e4410SEric Dumazet 58942e4e4410SEric Dumazet while (order) { 58952e4e4410SEric Dumazet if (npages >= 1 << order) { 5896d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 58972e4e4410SEric Dumazet __GFP_COMP | 5898d14b56f5SMichal Hocko __GFP_NOWARN, 58992e4e4410SEric Dumazet order); 59002e4e4410SEric Dumazet if (page) 59012e4e4410SEric Dumazet goto fill_page; 59022e4e4410SEric Dumazet /* Do not retry other high order allocations */ 59032e4e4410SEric Dumazet order = 1; 59042e4e4410SEric Dumazet max_page_order = 0; 59052e4e4410SEric Dumazet } 59062e4e4410SEric Dumazet order--; 59072e4e4410SEric Dumazet } 59082e4e4410SEric Dumazet page = alloc_page(gfp_mask); 59092e4e4410SEric Dumazet if (!page) 59102e4e4410SEric Dumazet goto failure; 59112e4e4410SEric Dumazet fill_page: 59122e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 59132e4e4410SEric Dumazet PAGE_SIZE << order); 59142e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 59152e4e4410SEric Dumazet data_len -= chunk; 59162e4e4410SEric Dumazet npages -= 1 << order; 59172e4e4410SEric Dumazet } 59182e4e4410SEric Dumazet return skb; 59192e4e4410SEric Dumazet 59202e4e4410SEric Dumazet failure: 59212e4e4410SEric Dumazet kfree_skb(skb); 59222e4e4410SEric Dumazet return NULL; 59232e4e4410SEric Dumazet } 59242e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 59256fa01ccdSSowmini Varadhan 59266fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 59276fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 59286fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 59296fa01ccdSSowmini Varadhan { 59306fa01ccdSSowmini Varadhan int i; 59316fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 59326fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 59336fa01ccdSSowmini Varadhan u8 *data; 59346fa01ccdSSowmini Varadhan 59356fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 59366fa01ccdSSowmini Varadhan 59376fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 59386fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 59396fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 59406fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 59416fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 59426fa01ccdSSowmini Varadhan if (!data) 59436fa01ccdSSowmini Varadhan return -ENOMEM; 59446fa01ccdSSowmini Varadhan 59456fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 59466fa01ccdSSowmini Varadhan 59476fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 59486fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 59496fa01ccdSSowmini Varadhan skb->len -= off; 59506fa01ccdSSowmini Varadhan 59516fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 59526fa01ccdSSowmini Varadhan skb_shinfo(skb), 59536fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 59546fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 59556fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 59566fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 59576fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 59586fa01ccdSSowmini Varadhan kfree(data); 59596fa01ccdSSowmini Varadhan return -ENOMEM; 59606fa01ccdSSowmini Varadhan } 59616fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 59626fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 59636fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 59646fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 59656fa01ccdSSowmini Varadhan skb_release_data(skb); 59666fa01ccdSSowmini Varadhan } else { 59676fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 59686fa01ccdSSowmini Varadhan * relocate values 59696fa01ccdSSowmini Varadhan */ 59706fa01ccdSSowmini Varadhan skb_free_head(skb); 59716fa01ccdSSowmini Varadhan } 59726fa01ccdSSowmini Varadhan 59736fa01ccdSSowmini Varadhan skb->head = data; 59746fa01ccdSSowmini Varadhan skb->data = data; 59756fa01ccdSSowmini Varadhan skb->head_frag = 0; 59766fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 59776fa01ccdSSowmini Varadhan skb->end = size; 59786fa01ccdSSowmini Varadhan #else 59796fa01ccdSSowmini Varadhan skb->end = skb->head + size; 59806fa01ccdSSowmini Varadhan #endif 59816fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 59826fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 59836fa01ccdSSowmini Varadhan skb->cloned = 0; 59846fa01ccdSSowmini Varadhan skb->hdr_len = 0; 59856fa01ccdSSowmini Varadhan skb->nohdr = 0; 59866fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 59876fa01ccdSSowmini Varadhan 59886fa01ccdSSowmini Varadhan return 0; 59896fa01ccdSSowmini Varadhan } 59906fa01ccdSSowmini Varadhan 59916fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 59926fa01ccdSSowmini Varadhan 59936fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 59946fa01ccdSSowmini Varadhan * pskb_carve() 59956fa01ccdSSowmini Varadhan */ 59966fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 59976fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 59986fa01ccdSSowmini Varadhan gfp_t gfp_mask) 59996fa01ccdSSowmini Varadhan { 60006fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 60016fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 60026fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 60036fa01ccdSSowmini Varadhan 60046fa01ccdSSowmini Varadhan do { 60056fa01ccdSSowmini Varadhan if (!list) { 60066fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 60076fa01ccdSSowmini Varadhan return -EFAULT; 60086fa01ccdSSowmini Varadhan } 60096fa01ccdSSowmini Varadhan if (list->len <= eat) { 60106fa01ccdSSowmini Varadhan /* Eaten as whole. */ 60116fa01ccdSSowmini Varadhan eat -= list->len; 60126fa01ccdSSowmini Varadhan list = list->next; 60136fa01ccdSSowmini Varadhan insp = list; 60146fa01ccdSSowmini Varadhan } else { 60156fa01ccdSSowmini Varadhan /* Eaten partially. */ 60166fa01ccdSSowmini Varadhan if (skb_shared(list)) { 60176fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 60186fa01ccdSSowmini Varadhan if (!clone) 60196fa01ccdSSowmini Varadhan return -ENOMEM; 60206fa01ccdSSowmini Varadhan insp = list->next; 60216fa01ccdSSowmini Varadhan list = clone; 60226fa01ccdSSowmini Varadhan } else { 60236fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 60246fa01ccdSSowmini Varadhan insp = list; 60256fa01ccdSSowmini Varadhan } 60266fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 60276fa01ccdSSowmini Varadhan kfree_skb(clone); 60286fa01ccdSSowmini Varadhan return -ENOMEM; 60296fa01ccdSSowmini Varadhan } 60306fa01ccdSSowmini Varadhan break; 60316fa01ccdSSowmini Varadhan } 60326fa01ccdSSowmini Varadhan } while (eat); 60336fa01ccdSSowmini Varadhan 60346fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 60356fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 60366fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 60376fa01ccdSSowmini Varadhan kfree_skb(list); 60386fa01ccdSSowmini Varadhan } 60396fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 60406fa01ccdSSowmini Varadhan if (clone) { 60416fa01ccdSSowmini Varadhan clone->next = list; 60426fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 60436fa01ccdSSowmini Varadhan } 60446fa01ccdSSowmini Varadhan return 0; 60456fa01ccdSSowmini Varadhan } 60466fa01ccdSSowmini Varadhan 60476fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 60486fa01ccdSSowmini Varadhan * non-linear part of skb 60496fa01ccdSSowmini Varadhan */ 60506fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 60516fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 60526fa01ccdSSowmini Varadhan { 60536fa01ccdSSowmini Varadhan int i, k = 0; 60546fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 60556fa01ccdSSowmini Varadhan u8 *data; 60566fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 60576fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 60586fa01ccdSSowmini Varadhan 60596fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 60606fa01ccdSSowmini Varadhan 60616fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 60626fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 60636fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 60646fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 60656fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 60666fa01ccdSSowmini Varadhan if (!data) 60676fa01ccdSSowmini Varadhan return -ENOMEM; 60686fa01ccdSSowmini Varadhan 60696fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 60706fa01ccdSSowmini Varadhan 60716fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6072e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 60736fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 60746fa01ccdSSowmini Varadhan kfree(data); 60756fa01ccdSSowmini Varadhan return -ENOMEM; 60766fa01ccdSSowmini Varadhan } 60776fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 60786fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 60796fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 60806fa01ccdSSowmini Varadhan 60816fa01ccdSSowmini Varadhan if (pos + fsize > off) { 60826fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 60836fa01ccdSSowmini Varadhan 60846fa01ccdSSowmini Varadhan if (pos < off) { 60856fa01ccdSSowmini Varadhan /* Split frag. 60866fa01ccdSSowmini Varadhan * We have two variants in this case: 60876fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 60886fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 60896fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 60906fa01ccdSSowmini Varadhan * where splitting is expensive. 60916fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 60926fa01ccdSSowmini Varadhan */ 6093b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 60946fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 60956fa01ccdSSowmini Varadhan } 60966fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 60976fa01ccdSSowmini Varadhan k++; 60986fa01ccdSSowmini Varadhan } 60996fa01ccdSSowmini Varadhan pos += fsize; 61006fa01ccdSSowmini Varadhan } 61016fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 61026fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 61036fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 61046fa01ccdSSowmini Varadhan 61056fa01ccdSSowmini Varadhan /* split line is in frag list */ 6106eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6107eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6108eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6109eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6110eabe8618SMiaohe Lin kfree(data); 6111eabe8618SMiaohe Lin return -ENOMEM; 61126fa01ccdSSowmini Varadhan } 61136fa01ccdSSowmini Varadhan skb_release_data(skb); 61146fa01ccdSSowmini Varadhan 61156fa01ccdSSowmini Varadhan skb->head = data; 61166fa01ccdSSowmini Varadhan skb->head_frag = 0; 61176fa01ccdSSowmini Varadhan skb->data = data; 61186fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 61196fa01ccdSSowmini Varadhan skb->end = size; 61206fa01ccdSSowmini Varadhan #else 61216fa01ccdSSowmini Varadhan skb->end = skb->head + size; 61226fa01ccdSSowmini Varadhan #endif 61236fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 61246fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 61256fa01ccdSSowmini Varadhan skb->cloned = 0; 61266fa01ccdSSowmini Varadhan skb->hdr_len = 0; 61276fa01ccdSSowmini Varadhan skb->nohdr = 0; 61286fa01ccdSSowmini Varadhan skb->len -= off; 61296fa01ccdSSowmini Varadhan skb->data_len = skb->len; 61306fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 61316fa01ccdSSowmini Varadhan return 0; 61326fa01ccdSSowmini Varadhan } 61336fa01ccdSSowmini Varadhan 61346fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 61356fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 61366fa01ccdSSowmini Varadhan { 61376fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 61386fa01ccdSSowmini Varadhan 61396fa01ccdSSowmini Varadhan if (len < headlen) 61406fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 61416fa01ccdSSowmini Varadhan else 61426fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 61436fa01ccdSSowmini Varadhan } 61446fa01ccdSSowmini Varadhan 61456fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 61466fa01ccdSSowmini Varadhan * a new skb 61476fa01ccdSSowmini Varadhan */ 61486fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 61496fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 61506fa01ccdSSowmini Varadhan { 61516fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 61526fa01ccdSSowmini Varadhan 61536fa01ccdSSowmini Varadhan if (!clone) 61546fa01ccdSSowmini Varadhan return NULL; 61556fa01ccdSSowmini Varadhan 61566fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 61576fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 61586fa01ccdSSowmini Varadhan kfree_skb(clone); 61596fa01ccdSSowmini Varadhan return NULL; 61606fa01ccdSSowmini Varadhan } 61616fa01ccdSSowmini Varadhan return clone; 61626fa01ccdSSowmini Varadhan } 61636fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6164c8c8b127SEric Dumazet 6165c8c8b127SEric Dumazet /** 6166c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6167c8c8b127SEric Dumazet * @skb: buffer 6168c8c8b127SEric Dumazet * 6169c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6170c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6171c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6172c8c8b127SEric Dumazet * truesize. 6173c8c8b127SEric Dumazet * Notes: 6174c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6175c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6176c8c8b127SEric Dumazet */ 6177c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6178c8c8b127SEric Dumazet { 61793174fed9SEric Dumazet if (skb->data_len) { 61803174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6181c8c8b127SEric Dumazet skb_cloned(skb)) 6182c8c8b127SEric Dumazet return; 6183c8c8b127SEric Dumazet 6184c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6185c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 61863174fed9SEric Dumazet } 61873174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 61883174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 61893174fed9SEric Dumazet * their truesize. 61903174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 61913174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 61923174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6193c8c8b127SEric Dumazet */ 6194c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6195c8c8b127SEric Dumazet } 6196df5042f4SFlorian Westphal 6197df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6198df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6199df5042f4SFlorian Westphal { 6200df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6201df5042f4SFlorian Westphal } 6202df5042f4SFlorian Westphal 62038b69a803SPaolo Abeni /** 62048b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 62058b69a803SPaolo Abeni * 62064930f483SFlorian Westphal * @flags: See kmalloc(). 62074930f483SFlorian Westphal * 62088b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 62098b69a803SPaolo Abeni * skb via __skb_ext_set(). 62108b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 62118b69a803SPaolo Abeni */ 62124930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6213df5042f4SFlorian Westphal { 62144930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6215df5042f4SFlorian Westphal 6216df5042f4SFlorian Westphal if (new) { 6217df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6218df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6219df5042f4SFlorian Westphal } 6220df5042f4SFlorian Westphal 6221df5042f4SFlorian Westphal return new; 6222df5042f4SFlorian Westphal } 6223df5042f4SFlorian Westphal 62244165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 62254165079bSFlorian Westphal unsigned int old_active) 6226df5042f4SFlorian Westphal { 6227df5042f4SFlorian Westphal struct skb_ext *new; 6228df5042f4SFlorian Westphal 6229df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6230df5042f4SFlorian Westphal return old; 6231df5042f4SFlorian Westphal 6232df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6233df5042f4SFlorian Westphal if (!new) 6234df5042f4SFlorian Westphal return NULL; 6235df5042f4SFlorian Westphal 6236df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6237df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6238df5042f4SFlorian Westphal 62394165079bSFlorian Westphal #ifdef CONFIG_XFRM 62404165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 62414165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 62424165079bSFlorian Westphal unsigned int i; 62434165079bSFlorian Westphal 62444165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 62454165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 62464165079bSFlorian Westphal } 62474165079bSFlorian Westphal #endif 6248df5042f4SFlorian Westphal __skb_ext_put(old); 6249df5042f4SFlorian Westphal return new; 6250df5042f4SFlorian Westphal } 6251df5042f4SFlorian Westphal 6252df5042f4SFlorian Westphal /** 62538b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 62548b69a803SPaolo Abeni * @skb: buffer 62558b69a803SPaolo Abeni * @id: extension id 62568b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 62578b69a803SPaolo Abeni * 62588b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 62598b69a803SPaolo Abeni * 62608b69a803SPaolo Abeni * Returns the pointer to the extension. 62618b69a803SPaolo Abeni */ 62628b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 62638b69a803SPaolo Abeni struct skb_ext *ext) 62648b69a803SPaolo Abeni { 62658b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 62668b69a803SPaolo Abeni 62678b69a803SPaolo Abeni skb_ext_put(skb); 62688b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 62698b69a803SPaolo Abeni ext->chunks = newlen; 62708b69a803SPaolo Abeni ext->offset[id] = newoff; 62718b69a803SPaolo Abeni skb->extensions = ext; 62728b69a803SPaolo Abeni skb->active_extensions = 1 << id; 62738b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 62748b69a803SPaolo Abeni } 62758b69a803SPaolo Abeni 62768b69a803SPaolo Abeni /** 6277df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6278df5042f4SFlorian Westphal * @skb: buffer 6279df5042f4SFlorian Westphal * @id: extension to allocate space for 6280df5042f4SFlorian Westphal * 6281df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6282df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6283df5042f4SFlorian Westphal * is returned. 6284df5042f4SFlorian Westphal * 6285df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6286df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6287df5042f4SFlorian Westphal * 6288df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6289df5042f4SFlorian Westphal */ 6290df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6291df5042f4SFlorian Westphal { 6292df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6293df5042f4SFlorian Westphal unsigned int newlen, newoff; 6294df5042f4SFlorian Westphal 6295df5042f4SFlorian Westphal if (skb->active_extensions) { 6296df5042f4SFlorian Westphal old = skb->extensions; 6297df5042f4SFlorian Westphal 62984165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6299df5042f4SFlorian Westphal if (!new) 6300df5042f4SFlorian Westphal return NULL; 6301df5042f4SFlorian Westphal 6302682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6303df5042f4SFlorian Westphal goto set_active; 6304df5042f4SFlorian Westphal 6305e94e50bdSPaolo Abeni newoff = new->chunks; 6306df5042f4SFlorian Westphal } else { 6307df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6308df5042f4SFlorian Westphal 63094930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6310df5042f4SFlorian Westphal if (!new) 6311df5042f4SFlorian Westphal return NULL; 6312df5042f4SFlorian Westphal } 6313df5042f4SFlorian Westphal 6314df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6315df5042f4SFlorian Westphal new->chunks = newlen; 6316df5042f4SFlorian Westphal new->offset[id] = newoff; 6317df5042f4SFlorian Westphal set_active: 6318682ec859SPaolo Abeni skb->extensions = new; 6319df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6320df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6321df5042f4SFlorian Westphal } 6322df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6323df5042f4SFlorian Westphal 63244165079bSFlorian Westphal #ifdef CONFIG_XFRM 63254165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 63264165079bSFlorian Westphal { 63274165079bSFlorian Westphal unsigned int i; 63284165079bSFlorian Westphal 63294165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 63304165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 63314165079bSFlorian Westphal } 63324165079bSFlorian Westphal #endif 63334165079bSFlorian Westphal 6334df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6335df5042f4SFlorian Westphal { 6336df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6337df5042f4SFlorian Westphal 6338df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6339df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6340df5042f4SFlorian Westphal skb->extensions = NULL; 6341df5042f4SFlorian Westphal __skb_ext_put(ext); 63424165079bSFlorian Westphal #ifdef CONFIG_XFRM 63434165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 63444165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 63454165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 63464165079bSFlorian Westphal 63474165079bSFlorian Westphal skb_ext_put_sp(sp); 63484165079bSFlorian Westphal sp->len = 0; 63494165079bSFlorian Westphal #endif 6350df5042f4SFlorian Westphal } 6351df5042f4SFlorian Westphal } 6352df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6353df5042f4SFlorian Westphal 6354df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6355df5042f4SFlorian Westphal { 6356df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6357df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6358df5042f4SFlorian Westphal */ 6359df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6360df5042f4SFlorian Westphal goto free_now; 6361df5042f4SFlorian Westphal 6362df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6363df5042f4SFlorian Westphal return; 6364df5042f4SFlorian Westphal free_now: 63654165079bSFlorian Westphal #ifdef CONFIG_XFRM 63664165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 63674165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 63684165079bSFlorian Westphal #endif 63694165079bSFlorian Westphal 6370df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6371df5042f4SFlorian Westphal } 6372df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6373df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 6374