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 { 504*3226b158SEric Dumazet struct napi_alloc_cache *nc; 505fd11a83dSAlexander Duyck struct sk_buff *skb; 5069451980aSAlexander Duyck void *data; 507fd11a83dSAlexander Duyck 5089451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 509fd11a83dSAlexander Duyck 510*3226b158SEric Dumazet /* If requested length is either too small or too big, 511*3226b158SEric Dumazet * we use kmalloc() for skb->head allocation. 512*3226b158SEric Dumazet */ 513*3226b158SEric Dumazet if (len <= SKB_WITH_OVERHEAD(1024) || 514*3226b158SEric Dumazet len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 515d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 516a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 517a080e7bdSAlexander Duyck if (!skb) 518a080e7bdSAlexander Duyck goto skb_fail; 519a080e7bdSAlexander Duyck goto skb_success; 520a080e7bdSAlexander Duyck } 5219451980aSAlexander Duyck 522*3226b158SEric Dumazet nc = this_cpu_ptr(&napi_alloc_cache); 5239451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5249451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5259451980aSAlexander Duyck 5269451980aSAlexander Duyck if (sk_memalloc_socks()) 5279451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5289451980aSAlexander Duyck 5298c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5309451980aSAlexander Duyck if (unlikely(!data)) 5319451980aSAlexander Duyck return NULL; 5329451980aSAlexander Duyck 5339451980aSAlexander Duyck skb = __build_skb(data, len); 5349451980aSAlexander Duyck if (unlikely(!skb)) { 535181edb2bSAlexander Duyck skb_free_frag(data); 5369451980aSAlexander Duyck return NULL; 5379451980aSAlexander Duyck } 5389451980aSAlexander Duyck 539795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5409451980aSAlexander Duyck skb->pfmemalloc = 1; 5419451980aSAlexander Duyck skb->head_frag = 1; 5429451980aSAlexander Duyck 543a080e7bdSAlexander Duyck skb_success: 544fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 545fd11a83dSAlexander Duyck skb->dev = napi->dev; 546fd11a83dSAlexander Duyck 547a080e7bdSAlexander Duyck skb_fail: 548fd11a83dSAlexander Duyck return skb; 549fd11a83dSAlexander Duyck } 550fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 551fd11a83dSAlexander Duyck 552654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 55350269e19SEric Dumazet int size, unsigned int truesize) 554654bed16SPeter Zijlstra { 555654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 556654bed16SPeter Zijlstra skb->len += size; 557654bed16SPeter Zijlstra skb->data_len += size; 55850269e19SEric Dumazet skb->truesize += truesize; 559654bed16SPeter Zijlstra } 560654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 561654bed16SPeter Zijlstra 562f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 563f8e617e1SJason Wang unsigned int truesize) 564f8e617e1SJason Wang { 565f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 566f8e617e1SJason Wang 567f8e617e1SJason Wang skb_frag_size_add(frag, size); 568f8e617e1SJason Wang skb->len += size; 569f8e617e1SJason Wang skb->data_len += size; 570f8e617e1SJason Wang skb->truesize += truesize; 571f8e617e1SJason Wang } 572f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 573f8e617e1SJason Wang 57427b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5751da177e4SLinus Torvalds { 576bd8a7036SEric Dumazet kfree_skb_list(*listp); 57727b437c8SHerbert Xu *listp = NULL; 5781da177e4SLinus Torvalds } 5791da177e4SLinus Torvalds 58027b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 58127b437c8SHerbert Xu { 58227b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 58327b437c8SHerbert Xu } 58427b437c8SHerbert Xu 5851da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5861da177e4SLinus Torvalds { 5871da177e4SLinus Torvalds struct sk_buff *list; 5881da177e4SLinus Torvalds 589fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5901da177e4SLinus Torvalds skb_get(list); 5911da177e4SLinus Torvalds } 5921da177e4SLinus Torvalds 593d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 594d3836f21SEric Dumazet { 595181edb2bSAlexander Duyck unsigned char *head = skb->head; 596181edb2bSAlexander Duyck 597d3836f21SEric Dumazet if (skb->head_frag) 598181edb2bSAlexander Duyck skb_free_frag(head); 599d3836f21SEric Dumazet else 600181edb2bSAlexander Duyck kfree(head); 601d3836f21SEric Dumazet } 602d3836f21SEric Dumazet 6035bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 6041da177e4SLinus Torvalds { 605ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6061da177e4SLinus Torvalds int i; 607ff04a771SEric Dumazet 608ff04a771SEric Dumazet if (skb->cloned && 609ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 610ff04a771SEric Dumazet &shinfo->dataref)) 611ff04a771SEric Dumazet return; 612ff04a771SEric Dumazet 613ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 614ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 6151da177e4SLinus Torvalds 616ff04a771SEric Dumazet if (shinfo->frag_list) 617ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6181da177e4SLinus Torvalds 6191f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 620d3836f21SEric Dumazet skb_free_head(skb); 6211da177e4SLinus Torvalds } 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds /* 6241da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6251da177e4SLinus Torvalds */ 6262d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6271da177e4SLinus Torvalds { 628d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 629d179cd12SDavid S. Miller 630d179cd12SDavid S. Miller switch (skb->fclone) { 631d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6321da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6336ffe75ebSEric Dumazet return; 634d179cd12SDavid S. Miller 635d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 636d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6376ffe75ebSEric Dumazet 6386ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6396ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6406ffe75ebSEric Dumazet * while here, branch prediction will be good. 6416ffe75ebSEric Dumazet */ 6422638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 6436ffe75ebSEric Dumazet goto fastpath; 644d179cd12SDavid S. Miller break; 645d179cd12SDavid S. Miller 6466ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 647d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 648d179cd12SDavid S. Miller break; 6493ff50b79SStephen Hemminger } 6502638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 6516ffe75ebSEric Dumazet return; 6526ffe75ebSEric Dumazet fastpath: 6536ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds 6560a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 6571da177e4SLinus Torvalds { 658adf30907SEric Dumazet skb_dst_drop(skb); 6591da177e4SLinus Torvalds if (skb->destructor) { 6609c2b3328SStephen Hemminger WARN_ON(in_irq()); 6611da177e4SLinus Torvalds skb->destructor(skb); 6621da177e4SLinus Torvalds } 663a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 664cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 6652fc72c7bSKOVACS Krisztian #endif 666df5042f4SFlorian Westphal skb_ext_put(skb); 66704a4bb55SLennert Buytenhek } 66804a4bb55SLennert Buytenhek 66904a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 67004a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 67104a4bb55SLennert Buytenhek { 67204a4bb55SLennert Buytenhek skb_release_head_state(skb); 673a28b1b90SFlorian Westphal if (likely(skb->head)) 6742d4baff8SHerbert Xu skb_release_data(skb); 6752d4baff8SHerbert Xu } 6761da177e4SLinus Torvalds 6772d4baff8SHerbert Xu /** 6782d4baff8SHerbert Xu * __kfree_skb - private function 6792d4baff8SHerbert Xu * @skb: buffer 6802d4baff8SHerbert Xu * 6812d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6822d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6832d4baff8SHerbert Xu * always call kfree_skb 6842d4baff8SHerbert Xu */ 6852d4baff8SHerbert Xu 6862d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6872d4baff8SHerbert Xu { 6882d4baff8SHerbert Xu skb_release_all(skb); 6891da177e4SLinus Torvalds kfree_skbmem(skb); 6901da177e4SLinus Torvalds } 691b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6921da177e4SLinus Torvalds 6931da177e4SLinus Torvalds /** 694231d06aeSJörn Engel * kfree_skb - free an sk_buff 695231d06aeSJörn Engel * @skb: buffer to free 696231d06aeSJörn Engel * 697231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 698231d06aeSJörn Engel * hit zero. 699231d06aeSJörn Engel */ 700231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 701231d06aeSJörn Engel { 7023889a803SPaolo Abeni if (!skb_unref(skb)) 703231d06aeSJörn Engel return; 7043889a803SPaolo Abeni 705ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 706231d06aeSJörn Engel __kfree_skb(skb); 707231d06aeSJörn Engel } 708b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 709231d06aeSJörn Engel 710bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 711bd8a7036SEric Dumazet { 712bd8a7036SEric Dumazet while (segs) { 713bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 714bd8a7036SEric Dumazet 715bd8a7036SEric Dumazet kfree_skb(segs); 716bd8a7036SEric Dumazet segs = next; 717bd8a7036SEric Dumazet } 718bd8a7036SEric Dumazet } 719bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 720bd8a7036SEric Dumazet 7216413139dSWillem de Bruijn /* Dump skb information and contents. 7226413139dSWillem de Bruijn * 7236413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 7246413139dSWillem de Bruijn * 725302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 7266413139dSWillem de Bruijn */ 7276413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 7286413139dSWillem de Bruijn { 7296413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 7306413139dSWillem de Bruijn struct net_device *dev = skb->dev; 7316413139dSWillem de Bruijn struct sock *sk = skb->sk; 7326413139dSWillem de Bruijn struct sk_buff *list_skb; 7336413139dSWillem de Bruijn bool has_mac, has_trans; 7346413139dSWillem de Bruijn int headroom, tailroom; 7356413139dSWillem de Bruijn int i, len, seg_len; 7366413139dSWillem de Bruijn 7376413139dSWillem de Bruijn if (full_pkt) 7386413139dSWillem de Bruijn len = skb->len; 7396413139dSWillem de Bruijn else 7406413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 7416413139dSWillem de Bruijn 7426413139dSWillem de Bruijn headroom = skb_headroom(skb); 7436413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 7446413139dSWillem de Bruijn 7456413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 7466413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 7476413139dSWillem de Bruijn 7486413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 7496413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 7506413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 7516413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 7526413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 7536413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 7546413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 7556413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 7566413139dSWillem de Bruijn skb->network_header, 7576413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 7586413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 7596413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 7606413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 7616413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 7626413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 7636413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 7646413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 7656413139dSWillem de Bruijn 7666413139dSWillem de Bruijn if (dev) 7676413139dSWillem de Bruijn printk("%sdev name=%s feat=0x%pNF\n", 7686413139dSWillem de Bruijn level, dev->name, &dev->features); 7696413139dSWillem de Bruijn if (sk) 770db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 7716413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 7726413139dSWillem de Bruijn 7736413139dSWillem de Bruijn if (full_pkt && headroom) 7746413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 7756413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 7766413139dSWillem de Bruijn 7776413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 7786413139dSWillem de Bruijn if (seg_len) 7796413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 7806413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 7816413139dSWillem de Bruijn len -= seg_len; 7826413139dSWillem de Bruijn 7836413139dSWillem de Bruijn if (full_pkt && tailroom) 7846413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 7856413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 7866413139dSWillem de Bruijn 7876413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 7886413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 7896413139dSWillem de Bruijn u32 p_off, p_len, copied; 7906413139dSWillem de Bruijn struct page *p; 7916413139dSWillem de Bruijn u8 *vaddr; 7926413139dSWillem de Bruijn 793b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 7946413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 7956413139dSWillem de Bruijn copied) { 7966413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 7976413139dSWillem de Bruijn vaddr = kmap_atomic(p); 7986413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 7996413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 8006413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 8016413139dSWillem de Bruijn kunmap_atomic(vaddr); 8026413139dSWillem de Bruijn len -= seg_len; 8036413139dSWillem de Bruijn if (!len) 8046413139dSWillem de Bruijn break; 8056413139dSWillem de Bruijn } 8066413139dSWillem de Bruijn } 8076413139dSWillem de Bruijn 8086413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8096413139dSWillem de Bruijn printk("skb fraglist:\n"); 8106413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8116413139dSWillem de Bruijn skb_dump(level, list_skb, true); 8126413139dSWillem de Bruijn } 8136413139dSWillem de Bruijn } 8146413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 8156413139dSWillem de Bruijn 816d1a203eaSStephen Hemminger /** 81725121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 81825121173SMichael S. Tsirkin * @skb: buffer that triggered an error 81925121173SMichael S. Tsirkin * 82025121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 82125121173SMichael S. Tsirkin * skb must be freed afterwards. 82225121173SMichael S. Tsirkin */ 82325121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 82425121173SMichael S. Tsirkin { 8251f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 82625121173SMichael S. Tsirkin } 82725121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 82825121173SMichael S. Tsirkin 829be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 83025121173SMichael S. Tsirkin /** 831ead2ceb0SNeil Horman * consume_skb - free an skbuff 832ead2ceb0SNeil Horman * @skb: buffer to free 833ead2ceb0SNeil Horman * 834ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 835ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 836ead2ceb0SNeil Horman * is being dropped after a failure and notes that 837ead2ceb0SNeil Horman */ 838ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 839ead2ceb0SNeil Horman { 8403889a803SPaolo Abeni if (!skb_unref(skb)) 841ead2ceb0SNeil Horman return; 8423889a803SPaolo Abeni 84307dc22e7SKoki Sanagi trace_consume_skb(skb); 844ead2ceb0SNeil Horman __kfree_skb(skb); 845ead2ceb0SNeil Horman } 846ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 847be769db2SHerbert Xu #endif 848ead2ceb0SNeil Horman 8490a463c78SPaolo Abeni /** 850c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 8510a463c78SPaolo Abeni * @skb: buffer to free 8520a463c78SPaolo Abeni * 853ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 854ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 8550a463c78SPaolo Abeni */ 856ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 8570a463c78SPaolo Abeni { 8580a463c78SPaolo Abeni trace_consume_skb(skb); 8590a463c78SPaolo Abeni skb_release_data(skb); 8600a463c78SPaolo Abeni kfree_skbmem(skb); 8610a463c78SPaolo Abeni } 8620a463c78SPaolo Abeni 863795bb1c0SJesper Dangaard Brouer void __kfree_skb_flush(void) 864795bb1c0SJesper Dangaard Brouer { 865795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 866795bb1c0SJesper Dangaard Brouer 867795bb1c0SJesper Dangaard Brouer /* flush skb_cache if containing objects */ 868795bb1c0SJesper Dangaard Brouer if (nc->skb_count) { 869795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 870795bb1c0SJesper Dangaard Brouer nc->skb_cache); 871795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 872795bb1c0SJesper Dangaard Brouer } 873795bb1c0SJesper Dangaard Brouer } 874795bb1c0SJesper Dangaard Brouer 87515fad714SJesper Dangaard Brouer static inline void _kfree_skb_defer(struct sk_buff *skb) 876795bb1c0SJesper Dangaard Brouer { 877795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 878795bb1c0SJesper Dangaard Brouer 879795bb1c0SJesper Dangaard Brouer /* drop skb->head and call any destructors for packet */ 880795bb1c0SJesper Dangaard Brouer skb_release_all(skb); 881795bb1c0SJesper Dangaard Brouer 882795bb1c0SJesper Dangaard Brouer /* record skb to CPU local list */ 883795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 884795bb1c0SJesper Dangaard Brouer 885795bb1c0SJesper Dangaard Brouer #ifdef CONFIG_SLUB 886795bb1c0SJesper Dangaard Brouer /* SLUB writes into objects when freeing */ 887795bb1c0SJesper Dangaard Brouer prefetchw(skb); 888795bb1c0SJesper Dangaard Brouer #endif 889795bb1c0SJesper Dangaard Brouer 890795bb1c0SJesper Dangaard Brouer /* flush skb_cache if it is filled */ 891795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 892795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 893795bb1c0SJesper Dangaard Brouer nc->skb_cache); 894795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 895795bb1c0SJesper Dangaard Brouer } 896795bb1c0SJesper Dangaard Brouer } 89715fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 89815fad714SJesper Dangaard Brouer { 89915fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 90015fad714SJesper Dangaard Brouer } 901795bb1c0SJesper Dangaard Brouer 902795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 903795bb1c0SJesper Dangaard Brouer { 904885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 905795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 906885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 907795bb1c0SJesper Dangaard Brouer return; 908795bb1c0SJesper Dangaard Brouer } 909795bb1c0SJesper Dangaard Brouer 9106454eca8SYunsheng Lin lockdep_assert_in_softirq(); 9116454eca8SYunsheng Lin 9127608894eSPaolo Abeni if (!skb_unref(skb)) 913795bb1c0SJesper Dangaard Brouer return; 9147608894eSPaolo Abeni 915795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 916795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 917795bb1c0SJesper Dangaard Brouer 918795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 919abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 920795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 921795bb1c0SJesper Dangaard Brouer return; 922795bb1c0SJesper Dangaard Brouer } 923795bb1c0SJesper Dangaard Brouer 92415fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 925795bb1c0SJesper Dangaard Brouer } 926795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 927795bb1c0SJesper Dangaard Brouer 928b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 929b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 930b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 931b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 932b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 933b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 934b1937227SEric Dumazet 935dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 936dec18810SHerbert Xu { 937dec18810SHerbert Xu new->tstamp = old->tstamp; 938b1937227SEric Dumazet /* We do not copy old->sk */ 939dec18810SHerbert Xu new->dev = old->dev; 940b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 9417fee226aSEric Dumazet skb_dst_copy(new, old); 942df5042f4SFlorian Westphal __skb_ext_copy(new, old); 943b1937227SEric Dumazet __nf_copy(new, old, false); 9446aa895b0SPatrick McHardy 945b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 946b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 947b1937227SEric Dumazet */ 948b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 94906021292SEliezer Tamir 950b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 951b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 952b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 953b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 954b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 955b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 956b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 957b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 958b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 959b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 960b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 961b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 962b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 963b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 964b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 965b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 966b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 967b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 968b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 969b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 970b1937227SEric Dumazet #endif 971e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 972b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 97306021292SEliezer Tamir #endif 9742bd82484SEric Dumazet #ifdef CONFIG_XPS 9752bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 9762bd82484SEric Dumazet #endif 977b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 978b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 979b1937227SEric Dumazet #endif 980b1937227SEric Dumazet 981dec18810SHerbert Xu } 982dec18810SHerbert Xu 98382c49a35SHerbert Xu /* 98482c49a35SHerbert Xu * You should not add any new code to this function. Add it to 98582c49a35SHerbert Xu * __copy_skb_header above instead. 98682c49a35SHerbert Xu */ 987e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 9881da177e4SLinus Torvalds { 9891da177e4SLinus Torvalds #define C(x) n->x = skb->x 9901da177e4SLinus Torvalds 9911da177e4SLinus Torvalds n->next = n->prev = NULL; 9921da177e4SLinus Torvalds n->sk = NULL; 993dec18810SHerbert Xu __copy_skb_header(n, skb); 994dec18810SHerbert Xu 9951da177e4SLinus Torvalds C(len); 9961da177e4SLinus Torvalds C(data_len); 9973e6b3b2eSAlexey Dobriyan C(mac_len); 998334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 99902f1c89dSPaul Moore n->cloned = 1; 10001da177e4SLinus Torvalds n->nohdr = 0; 1001b13dda9fSEric Dumazet n->peeked = 0; 1002e78bfb07SStefano Brivio C(pfmemalloc); 10031da177e4SLinus Torvalds n->destructor = NULL; 10041da177e4SLinus Torvalds C(tail); 10051da177e4SLinus Torvalds C(end); 100602f1c89dSPaul Moore C(head); 1007d3836f21SEric Dumazet C(head_frag); 100802f1c89dSPaul Moore C(data); 100902f1c89dSPaul Moore C(truesize); 101063354797SReshetova, Elena refcount_set(&n->users, 1); 10111da177e4SLinus Torvalds 10121da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 10131da177e4SLinus Torvalds skb->cloned = 1; 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds return n; 1016e0053ec0SHerbert Xu #undef C 1017e0053ec0SHerbert Xu } 1018e0053ec0SHerbert Xu 1019e0053ec0SHerbert Xu /** 1020da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1021da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1022da29e4b4SJakub Kicinski */ 1023da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1024da29e4b4SJakub Kicinski { 1025da29e4b4SJakub Kicinski struct sk_buff *n; 1026da29e4b4SJakub Kicinski 1027da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1028da29e4b4SJakub Kicinski if (!n) 1029da29e4b4SJakub Kicinski return NULL; 1030da29e4b4SJakub Kicinski 1031da29e4b4SJakub Kicinski n->len = first->len; 1032da29e4b4SJakub Kicinski n->data_len = first->len; 1033da29e4b4SJakub Kicinski n->truesize = first->truesize; 1034da29e4b4SJakub Kicinski 1035da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1036da29e4b4SJakub Kicinski 1037da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1038da29e4b4SJakub Kicinski n->destructor = NULL; 1039da29e4b4SJakub Kicinski 1040da29e4b4SJakub Kicinski return n; 1041da29e4b4SJakub Kicinski } 1042da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1043da29e4b4SJakub Kicinski 1044da29e4b4SJakub Kicinski /** 1045e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1046e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1047e0053ec0SHerbert Xu * @src: the skb to supply the contents 1048e0053ec0SHerbert Xu * 1049e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1050e0053ec0SHerbert Xu * supplied by the user. 1051e0053ec0SHerbert Xu * 1052e0053ec0SHerbert Xu * The target skb is returned upon exit. 1053e0053ec0SHerbert Xu */ 1054e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1055e0053ec0SHerbert Xu { 10562d4baff8SHerbert Xu skb_release_all(dst); 1057e0053ec0SHerbert Xu return __skb_clone(dst, src); 1058e0053ec0SHerbert Xu } 1059e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1060e0053ec0SHerbert Xu 10616f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1062a91dbff5SWillem de Bruijn { 1063a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1064a91dbff5SWillem de Bruijn struct user_struct *user; 1065a91dbff5SWillem de Bruijn 1066a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1067a91dbff5SWillem de Bruijn return 0; 1068a91dbff5SWillem de Bruijn 1069a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1070a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1071a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1072a91dbff5SWillem de Bruijn 1073a91dbff5SWillem de Bruijn do { 1074a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1075a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1076a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1077a91dbff5SWillem de Bruijn return -ENOBUFS; 1078a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1079a91dbff5SWillem de Bruijn old_pg); 1080a91dbff5SWillem de Bruijn 1081a91dbff5SWillem de Bruijn if (!mmp->user) { 1082a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1083a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1084a91dbff5SWillem de Bruijn } else { 1085a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1086a91dbff5SWillem de Bruijn } 1087a91dbff5SWillem de Bruijn 1088a91dbff5SWillem de Bruijn return 0; 1089a91dbff5SWillem de Bruijn } 10906f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1091a91dbff5SWillem de Bruijn 10926f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1093a91dbff5SWillem de Bruijn { 1094a91dbff5SWillem de Bruijn if (mmp->user) { 1095a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1096a91dbff5SWillem de Bruijn free_uid(mmp->user); 1097a91dbff5SWillem de Bruijn } 1098a91dbff5SWillem de Bruijn } 10996f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1100a91dbff5SWillem de Bruijn 110152267790SWillem de Bruijn struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size) 110252267790SWillem de Bruijn { 110352267790SWillem de Bruijn struct ubuf_info *uarg; 110452267790SWillem de Bruijn struct sk_buff *skb; 110552267790SWillem de Bruijn 110652267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 110752267790SWillem de Bruijn 110852267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 110952267790SWillem de Bruijn if (!skb) 111052267790SWillem de Bruijn return NULL; 111152267790SWillem de Bruijn 111252267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 111352267790SWillem de Bruijn uarg = (void *)skb->cb; 1114a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1115a91dbff5SWillem de Bruijn 1116a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1117a91dbff5SWillem de Bruijn kfree_skb(skb); 1118a91dbff5SWillem de Bruijn return NULL; 1119a91dbff5SWillem de Bruijn } 112052267790SWillem de Bruijn 112152267790SWillem de Bruijn uarg->callback = sock_zerocopy_callback; 11224ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 11234ab6c99dSWillem de Bruijn uarg->len = 1; 11244ab6c99dSWillem de Bruijn uarg->bytelen = size; 112552267790SWillem de Bruijn uarg->zerocopy = 1; 1126c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 112752267790SWillem de Bruijn sock_hold(sk); 112852267790SWillem de Bruijn 112952267790SWillem de Bruijn return uarg; 113052267790SWillem de Bruijn } 113152267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_alloc); 113252267790SWillem de Bruijn 113352267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 113452267790SWillem de Bruijn { 113552267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 113652267790SWillem de Bruijn } 113752267790SWillem de Bruijn 11384ab6c99dSWillem de Bruijn struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 11394ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 11404ab6c99dSWillem de Bruijn { 11414ab6c99dSWillem de Bruijn if (uarg) { 11424ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 11434ab6c99dSWillem de Bruijn u32 bytelen, next; 11444ab6c99dSWillem de Bruijn 11454ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 11464ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 11474ab6c99dSWillem de Bruijn */ 11484ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 11494ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 11504ab6c99dSWillem de Bruijn return NULL; 11514ab6c99dSWillem de Bruijn } 11524ab6c99dSWillem de Bruijn 11534ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 11544ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 11554ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 11564ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 11574ab6c99dSWillem de Bruijn goto new_alloc; 11584ab6c99dSWillem de Bruijn return NULL; 11594ab6c99dSWillem de Bruijn } 11604ab6c99dSWillem de Bruijn 11614ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 11624ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1163a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1164a91dbff5SWillem de Bruijn return NULL; 11654ab6c99dSWillem de Bruijn uarg->len++; 11664ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 11674ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1168100f6d8eSWillem de Bruijn 1169100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1170100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 1171db5bce32SEric Dumazet sock_zerocopy_get(uarg); 1172100f6d8eSWillem de Bruijn 11734ab6c99dSWillem de Bruijn return uarg; 11744ab6c99dSWillem de Bruijn } 11754ab6c99dSWillem de Bruijn } 11764ab6c99dSWillem de Bruijn 11774ab6c99dSWillem de Bruijn new_alloc: 11784ab6c99dSWillem de Bruijn return sock_zerocopy_alloc(sk, size); 11794ab6c99dSWillem de Bruijn } 11804ab6c99dSWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_realloc); 11814ab6c99dSWillem de Bruijn 11824ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 11834ab6c99dSWillem de Bruijn { 11844ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 11854ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 11864ab6c99dSWillem de Bruijn u64 sum_len; 11874ab6c99dSWillem de Bruijn 11884ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 11894ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 11904ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 11914ab6c99dSWillem de Bruijn 11924ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 11934ab6c99dSWillem de Bruijn return false; 11944ab6c99dSWillem de Bruijn 11954ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 11964ab6c99dSWillem de Bruijn return false; 11974ab6c99dSWillem de Bruijn 11984ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 11994ab6c99dSWillem de Bruijn return true; 12004ab6c99dSWillem de Bruijn } 12014ab6c99dSWillem de Bruijn 120252267790SWillem de Bruijn void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) 120352267790SWillem de Bruijn { 12044ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 120552267790SWillem de Bruijn struct sock_exterr_skb *serr; 120652267790SWillem de Bruijn struct sock *sk = skb->sk; 12074ab6c99dSWillem de Bruijn struct sk_buff_head *q; 12084ab6c99dSWillem de Bruijn unsigned long flags; 12094ab6c99dSWillem de Bruijn u32 lo, hi; 12104ab6c99dSWillem de Bruijn u16 len; 121152267790SWillem de Bruijn 1212ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1213ccaffff1SWillem de Bruijn 12144ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 12154ab6c99dSWillem de Bruijn * so do not queue a completion notification 12164ab6c99dSWillem de Bruijn */ 12174ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 121852267790SWillem de Bruijn goto release; 121952267790SWillem de Bruijn 12204ab6c99dSWillem de Bruijn len = uarg->len; 12214ab6c99dSWillem de Bruijn lo = uarg->id; 12224ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 12234ab6c99dSWillem de Bruijn 122452267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 122552267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 122652267790SWillem de Bruijn serr->ee.ee_errno = 0; 122752267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 12284ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 12294ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 123052267790SWillem de Bruijn if (!success) 123152267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 123252267790SWillem de Bruijn 12334ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 12344ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 12354ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 12364ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 12374ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 12384ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 123952267790SWillem de Bruijn skb = NULL; 12404ab6c99dSWillem de Bruijn } 12414ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 124252267790SWillem de Bruijn 124352267790SWillem de Bruijn sk->sk_error_report(sk); 124452267790SWillem de Bruijn 124552267790SWillem de Bruijn release: 124652267790SWillem de Bruijn consume_skb(skb); 124752267790SWillem de Bruijn sock_put(sk); 124852267790SWillem de Bruijn } 124952267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_callback); 125052267790SWillem de Bruijn 125152267790SWillem de Bruijn void sock_zerocopy_put(struct ubuf_info *uarg) 125252267790SWillem de Bruijn { 1253c1d1b437SEric Dumazet if (uarg && refcount_dec_and_test(&uarg->refcnt)) { 125452267790SWillem de Bruijn if (uarg->callback) 125552267790SWillem de Bruijn uarg->callback(uarg, uarg->zerocopy); 125652267790SWillem de Bruijn else 125752267790SWillem de Bruijn consume_skb(skb_from_uarg(uarg)); 125852267790SWillem de Bruijn } 125952267790SWillem de Bruijn } 126052267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put); 126152267790SWillem de Bruijn 126252900d22SWillem de Bruijn void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 126352267790SWillem de Bruijn { 126452267790SWillem de Bruijn if (uarg) { 126552267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 126652267790SWillem de Bruijn 126752267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 12684ab6c99dSWillem de Bruijn uarg->len--; 126952267790SWillem de Bruijn 127052900d22SWillem de Bruijn if (have_uref) 127152267790SWillem de Bruijn sock_zerocopy_put(uarg); 127252267790SWillem de Bruijn } 127352267790SWillem de Bruijn } 127452267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); 127552267790SWillem de Bruijn 1276b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1277b5947e5dSWillem de Bruijn { 1278b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1279b5947e5dSWillem de Bruijn } 1280b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1281b5947e5dSWillem de Bruijn 128252267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 128352267790SWillem de Bruijn struct msghdr *msg, int len, 128452267790SWillem de Bruijn struct ubuf_info *uarg) 128552267790SWillem de Bruijn { 12864ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 128752267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 128852267790SWillem de Bruijn int err, orig_len = skb->len; 128952267790SWillem de Bruijn 12904ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 12914ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 12924ab6c99dSWillem de Bruijn */ 12934ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 12944ab6c99dSWillem de Bruijn return -EEXIST; 12954ab6c99dSWillem de Bruijn 129652267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 129752267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 129854d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 129954d43117SWillem de Bruijn 130052267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 130152267790SWillem de Bruijn msg->msg_iter = orig_iter; 130254d43117SWillem de Bruijn skb->sk = sk; 130352267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 130454d43117SWillem de Bruijn skb->sk = save_sk; 130552267790SWillem de Bruijn return err; 130652267790SWillem de Bruijn } 130752267790SWillem de Bruijn 130852900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 130952267790SWillem de Bruijn return skb->len - orig_len; 131052267790SWillem de Bruijn } 131152267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 131252267790SWillem de Bruijn 13131f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 131452267790SWillem de Bruijn gfp_t gfp_mask) 131552267790SWillem de Bruijn { 131652267790SWillem de Bruijn if (skb_zcopy(orig)) { 131752267790SWillem de Bruijn if (skb_zcopy(nskb)) { 131852267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 131952267790SWillem de Bruijn if (!gfp_mask) { 132052267790SWillem de Bruijn WARN_ON_ONCE(1); 132152267790SWillem de Bruijn return -ENOMEM; 132252267790SWillem de Bruijn } 132352267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 132452267790SWillem de Bruijn return 0; 132552267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 132652267790SWillem de Bruijn return -EIO; 132752267790SWillem de Bruijn } 132852900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 132952267790SWillem de Bruijn } 133052267790SWillem de Bruijn return 0; 133152267790SWillem de Bruijn } 133252267790SWillem de Bruijn 13332c53040fSBen Hutchings /** 13342c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 133548c83012SMichael S. Tsirkin * @skb: the skb to modify 133648c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 133748c83012SMichael S. Tsirkin * 133848c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 133948c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 134048c83012SMichael S. Tsirkin * to userspace pages. 134148c83012SMichael S. Tsirkin * 134248c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 134348c83012SMichael S. Tsirkin * %GFP_ATOMIC. 134448c83012SMichael S. Tsirkin * 134548c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 134648c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 134748c83012SMichael S. Tsirkin */ 134848c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1349a6686f2fSShirley Ma { 1350a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1351a6686f2fSShirley Ma struct page *page, *head = NULL; 13523ece7826SWillem de Bruijn int i, new_frags; 13533ece7826SWillem de Bruijn u32 d_off; 1354a6686f2fSShirley Ma 13553ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 13563ece7826SWillem de Bruijn return -EINVAL; 13573ece7826SWillem de Bruijn 1358f72c4ac6SWillem de Bruijn if (!num_frags) 1359f72c4ac6SWillem de Bruijn goto release; 1360f72c4ac6SWillem de Bruijn 13613ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 13623ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 136302756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1364a6686f2fSShirley Ma if (!page) { 1365a6686f2fSShirley Ma while (head) { 136640dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1367a6686f2fSShirley Ma put_page(head); 1368a6686f2fSShirley Ma head = next; 1369a6686f2fSShirley Ma } 1370a6686f2fSShirley Ma return -ENOMEM; 1371a6686f2fSShirley Ma } 13723ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 13733ece7826SWillem de Bruijn head = page; 13743ece7826SWillem de Bruijn } 13753ece7826SWillem de Bruijn 13763ece7826SWillem de Bruijn page = head; 13773ece7826SWillem de Bruijn d_off = 0; 13783ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 13793ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 13803ece7826SWillem de Bruijn u32 p_off, p_len, copied; 13813ece7826SWillem de Bruijn struct page *p; 13823ece7826SWillem de Bruijn u8 *vaddr; 1383c613c209SWillem de Bruijn 1384b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1385c613c209SWillem de Bruijn p, p_off, p_len, copied) { 13863ece7826SWillem de Bruijn u32 copy, done = 0; 1387c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 13883ece7826SWillem de Bruijn 13893ece7826SWillem de Bruijn while (done < p_len) { 13903ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 13913ece7826SWillem de Bruijn d_off = 0; 13923ece7826SWillem de Bruijn page = (struct page *)page_private(page); 13933ece7826SWillem de Bruijn } 13943ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 13953ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 13963ece7826SWillem de Bruijn vaddr + p_off + done, copy); 13973ece7826SWillem de Bruijn done += copy; 13983ece7826SWillem de Bruijn d_off += copy; 13993ece7826SWillem de Bruijn } 140051c56b00SEric Dumazet kunmap_atomic(vaddr); 1401c613c209SWillem de Bruijn } 1402a6686f2fSShirley Ma } 1403a6686f2fSShirley Ma 1404a6686f2fSShirley Ma /* skb frags release userspace buffers */ 140502756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1406a8605c60SIan Campbell skb_frag_unref(skb, i); 1407a6686f2fSShirley Ma 1408a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 14093ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 14103ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 141140dadff2SSunghan Suh head = (struct page *)page_private(head); 1412a6686f2fSShirley Ma } 14133ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 14143ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 141548c83012SMichael S. Tsirkin 1416b90ddd56SWillem de Bruijn release: 14171f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1418a6686f2fSShirley Ma return 0; 1419a6686f2fSShirley Ma } 1420dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1421a6686f2fSShirley Ma 1422e0053ec0SHerbert Xu /** 1423e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1424e0053ec0SHerbert Xu * @skb: buffer to clone 1425e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1426e0053ec0SHerbert Xu * 1427e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1428e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1429e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1430e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1431e0053ec0SHerbert Xu * 1432e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1433e0053ec0SHerbert Xu * %GFP_ATOMIC. 1434e0053ec0SHerbert Xu */ 1435e0053ec0SHerbert Xu 1436e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1437e0053ec0SHerbert Xu { 1438d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1439d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1440d0bf4a9eSEric Dumazet skb1); 14416ffe75ebSEric Dumazet struct sk_buff *n; 1442e0053ec0SHerbert Xu 144370008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1444a6686f2fSShirley Ma return NULL; 1445a6686f2fSShirley Ma 1446e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 14472638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 14486ffe75ebSEric Dumazet n = &fclones->skb2; 14492638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1450e0053ec0SHerbert Xu } else { 1451c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1452c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1453c93bdd0eSMel Gorman 1454e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1455e0053ec0SHerbert Xu if (!n) 1456e0053ec0SHerbert Xu return NULL; 1457fe55f6d5SVegard Nossum 1458e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1459e0053ec0SHerbert Xu } 1460e0053ec0SHerbert Xu 1461e0053ec0SHerbert Xu return __skb_clone(n, skb); 14621da177e4SLinus Torvalds } 1463b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 14641da177e4SLinus Torvalds 1465b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1466f5b17294SPravin B Shelar { 1467030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1468030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1469030737bcSEric Dumazet skb->csum_start += off; 1470f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1471f5b17294SPravin B Shelar skb->transport_header += off; 1472f5b17294SPravin B Shelar skb->network_header += off; 1473f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1474f5b17294SPravin B Shelar skb->mac_header += off; 1475f5b17294SPravin B Shelar skb->inner_transport_header += off; 1476f5b17294SPravin B Shelar skb->inner_network_header += off; 1477aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1478f5b17294SPravin B Shelar } 1479b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1480f5b17294SPravin B Shelar 148108303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 14821da177e4SLinus Torvalds { 1483dec18810SHerbert Xu __copy_skb_header(new, old); 1484dec18810SHerbert Xu 14857967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 14867967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 14877967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 14881da177e4SLinus Torvalds } 148908303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 14901da177e4SLinus Torvalds 1491c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1492c93bdd0eSMel Gorman { 1493c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1494c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1495c93bdd0eSMel Gorman return 0; 1496c93bdd0eSMel Gorman } 1497c93bdd0eSMel Gorman 14981da177e4SLinus Torvalds /** 14991da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 15001da177e4SLinus Torvalds * @skb: buffer to copy 15011da177e4SLinus Torvalds * @gfp_mask: allocation priority 15021da177e4SLinus Torvalds * 15031da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 15041da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 15051da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 15061da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 15071da177e4SLinus Torvalds * 15081da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 15091da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 15101da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 15111da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 15121da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 15131da177e4SLinus Torvalds */ 15141da177e4SLinus Torvalds 1515dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 15161da177e4SLinus Torvalds { 15176602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1518ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1519c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1520c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 15216602cebbSEric Dumazet 15221da177e4SLinus Torvalds if (!n) 15231da177e4SLinus Torvalds return NULL; 15241da177e4SLinus Torvalds 15251da177e4SLinus Torvalds /* Set the data pointer */ 15261da177e4SLinus Torvalds skb_reserve(n, headerlen); 15271da177e4SLinus Torvalds /* Set the tail pointer and length */ 15281da177e4SLinus Torvalds skb_put(n, skb->len); 15291da177e4SLinus Torvalds 15309f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 15311da177e4SLinus Torvalds 153208303c18SIlya Lesokhin skb_copy_header(n, skb); 15331da177e4SLinus Torvalds return n; 15341da177e4SLinus Torvalds } 1535b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 15361da177e4SLinus Torvalds 15371da177e4SLinus Torvalds /** 1538bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 15391da177e4SLinus Torvalds * @skb: buffer to copy 1540117632e6SEric Dumazet * @headroom: headroom of new skb 15411da177e4SLinus Torvalds * @gfp_mask: allocation priority 1542bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1543bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1544bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 15451da177e4SLinus Torvalds * 15461da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 15471da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 15481da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 15491da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 15501da177e4SLinus Torvalds * or the pointer to the buffer on success. 15511da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 15521da177e4SLinus Torvalds */ 15531da177e4SLinus Torvalds 1554bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1555bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 15561da177e4SLinus Torvalds { 1557117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1558bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1559bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 15606602cebbSEric Dumazet 15611da177e4SLinus Torvalds if (!n) 15621da177e4SLinus Torvalds goto out; 15631da177e4SLinus Torvalds 15641da177e4SLinus Torvalds /* Set the data pointer */ 1565117632e6SEric Dumazet skb_reserve(n, headroom); 15661da177e4SLinus Torvalds /* Set the tail pointer and length */ 15671da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 15681da177e4SLinus Torvalds /* Copy the bytes */ 1569d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 15701da177e4SLinus Torvalds 157125f484a6SHerbert Xu n->truesize += skb->data_len; 15721da177e4SLinus Torvalds n->data_len = skb->data_len; 15731da177e4SLinus Torvalds n->len = skb->len; 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 15761da177e4SLinus Torvalds int i; 15771da177e4SLinus Torvalds 15781f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 15791f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 15801511022cSDan Carpenter kfree_skb(n); 15811511022cSDan Carpenter n = NULL; 1582a6686f2fSShirley Ma goto out; 1583a6686f2fSShirley Ma } 15841da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 15851da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1586ea2ab693SIan Campbell skb_frag_ref(skb, i); 15871da177e4SLinus Torvalds } 15881da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 15891da177e4SLinus Torvalds } 15901da177e4SLinus Torvalds 159121dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 15921da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 15931da177e4SLinus Torvalds skb_clone_fraglist(n); 15941da177e4SLinus Torvalds } 15951da177e4SLinus Torvalds 159608303c18SIlya Lesokhin skb_copy_header(n, skb); 15971da177e4SLinus Torvalds out: 15981da177e4SLinus Torvalds return n; 15991da177e4SLinus Torvalds } 1600bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds /** 16031da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 16041da177e4SLinus Torvalds * @skb: buffer to reallocate 16051da177e4SLinus Torvalds * @nhead: room to add at head 16061da177e4SLinus Torvalds * @ntail: room to add at tail 16071da177e4SLinus Torvalds * @gfp_mask: allocation priority 16081da177e4SLinus Torvalds * 1609bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1610bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 16111da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 16121da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 16131da177e4SLinus Torvalds * 16141da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 16151da177e4SLinus Torvalds * reloaded after call to this function. 16161da177e4SLinus Torvalds */ 16171da177e4SLinus Torvalds 161886a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1619dd0fc66fSAl Viro gfp_t gfp_mask) 16201da177e4SLinus Torvalds { 1621158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1622158f323bSEric Dumazet int size = osize + nhead + ntail; 16231da177e4SLinus Torvalds long off; 1624158f323bSEric Dumazet u8 *data; 16251da177e4SLinus Torvalds 16264edd87adSHerbert Xu BUG_ON(nhead < 0); 16274edd87adSHerbert Xu 16289f77fad3STim Hansen BUG_ON(skb_shared(skb)); 16291da177e4SLinus Torvalds 16301da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 16311da177e4SLinus Torvalds 1632c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1633c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1634c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1635c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 16361da177e4SLinus Torvalds if (!data) 16371da177e4SLinus Torvalds goto nodata; 163887151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 16391da177e4SLinus Torvalds 16401da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 16416602cebbSEric Dumazet * optimized for the cases when header is void. 16426602cebbSEric Dumazet */ 16436602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 16446602cebbSEric Dumazet 16456602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 16466602cebbSEric Dumazet skb_shinfo(skb), 1647fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 16481da177e4SLinus Torvalds 16493e24591aSAlexander Duyck /* 16503e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 16513e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 16523e24591aSAlexander Duyck * be since all we did is relocate the values 16533e24591aSAlexander Duyck */ 16543e24591aSAlexander Duyck if (skb_cloned(skb)) { 165570008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1656a6686f2fSShirley Ma goto nofrags; 16571f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1658c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 16591da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1660ea2ab693SIan Campbell skb_frag_ref(skb, i); 16611da177e4SLinus Torvalds 166221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 16631da177e4SLinus Torvalds skb_clone_fraglist(skb); 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds skb_release_data(skb); 16663e24591aSAlexander Duyck } else { 16673e24591aSAlexander Duyck skb_free_head(skb); 16681fd63041SEric Dumazet } 16691da177e4SLinus Torvalds off = (data + nhead) - skb->head; 16701da177e4SLinus Torvalds 16711da177e4SLinus Torvalds skb->head = data; 1672d3836f21SEric Dumazet skb->head_frag = 0; 16731da177e4SLinus Torvalds skb->data += off; 16744305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 16754305b541SArnaldo Carvalho de Melo skb->end = size; 167656eb8882SPatrick McHardy off = nhead; 16774305b541SArnaldo Carvalho de Melo #else 16784305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 167956eb8882SPatrick McHardy #endif 168027a884dcSArnaldo Carvalho de Melo skb->tail += off; 1681b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 16821da177e4SLinus Torvalds skb->cloned = 0; 1683334a8132SPatrick McHardy skb->hdr_len = 0; 16841da177e4SLinus Torvalds skb->nohdr = 0; 16851da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1686158f323bSEric Dumazet 1687de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1688de8f3a83SDaniel Borkmann 1689158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1690158f323bSEric Dumazet * For the moment, we really care of rx path, or 1691158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1692158f323bSEric Dumazet */ 1693158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1694158f323bSEric Dumazet skb->truesize += size - osize; 1695158f323bSEric Dumazet 16961da177e4SLinus Torvalds return 0; 16971da177e4SLinus Torvalds 1698a6686f2fSShirley Ma nofrags: 1699a6686f2fSShirley Ma kfree(data); 17001da177e4SLinus Torvalds nodata: 17011da177e4SLinus Torvalds return -ENOMEM; 17021da177e4SLinus Torvalds } 1703b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 17041da177e4SLinus Torvalds 17051da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 17061da177e4SLinus Torvalds 17071da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 17081da177e4SLinus Torvalds { 17091da177e4SLinus Torvalds struct sk_buff *skb2; 17101da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 17111da177e4SLinus Torvalds 17121da177e4SLinus Torvalds if (delta <= 0) 17131da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 17141da177e4SLinus Torvalds else { 17151da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 17161da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 17171da177e4SLinus Torvalds GFP_ATOMIC)) { 17181da177e4SLinus Torvalds kfree_skb(skb2); 17191da177e4SLinus Torvalds skb2 = NULL; 17201da177e4SLinus Torvalds } 17211da177e4SLinus Torvalds } 17221da177e4SLinus Torvalds return skb2; 17231da177e4SLinus Torvalds } 1724b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 17251da177e4SLinus Torvalds 17261da177e4SLinus Torvalds /** 17271da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 17281da177e4SLinus Torvalds * @skb: buffer to copy 17291da177e4SLinus Torvalds * @newheadroom: new free bytes at head 17301da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 17311da177e4SLinus Torvalds * @gfp_mask: allocation priority 17321da177e4SLinus Torvalds * 17331da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 17341da177e4SLinus Torvalds * allocate additional space. 17351da177e4SLinus Torvalds * 17361da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 17371da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 17381da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 17391da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 17401da177e4SLinus Torvalds * 17411da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 17421da177e4SLinus Torvalds * is called from an interrupt. 17431da177e4SLinus Torvalds */ 17441da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 174586a76cafSVictor Fusco int newheadroom, int newtailroom, 1746dd0fc66fSAl Viro gfp_t gfp_mask) 17471da177e4SLinus Torvalds { 17481da177e4SLinus Torvalds /* 17491da177e4SLinus Torvalds * Allocate the copy buffer 17501da177e4SLinus Torvalds */ 1751c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1752c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1753c93bdd0eSMel Gorman NUMA_NO_NODE); 1754efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 17551da177e4SLinus Torvalds int head_copy_len, head_copy_off; 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds if (!n) 17581da177e4SLinus Torvalds return NULL; 17591da177e4SLinus Torvalds 17601da177e4SLinus Torvalds skb_reserve(n, newheadroom); 17611da177e4SLinus Torvalds 17621da177e4SLinus Torvalds /* Set the tail pointer and length */ 17631da177e4SLinus Torvalds skb_put(n, skb->len); 17641da177e4SLinus Torvalds 1765efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 17661da177e4SLinus Torvalds head_copy_off = 0; 17671da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 17681da177e4SLinus Torvalds head_copy_len = newheadroom; 17691da177e4SLinus Torvalds else 17701da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 17711da177e4SLinus Torvalds 17721da177e4SLinus Torvalds /* Copy the linear header and data. */ 17739f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 17749f77fad3STim Hansen skb->len + head_copy_len)); 17751da177e4SLinus Torvalds 177608303c18SIlya Lesokhin skb_copy_header(n, skb); 17771da177e4SLinus Torvalds 1778030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1779efd1e8d5SPatrick McHardy 17801da177e4SLinus Torvalds return n; 17811da177e4SLinus Torvalds } 1782b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 17831da177e4SLinus Torvalds 17841da177e4SLinus Torvalds /** 1785cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 17861da177e4SLinus Torvalds * @skb: buffer to pad 17871da177e4SLinus Torvalds * @pad: space to pad 1788cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 17891da177e4SLinus Torvalds * 17901da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 17911da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 17921da177e4SLinus Torvalds * beyond the buffer end onto the wire. 17931da177e4SLinus Torvalds * 1794cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1795cd0a137aSFlorian Fainelli * if @free_on_error is true. 17961da177e4SLinus Torvalds */ 17971da177e4SLinus Torvalds 1798cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 17991da177e4SLinus Torvalds { 18005b057c6bSHerbert Xu int err; 18015b057c6bSHerbert Xu int ntail; 18021da177e4SLinus Torvalds 18031da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 18045b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 18051da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 18065b057c6bSHerbert Xu return 0; 18071da177e4SLinus Torvalds } 18081da177e4SLinus Torvalds 18094305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 18105b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 18115b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 18125b057c6bSHerbert Xu if (unlikely(err)) 18135b057c6bSHerbert Xu goto free_skb; 18145b057c6bSHerbert Xu } 18155b057c6bSHerbert Xu 18165b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 18175b057c6bSHerbert Xu * to be audited. 18185b057c6bSHerbert Xu */ 18195b057c6bSHerbert Xu err = skb_linearize(skb); 18205b057c6bSHerbert Xu if (unlikely(err)) 18215b057c6bSHerbert Xu goto free_skb; 18225b057c6bSHerbert Xu 18235b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 18245b057c6bSHerbert Xu return 0; 18255b057c6bSHerbert Xu 18265b057c6bSHerbert Xu free_skb: 1827cd0a137aSFlorian Fainelli if (free_on_error) 18281da177e4SLinus Torvalds kfree_skb(skb); 18295b057c6bSHerbert Xu return err; 18301da177e4SLinus Torvalds } 1831cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 18321da177e4SLinus Torvalds 18330dde3e16SIlpo Järvinen /** 18340c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 18350c7ddf36SMathias Krause * @skb: start of the buffer to use 18360c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 18370c7ddf36SMathias Krause * @len: amount of data to add 18380c7ddf36SMathias Krause * 18390c7ddf36SMathias Krause * This function extends the used data area of the potentially 18400c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 18410c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 18420c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 18430c7ddf36SMathias Krause * returned. 18440c7ddf36SMathias Krause */ 18450c7ddf36SMathias Krause 18464df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 18470c7ddf36SMathias Krause { 18480c7ddf36SMathias Krause if (tail != skb) { 18490c7ddf36SMathias Krause skb->data_len += len; 18500c7ddf36SMathias Krause skb->len += len; 18510c7ddf36SMathias Krause } 18520c7ddf36SMathias Krause return skb_put(tail, len); 18530c7ddf36SMathias Krause } 18540c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 18550c7ddf36SMathias Krause 18560c7ddf36SMathias Krause /** 18570dde3e16SIlpo Järvinen * skb_put - add data to a buffer 18580dde3e16SIlpo Järvinen * @skb: buffer to use 18590dde3e16SIlpo Järvinen * @len: amount of data to add 18600dde3e16SIlpo Järvinen * 18610dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 18620dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 18630dde3e16SIlpo Järvinen * first byte of the extra data is returned. 18640dde3e16SIlpo Järvinen */ 18654df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 18660dde3e16SIlpo Järvinen { 18674df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 18680dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 18690dde3e16SIlpo Järvinen skb->tail += len; 18700dde3e16SIlpo Järvinen skb->len += len; 18710dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 18720dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 18730dde3e16SIlpo Järvinen return tmp; 18740dde3e16SIlpo Järvinen } 18750dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 18760dde3e16SIlpo Järvinen 18776be8ac2fSIlpo Järvinen /** 1878c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1879c2aa270aSIlpo Järvinen * @skb: buffer to use 1880c2aa270aSIlpo Järvinen * @len: amount of data to add 1881c2aa270aSIlpo Järvinen * 1882c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1883c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1884c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1885c2aa270aSIlpo Järvinen */ 1886d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 1887c2aa270aSIlpo Järvinen { 1888c2aa270aSIlpo Järvinen skb->data -= len; 1889c2aa270aSIlpo Järvinen skb->len += len; 1890c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 1891c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1892c2aa270aSIlpo Järvinen return skb->data; 1893c2aa270aSIlpo Järvinen } 1894c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1895c2aa270aSIlpo Järvinen 1896c2aa270aSIlpo Järvinen /** 18976be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 18986be8ac2fSIlpo Järvinen * @skb: buffer to use 18996be8ac2fSIlpo Järvinen * @len: amount of data to remove 19006be8ac2fSIlpo Järvinen * 19016be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 19026be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 19036be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 19046be8ac2fSIlpo Järvinen * the old data. 19056be8ac2fSIlpo Järvinen */ 1906af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 19076be8ac2fSIlpo Järvinen { 190847d29646SDavid S. Miller return skb_pull_inline(skb, len); 19096be8ac2fSIlpo Järvinen } 19106be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 19116be8ac2fSIlpo Järvinen 1912419ae74eSIlpo Järvinen /** 1913419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1914419ae74eSIlpo Järvinen * @skb: buffer to alter 1915419ae74eSIlpo Järvinen * @len: new length 1916419ae74eSIlpo Järvinen * 1917419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1918419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1919419ae74eSIlpo Järvinen * The skb must be linear. 1920419ae74eSIlpo Järvinen */ 1921419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1922419ae74eSIlpo Järvinen { 1923419ae74eSIlpo Järvinen if (skb->len > len) 1924419ae74eSIlpo Järvinen __skb_trim(skb, len); 1925419ae74eSIlpo Järvinen } 1926419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1927419ae74eSIlpo Järvinen 19283cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 19291da177e4SLinus Torvalds */ 19301da177e4SLinus Torvalds 19313cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 19321da177e4SLinus Torvalds { 193327b437c8SHerbert Xu struct sk_buff **fragp; 193427b437c8SHerbert Xu struct sk_buff *frag; 19351da177e4SLinus Torvalds int offset = skb_headlen(skb); 19361da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 19371da177e4SLinus Torvalds int i; 193827b437c8SHerbert Xu int err; 193927b437c8SHerbert Xu 194027b437c8SHerbert Xu if (skb_cloned(skb) && 194127b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 194227b437c8SHerbert Xu return err; 19431da177e4SLinus Torvalds 1944f4d26fb3SHerbert Xu i = 0; 1945f4d26fb3SHerbert Xu if (offset >= len) 1946f4d26fb3SHerbert Xu goto drop_pages; 1947f4d26fb3SHerbert Xu 1948f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 19499e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 195027b437c8SHerbert Xu 195127b437c8SHerbert Xu if (end < len) { 19521da177e4SLinus Torvalds offset = end; 195327b437c8SHerbert Xu continue; 19541da177e4SLinus Torvalds } 19551da177e4SLinus Torvalds 19569e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 195727b437c8SHerbert Xu 1958f4d26fb3SHerbert Xu drop_pages: 195927b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 196027b437c8SHerbert Xu 196127b437c8SHerbert Xu for (; i < nfrags; i++) 1962ea2ab693SIan Campbell skb_frag_unref(skb, i); 196327b437c8SHerbert Xu 196421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 196527b437c8SHerbert Xu skb_drop_fraglist(skb); 1966f4d26fb3SHerbert Xu goto done; 196727b437c8SHerbert Xu } 196827b437c8SHerbert Xu 196927b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 197027b437c8SHerbert Xu fragp = &frag->next) { 197127b437c8SHerbert Xu int end = offset + frag->len; 197227b437c8SHerbert Xu 197327b437c8SHerbert Xu if (skb_shared(frag)) { 197427b437c8SHerbert Xu struct sk_buff *nfrag; 197527b437c8SHerbert Xu 197627b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 197727b437c8SHerbert Xu if (unlikely(!nfrag)) 197827b437c8SHerbert Xu return -ENOMEM; 197927b437c8SHerbert Xu 198027b437c8SHerbert Xu nfrag->next = frag->next; 198185bb2a60SEric Dumazet consume_skb(frag); 198227b437c8SHerbert Xu frag = nfrag; 198327b437c8SHerbert Xu *fragp = frag; 198427b437c8SHerbert Xu } 198527b437c8SHerbert Xu 198627b437c8SHerbert Xu if (end < len) { 198727b437c8SHerbert Xu offset = end; 198827b437c8SHerbert Xu continue; 198927b437c8SHerbert Xu } 199027b437c8SHerbert Xu 199127b437c8SHerbert Xu if (end > len && 199227b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 199327b437c8SHerbert Xu return err; 199427b437c8SHerbert Xu 199527b437c8SHerbert Xu if (frag->next) 199627b437c8SHerbert Xu skb_drop_list(&frag->next); 199727b437c8SHerbert Xu break; 199827b437c8SHerbert Xu } 199927b437c8SHerbert Xu 2000f4d26fb3SHerbert Xu done: 200127b437c8SHerbert Xu if (len > skb_headlen(skb)) { 20021da177e4SLinus Torvalds skb->data_len -= skb->len - len; 20031da177e4SLinus Torvalds skb->len = len; 20041da177e4SLinus Torvalds } else { 20051da177e4SLinus Torvalds skb->len = len; 20061da177e4SLinus Torvalds skb->data_len = 0; 200727a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 20081da177e4SLinus Torvalds } 20091da177e4SLinus Torvalds 2010c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2011c21b48ccSEric Dumazet skb_condense(skb); 20121da177e4SLinus Torvalds return 0; 20131da177e4SLinus Torvalds } 2014b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 20151da177e4SLinus Torvalds 201688078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 201788078d98SEric Dumazet */ 201888078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 201988078d98SEric Dumazet { 202088078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 202188078d98SEric Dumazet int delta = skb->len - len; 202288078d98SEric Dumazet 2023d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2024d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2025d55bef50SDimitris Michailidis len); 202654970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 202754970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 202854970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 202954970a2fSVasily Averin 203054970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 203154970a2fSVasily Averin return -EINVAL; 203288078d98SEric Dumazet } 203388078d98SEric Dumazet return __pskb_trim(skb, len); 203488078d98SEric Dumazet } 203588078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 203688078d98SEric Dumazet 20371da177e4SLinus Torvalds /** 20381da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 20391da177e4SLinus Torvalds * @skb: buffer to reallocate 20401da177e4SLinus Torvalds * @delta: number of bytes to advance tail 20411da177e4SLinus Torvalds * 20421da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 20431da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 20441da177e4SLinus Torvalds * data from fragmented part. 20451da177e4SLinus Torvalds * 20461da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 20471da177e4SLinus Torvalds * 20481da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 20491da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 20501da177e4SLinus Torvalds * 20511da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 20521da177e4SLinus Torvalds * reloaded after call to this function. 20531da177e4SLinus Torvalds */ 20541da177e4SLinus Torvalds 20551da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 20561da177e4SLinus Torvalds * when it is necessary. 20571da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 20581da177e4SLinus Torvalds * 2. It may change skb pointers. 20591da177e4SLinus Torvalds * 20601da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 20611da177e4SLinus Torvalds */ 2062af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 20631da177e4SLinus Torvalds { 20641da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 20651da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 20661da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 20671da177e4SLinus Torvalds */ 20684305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 20691da177e4SLinus Torvalds 20701da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 20711da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 20721da177e4SLinus Torvalds GFP_ATOMIC)) 20731da177e4SLinus Torvalds return NULL; 20741da177e4SLinus Torvalds } 20751da177e4SLinus Torvalds 20769f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 20779f77fad3STim Hansen skb_tail_pointer(skb), delta)); 20781da177e4SLinus Torvalds 20791da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 20801da177e4SLinus Torvalds * size of pulled pages. Superb. 20811da177e4SLinus Torvalds */ 208221dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 20831da177e4SLinus Torvalds goto pull_pages; 20841da177e4SLinus Torvalds 20851da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 20861da177e4SLinus Torvalds eat = delta; 20871da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20889e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 20899e903e08SEric Dumazet 20909e903e08SEric Dumazet if (size >= eat) 20911da177e4SLinus Torvalds goto pull_pages; 20929e903e08SEric Dumazet eat -= size; 20931da177e4SLinus Torvalds } 20941da177e4SLinus Torvalds 20951da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 209609001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 20971da177e4SLinus Torvalds * but taking into account that pulling is expected to 20981da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 20991da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 21001da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 21011da177e4SLinus Torvalds */ 21021da177e4SLinus Torvalds if (eat) { 21031da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 21041da177e4SLinus Torvalds struct sk_buff *clone = NULL; 21051da177e4SLinus Torvalds struct sk_buff *insp = NULL; 21061da177e4SLinus Torvalds 21071da177e4SLinus Torvalds do { 21081da177e4SLinus Torvalds if (list->len <= eat) { 21091da177e4SLinus Torvalds /* Eaten as whole. */ 21101da177e4SLinus Torvalds eat -= list->len; 21111da177e4SLinus Torvalds list = list->next; 21121da177e4SLinus Torvalds insp = list; 21131da177e4SLinus Torvalds } else { 21141da177e4SLinus Torvalds /* Eaten partially. */ 21151da177e4SLinus Torvalds 21161da177e4SLinus Torvalds if (skb_shared(list)) { 21171da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 21181da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 21191da177e4SLinus Torvalds if (!clone) 21201da177e4SLinus Torvalds return NULL; 21211da177e4SLinus Torvalds insp = list->next; 21221da177e4SLinus Torvalds list = clone; 21231da177e4SLinus Torvalds } else { 21241da177e4SLinus Torvalds /* This may be pulled without 21251da177e4SLinus Torvalds * problems. */ 21261da177e4SLinus Torvalds insp = list; 21271da177e4SLinus Torvalds } 21281da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 21291da177e4SLinus Torvalds kfree_skb(clone); 21301da177e4SLinus Torvalds return NULL; 21311da177e4SLinus Torvalds } 21321da177e4SLinus Torvalds break; 21331da177e4SLinus Torvalds } 21341da177e4SLinus Torvalds } while (eat); 21351da177e4SLinus Torvalds 21361da177e4SLinus Torvalds /* Free pulled out fragments. */ 21371da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 21381da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 21391da177e4SLinus Torvalds kfree_skb(list); 21401da177e4SLinus Torvalds } 21411da177e4SLinus Torvalds /* And insert new clone at head. */ 21421da177e4SLinus Torvalds if (clone) { 21431da177e4SLinus Torvalds clone->next = list; 21441da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds } 21471da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 21481da177e4SLinus Torvalds 21491da177e4SLinus Torvalds pull_pages: 21501da177e4SLinus Torvalds eat = delta; 21511da177e4SLinus Torvalds k = 0; 21521da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21539e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 21549e903e08SEric Dumazet 21559e903e08SEric Dumazet if (size <= eat) { 2156ea2ab693SIan Campbell skb_frag_unref(skb, i); 21579e903e08SEric Dumazet eat -= size; 21581da177e4SLinus Torvalds } else { 2159b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2160b54c9d5bSJonathan Lemon 2161b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 21621da177e4SLinus Torvalds if (eat) { 2163b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2164b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 21653ccc6c6fSlinzhang if (!i) 21663ccc6c6fSlinzhang goto end; 21671da177e4SLinus Torvalds eat = 0; 21681da177e4SLinus Torvalds } 21691da177e4SLinus Torvalds k++; 21701da177e4SLinus Torvalds } 21711da177e4SLinus Torvalds } 21721da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 21731da177e4SLinus Torvalds 21743ccc6c6fSlinzhang end: 21751da177e4SLinus Torvalds skb->tail += delta; 21761da177e4SLinus Torvalds skb->data_len -= delta; 21771da177e4SLinus Torvalds 21781f8b977aSWillem de Bruijn if (!skb->data_len) 21791f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 21801f8b977aSWillem de Bruijn 218127a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 21821da177e4SLinus Torvalds } 2183b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 21841da177e4SLinus Torvalds 218522019b17SEric Dumazet /** 218622019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 218722019b17SEric Dumazet * @skb: source skb 218822019b17SEric Dumazet * @offset: offset in source 218922019b17SEric Dumazet * @to: destination buffer 219022019b17SEric Dumazet * @len: number of bytes to copy 219122019b17SEric Dumazet * 219222019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 219322019b17SEric Dumazet * destination buffer. 219422019b17SEric Dumazet * 219522019b17SEric Dumazet * CAUTION ! : 219622019b17SEric Dumazet * If its prototype is ever changed, 219722019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 219822019b17SEric Dumazet * since it is called from BPF assembly code. 219922019b17SEric Dumazet */ 22001da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 22011da177e4SLinus Torvalds { 22021a028e50SDavid S. Miller int start = skb_headlen(skb); 2203fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2204fbb398a8SDavid S. Miller int i, copy; 22051da177e4SLinus Torvalds 22061da177e4SLinus Torvalds if (offset > (int)skb->len - len) 22071da177e4SLinus Torvalds goto fault; 22081da177e4SLinus Torvalds 22091da177e4SLinus Torvalds /* Copy header. */ 22101a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 22111da177e4SLinus Torvalds if (copy > len) 22121da177e4SLinus Torvalds copy = len; 2213d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 22141da177e4SLinus Torvalds if ((len -= copy) == 0) 22151da177e4SLinus Torvalds return 0; 22161da177e4SLinus Torvalds offset += copy; 22171da177e4SLinus Torvalds to += copy; 22181da177e4SLinus Torvalds } 22191da177e4SLinus Torvalds 22201da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22211a028e50SDavid S. Miller int end; 222251c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 22231da177e4SLinus Torvalds 2224547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22251a028e50SDavid S. Miller 222651c56b00SEric Dumazet end = start + skb_frag_size(f); 22271da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2228c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2229c613c209SWillem de Bruijn struct page *p; 22301da177e4SLinus Torvalds u8 *vaddr; 22311da177e4SLinus Torvalds 22321da177e4SLinus Torvalds if (copy > len) 22331da177e4SLinus Torvalds copy = len; 22341da177e4SLinus Torvalds 2235c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2236b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2237c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2238c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2239c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 224051c56b00SEric Dumazet kunmap_atomic(vaddr); 2241c613c209SWillem de Bruijn } 22421da177e4SLinus Torvalds 22431da177e4SLinus Torvalds if ((len -= copy) == 0) 22441da177e4SLinus Torvalds return 0; 22451da177e4SLinus Torvalds offset += copy; 22461da177e4SLinus Torvalds to += copy; 22471da177e4SLinus Torvalds } 22481a028e50SDavid S. Miller start = end; 22491da177e4SLinus Torvalds } 22501da177e4SLinus Torvalds 2251fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 22521a028e50SDavid S. Miller int end; 22531da177e4SLinus Torvalds 2254547b792cSIlpo Järvinen WARN_ON(start > offset + len); 22551a028e50SDavid S. Miller 2256fbb398a8SDavid S. Miller end = start + frag_iter->len; 22571da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 22581da177e4SLinus Torvalds if (copy > len) 22591da177e4SLinus Torvalds copy = len; 2260fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 22611da177e4SLinus Torvalds goto fault; 22621da177e4SLinus Torvalds if ((len -= copy) == 0) 22631da177e4SLinus Torvalds return 0; 22641da177e4SLinus Torvalds offset += copy; 22651da177e4SLinus Torvalds to += copy; 22661da177e4SLinus Torvalds } 22671a028e50SDavid S. Miller start = end; 22681da177e4SLinus Torvalds } 2269a6686f2fSShirley Ma 22701da177e4SLinus Torvalds if (!len) 22711da177e4SLinus Torvalds return 0; 22721da177e4SLinus Torvalds 22731da177e4SLinus Torvalds fault: 22741da177e4SLinus Torvalds return -EFAULT; 22751da177e4SLinus Torvalds } 2276b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 22771da177e4SLinus Torvalds 22789c55e01cSJens Axboe /* 22799c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 22809c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 22819c55e01cSJens Axboe */ 22829c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 22839c55e01cSJens Axboe { 22848b9d3728SJarek Poplawski put_page(spd->pages[i]); 22858b9d3728SJarek Poplawski } 22869c55e01cSJens Axboe 2287a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 22884fb66994SJarek Poplawski unsigned int *offset, 228918aafc62SEric Dumazet struct sock *sk) 22908b9d3728SJarek Poplawski { 22915640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 22928b9d3728SJarek Poplawski 22935640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 22948b9d3728SJarek Poplawski return NULL; 22954fb66994SJarek Poplawski 22965640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 22974fb66994SJarek Poplawski 22985640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 22995640f768SEric Dumazet page_address(page) + *offset, *len); 23005640f768SEric Dumazet *offset = pfrag->offset; 23015640f768SEric Dumazet pfrag->offset += *len; 23024fb66994SJarek Poplawski 23035640f768SEric Dumazet return pfrag->page; 23049c55e01cSJens Axboe } 23059c55e01cSJens Axboe 230641c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 230741c73a0dSEric Dumazet struct page *page, 230841c73a0dSEric Dumazet unsigned int offset) 230941c73a0dSEric Dumazet { 231041c73a0dSEric Dumazet return spd->nr_pages && 231141c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 231241c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 231341c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 231441c73a0dSEric Dumazet } 231541c73a0dSEric Dumazet 23169c55e01cSJens Axboe /* 23179c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 23189c55e01cSJens Axboe */ 2319a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 232035f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 23214fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 232218aafc62SEric Dumazet bool linear, 23237a67e56fSJarek Poplawski struct sock *sk) 23249c55e01cSJens Axboe { 232541c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2326a108d5f3SDavid S. Miller return true; 23279c55e01cSJens Axboe 23288b9d3728SJarek Poplawski if (linear) { 232918aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 23308b9d3728SJarek Poplawski if (!page) 2331a108d5f3SDavid S. Miller return true; 233241c73a0dSEric Dumazet } 233341c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 233441c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2335a108d5f3SDavid S. Miller return false; 233641c73a0dSEric Dumazet } 23378b9d3728SJarek Poplawski get_page(page); 23389c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 23394fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 23409c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 23419c55e01cSJens Axboe spd->nr_pages++; 23428b9d3728SJarek Poplawski 2343a108d5f3SDavid S. Miller return false; 23449c55e01cSJens Axboe } 23459c55e01cSJens Axboe 2346a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 23472870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 234818aafc62SEric Dumazet unsigned int *len, 2349d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 235035f3d14dSJens Axboe struct sock *sk, 235135f3d14dSJens Axboe struct pipe_inode_info *pipe) 23529c55e01cSJens Axboe { 23532870c43dSOctavian Purdila if (!*len) 2354a108d5f3SDavid S. Miller return true; 23559c55e01cSJens Axboe 23562870c43dSOctavian Purdila /* skip this segment if already processed */ 23572870c43dSOctavian Purdila if (*off >= plen) { 23582870c43dSOctavian Purdila *off -= plen; 2359a108d5f3SDavid S. Miller return false; 23602870c43dSOctavian Purdila } 23612870c43dSOctavian Purdila 23622870c43dSOctavian Purdila /* ignore any bits we already processed */ 23639ca1b22dSEric Dumazet poff += *off; 23649ca1b22dSEric Dumazet plen -= *off; 23652870c43dSOctavian Purdila *off = 0; 23662870c43dSOctavian Purdila 236718aafc62SEric Dumazet do { 236818aafc62SEric Dumazet unsigned int flen = min(*len, plen); 23692870c43dSOctavian Purdila 237018aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 237118aafc62SEric Dumazet linear, sk)) 2372a108d5f3SDavid S. Miller return true; 237318aafc62SEric Dumazet poff += flen; 237418aafc62SEric Dumazet plen -= flen; 23752870c43dSOctavian Purdila *len -= flen; 237618aafc62SEric Dumazet } while (*len && plen); 23772870c43dSOctavian Purdila 2378a108d5f3SDavid S. Miller return false; 2379db43a282SOctavian Purdila } 23809c55e01cSJens Axboe 23819c55e01cSJens Axboe /* 2382a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 23832870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 23849c55e01cSJens Axboe */ 2385a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 238635f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 238735f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 23882870c43dSOctavian Purdila { 23892870c43dSOctavian Purdila int seg; 2390fa9835e5STom Herbert struct sk_buff *iter; 23919c55e01cSJens Axboe 23921d0c0b32SEric Dumazet /* map the linear part : 23932996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 23942996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 23952996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 23969c55e01cSJens Axboe */ 23972870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 23982870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 23992870c43dSOctavian Purdila skb_headlen(skb), 240018aafc62SEric Dumazet offset, len, spd, 24013a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 24021d0c0b32SEric Dumazet sk, pipe)) 2403a108d5f3SDavid S. Miller return true; 24049c55e01cSJens Axboe 24059c55e01cSJens Axboe /* 24069c55e01cSJens Axboe * then map the fragments 24079c55e01cSJens Axboe */ 24089c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 24099c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 24109c55e01cSJens Axboe 2411ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2412b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 241318aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2414a108d5f3SDavid S. Miller return true; 24159c55e01cSJens Axboe } 24169c55e01cSJens Axboe 2417fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2418fa9835e5STom Herbert if (*offset >= iter->len) { 2419fa9835e5STom Herbert *offset -= iter->len; 2420fa9835e5STom Herbert continue; 2421fa9835e5STom Herbert } 2422fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2423fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2424fa9835e5STom Herbert * case. 2425fa9835e5STom Herbert */ 2426fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2427fa9835e5STom Herbert return true; 2428fa9835e5STom Herbert } 2429fa9835e5STom Herbert 2430a108d5f3SDavid S. Miller return false; 24319c55e01cSJens Axboe } 24329c55e01cSJens Axboe 24339c55e01cSJens Axboe /* 24349c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2435fa9835e5STom Herbert * the fragments, and the frag list. 24369c55e01cSJens Axboe */ 2437a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 24389c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 243925869262SAl Viro unsigned int flags) 24409c55e01cSJens Axboe { 244141c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 244241c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 24439c55e01cSJens Axboe struct splice_pipe_desc spd = { 24449c55e01cSJens Axboe .pages = pages, 24459c55e01cSJens Axboe .partial = partial, 2446047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 244728a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 24489c55e01cSJens Axboe .spd_release = sock_spd_release, 24499c55e01cSJens Axboe }; 245035f3d14dSJens Axboe int ret = 0; 245135f3d14dSJens Axboe 2452fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 24539c55e01cSJens Axboe 2454a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 245525869262SAl Viro ret = splice_to_pipe(pipe, &spd); 24569c55e01cSJens Axboe 245735f3d14dSJens Axboe return ret; 24589c55e01cSJens Axboe } 24592b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 24609c55e01cSJens Axboe 246120bf50deSTom Herbert /* Send skb data on a socket. Socket must be locked. */ 246220bf50deSTom Herbert int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 246320bf50deSTom Herbert int len) 246420bf50deSTom Herbert { 246520bf50deSTom Herbert unsigned int orig_len = len; 246620bf50deSTom Herbert struct sk_buff *head = skb; 246720bf50deSTom Herbert unsigned short fragidx; 246820bf50deSTom Herbert int slen, ret; 246920bf50deSTom Herbert 247020bf50deSTom Herbert do_frag_list: 247120bf50deSTom Herbert 247220bf50deSTom Herbert /* Deal with head data */ 247320bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 247420bf50deSTom Herbert struct kvec kv; 247520bf50deSTom Herbert struct msghdr msg; 247620bf50deSTom Herbert 247720bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 247820bf50deSTom Herbert kv.iov_base = skb->data + offset; 2479db5980d8SJohn Fastabend kv.iov_len = slen; 248020bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2481bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 248220bf50deSTom Herbert 248320bf50deSTom Herbert ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen); 248420bf50deSTom Herbert if (ret <= 0) 248520bf50deSTom Herbert goto error; 248620bf50deSTom Herbert 248720bf50deSTom Herbert offset += ret; 248820bf50deSTom Herbert len -= ret; 248920bf50deSTom Herbert } 249020bf50deSTom Herbert 249120bf50deSTom Herbert /* All the data was skb head? */ 249220bf50deSTom Herbert if (!len) 249320bf50deSTom Herbert goto out; 249420bf50deSTom Herbert 249520bf50deSTom Herbert /* Make offset relative to start of frags */ 249620bf50deSTom Herbert offset -= skb_headlen(skb); 249720bf50deSTom Herbert 249820bf50deSTom Herbert /* Find where we are in frag list */ 249920bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 250020bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 250120bf50deSTom Herbert 2502d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 250320bf50deSTom Herbert break; 250420bf50deSTom Herbert 2505d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 250620bf50deSTom Herbert } 250720bf50deSTom Herbert 250820bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 250920bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 251020bf50deSTom Herbert 2511d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 251220bf50deSTom Herbert 251320bf50deSTom Herbert while (slen) { 2514d8e18a51SMatthew Wilcox (Oracle) ret = kernel_sendpage_locked(sk, skb_frag_page(frag), 2515b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 251620bf50deSTom Herbert slen, MSG_DONTWAIT); 251720bf50deSTom Herbert if (ret <= 0) 251820bf50deSTom Herbert goto error; 251920bf50deSTom Herbert 252020bf50deSTom Herbert len -= ret; 252120bf50deSTom Herbert offset += ret; 252220bf50deSTom Herbert slen -= ret; 252320bf50deSTom Herbert } 252420bf50deSTom Herbert 252520bf50deSTom Herbert offset = 0; 252620bf50deSTom Herbert } 252720bf50deSTom Herbert 252820bf50deSTom Herbert if (len) { 252920bf50deSTom Herbert /* Process any frag lists */ 253020bf50deSTom Herbert 253120bf50deSTom Herbert if (skb == head) { 253220bf50deSTom Herbert if (skb_has_frag_list(skb)) { 253320bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 253420bf50deSTom Herbert goto do_frag_list; 253520bf50deSTom Herbert } 253620bf50deSTom Herbert } else if (skb->next) { 253720bf50deSTom Herbert skb = skb->next; 253820bf50deSTom Herbert goto do_frag_list; 253920bf50deSTom Herbert } 254020bf50deSTom Herbert } 254120bf50deSTom Herbert 254220bf50deSTom Herbert out: 254320bf50deSTom Herbert return orig_len - len; 254420bf50deSTom Herbert 254520bf50deSTom Herbert error: 254620bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 254720bf50deSTom Herbert } 254820bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 254920bf50deSTom Herbert 2550357b40a1SHerbert Xu /** 2551357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2552357b40a1SHerbert Xu * @skb: destination buffer 2553357b40a1SHerbert Xu * @offset: offset in destination 2554357b40a1SHerbert Xu * @from: source buffer 2555357b40a1SHerbert Xu * @len: number of bytes to copy 2556357b40a1SHerbert Xu * 2557357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2558357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2559357b40a1SHerbert Xu * traversing fragment lists and such. 2560357b40a1SHerbert Xu */ 2561357b40a1SHerbert Xu 25620c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2563357b40a1SHerbert Xu { 25641a028e50SDavid S. Miller int start = skb_headlen(skb); 2565fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2566fbb398a8SDavid S. Miller int i, copy; 2567357b40a1SHerbert Xu 2568357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2569357b40a1SHerbert Xu goto fault; 2570357b40a1SHerbert Xu 25711a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2572357b40a1SHerbert Xu if (copy > len) 2573357b40a1SHerbert Xu copy = len; 257427d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2575357b40a1SHerbert Xu if ((len -= copy) == 0) 2576357b40a1SHerbert Xu return 0; 2577357b40a1SHerbert Xu offset += copy; 2578357b40a1SHerbert Xu from += copy; 2579357b40a1SHerbert Xu } 2580357b40a1SHerbert Xu 2581357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2582357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 25831a028e50SDavid S. Miller int end; 2584357b40a1SHerbert Xu 2585547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25861a028e50SDavid S. Miller 25879e903e08SEric Dumazet end = start + skb_frag_size(frag); 2588357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2589c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2590c613c209SWillem de Bruijn struct page *p; 2591357b40a1SHerbert Xu u8 *vaddr; 2592357b40a1SHerbert Xu 2593357b40a1SHerbert Xu if (copy > len) 2594357b40a1SHerbert Xu copy = len; 2595357b40a1SHerbert Xu 2596c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2597b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2598c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2599c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2600c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 260151c56b00SEric Dumazet kunmap_atomic(vaddr); 2602c613c209SWillem de Bruijn } 2603357b40a1SHerbert Xu 2604357b40a1SHerbert Xu if ((len -= copy) == 0) 2605357b40a1SHerbert Xu return 0; 2606357b40a1SHerbert Xu offset += copy; 2607357b40a1SHerbert Xu from += copy; 2608357b40a1SHerbert Xu } 26091a028e50SDavid S. Miller start = end; 2610357b40a1SHerbert Xu } 2611357b40a1SHerbert Xu 2612fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 26131a028e50SDavid S. Miller int end; 2614357b40a1SHerbert Xu 2615547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26161a028e50SDavid S. Miller 2617fbb398a8SDavid S. Miller end = start + frag_iter->len; 2618357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2619357b40a1SHerbert Xu if (copy > len) 2620357b40a1SHerbert Xu copy = len; 2621fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 26221a028e50SDavid S. Miller from, copy)) 2623357b40a1SHerbert Xu goto fault; 2624357b40a1SHerbert Xu if ((len -= copy) == 0) 2625357b40a1SHerbert Xu return 0; 2626357b40a1SHerbert Xu offset += copy; 2627357b40a1SHerbert Xu from += copy; 2628357b40a1SHerbert Xu } 26291a028e50SDavid S. Miller start = end; 2630357b40a1SHerbert Xu } 2631357b40a1SHerbert Xu if (!len) 2632357b40a1SHerbert Xu return 0; 2633357b40a1SHerbert Xu 2634357b40a1SHerbert Xu fault: 2635357b40a1SHerbert Xu return -EFAULT; 2636357b40a1SHerbert Xu } 2637357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2638357b40a1SHerbert Xu 26391da177e4SLinus Torvalds /* Checksum skb data. */ 26402817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 26412817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 26421da177e4SLinus Torvalds { 26431a028e50SDavid S. Miller int start = skb_headlen(skb); 26441a028e50SDavid S. Miller int i, copy = start - offset; 2645fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 26461da177e4SLinus Torvalds int pos = 0; 26471da177e4SLinus Torvalds 26481da177e4SLinus Torvalds /* Checksum header. */ 26491da177e4SLinus Torvalds if (copy > 0) { 26501da177e4SLinus Torvalds if (copy > len) 26511da177e4SLinus Torvalds copy = len; 26522544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 26532544af03SMatteo Croce skb->data + offset, copy, csum); 26541da177e4SLinus Torvalds if ((len -= copy) == 0) 26551da177e4SLinus Torvalds return csum; 26561da177e4SLinus Torvalds offset += copy; 26571da177e4SLinus Torvalds pos = copy; 26581da177e4SLinus Torvalds } 26591da177e4SLinus Torvalds 26601da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 26611a028e50SDavid S. Miller int end; 266251c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 26631da177e4SLinus Torvalds 2664547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26651a028e50SDavid S. Miller 266651c56b00SEric Dumazet end = start + skb_frag_size(frag); 26671da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2668c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2669c613c209SWillem de Bruijn struct page *p; 267044bb9363SAl Viro __wsum csum2; 26711da177e4SLinus Torvalds u8 *vaddr; 26721da177e4SLinus Torvalds 26731da177e4SLinus Torvalds if (copy > len) 26741da177e4SLinus Torvalds copy = len; 2675c613c209SWillem de Bruijn 2676c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2677b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2678c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2679c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 26802544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 26812544af03SMatteo Croce csum_partial_ext, 26822544af03SMatteo Croce vaddr + p_off, p_len, 0); 268351c56b00SEric Dumazet kunmap_atomic(vaddr); 26842544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 26852544af03SMatteo Croce csum_block_add_ext, csum, 26862544af03SMatteo Croce csum2, pos, p_len); 2687c613c209SWillem de Bruijn pos += p_len; 2688c613c209SWillem de Bruijn } 2689c613c209SWillem de Bruijn 26901da177e4SLinus Torvalds if (!(len -= copy)) 26911da177e4SLinus Torvalds return csum; 26921da177e4SLinus Torvalds offset += copy; 26931da177e4SLinus Torvalds } 26941a028e50SDavid S. Miller start = end; 26951da177e4SLinus Torvalds } 26961da177e4SLinus Torvalds 2697fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 26981a028e50SDavid S. Miller int end; 26991da177e4SLinus Torvalds 2700547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27011a028e50SDavid S. Miller 2702fbb398a8SDavid S. Miller end = start + frag_iter->len; 27031da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 27045f92a738SAl Viro __wsum csum2; 27051da177e4SLinus Torvalds if (copy > len) 27061da177e4SLinus Torvalds copy = len; 27072817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 27082817a336SDaniel Borkmann copy, 0, ops); 27092544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 27102544af03SMatteo Croce csum, csum2, pos, copy); 27111da177e4SLinus Torvalds if ((len -= copy) == 0) 27121da177e4SLinus Torvalds return csum; 27131da177e4SLinus Torvalds offset += copy; 27141da177e4SLinus Torvalds pos += copy; 27151da177e4SLinus Torvalds } 27161a028e50SDavid S. Miller start = end; 27171da177e4SLinus Torvalds } 271809a62660SKris Katterjohn BUG_ON(len); 27191da177e4SLinus Torvalds 27201da177e4SLinus Torvalds return csum; 27211da177e4SLinus Torvalds } 27222817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 27232817a336SDaniel Borkmann 27242817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 27252817a336SDaniel Borkmann int len, __wsum csum) 27262817a336SDaniel Borkmann { 27272817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2728cea80ea8SDaniel Borkmann .update = csum_partial_ext, 27292817a336SDaniel Borkmann .combine = csum_block_add_ext, 27302817a336SDaniel Borkmann }; 27312817a336SDaniel Borkmann 27322817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 27332817a336SDaniel Borkmann } 2734b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 27351da177e4SLinus Torvalds 27361da177e4SLinus Torvalds /* Both of above in one bottle. */ 27371da177e4SLinus Torvalds 273881d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 27398d5930dfSAl Viro u8 *to, int len) 27401da177e4SLinus Torvalds { 27411a028e50SDavid S. Miller int start = skb_headlen(skb); 27421a028e50SDavid S. Miller int i, copy = start - offset; 2743fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 27441da177e4SLinus Torvalds int pos = 0; 27458d5930dfSAl Viro __wsum csum = 0; 27461da177e4SLinus Torvalds 27471da177e4SLinus Torvalds /* Copy header. */ 27481da177e4SLinus Torvalds if (copy > 0) { 27491da177e4SLinus Torvalds if (copy > len) 27501da177e4SLinus Torvalds copy = len; 27511da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2752cc44c17bSAl Viro copy); 27531da177e4SLinus Torvalds if ((len -= copy) == 0) 27541da177e4SLinus Torvalds return csum; 27551da177e4SLinus Torvalds offset += copy; 27561da177e4SLinus Torvalds to += copy; 27571da177e4SLinus Torvalds pos = copy; 27581da177e4SLinus Torvalds } 27591da177e4SLinus Torvalds 27601da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 27611a028e50SDavid S. Miller int end; 27621da177e4SLinus Torvalds 2763547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27641a028e50SDavid S. Miller 27659e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 27661da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2767c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2768c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2769c613c209SWillem de Bruijn struct page *p; 27705084205fSAl Viro __wsum csum2; 27711da177e4SLinus Torvalds u8 *vaddr; 27721da177e4SLinus Torvalds 27731da177e4SLinus Torvalds if (copy > len) 27741da177e4SLinus Torvalds copy = len; 2775c613c209SWillem de Bruijn 2776c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2777b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2778c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2779c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2780c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2781c613c209SWillem de Bruijn to + copied, 2782cc44c17bSAl Viro p_len); 278351c56b00SEric Dumazet kunmap_atomic(vaddr); 27841da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2785c613c209SWillem de Bruijn pos += p_len; 2786c613c209SWillem de Bruijn } 2787c613c209SWillem de Bruijn 27881da177e4SLinus Torvalds if (!(len -= copy)) 27891da177e4SLinus Torvalds return csum; 27901da177e4SLinus Torvalds offset += copy; 27911da177e4SLinus Torvalds to += copy; 27921da177e4SLinus Torvalds } 27931a028e50SDavid S. Miller start = end; 27941da177e4SLinus Torvalds } 27951da177e4SLinus Torvalds 2796fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 279781d77662SAl Viro __wsum csum2; 27981a028e50SDavid S. Miller int end; 27991da177e4SLinus Torvalds 2800547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28011a028e50SDavid S. Miller 2802fbb398a8SDavid S. Miller end = start + frag_iter->len; 28031da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 28041da177e4SLinus Torvalds if (copy > len) 28051da177e4SLinus Torvalds copy = len; 2806fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 28071a028e50SDavid S. Miller offset - start, 28088d5930dfSAl Viro to, copy); 28091da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 28101da177e4SLinus Torvalds if ((len -= copy) == 0) 28111da177e4SLinus Torvalds return csum; 28121da177e4SLinus Torvalds offset += copy; 28131da177e4SLinus Torvalds to += copy; 28141da177e4SLinus Torvalds pos += copy; 28151da177e4SLinus Torvalds } 28161a028e50SDavid S. Miller start = end; 28171da177e4SLinus Torvalds } 281809a62660SKris Katterjohn BUG_ON(len); 28191da177e4SLinus Torvalds return csum; 28201da177e4SLinus Torvalds } 2821b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 28221da177e4SLinus Torvalds 282349f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 282449f8e832SCong Wang { 282549f8e832SCong Wang __sum16 sum; 282649f8e832SCong Wang 282749f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 282814641931SCong Wang /* See comments in __skb_checksum_complete(). */ 282949f8e832SCong Wang if (likely(!sum)) { 283049f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 283149f8e832SCong Wang !skb->csum_complete_sw) 28327fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 283349f8e832SCong Wang } 283449f8e832SCong Wang if (!skb_shared(skb)) 283549f8e832SCong Wang skb->csum_valid = !sum; 283649f8e832SCong Wang return sum; 283749f8e832SCong Wang } 283849f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 283949f8e832SCong Wang 284014641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 284114641931SCong Wang * which has been changed from the hardware checksum, for example, by 284214641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 284314641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 284414641931SCong Wang * 284514641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 284614641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 284714641931SCong Wang * shared. 284814641931SCong Wang */ 284949f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 285049f8e832SCong Wang { 285149f8e832SCong Wang __wsum csum; 285249f8e832SCong Wang __sum16 sum; 285349f8e832SCong Wang 285449f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 285549f8e832SCong Wang 285649f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 285714641931SCong Wang /* This check is inverted, because we already knew the hardware 285814641931SCong Wang * checksum is invalid before calling this function. So, if the 285914641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 286014641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 286114641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 286214641931SCong Wang * when moving skb->data around. 286314641931SCong Wang */ 286449f8e832SCong Wang if (likely(!sum)) { 286549f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 286649f8e832SCong Wang !skb->csum_complete_sw) 28677fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 286849f8e832SCong Wang } 286949f8e832SCong Wang 287049f8e832SCong Wang if (!skb_shared(skb)) { 287149f8e832SCong Wang /* Save full packet checksum */ 287249f8e832SCong Wang skb->csum = csum; 287349f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 287449f8e832SCong Wang skb->csum_complete_sw = 1; 287549f8e832SCong Wang skb->csum_valid = !sum; 287649f8e832SCong Wang } 287749f8e832SCong Wang 287849f8e832SCong Wang return sum; 287949f8e832SCong Wang } 288049f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 288149f8e832SCong Wang 28829617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 28839617813dSDavide Caratti { 28849617813dSDavide Caratti net_warn_ratelimited( 28859617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 28869617813dSDavide Caratti __func__); 28879617813dSDavide Caratti return 0; 28889617813dSDavide Caratti } 28899617813dSDavide Caratti 28909617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 28919617813dSDavide Caratti int offset, int len) 28929617813dSDavide Caratti { 28939617813dSDavide Caratti net_warn_ratelimited( 28949617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 28959617813dSDavide Caratti __func__); 28969617813dSDavide Caratti return 0; 28979617813dSDavide Caratti } 28989617813dSDavide Caratti 28999617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 29009617813dSDavide Caratti .update = warn_crc32c_csum_update, 29019617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 29029617813dSDavide Caratti }; 29039617813dSDavide Caratti 29049617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 29059617813dSDavide Caratti &default_crc32c_ops; 29069617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 29079617813dSDavide Caratti 2908af2806f8SThomas Graf /** 2909af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2910af2806f8SThomas Graf * @from: source buffer 2911af2806f8SThomas Graf * 2912af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2913af2806f8SThomas Graf * into skb_zerocopy(). 2914af2806f8SThomas Graf */ 2915af2806f8SThomas Graf unsigned int 2916af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2917af2806f8SThomas Graf { 2918af2806f8SThomas Graf unsigned int hlen = 0; 2919af2806f8SThomas Graf 2920af2806f8SThomas Graf if (!from->head_frag || 2921af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2922af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2923af2806f8SThomas Graf hlen = skb_headlen(from); 2924af2806f8SThomas Graf 2925af2806f8SThomas Graf if (skb_has_frag_list(from)) 2926af2806f8SThomas Graf hlen = from->len; 2927af2806f8SThomas Graf 2928af2806f8SThomas Graf return hlen; 2929af2806f8SThomas Graf } 2930af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2931af2806f8SThomas Graf 2932af2806f8SThomas Graf /** 2933af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2934af2806f8SThomas Graf * @to: destination buffer 29357fceb4deSMasanari Iida * @from: source buffer 2936af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2937af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2938af2806f8SThomas Graf * 2939af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2940af2806f8SThomas Graf * to the frags in the source buffer. 2941af2806f8SThomas Graf * 2942af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2943af2806f8SThomas Graf * headroom in the `to` buffer. 294436d5fe6aSZoltan Kiss * 294536d5fe6aSZoltan Kiss * Return value: 294636d5fe6aSZoltan Kiss * 0: everything is OK 294736d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 294836d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2949af2806f8SThomas Graf */ 295036d5fe6aSZoltan Kiss int 295136d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2952af2806f8SThomas Graf { 2953af2806f8SThomas Graf int i, j = 0; 2954af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 295536d5fe6aSZoltan Kiss int ret; 2956af2806f8SThomas Graf struct page *page; 2957af2806f8SThomas Graf unsigned int offset; 2958af2806f8SThomas Graf 2959af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2960af2806f8SThomas Graf 2961af2806f8SThomas Graf /* dont bother with small payloads */ 296236d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 296336d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2964af2806f8SThomas Graf 2965af2806f8SThomas Graf if (hlen) { 296636d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 296736d5fe6aSZoltan Kiss if (unlikely(ret)) 296836d5fe6aSZoltan Kiss return ret; 2969af2806f8SThomas Graf len -= hlen; 2970af2806f8SThomas Graf } else { 2971af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2972af2806f8SThomas Graf if (plen) { 2973af2806f8SThomas Graf page = virt_to_head_page(from->head); 2974af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2975af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2976af2806f8SThomas Graf get_page(page); 2977af2806f8SThomas Graf j = 1; 2978af2806f8SThomas Graf len -= plen; 2979af2806f8SThomas Graf } 2980af2806f8SThomas Graf } 2981af2806f8SThomas Graf 2982af2806f8SThomas Graf to->truesize += len + plen; 2983af2806f8SThomas Graf to->len += len + plen; 2984af2806f8SThomas Graf to->data_len += len + plen; 2985af2806f8SThomas Graf 298636d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 298736d5fe6aSZoltan Kiss skb_tx_error(from); 298836d5fe6aSZoltan Kiss return -ENOMEM; 298936d5fe6aSZoltan Kiss } 29901f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 299136d5fe6aSZoltan Kiss 2992af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2993d8e18a51SMatthew Wilcox (Oracle) int size; 2994d8e18a51SMatthew Wilcox (Oracle) 2995af2806f8SThomas Graf if (!len) 2996af2806f8SThomas Graf break; 2997af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2998d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 2999d8e18a51SMatthew Wilcox (Oracle) len); 3000d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3001d8e18a51SMatthew Wilcox (Oracle) len -= size; 3002af2806f8SThomas Graf skb_frag_ref(to, j); 3003af2806f8SThomas Graf j++; 3004af2806f8SThomas Graf } 3005af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 300636d5fe6aSZoltan Kiss 300736d5fe6aSZoltan Kiss return 0; 3008af2806f8SThomas Graf } 3009af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3010af2806f8SThomas Graf 30111da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 30121da177e4SLinus Torvalds { 3013d3bc23e7SAl Viro __wsum csum; 30141da177e4SLinus Torvalds long csstart; 30151da177e4SLinus Torvalds 301684fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 301755508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 30181da177e4SLinus Torvalds else 30191da177e4SLinus Torvalds csstart = skb_headlen(skb); 30201da177e4SLinus Torvalds 302109a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 30221da177e4SLinus Torvalds 3023d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 30241da177e4SLinus Torvalds 30251da177e4SLinus Torvalds csum = 0; 30261da177e4SLinus Torvalds if (csstart != skb->len) 30271da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 30288d5930dfSAl Viro skb->len - csstart); 30291da177e4SLinus Torvalds 303084fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3031ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 30321da177e4SLinus Torvalds 3033d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 30341da177e4SLinus Torvalds } 30351da177e4SLinus Torvalds } 3036b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 30371da177e4SLinus Torvalds 30381da177e4SLinus Torvalds /** 30391da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 30401da177e4SLinus Torvalds * @list: list to dequeue from 30411da177e4SLinus Torvalds * 30421da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 30431da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 30441da177e4SLinus Torvalds * returned or %NULL if the list is empty. 30451da177e4SLinus Torvalds */ 30461da177e4SLinus Torvalds 30471da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 30481da177e4SLinus Torvalds { 30491da177e4SLinus Torvalds unsigned long flags; 30501da177e4SLinus Torvalds struct sk_buff *result; 30511da177e4SLinus Torvalds 30521da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30531da177e4SLinus Torvalds result = __skb_dequeue(list); 30541da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30551da177e4SLinus Torvalds return result; 30561da177e4SLinus Torvalds } 3057b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 30581da177e4SLinus Torvalds 30591da177e4SLinus Torvalds /** 30601da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 30611da177e4SLinus Torvalds * @list: list to dequeue from 30621da177e4SLinus Torvalds * 30631da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 30641da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 30651da177e4SLinus Torvalds * returned or %NULL if the list is empty. 30661da177e4SLinus Torvalds */ 30671da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 30681da177e4SLinus Torvalds { 30691da177e4SLinus Torvalds unsigned long flags; 30701da177e4SLinus Torvalds struct sk_buff *result; 30711da177e4SLinus Torvalds 30721da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30731da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 30741da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30751da177e4SLinus Torvalds return result; 30761da177e4SLinus Torvalds } 3077b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 30781da177e4SLinus Torvalds 30791da177e4SLinus Torvalds /** 30801da177e4SLinus Torvalds * skb_queue_purge - empty a list 30811da177e4SLinus Torvalds * @list: list to empty 30821da177e4SLinus Torvalds * 30831da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 30841da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 30851da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 30861da177e4SLinus Torvalds */ 30871da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 30881da177e4SLinus Torvalds { 30891da177e4SLinus Torvalds struct sk_buff *skb; 30901da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 30911da177e4SLinus Torvalds kfree_skb(skb); 30921da177e4SLinus Torvalds } 3093b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 30941da177e4SLinus Torvalds 30951da177e4SLinus Torvalds /** 30969f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 30979f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3098385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 30999f5afeaeSYaogong Wang * 31009f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 31019f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 31029f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 31039f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 31049f5afeaeSYaogong Wang */ 3105385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 31069f5afeaeSYaogong Wang { 31077c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3108385114deSPeter Oskolkov unsigned int sum = 0; 31099f5afeaeSYaogong Wang 31107c90584cSEric Dumazet while (p) { 31117c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 31127c90584cSEric Dumazet 31137c90584cSEric Dumazet p = rb_next(p); 31147c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3115385114deSPeter Oskolkov sum += skb->truesize; 31169f5afeaeSYaogong Wang kfree_skb(skb); 31177c90584cSEric Dumazet } 3118385114deSPeter Oskolkov return sum; 31199f5afeaeSYaogong Wang } 31209f5afeaeSYaogong Wang 31219f5afeaeSYaogong Wang /** 31221da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 31231da177e4SLinus Torvalds * @list: list to use 31241da177e4SLinus Torvalds * @newsk: buffer to queue 31251da177e4SLinus Torvalds * 31261da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 31271da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 31281da177e4SLinus Torvalds * safely. 31291da177e4SLinus Torvalds * 31301da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31311da177e4SLinus Torvalds */ 31321da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 31331da177e4SLinus Torvalds { 31341da177e4SLinus Torvalds unsigned long flags; 31351da177e4SLinus Torvalds 31361da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31371da177e4SLinus Torvalds __skb_queue_head(list, newsk); 31381da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31391da177e4SLinus Torvalds } 3140b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 31411da177e4SLinus Torvalds 31421da177e4SLinus Torvalds /** 31431da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 31441da177e4SLinus Torvalds * @list: list to use 31451da177e4SLinus Torvalds * @newsk: buffer to queue 31461da177e4SLinus Torvalds * 31471da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 31481da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 31491da177e4SLinus Torvalds * safely. 31501da177e4SLinus Torvalds * 31511da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31521da177e4SLinus Torvalds */ 31531da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 31541da177e4SLinus Torvalds { 31551da177e4SLinus Torvalds unsigned long flags; 31561da177e4SLinus Torvalds 31571da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31581da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 31591da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31601da177e4SLinus Torvalds } 3161b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 31628728b834SDavid S. Miller 31631da177e4SLinus Torvalds /** 31641da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 31651da177e4SLinus Torvalds * @skb: buffer to remove 31668728b834SDavid S. Miller * @list: list to use 31671da177e4SLinus Torvalds * 31688728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 31698728b834SDavid S. Miller * function is atomic with respect to other list locked calls 31701da177e4SLinus Torvalds * 31718728b834SDavid S. Miller * You must know what list the SKB is on. 31721da177e4SLinus Torvalds */ 31738728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 31741da177e4SLinus Torvalds { 31751da177e4SLinus Torvalds unsigned long flags; 31761da177e4SLinus Torvalds 31771da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 31788728b834SDavid S. Miller __skb_unlink(skb, list); 31791da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 31801da177e4SLinus Torvalds } 3181b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 31821da177e4SLinus Torvalds 31831da177e4SLinus Torvalds /** 31841da177e4SLinus Torvalds * skb_append - append a buffer 31851da177e4SLinus Torvalds * @old: buffer to insert after 31861da177e4SLinus Torvalds * @newsk: buffer to insert 31878728b834SDavid S. Miller * @list: list to use 31881da177e4SLinus Torvalds * 31891da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 31901da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 31911da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 31921da177e4SLinus Torvalds */ 31938728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 31941da177e4SLinus Torvalds { 31951da177e4SLinus Torvalds unsigned long flags; 31961da177e4SLinus Torvalds 31978728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 31987de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 31998728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 32001da177e4SLinus Torvalds } 3201b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 32021da177e4SLinus Torvalds 32031da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 32041da177e4SLinus Torvalds struct sk_buff* skb1, 32051da177e4SLinus Torvalds const u32 len, const int pos) 32061da177e4SLinus Torvalds { 32071da177e4SLinus Torvalds int i; 32081da177e4SLinus Torvalds 3209d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3210d626f62bSArnaldo Carvalho de Melo pos - len); 32111da177e4SLinus Torvalds /* And move data appendix as is. */ 32121da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 32131da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 32141da177e4SLinus Torvalds 32151da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 32161da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 32171da177e4SLinus Torvalds skb1->data_len = skb->data_len; 32181da177e4SLinus Torvalds skb1->len += skb1->data_len; 32191da177e4SLinus Torvalds skb->data_len = 0; 32201da177e4SLinus Torvalds skb->len = len; 322127a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 32221da177e4SLinus Torvalds } 32231da177e4SLinus Torvalds 32241da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 32251da177e4SLinus Torvalds struct sk_buff* skb1, 32261da177e4SLinus Torvalds const u32 len, int pos) 32271da177e4SLinus Torvalds { 32281da177e4SLinus Torvalds int i, k = 0; 32291da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 32301da177e4SLinus Torvalds 32311da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 32321da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 32331da177e4SLinus Torvalds skb->len = len; 32341da177e4SLinus Torvalds skb->data_len = len - pos; 32351da177e4SLinus Torvalds 32361da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 32379e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 32381da177e4SLinus Torvalds 32391da177e4SLinus Torvalds if (pos + size > len) { 32401da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 32411da177e4SLinus Torvalds 32421da177e4SLinus Torvalds if (pos < len) { 32431da177e4SLinus Torvalds /* Split frag. 32441da177e4SLinus Torvalds * We have two variants in this case: 32451da177e4SLinus Torvalds * 1. Move all the frag to the second 32461da177e4SLinus Torvalds * part, if it is possible. F.e. 32471da177e4SLinus Torvalds * this approach is mandatory for TUX, 32481da177e4SLinus Torvalds * where splitting is expensive. 32491da177e4SLinus Torvalds * 2. Split is accurately. We make this. 32501da177e4SLinus Torvalds */ 3251ea2ab693SIan Campbell skb_frag_ref(skb, i); 3252b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 32539e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 32549e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 32551da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 32561da177e4SLinus Torvalds } 32571da177e4SLinus Torvalds k++; 32581da177e4SLinus Torvalds } else 32591da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 32601da177e4SLinus Torvalds pos += size; 32611da177e4SLinus Torvalds } 32621da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 32631da177e4SLinus Torvalds } 32641da177e4SLinus Torvalds 32651da177e4SLinus Torvalds /** 32661da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 32671da177e4SLinus Torvalds * @skb: the buffer to split 32681da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 32691da177e4SLinus Torvalds * @len: new length for skb 32701da177e4SLinus Torvalds */ 32711da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 32721da177e4SLinus Torvalds { 32731da177e4SLinus Torvalds int pos = skb_headlen(skb); 32741da177e4SLinus Torvalds 3275fff88030SWillem de Bruijn skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags & 3276fff88030SWillem de Bruijn SKBTX_SHARED_FRAG; 32771f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 32781da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 32791da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 32801da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 32811da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 32821da177e4SLinus Torvalds } 3283b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 32841da177e4SLinus Torvalds 32859f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 32869f782db3SIlpo Järvinen * 32879f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 32889f782db3SIlpo Järvinen */ 3289832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3290832d11c5SIlpo Järvinen { 32910ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3292832d11c5SIlpo Järvinen } 3293832d11c5SIlpo Järvinen 3294832d11c5SIlpo Järvinen /** 3295832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3296832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3297832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3298832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3299832d11c5SIlpo Järvinen * 3300832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 330120e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3302832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3303832d11c5SIlpo Järvinen * 3304832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3305832d11c5SIlpo Järvinen * 3306832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3307832d11c5SIlpo Järvinen * to have non-paged data as well. 3308832d11c5SIlpo Järvinen * 3309832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3310832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3311832d11c5SIlpo Järvinen */ 3312832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3313832d11c5SIlpo Järvinen { 3314832d11c5SIlpo Järvinen int from, to, merge, todo; 3315d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3316832d11c5SIlpo Järvinen 3317832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3318f8071cdeSEric Dumazet 3319f8071cdeSEric Dumazet if (skb_headlen(skb)) 3320f8071cdeSEric Dumazet return 0; 33211f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 33221f8b977aSWillem de Bruijn return 0; 3323832d11c5SIlpo Järvinen 3324832d11c5SIlpo Järvinen todo = shiftlen; 3325832d11c5SIlpo Järvinen from = 0; 3326832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3327832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3328832d11c5SIlpo Järvinen 3329832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3330832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3331832d11c5SIlpo Järvinen */ 3332832d11c5SIlpo Järvinen if (!to || 3333ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3334b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3335832d11c5SIlpo Järvinen merge = -1; 3336832d11c5SIlpo Järvinen } else { 3337832d11c5SIlpo Järvinen merge = to - 1; 3338832d11c5SIlpo Järvinen 33399e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3340832d11c5SIlpo Järvinen if (todo < 0) { 3341832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3342832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3343832d11c5SIlpo Järvinen return 0; 3344832d11c5SIlpo Järvinen 33459f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 33469f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3347832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3348832d11c5SIlpo Järvinen 33499e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 33509e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3351b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3352832d11c5SIlpo Järvinen 3353832d11c5SIlpo Järvinen goto onlymerged; 3354832d11c5SIlpo Järvinen } 3355832d11c5SIlpo Järvinen 3356832d11c5SIlpo Järvinen from++; 3357832d11c5SIlpo Järvinen } 3358832d11c5SIlpo Järvinen 3359832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3360832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3361832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3362832d11c5SIlpo Järvinen return 0; 3363832d11c5SIlpo Järvinen 3364832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3365832d11c5SIlpo Järvinen return 0; 3366832d11c5SIlpo Järvinen 3367832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3368832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3369832d11c5SIlpo Järvinen return 0; 3370832d11c5SIlpo Järvinen 3371832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3372832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3373832d11c5SIlpo Järvinen 33749e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3375832d11c5SIlpo Järvinen *fragto = *fragfrom; 33769e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3377832d11c5SIlpo Järvinen from++; 3378832d11c5SIlpo Järvinen to++; 3379832d11c5SIlpo Järvinen 3380832d11c5SIlpo Järvinen } else { 3381ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3382b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3383b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 33849e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3385832d11c5SIlpo Järvinen 3386b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 33879e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3388832d11c5SIlpo Järvinen todo = 0; 3389832d11c5SIlpo Järvinen 3390832d11c5SIlpo Järvinen to++; 3391832d11c5SIlpo Järvinen break; 3392832d11c5SIlpo Järvinen } 3393832d11c5SIlpo Järvinen } 3394832d11c5SIlpo Järvinen 3395832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3396832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3397832d11c5SIlpo Järvinen 3398832d11c5SIlpo Järvinen if (merge >= 0) { 3399832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3400832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3401832d11c5SIlpo Järvinen 34029e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 3403ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 3404832d11c5SIlpo Järvinen } 3405832d11c5SIlpo Järvinen 3406832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3407832d11c5SIlpo Järvinen to = 0; 3408832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3409832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3410832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3411832d11c5SIlpo Järvinen 3412832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3413832d11c5SIlpo Järvinen 3414832d11c5SIlpo Järvinen onlymerged: 3415832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3416832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3417832d11c5SIlpo Järvinen */ 3418832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3419832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3420832d11c5SIlpo Järvinen 3421832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3422832d11c5SIlpo Järvinen skb->len -= shiftlen; 3423832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3424832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3425832d11c5SIlpo Järvinen tgt->len += shiftlen; 3426832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3427832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3428832d11c5SIlpo Järvinen 3429832d11c5SIlpo Järvinen return shiftlen; 3430832d11c5SIlpo Järvinen } 3431832d11c5SIlpo Järvinen 3432677e90edSThomas Graf /** 3433677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3434677e90edSThomas Graf * @skb: the buffer to read 3435677e90edSThomas Graf * @from: lower offset of data to be read 3436677e90edSThomas Graf * @to: upper offset of data to be read 3437677e90edSThomas Graf * @st: state variable 3438677e90edSThomas Graf * 3439677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3440677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3441677e90edSThomas Graf */ 3442677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3443677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3444677e90edSThomas Graf { 3445677e90edSThomas Graf st->lower_offset = from; 3446677e90edSThomas Graf st->upper_offset = to; 3447677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3448677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3449677e90edSThomas Graf st->frag_data = NULL; 345097550f6fSWillem de Bruijn st->frag_off = 0; 3451677e90edSThomas Graf } 3452b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3453677e90edSThomas Graf 3454677e90edSThomas Graf /** 3455677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3456677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3457677e90edSThomas Graf * @data: destination pointer for data to be returned 3458677e90edSThomas Graf * @st: state variable 3459677e90edSThomas Graf * 3460bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3461677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3462bc32383cSMathias Krause * the head of the data block to @data and returns the length 3463677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3464677e90edSThomas Graf * offset has been reached. 3465677e90edSThomas Graf * 3466677e90edSThomas Graf * The caller is not required to consume all of the data 3467bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3468677e90edSThomas Graf * of bytes already consumed and the next call to 3469677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3470677e90edSThomas Graf * 347125985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3472e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3473677e90edSThomas Graf * reads of potentially non linear data. 3474677e90edSThomas Graf * 3475bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3476677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3477677e90edSThomas Graf * a stack for this purpose. 3478677e90edSThomas Graf */ 3479677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3480677e90edSThomas Graf struct skb_seq_state *st) 3481677e90edSThomas Graf { 3482677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3483677e90edSThomas Graf skb_frag_t *frag; 3484677e90edSThomas Graf 3485aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3486aeb193eaSWedson Almeida Filho if (st->frag_data) { 3487aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3488aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3489aeb193eaSWedson Almeida Filho } 3490677e90edSThomas Graf return 0; 3491aeb193eaSWedson Almeida Filho } 3492677e90edSThomas Graf 3493677e90edSThomas Graf next_skb: 349495e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3495677e90edSThomas Graf 3496995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 349795e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3498677e90edSThomas Graf return block_limit - abs_offset; 3499677e90edSThomas Graf } 3500677e90edSThomas Graf 3501677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3502677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3503677e90edSThomas Graf 3504677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 350597550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3506677e90edSThomas Graf 350797550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 350897550f6fSWillem de Bruijn 350997550f6fSWillem de Bruijn pg_idx = 0; 351097550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 351197550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 351297550f6fSWillem de Bruijn 351397550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 351497550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 351597550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 351697550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 351797550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 351897550f6fSWillem de Bruijn } 351997550f6fSWillem de Bruijn 352097550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3521677e90edSThomas Graf if (abs_offset < block_limit) { 3522677e90edSThomas Graf if (!st->frag_data) 352397550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3524677e90edSThomas Graf 352597550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3526677e90edSThomas Graf (abs_offset - st->stepped_offset); 3527677e90edSThomas Graf 3528677e90edSThomas Graf return block_limit - abs_offset; 3529677e90edSThomas Graf } 3530677e90edSThomas Graf 3531677e90edSThomas Graf if (st->frag_data) { 353251c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3533677e90edSThomas Graf st->frag_data = NULL; 3534677e90edSThomas Graf } 3535677e90edSThomas Graf 353697550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 353797550f6fSWillem de Bruijn st->frag_off += pg_sz; 353897550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 353997550f6fSWillem de Bruijn st->frag_off = 0; 3540677e90edSThomas Graf st->frag_idx++; 354197550f6fSWillem de Bruijn } 3542677e90edSThomas Graf } 3543677e90edSThomas Graf 35445b5a60daSOlaf Kirch if (st->frag_data) { 354551c56b00SEric Dumazet kunmap_atomic(st->frag_data); 35465b5a60daSOlaf Kirch st->frag_data = NULL; 35475b5a60daSOlaf Kirch } 35485b5a60daSOlaf Kirch 354921dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3550677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 355195e3b24cSHerbert Xu st->frag_idx = 0; 3552677e90edSThomas Graf goto next_skb; 355371b3346dSShyam Iyer } else if (st->cur_skb->next) { 355471b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 355571b3346dSShyam Iyer st->frag_idx = 0; 3556677e90edSThomas Graf goto next_skb; 3557677e90edSThomas Graf } 3558677e90edSThomas Graf 3559677e90edSThomas Graf return 0; 3560677e90edSThomas Graf } 3561b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3562677e90edSThomas Graf 3563677e90edSThomas Graf /** 3564677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3565677e90edSThomas Graf * @st: state variable 3566677e90edSThomas Graf * 3567677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3568677e90edSThomas Graf * returned 0. 3569677e90edSThomas Graf */ 3570677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3571677e90edSThomas Graf { 3572677e90edSThomas Graf if (st->frag_data) 357351c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3574677e90edSThomas Graf } 3575b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3576677e90edSThomas Graf 35773fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 35783fc7e8a6SThomas Graf 35793fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 35803fc7e8a6SThomas Graf struct ts_config *conf, 35813fc7e8a6SThomas Graf struct ts_state *state) 35823fc7e8a6SThomas Graf { 35833fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 35843fc7e8a6SThomas Graf } 35853fc7e8a6SThomas Graf 35863fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 35873fc7e8a6SThomas Graf { 35883fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 35893fc7e8a6SThomas Graf } 35903fc7e8a6SThomas Graf 35913fc7e8a6SThomas Graf /** 35923fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 35933fc7e8a6SThomas Graf * @skb: the buffer to look in 35943fc7e8a6SThomas Graf * @from: search offset 35953fc7e8a6SThomas Graf * @to: search limit 35963fc7e8a6SThomas Graf * @config: textsearch configuration 35973fc7e8a6SThomas Graf * 35983fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 35993fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 36003fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 36013fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 36023fc7e8a6SThomas Graf */ 36033fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3604059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 36053fc7e8a6SThomas Graf { 3606059a2440SBojan Prtvar struct ts_state state; 3607f72b948dSPhil Oester unsigned int ret; 3608f72b948dSPhil Oester 36093fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 36103fc7e8a6SThomas Graf config->finish = skb_ts_finish; 36113fc7e8a6SThomas Graf 3612059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 36133fc7e8a6SThomas Graf 3614059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3615f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 36163fc7e8a6SThomas Graf } 3617b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 36183fc7e8a6SThomas Graf 3619be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3620be12a1feSHannes Frederic Sowa int offset, size_t size) 3621be12a1feSHannes Frederic Sowa { 3622be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3623be12a1feSHannes Frederic Sowa 3624be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3625be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3626be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3627be12a1feSHannes Frederic Sowa get_page(page); 3628be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3629be12a1feSHannes Frederic Sowa } else { 3630be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3631be12a1feSHannes Frederic Sowa } 3632be12a1feSHannes Frederic Sowa 3633be12a1feSHannes Frederic Sowa return 0; 3634be12a1feSHannes Frederic Sowa } 3635be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3636be12a1feSHannes Frederic Sowa 3637cbb042f9SHerbert Xu /** 3638cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3639cbb042f9SHerbert Xu * @skb: buffer to update 3640cbb042f9SHerbert Xu * @len: length of data pulled 3641cbb042f9SHerbert Xu * 3642cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3643fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 364484fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 364584fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 364684fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3647cbb042f9SHerbert Xu */ 3648af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3649cbb042f9SHerbert Xu { 365031b33dfbSPravin B Shelar unsigned char *data = skb->data; 365131b33dfbSPravin B Shelar 3652cbb042f9SHerbert Xu BUG_ON(len > skb->len); 365331b33dfbSPravin B Shelar __skb_pull(skb, len); 365431b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 365531b33dfbSPravin B Shelar return skb->data; 3656cbb042f9SHerbert Xu } 3657f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3658f94691acSArnaldo Carvalho de Melo 365913acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 366013acc94eSYonghong Song { 366113acc94eSYonghong Song skb_frag_t head_frag; 366213acc94eSYonghong Song struct page *page; 366313acc94eSYonghong Song 366413acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3665d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3666b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3667b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3668d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 366913acc94eSYonghong Song return head_frag; 367013acc94eSYonghong Song } 367113acc94eSYonghong Song 36723a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 36733a1296a3SSteffen Klassert netdev_features_t features, 36743a1296a3SSteffen Klassert unsigned int offset) 36753a1296a3SSteffen Klassert { 36763a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 36773a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 36783a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 36793a1296a3SSteffen Klassert unsigned int delta_len = 0; 36803a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 368153475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 368253475c5dSDongseok Yi int err; 36833a1296a3SSteffen Klassert 36843a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 36853a1296a3SSteffen Klassert 36863a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 36873a1296a3SSteffen Klassert 36883a1296a3SSteffen Klassert do { 36893a1296a3SSteffen Klassert nskb = list_skb; 36903a1296a3SSteffen Klassert list_skb = list_skb->next; 36913a1296a3SSteffen Klassert 369253475c5dSDongseok Yi err = 0; 369353475c5dSDongseok Yi if (skb_shared(nskb)) { 369453475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 369553475c5dSDongseok Yi if (tmp) { 369653475c5dSDongseok Yi consume_skb(nskb); 369753475c5dSDongseok Yi nskb = tmp; 369853475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 369953475c5dSDongseok Yi } else { 370053475c5dSDongseok Yi err = -ENOMEM; 370153475c5dSDongseok Yi } 370253475c5dSDongseok Yi } 370353475c5dSDongseok Yi 37043a1296a3SSteffen Klassert if (!tail) 37053a1296a3SSteffen Klassert skb->next = nskb; 37063a1296a3SSteffen Klassert else 37073a1296a3SSteffen Klassert tail->next = nskb; 37083a1296a3SSteffen Klassert 370953475c5dSDongseok Yi if (unlikely(err)) { 371053475c5dSDongseok Yi nskb->next = list_skb; 371153475c5dSDongseok Yi goto err_linearize; 371253475c5dSDongseok Yi } 371353475c5dSDongseok Yi 37143a1296a3SSteffen Klassert tail = nskb; 37153a1296a3SSteffen Klassert 37163a1296a3SSteffen Klassert delta_len += nskb->len; 37173a1296a3SSteffen Klassert delta_truesize += nskb->truesize; 37183a1296a3SSteffen Klassert 37193a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 37203a1296a3SSteffen Klassert 3721cf673ed0SFlorian Westphal skb_release_head_state(nskb); 37223a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 37233a1296a3SSteffen Klassert 37243a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 37253a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 37263a1296a3SSteffen Klassert nskb->data - tnl_hlen, 37273a1296a3SSteffen Klassert offset + tnl_hlen); 37283a1296a3SSteffen Klassert 37293a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 37303a1296a3SSteffen Klassert __skb_linearize(nskb)) 37313a1296a3SSteffen Klassert goto err_linearize; 37323a1296a3SSteffen Klassert 37333a1296a3SSteffen Klassert } while (list_skb); 37343a1296a3SSteffen Klassert 37353a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 37363a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 37373a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 37383a1296a3SSteffen Klassert 37393a1296a3SSteffen Klassert skb_gso_reset(skb); 37403a1296a3SSteffen Klassert 37413a1296a3SSteffen Klassert skb->prev = tail; 37423a1296a3SSteffen Klassert 37433a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 37443a1296a3SSteffen Klassert __skb_linearize(skb)) 37453a1296a3SSteffen Klassert goto err_linearize; 37463a1296a3SSteffen Klassert 37473a1296a3SSteffen Klassert skb_get(skb); 37483a1296a3SSteffen Klassert 37493a1296a3SSteffen Klassert return skb; 37503a1296a3SSteffen Klassert 37513a1296a3SSteffen Klassert err_linearize: 37523a1296a3SSteffen Klassert kfree_skb_list(skb->next); 37533a1296a3SSteffen Klassert skb->next = NULL; 37543a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 37553a1296a3SSteffen Klassert } 37563a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 37573a1296a3SSteffen Klassert 37583a1296a3SSteffen Klassert int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb) 37593a1296a3SSteffen Klassert { 37603a1296a3SSteffen Klassert if (unlikely(p->len + skb->len >= 65536)) 37613a1296a3SSteffen Klassert return -E2BIG; 37623a1296a3SSteffen Klassert 37633a1296a3SSteffen Klassert if (NAPI_GRO_CB(p)->last == p) 37643a1296a3SSteffen Klassert skb_shinfo(p)->frag_list = skb; 37653a1296a3SSteffen Klassert else 37663a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last->next = skb; 37673a1296a3SSteffen Klassert 37683a1296a3SSteffen Klassert skb_pull(skb, skb_gro_offset(skb)); 37693a1296a3SSteffen Klassert 37703a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last = skb; 37713a1296a3SSteffen Klassert NAPI_GRO_CB(p)->count++; 37723a1296a3SSteffen Klassert p->data_len += skb->len; 37733a1296a3SSteffen Klassert p->truesize += skb->truesize; 37743a1296a3SSteffen Klassert p->len += skb->len; 37753a1296a3SSteffen Klassert 37763a1296a3SSteffen Klassert NAPI_GRO_CB(skb)->same_flow = 1; 37773a1296a3SSteffen Klassert 37783a1296a3SSteffen Klassert return 0; 37793a1296a3SSteffen Klassert } 37803a1296a3SSteffen Klassert 3781f4c50d99SHerbert Xu /** 3782f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3783df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3784576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3785f4c50d99SHerbert Xu * 3786f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 37874c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 37884c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3789f4c50d99SHerbert Xu */ 3790df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3791df5771ffSMichael S. Tsirkin netdev_features_t features) 3792f4c50d99SHerbert Xu { 3793f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3794f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 37951a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3796df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3797df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3798df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 37991fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3800f4c50d99SHerbert Xu unsigned int offset = doffset; 3801df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3802802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3803f4c50d99SHerbert Xu unsigned int headroom; 3804802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3805ec5f0615SPravin B Shelar __be16 proto; 380636c98382SAlexander Duyck bool csum, sg; 3807df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3808f4c50d99SHerbert Xu int err = -ENOMEM; 3809f4c50d99SHerbert Xu int i = 0; 3810f4c50d99SHerbert Xu int pos; 3811f4c50d99SHerbert Xu 38123dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 38133dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 38143dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 38153dcbdb13SShmulik Ladkani * non head_frag head. 38163dcbdb13SShmulik Ladkani * 38173dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 38183dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 38193dcbdb13SShmulik Ladkani * head, then the first one has too). 38203dcbdb13SShmulik Ladkani * 38213dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 38223dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 38233dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 38243dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 38253dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 38263dcbdb13SShmulik Ladkani */ 38273dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 38283dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 38293dcbdb13SShmulik Ladkani } 38303dcbdb13SShmulik Ladkani 38315882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 38322f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 3833ec5f0615SPravin B Shelar if (unlikely(!proto)) 3834ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3835ec5f0615SPravin B Shelar 383636c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3837f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 38387e2b10c1STom Herbert 383907b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 384007b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 384107b26c94SSteffen Klassert struct sk_buff *iter; 384243170c4eSIlan Tayari unsigned int frag_len; 384307b26c94SSteffen Klassert 384407b26c94SSteffen Klassert if (!list_skb || 384507b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 384607b26c94SSteffen Klassert goto normal; 384707b26c94SSteffen Klassert 384843170c4eSIlan Tayari /* If we get here then all the required 384943170c4eSIlan Tayari * GSO features except frag_list are supported. 385043170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 385143170c4eSIlan Tayari * with no frag_list. 385243170c4eSIlan Tayari * Currently we can do that only when the buffers don't 385343170c4eSIlan Tayari * have a linear part and all the buffers except 385443170c4eSIlan Tayari * the last are of the same length. 385507b26c94SSteffen Klassert */ 385643170c4eSIlan Tayari frag_len = list_skb->len; 385707b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 385843170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 385943170c4eSIlan Tayari goto normal; 3860eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 386107b26c94SSteffen Klassert goto normal; 386207b26c94SSteffen Klassert 386307b26c94SSteffen Klassert len -= iter->len; 386407b26c94SSteffen Klassert } 386543170c4eSIlan Tayari 386643170c4eSIlan Tayari if (len != frag_len) 386743170c4eSIlan Tayari goto normal; 386807b26c94SSteffen Klassert } 386907b26c94SSteffen Klassert 3870802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 3871802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 3872802ab55aSAlexander Duyck * now. 3873802ab55aSAlexander Duyck */ 3874802ab55aSAlexander Duyck partial_segs = len / mss; 3875d7fb5a80SAlexander Duyck if (partial_segs > 1) 3876802ab55aSAlexander Duyck mss *= partial_segs; 3877d7fb5a80SAlexander Duyck else 3878d7fb5a80SAlexander Duyck partial_segs = 0; 3879802ab55aSAlexander Duyck } 3880802ab55aSAlexander Duyck 388107b26c94SSteffen Klassert normal: 3882df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3883df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3884f4c50d99SHerbert Xu 3885f4c50d99SHerbert Xu do { 3886f4c50d99SHerbert Xu struct sk_buff *nskb; 38878cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3888c8884eddSHerbert Xu int hsize; 3889f4c50d99SHerbert Xu int size; 3890f4c50d99SHerbert Xu 38913953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 38923953c46cSMarcelo Ricardo Leitner len = list_skb->len; 38933953c46cSMarcelo Ricardo Leitner } else { 3894df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3895f4c50d99SHerbert Xu if (len > mss) 3896f4c50d99SHerbert Xu len = mss; 38973953c46cSMarcelo Ricardo Leitner } 3898f4c50d99SHerbert Xu 3899df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3900f4c50d99SHerbert Xu if (hsize < 0) 3901f4c50d99SHerbert Xu hsize = 0; 3902c8884eddSHerbert Xu if (hsize > len || !sg) 3903c8884eddSHerbert Xu hsize = len; 3904f4c50d99SHerbert Xu 39051a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 39061a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 39071a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 390889319d38SHerbert Xu 39099d8506ccSHerbert Xu i = 0; 39101a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 39111a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 39121fd819ecSMichael S. Tsirkin frag_skb = list_skb; 39131a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 39149d8506ccSHerbert Xu 39159d8506ccSHerbert Xu while (pos < offset + len) { 39169d8506ccSHerbert Xu BUG_ON(i >= nfrags); 39179d8506ccSHerbert Xu 39184e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 39199d8506ccSHerbert Xu if (pos + size > offset + len) 39209d8506ccSHerbert Xu break; 39219d8506ccSHerbert Xu 39229d8506ccSHerbert Xu i++; 39239d8506ccSHerbert Xu pos += size; 39244e1beba1SMichael S. Tsirkin frag++; 39259d8506ccSHerbert Xu } 39269d8506ccSHerbert Xu 39271a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 39281a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 392989319d38SHerbert Xu 3930f4c50d99SHerbert Xu if (unlikely(!nskb)) 3931f4c50d99SHerbert Xu goto err; 3932f4c50d99SHerbert Xu 39339d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 39349d8506ccSHerbert Xu kfree_skb(nskb); 39359d8506ccSHerbert Xu goto err; 39369d8506ccSHerbert Xu } 39379d8506ccSHerbert Xu 3938ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 393989319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 394089319d38SHerbert Xu kfree_skb(nskb); 394189319d38SHerbert Xu goto err; 394289319d38SHerbert Xu } 394389319d38SHerbert Xu 3944ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 394589319d38SHerbert Xu skb_release_head_state(nskb); 394689319d38SHerbert Xu __skb_push(nskb, doffset); 394789319d38SHerbert Xu } else { 3948c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3949df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3950c93bdd0eSMel Gorman NUMA_NO_NODE); 395189319d38SHerbert Xu 395289319d38SHerbert Xu if (unlikely(!nskb)) 395389319d38SHerbert Xu goto err; 395489319d38SHerbert Xu 395589319d38SHerbert Xu skb_reserve(nskb, headroom); 395689319d38SHerbert Xu __skb_put(nskb, doffset); 395789319d38SHerbert Xu } 395889319d38SHerbert Xu 3959f4c50d99SHerbert Xu if (segs) 3960f4c50d99SHerbert Xu tail->next = nskb; 3961f4c50d99SHerbert Xu else 3962f4c50d99SHerbert Xu segs = nskb; 3963f4c50d99SHerbert Xu tail = nskb; 3964f4c50d99SHerbert Xu 3965df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3966f4c50d99SHerbert Xu 3967030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3968fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 396968c33163SPravin B Shelar 3970df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 397168c33163SPravin B Shelar nskb->data - tnl_hlen, 397268c33163SPravin B Shelar doffset + tnl_hlen); 397389319d38SHerbert Xu 39749d8506ccSHerbert Xu if (nskb->len == len + doffset) 39751cdbcb79SSimon Horman goto perform_csum_check; 397689319d38SHerbert Xu 39777fbeffedSAlexander Duyck if (!sg) { 39781454c9faSYadu Kishore if (!csum) { 39797fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 39806f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 398176443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 398276443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 39831454c9faSYadu Kishore skb_put(nskb, 39841454c9faSYadu Kishore len), 39858d5930dfSAl Viro len); 39867e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3987de843723STom Herbert skb_headroom(nskb) + doffset; 39881454c9faSYadu Kishore } else { 39891454c9faSYadu Kishore skb_copy_bits(head_skb, offset, 39901454c9faSYadu Kishore skb_put(nskb, len), 39911454c9faSYadu Kishore len); 39921454c9faSYadu Kishore } 3993f4c50d99SHerbert Xu continue; 3994f4c50d99SHerbert Xu } 3995f4c50d99SHerbert Xu 39968cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3997f4c50d99SHerbert Xu 3998df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3999d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4000f4c50d99SHerbert Xu 4001fff88030SWillem de Bruijn skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags & 4002df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 4003cef401deSEric Dumazet 4004bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4005bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4006bf5c25d6SWillem de Bruijn goto err; 4007bf5c25d6SWillem de Bruijn 40089d8506ccSHerbert Xu while (pos < offset + len) { 40099d8506ccSHerbert Xu if (i >= nfrags) { 40109d8506ccSHerbert Xu i = 0; 40111a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 40121a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 40131fd819ecSMichael S. Tsirkin frag_skb = list_skb; 401413acc94eSYonghong Song if (!skb_headlen(list_skb)) { 40159d8506ccSHerbert Xu BUG_ON(!nfrags); 401613acc94eSYonghong Song } else { 401713acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 40189d8506ccSHerbert Xu 401913acc94eSYonghong Song /* to make room for head_frag. */ 402013acc94eSYonghong Song i--; 402113acc94eSYonghong Song frag--; 402213acc94eSYonghong Song } 4023bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4024bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4025bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4026bf5c25d6SWillem de Bruijn goto err; 4027bf5c25d6SWillem de Bruijn 40281a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 40299d8506ccSHerbert Xu } 40309d8506ccSHerbert Xu 40319d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 40329d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 40339d8506ccSHerbert Xu net_warn_ratelimited( 40349d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 40359d8506ccSHerbert Xu pos, mss); 4036ff907a11SEric Dumazet err = -EINVAL; 40379d8506ccSHerbert Xu goto err; 40389d8506ccSHerbert Xu } 40399d8506ccSHerbert Xu 404013acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 40418cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 40428cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4043f4c50d99SHerbert Xu 4044f4c50d99SHerbert Xu if (pos < offset) { 4045b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 40468cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4047f4c50d99SHerbert Xu } 4048f4c50d99SHerbert Xu 404989319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4050f4c50d99SHerbert Xu 4051f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4052f4c50d99SHerbert Xu i++; 40534e1beba1SMichael S. Tsirkin frag++; 4054f4c50d99SHerbert Xu pos += size; 4055f4c50d99SHerbert Xu } else { 40568cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 405789319d38SHerbert Xu goto skip_fraglist; 4058f4c50d99SHerbert Xu } 4059f4c50d99SHerbert Xu 40608cb19905SMichael S. Tsirkin nskb_frag++; 4061f4c50d99SHerbert Xu } 4062f4c50d99SHerbert Xu 406389319d38SHerbert Xu skip_fraglist: 4064f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4065f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4066f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4067ec5f0615SPravin B Shelar 40681cdbcb79SSimon Horman perform_csum_check: 40697fbeffedSAlexander Duyck if (!csum) { 4070ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4071ff907a11SEric Dumazet __skb_linearize(nskb)) 4072ddff00d4SAlexander Duyck goto err; 4073ff907a11SEric Dumazet 40747fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4075ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 407676443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 407776443456SAlexander Duyck skb_checksum(nskb, doffset, 407876443456SAlexander Duyck nskb->len - doffset, 0); 40797e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 40807e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4081ec5f0615SPravin B Shelar } 4082df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4083f4c50d99SHerbert Xu 4084bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4085bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4086bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4087bec3cfdcSEric Dumazet */ 4088bec3cfdcSEric Dumazet segs->prev = tail; 4089432c856fSToshiaki Makita 4090802ab55aSAlexander Duyck if (partial_segs) { 409107b26c94SSteffen Klassert struct sk_buff *iter; 4092802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 409307b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4094802ab55aSAlexander Duyck 4095802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 409607b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4097802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4098802ab55aSAlexander Duyck 4099802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4100802ab55aSAlexander Duyck * our way back down the stack of protocols. 4101802ab55aSAlexander Duyck */ 410207b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 410307b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 410407b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 410507b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 410607b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 410707b26c94SSteffen Klassert } 410807b26c94SSteffen Klassert 410907b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 411007b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 411107b26c94SSteffen Klassert else if (tail != segs) 411207b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4113802ab55aSAlexander Duyck } 4114802ab55aSAlexander Duyck 4115432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4116432c856fSToshiaki Makita * using skb_set_owner_w(). 4117432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4118432c856fSToshiaki Makita */ 4119432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4120432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4121432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4122432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4123432c856fSToshiaki Makita } 4124f4c50d99SHerbert Xu return segs; 4125f4c50d99SHerbert Xu 4126f4c50d99SHerbert Xu err: 4127289dccbeSEric Dumazet kfree_skb_list(segs); 4128f4c50d99SHerbert Xu return ERR_PTR(err); 4129f4c50d99SHerbert Xu } 4130f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4131f4c50d99SHerbert Xu 4132d4546c25SDavid Miller int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb) 413371d93b39SHerbert Xu { 41348a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 413567147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 413667147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 41378a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 4138715dc1f3SEric Dumazet unsigned int delta_truesize; 4139d4546c25SDavid Miller struct sk_buff *lp; 414071d93b39SHerbert Xu 41410ab03f35SSteffen Klassert if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush)) 414271d93b39SHerbert Xu return -E2BIG; 414371d93b39SHerbert Xu 414429e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 41458a29111cSEric Dumazet pinfo = skb_shinfo(lp); 41468a29111cSEric Dumazet 41478a29111cSEric Dumazet if (headlen <= offset) { 414842da6994SHerbert Xu skb_frag_t *frag; 414966e92fcfSHerbert Xu skb_frag_t *frag2; 41509aaa156cSHerbert Xu int i = skbinfo->nr_frags; 41519aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 415242da6994SHerbert Xu 415366e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 41548a29111cSEric Dumazet goto merge; 415581705ad1SHerbert Xu 41568a29111cSEric Dumazet offset -= headlen; 41579aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 41589aaa156cSHerbert Xu skbinfo->nr_frags = 0; 4159f5572068SHerbert Xu 41609aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 41619aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 416266e92fcfSHerbert Xu do { 416366e92fcfSHerbert Xu *--frag = *--frag2; 416466e92fcfSHerbert Xu } while (--i); 416566e92fcfSHerbert Xu 4166b54c9d5bSJonathan Lemon skb_frag_off_add(frag, offset); 41679e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 416866e92fcfSHerbert Xu 4169715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 4170ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 4171ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 4172715dc1f3SEric Dumazet 4173f5572068SHerbert Xu skb->truesize -= skb->data_len; 4174f5572068SHerbert Xu skb->len -= skb->data_len; 4175f5572068SHerbert Xu skb->data_len = 0; 4176f5572068SHerbert Xu 4177715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 41785d38a079SHerbert Xu goto done; 4179d7e8883cSEric Dumazet } else if (skb->head_frag) { 4180d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 4181d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 4182d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 4183d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 4184d7e8883cSEric Dumazet unsigned int first_offset; 4185d7e8883cSEric Dumazet 4186d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 41878a29111cSEric Dumazet goto merge; 4188d7e8883cSEric Dumazet 4189d7e8883cSEric Dumazet first_offset = skb->data - 4190d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 4191d7e8883cSEric Dumazet offset; 4192d7e8883cSEric Dumazet 4193d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 4194d7e8883cSEric Dumazet 4195d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(frag, page); 4196b54c9d5bSJonathan Lemon skb_frag_off_set(frag, first_offset); 4197d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 4198d7e8883cSEric Dumazet 4199d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 4200d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 4201d7e8883cSEric Dumazet 4202715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4203d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 4204d7e8883cSEric Dumazet goto done; 42058a29111cSEric Dumazet } 420671d93b39SHerbert Xu 420771d93b39SHerbert Xu merge: 4208715dc1f3SEric Dumazet delta_truesize = skb->truesize; 420967147ba9SHerbert Xu if (offset > headlen) { 4210d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 4211d1dc7abfSMichal Schmidt 4212b54c9d5bSJonathan Lemon skb_frag_off_add(&skbinfo->frags[0], eat); 42139e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 4214d1dc7abfSMichal Schmidt skb->data_len -= eat; 4215d1dc7abfSMichal Schmidt skb->len -= eat; 421667147ba9SHerbert Xu offset = headlen; 421756035022SHerbert Xu } 421856035022SHerbert Xu 421967147ba9SHerbert Xu __skb_pull(skb, offset); 422056035022SHerbert Xu 422129e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 42228a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 42238a29111cSEric Dumazet else 4224c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 4225c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 4226f4a775d1SEric Dumazet __skb_header_release(skb); 42278a29111cSEric Dumazet lp = p; 422871d93b39SHerbert Xu 42295d38a079SHerbert Xu done: 42305d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 423137fe4732SHerbert Xu p->data_len += len; 4232715dc1f3SEric Dumazet p->truesize += delta_truesize; 423337fe4732SHerbert Xu p->len += len; 42348a29111cSEric Dumazet if (lp != p) { 42358a29111cSEric Dumazet lp->data_len += len; 42368a29111cSEric Dumazet lp->truesize += delta_truesize; 42378a29111cSEric Dumazet lp->len += len; 42388a29111cSEric Dumazet } 423971d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 424071d93b39SHerbert Xu return 0; 424171d93b39SHerbert Xu } 424271d93b39SHerbert Xu 4243df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4244df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4245df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4246df5042f4SFlorian Westphal 4247df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4248df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4249df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4250df5042f4SFlorian Westphal #endif 42514165079bSFlorian Westphal #ifdef CONFIG_XFRM 42524165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 42534165079bSFlorian Westphal #endif 425495a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 425595a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 425695a7233cSPaul Blakey #endif 42573ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 42583ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 42593ee17bc7SMat Martineau #endif 4260df5042f4SFlorian Westphal }; 4261df5042f4SFlorian Westphal 4262df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4263df5042f4SFlorian Westphal { 4264df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4265df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4266df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4267df5042f4SFlorian Westphal #endif 42684165079bSFlorian Westphal #ifdef CONFIG_XFRM 42694165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 42704165079bSFlorian Westphal #endif 427195a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 427295a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 427395a7233cSPaul Blakey #endif 42743ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 42753ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 42763ee17bc7SMat Martineau #endif 4277df5042f4SFlorian Westphal 0; 4278df5042f4SFlorian Westphal } 4279df5042f4SFlorian Westphal 4280df5042f4SFlorian Westphal static void skb_extensions_init(void) 4281df5042f4SFlorian Westphal { 4282df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4283df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4284df5042f4SFlorian Westphal 4285df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4286df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4287df5042f4SFlorian Westphal 0, 4288df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4289df5042f4SFlorian Westphal NULL); 4290df5042f4SFlorian Westphal } 4291df5042f4SFlorian Westphal #else 4292df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4293df5042f4SFlorian Westphal #endif 4294df5042f4SFlorian Westphal 42951da177e4SLinus Torvalds void __init skb_init(void) 42961da177e4SLinus Torvalds { 429779a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 42981da177e4SLinus Torvalds sizeof(struct sk_buff), 42991da177e4SLinus Torvalds 0, 4300e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 430179a8a642SKees Cook offsetof(struct sk_buff, cb), 430279a8a642SKees Cook sizeof_field(struct sk_buff, cb), 430320c2df83SPaul Mundt NULL); 4304d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4305d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4306d179cd12SDavid S. Miller 0, 4307e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 430820c2df83SPaul Mundt NULL); 4309df5042f4SFlorian Westphal skb_extensions_init(); 43101da177e4SLinus Torvalds } 43111da177e4SLinus Torvalds 431251c739d1SDavid S. Miller static int 431348a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 431448a1df65SJason A. Donenfeld unsigned int recursion_level) 4315716ea3a7SDavid Howells { 43161a028e50SDavid S. Miller int start = skb_headlen(skb); 43171a028e50SDavid S. Miller int i, copy = start - offset; 4318fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4319716ea3a7SDavid Howells int elt = 0; 4320716ea3a7SDavid Howells 432148a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 432248a1df65SJason A. Donenfeld return -EMSGSIZE; 432348a1df65SJason A. Donenfeld 4324716ea3a7SDavid Howells if (copy > 0) { 4325716ea3a7SDavid Howells if (copy > len) 4326716ea3a7SDavid Howells copy = len; 4327642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4328716ea3a7SDavid Howells elt++; 4329716ea3a7SDavid Howells if ((len -= copy) == 0) 4330716ea3a7SDavid Howells return elt; 4331716ea3a7SDavid Howells offset += copy; 4332716ea3a7SDavid Howells } 4333716ea3a7SDavid Howells 4334716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 43351a028e50SDavid S. Miller int end; 4336716ea3a7SDavid Howells 4337547b792cSIlpo Järvinen WARN_ON(start > offset + len); 43381a028e50SDavid S. Miller 43399e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4340716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4341716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 434248a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 434348a1df65SJason A. Donenfeld return -EMSGSIZE; 4344716ea3a7SDavid Howells 4345716ea3a7SDavid Howells if (copy > len) 4346716ea3a7SDavid Howells copy = len; 4347ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4348b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4349716ea3a7SDavid Howells elt++; 4350716ea3a7SDavid Howells if (!(len -= copy)) 4351716ea3a7SDavid Howells return elt; 4352716ea3a7SDavid Howells offset += copy; 4353716ea3a7SDavid Howells } 43541a028e50SDavid S. Miller start = end; 4355716ea3a7SDavid Howells } 4356716ea3a7SDavid Howells 4357fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 435848a1df65SJason A. Donenfeld int end, ret; 4359716ea3a7SDavid Howells 4360547b792cSIlpo Järvinen WARN_ON(start > offset + len); 43611a028e50SDavid S. Miller 4362fbb398a8SDavid S. Miller end = start + frag_iter->len; 4363716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 436448a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 436548a1df65SJason A. Donenfeld return -EMSGSIZE; 436648a1df65SJason A. Donenfeld 4367716ea3a7SDavid Howells if (copy > len) 4368716ea3a7SDavid Howells copy = len; 436948a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 437048a1df65SJason A. Donenfeld copy, recursion_level + 1); 437148a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 437248a1df65SJason A. Donenfeld return ret; 437348a1df65SJason A. Donenfeld elt += ret; 4374716ea3a7SDavid Howells if ((len -= copy) == 0) 4375716ea3a7SDavid Howells return elt; 4376716ea3a7SDavid Howells offset += copy; 4377716ea3a7SDavid Howells } 43781a028e50SDavid S. Miller start = end; 4379716ea3a7SDavid Howells } 4380716ea3a7SDavid Howells BUG_ON(len); 4381716ea3a7SDavid Howells return elt; 4382716ea3a7SDavid Howells } 4383716ea3a7SDavid Howells 438448a1df65SJason A. Donenfeld /** 438548a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 438648a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 438748a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 438848a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 438948a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 439048a1df65SJason A. Donenfeld * 439148a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 439248a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 439348a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 439448a1df65SJason A. Donenfeld * could not fit. 439548a1df65SJason A. Donenfeld */ 439648a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 439748a1df65SJason A. Donenfeld { 439848a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 439948a1df65SJason A. Donenfeld 440048a1df65SJason A. Donenfeld if (nsg <= 0) 440148a1df65SJason A. Donenfeld return nsg; 440248a1df65SJason A. Donenfeld 440348a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 440448a1df65SJason A. Donenfeld 440548a1df65SJason A. Donenfeld return nsg; 440648a1df65SJason A. Donenfeld } 440748a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 440848a1df65SJason A. Donenfeld 440925a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 441025a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 441125a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 441225a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 441325a91d8dSFan Du * 441425a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 441525a91d8dSFan Du * 1. sg_init_table 441625a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 441725a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 441825a91d8dSFan Du * 441925a91d8dSFan Du * This is equivalent to: 442025a91d8dSFan Du * 1. sg_init_table 442125a91d8dSFan Du * 2. skb_to_sgvec(payload1) 442225a91d8dSFan Du * 3. sg_unmark_end 442325a91d8dSFan Du * 4. skb_to_sgvec(payload2) 442425a91d8dSFan Du * 442525a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 442625a91d8dSFan Du * is more preferable. 442725a91d8dSFan Du */ 442825a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 442925a91d8dSFan Du int offset, int len) 443025a91d8dSFan Du { 443148a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 443225a91d8dSFan Du } 443325a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 443425a91d8dSFan Du 443551c739d1SDavid S. Miller 443651c739d1SDavid S. Miller 4437716ea3a7SDavid Howells /** 4438716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4439716ea3a7SDavid Howells * @skb: The socket buffer to check. 4440716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4441716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4442716ea3a7SDavid Howells * 4443716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4444716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4445716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4446716ea3a7SDavid Howells * 4447716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4448716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4449716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4450716ea3a7SDavid Howells * 4451716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4452716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4453716ea3a7SDavid Howells */ 4454716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4455716ea3a7SDavid Howells { 4456716ea3a7SDavid Howells int copyflag; 4457716ea3a7SDavid Howells int elt; 4458716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4459716ea3a7SDavid Howells 4460716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4461716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4462716ea3a7SDavid Howells * at the moment even if they are anonymous). 4463716ea3a7SDavid Howells */ 4464716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4465c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4466716ea3a7SDavid Howells return -ENOMEM; 4467716ea3a7SDavid Howells 4468716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 446921dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4470716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4471716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4472716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4473716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4474716ea3a7SDavid Howells 4475716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4476716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4477716ea3a7SDavid Howells return -ENOMEM; 4478716ea3a7SDavid Howells 4479716ea3a7SDavid Howells /* Voila! */ 4480716ea3a7SDavid Howells *trailer = skb; 4481716ea3a7SDavid Howells return 1; 4482716ea3a7SDavid Howells } 4483716ea3a7SDavid Howells 4484716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4485716ea3a7SDavid Howells 4486716ea3a7SDavid Howells elt = 1; 4487716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4488716ea3a7SDavid Howells copyflag = 0; 4489716ea3a7SDavid Howells 4490716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4491716ea3a7SDavid Howells int ntail = 0; 4492716ea3a7SDavid Howells 4493716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4494716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4495716ea3a7SDavid Howells * after it. */ 4496716ea3a7SDavid Howells 4497716ea3a7SDavid Howells if (skb_shared(skb1)) 4498716ea3a7SDavid Howells copyflag = 1; 4499716ea3a7SDavid Howells 4500716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4501716ea3a7SDavid Howells 4502716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4503716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 450421dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4505716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4506716ea3a7SDavid Howells ntail = tailbits + 128; 4507716ea3a7SDavid Howells } 4508716ea3a7SDavid Howells 4509716ea3a7SDavid Howells if (copyflag || 4510716ea3a7SDavid Howells skb_cloned(skb1) || 4511716ea3a7SDavid Howells ntail || 4512716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 451321dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4514716ea3a7SDavid Howells struct sk_buff *skb2; 4515716ea3a7SDavid Howells 4516716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4517716ea3a7SDavid Howells if (ntail == 0) 4518716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4519716ea3a7SDavid Howells else 4520716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4521716ea3a7SDavid Howells skb_headroom(skb1), 4522716ea3a7SDavid Howells ntail, 4523716ea3a7SDavid Howells GFP_ATOMIC); 4524716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4525716ea3a7SDavid Howells return -ENOMEM; 4526716ea3a7SDavid Howells 4527716ea3a7SDavid Howells if (skb1->sk) 4528716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4529716ea3a7SDavid Howells 4530716ea3a7SDavid Howells /* Looking around. Are we still alive? 4531716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4532716ea3a7SDavid Howells 4533716ea3a7SDavid Howells skb2->next = skb1->next; 4534716ea3a7SDavid Howells *skb_p = skb2; 4535716ea3a7SDavid Howells kfree_skb(skb1); 4536716ea3a7SDavid Howells skb1 = skb2; 4537716ea3a7SDavid Howells } 4538716ea3a7SDavid Howells elt++; 4539716ea3a7SDavid Howells *trailer = skb1; 4540716ea3a7SDavid Howells skb_p = &skb1->next; 4541716ea3a7SDavid Howells } 4542716ea3a7SDavid Howells 4543716ea3a7SDavid Howells return elt; 4544716ea3a7SDavid Howells } 4545b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4546716ea3a7SDavid Howells 4547b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4548b1faf566SEric Dumazet { 4549b1faf566SEric Dumazet struct sock *sk = skb->sk; 4550b1faf566SEric Dumazet 4551b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4552b1faf566SEric Dumazet } 4553b1faf566SEric Dumazet 45548605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 45558605330aSSoheil Hassas Yeganeh { 45568605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 45578605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 45588605330aSSoheil Hassas Yeganeh */ 45598605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 45608605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 45618605330aSSoheil Hassas Yeganeh } 45628605330aSSoheil Hassas Yeganeh 4563b1faf566SEric Dumazet /* 4564b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4565b1faf566SEric Dumazet */ 4566b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4567b1faf566SEric Dumazet { 4568b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4569ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4570b1faf566SEric Dumazet return -ENOMEM; 4571b1faf566SEric Dumazet 4572b1faf566SEric Dumazet skb_orphan(skb); 4573b1faf566SEric Dumazet skb->sk = sk; 4574b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4575b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 45768605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4577b1faf566SEric Dumazet 4578abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4579abb57ea4SEric Dumazet skb_dst_force(skb); 4580abb57ea4SEric Dumazet 4581b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4582b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 45836e5d58fdSVinicius Costa Gomes sk->sk_error_report(sk); 4584b1faf566SEric Dumazet return 0; 4585b1faf566SEric Dumazet } 4586b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4587b1faf566SEric Dumazet 458883a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 458983a1a1a7SSoheil Hassas Yeganeh { 459083a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 459183a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 459283a1a1a7SSoheil Hassas Yeganeh } 459383a1a1a7SSoheil Hassas Yeganeh 4594364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4595364a9e93SWillem de Bruijn { 4596364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 459783a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 459883a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4599997d5c3fSEric Dumazet unsigned long flags; 4600364a9e93SWillem de Bruijn 4601997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4602364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 460338b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 460483a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 460538b25793SSoheil Hassas Yeganeh if (icmp_next) 4606985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 460738b25793SSoheil Hassas Yeganeh } 4608997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4609364a9e93SWillem de Bruijn 461083a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 461183a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 461283a1a1a7SSoheil Hassas Yeganeh 461383a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4614364a9e93SWillem de Bruijn sk->sk_error_report(sk); 4615364a9e93SWillem de Bruijn 4616364a9e93SWillem de Bruijn return skb; 4617364a9e93SWillem de Bruijn } 4618364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4619364a9e93SWillem de Bruijn 4620cab41c47SAlexander Duyck /** 4621cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4622cab41c47SAlexander Duyck * @skb: the skb to clone 4623cab41c47SAlexander Duyck * 4624cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4625cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4626cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4627cab41c47SAlexander Duyck * 4628cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4629cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4630cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4631cab41c47SAlexander Duyck * the sk_error_queue. 4632cab41c47SAlexander Duyck */ 463362bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 463462bccb8cSAlexander Duyck { 463562bccb8cSAlexander Duyck struct sock *sk = skb->sk; 463662bccb8cSAlexander Duyck struct sk_buff *clone; 463762bccb8cSAlexander Duyck 463841c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 463962bccb8cSAlexander Duyck return NULL; 464062bccb8cSAlexander Duyck 464162bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 464262bccb8cSAlexander Duyck if (!clone) { 464362bccb8cSAlexander Duyck sock_put(sk); 464462bccb8cSAlexander Duyck return NULL; 464562bccb8cSAlexander Duyck } 464662bccb8cSAlexander Duyck 464762bccb8cSAlexander Duyck clone->sk = sk; 464862bccb8cSAlexander Duyck clone->destructor = sock_efree; 464962bccb8cSAlexander Duyck 465062bccb8cSAlexander Duyck return clone; 465162bccb8cSAlexander Duyck } 465262bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 465362bccb8cSAlexander Duyck 465437846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 465537846ef0SAlexander Duyck struct sock *sk, 46564ef1b286SSoheil Hassas Yeganeh int tstype, 46574ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4658ac45f602SPatrick Ohly { 4659ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4660ac45f602SPatrick Ohly int err; 4661ac45f602SPatrick Ohly 46624ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 46634ef1b286SSoheil Hassas Yeganeh 4664ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4665ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4666ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4667ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4668e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 46694ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 46701862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 46714ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 467209c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 4673ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 4674ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 46754ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 46764ed2d765SWillem de Bruijn } 467729030374SEric Dumazet 4678ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 467929030374SEric Dumazet 4680ac45f602SPatrick Ohly if (err) 4681ac45f602SPatrick Ohly kfree_skb(skb); 4682ac45f602SPatrick Ohly } 468337846ef0SAlexander Duyck 4684b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4685b245be1fSWillem de Bruijn { 4686b245be1fSWillem de Bruijn bool ret; 4687b245be1fSWillem de Bruijn 4688b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4689b245be1fSWillem de Bruijn return true; 4690b245be1fSWillem de Bruijn 4691b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4692b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4693b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4694b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4695b245be1fSWillem de Bruijn return ret; 4696b245be1fSWillem de Bruijn } 4697b245be1fSWillem de Bruijn 469837846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 469937846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 470037846ef0SAlexander Duyck { 470137846ef0SAlexander Duyck struct sock *sk = skb->sk; 470237846ef0SAlexander Duyck 4703b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 470435b99dffSWillem de Bruijn goto err; 4705b245be1fSWillem de Bruijn 47069ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 47079ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 47089ac25fc0SEric Dumazet */ 470941c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 471037846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 47114ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 471237846ef0SAlexander Duyck sock_put(sk); 471335b99dffSWillem de Bruijn return; 471437846ef0SAlexander Duyck } 471535b99dffSWillem de Bruijn 471635b99dffSWillem de Bruijn err: 471735b99dffSWillem de Bruijn kfree_skb(skb); 47189ac25fc0SEric Dumazet } 471937846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 472037846ef0SAlexander Duyck 472137846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 472237846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 472337846ef0SAlexander Duyck struct sock *sk, int tstype) 472437846ef0SAlexander Duyck { 472537846ef0SAlexander Duyck struct sk_buff *skb; 47264ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 472737846ef0SAlexander Duyck 47283a8dd971SWillem de Bruijn if (!sk) 47293a8dd971SWillem de Bruijn return; 47303a8dd971SWillem de Bruijn 4731b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4732b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4733b50a5c70SMiroslav Lichvar return; 4734b50a5c70SMiroslav Lichvar 47353a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 47363a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 473737846ef0SAlexander Duyck return; 473837846ef0SAlexander Duyck 47391c885808SFrancis Yan if (tsonly) { 47401c885808SFrancis Yan #ifdef CONFIG_INET 47411c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 47421c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 47434ef1b286SSoheil Hassas Yeganeh sk->sk_type == SOCK_STREAM) { 474448040793SYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb); 47454ef1b286SSoheil Hassas Yeganeh opt_stats = true; 47464ef1b286SSoheil Hassas Yeganeh } else 47471c885808SFrancis Yan #endif 47481c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 47491c885808SFrancis Yan } else { 475037846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 47511c885808SFrancis Yan } 475237846ef0SAlexander Duyck if (!skb) 475337846ef0SAlexander Duyck return; 475437846ef0SAlexander Duyck 475549ca0d8bSWillem de Bruijn if (tsonly) { 4756fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4757fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 475849ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 475949ca0d8bSWillem de Bruijn } 476049ca0d8bSWillem de Bruijn 476149ca0d8bSWillem de Bruijn if (hwtstamps) 476249ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 476349ca0d8bSWillem de Bruijn else 476449ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 476549ca0d8bSWillem de Bruijn 47664ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 476737846ef0SAlexander Duyck } 4768e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4769e7fd2885SWillem de Bruijn 4770e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4771e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4772e7fd2885SWillem de Bruijn { 4773e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 4774e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4775e7fd2885SWillem de Bruijn } 4776ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4777ac45f602SPatrick Ohly 47786e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 47796e3e939fSJohannes Berg { 47806e3e939fSJohannes Berg struct sock *sk = skb->sk; 47816e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4782dd4f1072SEric Dumazet int err = 1; 47836e3e939fSJohannes Berg 47846e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 47856e3e939fSJohannes Berg skb->wifi_acked = acked; 47866e3e939fSJohannes Berg 47876e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 47886e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 47896e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 47906e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 47916e3e939fSJohannes Berg 4792dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4793dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4794dd4f1072SEric Dumazet */ 479541c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 47966e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4797dd4f1072SEric Dumazet sock_put(sk); 4798dd4f1072SEric Dumazet } 47996e3e939fSJohannes Berg if (err) 48006e3e939fSJohannes Berg kfree_skb(skb); 48016e3e939fSJohannes Berg } 48026e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 48036e3e939fSJohannes Berg 4804f35d9d8aSRusty Russell /** 4805f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4806f35d9d8aSRusty Russell * @skb: the skb to set 4807f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4808f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4809f35d9d8aSRusty Russell * 4810f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4811f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4812f35d9d8aSRusty Russell * 4813f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4814f35d9d8aSRusty Russell * returns false you should drop the packet. 4815f35d9d8aSRusty Russell */ 4816f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4817f35d9d8aSRusty Russell { 481852b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 481952b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 482052b5d6f5SEric Dumazet 482152b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 482252b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 482352b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4824f35d9d8aSRusty Russell return false; 4825f35d9d8aSRusty Russell } 4826f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 482752b5d6f5SEric Dumazet skb->csum_start = csum_start; 4828f35d9d8aSRusty Russell skb->csum_offset = off; 4829e5d5decaSJason Wang skb_set_transport_header(skb, start); 4830f35d9d8aSRusty Russell return true; 4831f35d9d8aSRusty Russell } 4832b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4833f35d9d8aSRusty Russell 4834ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4835ed1f50c3SPaul Durrant unsigned int max) 4836ed1f50c3SPaul Durrant { 4837ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4838ed1f50c3SPaul Durrant return 0; 4839ed1f50c3SPaul Durrant 4840ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4841ed1f50c3SPaul Durrant * won't need to do it again. 4842ed1f50c3SPaul Durrant */ 4843ed1f50c3SPaul Durrant if (max > skb->len) 4844ed1f50c3SPaul Durrant max = skb->len; 4845ed1f50c3SPaul Durrant 4846ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4847ed1f50c3SPaul Durrant return -ENOMEM; 4848ed1f50c3SPaul Durrant 4849ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4850ed1f50c3SPaul Durrant return -EPROTO; 4851ed1f50c3SPaul Durrant 4852ed1f50c3SPaul Durrant return 0; 4853ed1f50c3SPaul Durrant } 4854ed1f50c3SPaul Durrant 4855f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4856f9708b43SJan Beulich 4857f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4858f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4859f9708b43SJan Beulich unsigned int off) 4860f9708b43SJan Beulich { 4861f9708b43SJan Beulich int err; 4862f9708b43SJan Beulich 4863161d1792SKees Cook switch (proto) { 4864f9708b43SJan Beulich case IPPROTO_TCP: 4865f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4866f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4867f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4868f9708b43SJan Beulich offsetof(struct tcphdr, 4869f9708b43SJan Beulich check))) 4870f9708b43SJan Beulich err = -EPROTO; 4871f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4872f9708b43SJan Beulich 4873f9708b43SJan Beulich case IPPROTO_UDP: 4874f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4875f9708b43SJan Beulich off + sizeof(struct udphdr)); 4876f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4877f9708b43SJan Beulich offsetof(struct udphdr, 4878f9708b43SJan Beulich check))) 4879f9708b43SJan Beulich err = -EPROTO; 4880f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4881f9708b43SJan Beulich } 4882f9708b43SJan Beulich 4883f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4884f9708b43SJan Beulich } 4885f9708b43SJan Beulich 4886ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4887ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4888ed1f50c3SPaul Durrant */ 4889ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4890ed1f50c3SPaul Durrant 4891f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4892ed1f50c3SPaul Durrant { 4893ed1f50c3SPaul Durrant unsigned int off; 4894ed1f50c3SPaul Durrant bool fragment; 4895f9708b43SJan Beulich __sum16 *csum; 4896ed1f50c3SPaul Durrant int err; 4897ed1f50c3SPaul Durrant 4898ed1f50c3SPaul Durrant fragment = false; 4899ed1f50c3SPaul Durrant 4900ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4901ed1f50c3SPaul Durrant sizeof(struct iphdr), 4902ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4903ed1f50c3SPaul Durrant if (err < 0) 4904ed1f50c3SPaul Durrant goto out; 4905ed1f50c3SPaul Durrant 490611f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 4907ed1f50c3SPaul Durrant fragment = true; 4908ed1f50c3SPaul Durrant 4909ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 4910ed1f50c3SPaul Durrant 4911ed1f50c3SPaul Durrant err = -EPROTO; 4912ed1f50c3SPaul Durrant 4913ed1f50c3SPaul Durrant if (fragment) 4914ed1f50c3SPaul Durrant goto out; 4915ed1f50c3SPaul Durrant 4916f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4917f9708b43SJan Beulich if (IS_ERR(csum)) 4918f9708b43SJan Beulich return PTR_ERR(csum); 4919ed1f50c3SPaul Durrant 4920ed1f50c3SPaul Durrant if (recalculate) 4921f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4922ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4923ed1f50c3SPaul Durrant skb->len - off, 4924f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4925ed1f50c3SPaul Durrant err = 0; 4926ed1f50c3SPaul Durrant 4927ed1f50c3SPaul Durrant out: 4928ed1f50c3SPaul Durrant return err; 4929ed1f50c3SPaul Durrant } 4930ed1f50c3SPaul Durrant 4931ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4932ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4933ed1f50c3SPaul Durrant */ 4934ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4935ed1f50c3SPaul Durrant 4936ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4937ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4938ed1f50c3SPaul Durrant 4939ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4940ed1f50c3SPaul Durrant { 4941ed1f50c3SPaul Durrant int err; 4942ed1f50c3SPaul Durrant u8 nexthdr; 4943ed1f50c3SPaul Durrant unsigned int off; 4944ed1f50c3SPaul Durrant unsigned int len; 4945ed1f50c3SPaul Durrant bool fragment; 4946ed1f50c3SPaul Durrant bool done; 4947f9708b43SJan Beulich __sum16 *csum; 4948ed1f50c3SPaul Durrant 4949ed1f50c3SPaul Durrant fragment = false; 4950ed1f50c3SPaul Durrant done = false; 4951ed1f50c3SPaul Durrant 4952ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4953ed1f50c3SPaul Durrant 4954ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4955ed1f50c3SPaul Durrant if (err < 0) 4956ed1f50c3SPaul Durrant goto out; 4957ed1f50c3SPaul Durrant 4958ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4959ed1f50c3SPaul Durrant 4960ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4961ed1f50c3SPaul Durrant while (off <= len && !done) { 4962ed1f50c3SPaul Durrant switch (nexthdr) { 4963ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4964ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4965ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4966ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 4967ed1f50c3SPaul Durrant 4968ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4969ed1f50c3SPaul Durrant off + 4970ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 4971ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4972ed1f50c3SPaul Durrant if (err < 0) 4973ed1f50c3SPaul Durrant goto out; 4974ed1f50c3SPaul Durrant 4975ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4976ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4977ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 4978ed1f50c3SPaul Durrant break; 4979ed1f50c3SPaul Durrant } 4980ed1f50c3SPaul Durrant case IPPROTO_AH: { 4981ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 4982ed1f50c3SPaul Durrant 4983ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4984ed1f50c3SPaul Durrant off + 4985ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 4986ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4987ed1f50c3SPaul Durrant if (err < 0) 4988ed1f50c3SPaul Durrant goto out; 4989ed1f50c3SPaul Durrant 4990ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4991ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4992ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 4993ed1f50c3SPaul Durrant break; 4994ed1f50c3SPaul Durrant } 4995ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 4996ed1f50c3SPaul Durrant struct frag_hdr *hp; 4997ed1f50c3SPaul Durrant 4998ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4999ed1f50c3SPaul Durrant off + 5000ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5001ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5002ed1f50c3SPaul Durrant if (err < 0) 5003ed1f50c3SPaul Durrant goto out; 5004ed1f50c3SPaul Durrant 5005ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5006ed1f50c3SPaul Durrant 5007ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5008ed1f50c3SPaul Durrant fragment = true; 5009ed1f50c3SPaul Durrant 5010ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5011ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5012ed1f50c3SPaul Durrant break; 5013ed1f50c3SPaul Durrant } 5014ed1f50c3SPaul Durrant default: 5015ed1f50c3SPaul Durrant done = true; 5016ed1f50c3SPaul Durrant break; 5017ed1f50c3SPaul Durrant } 5018ed1f50c3SPaul Durrant } 5019ed1f50c3SPaul Durrant 5020ed1f50c3SPaul Durrant err = -EPROTO; 5021ed1f50c3SPaul Durrant 5022ed1f50c3SPaul Durrant if (!done || fragment) 5023ed1f50c3SPaul Durrant goto out; 5024ed1f50c3SPaul Durrant 5025f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5026f9708b43SJan Beulich if (IS_ERR(csum)) 5027f9708b43SJan Beulich return PTR_ERR(csum); 5028ed1f50c3SPaul Durrant 5029ed1f50c3SPaul Durrant if (recalculate) 5030f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5031ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5032f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5033ed1f50c3SPaul Durrant err = 0; 5034ed1f50c3SPaul Durrant 5035ed1f50c3SPaul Durrant out: 5036ed1f50c3SPaul Durrant return err; 5037ed1f50c3SPaul Durrant } 5038ed1f50c3SPaul Durrant 5039ed1f50c3SPaul Durrant /** 5040ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5041ed1f50c3SPaul Durrant * @skb: the skb to set up 5042ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5043ed1f50c3SPaul Durrant */ 5044ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5045ed1f50c3SPaul Durrant { 5046ed1f50c3SPaul Durrant int err; 5047ed1f50c3SPaul Durrant 5048ed1f50c3SPaul Durrant switch (skb->protocol) { 5049ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5050f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5051ed1f50c3SPaul Durrant break; 5052ed1f50c3SPaul Durrant 5053ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5054ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5055ed1f50c3SPaul Durrant break; 5056ed1f50c3SPaul Durrant 5057ed1f50c3SPaul Durrant default: 5058ed1f50c3SPaul Durrant err = -EPROTO; 5059ed1f50c3SPaul Durrant break; 5060ed1f50c3SPaul Durrant } 5061ed1f50c3SPaul Durrant 5062ed1f50c3SPaul Durrant return err; 5063ed1f50c3SPaul Durrant } 5064ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5065ed1f50c3SPaul Durrant 50669afd85c9SLinus Lüssing /** 50679afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 50689afd85c9SLinus Lüssing * @skb: the skb to check 50699afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 50709afd85c9SLinus Lüssing * 50719afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 50729afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 50739afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 50749afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 50759afd85c9SLinus Lüssing * 5076a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5077a516993fSLinus Lüssing * differs from the provided skb. 50789afd85c9SLinus Lüssing */ 50799afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 50809afd85c9SLinus Lüssing unsigned int transport_len) 50819afd85c9SLinus Lüssing { 50829afd85c9SLinus Lüssing struct sk_buff *skb_chk; 50839afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 50849afd85c9SLinus Lüssing int ret; 50859afd85c9SLinus Lüssing 5086a516993fSLinus Lüssing if (skb->len < len) 50879afd85c9SLinus Lüssing return NULL; 5088a516993fSLinus Lüssing else if (skb->len == len) 50899afd85c9SLinus Lüssing return skb; 50909afd85c9SLinus Lüssing 50919afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 50929afd85c9SLinus Lüssing if (!skb_chk) 50939afd85c9SLinus Lüssing return NULL; 50949afd85c9SLinus Lüssing 50959afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 50969afd85c9SLinus Lüssing if (ret) { 50979afd85c9SLinus Lüssing kfree_skb(skb_chk); 50989afd85c9SLinus Lüssing return NULL; 50999afd85c9SLinus Lüssing } 51009afd85c9SLinus Lüssing 51019afd85c9SLinus Lüssing return skb_chk; 51029afd85c9SLinus Lüssing } 51039afd85c9SLinus Lüssing 51049afd85c9SLinus Lüssing /** 51059afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 51069afd85c9SLinus Lüssing * @skb: the skb to check 51079afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51089afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 51099afd85c9SLinus Lüssing * 51109afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 51119afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 51129afd85c9SLinus Lüssing * 51139afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 51149afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 51159afd85c9SLinus Lüssing * 5116a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5117a516993fSLinus Lüssing * differs from the provided skb. 51189afd85c9SLinus Lüssing */ 51199afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 51209afd85c9SLinus Lüssing unsigned int transport_len, 51219afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 51229afd85c9SLinus Lüssing { 51239afd85c9SLinus Lüssing struct sk_buff *skb_chk; 51249afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5125fcba67c9SLinus Lüssing __sum16 ret; 51269afd85c9SLinus Lüssing 51279afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 51289afd85c9SLinus Lüssing if (!skb_chk) 5129a516993fSLinus Lüssing goto err; 51309afd85c9SLinus Lüssing 5131a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5132a516993fSLinus Lüssing goto err; 51339afd85c9SLinus Lüssing 51349b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 51359afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 51369b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 51379afd85c9SLinus Lüssing 5138a516993fSLinus Lüssing if (ret) 5139a516993fSLinus Lüssing goto err; 51409afd85c9SLinus Lüssing 51419afd85c9SLinus Lüssing return skb_chk; 5142a516993fSLinus Lüssing 5143a516993fSLinus Lüssing err: 5144a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5145a516993fSLinus Lüssing kfree_skb(skb_chk); 5146a516993fSLinus Lüssing 5147a516993fSLinus Lüssing return NULL; 5148a516993fSLinus Lüssing 51499afd85c9SLinus Lüssing } 51509afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 51519afd85c9SLinus Lüssing 51524497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 51534497b076SBen Hutchings { 5154e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5155e87cc472SJoe Perches skb->dev->name); 51564497b076SBen Hutchings } 51574497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5158bad43ca8SEric Dumazet 5159bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5160bad43ca8SEric Dumazet { 51613d861f66SEric Dumazet if (head_stolen) { 51623d861f66SEric Dumazet skb_release_head_state(skb); 5163bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 51643d861f66SEric Dumazet } else { 5165bad43ca8SEric Dumazet __kfree_skb(skb); 5166bad43ca8SEric Dumazet } 51673d861f66SEric Dumazet } 5168bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5169bad43ca8SEric Dumazet 5170bad43ca8SEric Dumazet /** 5171bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5172bad43ca8SEric Dumazet * @to: prior buffer 5173bad43ca8SEric Dumazet * @from: buffer to add 5174bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5175c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5176bad43ca8SEric Dumazet */ 5177bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5178bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5179bad43ca8SEric Dumazet { 5180c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5181bad43ca8SEric Dumazet int i, delta, len = from->len; 5182bad43ca8SEric Dumazet 5183bad43ca8SEric Dumazet *fragstolen = false; 5184bad43ca8SEric Dumazet 5185bad43ca8SEric Dumazet if (skb_cloned(to)) 5186bad43ca8SEric Dumazet return false; 5187bad43ca8SEric Dumazet 5188bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5189e93a0435SEric Dumazet if (len) 5190bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5191bad43ca8SEric Dumazet *delta_truesize = 0; 5192bad43ca8SEric Dumazet return true; 5193bad43ca8SEric Dumazet } 5194bad43ca8SEric Dumazet 5195c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5196c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5197c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5198bad43ca8SEric Dumazet return false; 51991f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 52001f8b977aSWillem de Bruijn return false; 5201bad43ca8SEric Dumazet 5202bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5203bad43ca8SEric Dumazet struct page *page; 5204bad43ca8SEric Dumazet unsigned int offset; 5205bad43ca8SEric Dumazet 5206c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5207c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5208bad43ca8SEric Dumazet return false; 5209bad43ca8SEric Dumazet 5210bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5211bad43ca8SEric Dumazet return false; 5212bad43ca8SEric Dumazet 5213bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5214bad43ca8SEric Dumazet 5215bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5216bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5217bad43ca8SEric Dumazet 5218c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5219bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5220bad43ca8SEric Dumazet *fragstolen = true; 5221bad43ca8SEric Dumazet } else { 5222c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5223c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5224bad43ca8SEric Dumazet return false; 5225bad43ca8SEric Dumazet 5226f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5227bad43ca8SEric Dumazet } 5228bad43ca8SEric Dumazet 5229bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5230bad43ca8SEric Dumazet 5231c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5232c818fa9eSEric Dumazet from_shinfo->frags, 5233c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5234c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5235bad43ca8SEric Dumazet 5236bad43ca8SEric Dumazet if (!skb_cloned(from)) 5237c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5238bad43ca8SEric Dumazet 52398ea853fdSLi RongQing /* if the skb is not cloned this does nothing 52408ea853fdSLi RongQing * since we set nr_frags to 0. 52418ea853fdSLi RongQing */ 5242c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5243c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5244bad43ca8SEric Dumazet 5245bad43ca8SEric Dumazet to->truesize += delta; 5246bad43ca8SEric Dumazet to->len += len; 5247bad43ca8SEric Dumazet to->data_len += len; 5248bad43ca8SEric Dumazet 5249bad43ca8SEric Dumazet *delta_truesize = delta; 5250bad43ca8SEric Dumazet return true; 5251bad43ca8SEric Dumazet } 5252bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5253621e84d6SNicolas Dichtel 5254621e84d6SNicolas Dichtel /** 52558b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5256621e84d6SNicolas Dichtel * 5257621e84d6SNicolas Dichtel * @skb: buffer to clean 52588b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5259621e84d6SNicolas Dichtel * 52608b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 52618b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 52628b27f277SNicolas Dichtel * operations. 52638b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 52648b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 52658b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5266621e84d6SNicolas Dichtel */ 52678b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5268621e84d6SNicolas Dichtel { 5269621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5270621e84d6SNicolas Dichtel skb->skb_iif = 0; 527160ff7467SWANG Cong skb->ignore_df = 0; 5272621e84d6SNicolas Dichtel skb_dst_drop(skb); 5273174e2381SFlorian Westphal skb_ext_reset(skb); 5274895b5c9fSFlorian Westphal nf_reset_ct(skb); 5275621e84d6SNicolas Dichtel nf_reset_trace(skb); 5276213dd74aSHerbert Xu 52776f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 52786f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5279875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 52806f9a5069SPetr Machata #endif 52816f9a5069SPetr Machata 5282213dd74aSHerbert Xu if (!xnet) 5283213dd74aSHerbert Xu return; 5284213dd74aSHerbert Xu 52852b5ec1a5SYe Yin ipvs_reset(skb); 5286213dd74aSHerbert Xu skb->mark = 0; 5287c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 5288621e84d6SNicolas Dichtel } 5289621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5290de960aa9SFlorian Westphal 5291de960aa9SFlorian Westphal /** 5292de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5293de960aa9SFlorian Westphal * 5294de960aa9SFlorian Westphal * @skb: GSO skb 5295de960aa9SFlorian Westphal * 5296de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5297de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5298de960aa9SFlorian Westphal * 5299de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5300de960aa9SFlorian Westphal */ 5301a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5302de960aa9SFlorian Westphal { 5303de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5304f993bc25SFlorian Westphal unsigned int thlen = 0; 5305f993bc25SFlorian Westphal 5306f993bc25SFlorian Westphal if (skb->encapsulation) { 5307f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5308f993bc25SFlorian Westphal skb_transport_header(skb); 5309de960aa9SFlorian Westphal 5310de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5311f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5312f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5313f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 53141dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 531590017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5316ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5317ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5318f993bc25SFlorian Westphal } 53196d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 53206d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 53216d39d589SFlorian Westphal * accounted for. 53226d39d589SFlorian Westphal */ 5323f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5324de960aa9SFlorian Westphal } 5325a4a77718SDaniel Axtens 5326a4a77718SDaniel Axtens /** 5327a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5328a4a77718SDaniel Axtens * 5329a4a77718SDaniel Axtens * @skb: GSO skb 5330a4a77718SDaniel Axtens * 5331a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5332a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5333a4a77718SDaniel Axtens * 5334a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5335a4a77718SDaniel Axtens */ 5336a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5337a4a77718SDaniel Axtens { 5338a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5339a4a77718SDaniel Axtens skb_network_header(skb); 5340a4a77718SDaniel Axtens 5341a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5342a4a77718SDaniel Axtens } 5343a4a77718SDaniel Axtens 5344a4a77718SDaniel Axtens /** 5345a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5346a4a77718SDaniel Axtens * 5347a4a77718SDaniel Axtens * @skb: GSO skb 5348a4a77718SDaniel Axtens * 5349a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5350a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5351a4a77718SDaniel Axtens * headers (TCP/UDP). 5352a4a77718SDaniel Axtens */ 5353a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5354a4a77718SDaniel Axtens { 5355a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5356a4a77718SDaniel Axtens 5357a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5358a4a77718SDaniel Axtens } 53590d5501c1SVlad Yasevich 5360ae7ef81eSMarcelo Ricardo Leitner /** 53612b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 53622b16f048SDaniel Axtens * 53632b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 53642b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 53652b16f048SDaniel Axtens * 53662b16f048SDaniel Axtens * We might want to check: 53672b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 53682b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 53692b16f048SDaniel Axtens * 53702b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 53712b16f048SDaniel Axtens * 537249682bfaSMathieu Malaterre * @skb: GSO skb 537349682bfaSMathieu Malaterre * 53742b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 53752b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 53762b16f048SDaniel Axtens * 53772b16f048SDaniel Axtens * @max_len: The maximum permissible length. 53782b16f048SDaniel Axtens * 53792b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 53802b16f048SDaniel Axtens */ 53812b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 53822b16f048SDaniel Axtens unsigned int seg_len, 53832b16f048SDaniel Axtens unsigned int max_len) { 53842b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 53852b16f048SDaniel Axtens const struct sk_buff *iter; 53862b16f048SDaniel Axtens 53872b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 53882b16f048SDaniel Axtens return seg_len <= max_len; 53892b16f048SDaniel Axtens 53902b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 53912b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 53922b16f048SDaniel Axtens 53932b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 53942b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 53952b16f048SDaniel Axtens return false; 53962b16f048SDaniel Axtens } 53972b16f048SDaniel Axtens 53982b16f048SDaniel Axtens return true; 53992b16f048SDaniel Axtens } 54002b16f048SDaniel Axtens 54012b16f048SDaniel Axtens /** 5402779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5403ae7ef81eSMarcelo Ricardo Leitner * 5404ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 540576f21b99SDavid S. Miller * @mtu: MTU to validate against 5406ae7ef81eSMarcelo Ricardo Leitner * 5407779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5408779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5409779b7931SDaniel Axtens * payload. 5410ae7ef81eSMarcelo Ricardo Leitner */ 5411779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5412ae7ef81eSMarcelo Ricardo Leitner { 54132b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5414ae7ef81eSMarcelo Ricardo Leitner } 5415779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5416ae7ef81eSMarcelo Ricardo Leitner 54172b16f048SDaniel Axtens /** 54182b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 54192b16f048SDaniel Axtens * 54202b16f048SDaniel Axtens * @skb: GSO skb 54212b16f048SDaniel Axtens * @len: length to validate against 54222b16f048SDaniel Axtens * 54232b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 54242b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 54252b16f048SDaniel Axtens */ 54262b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 54272b16f048SDaniel Axtens { 54282b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 54292b16f048SDaniel Axtens } 54302b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 54312b16f048SDaniel Axtens 54320d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 54330d5501c1SVlad Yasevich { 5434d85e8be2SYuya Kusakabe int mac_len, meta_len; 5435d85e8be2SYuya Kusakabe void *meta; 54364bbb3e0eSToshiaki Makita 54370d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 54380d5501c1SVlad Yasevich kfree_skb(skb); 54390d5501c1SVlad Yasevich return NULL; 54400d5501c1SVlad Yasevich } 54410d5501c1SVlad Yasevich 54424bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5443ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 54444bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 54454bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5446ae474573SToshiaki Makita } 5447d85e8be2SYuya Kusakabe 5448d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5449d85e8be2SYuya Kusakabe if (meta_len) { 5450d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5451d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5452d85e8be2SYuya Kusakabe } 5453d85e8be2SYuya Kusakabe 54540d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 54550d5501c1SVlad Yasevich return skb; 54560d5501c1SVlad Yasevich } 54570d5501c1SVlad Yasevich 54580d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 54590d5501c1SVlad Yasevich { 54600d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 54610d5501c1SVlad Yasevich u16 vlan_tci; 54620d5501c1SVlad Yasevich 5463df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 54640d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 54650d5501c1SVlad Yasevich return skb; 54660d5501c1SVlad Yasevich } 54670d5501c1SVlad Yasevich 54680d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 54690d5501c1SVlad Yasevich if (unlikely(!skb)) 54700d5501c1SVlad Yasevich goto err_free; 547155eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 547255eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 54730d5501c1SVlad Yasevich goto err_free; 54740d5501c1SVlad Yasevich 54750d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 54760d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 54770d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 54780d5501c1SVlad Yasevich 54790d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 54800d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 54810d5501c1SVlad Yasevich 54820d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 54830d5501c1SVlad Yasevich if (unlikely(!skb)) 54840d5501c1SVlad Yasevich goto err_free; 54850d5501c1SVlad Yasevich 54860d5501c1SVlad Yasevich skb_reset_network_header(skb); 54878be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 54880d5501c1SVlad Yasevich skb_reset_transport_header(skb); 54890d5501c1SVlad Yasevich skb_reset_mac_len(skb); 54900d5501c1SVlad Yasevich 54910d5501c1SVlad Yasevich return skb; 54920d5501c1SVlad Yasevich 54930d5501c1SVlad Yasevich err_free: 54940d5501c1SVlad Yasevich kfree_skb(skb); 54950d5501c1SVlad Yasevich return NULL; 54960d5501c1SVlad Yasevich } 54970d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 54982e4e4410SEric Dumazet 5499e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5500e2195121SJiri Pirko { 5501e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5502e2195121SJiri Pirko return -ENOMEM; 5503e2195121SJiri Pirko 5504e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5505e2195121SJiri Pirko return 0; 5506e2195121SJiri Pirko 5507e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5508e2195121SJiri Pirko } 5509e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5510e2195121SJiri Pirko 5511bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5512bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5513bfca4c52SShmulik Ladkani */ 5514bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 551593515d53SJiri Pirko { 551693515d53SJiri Pirko struct vlan_hdr *vhdr; 5517b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 551893515d53SJiri Pirko int err; 551993515d53SJiri Pirko 5520b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5521b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5522b6a79208SShmulik Ladkani offset)) { 5523b6a79208SShmulik Ladkani return -EINVAL; 5524b6a79208SShmulik Ladkani } 5525b6a79208SShmulik Ladkani 552693515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 552793515d53SJiri Pirko if (unlikely(err)) 5528b6a79208SShmulik Ladkani return err; 552993515d53SJiri Pirko 553093515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 553193515d53SJiri Pirko 553293515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 553393515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 553493515d53SJiri Pirko 553593515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 553693515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 553793515d53SJiri Pirko 553893515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 553993515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 554093515d53SJiri Pirko 554193515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 554293515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 554393515d53SJiri Pirko 554493515d53SJiri Pirko skb_reset_mac_len(skb); 554593515d53SJiri Pirko 554693515d53SJiri Pirko return err; 554793515d53SJiri Pirko } 5548bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 554993515d53SJiri Pirko 5550b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5551b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5552b6a79208SShmulik Ladkani */ 555393515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 555493515d53SJiri Pirko { 555593515d53SJiri Pirko u16 vlan_tci; 555693515d53SJiri Pirko __be16 vlan_proto; 555793515d53SJiri Pirko int err; 555893515d53SJiri Pirko 5559df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5560b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 556193515d53SJiri Pirko } else { 5562ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 556393515d53SJiri Pirko return 0; 556493515d53SJiri Pirko 556593515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 556693515d53SJiri Pirko if (err) 556793515d53SJiri Pirko return err; 556893515d53SJiri Pirko } 556993515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5570ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 557193515d53SJiri Pirko return 0; 557293515d53SJiri Pirko 557393515d53SJiri Pirko vlan_proto = skb->protocol; 557493515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 557593515d53SJiri Pirko if (unlikely(err)) 557693515d53SJiri Pirko return err; 557793515d53SJiri Pirko 557893515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 557993515d53SJiri Pirko return 0; 558093515d53SJiri Pirko } 558193515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 558293515d53SJiri Pirko 5583b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5584b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5585b6a79208SShmulik Ladkani */ 558693515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 558793515d53SJiri Pirko { 5588df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5589b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 559093515d53SJiri Pirko int err; 559193515d53SJiri Pirko 5592b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5593b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5594b6a79208SShmulik Ladkani offset)) { 5595b6a79208SShmulik Ladkani return -EINVAL; 5596b6a79208SShmulik Ladkani } 5597b6a79208SShmulik Ladkani 559893515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5599df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5600b6a79208SShmulik Ladkani if (err) 560193515d53SJiri Pirko return err; 56029241e2dfSDaniel Borkmann 560393515d53SJiri Pirko skb->protocol = skb->vlan_proto; 560493515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 560593515d53SJiri Pirko 56066b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 560793515d53SJiri Pirko } 560893515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 560993515d53SJiri Pirko return 0; 561093515d53SJiri Pirko } 561193515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 561293515d53SJiri Pirko 561319fbcb36SGuillaume Nault /** 561419fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 561519fbcb36SGuillaume Nault * 561619fbcb36SGuillaume Nault * @skb: Socket buffer to modify 561719fbcb36SGuillaume Nault * 561819fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 561919fbcb36SGuillaume Nault * 562019fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 562119fbcb36SGuillaume Nault * present. 562219fbcb36SGuillaume Nault * 562319fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 562419fbcb36SGuillaume Nault */ 562519fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 562619fbcb36SGuillaume Nault { 562719fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 562819fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 562919fbcb36SGuillaume Nault return -EPROTO; 563019fbcb36SGuillaume Nault 563119fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 563219fbcb36SGuillaume Nault skb_reset_mac_header(skb); 563319fbcb36SGuillaume Nault skb_reset_mac_len(skb); 563419fbcb36SGuillaume Nault 563519fbcb36SGuillaume Nault return 0; 563619fbcb36SGuillaume Nault } 563719fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 563819fbcb36SGuillaume Nault 563919fbcb36SGuillaume Nault /** 564019fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 564119fbcb36SGuillaume Nault * 564219fbcb36SGuillaume Nault * @skb: Socket buffer to modify 564319fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 564419fbcb36SGuillaume Nault * @src: Source MAC address of the new header 564519fbcb36SGuillaume Nault * 564619fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 564719fbcb36SGuillaume Nault * 564819fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 564919fbcb36SGuillaume Nault * 565019fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 565119fbcb36SGuillaume Nault */ 565219fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 565319fbcb36SGuillaume Nault const unsigned char *src) 565419fbcb36SGuillaume Nault { 565519fbcb36SGuillaume Nault struct ethhdr *eth; 565619fbcb36SGuillaume Nault int err; 565719fbcb36SGuillaume Nault 565819fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 565919fbcb36SGuillaume Nault return -EPROTO; 566019fbcb36SGuillaume Nault 566119fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 566219fbcb36SGuillaume Nault if (err < 0) 566319fbcb36SGuillaume Nault return err; 566419fbcb36SGuillaume Nault 566519fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 566619fbcb36SGuillaume Nault skb_reset_mac_header(skb); 566719fbcb36SGuillaume Nault skb_reset_mac_len(skb); 566819fbcb36SGuillaume Nault 566919fbcb36SGuillaume Nault eth = eth_hdr(skb); 567019fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 567119fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 567219fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 567319fbcb36SGuillaume Nault 567419fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 567519fbcb36SGuillaume Nault 567619fbcb36SGuillaume Nault return 0; 567719fbcb36SGuillaume Nault } 567819fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 567919fbcb36SGuillaume Nault 56808822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 56818822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 56828822e270SJohn Hurley __be16 ethertype) 56838822e270SJohn Hurley { 56848822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 56858822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 56868822e270SJohn Hurley 56878822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 56888822e270SJohn Hurley } 56898822e270SJohn Hurley 56908822e270SJohn Hurley hdr->h_proto = ethertype; 56918822e270SJohn Hurley } 56928822e270SJohn Hurley 56938822e270SJohn Hurley /** 5694e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5695e7dbfed1SMartin Varghese * the packet 56968822e270SJohn Hurley * 56978822e270SJohn Hurley * @skb: buffer 56988822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 56998822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5700fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5701e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5702e7dbfed1SMartin Varghese * ethernet 57038822e270SJohn Hurley * 57048822e270SJohn Hurley * Expects skb->data at mac header. 57058822e270SJohn Hurley * 57068822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 57078822e270SJohn Hurley */ 5708fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5709d04ac224SMartin Varghese int mac_len, bool ethernet) 57108822e270SJohn Hurley { 57118822e270SJohn Hurley struct mpls_shim_hdr *lse; 57128822e270SJohn Hurley int err; 57138822e270SJohn Hurley 57148822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 57158822e270SJohn Hurley return -EINVAL; 57168822e270SJohn Hurley 57178822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 57188822e270SJohn Hurley if (skb->encapsulation) 57198822e270SJohn Hurley return -EINVAL; 57208822e270SJohn Hurley 57218822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 57228822e270SJohn Hurley if (unlikely(err)) 57238822e270SJohn Hurley return err; 57248822e270SJohn Hurley 57258822e270SJohn Hurley if (!skb->inner_protocol) { 5726e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 57278822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 57288822e270SJohn Hurley } 57298822e270SJohn Hurley 57308822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 57318822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5732fa4e0f88SDavide Caratti mac_len); 57338822e270SJohn Hurley skb_reset_mac_header(skb); 5734fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5735e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 57368822e270SJohn Hurley 57378822e270SJohn Hurley lse = mpls_hdr(skb); 57388822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 57398822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 57408822e270SJohn Hurley 57414296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 57428822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 57438822e270SJohn Hurley skb->protocol = mpls_proto; 57448822e270SJohn Hurley 57458822e270SJohn Hurley return 0; 57468822e270SJohn Hurley } 57478822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 57488822e270SJohn Hurley 57492e4e4410SEric Dumazet /** 5750ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5751ed246ceeSJohn Hurley * 5752ed246ceeSJohn Hurley * @skb: buffer 5753ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5754fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 575576f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5756ed246ceeSJohn Hurley * 5757ed246ceeSJohn Hurley * Expects skb->data at mac header. 5758ed246ceeSJohn Hurley * 5759ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5760ed246ceeSJohn Hurley */ 5761040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5762040b5cfbSMartin Varghese bool ethernet) 5763ed246ceeSJohn Hurley { 5764ed246ceeSJohn Hurley int err; 5765ed246ceeSJohn Hurley 5766ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5767dedc5a08SDavide Caratti return 0; 5768ed246ceeSJohn Hurley 5769fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5770ed246ceeSJohn Hurley if (unlikely(err)) 5771ed246ceeSJohn Hurley return err; 5772ed246ceeSJohn Hurley 5773ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5774ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5775fa4e0f88SDavide Caratti mac_len); 5776ed246ceeSJohn Hurley 5777ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5778ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5779fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5780ed246ceeSJohn Hurley 57814296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5782ed246ceeSJohn Hurley struct ethhdr *hdr; 5783ed246ceeSJohn Hurley 5784ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5785ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5786ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5787ed246ceeSJohn Hurley } 5788ed246ceeSJohn Hurley skb->protocol = next_proto; 5789ed246ceeSJohn Hurley 5790ed246ceeSJohn Hurley return 0; 5791ed246ceeSJohn Hurley } 5792ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5793ed246ceeSJohn Hurley 5794ed246ceeSJohn Hurley /** 5795d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5796d27cf5c5SJohn Hurley * 5797d27cf5c5SJohn Hurley * @skb: buffer 5798d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5799d27cf5c5SJohn Hurley * 5800d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5801d27cf5c5SJohn Hurley * 5802d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5803d27cf5c5SJohn Hurley */ 5804d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5805d27cf5c5SJohn Hurley { 5806d27cf5c5SJohn Hurley int err; 5807d27cf5c5SJohn Hurley 5808d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5809d27cf5c5SJohn Hurley return -EINVAL; 5810d27cf5c5SJohn Hurley 5811d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5812d27cf5c5SJohn Hurley if (unlikely(err)) 5813d27cf5c5SJohn Hurley return err; 5814d27cf5c5SJohn Hurley 5815d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5816d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5817d27cf5c5SJohn Hurley 5818d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5819d27cf5c5SJohn Hurley } 5820d27cf5c5SJohn Hurley 5821d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 5822d27cf5c5SJohn Hurley 5823d27cf5c5SJohn Hurley return 0; 5824d27cf5c5SJohn Hurley } 5825d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 5826d27cf5c5SJohn Hurley 5827d27cf5c5SJohn Hurley /** 58282a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 58292a2ea508SJohn Hurley * 58302a2ea508SJohn Hurley * @skb: buffer 58312a2ea508SJohn Hurley * 58322a2ea508SJohn Hurley * Expects skb->data at mac header. 58332a2ea508SJohn Hurley * 58342a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 58352a2ea508SJohn Hurley */ 58362a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 58372a2ea508SJohn Hurley { 58382a2ea508SJohn Hurley u32 lse; 58392a2ea508SJohn Hurley u8 ttl; 58402a2ea508SJohn Hurley 58412a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 58422a2ea508SJohn Hurley return -EINVAL; 58432a2ea508SJohn Hurley 584413de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 584513de4ed9SDavide Caratti return -ENOMEM; 584613de4ed9SDavide Caratti 58472a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 58482a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 58492a2ea508SJohn Hurley if (!--ttl) 58502a2ea508SJohn Hurley return -EINVAL; 58512a2ea508SJohn Hurley 58522a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 58532a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 58542a2ea508SJohn Hurley 58552a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 58562a2ea508SJohn Hurley } 58572a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 58582a2ea508SJohn Hurley 58592a2ea508SJohn Hurley /** 58602e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 58612e4e4410SEric Dumazet * 5862de3f0d0eSMasanari Iida * @header_len: size of linear part 5863de3f0d0eSMasanari Iida * @data_len: needed length in frags 5864de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5865de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5866de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 58672e4e4410SEric Dumazet * 58682e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 58692e4e4410SEric Dumazet */ 58702e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 58712e4e4410SEric Dumazet unsigned long data_len, 58722e4e4410SEric Dumazet int max_page_order, 58732e4e4410SEric Dumazet int *errcode, 58742e4e4410SEric Dumazet gfp_t gfp_mask) 58752e4e4410SEric Dumazet { 58762e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 58772e4e4410SEric Dumazet unsigned long chunk; 58782e4e4410SEric Dumazet struct sk_buff *skb; 58792e4e4410SEric Dumazet struct page *page; 58802e4e4410SEric Dumazet int i; 58812e4e4410SEric Dumazet 58822e4e4410SEric Dumazet *errcode = -EMSGSIZE; 58832e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 58842e4e4410SEric Dumazet * high order pages... 58852e4e4410SEric Dumazet */ 58862e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 58872e4e4410SEric Dumazet return NULL; 58882e4e4410SEric Dumazet 58892e4e4410SEric Dumazet *errcode = -ENOBUFS; 5890f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 58912e4e4410SEric Dumazet if (!skb) 58922e4e4410SEric Dumazet return NULL; 58932e4e4410SEric Dumazet 58942e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 58952e4e4410SEric Dumazet 58962e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 58972e4e4410SEric Dumazet int order = max_page_order; 58982e4e4410SEric Dumazet 58992e4e4410SEric Dumazet while (order) { 59002e4e4410SEric Dumazet if (npages >= 1 << order) { 5901d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 59022e4e4410SEric Dumazet __GFP_COMP | 5903d14b56f5SMichal Hocko __GFP_NOWARN, 59042e4e4410SEric Dumazet order); 59052e4e4410SEric Dumazet if (page) 59062e4e4410SEric Dumazet goto fill_page; 59072e4e4410SEric Dumazet /* Do not retry other high order allocations */ 59082e4e4410SEric Dumazet order = 1; 59092e4e4410SEric Dumazet max_page_order = 0; 59102e4e4410SEric Dumazet } 59112e4e4410SEric Dumazet order--; 59122e4e4410SEric Dumazet } 59132e4e4410SEric Dumazet page = alloc_page(gfp_mask); 59142e4e4410SEric Dumazet if (!page) 59152e4e4410SEric Dumazet goto failure; 59162e4e4410SEric Dumazet fill_page: 59172e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 59182e4e4410SEric Dumazet PAGE_SIZE << order); 59192e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 59202e4e4410SEric Dumazet data_len -= chunk; 59212e4e4410SEric Dumazet npages -= 1 << order; 59222e4e4410SEric Dumazet } 59232e4e4410SEric Dumazet return skb; 59242e4e4410SEric Dumazet 59252e4e4410SEric Dumazet failure: 59262e4e4410SEric Dumazet kfree_skb(skb); 59272e4e4410SEric Dumazet return NULL; 59282e4e4410SEric Dumazet } 59292e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 59306fa01ccdSSowmini Varadhan 59316fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 59326fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 59336fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 59346fa01ccdSSowmini Varadhan { 59356fa01ccdSSowmini Varadhan int i; 59366fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 59376fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 59386fa01ccdSSowmini Varadhan u8 *data; 59396fa01ccdSSowmini Varadhan 59406fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 59416fa01ccdSSowmini Varadhan 59426fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 59436fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 59446fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 59456fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 59466fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 59476fa01ccdSSowmini Varadhan if (!data) 59486fa01ccdSSowmini Varadhan return -ENOMEM; 59496fa01ccdSSowmini Varadhan 59506fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 59516fa01ccdSSowmini Varadhan 59526fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 59536fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 59546fa01ccdSSowmini Varadhan skb->len -= off; 59556fa01ccdSSowmini Varadhan 59566fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 59576fa01ccdSSowmini Varadhan skb_shinfo(skb), 59586fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 59596fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 59606fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 59616fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 59626fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 59636fa01ccdSSowmini Varadhan kfree(data); 59646fa01ccdSSowmini Varadhan return -ENOMEM; 59656fa01ccdSSowmini Varadhan } 59666fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 59676fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 59686fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 59696fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 59706fa01ccdSSowmini Varadhan skb_release_data(skb); 59716fa01ccdSSowmini Varadhan } else { 59726fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 59736fa01ccdSSowmini Varadhan * relocate values 59746fa01ccdSSowmini Varadhan */ 59756fa01ccdSSowmini Varadhan skb_free_head(skb); 59766fa01ccdSSowmini Varadhan } 59776fa01ccdSSowmini Varadhan 59786fa01ccdSSowmini Varadhan skb->head = data; 59796fa01ccdSSowmini Varadhan skb->data = data; 59806fa01ccdSSowmini Varadhan skb->head_frag = 0; 59816fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 59826fa01ccdSSowmini Varadhan skb->end = size; 59836fa01ccdSSowmini Varadhan #else 59846fa01ccdSSowmini Varadhan skb->end = skb->head + size; 59856fa01ccdSSowmini Varadhan #endif 59866fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 59876fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 59886fa01ccdSSowmini Varadhan skb->cloned = 0; 59896fa01ccdSSowmini Varadhan skb->hdr_len = 0; 59906fa01ccdSSowmini Varadhan skb->nohdr = 0; 59916fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 59926fa01ccdSSowmini Varadhan 59936fa01ccdSSowmini Varadhan return 0; 59946fa01ccdSSowmini Varadhan } 59956fa01ccdSSowmini Varadhan 59966fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 59976fa01ccdSSowmini Varadhan 59986fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 59996fa01ccdSSowmini Varadhan * pskb_carve() 60006fa01ccdSSowmini Varadhan */ 60016fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 60026fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 60036fa01ccdSSowmini Varadhan gfp_t gfp_mask) 60046fa01ccdSSowmini Varadhan { 60056fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 60066fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 60076fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 60086fa01ccdSSowmini Varadhan 60096fa01ccdSSowmini Varadhan do { 60106fa01ccdSSowmini Varadhan if (!list) { 60116fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 60126fa01ccdSSowmini Varadhan return -EFAULT; 60136fa01ccdSSowmini Varadhan } 60146fa01ccdSSowmini Varadhan if (list->len <= eat) { 60156fa01ccdSSowmini Varadhan /* Eaten as whole. */ 60166fa01ccdSSowmini Varadhan eat -= list->len; 60176fa01ccdSSowmini Varadhan list = list->next; 60186fa01ccdSSowmini Varadhan insp = list; 60196fa01ccdSSowmini Varadhan } else { 60206fa01ccdSSowmini Varadhan /* Eaten partially. */ 60216fa01ccdSSowmini Varadhan if (skb_shared(list)) { 60226fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 60236fa01ccdSSowmini Varadhan if (!clone) 60246fa01ccdSSowmini Varadhan return -ENOMEM; 60256fa01ccdSSowmini Varadhan insp = list->next; 60266fa01ccdSSowmini Varadhan list = clone; 60276fa01ccdSSowmini Varadhan } else { 60286fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 60296fa01ccdSSowmini Varadhan insp = list; 60306fa01ccdSSowmini Varadhan } 60316fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 60326fa01ccdSSowmini Varadhan kfree_skb(clone); 60336fa01ccdSSowmini Varadhan return -ENOMEM; 60346fa01ccdSSowmini Varadhan } 60356fa01ccdSSowmini Varadhan break; 60366fa01ccdSSowmini Varadhan } 60376fa01ccdSSowmini Varadhan } while (eat); 60386fa01ccdSSowmini Varadhan 60396fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 60406fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 60416fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 60426fa01ccdSSowmini Varadhan kfree_skb(list); 60436fa01ccdSSowmini Varadhan } 60446fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 60456fa01ccdSSowmini Varadhan if (clone) { 60466fa01ccdSSowmini Varadhan clone->next = list; 60476fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 60486fa01ccdSSowmini Varadhan } 60496fa01ccdSSowmini Varadhan return 0; 60506fa01ccdSSowmini Varadhan } 60516fa01ccdSSowmini Varadhan 60526fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 60536fa01ccdSSowmini Varadhan * non-linear part of skb 60546fa01ccdSSowmini Varadhan */ 60556fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 60566fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 60576fa01ccdSSowmini Varadhan { 60586fa01ccdSSowmini Varadhan int i, k = 0; 60596fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 60606fa01ccdSSowmini Varadhan u8 *data; 60616fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 60626fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 60636fa01ccdSSowmini Varadhan 60646fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 60656fa01ccdSSowmini Varadhan 60666fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 60676fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 60686fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 60696fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 60706fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 60716fa01ccdSSowmini Varadhan if (!data) 60726fa01ccdSSowmini Varadhan return -ENOMEM; 60736fa01ccdSSowmini Varadhan 60746fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 60756fa01ccdSSowmini Varadhan 60766fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6077e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 60786fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 60796fa01ccdSSowmini Varadhan kfree(data); 60806fa01ccdSSowmini Varadhan return -ENOMEM; 60816fa01ccdSSowmini Varadhan } 60826fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 60836fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 60846fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 60856fa01ccdSSowmini Varadhan 60866fa01ccdSSowmini Varadhan if (pos + fsize > off) { 60876fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 60886fa01ccdSSowmini Varadhan 60896fa01ccdSSowmini Varadhan if (pos < off) { 60906fa01ccdSSowmini Varadhan /* Split frag. 60916fa01ccdSSowmini Varadhan * We have two variants in this case: 60926fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 60936fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 60946fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 60956fa01ccdSSowmini Varadhan * where splitting is expensive. 60966fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 60976fa01ccdSSowmini Varadhan */ 6098b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 60996fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 61006fa01ccdSSowmini Varadhan } 61016fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 61026fa01ccdSSowmini Varadhan k++; 61036fa01ccdSSowmini Varadhan } 61046fa01ccdSSowmini Varadhan pos += fsize; 61056fa01ccdSSowmini Varadhan } 61066fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 61076fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 61086fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 61096fa01ccdSSowmini Varadhan 61106fa01ccdSSowmini Varadhan /* split line is in frag list */ 6111eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6112eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6113eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6114eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6115eabe8618SMiaohe Lin kfree(data); 6116eabe8618SMiaohe Lin return -ENOMEM; 61176fa01ccdSSowmini Varadhan } 61186fa01ccdSSowmini Varadhan skb_release_data(skb); 61196fa01ccdSSowmini Varadhan 61206fa01ccdSSowmini Varadhan skb->head = data; 61216fa01ccdSSowmini Varadhan skb->head_frag = 0; 61226fa01ccdSSowmini Varadhan skb->data = data; 61236fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 61246fa01ccdSSowmini Varadhan skb->end = size; 61256fa01ccdSSowmini Varadhan #else 61266fa01ccdSSowmini Varadhan skb->end = skb->head + size; 61276fa01ccdSSowmini Varadhan #endif 61286fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 61296fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 61306fa01ccdSSowmini Varadhan skb->cloned = 0; 61316fa01ccdSSowmini Varadhan skb->hdr_len = 0; 61326fa01ccdSSowmini Varadhan skb->nohdr = 0; 61336fa01ccdSSowmini Varadhan skb->len -= off; 61346fa01ccdSSowmini Varadhan skb->data_len = skb->len; 61356fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 61366fa01ccdSSowmini Varadhan return 0; 61376fa01ccdSSowmini Varadhan } 61386fa01ccdSSowmini Varadhan 61396fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 61406fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 61416fa01ccdSSowmini Varadhan { 61426fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 61436fa01ccdSSowmini Varadhan 61446fa01ccdSSowmini Varadhan if (len < headlen) 61456fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 61466fa01ccdSSowmini Varadhan else 61476fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 61486fa01ccdSSowmini Varadhan } 61496fa01ccdSSowmini Varadhan 61506fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 61516fa01ccdSSowmini Varadhan * a new skb 61526fa01ccdSSowmini Varadhan */ 61536fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 61546fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 61556fa01ccdSSowmini Varadhan { 61566fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 61576fa01ccdSSowmini Varadhan 61586fa01ccdSSowmini Varadhan if (!clone) 61596fa01ccdSSowmini Varadhan return NULL; 61606fa01ccdSSowmini Varadhan 61616fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 61626fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 61636fa01ccdSSowmini Varadhan kfree_skb(clone); 61646fa01ccdSSowmini Varadhan return NULL; 61656fa01ccdSSowmini Varadhan } 61666fa01ccdSSowmini Varadhan return clone; 61676fa01ccdSSowmini Varadhan } 61686fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6169c8c8b127SEric Dumazet 6170c8c8b127SEric Dumazet /** 6171c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6172c8c8b127SEric Dumazet * @skb: buffer 6173c8c8b127SEric Dumazet * 6174c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6175c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6176c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6177c8c8b127SEric Dumazet * truesize. 6178c8c8b127SEric Dumazet * Notes: 6179c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6180c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6181c8c8b127SEric Dumazet */ 6182c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6183c8c8b127SEric Dumazet { 61843174fed9SEric Dumazet if (skb->data_len) { 61853174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6186c8c8b127SEric Dumazet skb_cloned(skb)) 6187c8c8b127SEric Dumazet return; 6188c8c8b127SEric Dumazet 6189c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6190c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 61913174fed9SEric Dumazet } 61923174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 61933174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 61943174fed9SEric Dumazet * their truesize. 61953174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 61963174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 61973174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6198c8c8b127SEric Dumazet */ 6199c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6200c8c8b127SEric Dumazet } 6201df5042f4SFlorian Westphal 6202df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6203df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6204df5042f4SFlorian Westphal { 6205df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6206df5042f4SFlorian Westphal } 6207df5042f4SFlorian Westphal 62088b69a803SPaolo Abeni /** 62098b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 62108b69a803SPaolo Abeni * 62114930f483SFlorian Westphal * @flags: See kmalloc(). 62124930f483SFlorian Westphal * 62138b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 62148b69a803SPaolo Abeni * skb via __skb_ext_set(). 62158b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 62168b69a803SPaolo Abeni */ 62174930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6218df5042f4SFlorian Westphal { 62194930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6220df5042f4SFlorian Westphal 6221df5042f4SFlorian Westphal if (new) { 6222df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6223df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6224df5042f4SFlorian Westphal } 6225df5042f4SFlorian Westphal 6226df5042f4SFlorian Westphal return new; 6227df5042f4SFlorian Westphal } 6228df5042f4SFlorian Westphal 62294165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 62304165079bSFlorian Westphal unsigned int old_active) 6231df5042f4SFlorian Westphal { 6232df5042f4SFlorian Westphal struct skb_ext *new; 6233df5042f4SFlorian Westphal 6234df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6235df5042f4SFlorian Westphal return old; 6236df5042f4SFlorian Westphal 6237df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6238df5042f4SFlorian Westphal if (!new) 6239df5042f4SFlorian Westphal return NULL; 6240df5042f4SFlorian Westphal 6241df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6242df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6243df5042f4SFlorian Westphal 62444165079bSFlorian Westphal #ifdef CONFIG_XFRM 62454165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 62464165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 62474165079bSFlorian Westphal unsigned int i; 62484165079bSFlorian Westphal 62494165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 62504165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 62514165079bSFlorian Westphal } 62524165079bSFlorian Westphal #endif 6253df5042f4SFlorian Westphal __skb_ext_put(old); 6254df5042f4SFlorian Westphal return new; 6255df5042f4SFlorian Westphal } 6256df5042f4SFlorian Westphal 6257df5042f4SFlorian Westphal /** 62588b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 62598b69a803SPaolo Abeni * @skb: buffer 62608b69a803SPaolo Abeni * @id: extension id 62618b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 62628b69a803SPaolo Abeni * 62638b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 62648b69a803SPaolo Abeni * 62658b69a803SPaolo Abeni * Returns the pointer to the extension. 62668b69a803SPaolo Abeni */ 62678b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 62688b69a803SPaolo Abeni struct skb_ext *ext) 62698b69a803SPaolo Abeni { 62708b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 62718b69a803SPaolo Abeni 62728b69a803SPaolo Abeni skb_ext_put(skb); 62738b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 62748b69a803SPaolo Abeni ext->chunks = newlen; 62758b69a803SPaolo Abeni ext->offset[id] = newoff; 62768b69a803SPaolo Abeni skb->extensions = ext; 62778b69a803SPaolo Abeni skb->active_extensions = 1 << id; 62788b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 62798b69a803SPaolo Abeni } 62808b69a803SPaolo Abeni 62818b69a803SPaolo Abeni /** 6282df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6283df5042f4SFlorian Westphal * @skb: buffer 6284df5042f4SFlorian Westphal * @id: extension to allocate space for 6285df5042f4SFlorian Westphal * 6286df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6287df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6288df5042f4SFlorian Westphal * is returned. 6289df5042f4SFlorian Westphal * 6290df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6291df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6292df5042f4SFlorian Westphal * 6293df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6294df5042f4SFlorian Westphal */ 6295df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6296df5042f4SFlorian Westphal { 6297df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6298df5042f4SFlorian Westphal unsigned int newlen, newoff; 6299df5042f4SFlorian Westphal 6300df5042f4SFlorian Westphal if (skb->active_extensions) { 6301df5042f4SFlorian Westphal old = skb->extensions; 6302df5042f4SFlorian Westphal 63034165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6304df5042f4SFlorian Westphal if (!new) 6305df5042f4SFlorian Westphal return NULL; 6306df5042f4SFlorian Westphal 6307682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6308df5042f4SFlorian Westphal goto set_active; 6309df5042f4SFlorian Westphal 6310e94e50bdSPaolo Abeni newoff = new->chunks; 6311df5042f4SFlorian Westphal } else { 6312df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6313df5042f4SFlorian Westphal 63144930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6315df5042f4SFlorian Westphal if (!new) 6316df5042f4SFlorian Westphal return NULL; 6317df5042f4SFlorian Westphal } 6318df5042f4SFlorian Westphal 6319df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6320df5042f4SFlorian Westphal new->chunks = newlen; 6321df5042f4SFlorian Westphal new->offset[id] = newoff; 6322df5042f4SFlorian Westphal set_active: 6323682ec859SPaolo Abeni skb->extensions = new; 6324df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6325df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6326df5042f4SFlorian Westphal } 6327df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6328df5042f4SFlorian Westphal 63294165079bSFlorian Westphal #ifdef CONFIG_XFRM 63304165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 63314165079bSFlorian Westphal { 63324165079bSFlorian Westphal unsigned int i; 63334165079bSFlorian Westphal 63344165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 63354165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 63364165079bSFlorian Westphal } 63374165079bSFlorian Westphal #endif 63384165079bSFlorian Westphal 6339df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6340df5042f4SFlorian Westphal { 6341df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6342df5042f4SFlorian Westphal 6343df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6344df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6345df5042f4SFlorian Westphal skb->extensions = NULL; 6346df5042f4SFlorian Westphal __skb_ext_put(ext); 63474165079bSFlorian Westphal #ifdef CONFIG_XFRM 63484165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 63494165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 63504165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 63514165079bSFlorian Westphal 63524165079bSFlorian Westphal skb_ext_put_sp(sp); 63534165079bSFlorian Westphal sp->len = 0; 63544165079bSFlorian Westphal #endif 6355df5042f4SFlorian Westphal } 6356df5042f4SFlorian Westphal } 6357df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6358df5042f4SFlorian Westphal 6359df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6360df5042f4SFlorian Westphal { 6361df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6362df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6363df5042f4SFlorian Westphal */ 6364df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6365df5042f4SFlorian Westphal goto free_now; 6366df5042f4SFlorian Westphal 6367df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6368df5042f4SFlorian Westphal return; 6369df5042f4SFlorian Westphal free_now: 63704165079bSFlorian Westphal #ifdef CONFIG_XFRM 63714165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 63724165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 63734165079bSFlorian Westphal #endif 63744165079bSFlorian Westphal 6375df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6376df5042f4SFlorian Westphal } 6377df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6378df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 6379