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> 621da177e4SLinus Torvalds 631da177e4SLinus Torvalds #include <net/protocol.h> 641da177e4SLinus Torvalds #include <net/dst.h> 651da177e4SLinus Torvalds #include <net/sock.h> 661da177e4SLinus Torvalds #include <net/checksum.h> 67ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 681da177e4SLinus Torvalds #include <net/xfrm.h> 691da177e4SLinus Torvalds 707c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 71ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7251c56b00SEric Dumazet #include <linux/highmem.h> 73b245be1fSWillem de Bruijn #include <linux/capability.h> 74b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 75a1f8e7f7SAl Viro 767b7ed885SBart Van Assche #include "datagram.h" 777b7ed885SBart Van Assche 7808009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 7908009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 80df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 81df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 82df5042f4SFlorian Westphal #endif 835f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 845f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 851da177e4SLinus Torvalds 861da177e4SLinus Torvalds /** 87f05de73bSJean Sacren * skb_panic - private function for out-of-line support 881da177e4SLinus Torvalds * @skb: buffer 891da177e4SLinus Torvalds * @sz: size 90f05de73bSJean Sacren * @addr: address 9199d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 921da177e4SLinus Torvalds * 93f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 94f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 95f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 96f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 971da177e4SLinus Torvalds */ 98f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 9999d5851eSJames Hogan const char msg[]) 1001da177e4SLinus Torvalds { 101e005d193SJoe Perches pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 10299d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1034305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 10426095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1051da177e4SLinus Torvalds BUG(); 1061da177e4SLinus Torvalds } 1071da177e4SLinus Torvalds 108f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1091da177e4SLinus Torvalds { 110f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1111da177e4SLinus Torvalds } 1121da177e4SLinus Torvalds 113f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 114f05de73bSJean Sacren { 115f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 116f05de73bSJean Sacren } 117c93bdd0eSMel Gorman 118c93bdd0eSMel Gorman /* 119c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 120c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 121c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 122c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 123c93bdd0eSMel Gorman * memory is free 124c93bdd0eSMel Gorman */ 125c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 126c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 12761c5e88aSstephen hemminger 12861c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 12961c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 130c93bdd0eSMel Gorman { 131c93bdd0eSMel Gorman void *obj; 132c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 133c93bdd0eSMel Gorman 134c93bdd0eSMel Gorman /* 135c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 136c93bdd0eSMel Gorman * to the reserves, fail. 137c93bdd0eSMel Gorman */ 138c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 139c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 140c93bdd0eSMel Gorman node); 141c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 142c93bdd0eSMel Gorman goto out; 143c93bdd0eSMel Gorman 144c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 145c93bdd0eSMel Gorman ret_pfmemalloc = true; 146c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 147c93bdd0eSMel Gorman 148c93bdd0eSMel Gorman out: 149c93bdd0eSMel Gorman if (pfmemalloc) 150c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 151c93bdd0eSMel Gorman 152c93bdd0eSMel Gorman return obj; 153c93bdd0eSMel Gorman } 154c93bdd0eSMel Gorman 1551da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1561da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1571da177e4SLinus Torvalds * [BEEP] leaks. 1581da177e4SLinus Torvalds * 1591da177e4SLinus Torvalds */ 1601da177e4SLinus Torvalds 1611da177e4SLinus Torvalds /** 162d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1631da177e4SLinus Torvalds * @size: size to allocate 1641da177e4SLinus Torvalds * @gfp_mask: allocation mask 165c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 166c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 167c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 168c93bdd0eSMel Gorman * allocations in case the data is required for writeback 169b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1701da177e4SLinus Torvalds * 1711da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 17294b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 17394b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 1741da177e4SLinus Torvalds * 1751da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1761da177e4SLinus Torvalds * %GFP_ATOMIC. 1771da177e4SLinus Torvalds */ 178dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 179c93bdd0eSMel Gorman int flags, int node) 1801da177e4SLinus Torvalds { 181e18b890bSChristoph Lameter struct kmem_cache *cache; 1824947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1831da177e4SLinus Torvalds struct sk_buff *skb; 1841da177e4SLinus Torvalds u8 *data; 185c93bdd0eSMel Gorman bool pfmemalloc; 1861da177e4SLinus Torvalds 187c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 188c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 189c93bdd0eSMel Gorman 190c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 191c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1928798b3fbSHerbert Xu 1931da177e4SLinus Torvalds /* Get the HEAD */ 194b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1951da177e4SLinus Torvalds if (!skb) 1961da177e4SLinus Torvalds goto out; 197ec7d2f2cSEric Dumazet prefetchw(skb); 1981da177e4SLinus Torvalds 19987fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 20087fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 20187fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 20287fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 20387fb4b7bSEric Dumazet */ 204bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 20587fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 206c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2071da177e4SLinus Torvalds if (!data) 2081da177e4SLinus Torvalds goto nodata; 20987fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 21087fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 21187fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 21287fb4b7bSEric Dumazet */ 21387fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 214ec7d2f2cSEric Dumazet prefetchw(data + size); 2151da177e4SLinus Torvalds 216ca0605a7SArnaldo Carvalho de Melo /* 217c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 218c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 219c8005785SJohannes Berg * the tail pointer in struct sk_buff! 220ca0605a7SArnaldo Carvalho de Melo */ 221ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 22287fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 22387fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 224c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 22563354797SReshetova, Elena refcount_set(&skb->users, 1); 2261da177e4SLinus Torvalds skb->head = data; 2271da177e4SLinus Torvalds skb->data = data; 22827a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2294305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 23035d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 23135d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 23219633e12SStephen Hemminger 2334947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2344947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 235ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2364947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 2374947d3efSBenjamin LaHaise 238c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 239d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 2401da177e4SLinus Torvalds 241d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 242d0bf4a9eSEric Dumazet 243d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 2442638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 1); 245d179cd12SDavid S. Miller 2466ffe75ebSEric Dumazet fclones->skb2.fclone = SKB_FCLONE_CLONE; 247d179cd12SDavid S. Miller } 2481da177e4SLinus Torvalds out: 2491da177e4SLinus Torvalds return skb; 2501da177e4SLinus Torvalds nodata: 2518798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2521da177e4SLinus Torvalds skb = NULL; 2531da177e4SLinus Torvalds goto out; 2541da177e4SLinus Torvalds } 255b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2561da177e4SLinus Torvalds 257ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 258ba0509b6SJesper Dangaard Brouer static struct sk_buff *__build_skb_around(struct sk_buff *skb, 259ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 260ba0509b6SJesper Dangaard Brouer { 261ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 262ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 263ba0509b6SJesper Dangaard Brouer 264ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 265ba0509b6SJesper Dangaard Brouer 266ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 267ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 268ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 269ba0509b6SJesper Dangaard Brouer skb->head = data; 270ba0509b6SJesper Dangaard Brouer skb->data = data; 271ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 272ba0509b6SJesper Dangaard Brouer skb->end = skb->tail + size; 273ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 274ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 275ba0509b6SJesper Dangaard Brouer 276ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 277ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 278ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 279ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 280ba0509b6SJesper Dangaard Brouer 281ba0509b6SJesper Dangaard Brouer return skb; 282ba0509b6SJesper Dangaard Brouer } 283ba0509b6SJesper Dangaard Brouer 2841da177e4SLinus Torvalds /** 2852ea2f62cSEric Dumazet * __build_skb - build a network buffer 286b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2872ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 288b2b5ce9dSEric Dumazet * 289b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 290deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2912ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2922ea2f62cSEric Dumazet * or vmalloc() 293b2b5ce9dSEric Dumazet * The return is the new skb buffer. 294b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 295b2b5ce9dSEric Dumazet * Notes : 296b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 297b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 298b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 299b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 300b2b5ce9dSEric Dumazet * before giving packet to stack. 301b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 302b2b5ce9dSEric Dumazet */ 3032ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 304b2b5ce9dSEric Dumazet { 305b2b5ce9dSEric Dumazet struct sk_buff *skb; 306b2b5ce9dSEric Dumazet 307b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 308ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 309b2b5ce9dSEric Dumazet return NULL; 310b2b5ce9dSEric Dumazet 311b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 312b2b5ce9dSEric Dumazet 313ba0509b6SJesper Dangaard Brouer return __build_skb_around(skb, data, frag_size); 314b2b5ce9dSEric Dumazet } 3152ea2f62cSEric Dumazet 3162ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3172ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3182ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 3192ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 3202ea2f62cSEric Dumazet */ 3212ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3222ea2f62cSEric Dumazet { 3232ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3242ea2f62cSEric Dumazet 3252ea2f62cSEric Dumazet if (skb && frag_size) { 3262ea2f62cSEric Dumazet skb->head_frag = 1; 3272f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 3282ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3292ea2f62cSEric Dumazet } 3302ea2f62cSEric Dumazet return skb; 3312ea2f62cSEric Dumazet } 332b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 333b2b5ce9dSEric Dumazet 334ba0509b6SJesper Dangaard Brouer /** 335ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 336ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 337ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 338ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 339ba0509b6SJesper Dangaard Brouer */ 340ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 341ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 342ba0509b6SJesper Dangaard Brouer { 343ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 344ba0509b6SJesper Dangaard Brouer return NULL; 345ba0509b6SJesper Dangaard Brouer 346ba0509b6SJesper Dangaard Brouer skb = __build_skb_around(skb, data, frag_size); 347ba0509b6SJesper Dangaard Brouer 348ba0509b6SJesper Dangaard Brouer if (skb && frag_size) { 349ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 350ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 351ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 352ba0509b6SJesper Dangaard Brouer } 353ba0509b6SJesper Dangaard Brouer return skb; 354ba0509b6SJesper Dangaard Brouer } 355ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 356ba0509b6SJesper Dangaard Brouer 357795bb1c0SJesper Dangaard Brouer #define NAPI_SKB_CACHE_SIZE 64 358795bb1c0SJesper Dangaard Brouer 359795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache { 360795bb1c0SJesper Dangaard Brouer struct page_frag_cache page; 361e0d7924aSAlexey Dobriyan unsigned int skb_count; 362795bb1c0SJesper Dangaard Brouer void *skb_cache[NAPI_SKB_CACHE_SIZE]; 363795bb1c0SJesper Dangaard Brouer }; 364795bb1c0SJesper Dangaard Brouer 365b63ae8caSAlexander Duyck static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 366795bb1c0SJesper Dangaard Brouer static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 367ffde7328SAlexander Duyck 368ffde7328SAlexander Duyck static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 369ffde7328SAlexander Duyck { 370b63ae8caSAlexander Duyck struct page_frag_cache *nc; 371ffde7328SAlexander Duyck unsigned long flags; 372ffde7328SAlexander Duyck void *data; 373ffde7328SAlexander Duyck 374ffde7328SAlexander Duyck local_irq_save(flags); 3759451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 3768c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, fragsz, gfp_mask); 3776f532612SEric Dumazet local_irq_restore(flags); 3786f532612SEric Dumazet return data; 3796f532612SEric Dumazet } 380c93bdd0eSMel Gorman 381c93bdd0eSMel Gorman /** 382c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 383c93bdd0eSMel Gorman * @fragsz: fragment size 384c93bdd0eSMel Gorman * 385c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 386c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 387c93bdd0eSMel Gorman */ 388c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 389c93bdd0eSMel Gorman { 3903bed3cc4SAlexander Duyck fragsz = SKB_DATA_ALIGN(fragsz); 3913bed3cc4SAlexander Duyck 392453f85d4SMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC); 393c93bdd0eSMel Gorman } 3946f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 3956f532612SEric Dumazet 396ffde7328SAlexander Duyck static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 397ffde7328SAlexander Duyck { 398795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 3999451980aSAlexander Duyck 4008c2dd3e4SAlexander Duyck return page_frag_alloc(&nc->page, fragsz, gfp_mask); 401ffde7328SAlexander Duyck } 402ffde7328SAlexander Duyck 403ffde7328SAlexander Duyck void *napi_alloc_frag(unsigned int fragsz) 404ffde7328SAlexander Duyck { 4053bed3cc4SAlexander Duyck fragsz = SKB_DATA_ALIGN(fragsz); 4063bed3cc4SAlexander Duyck 407453f85d4SMel Gorman return __napi_alloc_frag(fragsz, GFP_ATOMIC); 408ffde7328SAlexander Duyck } 409ffde7328SAlexander Duyck EXPORT_SYMBOL(napi_alloc_frag); 410ffde7328SAlexander Duyck 4116f532612SEric Dumazet /** 412fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 413fd11a83dSAlexander Duyck * @dev: network device to receive on 414d7499160SMasanari Iida * @len: length to allocate 415fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 416fd11a83dSAlexander Duyck * 417fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 418fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 419fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 420fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 421fd11a83dSAlexander Duyck * 422fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 423fd11a83dSAlexander Duyck */ 4249451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4259451980aSAlexander Duyck gfp_t gfp_mask) 426fd11a83dSAlexander Duyck { 427b63ae8caSAlexander Duyck struct page_frag_cache *nc; 4289451980aSAlexander Duyck unsigned long flags; 429fd11a83dSAlexander Duyck struct sk_buff *skb; 4309451980aSAlexander Duyck bool pfmemalloc; 4319451980aSAlexander Duyck void *data; 432fd11a83dSAlexander Duyck 4339451980aSAlexander Duyck len += NET_SKB_PAD; 434fd11a83dSAlexander Duyck 4359451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 436d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 437a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 438a080e7bdSAlexander Duyck if (!skb) 439a080e7bdSAlexander Duyck goto skb_fail; 440a080e7bdSAlexander Duyck goto skb_success; 441a080e7bdSAlexander Duyck } 4429451980aSAlexander Duyck 4439451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4449451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 4459451980aSAlexander Duyck 4469451980aSAlexander Duyck if (sk_memalloc_socks()) 4479451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 4489451980aSAlexander Duyck 4499451980aSAlexander Duyck local_irq_save(flags); 4509451980aSAlexander Duyck 4519451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 4528c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 4539451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 4549451980aSAlexander Duyck 4559451980aSAlexander Duyck local_irq_restore(flags); 4569451980aSAlexander Duyck 4579451980aSAlexander Duyck if (unlikely(!data)) 4589451980aSAlexander Duyck return NULL; 4599451980aSAlexander Duyck 4609451980aSAlexander Duyck skb = __build_skb(data, len); 4619451980aSAlexander Duyck if (unlikely(!skb)) { 462181edb2bSAlexander Duyck skb_free_frag(data); 4639451980aSAlexander Duyck return NULL; 4649451980aSAlexander Duyck } 4659451980aSAlexander Duyck 4669451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 4679451980aSAlexander Duyck if (pfmemalloc) 4689451980aSAlexander Duyck skb->pfmemalloc = 1; 4699451980aSAlexander Duyck skb->head_frag = 1; 4709451980aSAlexander Duyck 471a080e7bdSAlexander Duyck skb_success: 4728af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4737b2e497aSChristoph Hellwig skb->dev = dev; 474fd11a83dSAlexander Duyck 475a080e7bdSAlexander Duyck skb_fail: 4768af27456SChristoph Hellwig return skb; 4778af27456SChristoph Hellwig } 478b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4791da177e4SLinus Torvalds 480fd11a83dSAlexander Duyck /** 481fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 482fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 483d7499160SMasanari Iida * @len: length to allocate 484fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 485fd11a83dSAlexander Duyck * 486fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 487fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 488fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 489fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 490fd11a83dSAlexander Duyck * 491fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 492fd11a83dSAlexander Duyck */ 4939451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 4949451980aSAlexander Duyck gfp_t gfp_mask) 495fd11a83dSAlexander Duyck { 496795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 497fd11a83dSAlexander Duyck struct sk_buff *skb; 4989451980aSAlexander Duyck void *data; 499fd11a83dSAlexander Duyck 5009451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 501fd11a83dSAlexander Duyck 5029451980aSAlexander Duyck if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 503d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 504a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 505a080e7bdSAlexander Duyck if (!skb) 506a080e7bdSAlexander Duyck goto skb_fail; 507a080e7bdSAlexander Duyck goto skb_success; 508a080e7bdSAlexander Duyck } 5099451980aSAlexander Duyck 5109451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5119451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5129451980aSAlexander Duyck 5139451980aSAlexander Duyck if (sk_memalloc_socks()) 5149451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5159451980aSAlexander Duyck 5168c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5179451980aSAlexander Duyck if (unlikely(!data)) 5189451980aSAlexander Duyck return NULL; 5199451980aSAlexander Duyck 5209451980aSAlexander Duyck skb = __build_skb(data, len); 5219451980aSAlexander Duyck if (unlikely(!skb)) { 522181edb2bSAlexander Duyck skb_free_frag(data); 5239451980aSAlexander Duyck return NULL; 5249451980aSAlexander Duyck } 5259451980aSAlexander Duyck 5269451980aSAlexander Duyck /* use OR instead of assignment to avoid clearing of bits in mask */ 527795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5289451980aSAlexander Duyck skb->pfmemalloc = 1; 5299451980aSAlexander Duyck skb->head_frag = 1; 5309451980aSAlexander Duyck 531a080e7bdSAlexander Duyck skb_success: 532fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 533fd11a83dSAlexander Duyck skb->dev = napi->dev; 534fd11a83dSAlexander Duyck 535a080e7bdSAlexander Duyck skb_fail: 536fd11a83dSAlexander Duyck return skb; 537fd11a83dSAlexander Duyck } 538fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 539fd11a83dSAlexander Duyck 540654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 54150269e19SEric Dumazet int size, unsigned int truesize) 542654bed16SPeter Zijlstra { 543654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 544654bed16SPeter Zijlstra skb->len += size; 545654bed16SPeter Zijlstra skb->data_len += size; 54650269e19SEric Dumazet skb->truesize += truesize; 547654bed16SPeter Zijlstra } 548654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 549654bed16SPeter Zijlstra 550f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 551f8e617e1SJason Wang unsigned int truesize) 552f8e617e1SJason Wang { 553f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 554f8e617e1SJason Wang 555f8e617e1SJason Wang skb_frag_size_add(frag, size); 556f8e617e1SJason Wang skb->len += size; 557f8e617e1SJason Wang skb->data_len += size; 558f8e617e1SJason Wang skb->truesize += truesize; 559f8e617e1SJason Wang } 560f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 561f8e617e1SJason Wang 56227b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 5631da177e4SLinus Torvalds { 564bd8a7036SEric Dumazet kfree_skb_list(*listp); 56527b437c8SHerbert Xu *listp = NULL; 5661da177e4SLinus Torvalds } 5671da177e4SLinus Torvalds 56827b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 56927b437c8SHerbert Xu { 57027b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 57127b437c8SHerbert Xu } 57227b437c8SHerbert Xu 5731da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5741da177e4SLinus Torvalds { 5751da177e4SLinus Torvalds struct sk_buff *list; 5761da177e4SLinus Torvalds 577fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5781da177e4SLinus Torvalds skb_get(list); 5791da177e4SLinus Torvalds } 5801da177e4SLinus Torvalds 581d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 582d3836f21SEric Dumazet { 583181edb2bSAlexander Duyck unsigned char *head = skb->head; 584181edb2bSAlexander Duyck 585d3836f21SEric Dumazet if (skb->head_frag) 586181edb2bSAlexander Duyck skb_free_frag(head); 587d3836f21SEric Dumazet else 588181edb2bSAlexander Duyck kfree(head); 589d3836f21SEric Dumazet } 590d3836f21SEric Dumazet 5915bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5921da177e4SLinus Torvalds { 593ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 5941da177e4SLinus Torvalds int i; 595ff04a771SEric Dumazet 596ff04a771SEric Dumazet if (skb->cloned && 597ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 598ff04a771SEric Dumazet &shinfo->dataref)) 599ff04a771SEric Dumazet return; 600ff04a771SEric Dumazet 601ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 602ff04a771SEric Dumazet __skb_frag_unref(&shinfo->frags[i]); 6031da177e4SLinus Torvalds 604ff04a771SEric Dumazet if (shinfo->frag_list) 605ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6061da177e4SLinus Torvalds 6071f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 608d3836f21SEric Dumazet skb_free_head(skb); 6091da177e4SLinus Torvalds } 6101da177e4SLinus Torvalds 6111da177e4SLinus Torvalds /* 6121da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6131da177e4SLinus Torvalds */ 6142d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6151da177e4SLinus Torvalds { 616d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 617d179cd12SDavid S. Miller 618d179cd12SDavid S. Miller switch (skb->fclone) { 619d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6201da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6216ffe75ebSEric Dumazet return; 622d179cd12SDavid S. Miller 623d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 624d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 6256ffe75ebSEric Dumazet 6266ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 6276ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 6286ffe75ebSEric Dumazet * while here, branch prediction will be good. 6296ffe75ebSEric Dumazet */ 6302638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 6316ffe75ebSEric Dumazet goto fastpath; 632d179cd12SDavid S. Miller break; 633d179cd12SDavid S. Miller 6346ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 635d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 636d179cd12SDavid S. Miller break; 6373ff50b79SStephen Hemminger } 6382638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 6396ffe75ebSEric Dumazet return; 6406ffe75ebSEric Dumazet fastpath: 6416ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 6421da177e4SLinus Torvalds } 6431da177e4SLinus Torvalds 6440a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 6451da177e4SLinus Torvalds { 646adf30907SEric Dumazet skb_dst_drop(skb); 6471da177e4SLinus Torvalds if (skb->destructor) { 6489c2b3328SStephen Hemminger WARN_ON(in_irq()); 6491da177e4SLinus Torvalds skb->destructor(skb); 6501da177e4SLinus Torvalds } 651a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 652cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 6532fc72c7bSKOVACS Krisztian #endif 654df5042f4SFlorian Westphal skb_ext_put(skb); 65504a4bb55SLennert Buytenhek } 65604a4bb55SLennert Buytenhek 65704a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 65804a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 65904a4bb55SLennert Buytenhek { 66004a4bb55SLennert Buytenhek skb_release_head_state(skb); 661a28b1b90SFlorian Westphal if (likely(skb->head)) 6622d4baff8SHerbert Xu skb_release_data(skb); 6632d4baff8SHerbert Xu } 6641da177e4SLinus Torvalds 6652d4baff8SHerbert Xu /** 6662d4baff8SHerbert Xu * __kfree_skb - private function 6672d4baff8SHerbert Xu * @skb: buffer 6682d4baff8SHerbert Xu * 6692d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6702d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6712d4baff8SHerbert Xu * always call kfree_skb 6722d4baff8SHerbert Xu */ 6732d4baff8SHerbert Xu 6742d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6752d4baff8SHerbert Xu { 6762d4baff8SHerbert Xu skb_release_all(skb); 6771da177e4SLinus Torvalds kfree_skbmem(skb); 6781da177e4SLinus Torvalds } 679b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6801da177e4SLinus Torvalds 6811da177e4SLinus Torvalds /** 682231d06aeSJörn Engel * kfree_skb - free an sk_buff 683231d06aeSJörn Engel * @skb: buffer to free 684231d06aeSJörn Engel * 685231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 686231d06aeSJörn Engel * hit zero. 687231d06aeSJörn Engel */ 688231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 689231d06aeSJörn Engel { 6903889a803SPaolo Abeni if (!skb_unref(skb)) 691231d06aeSJörn Engel return; 6923889a803SPaolo Abeni 693ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 694231d06aeSJörn Engel __kfree_skb(skb); 695231d06aeSJörn Engel } 696b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 697231d06aeSJörn Engel 698bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 699bd8a7036SEric Dumazet { 700bd8a7036SEric Dumazet while (segs) { 701bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 702bd8a7036SEric Dumazet 703bd8a7036SEric Dumazet kfree_skb(segs); 704bd8a7036SEric Dumazet segs = next; 705bd8a7036SEric Dumazet } 706bd8a7036SEric Dumazet } 707bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 708bd8a7036SEric Dumazet 709d1a203eaSStephen Hemminger /** 71025121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 71125121173SMichael S. Tsirkin * @skb: buffer that triggered an error 71225121173SMichael S. Tsirkin * 71325121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 71425121173SMichael S. Tsirkin * skb must be freed afterwards. 71525121173SMichael S. Tsirkin */ 71625121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 71725121173SMichael S. Tsirkin { 7181f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 71925121173SMichael S. Tsirkin } 72025121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 72125121173SMichael S. Tsirkin 72225121173SMichael S. Tsirkin /** 723ead2ceb0SNeil Horman * consume_skb - free an skbuff 724ead2ceb0SNeil Horman * @skb: buffer to free 725ead2ceb0SNeil Horman * 726ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 727ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 728ead2ceb0SNeil Horman * is being dropped after a failure and notes that 729ead2ceb0SNeil Horman */ 730ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 731ead2ceb0SNeil Horman { 7323889a803SPaolo Abeni if (!skb_unref(skb)) 733ead2ceb0SNeil Horman return; 7343889a803SPaolo Abeni 73507dc22e7SKoki Sanagi trace_consume_skb(skb); 736ead2ceb0SNeil Horman __kfree_skb(skb); 737ead2ceb0SNeil Horman } 738ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 739ead2ceb0SNeil Horman 7400a463c78SPaolo Abeni /** 7410a463c78SPaolo Abeni * consume_stateless_skb - free an skbuff, assuming it is stateless 7420a463c78SPaolo Abeni * @skb: buffer to free 7430a463c78SPaolo Abeni * 744ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 745ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 7460a463c78SPaolo Abeni */ 747ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 7480a463c78SPaolo Abeni { 7490a463c78SPaolo Abeni trace_consume_skb(skb); 7500a463c78SPaolo Abeni skb_release_data(skb); 7510a463c78SPaolo Abeni kfree_skbmem(skb); 7520a463c78SPaolo Abeni } 7530a463c78SPaolo Abeni 754795bb1c0SJesper Dangaard Brouer void __kfree_skb_flush(void) 755795bb1c0SJesper Dangaard Brouer { 756795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 757795bb1c0SJesper Dangaard Brouer 758795bb1c0SJesper Dangaard Brouer /* flush skb_cache if containing objects */ 759795bb1c0SJesper Dangaard Brouer if (nc->skb_count) { 760795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 761795bb1c0SJesper Dangaard Brouer nc->skb_cache); 762795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 763795bb1c0SJesper Dangaard Brouer } 764795bb1c0SJesper Dangaard Brouer } 765795bb1c0SJesper Dangaard Brouer 76615fad714SJesper Dangaard Brouer static inline void _kfree_skb_defer(struct sk_buff *skb) 767795bb1c0SJesper Dangaard Brouer { 768795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 769795bb1c0SJesper Dangaard Brouer 770795bb1c0SJesper Dangaard Brouer /* drop skb->head and call any destructors for packet */ 771795bb1c0SJesper Dangaard Brouer skb_release_all(skb); 772795bb1c0SJesper Dangaard Brouer 773795bb1c0SJesper Dangaard Brouer /* record skb to CPU local list */ 774795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 775795bb1c0SJesper Dangaard Brouer 776795bb1c0SJesper Dangaard Brouer #ifdef CONFIG_SLUB 777795bb1c0SJesper Dangaard Brouer /* SLUB writes into objects when freeing */ 778795bb1c0SJesper Dangaard Brouer prefetchw(skb); 779795bb1c0SJesper Dangaard Brouer #endif 780795bb1c0SJesper Dangaard Brouer 781795bb1c0SJesper Dangaard Brouer /* flush skb_cache if it is filled */ 782795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 783795bb1c0SJesper Dangaard Brouer kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 784795bb1c0SJesper Dangaard Brouer nc->skb_cache); 785795bb1c0SJesper Dangaard Brouer nc->skb_count = 0; 786795bb1c0SJesper Dangaard Brouer } 787795bb1c0SJesper Dangaard Brouer } 78815fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 78915fad714SJesper Dangaard Brouer { 79015fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 79115fad714SJesper Dangaard Brouer } 792795bb1c0SJesper Dangaard Brouer 793795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 794795bb1c0SJesper Dangaard Brouer { 795795bb1c0SJesper Dangaard Brouer if (unlikely(!skb)) 796795bb1c0SJesper Dangaard Brouer return; 797795bb1c0SJesper Dangaard Brouer 798885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 799795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 800885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 801795bb1c0SJesper Dangaard Brouer return; 802795bb1c0SJesper Dangaard Brouer } 803795bb1c0SJesper Dangaard Brouer 8047608894eSPaolo Abeni if (!skb_unref(skb)) 805795bb1c0SJesper Dangaard Brouer return; 8067608894eSPaolo Abeni 807795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 808795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 809795bb1c0SJesper Dangaard Brouer 810795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 811abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 812795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 813795bb1c0SJesper Dangaard Brouer return; 814795bb1c0SJesper Dangaard Brouer } 815795bb1c0SJesper Dangaard Brouer 81615fad714SJesper Dangaard Brouer _kfree_skb_defer(skb); 817795bb1c0SJesper Dangaard Brouer } 818795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 819795bb1c0SJesper Dangaard Brouer 820b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 821b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 822b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 823b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 824b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 825b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 826b1937227SEric Dumazet 827dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 828dec18810SHerbert Xu { 829dec18810SHerbert Xu new->tstamp = old->tstamp; 830b1937227SEric Dumazet /* We do not copy old->sk */ 831dec18810SHerbert Xu new->dev = old->dev; 832b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 8337fee226aSEric Dumazet skb_dst_copy(new, old); 834df5042f4SFlorian Westphal __skb_ext_copy(new, old); 835b1937227SEric Dumazet __nf_copy(new, old, false); 8366aa895b0SPatrick McHardy 837b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 838b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 839b1937227SEric Dumazet */ 840b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 84106021292SEliezer Tamir 842b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 843b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 844b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 845b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 846b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 847b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 848b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 849b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 850b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 851b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 852b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 853b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 854b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 855b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 856b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 857b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 858b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 859b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 860b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 861b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 862b1937227SEric Dumazet #endif 863e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 864b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 86506021292SEliezer Tamir #endif 8662bd82484SEric Dumazet #ifdef CONFIG_XPS 8672bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 8682bd82484SEric Dumazet #endif 869b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 870b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 871b1937227SEric Dumazet #endif 872b1937227SEric Dumazet 873dec18810SHerbert Xu } 874dec18810SHerbert Xu 87582c49a35SHerbert Xu /* 87682c49a35SHerbert Xu * You should not add any new code to this function. Add it to 87782c49a35SHerbert Xu * __copy_skb_header above instead. 87882c49a35SHerbert Xu */ 879e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 8801da177e4SLinus Torvalds { 8811da177e4SLinus Torvalds #define C(x) n->x = skb->x 8821da177e4SLinus Torvalds 8831da177e4SLinus Torvalds n->next = n->prev = NULL; 8841da177e4SLinus Torvalds n->sk = NULL; 885dec18810SHerbert Xu __copy_skb_header(n, skb); 886dec18810SHerbert Xu 8871da177e4SLinus Torvalds C(len); 8881da177e4SLinus Torvalds C(data_len); 8893e6b3b2eSAlexey Dobriyan C(mac_len); 890334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 89102f1c89dSPaul Moore n->cloned = 1; 8921da177e4SLinus Torvalds n->nohdr = 0; 893b13dda9fSEric Dumazet n->peeked = 0; 894e78bfb07SStefano Brivio C(pfmemalloc); 8951da177e4SLinus Torvalds n->destructor = NULL; 8961da177e4SLinus Torvalds C(tail); 8971da177e4SLinus Torvalds C(end); 89802f1c89dSPaul Moore C(head); 899d3836f21SEric Dumazet C(head_frag); 90002f1c89dSPaul Moore C(data); 90102f1c89dSPaul Moore C(truesize); 90263354797SReshetova, Elena refcount_set(&n->users, 1); 9031da177e4SLinus Torvalds 9041da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 9051da177e4SLinus Torvalds skb->cloned = 1; 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds return n; 908e0053ec0SHerbert Xu #undef C 909e0053ec0SHerbert Xu } 910e0053ec0SHerbert Xu 911e0053ec0SHerbert Xu /** 912e0053ec0SHerbert Xu * skb_morph - morph one skb into another 913e0053ec0SHerbert Xu * @dst: the skb to receive the contents 914e0053ec0SHerbert Xu * @src: the skb to supply the contents 915e0053ec0SHerbert Xu * 916e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 917e0053ec0SHerbert Xu * supplied by the user. 918e0053ec0SHerbert Xu * 919e0053ec0SHerbert Xu * The target skb is returned upon exit. 920e0053ec0SHerbert Xu */ 921e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 922e0053ec0SHerbert Xu { 9232d4baff8SHerbert Xu skb_release_all(dst); 924e0053ec0SHerbert Xu return __skb_clone(dst, src); 925e0053ec0SHerbert Xu } 926e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 927e0053ec0SHerbert Xu 9286f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 929a91dbff5SWillem de Bruijn { 930a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 931a91dbff5SWillem de Bruijn struct user_struct *user; 932a91dbff5SWillem de Bruijn 933a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 934a91dbff5SWillem de Bruijn return 0; 935a91dbff5SWillem de Bruijn 936a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 937a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 938a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 939a91dbff5SWillem de Bruijn 940a91dbff5SWillem de Bruijn do { 941a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 942a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 943a91dbff5SWillem de Bruijn if (new_pg > max_pg) 944a91dbff5SWillem de Bruijn return -ENOBUFS; 945a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 946a91dbff5SWillem de Bruijn old_pg); 947a91dbff5SWillem de Bruijn 948a91dbff5SWillem de Bruijn if (!mmp->user) { 949a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 950a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 951a91dbff5SWillem de Bruijn } else { 952a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 953a91dbff5SWillem de Bruijn } 954a91dbff5SWillem de Bruijn 955a91dbff5SWillem de Bruijn return 0; 956a91dbff5SWillem de Bruijn } 9576f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 958a91dbff5SWillem de Bruijn 9596f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 960a91dbff5SWillem de Bruijn { 961a91dbff5SWillem de Bruijn if (mmp->user) { 962a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 963a91dbff5SWillem de Bruijn free_uid(mmp->user); 964a91dbff5SWillem de Bruijn } 965a91dbff5SWillem de Bruijn } 9666f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 967a91dbff5SWillem de Bruijn 96852267790SWillem de Bruijn struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size) 96952267790SWillem de Bruijn { 97052267790SWillem de Bruijn struct ubuf_info *uarg; 97152267790SWillem de Bruijn struct sk_buff *skb; 97252267790SWillem de Bruijn 97352267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 97452267790SWillem de Bruijn 97552267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 97652267790SWillem de Bruijn if (!skb) 97752267790SWillem de Bruijn return NULL; 97852267790SWillem de Bruijn 97952267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 98052267790SWillem de Bruijn uarg = (void *)skb->cb; 981a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 982a91dbff5SWillem de Bruijn 983a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 984a91dbff5SWillem de Bruijn kfree_skb(skb); 985a91dbff5SWillem de Bruijn return NULL; 986a91dbff5SWillem de Bruijn } 98752267790SWillem de Bruijn 98852267790SWillem de Bruijn uarg->callback = sock_zerocopy_callback; 9894ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 9904ab6c99dSWillem de Bruijn uarg->len = 1; 9914ab6c99dSWillem de Bruijn uarg->bytelen = size; 99252267790SWillem de Bruijn uarg->zerocopy = 1; 993c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 99452267790SWillem de Bruijn sock_hold(sk); 99552267790SWillem de Bruijn 99652267790SWillem de Bruijn return uarg; 99752267790SWillem de Bruijn } 99852267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_alloc); 99952267790SWillem de Bruijn 100052267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 100152267790SWillem de Bruijn { 100252267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 100352267790SWillem de Bruijn } 100452267790SWillem de Bruijn 10054ab6c99dSWillem de Bruijn struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 10064ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 10074ab6c99dSWillem de Bruijn { 10084ab6c99dSWillem de Bruijn if (uarg) { 10094ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 10104ab6c99dSWillem de Bruijn u32 bytelen, next; 10114ab6c99dSWillem de Bruijn 10124ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 10134ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 10144ab6c99dSWillem de Bruijn */ 10154ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 10164ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 10174ab6c99dSWillem de Bruijn return NULL; 10184ab6c99dSWillem de Bruijn } 10194ab6c99dSWillem de Bruijn 10204ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 10214ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 10224ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 10234ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 10244ab6c99dSWillem de Bruijn goto new_alloc; 10254ab6c99dSWillem de Bruijn return NULL; 10264ab6c99dSWillem de Bruijn } 10274ab6c99dSWillem de Bruijn 10284ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 10294ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1030a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1031a91dbff5SWillem de Bruijn return NULL; 10324ab6c99dSWillem de Bruijn uarg->len++; 10334ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 10344ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1035db5bce32SEric Dumazet sock_zerocopy_get(uarg); 10364ab6c99dSWillem de Bruijn return uarg; 10374ab6c99dSWillem de Bruijn } 10384ab6c99dSWillem de Bruijn } 10394ab6c99dSWillem de Bruijn 10404ab6c99dSWillem de Bruijn new_alloc: 10414ab6c99dSWillem de Bruijn return sock_zerocopy_alloc(sk, size); 10424ab6c99dSWillem de Bruijn } 10434ab6c99dSWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_realloc); 10444ab6c99dSWillem de Bruijn 10454ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 10464ab6c99dSWillem de Bruijn { 10474ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 10484ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 10494ab6c99dSWillem de Bruijn u64 sum_len; 10504ab6c99dSWillem de Bruijn 10514ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 10524ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 10534ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 10544ab6c99dSWillem de Bruijn 10554ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 10564ab6c99dSWillem de Bruijn return false; 10574ab6c99dSWillem de Bruijn 10584ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 10594ab6c99dSWillem de Bruijn return false; 10604ab6c99dSWillem de Bruijn 10614ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 10624ab6c99dSWillem de Bruijn return true; 10634ab6c99dSWillem de Bruijn } 10644ab6c99dSWillem de Bruijn 106552267790SWillem de Bruijn void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) 106652267790SWillem de Bruijn { 10674ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 106852267790SWillem de Bruijn struct sock_exterr_skb *serr; 106952267790SWillem de Bruijn struct sock *sk = skb->sk; 10704ab6c99dSWillem de Bruijn struct sk_buff_head *q; 10714ab6c99dSWillem de Bruijn unsigned long flags; 10724ab6c99dSWillem de Bruijn u32 lo, hi; 10734ab6c99dSWillem de Bruijn u16 len; 107452267790SWillem de Bruijn 1075ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1076ccaffff1SWillem de Bruijn 10774ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 10784ab6c99dSWillem de Bruijn * so do not queue a completion notification 10794ab6c99dSWillem de Bruijn */ 10804ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 108152267790SWillem de Bruijn goto release; 108252267790SWillem de Bruijn 10834ab6c99dSWillem de Bruijn len = uarg->len; 10844ab6c99dSWillem de Bruijn lo = uarg->id; 10854ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 10864ab6c99dSWillem de Bruijn 108752267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 108852267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 108952267790SWillem de Bruijn serr->ee.ee_errno = 0; 109052267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 10914ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 10924ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 109352267790SWillem de Bruijn if (!success) 109452267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 109552267790SWillem de Bruijn 10964ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 10974ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 10984ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 10994ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 11004ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 11014ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 110252267790SWillem de Bruijn skb = NULL; 11034ab6c99dSWillem de Bruijn } 11044ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 110552267790SWillem de Bruijn 110652267790SWillem de Bruijn sk->sk_error_report(sk); 110752267790SWillem de Bruijn 110852267790SWillem de Bruijn release: 110952267790SWillem de Bruijn consume_skb(skb); 111052267790SWillem de Bruijn sock_put(sk); 111152267790SWillem de Bruijn } 111252267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_callback); 111352267790SWillem de Bruijn 111452267790SWillem de Bruijn void sock_zerocopy_put(struct ubuf_info *uarg) 111552267790SWillem de Bruijn { 1116c1d1b437SEric Dumazet if (uarg && refcount_dec_and_test(&uarg->refcnt)) { 111752267790SWillem de Bruijn if (uarg->callback) 111852267790SWillem de Bruijn uarg->callback(uarg, uarg->zerocopy); 111952267790SWillem de Bruijn else 112052267790SWillem de Bruijn consume_skb(skb_from_uarg(uarg)); 112152267790SWillem de Bruijn } 112252267790SWillem de Bruijn } 112352267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put); 112452267790SWillem de Bruijn 112552900d22SWillem de Bruijn void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 112652267790SWillem de Bruijn { 112752267790SWillem de Bruijn if (uarg) { 112852267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 112952267790SWillem de Bruijn 113052267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 11314ab6c99dSWillem de Bruijn uarg->len--; 113252267790SWillem de Bruijn 113352900d22SWillem de Bruijn if (have_uref) 113452267790SWillem de Bruijn sock_zerocopy_put(uarg); 113552267790SWillem de Bruijn } 113652267790SWillem de Bruijn } 113752267790SWillem de Bruijn EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); 113852267790SWillem de Bruijn 1139b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1140b5947e5dSWillem de Bruijn { 1141b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1142b5947e5dSWillem de Bruijn } 1143b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1144b5947e5dSWillem de Bruijn 114552267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 114652267790SWillem de Bruijn struct msghdr *msg, int len, 114752267790SWillem de Bruijn struct ubuf_info *uarg) 114852267790SWillem de Bruijn { 11494ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 115052267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 115152267790SWillem de Bruijn int err, orig_len = skb->len; 115252267790SWillem de Bruijn 11534ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 11544ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 11554ab6c99dSWillem de Bruijn */ 11564ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 11574ab6c99dSWillem de Bruijn return -EEXIST; 11584ab6c99dSWillem de Bruijn 115952267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 116052267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 116154d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 116254d43117SWillem de Bruijn 116352267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 116452267790SWillem de Bruijn msg->msg_iter = orig_iter; 116554d43117SWillem de Bruijn skb->sk = sk; 116652267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 116754d43117SWillem de Bruijn skb->sk = save_sk; 116852267790SWillem de Bruijn return err; 116952267790SWillem de Bruijn } 117052267790SWillem de Bruijn 117152900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 117252267790SWillem de Bruijn return skb->len - orig_len; 117352267790SWillem de Bruijn } 117452267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 117552267790SWillem de Bruijn 11761f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 117752267790SWillem de Bruijn gfp_t gfp_mask) 117852267790SWillem de Bruijn { 117952267790SWillem de Bruijn if (skb_zcopy(orig)) { 118052267790SWillem de Bruijn if (skb_zcopy(nskb)) { 118152267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 118252267790SWillem de Bruijn if (!gfp_mask) { 118352267790SWillem de Bruijn WARN_ON_ONCE(1); 118452267790SWillem de Bruijn return -ENOMEM; 118552267790SWillem de Bruijn } 118652267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 118752267790SWillem de Bruijn return 0; 118852267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 118952267790SWillem de Bruijn return -EIO; 119052267790SWillem de Bruijn } 119152900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 119252267790SWillem de Bruijn } 119352267790SWillem de Bruijn return 0; 119452267790SWillem de Bruijn } 119552267790SWillem de Bruijn 11962c53040fSBen Hutchings /** 11972c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 119848c83012SMichael S. Tsirkin * @skb: the skb to modify 119948c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 120048c83012SMichael S. Tsirkin * 120148c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 120248c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 120348c83012SMichael S. Tsirkin * to userspace pages. 120448c83012SMichael S. Tsirkin * 120548c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 120648c83012SMichael S. Tsirkin * %GFP_ATOMIC. 120748c83012SMichael S. Tsirkin * 120848c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 120948c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 121048c83012SMichael S. Tsirkin */ 121148c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1212a6686f2fSShirley Ma { 1213a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1214a6686f2fSShirley Ma struct page *page, *head = NULL; 12153ece7826SWillem de Bruijn int i, new_frags; 12163ece7826SWillem de Bruijn u32 d_off; 1217a6686f2fSShirley Ma 12183ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 12193ece7826SWillem de Bruijn return -EINVAL; 12203ece7826SWillem de Bruijn 1221f72c4ac6SWillem de Bruijn if (!num_frags) 1222f72c4ac6SWillem de Bruijn goto release; 1223f72c4ac6SWillem de Bruijn 12243ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 12253ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 122602756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1227a6686f2fSShirley Ma if (!page) { 1228a6686f2fSShirley Ma while (head) { 122940dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1230a6686f2fSShirley Ma put_page(head); 1231a6686f2fSShirley Ma head = next; 1232a6686f2fSShirley Ma } 1233a6686f2fSShirley Ma return -ENOMEM; 1234a6686f2fSShirley Ma } 12353ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 12363ece7826SWillem de Bruijn head = page; 12373ece7826SWillem de Bruijn } 12383ece7826SWillem de Bruijn 12393ece7826SWillem de Bruijn page = head; 12403ece7826SWillem de Bruijn d_off = 0; 12413ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 12423ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 12433ece7826SWillem de Bruijn u32 p_off, p_len, copied; 12443ece7826SWillem de Bruijn struct page *p; 12453ece7826SWillem de Bruijn u8 *vaddr; 1246c613c209SWillem de Bruijn 1247c613c209SWillem de Bruijn skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f), 1248c613c209SWillem de Bruijn p, p_off, p_len, copied) { 12493ece7826SWillem de Bruijn u32 copy, done = 0; 1250c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 12513ece7826SWillem de Bruijn 12523ece7826SWillem de Bruijn while (done < p_len) { 12533ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 12543ece7826SWillem de Bruijn d_off = 0; 12553ece7826SWillem de Bruijn page = (struct page *)page_private(page); 12563ece7826SWillem de Bruijn } 12573ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 12583ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 12593ece7826SWillem de Bruijn vaddr + p_off + done, copy); 12603ece7826SWillem de Bruijn done += copy; 12613ece7826SWillem de Bruijn d_off += copy; 12623ece7826SWillem de Bruijn } 126351c56b00SEric Dumazet kunmap_atomic(vaddr); 1264c613c209SWillem de Bruijn } 1265a6686f2fSShirley Ma } 1266a6686f2fSShirley Ma 1267a6686f2fSShirley Ma /* skb frags release userspace buffers */ 126802756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1269a8605c60SIan Campbell skb_frag_unref(skb, i); 1270a6686f2fSShirley Ma 1271a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 12723ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 12733ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 127440dadff2SSunghan Suh head = (struct page *)page_private(head); 1275a6686f2fSShirley Ma } 12763ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 12773ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 127848c83012SMichael S. Tsirkin 1279b90ddd56SWillem de Bruijn release: 12801f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1281a6686f2fSShirley Ma return 0; 1282a6686f2fSShirley Ma } 1283dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1284a6686f2fSShirley Ma 1285e0053ec0SHerbert Xu /** 1286e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1287e0053ec0SHerbert Xu * @skb: buffer to clone 1288e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1289e0053ec0SHerbert Xu * 1290e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1291e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1292e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1293e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1294e0053ec0SHerbert Xu * 1295e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1296e0053ec0SHerbert Xu * %GFP_ATOMIC. 1297e0053ec0SHerbert Xu */ 1298e0053ec0SHerbert Xu 1299e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1300e0053ec0SHerbert Xu { 1301d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1302d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1303d0bf4a9eSEric Dumazet skb1); 13046ffe75ebSEric Dumazet struct sk_buff *n; 1305e0053ec0SHerbert Xu 130670008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1307a6686f2fSShirley Ma return NULL; 1308a6686f2fSShirley Ma 1309e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 13102638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 13116ffe75ebSEric Dumazet n = &fclones->skb2; 13122638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1313e0053ec0SHerbert Xu } else { 1314c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1315c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1316c93bdd0eSMel Gorman 1317e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1318e0053ec0SHerbert Xu if (!n) 1319e0053ec0SHerbert Xu return NULL; 1320fe55f6d5SVegard Nossum 1321e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1322e0053ec0SHerbert Xu } 1323e0053ec0SHerbert Xu 1324e0053ec0SHerbert Xu return __skb_clone(n, skb); 13251da177e4SLinus Torvalds } 1326b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 13271da177e4SLinus Torvalds 1328b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1329f5b17294SPravin B Shelar { 1330030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1331030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1332030737bcSEric Dumazet skb->csum_start += off; 1333f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1334f5b17294SPravin B Shelar skb->transport_header += off; 1335f5b17294SPravin B Shelar skb->network_header += off; 1336f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1337f5b17294SPravin B Shelar skb->mac_header += off; 1338f5b17294SPravin B Shelar skb->inner_transport_header += off; 1339f5b17294SPravin B Shelar skb->inner_network_header += off; 1340aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1341f5b17294SPravin B Shelar } 1342b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1343f5b17294SPravin B Shelar 134408303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 13451da177e4SLinus Torvalds { 1346dec18810SHerbert Xu __copy_skb_header(new, old); 1347dec18810SHerbert Xu 13487967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 13497967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 13507967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 13511da177e4SLinus Torvalds } 135208303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 13531da177e4SLinus Torvalds 1354c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1355c93bdd0eSMel Gorman { 1356c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1357c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1358c93bdd0eSMel Gorman return 0; 1359c93bdd0eSMel Gorman } 1360c93bdd0eSMel Gorman 13611da177e4SLinus Torvalds /** 13621da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 13631da177e4SLinus Torvalds * @skb: buffer to copy 13641da177e4SLinus Torvalds * @gfp_mask: allocation priority 13651da177e4SLinus Torvalds * 13661da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 13671da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 13681da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 13691da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 13701da177e4SLinus Torvalds * 13711da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 13721da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 13731da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 13741da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 13751da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 13761da177e4SLinus Torvalds */ 13771da177e4SLinus Torvalds 1378dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 13791da177e4SLinus Torvalds { 13806602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1381ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1382c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1383c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 13846602cebbSEric Dumazet 13851da177e4SLinus Torvalds if (!n) 13861da177e4SLinus Torvalds return NULL; 13871da177e4SLinus Torvalds 13881da177e4SLinus Torvalds /* Set the data pointer */ 13891da177e4SLinus Torvalds skb_reserve(n, headerlen); 13901da177e4SLinus Torvalds /* Set the tail pointer and length */ 13911da177e4SLinus Torvalds skb_put(n, skb->len); 13921da177e4SLinus Torvalds 13939f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 13941da177e4SLinus Torvalds 139508303c18SIlya Lesokhin skb_copy_header(n, skb); 13961da177e4SLinus Torvalds return n; 13971da177e4SLinus Torvalds } 1398b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds /** 1401bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 14021da177e4SLinus Torvalds * @skb: buffer to copy 1403117632e6SEric Dumazet * @headroom: headroom of new skb 14041da177e4SLinus Torvalds * @gfp_mask: allocation priority 1405bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1406bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1407bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 14081da177e4SLinus Torvalds * 14091da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 14101da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 14111da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 14121da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 14131da177e4SLinus Torvalds * or the pointer to the buffer on success. 14141da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 14151da177e4SLinus Torvalds */ 14161da177e4SLinus Torvalds 1417bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1418bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 14191da177e4SLinus Torvalds { 1420117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1421bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1422bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 14236602cebbSEric Dumazet 14241da177e4SLinus Torvalds if (!n) 14251da177e4SLinus Torvalds goto out; 14261da177e4SLinus Torvalds 14271da177e4SLinus Torvalds /* Set the data pointer */ 1428117632e6SEric Dumazet skb_reserve(n, headroom); 14291da177e4SLinus Torvalds /* Set the tail pointer and length */ 14301da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 14311da177e4SLinus Torvalds /* Copy the bytes */ 1432d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 14331da177e4SLinus Torvalds 143425f484a6SHerbert Xu n->truesize += skb->data_len; 14351da177e4SLinus Torvalds n->data_len = skb->data_len; 14361da177e4SLinus Torvalds n->len = skb->len; 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 14391da177e4SLinus Torvalds int i; 14401da177e4SLinus Torvalds 14411f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 14421f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 14431511022cSDan Carpenter kfree_skb(n); 14441511022cSDan Carpenter n = NULL; 1445a6686f2fSShirley Ma goto out; 1446a6686f2fSShirley Ma } 14471da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 14481da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1449ea2ab693SIan Campbell skb_frag_ref(skb, i); 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 14521da177e4SLinus Torvalds } 14531da177e4SLinus Torvalds 145421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 14551da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 14561da177e4SLinus Torvalds skb_clone_fraglist(n); 14571da177e4SLinus Torvalds } 14581da177e4SLinus Torvalds 145908303c18SIlya Lesokhin skb_copy_header(n, skb); 14601da177e4SLinus Torvalds out: 14611da177e4SLinus Torvalds return n; 14621da177e4SLinus Torvalds } 1463bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 14641da177e4SLinus Torvalds 14651da177e4SLinus Torvalds /** 14661da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 14671da177e4SLinus Torvalds * @skb: buffer to reallocate 14681da177e4SLinus Torvalds * @nhead: room to add at head 14691da177e4SLinus Torvalds * @ntail: room to add at tail 14701da177e4SLinus Torvalds * @gfp_mask: allocation priority 14711da177e4SLinus Torvalds * 1472bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1473bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 14741da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 14751da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 14761da177e4SLinus Torvalds * 14771da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 14781da177e4SLinus Torvalds * reloaded after call to this function. 14791da177e4SLinus Torvalds */ 14801da177e4SLinus Torvalds 148186a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1482dd0fc66fSAl Viro gfp_t gfp_mask) 14831da177e4SLinus Torvalds { 1484158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1485158f323bSEric Dumazet int size = osize + nhead + ntail; 14861da177e4SLinus Torvalds long off; 1487158f323bSEric Dumazet u8 *data; 14881da177e4SLinus Torvalds 14894edd87adSHerbert Xu BUG_ON(nhead < 0); 14904edd87adSHerbert Xu 14919f77fad3STim Hansen BUG_ON(skb_shared(skb)); 14921da177e4SLinus Torvalds 14931da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 14941da177e4SLinus Torvalds 1495c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1496c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1497c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1498c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 14991da177e4SLinus Torvalds if (!data) 15001da177e4SLinus Torvalds goto nodata; 150187151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 15021da177e4SLinus Torvalds 15031da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 15046602cebbSEric Dumazet * optimized for the cases when header is void. 15056602cebbSEric Dumazet */ 15066602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 15076602cebbSEric Dumazet 15086602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 15096602cebbSEric Dumazet skb_shinfo(skb), 1510fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 15111da177e4SLinus Torvalds 15123e24591aSAlexander Duyck /* 15133e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 15143e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 15153e24591aSAlexander Duyck * be since all we did is relocate the values 15163e24591aSAlexander Duyck */ 15173e24591aSAlexander Duyck if (skb_cloned(skb)) { 151870008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1519a6686f2fSShirley Ma goto nofrags; 15201f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1521c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 15221da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1523ea2ab693SIan Campbell skb_frag_ref(skb, i); 15241da177e4SLinus Torvalds 152521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 15261da177e4SLinus Torvalds skb_clone_fraglist(skb); 15271da177e4SLinus Torvalds 15281da177e4SLinus Torvalds skb_release_data(skb); 15293e24591aSAlexander Duyck } else { 15303e24591aSAlexander Duyck skb_free_head(skb); 15311fd63041SEric Dumazet } 15321da177e4SLinus Torvalds off = (data + nhead) - skb->head; 15331da177e4SLinus Torvalds 15341da177e4SLinus Torvalds skb->head = data; 1535d3836f21SEric Dumazet skb->head_frag = 0; 15361da177e4SLinus Torvalds skb->data += off; 15374305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 15384305b541SArnaldo Carvalho de Melo skb->end = size; 153956eb8882SPatrick McHardy off = nhead; 15404305b541SArnaldo Carvalho de Melo #else 15414305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 154256eb8882SPatrick McHardy #endif 154327a884dcSArnaldo Carvalho de Melo skb->tail += off; 1544b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 15451da177e4SLinus Torvalds skb->cloned = 0; 1546334a8132SPatrick McHardy skb->hdr_len = 0; 15471da177e4SLinus Torvalds skb->nohdr = 0; 15481da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1549158f323bSEric Dumazet 1550de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1551de8f3a83SDaniel Borkmann 1552158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1553158f323bSEric Dumazet * For the moment, we really care of rx path, or 1554158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1555158f323bSEric Dumazet */ 1556158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1557158f323bSEric Dumazet skb->truesize += size - osize; 1558158f323bSEric Dumazet 15591da177e4SLinus Torvalds return 0; 15601da177e4SLinus Torvalds 1561a6686f2fSShirley Ma nofrags: 1562a6686f2fSShirley Ma kfree(data); 15631da177e4SLinus Torvalds nodata: 15641da177e4SLinus Torvalds return -ENOMEM; 15651da177e4SLinus Torvalds } 1566b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 15671da177e4SLinus Torvalds 15681da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 15711da177e4SLinus Torvalds { 15721da177e4SLinus Torvalds struct sk_buff *skb2; 15731da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds if (delta <= 0) 15761da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 15771da177e4SLinus Torvalds else { 15781da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 15791da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 15801da177e4SLinus Torvalds GFP_ATOMIC)) { 15811da177e4SLinus Torvalds kfree_skb(skb2); 15821da177e4SLinus Torvalds skb2 = NULL; 15831da177e4SLinus Torvalds } 15841da177e4SLinus Torvalds } 15851da177e4SLinus Torvalds return skb2; 15861da177e4SLinus Torvalds } 1587b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 15881da177e4SLinus Torvalds 15891da177e4SLinus Torvalds /** 15901da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 15911da177e4SLinus Torvalds * @skb: buffer to copy 15921da177e4SLinus Torvalds * @newheadroom: new free bytes at head 15931da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 15941da177e4SLinus Torvalds * @gfp_mask: allocation priority 15951da177e4SLinus Torvalds * 15961da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 15971da177e4SLinus Torvalds * allocate additional space. 15981da177e4SLinus Torvalds * 15991da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 16001da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 16011da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 16021da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 16031da177e4SLinus Torvalds * 16041da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 16051da177e4SLinus Torvalds * is called from an interrupt. 16061da177e4SLinus Torvalds */ 16071da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 160886a76cafSVictor Fusco int newheadroom, int newtailroom, 1609dd0fc66fSAl Viro gfp_t gfp_mask) 16101da177e4SLinus Torvalds { 16111da177e4SLinus Torvalds /* 16121da177e4SLinus Torvalds * Allocate the copy buffer 16131da177e4SLinus Torvalds */ 1614c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1615c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1616c93bdd0eSMel Gorman NUMA_NO_NODE); 1617efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 16181da177e4SLinus Torvalds int head_copy_len, head_copy_off; 16191da177e4SLinus Torvalds 16201da177e4SLinus Torvalds if (!n) 16211da177e4SLinus Torvalds return NULL; 16221da177e4SLinus Torvalds 16231da177e4SLinus Torvalds skb_reserve(n, newheadroom); 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds /* Set the tail pointer and length */ 16261da177e4SLinus Torvalds skb_put(n, skb->len); 16271da177e4SLinus Torvalds 1628efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 16291da177e4SLinus Torvalds head_copy_off = 0; 16301da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 16311da177e4SLinus Torvalds head_copy_len = newheadroom; 16321da177e4SLinus Torvalds else 16331da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds /* Copy the linear header and data. */ 16369f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 16379f77fad3STim Hansen skb->len + head_copy_len)); 16381da177e4SLinus Torvalds 163908303c18SIlya Lesokhin skb_copy_header(n, skb); 16401da177e4SLinus Torvalds 1641030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1642efd1e8d5SPatrick McHardy 16431da177e4SLinus Torvalds return n; 16441da177e4SLinus Torvalds } 1645b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 16461da177e4SLinus Torvalds 16471da177e4SLinus Torvalds /** 1648cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 16491da177e4SLinus Torvalds * @skb: buffer to pad 16501da177e4SLinus Torvalds * @pad: space to pad 1651cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 16521da177e4SLinus Torvalds * 16531da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 16541da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 16551da177e4SLinus Torvalds * beyond the buffer end onto the wire. 16561da177e4SLinus Torvalds * 1657cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1658cd0a137aSFlorian Fainelli * if @free_on_error is true. 16591da177e4SLinus Torvalds */ 16601da177e4SLinus Torvalds 1661cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 16621da177e4SLinus Torvalds { 16635b057c6bSHerbert Xu int err; 16645b057c6bSHerbert Xu int ntail; 16651da177e4SLinus Torvalds 16661da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 16675b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 16681da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 16695b057c6bSHerbert Xu return 0; 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds 16724305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 16735b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 16745b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 16755b057c6bSHerbert Xu if (unlikely(err)) 16765b057c6bSHerbert Xu goto free_skb; 16775b057c6bSHerbert Xu } 16785b057c6bSHerbert Xu 16795b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 16805b057c6bSHerbert Xu * to be audited. 16815b057c6bSHerbert Xu */ 16825b057c6bSHerbert Xu err = skb_linearize(skb); 16835b057c6bSHerbert Xu if (unlikely(err)) 16845b057c6bSHerbert Xu goto free_skb; 16855b057c6bSHerbert Xu 16865b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 16875b057c6bSHerbert Xu return 0; 16885b057c6bSHerbert Xu 16895b057c6bSHerbert Xu free_skb: 1690cd0a137aSFlorian Fainelli if (free_on_error) 16911da177e4SLinus Torvalds kfree_skb(skb); 16925b057c6bSHerbert Xu return err; 16931da177e4SLinus Torvalds } 1694cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 16951da177e4SLinus Torvalds 16960dde3e16SIlpo Järvinen /** 16970c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 16980c7ddf36SMathias Krause * @skb: start of the buffer to use 16990c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 17000c7ddf36SMathias Krause * @len: amount of data to add 17010c7ddf36SMathias Krause * 17020c7ddf36SMathias Krause * This function extends the used data area of the potentially 17030c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 17040c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 17050c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 17060c7ddf36SMathias Krause * returned. 17070c7ddf36SMathias Krause */ 17080c7ddf36SMathias Krause 17094df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 17100c7ddf36SMathias Krause { 17110c7ddf36SMathias Krause if (tail != skb) { 17120c7ddf36SMathias Krause skb->data_len += len; 17130c7ddf36SMathias Krause skb->len += len; 17140c7ddf36SMathias Krause } 17150c7ddf36SMathias Krause return skb_put(tail, len); 17160c7ddf36SMathias Krause } 17170c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 17180c7ddf36SMathias Krause 17190c7ddf36SMathias Krause /** 17200dde3e16SIlpo Järvinen * skb_put - add data to a buffer 17210dde3e16SIlpo Järvinen * @skb: buffer to use 17220dde3e16SIlpo Järvinen * @len: amount of data to add 17230dde3e16SIlpo Järvinen * 17240dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 17250dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 17260dde3e16SIlpo Järvinen * first byte of the extra data is returned. 17270dde3e16SIlpo Järvinen */ 17284df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 17290dde3e16SIlpo Järvinen { 17304df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 17310dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 17320dde3e16SIlpo Järvinen skb->tail += len; 17330dde3e16SIlpo Järvinen skb->len += len; 17340dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 17350dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 17360dde3e16SIlpo Järvinen return tmp; 17370dde3e16SIlpo Järvinen } 17380dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 17390dde3e16SIlpo Järvinen 17406be8ac2fSIlpo Järvinen /** 1741c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1742c2aa270aSIlpo Järvinen * @skb: buffer to use 1743c2aa270aSIlpo Järvinen * @len: amount of data to add 1744c2aa270aSIlpo Järvinen * 1745c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1746c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1747c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1748c2aa270aSIlpo Järvinen */ 1749d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 1750c2aa270aSIlpo Järvinen { 1751c2aa270aSIlpo Järvinen skb->data -= len; 1752c2aa270aSIlpo Järvinen skb->len += len; 1753c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 1754c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1755c2aa270aSIlpo Järvinen return skb->data; 1756c2aa270aSIlpo Järvinen } 1757c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1758c2aa270aSIlpo Järvinen 1759c2aa270aSIlpo Järvinen /** 17606be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 17616be8ac2fSIlpo Järvinen * @skb: buffer to use 17626be8ac2fSIlpo Järvinen * @len: amount of data to remove 17636be8ac2fSIlpo Järvinen * 17646be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 17656be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 17666be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 17676be8ac2fSIlpo Järvinen * the old data. 17686be8ac2fSIlpo Järvinen */ 1769af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 17706be8ac2fSIlpo Järvinen { 177147d29646SDavid S. Miller return skb_pull_inline(skb, len); 17726be8ac2fSIlpo Järvinen } 17736be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 17746be8ac2fSIlpo Järvinen 1775419ae74eSIlpo Järvinen /** 1776419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1777419ae74eSIlpo Järvinen * @skb: buffer to alter 1778419ae74eSIlpo Järvinen * @len: new length 1779419ae74eSIlpo Järvinen * 1780419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1781419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1782419ae74eSIlpo Järvinen * The skb must be linear. 1783419ae74eSIlpo Järvinen */ 1784419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1785419ae74eSIlpo Järvinen { 1786419ae74eSIlpo Järvinen if (skb->len > len) 1787419ae74eSIlpo Järvinen __skb_trim(skb, len); 1788419ae74eSIlpo Järvinen } 1789419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1790419ae74eSIlpo Järvinen 17913cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 17921da177e4SLinus Torvalds */ 17931da177e4SLinus Torvalds 17943cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 17951da177e4SLinus Torvalds { 179627b437c8SHerbert Xu struct sk_buff **fragp; 179727b437c8SHerbert Xu struct sk_buff *frag; 17981da177e4SLinus Torvalds int offset = skb_headlen(skb); 17991da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 18001da177e4SLinus Torvalds int i; 180127b437c8SHerbert Xu int err; 180227b437c8SHerbert Xu 180327b437c8SHerbert Xu if (skb_cloned(skb) && 180427b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 180527b437c8SHerbert Xu return err; 18061da177e4SLinus Torvalds 1807f4d26fb3SHerbert Xu i = 0; 1808f4d26fb3SHerbert Xu if (offset >= len) 1809f4d26fb3SHerbert Xu goto drop_pages; 1810f4d26fb3SHerbert Xu 1811f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 18129e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 181327b437c8SHerbert Xu 181427b437c8SHerbert Xu if (end < len) { 18151da177e4SLinus Torvalds offset = end; 181627b437c8SHerbert Xu continue; 18171da177e4SLinus Torvalds } 18181da177e4SLinus Torvalds 18199e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 182027b437c8SHerbert Xu 1821f4d26fb3SHerbert Xu drop_pages: 182227b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 182327b437c8SHerbert Xu 182427b437c8SHerbert Xu for (; i < nfrags; i++) 1825ea2ab693SIan Campbell skb_frag_unref(skb, i); 182627b437c8SHerbert Xu 182721dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 182827b437c8SHerbert Xu skb_drop_fraglist(skb); 1829f4d26fb3SHerbert Xu goto done; 183027b437c8SHerbert Xu } 183127b437c8SHerbert Xu 183227b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 183327b437c8SHerbert Xu fragp = &frag->next) { 183427b437c8SHerbert Xu int end = offset + frag->len; 183527b437c8SHerbert Xu 183627b437c8SHerbert Xu if (skb_shared(frag)) { 183727b437c8SHerbert Xu struct sk_buff *nfrag; 183827b437c8SHerbert Xu 183927b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 184027b437c8SHerbert Xu if (unlikely(!nfrag)) 184127b437c8SHerbert Xu return -ENOMEM; 184227b437c8SHerbert Xu 184327b437c8SHerbert Xu nfrag->next = frag->next; 184485bb2a60SEric Dumazet consume_skb(frag); 184527b437c8SHerbert Xu frag = nfrag; 184627b437c8SHerbert Xu *fragp = frag; 184727b437c8SHerbert Xu } 184827b437c8SHerbert Xu 184927b437c8SHerbert Xu if (end < len) { 185027b437c8SHerbert Xu offset = end; 185127b437c8SHerbert Xu continue; 185227b437c8SHerbert Xu } 185327b437c8SHerbert Xu 185427b437c8SHerbert Xu if (end > len && 185527b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 185627b437c8SHerbert Xu return err; 185727b437c8SHerbert Xu 185827b437c8SHerbert Xu if (frag->next) 185927b437c8SHerbert Xu skb_drop_list(&frag->next); 186027b437c8SHerbert Xu break; 186127b437c8SHerbert Xu } 186227b437c8SHerbert Xu 1863f4d26fb3SHerbert Xu done: 186427b437c8SHerbert Xu if (len > skb_headlen(skb)) { 18651da177e4SLinus Torvalds skb->data_len -= skb->len - len; 18661da177e4SLinus Torvalds skb->len = len; 18671da177e4SLinus Torvalds } else { 18681da177e4SLinus Torvalds skb->len = len; 18691da177e4SLinus Torvalds skb->data_len = 0; 187027a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 18711da177e4SLinus Torvalds } 18721da177e4SLinus Torvalds 1873c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1874c21b48ccSEric Dumazet skb_condense(skb); 18751da177e4SLinus Torvalds return 0; 18761da177e4SLinus Torvalds } 1877b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 18781da177e4SLinus Torvalds 187988078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 188088078d98SEric Dumazet */ 188188078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 188288078d98SEric Dumazet { 188388078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 188488078d98SEric Dumazet int delta = skb->len - len; 188588078d98SEric Dumazet 1886d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 1887d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 1888d55bef50SDimitris Michailidis len); 188988078d98SEric Dumazet } 189088078d98SEric Dumazet return __pskb_trim(skb, len); 189188078d98SEric Dumazet } 189288078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 189388078d98SEric Dumazet 18941da177e4SLinus Torvalds /** 18951da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 18961da177e4SLinus Torvalds * @skb: buffer to reallocate 18971da177e4SLinus Torvalds * @delta: number of bytes to advance tail 18981da177e4SLinus Torvalds * 18991da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 19001da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 19011da177e4SLinus Torvalds * data from fragmented part. 19021da177e4SLinus Torvalds * 19031da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 19041da177e4SLinus Torvalds * 19051da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 19061da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 19071da177e4SLinus Torvalds * 19081da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 19091da177e4SLinus Torvalds * reloaded after call to this function. 19101da177e4SLinus Torvalds */ 19111da177e4SLinus Torvalds 19121da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 19131da177e4SLinus Torvalds * when it is necessary. 19141da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 19151da177e4SLinus Torvalds * 2. It may change skb pointers. 19161da177e4SLinus Torvalds * 19171da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 19181da177e4SLinus Torvalds */ 1919af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 19201da177e4SLinus Torvalds { 19211da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 19221da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 19231da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 19241da177e4SLinus Torvalds */ 19254305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 19261da177e4SLinus Torvalds 19271da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 19281da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 19291da177e4SLinus Torvalds GFP_ATOMIC)) 19301da177e4SLinus Torvalds return NULL; 19311da177e4SLinus Torvalds } 19321da177e4SLinus Torvalds 19339f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 19349f77fad3STim Hansen skb_tail_pointer(skb), delta)); 19351da177e4SLinus Torvalds 19361da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 19371da177e4SLinus Torvalds * size of pulled pages. Superb. 19381da177e4SLinus Torvalds */ 193921dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 19401da177e4SLinus Torvalds goto pull_pages; 19411da177e4SLinus Torvalds 19421da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 19431da177e4SLinus Torvalds eat = delta; 19441da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 19459e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 19469e903e08SEric Dumazet 19479e903e08SEric Dumazet if (size >= eat) 19481da177e4SLinus Torvalds goto pull_pages; 19499e903e08SEric Dumazet eat -= size; 19501da177e4SLinus Torvalds } 19511da177e4SLinus Torvalds 19521da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 195309001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 19541da177e4SLinus Torvalds * but taking into account that pulling is expected to 19551da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 19561da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 19571da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 19581da177e4SLinus Torvalds */ 19591da177e4SLinus Torvalds if (eat) { 19601da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 19611da177e4SLinus Torvalds struct sk_buff *clone = NULL; 19621da177e4SLinus Torvalds struct sk_buff *insp = NULL; 19631da177e4SLinus Torvalds 19641da177e4SLinus Torvalds do { 19651da177e4SLinus Torvalds if (list->len <= eat) { 19661da177e4SLinus Torvalds /* Eaten as whole. */ 19671da177e4SLinus Torvalds eat -= list->len; 19681da177e4SLinus Torvalds list = list->next; 19691da177e4SLinus Torvalds insp = list; 19701da177e4SLinus Torvalds } else { 19711da177e4SLinus Torvalds /* Eaten partially. */ 19721da177e4SLinus Torvalds 19731da177e4SLinus Torvalds if (skb_shared(list)) { 19741da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 19751da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 19761da177e4SLinus Torvalds if (!clone) 19771da177e4SLinus Torvalds return NULL; 19781da177e4SLinus Torvalds insp = list->next; 19791da177e4SLinus Torvalds list = clone; 19801da177e4SLinus Torvalds } else { 19811da177e4SLinus Torvalds /* This may be pulled without 19821da177e4SLinus Torvalds * problems. */ 19831da177e4SLinus Torvalds insp = list; 19841da177e4SLinus Torvalds } 19851da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 19861da177e4SLinus Torvalds kfree_skb(clone); 19871da177e4SLinus Torvalds return NULL; 19881da177e4SLinus Torvalds } 19891da177e4SLinus Torvalds break; 19901da177e4SLinus Torvalds } 19911da177e4SLinus Torvalds } while (eat); 19921da177e4SLinus Torvalds 19931da177e4SLinus Torvalds /* Free pulled out fragments. */ 19941da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 19951da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 19961da177e4SLinus Torvalds kfree_skb(list); 19971da177e4SLinus Torvalds } 19981da177e4SLinus Torvalds /* And insert new clone at head. */ 19991da177e4SLinus Torvalds if (clone) { 20001da177e4SLinus Torvalds clone->next = list; 20011da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 20021da177e4SLinus Torvalds } 20031da177e4SLinus Torvalds } 20041da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 20051da177e4SLinus Torvalds 20061da177e4SLinus Torvalds pull_pages: 20071da177e4SLinus Torvalds eat = delta; 20081da177e4SLinus Torvalds k = 0; 20091da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20109e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 20119e903e08SEric Dumazet 20129e903e08SEric Dumazet if (size <= eat) { 2013ea2ab693SIan Campbell skb_frag_unref(skb, i); 20149e903e08SEric Dumazet eat -= size; 20151da177e4SLinus Torvalds } else { 20161da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 20171da177e4SLinus Torvalds if (eat) { 20181da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 20199e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 20203ccc6c6fSlinzhang if (!i) 20213ccc6c6fSlinzhang goto end; 20221da177e4SLinus Torvalds eat = 0; 20231da177e4SLinus Torvalds } 20241da177e4SLinus Torvalds k++; 20251da177e4SLinus Torvalds } 20261da177e4SLinus Torvalds } 20271da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 20281da177e4SLinus Torvalds 20293ccc6c6fSlinzhang end: 20301da177e4SLinus Torvalds skb->tail += delta; 20311da177e4SLinus Torvalds skb->data_len -= delta; 20321da177e4SLinus Torvalds 20331f8b977aSWillem de Bruijn if (!skb->data_len) 20341f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 20351f8b977aSWillem de Bruijn 203627a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 20371da177e4SLinus Torvalds } 2038b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 20391da177e4SLinus Torvalds 204022019b17SEric Dumazet /** 204122019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 204222019b17SEric Dumazet * @skb: source skb 204322019b17SEric Dumazet * @offset: offset in source 204422019b17SEric Dumazet * @to: destination buffer 204522019b17SEric Dumazet * @len: number of bytes to copy 204622019b17SEric Dumazet * 204722019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 204822019b17SEric Dumazet * destination buffer. 204922019b17SEric Dumazet * 205022019b17SEric Dumazet * CAUTION ! : 205122019b17SEric Dumazet * If its prototype is ever changed, 205222019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 205322019b17SEric Dumazet * since it is called from BPF assembly code. 205422019b17SEric Dumazet */ 20551da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 20561da177e4SLinus Torvalds { 20571a028e50SDavid S. Miller int start = skb_headlen(skb); 2058fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2059fbb398a8SDavid S. Miller int i, copy; 20601da177e4SLinus Torvalds 20611da177e4SLinus Torvalds if (offset > (int)skb->len - len) 20621da177e4SLinus Torvalds goto fault; 20631da177e4SLinus Torvalds 20641da177e4SLinus Torvalds /* Copy header. */ 20651a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 20661da177e4SLinus Torvalds if (copy > len) 20671da177e4SLinus Torvalds copy = len; 2068d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 20691da177e4SLinus Torvalds if ((len -= copy) == 0) 20701da177e4SLinus Torvalds return 0; 20711da177e4SLinus Torvalds offset += copy; 20721da177e4SLinus Torvalds to += copy; 20731da177e4SLinus Torvalds } 20741da177e4SLinus Torvalds 20751da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20761a028e50SDavid S. Miller int end; 207751c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 20781da177e4SLinus Torvalds 2079547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20801a028e50SDavid S. Miller 208151c56b00SEric Dumazet end = start + skb_frag_size(f); 20821da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2083c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2084c613c209SWillem de Bruijn struct page *p; 20851da177e4SLinus Torvalds u8 *vaddr; 20861da177e4SLinus Torvalds 20871da177e4SLinus Torvalds if (copy > len) 20881da177e4SLinus Torvalds copy = len; 20891da177e4SLinus Torvalds 2090c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2091c613c209SWillem de Bruijn f->page_offset + offset - start, 2092c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2093c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2094c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 209551c56b00SEric Dumazet kunmap_atomic(vaddr); 2096c613c209SWillem de Bruijn } 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds if ((len -= copy) == 0) 20991da177e4SLinus Torvalds return 0; 21001da177e4SLinus Torvalds offset += copy; 21011da177e4SLinus Torvalds to += copy; 21021da177e4SLinus Torvalds } 21031a028e50SDavid S. Miller start = end; 21041da177e4SLinus Torvalds } 21051da177e4SLinus Torvalds 2106fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 21071a028e50SDavid S. Miller int end; 21081da177e4SLinus Torvalds 2109547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21101a028e50SDavid S. Miller 2111fbb398a8SDavid S. Miller end = start + frag_iter->len; 21121da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21131da177e4SLinus Torvalds if (copy > len) 21141da177e4SLinus Torvalds copy = len; 2115fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 21161da177e4SLinus Torvalds goto fault; 21171da177e4SLinus Torvalds if ((len -= copy) == 0) 21181da177e4SLinus Torvalds return 0; 21191da177e4SLinus Torvalds offset += copy; 21201da177e4SLinus Torvalds to += copy; 21211da177e4SLinus Torvalds } 21221a028e50SDavid S. Miller start = end; 21231da177e4SLinus Torvalds } 2124a6686f2fSShirley Ma 21251da177e4SLinus Torvalds if (!len) 21261da177e4SLinus Torvalds return 0; 21271da177e4SLinus Torvalds 21281da177e4SLinus Torvalds fault: 21291da177e4SLinus Torvalds return -EFAULT; 21301da177e4SLinus Torvalds } 2131b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 21321da177e4SLinus Torvalds 21339c55e01cSJens Axboe /* 21349c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 21359c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 21369c55e01cSJens Axboe */ 21379c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 21389c55e01cSJens Axboe { 21398b9d3728SJarek Poplawski put_page(spd->pages[i]); 21408b9d3728SJarek Poplawski } 21419c55e01cSJens Axboe 2142a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 21434fb66994SJarek Poplawski unsigned int *offset, 214418aafc62SEric Dumazet struct sock *sk) 21458b9d3728SJarek Poplawski { 21465640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 21478b9d3728SJarek Poplawski 21485640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 21498b9d3728SJarek Poplawski return NULL; 21504fb66994SJarek Poplawski 21515640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 21524fb66994SJarek Poplawski 21535640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 21545640f768SEric Dumazet page_address(page) + *offset, *len); 21555640f768SEric Dumazet *offset = pfrag->offset; 21565640f768SEric Dumazet pfrag->offset += *len; 21574fb66994SJarek Poplawski 21585640f768SEric Dumazet return pfrag->page; 21599c55e01cSJens Axboe } 21609c55e01cSJens Axboe 216141c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 216241c73a0dSEric Dumazet struct page *page, 216341c73a0dSEric Dumazet unsigned int offset) 216441c73a0dSEric Dumazet { 216541c73a0dSEric Dumazet return spd->nr_pages && 216641c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 216741c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 216841c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 216941c73a0dSEric Dumazet } 217041c73a0dSEric Dumazet 21719c55e01cSJens Axboe /* 21729c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 21739c55e01cSJens Axboe */ 2174a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 217535f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 21764fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 217718aafc62SEric Dumazet bool linear, 21787a67e56fSJarek Poplawski struct sock *sk) 21799c55e01cSJens Axboe { 218041c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2181a108d5f3SDavid S. Miller return true; 21829c55e01cSJens Axboe 21838b9d3728SJarek Poplawski if (linear) { 218418aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 21858b9d3728SJarek Poplawski if (!page) 2186a108d5f3SDavid S. Miller return true; 218741c73a0dSEric Dumazet } 218841c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 218941c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2190a108d5f3SDavid S. Miller return false; 219141c73a0dSEric Dumazet } 21928b9d3728SJarek Poplawski get_page(page); 21939c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 21944fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 21959c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 21969c55e01cSJens Axboe spd->nr_pages++; 21978b9d3728SJarek Poplawski 2198a108d5f3SDavid S. Miller return false; 21999c55e01cSJens Axboe } 22009c55e01cSJens Axboe 2201a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 22022870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 220318aafc62SEric Dumazet unsigned int *len, 2204d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 220535f3d14dSJens Axboe struct sock *sk, 220635f3d14dSJens Axboe struct pipe_inode_info *pipe) 22079c55e01cSJens Axboe { 22082870c43dSOctavian Purdila if (!*len) 2209a108d5f3SDavid S. Miller return true; 22109c55e01cSJens Axboe 22112870c43dSOctavian Purdila /* skip this segment if already processed */ 22122870c43dSOctavian Purdila if (*off >= plen) { 22132870c43dSOctavian Purdila *off -= plen; 2214a108d5f3SDavid S. Miller return false; 22152870c43dSOctavian Purdila } 22162870c43dSOctavian Purdila 22172870c43dSOctavian Purdila /* ignore any bits we already processed */ 22189ca1b22dSEric Dumazet poff += *off; 22199ca1b22dSEric Dumazet plen -= *off; 22202870c43dSOctavian Purdila *off = 0; 22212870c43dSOctavian Purdila 222218aafc62SEric Dumazet do { 222318aafc62SEric Dumazet unsigned int flen = min(*len, plen); 22242870c43dSOctavian Purdila 222518aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 222618aafc62SEric Dumazet linear, sk)) 2227a108d5f3SDavid S. Miller return true; 222818aafc62SEric Dumazet poff += flen; 222918aafc62SEric Dumazet plen -= flen; 22302870c43dSOctavian Purdila *len -= flen; 223118aafc62SEric Dumazet } while (*len && plen); 22322870c43dSOctavian Purdila 2233a108d5f3SDavid S. Miller return false; 2234db43a282SOctavian Purdila } 22359c55e01cSJens Axboe 22369c55e01cSJens Axboe /* 2237a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 22382870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 22399c55e01cSJens Axboe */ 2240a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 224135f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 224235f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 22432870c43dSOctavian Purdila { 22442870c43dSOctavian Purdila int seg; 2245fa9835e5STom Herbert struct sk_buff *iter; 22469c55e01cSJens Axboe 22471d0c0b32SEric Dumazet /* map the linear part : 22482996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 22492996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 22502996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 22519c55e01cSJens Axboe */ 22522870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 22532870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 22542870c43dSOctavian Purdila skb_headlen(skb), 225518aafc62SEric Dumazet offset, len, spd, 22563a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 22571d0c0b32SEric Dumazet sk, pipe)) 2258a108d5f3SDavid S. Miller return true; 22599c55e01cSJens Axboe 22609c55e01cSJens Axboe /* 22619c55e01cSJens Axboe * then map the fragments 22629c55e01cSJens Axboe */ 22639c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 22649c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 22659c55e01cSJens Axboe 2266ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 22679e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 226818aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2269a108d5f3SDavid S. Miller return true; 22709c55e01cSJens Axboe } 22719c55e01cSJens Axboe 2272fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2273fa9835e5STom Herbert if (*offset >= iter->len) { 2274fa9835e5STom Herbert *offset -= iter->len; 2275fa9835e5STom Herbert continue; 2276fa9835e5STom Herbert } 2277fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2278fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2279fa9835e5STom Herbert * case. 2280fa9835e5STom Herbert */ 2281fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2282fa9835e5STom Herbert return true; 2283fa9835e5STom Herbert } 2284fa9835e5STom Herbert 2285a108d5f3SDavid S. Miller return false; 22869c55e01cSJens Axboe } 22879c55e01cSJens Axboe 22889c55e01cSJens Axboe /* 22899c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2290fa9835e5STom Herbert * the fragments, and the frag list. 22919c55e01cSJens Axboe */ 2292a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 22939c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 229425869262SAl Viro unsigned int flags) 22959c55e01cSJens Axboe { 229641c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 229741c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 22989c55e01cSJens Axboe struct splice_pipe_desc spd = { 22999c55e01cSJens Axboe .pages = pages, 23009c55e01cSJens Axboe .partial = partial, 2301047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 230228a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 23039c55e01cSJens Axboe .spd_release = sock_spd_release, 23049c55e01cSJens Axboe }; 230535f3d14dSJens Axboe int ret = 0; 230635f3d14dSJens Axboe 2307fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 23089c55e01cSJens Axboe 2309a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 231025869262SAl Viro ret = splice_to_pipe(pipe, &spd); 23119c55e01cSJens Axboe 231235f3d14dSJens Axboe return ret; 23139c55e01cSJens Axboe } 23142b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 23159c55e01cSJens Axboe 231620bf50deSTom Herbert /* Send skb data on a socket. Socket must be locked. */ 231720bf50deSTom Herbert int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 231820bf50deSTom Herbert int len) 231920bf50deSTom Herbert { 232020bf50deSTom Herbert unsigned int orig_len = len; 232120bf50deSTom Herbert struct sk_buff *head = skb; 232220bf50deSTom Herbert unsigned short fragidx; 232320bf50deSTom Herbert int slen, ret; 232420bf50deSTom Herbert 232520bf50deSTom Herbert do_frag_list: 232620bf50deSTom Herbert 232720bf50deSTom Herbert /* Deal with head data */ 232820bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 232920bf50deSTom Herbert struct kvec kv; 233020bf50deSTom Herbert struct msghdr msg; 233120bf50deSTom Herbert 233220bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 233320bf50deSTom Herbert kv.iov_base = skb->data + offset; 2334db5980d8SJohn Fastabend kv.iov_len = slen; 233520bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 233620bf50deSTom Herbert 233720bf50deSTom Herbert ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen); 233820bf50deSTom Herbert if (ret <= 0) 233920bf50deSTom Herbert goto error; 234020bf50deSTom Herbert 234120bf50deSTom Herbert offset += ret; 234220bf50deSTom Herbert len -= ret; 234320bf50deSTom Herbert } 234420bf50deSTom Herbert 234520bf50deSTom Herbert /* All the data was skb head? */ 234620bf50deSTom Herbert if (!len) 234720bf50deSTom Herbert goto out; 234820bf50deSTom Herbert 234920bf50deSTom Herbert /* Make offset relative to start of frags */ 235020bf50deSTom Herbert offset -= skb_headlen(skb); 235120bf50deSTom Herbert 235220bf50deSTom Herbert /* Find where we are in frag list */ 235320bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 235420bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 235520bf50deSTom Herbert 235620bf50deSTom Herbert if (offset < frag->size) 235720bf50deSTom Herbert break; 235820bf50deSTom Herbert 235920bf50deSTom Herbert offset -= frag->size; 236020bf50deSTom Herbert } 236120bf50deSTom Herbert 236220bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 236320bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 236420bf50deSTom Herbert 236520bf50deSTom Herbert slen = min_t(size_t, len, frag->size - offset); 236620bf50deSTom Herbert 236720bf50deSTom Herbert while (slen) { 236820bf50deSTom Herbert ret = kernel_sendpage_locked(sk, frag->page.p, 236920bf50deSTom Herbert frag->page_offset + offset, 237020bf50deSTom Herbert slen, MSG_DONTWAIT); 237120bf50deSTom Herbert if (ret <= 0) 237220bf50deSTom Herbert goto error; 237320bf50deSTom Herbert 237420bf50deSTom Herbert len -= ret; 237520bf50deSTom Herbert offset += ret; 237620bf50deSTom Herbert slen -= ret; 237720bf50deSTom Herbert } 237820bf50deSTom Herbert 237920bf50deSTom Herbert offset = 0; 238020bf50deSTom Herbert } 238120bf50deSTom Herbert 238220bf50deSTom Herbert if (len) { 238320bf50deSTom Herbert /* Process any frag lists */ 238420bf50deSTom Herbert 238520bf50deSTom Herbert if (skb == head) { 238620bf50deSTom Herbert if (skb_has_frag_list(skb)) { 238720bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 238820bf50deSTom Herbert goto do_frag_list; 238920bf50deSTom Herbert } 239020bf50deSTom Herbert } else if (skb->next) { 239120bf50deSTom Herbert skb = skb->next; 239220bf50deSTom Herbert goto do_frag_list; 239320bf50deSTom Herbert } 239420bf50deSTom Herbert } 239520bf50deSTom Herbert 239620bf50deSTom Herbert out: 239720bf50deSTom Herbert return orig_len - len; 239820bf50deSTom Herbert 239920bf50deSTom Herbert error: 240020bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 240120bf50deSTom Herbert } 240220bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 240320bf50deSTom Herbert 2404357b40a1SHerbert Xu /** 2405357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2406357b40a1SHerbert Xu * @skb: destination buffer 2407357b40a1SHerbert Xu * @offset: offset in destination 2408357b40a1SHerbert Xu * @from: source buffer 2409357b40a1SHerbert Xu * @len: number of bytes to copy 2410357b40a1SHerbert Xu * 2411357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2412357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2413357b40a1SHerbert Xu * traversing fragment lists and such. 2414357b40a1SHerbert Xu */ 2415357b40a1SHerbert Xu 24160c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2417357b40a1SHerbert Xu { 24181a028e50SDavid S. Miller int start = skb_headlen(skb); 2419fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2420fbb398a8SDavid S. Miller int i, copy; 2421357b40a1SHerbert Xu 2422357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2423357b40a1SHerbert Xu goto fault; 2424357b40a1SHerbert Xu 24251a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2426357b40a1SHerbert Xu if (copy > len) 2427357b40a1SHerbert Xu copy = len; 242827d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2429357b40a1SHerbert Xu if ((len -= copy) == 0) 2430357b40a1SHerbert Xu return 0; 2431357b40a1SHerbert Xu offset += copy; 2432357b40a1SHerbert Xu from += copy; 2433357b40a1SHerbert Xu } 2434357b40a1SHerbert Xu 2435357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2436357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 24371a028e50SDavid S. Miller int end; 2438357b40a1SHerbert Xu 2439547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24401a028e50SDavid S. Miller 24419e903e08SEric Dumazet end = start + skb_frag_size(frag); 2442357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2443c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2444c613c209SWillem de Bruijn struct page *p; 2445357b40a1SHerbert Xu u8 *vaddr; 2446357b40a1SHerbert Xu 2447357b40a1SHerbert Xu if (copy > len) 2448357b40a1SHerbert Xu copy = len; 2449357b40a1SHerbert Xu 2450c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2451c613c209SWillem de Bruijn frag->page_offset + offset - start, 2452c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2453c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2454c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 245551c56b00SEric Dumazet kunmap_atomic(vaddr); 2456c613c209SWillem de Bruijn } 2457357b40a1SHerbert Xu 2458357b40a1SHerbert Xu if ((len -= copy) == 0) 2459357b40a1SHerbert Xu return 0; 2460357b40a1SHerbert Xu offset += copy; 2461357b40a1SHerbert Xu from += copy; 2462357b40a1SHerbert Xu } 24631a028e50SDavid S. Miller start = end; 2464357b40a1SHerbert Xu } 2465357b40a1SHerbert Xu 2466fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 24671a028e50SDavid S. Miller int end; 2468357b40a1SHerbert Xu 2469547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24701a028e50SDavid S. Miller 2471fbb398a8SDavid S. Miller end = start + frag_iter->len; 2472357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2473357b40a1SHerbert Xu if (copy > len) 2474357b40a1SHerbert Xu copy = len; 2475fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 24761a028e50SDavid S. Miller from, copy)) 2477357b40a1SHerbert Xu goto fault; 2478357b40a1SHerbert Xu if ((len -= copy) == 0) 2479357b40a1SHerbert Xu return 0; 2480357b40a1SHerbert Xu offset += copy; 2481357b40a1SHerbert Xu from += copy; 2482357b40a1SHerbert Xu } 24831a028e50SDavid S. Miller start = end; 2484357b40a1SHerbert Xu } 2485357b40a1SHerbert Xu if (!len) 2486357b40a1SHerbert Xu return 0; 2487357b40a1SHerbert Xu 2488357b40a1SHerbert Xu fault: 2489357b40a1SHerbert Xu return -EFAULT; 2490357b40a1SHerbert Xu } 2491357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2492357b40a1SHerbert Xu 24931da177e4SLinus Torvalds /* Checksum skb data. */ 24942817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 24952817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 24961da177e4SLinus Torvalds { 24971a028e50SDavid S. Miller int start = skb_headlen(skb); 24981a028e50SDavid S. Miller int i, copy = start - offset; 2499fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 25001da177e4SLinus Torvalds int pos = 0; 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds /* Checksum header. */ 25031da177e4SLinus Torvalds if (copy > 0) { 25041da177e4SLinus Torvalds if (copy > len) 25051da177e4SLinus Torvalds copy = len; 25062817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 25071da177e4SLinus Torvalds if ((len -= copy) == 0) 25081da177e4SLinus Torvalds return csum; 25091da177e4SLinus Torvalds offset += copy; 25101da177e4SLinus Torvalds pos = copy; 25111da177e4SLinus Torvalds } 25121da177e4SLinus Torvalds 25131da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 25141a028e50SDavid S. Miller int end; 251551c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 25161da177e4SLinus Torvalds 2517547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25181a028e50SDavid S. Miller 251951c56b00SEric Dumazet end = start + skb_frag_size(frag); 25201da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2521c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2522c613c209SWillem de Bruijn struct page *p; 252344bb9363SAl Viro __wsum csum2; 25241da177e4SLinus Torvalds u8 *vaddr; 25251da177e4SLinus Torvalds 25261da177e4SLinus Torvalds if (copy > len) 25271da177e4SLinus Torvalds copy = len; 2528c613c209SWillem de Bruijn 2529c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2530c613c209SWillem de Bruijn frag->page_offset + offset - start, 2531c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2532c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2533c613c209SWillem de Bruijn csum2 = ops->update(vaddr + p_off, p_len, 0); 253451c56b00SEric Dumazet kunmap_atomic(vaddr); 2535c613c209SWillem de Bruijn csum = ops->combine(csum, csum2, pos, p_len); 2536c613c209SWillem de Bruijn pos += p_len; 2537c613c209SWillem de Bruijn } 2538c613c209SWillem de Bruijn 25391da177e4SLinus Torvalds if (!(len -= copy)) 25401da177e4SLinus Torvalds return csum; 25411da177e4SLinus Torvalds offset += copy; 25421da177e4SLinus Torvalds } 25431a028e50SDavid S. Miller start = end; 25441da177e4SLinus Torvalds } 25451da177e4SLinus Torvalds 2546fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 25471a028e50SDavid S. Miller int end; 25481da177e4SLinus Torvalds 2549547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25501a028e50SDavid S. Miller 2551fbb398a8SDavid S. Miller end = start + frag_iter->len; 25521da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 25535f92a738SAl Viro __wsum csum2; 25541da177e4SLinus Torvalds if (copy > len) 25551da177e4SLinus Torvalds copy = len; 25562817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 25572817a336SDaniel Borkmann copy, 0, ops); 25582817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 25591da177e4SLinus Torvalds if ((len -= copy) == 0) 25601da177e4SLinus Torvalds return csum; 25611da177e4SLinus Torvalds offset += copy; 25621da177e4SLinus Torvalds pos += copy; 25631da177e4SLinus Torvalds } 25641a028e50SDavid S. Miller start = end; 25651da177e4SLinus Torvalds } 256609a62660SKris Katterjohn BUG_ON(len); 25671da177e4SLinus Torvalds 25681da177e4SLinus Torvalds return csum; 25691da177e4SLinus Torvalds } 25702817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 25712817a336SDaniel Borkmann 25722817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 25732817a336SDaniel Borkmann int len, __wsum csum) 25742817a336SDaniel Borkmann { 25752817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2576cea80ea8SDaniel Borkmann .update = csum_partial_ext, 25772817a336SDaniel Borkmann .combine = csum_block_add_ext, 25782817a336SDaniel Borkmann }; 25792817a336SDaniel Borkmann 25802817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 25812817a336SDaniel Borkmann } 2582b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 25831da177e4SLinus Torvalds 25841da177e4SLinus Torvalds /* Both of above in one bottle. */ 25851da177e4SLinus Torvalds 258681d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 258781d77662SAl Viro u8 *to, int len, __wsum csum) 25881da177e4SLinus Torvalds { 25891a028e50SDavid S. Miller int start = skb_headlen(skb); 25901a028e50SDavid S. Miller int i, copy = start - offset; 2591fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 25921da177e4SLinus Torvalds int pos = 0; 25931da177e4SLinus Torvalds 25941da177e4SLinus Torvalds /* Copy header. */ 25951da177e4SLinus Torvalds if (copy > 0) { 25961da177e4SLinus Torvalds if (copy > len) 25971da177e4SLinus Torvalds copy = len; 25981da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 25991da177e4SLinus Torvalds copy, csum); 26001da177e4SLinus Torvalds if ((len -= copy) == 0) 26011da177e4SLinus Torvalds return csum; 26021da177e4SLinus Torvalds offset += copy; 26031da177e4SLinus Torvalds to += copy; 26041da177e4SLinus Torvalds pos = copy; 26051da177e4SLinus Torvalds } 26061da177e4SLinus Torvalds 26071da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 26081a028e50SDavid S. Miller int end; 26091da177e4SLinus Torvalds 2610547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26111a028e50SDavid S. Miller 26129e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 26131da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2614c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2615c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2616c613c209SWillem de Bruijn struct page *p; 26175084205fSAl Viro __wsum csum2; 26181da177e4SLinus Torvalds u8 *vaddr; 26191da177e4SLinus Torvalds 26201da177e4SLinus Torvalds if (copy > len) 26211da177e4SLinus Torvalds copy = len; 2622c613c209SWillem de Bruijn 2623c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2624c613c209SWillem de Bruijn frag->page_offset + offset - start, 2625c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2626c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2627c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2628c613c209SWillem de Bruijn to + copied, 2629c613c209SWillem de Bruijn p_len, 0); 263051c56b00SEric Dumazet kunmap_atomic(vaddr); 26311da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2632c613c209SWillem de Bruijn pos += p_len; 2633c613c209SWillem de Bruijn } 2634c613c209SWillem de Bruijn 26351da177e4SLinus Torvalds if (!(len -= copy)) 26361da177e4SLinus Torvalds return csum; 26371da177e4SLinus Torvalds offset += copy; 26381da177e4SLinus Torvalds to += copy; 26391da177e4SLinus Torvalds } 26401a028e50SDavid S. Miller start = end; 26411da177e4SLinus Torvalds } 26421da177e4SLinus Torvalds 2643fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 264481d77662SAl Viro __wsum csum2; 26451a028e50SDavid S. Miller int end; 26461da177e4SLinus Torvalds 2647547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26481a028e50SDavid S. Miller 2649fbb398a8SDavid S. Miller end = start + frag_iter->len; 26501da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 26511da177e4SLinus Torvalds if (copy > len) 26521da177e4SLinus Torvalds copy = len; 2653fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 26541a028e50SDavid S. Miller offset - start, 26551da177e4SLinus Torvalds to, copy, 0); 26561da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 26571da177e4SLinus Torvalds if ((len -= copy) == 0) 26581da177e4SLinus Torvalds return csum; 26591da177e4SLinus Torvalds offset += copy; 26601da177e4SLinus Torvalds to += copy; 26611da177e4SLinus Torvalds pos += copy; 26621da177e4SLinus Torvalds } 26631a028e50SDavid S. Miller start = end; 26641da177e4SLinus Torvalds } 266509a62660SKris Katterjohn BUG_ON(len); 26661da177e4SLinus Torvalds return csum; 26671da177e4SLinus Torvalds } 2668b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 26691da177e4SLinus Torvalds 267049f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 267149f8e832SCong Wang { 267249f8e832SCong Wang __sum16 sum; 267349f8e832SCong Wang 267449f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 267514641931SCong Wang /* See comments in __skb_checksum_complete(). */ 267649f8e832SCong Wang if (likely(!sum)) { 267749f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 267849f8e832SCong Wang !skb->csum_complete_sw) 26797fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 268049f8e832SCong Wang } 268149f8e832SCong Wang if (!skb_shared(skb)) 268249f8e832SCong Wang skb->csum_valid = !sum; 268349f8e832SCong Wang return sum; 268449f8e832SCong Wang } 268549f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 268649f8e832SCong Wang 268714641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 268814641931SCong Wang * which has been changed from the hardware checksum, for example, by 268914641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 269014641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 269114641931SCong Wang * 269214641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 269314641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 269414641931SCong Wang * shared. 269514641931SCong Wang */ 269649f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 269749f8e832SCong Wang { 269849f8e832SCong Wang __wsum csum; 269949f8e832SCong Wang __sum16 sum; 270049f8e832SCong Wang 270149f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 270249f8e832SCong Wang 270349f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 270414641931SCong Wang /* This check is inverted, because we already knew the hardware 270514641931SCong Wang * checksum is invalid before calling this function. So, if the 270614641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 270714641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 270814641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 270914641931SCong Wang * when moving skb->data around. 271014641931SCong Wang */ 271149f8e832SCong Wang if (likely(!sum)) { 271249f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 271349f8e832SCong Wang !skb->csum_complete_sw) 27147fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 271549f8e832SCong Wang } 271649f8e832SCong Wang 271749f8e832SCong Wang if (!skb_shared(skb)) { 271849f8e832SCong Wang /* Save full packet checksum */ 271949f8e832SCong Wang skb->csum = csum; 272049f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 272149f8e832SCong Wang skb->csum_complete_sw = 1; 272249f8e832SCong Wang skb->csum_valid = !sum; 272349f8e832SCong Wang } 272449f8e832SCong Wang 272549f8e832SCong Wang return sum; 272649f8e832SCong Wang } 272749f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 272849f8e832SCong Wang 27299617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 27309617813dSDavide Caratti { 27319617813dSDavide Caratti net_warn_ratelimited( 27329617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 27339617813dSDavide Caratti __func__); 27349617813dSDavide Caratti return 0; 27359617813dSDavide Caratti } 27369617813dSDavide Caratti 27379617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 27389617813dSDavide Caratti int offset, int len) 27399617813dSDavide Caratti { 27409617813dSDavide Caratti net_warn_ratelimited( 27419617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 27429617813dSDavide Caratti __func__); 27439617813dSDavide Caratti return 0; 27449617813dSDavide Caratti } 27459617813dSDavide Caratti 27469617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 27479617813dSDavide Caratti .update = warn_crc32c_csum_update, 27489617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 27499617813dSDavide Caratti }; 27509617813dSDavide Caratti 27519617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 27529617813dSDavide Caratti &default_crc32c_ops; 27539617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 27549617813dSDavide Caratti 2755af2806f8SThomas Graf /** 2756af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2757af2806f8SThomas Graf * @from: source buffer 2758af2806f8SThomas Graf * 2759af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 2760af2806f8SThomas Graf * into skb_zerocopy(). 2761af2806f8SThomas Graf */ 2762af2806f8SThomas Graf unsigned int 2763af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 2764af2806f8SThomas Graf { 2765af2806f8SThomas Graf unsigned int hlen = 0; 2766af2806f8SThomas Graf 2767af2806f8SThomas Graf if (!from->head_frag || 2768af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 2769af2806f8SThomas Graf skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2770af2806f8SThomas Graf hlen = skb_headlen(from); 2771af2806f8SThomas Graf 2772af2806f8SThomas Graf if (skb_has_frag_list(from)) 2773af2806f8SThomas Graf hlen = from->len; 2774af2806f8SThomas Graf 2775af2806f8SThomas Graf return hlen; 2776af2806f8SThomas Graf } 2777af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2778af2806f8SThomas Graf 2779af2806f8SThomas Graf /** 2780af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 2781af2806f8SThomas Graf * @to: destination buffer 27827fceb4deSMasanari Iida * @from: source buffer 2783af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 2784af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 2785af2806f8SThomas Graf * 2786af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 2787af2806f8SThomas Graf * to the frags in the source buffer. 2788af2806f8SThomas Graf * 2789af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2790af2806f8SThomas Graf * headroom in the `to` buffer. 279136d5fe6aSZoltan Kiss * 279236d5fe6aSZoltan Kiss * Return value: 279336d5fe6aSZoltan Kiss * 0: everything is OK 279436d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 279536d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 2796af2806f8SThomas Graf */ 279736d5fe6aSZoltan Kiss int 279836d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2799af2806f8SThomas Graf { 2800af2806f8SThomas Graf int i, j = 0; 2801af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 280236d5fe6aSZoltan Kiss int ret; 2803af2806f8SThomas Graf struct page *page; 2804af2806f8SThomas Graf unsigned int offset; 2805af2806f8SThomas Graf 2806af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 2807af2806f8SThomas Graf 2808af2806f8SThomas Graf /* dont bother with small payloads */ 280936d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 281036d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 2811af2806f8SThomas Graf 2812af2806f8SThomas Graf if (hlen) { 281336d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 281436d5fe6aSZoltan Kiss if (unlikely(ret)) 281536d5fe6aSZoltan Kiss return ret; 2816af2806f8SThomas Graf len -= hlen; 2817af2806f8SThomas Graf } else { 2818af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 2819af2806f8SThomas Graf if (plen) { 2820af2806f8SThomas Graf page = virt_to_head_page(from->head); 2821af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 2822af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 2823af2806f8SThomas Graf get_page(page); 2824af2806f8SThomas Graf j = 1; 2825af2806f8SThomas Graf len -= plen; 2826af2806f8SThomas Graf } 2827af2806f8SThomas Graf } 2828af2806f8SThomas Graf 2829af2806f8SThomas Graf to->truesize += len + plen; 2830af2806f8SThomas Graf to->len += len + plen; 2831af2806f8SThomas Graf to->data_len += len + plen; 2832af2806f8SThomas Graf 283336d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 283436d5fe6aSZoltan Kiss skb_tx_error(from); 283536d5fe6aSZoltan Kiss return -ENOMEM; 283636d5fe6aSZoltan Kiss } 28371f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 283836d5fe6aSZoltan Kiss 2839af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2840af2806f8SThomas Graf if (!len) 2841af2806f8SThomas Graf break; 2842af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2843af2806f8SThomas Graf skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2844af2806f8SThomas Graf len -= skb_shinfo(to)->frags[j].size; 2845af2806f8SThomas Graf skb_frag_ref(to, j); 2846af2806f8SThomas Graf j++; 2847af2806f8SThomas Graf } 2848af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 284936d5fe6aSZoltan Kiss 285036d5fe6aSZoltan Kiss return 0; 2851af2806f8SThomas Graf } 2852af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 2853af2806f8SThomas Graf 28541da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 28551da177e4SLinus Torvalds { 2856d3bc23e7SAl Viro __wsum csum; 28571da177e4SLinus Torvalds long csstart; 28581da177e4SLinus Torvalds 285984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 286055508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 28611da177e4SLinus Torvalds else 28621da177e4SLinus Torvalds csstart = skb_headlen(skb); 28631da177e4SLinus Torvalds 286409a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 28651da177e4SLinus Torvalds 2866d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 28671da177e4SLinus Torvalds 28681da177e4SLinus Torvalds csum = 0; 28691da177e4SLinus Torvalds if (csstart != skb->len) 28701da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 28711da177e4SLinus Torvalds skb->len - csstart, 0); 28721da177e4SLinus Torvalds 287384fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2874ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 28751da177e4SLinus Torvalds 2876d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 28771da177e4SLinus Torvalds } 28781da177e4SLinus Torvalds } 2879b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 28801da177e4SLinus Torvalds 28811da177e4SLinus Torvalds /** 28821da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 28831da177e4SLinus Torvalds * @list: list to dequeue from 28841da177e4SLinus Torvalds * 28851da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 28861da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 28871da177e4SLinus Torvalds * returned or %NULL if the list is empty. 28881da177e4SLinus Torvalds */ 28891da177e4SLinus Torvalds 28901da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 28911da177e4SLinus Torvalds { 28921da177e4SLinus Torvalds unsigned long flags; 28931da177e4SLinus Torvalds struct sk_buff *result; 28941da177e4SLinus Torvalds 28951da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 28961da177e4SLinus Torvalds result = __skb_dequeue(list); 28971da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 28981da177e4SLinus Torvalds return result; 28991da177e4SLinus Torvalds } 2900b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 29011da177e4SLinus Torvalds 29021da177e4SLinus Torvalds /** 29031da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 29041da177e4SLinus Torvalds * @list: list to dequeue from 29051da177e4SLinus Torvalds * 29061da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 29071da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 29081da177e4SLinus Torvalds * returned or %NULL if the list is empty. 29091da177e4SLinus Torvalds */ 29101da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 29111da177e4SLinus Torvalds { 29121da177e4SLinus Torvalds unsigned long flags; 29131da177e4SLinus Torvalds struct sk_buff *result; 29141da177e4SLinus Torvalds 29151da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 29161da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 29171da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 29181da177e4SLinus Torvalds return result; 29191da177e4SLinus Torvalds } 2920b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 29211da177e4SLinus Torvalds 29221da177e4SLinus Torvalds /** 29231da177e4SLinus Torvalds * skb_queue_purge - empty a list 29241da177e4SLinus Torvalds * @list: list to empty 29251da177e4SLinus Torvalds * 29261da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 29271da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 29281da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 29291da177e4SLinus Torvalds */ 29301da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 29311da177e4SLinus Torvalds { 29321da177e4SLinus Torvalds struct sk_buff *skb; 29331da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 29341da177e4SLinus Torvalds kfree_skb(skb); 29351da177e4SLinus Torvalds } 2936b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 29371da177e4SLinus Torvalds 29381da177e4SLinus Torvalds /** 29399f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 29409f5afeaeSYaogong Wang * @root: root of the rbtree to empty 2941385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 29429f5afeaeSYaogong Wang * 29439f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 29449f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 29459f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 29469f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 29479f5afeaeSYaogong Wang */ 2948385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 29499f5afeaeSYaogong Wang { 29507c90584cSEric Dumazet struct rb_node *p = rb_first(root); 2951385114deSPeter Oskolkov unsigned int sum = 0; 29529f5afeaeSYaogong Wang 29537c90584cSEric Dumazet while (p) { 29547c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 29557c90584cSEric Dumazet 29567c90584cSEric Dumazet p = rb_next(p); 29577c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 2958385114deSPeter Oskolkov sum += skb->truesize; 29599f5afeaeSYaogong Wang kfree_skb(skb); 29607c90584cSEric Dumazet } 2961385114deSPeter Oskolkov return sum; 29629f5afeaeSYaogong Wang } 29639f5afeaeSYaogong Wang 29649f5afeaeSYaogong Wang /** 29651da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 29661da177e4SLinus Torvalds * @list: list to use 29671da177e4SLinus Torvalds * @newsk: buffer to queue 29681da177e4SLinus Torvalds * 29691da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 29701da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 29711da177e4SLinus Torvalds * safely. 29721da177e4SLinus Torvalds * 29731da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 29741da177e4SLinus Torvalds */ 29751da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 29761da177e4SLinus Torvalds { 29771da177e4SLinus Torvalds unsigned long flags; 29781da177e4SLinus Torvalds 29791da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 29801da177e4SLinus Torvalds __skb_queue_head(list, newsk); 29811da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 29821da177e4SLinus Torvalds } 2983b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds /** 29861da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 29871da177e4SLinus Torvalds * @list: list to use 29881da177e4SLinus Torvalds * @newsk: buffer to queue 29891da177e4SLinus Torvalds * 29901da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 29911da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 29921da177e4SLinus Torvalds * safely. 29931da177e4SLinus Torvalds * 29941da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 29951da177e4SLinus Torvalds */ 29961da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 29971da177e4SLinus Torvalds { 29981da177e4SLinus Torvalds unsigned long flags; 29991da177e4SLinus Torvalds 30001da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30011da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 30021da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30031da177e4SLinus Torvalds } 3004b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 30058728b834SDavid S. Miller 30061da177e4SLinus Torvalds /** 30071da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 30081da177e4SLinus Torvalds * @skb: buffer to remove 30098728b834SDavid S. Miller * @list: list to use 30101da177e4SLinus Torvalds * 30118728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 30128728b834SDavid S. Miller * function is atomic with respect to other list locked calls 30131da177e4SLinus Torvalds * 30148728b834SDavid S. Miller * You must know what list the SKB is on. 30151da177e4SLinus Torvalds */ 30168728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 30171da177e4SLinus Torvalds { 30181da177e4SLinus Torvalds unsigned long flags; 30191da177e4SLinus Torvalds 30201da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 30218728b834SDavid S. Miller __skb_unlink(skb, list); 30221da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 30231da177e4SLinus Torvalds } 3024b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 30251da177e4SLinus Torvalds 30261da177e4SLinus Torvalds /** 30271da177e4SLinus Torvalds * skb_append - append a buffer 30281da177e4SLinus Torvalds * @old: buffer to insert after 30291da177e4SLinus Torvalds * @newsk: buffer to insert 30308728b834SDavid S. Miller * @list: list to use 30311da177e4SLinus Torvalds * 30321da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 30331da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 30341da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 30351da177e4SLinus Torvalds */ 30368728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 30371da177e4SLinus Torvalds { 30381da177e4SLinus Torvalds unsigned long flags; 30391da177e4SLinus Torvalds 30408728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 30417de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 30428728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 30431da177e4SLinus Torvalds } 3044b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 30451da177e4SLinus Torvalds 30461da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 30471da177e4SLinus Torvalds struct sk_buff* skb1, 30481da177e4SLinus Torvalds const u32 len, const int pos) 30491da177e4SLinus Torvalds { 30501da177e4SLinus Torvalds int i; 30511da177e4SLinus Torvalds 3052d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3053d626f62bSArnaldo Carvalho de Melo pos - len); 30541da177e4SLinus Torvalds /* And move data appendix as is. */ 30551da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 30561da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 30571da177e4SLinus Torvalds 30581da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 30591da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 30601da177e4SLinus Torvalds skb1->data_len = skb->data_len; 30611da177e4SLinus Torvalds skb1->len += skb1->data_len; 30621da177e4SLinus Torvalds skb->data_len = 0; 30631da177e4SLinus Torvalds skb->len = len; 306427a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 30651da177e4SLinus Torvalds } 30661da177e4SLinus Torvalds 30671da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 30681da177e4SLinus Torvalds struct sk_buff* skb1, 30691da177e4SLinus Torvalds const u32 len, int pos) 30701da177e4SLinus Torvalds { 30711da177e4SLinus Torvalds int i, k = 0; 30721da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 30731da177e4SLinus Torvalds 30741da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 30751da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 30761da177e4SLinus Torvalds skb->len = len; 30771da177e4SLinus Torvalds skb->data_len = len - pos; 30781da177e4SLinus Torvalds 30791da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 30809e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 30811da177e4SLinus Torvalds 30821da177e4SLinus Torvalds if (pos + size > len) { 30831da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 30841da177e4SLinus Torvalds 30851da177e4SLinus Torvalds if (pos < len) { 30861da177e4SLinus Torvalds /* Split frag. 30871da177e4SLinus Torvalds * We have two variants in this case: 30881da177e4SLinus Torvalds * 1. Move all the frag to the second 30891da177e4SLinus Torvalds * part, if it is possible. F.e. 30901da177e4SLinus Torvalds * this approach is mandatory for TUX, 30911da177e4SLinus Torvalds * where splitting is expensive. 30921da177e4SLinus Torvalds * 2. Split is accurately. We make this. 30931da177e4SLinus Torvalds */ 3094ea2ab693SIan Campbell skb_frag_ref(skb, i); 30951da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 30969e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 30979e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 30981da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 30991da177e4SLinus Torvalds } 31001da177e4SLinus Torvalds k++; 31011da177e4SLinus Torvalds } else 31021da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 31031da177e4SLinus Torvalds pos += size; 31041da177e4SLinus Torvalds } 31051da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 31061da177e4SLinus Torvalds } 31071da177e4SLinus Torvalds 31081da177e4SLinus Torvalds /** 31091da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 31101da177e4SLinus Torvalds * @skb: the buffer to split 31111da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 31121da177e4SLinus Torvalds * @len: new length for skb 31131da177e4SLinus Torvalds */ 31141da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 31151da177e4SLinus Torvalds { 31161da177e4SLinus Torvalds int pos = skb_headlen(skb); 31171da177e4SLinus Torvalds 3118fff88030SWillem de Bruijn skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags & 3119fff88030SWillem de Bruijn SKBTX_SHARED_FRAG; 31201f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 31211da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 31221da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 31231da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 31241da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 31251da177e4SLinus Torvalds } 3126b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 31271da177e4SLinus Torvalds 31289f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 31299f782db3SIlpo Järvinen * 31309f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 31319f782db3SIlpo Järvinen */ 3132832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3133832d11c5SIlpo Järvinen { 31340ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3135832d11c5SIlpo Järvinen } 3136832d11c5SIlpo Järvinen 3137832d11c5SIlpo Järvinen /** 3138832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3139832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3140832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3141832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3142832d11c5SIlpo Järvinen * 3143832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 314420e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3145832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3146832d11c5SIlpo Järvinen * 3147832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3148832d11c5SIlpo Järvinen * 3149832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3150832d11c5SIlpo Järvinen * to have non-paged data as well. 3151832d11c5SIlpo Järvinen * 3152832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3153832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3154832d11c5SIlpo Järvinen */ 3155832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3156832d11c5SIlpo Järvinen { 3157832d11c5SIlpo Järvinen int from, to, merge, todo; 3158832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 3159832d11c5SIlpo Järvinen 3160832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3161f8071cdeSEric Dumazet 3162f8071cdeSEric Dumazet if (skb_headlen(skb)) 3163f8071cdeSEric Dumazet return 0; 31641f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 31651f8b977aSWillem de Bruijn return 0; 3166832d11c5SIlpo Järvinen 3167832d11c5SIlpo Järvinen todo = shiftlen; 3168832d11c5SIlpo Järvinen from = 0; 3169832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3170832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3171832d11c5SIlpo Järvinen 3172832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3173832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3174832d11c5SIlpo Järvinen */ 3175832d11c5SIlpo Järvinen if (!to || 3176ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3177ea2ab693SIan Campbell fragfrom->page_offset)) { 3178832d11c5SIlpo Järvinen merge = -1; 3179832d11c5SIlpo Järvinen } else { 3180832d11c5SIlpo Järvinen merge = to - 1; 3181832d11c5SIlpo Järvinen 31829e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3183832d11c5SIlpo Järvinen if (todo < 0) { 3184832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3185832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3186832d11c5SIlpo Järvinen return 0; 3187832d11c5SIlpo Järvinen 31889f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 31899f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3190832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3191832d11c5SIlpo Järvinen 31929e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 31939e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3194832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 3195832d11c5SIlpo Järvinen 3196832d11c5SIlpo Järvinen goto onlymerged; 3197832d11c5SIlpo Järvinen } 3198832d11c5SIlpo Järvinen 3199832d11c5SIlpo Järvinen from++; 3200832d11c5SIlpo Järvinen } 3201832d11c5SIlpo Järvinen 3202832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3203832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3204832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3205832d11c5SIlpo Järvinen return 0; 3206832d11c5SIlpo Järvinen 3207832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3208832d11c5SIlpo Järvinen return 0; 3209832d11c5SIlpo Järvinen 3210832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3211832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3212832d11c5SIlpo Järvinen return 0; 3213832d11c5SIlpo Järvinen 3214832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3215832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3216832d11c5SIlpo Järvinen 32179e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3218832d11c5SIlpo Järvinen *fragto = *fragfrom; 32199e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3220832d11c5SIlpo Järvinen from++; 3221832d11c5SIlpo Järvinen to++; 3222832d11c5SIlpo Järvinen 3223832d11c5SIlpo Järvinen } else { 3224ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3225832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 3226832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 32279e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3228832d11c5SIlpo Järvinen 3229832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 32309e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3231832d11c5SIlpo Järvinen todo = 0; 3232832d11c5SIlpo Järvinen 3233832d11c5SIlpo Järvinen to++; 3234832d11c5SIlpo Järvinen break; 3235832d11c5SIlpo Järvinen } 3236832d11c5SIlpo Järvinen } 3237832d11c5SIlpo Järvinen 3238832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3239832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3240832d11c5SIlpo Järvinen 3241832d11c5SIlpo Järvinen if (merge >= 0) { 3242832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3243832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3244832d11c5SIlpo Järvinen 32459e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 3246ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 3247832d11c5SIlpo Järvinen } 3248832d11c5SIlpo Järvinen 3249832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3250832d11c5SIlpo Järvinen to = 0; 3251832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3252832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3253832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3254832d11c5SIlpo Järvinen 3255832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3256832d11c5SIlpo Järvinen 3257832d11c5SIlpo Järvinen onlymerged: 3258832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3259832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3260832d11c5SIlpo Järvinen */ 3261832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3262832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3263832d11c5SIlpo Järvinen 3264832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3265832d11c5SIlpo Järvinen skb->len -= shiftlen; 3266832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3267832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3268832d11c5SIlpo Järvinen tgt->len += shiftlen; 3269832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3270832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3271832d11c5SIlpo Järvinen 3272832d11c5SIlpo Järvinen return shiftlen; 3273832d11c5SIlpo Järvinen } 3274832d11c5SIlpo Järvinen 3275677e90edSThomas Graf /** 3276677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3277677e90edSThomas Graf * @skb: the buffer to read 3278677e90edSThomas Graf * @from: lower offset of data to be read 3279677e90edSThomas Graf * @to: upper offset of data to be read 3280677e90edSThomas Graf * @st: state variable 3281677e90edSThomas Graf * 3282677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3283677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3284677e90edSThomas Graf */ 3285677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3286677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3287677e90edSThomas Graf { 3288677e90edSThomas Graf st->lower_offset = from; 3289677e90edSThomas Graf st->upper_offset = to; 3290677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3291677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3292677e90edSThomas Graf st->frag_data = NULL; 3293677e90edSThomas Graf } 3294b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3295677e90edSThomas Graf 3296677e90edSThomas Graf /** 3297677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3298677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3299677e90edSThomas Graf * @data: destination pointer for data to be returned 3300677e90edSThomas Graf * @st: state variable 3301677e90edSThomas Graf * 3302bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3303677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3304bc32383cSMathias Krause * the head of the data block to @data and returns the length 3305677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3306677e90edSThomas Graf * offset has been reached. 3307677e90edSThomas Graf * 3308677e90edSThomas Graf * The caller is not required to consume all of the data 3309bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3310677e90edSThomas Graf * of bytes already consumed and the next call to 3311677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3312677e90edSThomas Graf * 331325985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3314e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3315677e90edSThomas Graf * reads of potentially non linear data. 3316677e90edSThomas Graf * 3317bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3318677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3319677e90edSThomas Graf * a stack for this purpose. 3320677e90edSThomas Graf */ 3321677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3322677e90edSThomas Graf struct skb_seq_state *st) 3323677e90edSThomas Graf { 3324677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3325677e90edSThomas Graf skb_frag_t *frag; 3326677e90edSThomas Graf 3327aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3328aeb193eaSWedson Almeida Filho if (st->frag_data) { 3329aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3330aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3331aeb193eaSWedson Almeida Filho } 3332677e90edSThomas Graf return 0; 3333aeb193eaSWedson Almeida Filho } 3334677e90edSThomas Graf 3335677e90edSThomas Graf next_skb: 333695e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3337677e90edSThomas Graf 3338995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 333995e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3340677e90edSThomas Graf return block_limit - abs_offset; 3341677e90edSThomas Graf } 3342677e90edSThomas Graf 3343677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3344677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3345677e90edSThomas Graf 3346677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 3347677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 33489e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 3349677e90edSThomas Graf 3350677e90edSThomas Graf if (abs_offset < block_limit) { 3351677e90edSThomas Graf if (!st->frag_data) 335251c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 3353677e90edSThomas Graf 3354677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 3355677e90edSThomas Graf (abs_offset - st->stepped_offset); 3356677e90edSThomas Graf 3357677e90edSThomas Graf return block_limit - abs_offset; 3358677e90edSThomas Graf } 3359677e90edSThomas Graf 3360677e90edSThomas Graf if (st->frag_data) { 336151c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3362677e90edSThomas Graf st->frag_data = NULL; 3363677e90edSThomas Graf } 3364677e90edSThomas Graf 3365677e90edSThomas Graf st->frag_idx++; 33669e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 3367677e90edSThomas Graf } 3368677e90edSThomas Graf 33695b5a60daSOlaf Kirch if (st->frag_data) { 337051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 33715b5a60daSOlaf Kirch st->frag_data = NULL; 33725b5a60daSOlaf Kirch } 33735b5a60daSOlaf Kirch 337421dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3375677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 337695e3b24cSHerbert Xu st->frag_idx = 0; 3377677e90edSThomas Graf goto next_skb; 337871b3346dSShyam Iyer } else if (st->cur_skb->next) { 337971b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 338071b3346dSShyam Iyer st->frag_idx = 0; 3381677e90edSThomas Graf goto next_skb; 3382677e90edSThomas Graf } 3383677e90edSThomas Graf 3384677e90edSThomas Graf return 0; 3385677e90edSThomas Graf } 3386b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3387677e90edSThomas Graf 3388677e90edSThomas Graf /** 3389677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3390677e90edSThomas Graf * @st: state variable 3391677e90edSThomas Graf * 3392677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3393677e90edSThomas Graf * returned 0. 3394677e90edSThomas Graf */ 3395677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3396677e90edSThomas Graf { 3397677e90edSThomas Graf if (st->frag_data) 339851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3399677e90edSThomas Graf } 3400b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3401677e90edSThomas Graf 34023fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 34033fc7e8a6SThomas Graf 34043fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 34053fc7e8a6SThomas Graf struct ts_config *conf, 34063fc7e8a6SThomas Graf struct ts_state *state) 34073fc7e8a6SThomas Graf { 34083fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 34093fc7e8a6SThomas Graf } 34103fc7e8a6SThomas Graf 34113fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 34123fc7e8a6SThomas Graf { 34133fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 34143fc7e8a6SThomas Graf } 34153fc7e8a6SThomas Graf 34163fc7e8a6SThomas Graf /** 34173fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 34183fc7e8a6SThomas Graf * @skb: the buffer to look in 34193fc7e8a6SThomas Graf * @from: search offset 34203fc7e8a6SThomas Graf * @to: search limit 34213fc7e8a6SThomas Graf * @config: textsearch configuration 34223fc7e8a6SThomas Graf * 34233fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 34243fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 34253fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 34263fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 34273fc7e8a6SThomas Graf */ 34283fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3429059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 34303fc7e8a6SThomas Graf { 3431059a2440SBojan Prtvar struct ts_state state; 3432f72b948dSPhil Oester unsigned int ret; 3433f72b948dSPhil Oester 34343fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 34353fc7e8a6SThomas Graf config->finish = skb_ts_finish; 34363fc7e8a6SThomas Graf 3437059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 34383fc7e8a6SThomas Graf 3439059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3440f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 34413fc7e8a6SThomas Graf } 3442b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 34433fc7e8a6SThomas Graf 3444be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3445be12a1feSHannes Frederic Sowa int offset, size_t size) 3446be12a1feSHannes Frederic Sowa { 3447be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3448be12a1feSHannes Frederic Sowa 3449be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3450be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3451be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3452be12a1feSHannes Frederic Sowa get_page(page); 3453be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3454be12a1feSHannes Frederic Sowa } else { 3455be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3456be12a1feSHannes Frederic Sowa } 3457be12a1feSHannes Frederic Sowa 3458be12a1feSHannes Frederic Sowa return 0; 3459be12a1feSHannes Frederic Sowa } 3460be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3461be12a1feSHannes Frederic Sowa 3462cbb042f9SHerbert Xu /** 3463cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3464cbb042f9SHerbert Xu * @skb: buffer to update 3465cbb042f9SHerbert Xu * @len: length of data pulled 3466cbb042f9SHerbert Xu * 3467cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3468fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 346984fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 347084fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 347184fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3472cbb042f9SHerbert Xu */ 3473af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3474cbb042f9SHerbert Xu { 347531b33dfbSPravin B Shelar unsigned char *data = skb->data; 347631b33dfbSPravin B Shelar 3477cbb042f9SHerbert Xu BUG_ON(len > skb->len); 347831b33dfbSPravin B Shelar __skb_pull(skb, len); 347931b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 348031b33dfbSPravin B Shelar return skb->data; 3481cbb042f9SHerbert Xu } 3482f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3483f94691acSArnaldo Carvalho de Melo 348413acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 348513acc94eSYonghong Song { 348613acc94eSYonghong Song skb_frag_t head_frag; 348713acc94eSYonghong Song struct page *page; 348813acc94eSYonghong Song 348913acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 349013acc94eSYonghong Song head_frag.page.p = page; 349113acc94eSYonghong Song head_frag.page_offset = frag_skb->data - 349213acc94eSYonghong Song (unsigned char *)page_address(page); 349313acc94eSYonghong Song head_frag.size = skb_headlen(frag_skb); 349413acc94eSYonghong Song return head_frag; 349513acc94eSYonghong Song } 349613acc94eSYonghong Song 3497f4c50d99SHerbert Xu /** 3498f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3499df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3500576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3501f4c50d99SHerbert Xu * 3502f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 35034c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 35044c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3505f4c50d99SHerbert Xu */ 3506df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3507df5771ffSMichael S. Tsirkin netdev_features_t features) 3508f4c50d99SHerbert Xu { 3509f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3510f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 35111a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3512df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3513df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3514df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 35151fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3516f4c50d99SHerbert Xu unsigned int offset = doffset; 3517df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3518802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3519f4c50d99SHerbert Xu unsigned int headroom; 3520802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3521ec5f0615SPravin B Shelar __be16 proto; 352236c98382SAlexander Duyck bool csum, sg; 3523df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3524f4c50d99SHerbert Xu int err = -ENOMEM; 3525f4c50d99SHerbert Xu int i = 0; 3526f4c50d99SHerbert Xu int pos; 352753d6471cSVlad Yasevich int dummy; 3528f4c50d99SHerbert Xu 35295882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 353053d6471cSVlad Yasevich proto = skb_network_protocol(head_skb, &dummy); 3531ec5f0615SPravin B Shelar if (unlikely(!proto)) 3532ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3533ec5f0615SPravin B Shelar 353436c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3535f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 35367e2b10c1STom Herbert 353707b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 353807b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 353907b26c94SSteffen Klassert struct sk_buff *iter; 354043170c4eSIlan Tayari unsigned int frag_len; 354107b26c94SSteffen Klassert 354207b26c94SSteffen Klassert if (!list_skb || 354307b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 354407b26c94SSteffen Klassert goto normal; 354507b26c94SSteffen Klassert 354643170c4eSIlan Tayari /* If we get here then all the required 354743170c4eSIlan Tayari * GSO features except frag_list are supported. 354843170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 354943170c4eSIlan Tayari * with no frag_list. 355043170c4eSIlan Tayari * Currently we can do that only when the buffers don't 355143170c4eSIlan Tayari * have a linear part and all the buffers except 355243170c4eSIlan Tayari * the last are of the same length. 355307b26c94SSteffen Klassert */ 355443170c4eSIlan Tayari frag_len = list_skb->len; 355507b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 355643170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 355743170c4eSIlan Tayari goto normal; 3558eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 355907b26c94SSteffen Klassert goto normal; 356007b26c94SSteffen Klassert 356107b26c94SSteffen Klassert len -= iter->len; 356207b26c94SSteffen Klassert } 356343170c4eSIlan Tayari 356443170c4eSIlan Tayari if (len != frag_len) 356543170c4eSIlan Tayari goto normal; 356607b26c94SSteffen Klassert } 356707b26c94SSteffen Klassert 3568802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 3569802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 3570802ab55aSAlexander Duyck * now. 3571802ab55aSAlexander Duyck */ 3572802ab55aSAlexander Duyck partial_segs = len / mss; 3573d7fb5a80SAlexander Duyck if (partial_segs > 1) 3574802ab55aSAlexander Duyck mss *= partial_segs; 3575d7fb5a80SAlexander Duyck else 3576d7fb5a80SAlexander Duyck partial_segs = 0; 3577802ab55aSAlexander Duyck } 3578802ab55aSAlexander Duyck 357907b26c94SSteffen Klassert normal: 3580df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 3581df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 3582f4c50d99SHerbert Xu 3583f4c50d99SHerbert Xu do { 3584f4c50d99SHerbert Xu struct sk_buff *nskb; 35858cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 3586c8884eddSHerbert Xu int hsize; 3587f4c50d99SHerbert Xu int size; 3588f4c50d99SHerbert Xu 35893953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 35903953c46cSMarcelo Ricardo Leitner len = list_skb->len; 35913953c46cSMarcelo Ricardo Leitner } else { 3592df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 3593f4c50d99SHerbert Xu if (len > mss) 3594f4c50d99SHerbert Xu len = mss; 35953953c46cSMarcelo Ricardo Leitner } 3596f4c50d99SHerbert Xu 3597df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 3598f4c50d99SHerbert Xu if (hsize < 0) 3599f4c50d99SHerbert Xu hsize = 0; 3600c8884eddSHerbert Xu if (hsize > len || !sg) 3601c8884eddSHerbert Xu hsize = len; 3602f4c50d99SHerbert Xu 36031a4cedafSMichael S. Tsirkin if (!hsize && i >= nfrags && skb_headlen(list_skb) && 36041a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 36051a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 360689319d38SHerbert Xu 36079d8506ccSHerbert Xu i = 0; 36081a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 36091a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 36101fd819ecSMichael S. Tsirkin frag_skb = list_skb; 36111a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 36129d8506ccSHerbert Xu 36139d8506ccSHerbert Xu while (pos < offset + len) { 36149d8506ccSHerbert Xu BUG_ON(i >= nfrags); 36159d8506ccSHerbert Xu 36164e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 36179d8506ccSHerbert Xu if (pos + size > offset + len) 36189d8506ccSHerbert Xu break; 36199d8506ccSHerbert Xu 36209d8506ccSHerbert Xu i++; 36219d8506ccSHerbert Xu pos += size; 36224e1beba1SMichael S. Tsirkin frag++; 36239d8506ccSHerbert Xu } 36249d8506ccSHerbert Xu 36251a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 36261a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 362789319d38SHerbert Xu 3628f4c50d99SHerbert Xu if (unlikely(!nskb)) 3629f4c50d99SHerbert Xu goto err; 3630f4c50d99SHerbert Xu 36319d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 36329d8506ccSHerbert Xu kfree_skb(nskb); 36339d8506ccSHerbert Xu goto err; 36349d8506ccSHerbert Xu } 36359d8506ccSHerbert Xu 3636ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 363789319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 363889319d38SHerbert Xu kfree_skb(nskb); 363989319d38SHerbert Xu goto err; 364089319d38SHerbert Xu } 364189319d38SHerbert Xu 3642ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 364389319d38SHerbert Xu skb_release_head_state(nskb); 364489319d38SHerbert Xu __skb_push(nskb, doffset); 364589319d38SHerbert Xu } else { 3646c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 3647df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3648c93bdd0eSMel Gorman NUMA_NO_NODE); 364989319d38SHerbert Xu 365089319d38SHerbert Xu if (unlikely(!nskb)) 365189319d38SHerbert Xu goto err; 365289319d38SHerbert Xu 365389319d38SHerbert Xu skb_reserve(nskb, headroom); 365489319d38SHerbert Xu __skb_put(nskb, doffset); 365589319d38SHerbert Xu } 365689319d38SHerbert Xu 3657f4c50d99SHerbert Xu if (segs) 3658f4c50d99SHerbert Xu tail->next = nskb; 3659f4c50d99SHerbert Xu else 3660f4c50d99SHerbert Xu segs = nskb; 3661f4c50d99SHerbert Xu tail = nskb; 3662f4c50d99SHerbert Xu 3663df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 3664f4c50d99SHerbert Xu 3665030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3666fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 366768c33163SPravin B Shelar 3668df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 366968c33163SPravin B Shelar nskb->data - tnl_hlen, 367068c33163SPravin B Shelar doffset + tnl_hlen); 367189319d38SHerbert Xu 36729d8506ccSHerbert Xu if (nskb->len == len + doffset) 36731cdbcb79SSimon Horman goto perform_csum_check; 367489319d38SHerbert Xu 36757fbeffedSAlexander Duyck if (!sg) { 36767fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 36776f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 367876443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 367976443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 3680f4c50d99SHerbert Xu skb_put(nskb, len), 3681f4c50d99SHerbert Xu len, 0); 36827e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 3683de843723STom Herbert skb_headroom(nskb) + doffset; 3684f4c50d99SHerbert Xu continue; 3685f4c50d99SHerbert Xu } 3686f4c50d99SHerbert Xu 36878cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 3688f4c50d99SHerbert Xu 3689df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 3690d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 3691f4c50d99SHerbert Xu 3692fff88030SWillem de Bruijn skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags & 3693df5771ffSMichael S. Tsirkin SKBTX_SHARED_FRAG; 3694cef401deSEric Dumazet 3695bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 3696bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 3697bf5c25d6SWillem de Bruijn goto err; 3698bf5c25d6SWillem de Bruijn 36999d8506ccSHerbert Xu while (pos < offset + len) { 37009d8506ccSHerbert Xu if (i >= nfrags) { 37019d8506ccSHerbert Xu i = 0; 37021a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 37031a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 37041fd819ecSMichael S. Tsirkin frag_skb = list_skb; 370513acc94eSYonghong Song if (!skb_headlen(list_skb)) { 37069d8506ccSHerbert Xu BUG_ON(!nfrags); 370713acc94eSYonghong Song } else { 370813acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 37099d8506ccSHerbert Xu 371013acc94eSYonghong Song /* to make room for head_frag. */ 371113acc94eSYonghong Song i--; 371213acc94eSYonghong Song frag--; 371313acc94eSYonghong Song } 3714bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 3715bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 3716bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 3717bf5c25d6SWillem de Bruijn goto err; 3718bf5c25d6SWillem de Bruijn 37191a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 37209d8506ccSHerbert Xu } 37219d8506ccSHerbert Xu 37229d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 37239d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 37249d8506ccSHerbert Xu net_warn_ratelimited( 37259d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 37269d8506ccSHerbert Xu pos, mss); 3727ff907a11SEric Dumazet err = -EINVAL; 37289d8506ccSHerbert Xu goto err; 37299d8506ccSHerbert Xu } 37309d8506ccSHerbert Xu 373113acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 37328cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 37338cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 3734f4c50d99SHerbert Xu 3735f4c50d99SHerbert Xu if (pos < offset) { 37368cb19905SMichael S. Tsirkin nskb_frag->page_offset += offset - pos; 37378cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 3738f4c50d99SHerbert Xu } 3739f4c50d99SHerbert Xu 374089319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 3741f4c50d99SHerbert Xu 3742f4c50d99SHerbert Xu if (pos + size <= offset + len) { 3743f4c50d99SHerbert Xu i++; 37444e1beba1SMichael S. Tsirkin frag++; 3745f4c50d99SHerbert Xu pos += size; 3746f4c50d99SHerbert Xu } else { 37478cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 374889319d38SHerbert Xu goto skip_fraglist; 3749f4c50d99SHerbert Xu } 3750f4c50d99SHerbert Xu 37518cb19905SMichael S. Tsirkin nskb_frag++; 3752f4c50d99SHerbert Xu } 3753f4c50d99SHerbert Xu 375489319d38SHerbert Xu skip_fraglist: 3755f4c50d99SHerbert Xu nskb->data_len = len - hsize; 3756f4c50d99SHerbert Xu nskb->len += nskb->data_len; 3757f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 3758ec5f0615SPravin B Shelar 37591cdbcb79SSimon Horman perform_csum_check: 37607fbeffedSAlexander Duyck if (!csum) { 3761ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 3762ff907a11SEric Dumazet __skb_linearize(nskb)) 3763ddff00d4SAlexander Duyck goto err; 3764ff907a11SEric Dumazet 37657fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 3766ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 376776443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 376876443456SAlexander Duyck skb_checksum(nskb, doffset, 376976443456SAlexander Duyck nskb->len - doffset, 0); 37707e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 37717e2b10c1STom Herbert skb_headroom(nskb) + doffset; 3772ec5f0615SPravin B Shelar } 3773df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 3774f4c50d99SHerbert Xu 3775bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 3776bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 3777bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 3778bec3cfdcSEric Dumazet */ 3779bec3cfdcSEric Dumazet segs->prev = tail; 3780432c856fSToshiaki Makita 3781802ab55aSAlexander Duyck if (partial_segs) { 378207b26c94SSteffen Klassert struct sk_buff *iter; 3783802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 378407b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 3785802ab55aSAlexander Duyck 3786802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 378707b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 3788802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 3789802ab55aSAlexander Duyck 3790802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 3791802ab55aSAlexander Duyck * our way back down the stack of protocols. 3792802ab55aSAlexander Duyck */ 379307b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 379407b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 379507b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 379607b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 379707b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 379807b26c94SSteffen Klassert } 379907b26c94SSteffen Klassert 380007b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 380107b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 380207b26c94SSteffen Klassert else if (tail != segs) 380307b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 3804802ab55aSAlexander Duyck } 3805802ab55aSAlexander Duyck 3806432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 3807432c856fSToshiaki Makita * using skb_set_owner_w(). 3808432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 3809432c856fSToshiaki Makita */ 3810432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 3811432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 3812432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 3813432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 3814432c856fSToshiaki Makita } 3815f4c50d99SHerbert Xu return segs; 3816f4c50d99SHerbert Xu 3817f4c50d99SHerbert Xu err: 3818289dccbeSEric Dumazet kfree_skb_list(segs); 3819f4c50d99SHerbert Xu return ERR_PTR(err); 3820f4c50d99SHerbert Xu } 3821f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 3822f4c50d99SHerbert Xu 3823d4546c25SDavid Miller int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb) 382471d93b39SHerbert Xu { 38258a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 382667147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 382767147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 38288a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 3829715dc1f3SEric Dumazet unsigned int delta_truesize; 3830d4546c25SDavid Miller struct sk_buff *lp; 383171d93b39SHerbert Xu 38320ab03f35SSteffen Klassert if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush)) 383371d93b39SHerbert Xu return -E2BIG; 383471d93b39SHerbert Xu 383529e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 38368a29111cSEric Dumazet pinfo = skb_shinfo(lp); 38378a29111cSEric Dumazet 38388a29111cSEric Dumazet if (headlen <= offset) { 383942da6994SHerbert Xu skb_frag_t *frag; 384066e92fcfSHerbert Xu skb_frag_t *frag2; 38419aaa156cSHerbert Xu int i = skbinfo->nr_frags; 38429aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 384342da6994SHerbert Xu 384466e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 38458a29111cSEric Dumazet goto merge; 384681705ad1SHerbert Xu 38478a29111cSEric Dumazet offset -= headlen; 38489aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 38499aaa156cSHerbert Xu skbinfo->nr_frags = 0; 3850f5572068SHerbert Xu 38519aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 38529aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 385366e92fcfSHerbert Xu do { 385466e92fcfSHerbert Xu *--frag = *--frag2; 385566e92fcfSHerbert Xu } while (--i); 385666e92fcfSHerbert Xu 385766e92fcfSHerbert Xu frag->page_offset += offset; 38589e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 385966e92fcfSHerbert Xu 3860715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3861ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3862ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3863715dc1f3SEric Dumazet 3864f5572068SHerbert Xu skb->truesize -= skb->data_len; 3865f5572068SHerbert Xu skb->len -= skb->data_len; 3866f5572068SHerbert Xu skb->data_len = 0; 3867f5572068SHerbert Xu 3868715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 38695d38a079SHerbert Xu goto done; 3870d7e8883cSEric Dumazet } else if (skb->head_frag) { 3871d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3872d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3873d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3874d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3875d7e8883cSEric Dumazet unsigned int first_offset; 3876d7e8883cSEric Dumazet 3877d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 38788a29111cSEric Dumazet goto merge; 3879d7e8883cSEric Dumazet 3880d7e8883cSEric Dumazet first_offset = skb->data - 3881d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3882d7e8883cSEric Dumazet offset; 3883d7e8883cSEric Dumazet 3884d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3885d7e8883cSEric Dumazet 3886d7e8883cSEric Dumazet frag->page.p = page; 3887d7e8883cSEric Dumazet frag->page_offset = first_offset; 3888d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3889d7e8883cSEric Dumazet 3890d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3891d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3892d7e8883cSEric Dumazet 3893715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3894d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3895d7e8883cSEric Dumazet goto done; 38968a29111cSEric Dumazet } 389771d93b39SHerbert Xu 389871d93b39SHerbert Xu merge: 3899715dc1f3SEric Dumazet delta_truesize = skb->truesize; 390067147ba9SHerbert Xu if (offset > headlen) { 3901d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3902d1dc7abfSMichal Schmidt 3903d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 39049e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3905d1dc7abfSMichal Schmidt skb->data_len -= eat; 3906d1dc7abfSMichal Schmidt skb->len -= eat; 390767147ba9SHerbert Xu offset = headlen; 390856035022SHerbert Xu } 390956035022SHerbert Xu 391067147ba9SHerbert Xu __skb_pull(skb, offset); 391156035022SHerbert Xu 391229e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 39138a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 39148a29111cSEric Dumazet else 3915c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3916c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 3917f4a775d1SEric Dumazet __skb_header_release(skb); 39188a29111cSEric Dumazet lp = p; 391971d93b39SHerbert Xu 39205d38a079SHerbert Xu done: 39215d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 392237fe4732SHerbert Xu p->data_len += len; 3923715dc1f3SEric Dumazet p->truesize += delta_truesize; 392437fe4732SHerbert Xu p->len += len; 39258a29111cSEric Dumazet if (lp != p) { 39268a29111cSEric Dumazet lp->data_len += len; 39278a29111cSEric Dumazet lp->truesize += delta_truesize; 39288a29111cSEric Dumazet lp->len += len; 39298a29111cSEric Dumazet } 393071d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 393171d93b39SHerbert Xu return 0; 393271d93b39SHerbert Xu } 393357c05650SMarcelo Ricardo Leitner EXPORT_SYMBOL_GPL(skb_gro_receive); 393471d93b39SHerbert Xu 3935df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 3936df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 3937df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 3938df5042f4SFlorian Westphal 3939df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 3940df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3941df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 3942df5042f4SFlorian Westphal #endif 39434165079bSFlorian Westphal #ifdef CONFIG_XFRM 39444165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 39454165079bSFlorian Westphal #endif 3946df5042f4SFlorian Westphal }; 3947df5042f4SFlorian Westphal 3948df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 3949df5042f4SFlorian Westphal { 3950df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 3951df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3952df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 3953df5042f4SFlorian Westphal #endif 39544165079bSFlorian Westphal #ifdef CONFIG_XFRM 39554165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 39564165079bSFlorian Westphal #endif 3957df5042f4SFlorian Westphal 0; 3958df5042f4SFlorian Westphal } 3959df5042f4SFlorian Westphal 3960df5042f4SFlorian Westphal static void skb_extensions_init(void) 3961df5042f4SFlorian Westphal { 3962df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 3963df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 3964df5042f4SFlorian Westphal 3965df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 3966df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 3967df5042f4SFlorian Westphal 0, 3968df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 3969df5042f4SFlorian Westphal NULL); 3970df5042f4SFlorian Westphal } 3971df5042f4SFlorian Westphal #else 3972df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 3973df5042f4SFlorian Westphal #endif 3974df5042f4SFlorian Westphal 39751da177e4SLinus Torvalds void __init skb_init(void) 39761da177e4SLinus Torvalds { 397779a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 39781da177e4SLinus Torvalds sizeof(struct sk_buff), 39791da177e4SLinus Torvalds 0, 3980e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 398179a8a642SKees Cook offsetof(struct sk_buff, cb), 398279a8a642SKees Cook sizeof_field(struct sk_buff, cb), 398320c2df83SPaul Mundt NULL); 3984d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3985d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 3986d179cd12SDavid S. Miller 0, 3987e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 398820c2df83SPaul Mundt NULL); 3989df5042f4SFlorian Westphal skb_extensions_init(); 39901da177e4SLinus Torvalds } 39911da177e4SLinus Torvalds 399251c739d1SDavid S. Miller static int 399348a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 399448a1df65SJason A. Donenfeld unsigned int recursion_level) 3995716ea3a7SDavid Howells { 39961a028e50SDavid S. Miller int start = skb_headlen(skb); 39971a028e50SDavid S. Miller int i, copy = start - offset; 3998fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3999716ea3a7SDavid Howells int elt = 0; 4000716ea3a7SDavid Howells 400148a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 400248a1df65SJason A. Donenfeld return -EMSGSIZE; 400348a1df65SJason A. Donenfeld 4004716ea3a7SDavid Howells if (copy > 0) { 4005716ea3a7SDavid Howells if (copy > len) 4006716ea3a7SDavid Howells copy = len; 4007642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4008716ea3a7SDavid Howells elt++; 4009716ea3a7SDavid Howells if ((len -= copy) == 0) 4010716ea3a7SDavid Howells return elt; 4011716ea3a7SDavid Howells offset += copy; 4012716ea3a7SDavid Howells } 4013716ea3a7SDavid Howells 4014716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 40151a028e50SDavid S. Miller int end; 4016716ea3a7SDavid Howells 4017547b792cSIlpo Järvinen WARN_ON(start > offset + len); 40181a028e50SDavid S. Miller 40199e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4020716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4021716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 402248a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 402348a1df65SJason A. Donenfeld return -EMSGSIZE; 4024716ea3a7SDavid Howells 4025716ea3a7SDavid Howells if (copy > len) 4026716ea3a7SDavid Howells copy = len; 4027ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4028642f1490SJens Axboe frag->page_offset+offset-start); 4029716ea3a7SDavid Howells elt++; 4030716ea3a7SDavid Howells if (!(len -= copy)) 4031716ea3a7SDavid Howells return elt; 4032716ea3a7SDavid Howells offset += copy; 4033716ea3a7SDavid Howells } 40341a028e50SDavid S. Miller start = end; 4035716ea3a7SDavid Howells } 4036716ea3a7SDavid Howells 4037fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 403848a1df65SJason A. Donenfeld int end, ret; 4039716ea3a7SDavid Howells 4040547b792cSIlpo Järvinen WARN_ON(start > offset + len); 40411a028e50SDavid S. Miller 4042fbb398a8SDavid S. Miller end = start + frag_iter->len; 4043716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 404448a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 404548a1df65SJason A. Donenfeld return -EMSGSIZE; 404648a1df65SJason A. Donenfeld 4047716ea3a7SDavid Howells if (copy > len) 4048716ea3a7SDavid Howells copy = len; 404948a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 405048a1df65SJason A. Donenfeld copy, recursion_level + 1); 405148a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 405248a1df65SJason A. Donenfeld return ret; 405348a1df65SJason A. Donenfeld elt += ret; 4054716ea3a7SDavid Howells if ((len -= copy) == 0) 4055716ea3a7SDavid Howells return elt; 4056716ea3a7SDavid Howells offset += copy; 4057716ea3a7SDavid Howells } 40581a028e50SDavid S. Miller start = end; 4059716ea3a7SDavid Howells } 4060716ea3a7SDavid Howells BUG_ON(len); 4061716ea3a7SDavid Howells return elt; 4062716ea3a7SDavid Howells } 4063716ea3a7SDavid Howells 406448a1df65SJason A. Donenfeld /** 406548a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 406648a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 406748a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 406848a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 406948a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 407048a1df65SJason A. Donenfeld * 407148a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 407248a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 407348a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 407448a1df65SJason A. Donenfeld * could not fit. 407548a1df65SJason A. Donenfeld */ 407648a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 407748a1df65SJason A. Donenfeld { 407848a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 407948a1df65SJason A. Donenfeld 408048a1df65SJason A. Donenfeld if (nsg <= 0) 408148a1df65SJason A. Donenfeld return nsg; 408248a1df65SJason A. Donenfeld 408348a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 408448a1df65SJason A. Donenfeld 408548a1df65SJason A. Donenfeld return nsg; 408648a1df65SJason A. Donenfeld } 408748a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 408848a1df65SJason A. Donenfeld 408925a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 409025a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 409125a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 409225a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 409325a91d8dSFan Du * 409425a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 409525a91d8dSFan Du * 1. sg_init_table 409625a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 409725a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 409825a91d8dSFan Du * 409925a91d8dSFan Du * This is equivalent to: 410025a91d8dSFan Du * 1. sg_init_table 410125a91d8dSFan Du * 2. skb_to_sgvec(payload1) 410225a91d8dSFan Du * 3. sg_unmark_end 410325a91d8dSFan Du * 4. skb_to_sgvec(payload2) 410425a91d8dSFan Du * 410525a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 410625a91d8dSFan Du * is more preferable. 410725a91d8dSFan Du */ 410825a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 410925a91d8dSFan Du int offset, int len) 411025a91d8dSFan Du { 411148a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 411225a91d8dSFan Du } 411325a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 411425a91d8dSFan Du 411551c739d1SDavid S. Miller 411651c739d1SDavid S. Miller 4117716ea3a7SDavid Howells /** 4118716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4119716ea3a7SDavid Howells * @skb: The socket buffer to check. 4120716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4121716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4122716ea3a7SDavid Howells * 4123716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4124716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4125716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4126716ea3a7SDavid Howells * 4127716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4128716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4129716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4130716ea3a7SDavid Howells * 4131716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4132716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4133716ea3a7SDavid Howells */ 4134716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4135716ea3a7SDavid Howells { 4136716ea3a7SDavid Howells int copyflag; 4137716ea3a7SDavid Howells int elt; 4138716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4139716ea3a7SDavid Howells 4140716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4141716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4142716ea3a7SDavid Howells * at the moment even if they are anonymous). 4143716ea3a7SDavid Howells */ 4144716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4145716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 4146716ea3a7SDavid Howells return -ENOMEM; 4147716ea3a7SDavid Howells 4148716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 414921dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4150716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4151716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4152716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4153716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4154716ea3a7SDavid Howells 4155716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4156716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4157716ea3a7SDavid Howells return -ENOMEM; 4158716ea3a7SDavid Howells 4159716ea3a7SDavid Howells /* Voila! */ 4160716ea3a7SDavid Howells *trailer = skb; 4161716ea3a7SDavid Howells return 1; 4162716ea3a7SDavid Howells } 4163716ea3a7SDavid Howells 4164716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4165716ea3a7SDavid Howells 4166716ea3a7SDavid Howells elt = 1; 4167716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4168716ea3a7SDavid Howells copyflag = 0; 4169716ea3a7SDavid Howells 4170716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4171716ea3a7SDavid Howells int ntail = 0; 4172716ea3a7SDavid Howells 4173716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4174716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4175716ea3a7SDavid Howells * after it. */ 4176716ea3a7SDavid Howells 4177716ea3a7SDavid Howells if (skb_shared(skb1)) 4178716ea3a7SDavid Howells copyflag = 1; 4179716ea3a7SDavid Howells 4180716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4181716ea3a7SDavid Howells 4182716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4183716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 418421dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4185716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4186716ea3a7SDavid Howells ntail = tailbits + 128; 4187716ea3a7SDavid Howells } 4188716ea3a7SDavid Howells 4189716ea3a7SDavid Howells if (copyflag || 4190716ea3a7SDavid Howells skb_cloned(skb1) || 4191716ea3a7SDavid Howells ntail || 4192716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 419321dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4194716ea3a7SDavid Howells struct sk_buff *skb2; 4195716ea3a7SDavid Howells 4196716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4197716ea3a7SDavid Howells if (ntail == 0) 4198716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4199716ea3a7SDavid Howells else 4200716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4201716ea3a7SDavid Howells skb_headroom(skb1), 4202716ea3a7SDavid Howells ntail, 4203716ea3a7SDavid Howells GFP_ATOMIC); 4204716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4205716ea3a7SDavid Howells return -ENOMEM; 4206716ea3a7SDavid Howells 4207716ea3a7SDavid Howells if (skb1->sk) 4208716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4209716ea3a7SDavid Howells 4210716ea3a7SDavid Howells /* Looking around. Are we still alive? 4211716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4212716ea3a7SDavid Howells 4213716ea3a7SDavid Howells skb2->next = skb1->next; 4214716ea3a7SDavid Howells *skb_p = skb2; 4215716ea3a7SDavid Howells kfree_skb(skb1); 4216716ea3a7SDavid Howells skb1 = skb2; 4217716ea3a7SDavid Howells } 4218716ea3a7SDavid Howells elt++; 4219716ea3a7SDavid Howells *trailer = skb1; 4220716ea3a7SDavid Howells skb_p = &skb1->next; 4221716ea3a7SDavid Howells } 4222716ea3a7SDavid Howells 4223716ea3a7SDavid Howells return elt; 4224716ea3a7SDavid Howells } 4225b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4226716ea3a7SDavid Howells 4227b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4228b1faf566SEric Dumazet { 4229b1faf566SEric Dumazet struct sock *sk = skb->sk; 4230b1faf566SEric Dumazet 4231b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4232b1faf566SEric Dumazet } 4233b1faf566SEric Dumazet 42348605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 42358605330aSSoheil Hassas Yeganeh { 42368605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 42378605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 42388605330aSSoheil Hassas Yeganeh */ 42398605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 42408605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 42418605330aSSoheil Hassas Yeganeh } 42428605330aSSoheil Hassas Yeganeh 4243b1faf566SEric Dumazet /* 4244b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4245b1faf566SEric Dumazet */ 4246b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4247b1faf566SEric Dumazet { 4248b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 424995c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 4250b1faf566SEric Dumazet return -ENOMEM; 4251b1faf566SEric Dumazet 4252b1faf566SEric Dumazet skb_orphan(skb); 4253b1faf566SEric Dumazet skb->sk = sk; 4254b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4255b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 42568605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4257b1faf566SEric Dumazet 4258abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4259abb57ea4SEric Dumazet skb_dst_force(skb); 4260abb57ea4SEric Dumazet 4261b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4262b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 42636e5d58fdSVinicius Costa Gomes sk->sk_error_report(sk); 4264b1faf566SEric Dumazet return 0; 4265b1faf566SEric Dumazet } 4266b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4267b1faf566SEric Dumazet 426883a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 426983a1a1a7SSoheil Hassas Yeganeh { 427083a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 427183a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 427283a1a1a7SSoheil Hassas Yeganeh } 427383a1a1a7SSoheil Hassas Yeganeh 4274364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4275364a9e93SWillem de Bruijn { 4276364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 427783a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 427883a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4279997d5c3fSEric Dumazet unsigned long flags; 4280364a9e93SWillem de Bruijn 4281997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4282364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 428338b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 428483a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 428538b25793SSoheil Hassas Yeganeh if (icmp_next) 428638b25793SSoheil Hassas Yeganeh sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin; 428738b25793SSoheil Hassas Yeganeh } 4288997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4289364a9e93SWillem de Bruijn 429083a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 429183a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 429283a1a1a7SSoheil Hassas Yeganeh 429383a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4294364a9e93SWillem de Bruijn sk->sk_error_report(sk); 4295364a9e93SWillem de Bruijn 4296364a9e93SWillem de Bruijn return skb; 4297364a9e93SWillem de Bruijn } 4298364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4299364a9e93SWillem de Bruijn 4300cab41c47SAlexander Duyck /** 4301cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4302cab41c47SAlexander Duyck * @skb: the skb to clone 4303cab41c47SAlexander Duyck * 4304cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4305cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4306cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4307cab41c47SAlexander Duyck * 4308cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4309cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4310cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4311cab41c47SAlexander Duyck * the sk_error_queue. 4312cab41c47SAlexander Duyck */ 431362bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 431462bccb8cSAlexander Duyck { 431562bccb8cSAlexander Duyck struct sock *sk = skb->sk; 431662bccb8cSAlexander Duyck struct sk_buff *clone; 431762bccb8cSAlexander Duyck 431841c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 431962bccb8cSAlexander Duyck return NULL; 432062bccb8cSAlexander Duyck 432162bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 432262bccb8cSAlexander Duyck if (!clone) { 432362bccb8cSAlexander Duyck sock_put(sk); 432462bccb8cSAlexander Duyck return NULL; 432562bccb8cSAlexander Duyck } 432662bccb8cSAlexander Duyck 432762bccb8cSAlexander Duyck clone->sk = sk; 432862bccb8cSAlexander Duyck clone->destructor = sock_efree; 432962bccb8cSAlexander Duyck 433062bccb8cSAlexander Duyck return clone; 433162bccb8cSAlexander Duyck } 433262bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 433362bccb8cSAlexander Duyck 433437846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 433537846ef0SAlexander Duyck struct sock *sk, 43364ef1b286SSoheil Hassas Yeganeh int tstype, 43374ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4338ac45f602SPatrick Ohly { 4339ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4340ac45f602SPatrick Ohly int err; 4341ac45f602SPatrick Ohly 43424ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 43434ef1b286SSoheil Hassas Yeganeh 4344ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4345ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4346ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4347ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4348e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 43494ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 43501862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 43514ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 435209c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 4353ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 4354ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 43554ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 43564ed2d765SWillem de Bruijn } 435729030374SEric Dumazet 4358ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 435929030374SEric Dumazet 4360ac45f602SPatrick Ohly if (err) 4361ac45f602SPatrick Ohly kfree_skb(skb); 4362ac45f602SPatrick Ohly } 436337846ef0SAlexander Duyck 4364b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4365b245be1fSWillem de Bruijn { 4366b245be1fSWillem de Bruijn bool ret; 4367b245be1fSWillem de Bruijn 4368b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4369b245be1fSWillem de Bruijn return true; 4370b245be1fSWillem de Bruijn 4371b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4372b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4373b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4374b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4375b245be1fSWillem de Bruijn return ret; 4376b245be1fSWillem de Bruijn } 4377b245be1fSWillem de Bruijn 437837846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 437937846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 438037846ef0SAlexander Duyck { 438137846ef0SAlexander Duyck struct sock *sk = skb->sk; 438237846ef0SAlexander Duyck 4383b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 438435b99dffSWillem de Bruijn goto err; 4385b245be1fSWillem de Bruijn 43869ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 43879ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 43889ac25fc0SEric Dumazet */ 438941c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 439037846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 43914ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 439237846ef0SAlexander Duyck sock_put(sk); 439335b99dffSWillem de Bruijn return; 439437846ef0SAlexander Duyck } 439535b99dffSWillem de Bruijn 439635b99dffSWillem de Bruijn err: 439735b99dffSWillem de Bruijn kfree_skb(skb); 43989ac25fc0SEric Dumazet } 439937846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 440037846ef0SAlexander Duyck 440137846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 440237846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 440337846ef0SAlexander Duyck struct sock *sk, int tstype) 440437846ef0SAlexander Duyck { 440537846ef0SAlexander Duyck struct sk_buff *skb; 44064ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 440737846ef0SAlexander Duyck 44083a8dd971SWillem de Bruijn if (!sk) 44093a8dd971SWillem de Bruijn return; 44103a8dd971SWillem de Bruijn 4411b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4412b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4413b50a5c70SMiroslav Lichvar return; 4414b50a5c70SMiroslav Lichvar 44153a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 44163a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 441737846ef0SAlexander Duyck return; 441837846ef0SAlexander Duyck 44191c885808SFrancis Yan if (tsonly) { 44201c885808SFrancis Yan #ifdef CONFIG_INET 44211c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 44221c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 44234ef1b286SSoheil Hassas Yeganeh sk->sk_type == SOCK_STREAM) { 44241c885808SFrancis Yan skb = tcp_get_timestamping_opt_stats(sk); 44254ef1b286SSoheil Hassas Yeganeh opt_stats = true; 44264ef1b286SSoheil Hassas Yeganeh } else 44271c885808SFrancis Yan #endif 44281c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 44291c885808SFrancis Yan } else { 443037846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 44311c885808SFrancis Yan } 443237846ef0SAlexander Duyck if (!skb) 443337846ef0SAlexander Duyck return; 443437846ef0SAlexander Duyck 443549ca0d8bSWillem de Bruijn if (tsonly) { 4436fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4437fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 443849ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 443949ca0d8bSWillem de Bruijn } 444049ca0d8bSWillem de Bruijn 444149ca0d8bSWillem de Bruijn if (hwtstamps) 444249ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 444349ca0d8bSWillem de Bruijn else 444449ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 444549ca0d8bSWillem de Bruijn 44464ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 444737846ef0SAlexander Duyck } 4448e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4449e7fd2885SWillem de Bruijn 4450e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4451e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4452e7fd2885SWillem de Bruijn { 4453e7fd2885SWillem de Bruijn return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 4454e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4455e7fd2885SWillem de Bruijn } 4456ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4457ac45f602SPatrick Ohly 44586e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 44596e3e939fSJohannes Berg { 44606e3e939fSJohannes Berg struct sock *sk = skb->sk; 44616e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4462dd4f1072SEric Dumazet int err = 1; 44636e3e939fSJohannes Berg 44646e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 44656e3e939fSJohannes Berg skb->wifi_acked = acked; 44666e3e939fSJohannes Berg 44676e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 44686e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 44696e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 44706e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 44716e3e939fSJohannes Berg 4472dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4473dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4474dd4f1072SEric Dumazet */ 447541c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 44766e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4477dd4f1072SEric Dumazet sock_put(sk); 4478dd4f1072SEric Dumazet } 44796e3e939fSJohannes Berg if (err) 44806e3e939fSJohannes Berg kfree_skb(skb); 44816e3e939fSJohannes Berg } 44826e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 44836e3e939fSJohannes Berg 4484f35d9d8aSRusty Russell /** 4485f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4486f35d9d8aSRusty Russell * @skb: the skb to set 4487f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4488f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4489f35d9d8aSRusty Russell * 4490f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4491f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4492f35d9d8aSRusty Russell * 4493f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4494f35d9d8aSRusty Russell * returns false you should drop the packet. 4495f35d9d8aSRusty Russell */ 4496f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4497f35d9d8aSRusty Russell { 449852b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 449952b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 450052b5d6f5SEric Dumazet 450152b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 450252b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 450352b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4504f35d9d8aSRusty Russell return false; 4505f35d9d8aSRusty Russell } 4506f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 450752b5d6f5SEric Dumazet skb->csum_start = csum_start; 4508f35d9d8aSRusty Russell skb->csum_offset = off; 4509e5d5decaSJason Wang skb_set_transport_header(skb, start); 4510f35d9d8aSRusty Russell return true; 4511f35d9d8aSRusty Russell } 4512b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4513f35d9d8aSRusty Russell 4514ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4515ed1f50c3SPaul Durrant unsigned int max) 4516ed1f50c3SPaul Durrant { 4517ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4518ed1f50c3SPaul Durrant return 0; 4519ed1f50c3SPaul Durrant 4520ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4521ed1f50c3SPaul Durrant * won't need to do it again. 4522ed1f50c3SPaul Durrant */ 4523ed1f50c3SPaul Durrant if (max > skb->len) 4524ed1f50c3SPaul Durrant max = skb->len; 4525ed1f50c3SPaul Durrant 4526ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4527ed1f50c3SPaul Durrant return -ENOMEM; 4528ed1f50c3SPaul Durrant 4529ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4530ed1f50c3SPaul Durrant return -EPROTO; 4531ed1f50c3SPaul Durrant 4532ed1f50c3SPaul Durrant return 0; 4533ed1f50c3SPaul Durrant } 4534ed1f50c3SPaul Durrant 4535f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4536f9708b43SJan Beulich 4537f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4538f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4539f9708b43SJan Beulich unsigned int off) 4540f9708b43SJan Beulich { 4541f9708b43SJan Beulich switch (proto) { 4542f9708b43SJan Beulich int err; 4543f9708b43SJan Beulich 4544f9708b43SJan Beulich case IPPROTO_TCP: 4545f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4546f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4547f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4548f9708b43SJan Beulich offsetof(struct tcphdr, 4549f9708b43SJan Beulich check))) 4550f9708b43SJan Beulich err = -EPROTO; 4551f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4552f9708b43SJan Beulich 4553f9708b43SJan Beulich case IPPROTO_UDP: 4554f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4555f9708b43SJan Beulich off + sizeof(struct udphdr)); 4556f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4557f9708b43SJan Beulich offsetof(struct udphdr, 4558f9708b43SJan Beulich check))) 4559f9708b43SJan Beulich err = -EPROTO; 4560f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4561f9708b43SJan Beulich } 4562f9708b43SJan Beulich 4563f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4564f9708b43SJan Beulich } 4565f9708b43SJan Beulich 4566ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4567ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4568ed1f50c3SPaul Durrant */ 4569ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4570ed1f50c3SPaul Durrant 4571f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4572ed1f50c3SPaul Durrant { 4573ed1f50c3SPaul Durrant unsigned int off; 4574ed1f50c3SPaul Durrant bool fragment; 4575f9708b43SJan Beulich __sum16 *csum; 4576ed1f50c3SPaul Durrant int err; 4577ed1f50c3SPaul Durrant 4578ed1f50c3SPaul Durrant fragment = false; 4579ed1f50c3SPaul Durrant 4580ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4581ed1f50c3SPaul Durrant sizeof(struct iphdr), 4582ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4583ed1f50c3SPaul Durrant if (err < 0) 4584ed1f50c3SPaul Durrant goto out; 4585ed1f50c3SPaul Durrant 4586ed1f50c3SPaul Durrant if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 4587ed1f50c3SPaul Durrant fragment = true; 4588ed1f50c3SPaul Durrant 4589ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 4590ed1f50c3SPaul Durrant 4591ed1f50c3SPaul Durrant err = -EPROTO; 4592ed1f50c3SPaul Durrant 4593ed1f50c3SPaul Durrant if (fragment) 4594ed1f50c3SPaul Durrant goto out; 4595ed1f50c3SPaul Durrant 4596f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4597f9708b43SJan Beulich if (IS_ERR(csum)) 4598f9708b43SJan Beulich return PTR_ERR(csum); 4599ed1f50c3SPaul Durrant 4600ed1f50c3SPaul Durrant if (recalculate) 4601f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4602ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4603ed1f50c3SPaul Durrant skb->len - off, 4604f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4605ed1f50c3SPaul Durrant err = 0; 4606ed1f50c3SPaul Durrant 4607ed1f50c3SPaul Durrant out: 4608ed1f50c3SPaul Durrant return err; 4609ed1f50c3SPaul Durrant } 4610ed1f50c3SPaul Durrant 4611ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4612ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4613ed1f50c3SPaul Durrant */ 4614ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4615ed1f50c3SPaul Durrant 4616ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4617ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4618ed1f50c3SPaul Durrant 4619ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4620ed1f50c3SPaul Durrant { 4621ed1f50c3SPaul Durrant int err; 4622ed1f50c3SPaul Durrant u8 nexthdr; 4623ed1f50c3SPaul Durrant unsigned int off; 4624ed1f50c3SPaul Durrant unsigned int len; 4625ed1f50c3SPaul Durrant bool fragment; 4626ed1f50c3SPaul Durrant bool done; 4627f9708b43SJan Beulich __sum16 *csum; 4628ed1f50c3SPaul Durrant 4629ed1f50c3SPaul Durrant fragment = false; 4630ed1f50c3SPaul Durrant done = false; 4631ed1f50c3SPaul Durrant 4632ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4633ed1f50c3SPaul Durrant 4634ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4635ed1f50c3SPaul Durrant if (err < 0) 4636ed1f50c3SPaul Durrant goto out; 4637ed1f50c3SPaul Durrant 4638ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4639ed1f50c3SPaul Durrant 4640ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4641ed1f50c3SPaul Durrant while (off <= len && !done) { 4642ed1f50c3SPaul Durrant switch (nexthdr) { 4643ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4644ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4645ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4646ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 4647ed1f50c3SPaul Durrant 4648ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4649ed1f50c3SPaul Durrant off + 4650ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 4651ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4652ed1f50c3SPaul Durrant if (err < 0) 4653ed1f50c3SPaul Durrant goto out; 4654ed1f50c3SPaul Durrant 4655ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4656ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4657ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 4658ed1f50c3SPaul Durrant break; 4659ed1f50c3SPaul Durrant } 4660ed1f50c3SPaul Durrant case IPPROTO_AH: { 4661ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 4662ed1f50c3SPaul Durrant 4663ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4664ed1f50c3SPaul Durrant off + 4665ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 4666ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4667ed1f50c3SPaul Durrant if (err < 0) 4668ed1f50c3SPaul Durrant goto out; 4669ed1f50c3SPaul Durrant 4670ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4671ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4672ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 4673ed1f50c3SPaul Durrant break; 4674ed1f50c3SPaul Durrant } 4675ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 4676ed1f50c3SPaul Durrant struct frag_hdr *hp; 4677ed1f50c3SPaul Durrant 4678ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4679ed1f50c3SPaul Durrant off + 4680ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 4681ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 4682ed1f50c3SPaul Durrant if (err < 0) 4683ed1f50c3SPaul Durrant goto out; 4684ed1f50c3SPaul Durrant 4685ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 4686ed1f50c3SPaul Durrant 4687ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 4688ed1f50c3SPaul Durrant fragment = true; 4689ed1f50c3SPaul Durrant 4690ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 4691ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 4692ed1f50c3SPaul Durrant break; 4693ed1f50c3SPaul Durrant } 4694ed1f50c3SPaul Durrant default: 4695ed1f50c3SPaul Durrant done = true; 4696ed1f50c3SPaul Durrant break; 4697ed1f50c3SPaul Durrant } 4698ed1f50c3SPaul Durrant } 4699ed1f50c3SPaul Durrant 4700ed1f50c3SPaul Durrant err = -EPROTO; 4701ed1f50c3SPaul Durrant 4702ed1f50c3SPaul Durrant if (!done || fragment) 4703ed1f50c3SPaul Durrant goto out; 4704ed1f50c3SPaul Durrant 4705f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 4706f9708b43SJan Beulich if (IS_ERR(csum)) 4707f9708b43SJan Beulich return PTR_ERR(csum); 4708ed1f50c3SPaul Durrant 4709ed1f50c3SPaul Durrant if (recalculate) 4710f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 4711ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 4712f9708b43SJan Beulich skb->len - off, nexthdr, 0); 4713ed1f50c3SPaul Durrant err = 0; 4714ed1f50c3SPaul Durrant 4715ed1f50c3SPaul Durrant out: 4716ed1f50c3SPaul Durrant return err; 4717ed1f50c3SPaul Durrant } 4718ed1f50c3SPaul Durrant 4719ed1f50c3SPaul Durrant /** 4720ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 4721ed1f50c3SPaul Durrant * @skb: the skb to set up 4722ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 4723ed1f50c3SPaul Durrant */ 4724ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 4725ed1f50c3SPaul Durrant { 4726ed1f50c3SPaul Durrant int err; 4727ed1f50c3SPaul Durrant 4728ed1f50c3SPaul Durrant switch (skb->protocol) { 4729ed1f50c3SPaul Durrant case htons(ETH_P_IP): 4730f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 4731ed1f50c3SPaul Durrant break; 4732ed1f50c3SPaul Durrant 4733ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 4734ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 4735ed1f50c3SPaul Durrant break; 4736ed1f50c3SPaul Durrant 4737ed1f50c3SPaul Durrant default: 4738ed1f50c3SPaul Durrant err = -EPROTO; 4739ed1f50c3SPaul Durrant break; 4740ed1f50c3SPaul Durrant } 4741ed1f50c3SPaul Durrant 4742ed1f50c3SPaul Durrant return err; 4743ed1f50c3SPaul Durrant } 4744ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 4745ed1f50c3SPaul Durrant 47469afd85c9SLinus Lüssing /** 47479afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 47489afd85c9SLinus Lüssing * @skb: the skb to check 47499afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 47509afd85c9SLinus Lüssing * 47519afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 47529afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 47539afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 47549afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 47559afd85c9SLinus Lüssing * 4756a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4757a516993fSLinus Lüssing * differs from the provided skb. 47589afd85c9SLinus Lüssing */ 47599afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 47609afd85c9SLinus Lüssing unsigned int transport_len) 47619afd85c9SLinus Lüssing { 47629afd85c9SLinus Lüssing struct sk_buff *skb_chk; 47639afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 47649afd85c9SLinus Lüssing int ret; 47659afd85c9SLinus Lüssing 4766a516993fSLinus Lüssing if (skb->len < len) 47679afd85c9SLinus Lüssing return NULL; 4768a516993fSLinus Lüssing else if (skb->len == len) 47699afd85c9SLinus Lüssing return skb; 47709afd85c9SLinus Lüssing 47719afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 47729afd85c9SLinus Lüssing if (!skb_chk) 47739afd85c9SLinus Lüssing return NULL; 47749afd85c9SLinus Lüssing 47759afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 47769afd85c9SLinus Lüssing if (ret) { 47779afd85c9SLinus Lüssing kfree_skb(skb_chk); 47789afd85c9SLinus Lüssing return NULL; 47799afd85c9SLinus Lüssing } 47809afd85c9SLinus Lüssing 47819afd85c9SLinus Lüssing return skb_chk; 47829afd85c9SLinus Lüssing } 47839afd85c9SLinus Lüssing 47849afd85c9SLinus Lüssing /** 47859afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 47869afd85c9SLinus Lüssing * @skb: the skb to check 47879afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 47889afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 47899afd85c9SLinus Lüssing * 47909afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 47919afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 47929afd85c9SLinus Lüssing * 47939afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 47949afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 47959afd85c9SLinus Lüssing * 4796a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 4797a516993fSLinus Lüssing * differs from the provided skb. 47989afd85c9SLinus Lüssing */ 47999afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 48009afd85c9SLinus Lüssing unsigned int transport_len, 48019afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 48029afd85c9SLinus Lüssing { 48039afd85c9SLinus Lüssing struct sk_buff *skb_chk; 48049afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 4805fcba67c9SLinus Lüssing __sum16 ret; 48069afd85c9SLinus Lüssing 48079afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 48089afd85c9SLinus Lüssing if (!skb_chk) 4809a516993fSLinus Lüssing goto err; 48109afd85c9SLinus Lüssing 4811a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 4812a516993fSLinus Lüssing goto err; 48139afd85c9SLinus Lüssing 48149b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 48159afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 48169b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 48179afd85c9SLinus Lüssing 4818a516993fSLinus Lüssing if (ret) 4819a516993fSLinus Lüssing goto err; 48209afd85c9SLinus Lüssing 48219afd85c9SLinus Lüssing return skb_chk; 4822a516993fSLinus Lüssing 4823a516993fSLinus Lüssing err: 4824a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 4825a516993fSLinus Lüssing kfree_skb(skb_chk); 4826a516993fSLinus Lüssing 4827a516993fSLinus Lüssing return NULL; 4828a516993fSLinus Lüssing 48299afd85c9SLinus Lüssing } 48309afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 48319afd85c9SLinus Lüssing 48324497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 48334497b076SBen Hutchings { 4834e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 4835e87cc472SJoe Perches skb->dev->name); 48364497b076SBen Hutchings } 48374497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 4838bad43ca8SEric Dumazet 4839bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 4840bad43ca8SEric Dumazet { 48413d861f66SEric Dumazet if (head_stolen) { 48423d861f66SEric Dumazet skb_release_head_state(skb); 4843bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 48443d861f66SEric Dumazet } else { 4845bad43ca8SEric Dumazet __kfree_skb(skb); 4846bad43ca8SEric Dumazet } 48473d861f66SEric Dumazet } 4848bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 4849bad43ca8SEric Dumazet 4850bad43ca8SEric Dumazet /** 4851bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 4852bad43ca8SEric Dumazet * @to: prior buffer 4853bad43ca8SEric Dumazet * @from: buffer to add 4854bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 4855c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 4856bad43ca8SEric Dumazet */ 4857bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 4858bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 4859bad43ca8SEric Dumazet { 4860c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 4861bad43ca8SEric Dumazet int i, delta, len = from->len; 4862bad43ca8SEric Dumazet 4863bad43ca8SEric Dumazet *fragstolen = false; 4864bad43ca8SEric Dumazet 4865bad43ca8SEric Dumazet if (skb_cloned(to)) 4866bad43ca8SEric Dumazet return false; 4867bad43ca8SEric Dumazet 4868bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 4869e93a0435SEric Dumazet if (len) 4870bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 4871bad43ca8SEric Dumazet *delta_truesize = 0; 4872bad43ca8SEric Dumazet return true; 4873bad43ca8SEric Dumazet } 4874bad43ca8SEric Dumazet 4875c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 4876c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 4877c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 4878bad43ca8SEric Dumazet return false; 48791f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 48801f8b977aSWillem de Bruijn return false; 4881bad43ca8SEric Dumazet 4882bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 4883bad43ca8SEric Dumazet struct page *page; 4884bad43ca8SEric Dumazet unsigned int offset; 4885bad43ca8SEric Dumazet 4886c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 4887c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 4888bad43ca8SEric Dumazet return false; 4889bad43ca8SEric Dumazet 4890bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 4891bad43ca8SEric Dumazet return false; 4892bad43ca8SEric Dumazet 4893bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4894bad43ca8SEric Dumazet 4895bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 4896bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 4897bad43ca8SEric Dumazet 4898c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 4899bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 4900bad43ca8SEric Dumazet *fragstolen = true; 4901bad43ca8SEric Dumazet } else { 4902c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 4903c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 4904bad43ca8SEric Dumazet return false; 4905bad43ca8SEric Dumazet 4906f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 4907bad43ca8SEric Dumazet } 4908bad43ca8SEric Dumazet 4909bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 4910bad43ca8SEric Dumazet 4911c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 4912c818fa9eSEric Dumazet from_shinfo->frags, 4913c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 4914c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 4915bad43ca8SEric Dumazet 4916bad43ca8SEric Dumazet if (!skb_cloned(from)) 4917c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 4918bad43ca8SEric Dumazet 49198ea853fdSLi RongQing /* if the skb is not cloned this does nothing 49208ea853fdSLi RongQing * since we set nr_frags to 0. 49218ea853fdSLi RongQing */ 4922c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 4923c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 4924bad43ca8SEric Dumazet 4925bad43ca8SEric Dumazet to->truesize += delta; 4926bad43ca8SEric Dumazet to->len += len; 4927bad43ca8SEric Dumazet to->data_len += len; 4928bad43ca8SEric Dumazet 4929bad43ca8SEric Dumazet *delta_truesize = delta; 4930bad43ca8SEric Dumazet return true; 4931bad43ca8SEric Dumazet } 4932bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 4933621e84d6SNicolas Dichtel 4934621e84d6SNicolas Dichtel /** 49358b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 4936621e84d6SNicolas Dichtel * 4937621e84d6SNicolas Dichtel * @skb: buffer to clean 49388b27f277SNicolas Dichtel * @xnet: packet is crossing netns 4939621e84d6SNicolas Dichtel * 49408b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 49418b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 49428b27f277SNicolas Dichtel * operations. 49438b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 49448b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 49458b27f277SNicolas Dichtel * skb that could impact namespace isolation. 4946621e84d6SNicolas Dichtel */ 49478b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4948621e84d6SNicolas Dichtel { 4949621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 4950621e84d6SNicolas Dichtel skb->skb_iif = 0; 495160ff7467SWANG Cong skb->ignore_df = 0; 4952621e84d6SNicolas Dichtel skb_dst_drop(skb); 4953621e84d6SNicolas Dichtel secpath_reset(skb); 4954621e84d6SNicolas Dichtel nf_reset(skb); 4955621e84d6SNicolas Dichtel nf_reset_trace(skb); 4956213dd74aSHerbert Xu 49576f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 49586f9a5069SPetr Machata skb->offload_fwd_mark = 0; 4959875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 49606f9a5069SPetr Machata #endif 49616f9a5069SPetr Machata 4962213dd74aSHerbert Xu if (!xnet) 4963213dd74aSHerbert Xu return; 4964213dd74aSHerbert Xu 49652b5ec1a5SYe Yin ipvs_reset(skb); 4966213dd74aSHerbert Xu skb->mark = 0; 4967c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 4968621e84d6SNicolas Dichtel } 4969621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 4970de960aa9SFlorian Westphal 4971de960aa9SFlorian Westphal /** 4972de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4973de960aa9SFlorian Westphal * 4974de960aa9SFlorian Westphal * @skb: GSO skb 4975de960aa9SFlorian Westphal * 4976de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 4977de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 4978de960aa9SFlorian Westphal * 4979de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4980de960aa9SFlorian Westphal */ 4981a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4982de960aa9SFlorian Westphal { 4983de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 4984f993bc25SFlorian Westphal unsigned int thlen = 0; 4985f993bc25SFlorian Westphal 4986f993bc25SFlorian Westphal if (skb->encapsulation) { 4987f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 4988f993bc25SFlorian Westphal skb_transport_header(skb); 4989de960aa9SFlorian Westphal 4990de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 4991f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 4992f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 4993f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 49941dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 499590017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 4996ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 4997ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 4998f993bc25SFlorian Westphal } 49996d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 50006d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 50016d39d589SFlorian Westphal * accounted for. 50026d39d589SFlorian Westphal */ 5003f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5004de960aa9SFlorian Westphal } 5005a4a77718SDaniel Axtens 5006a4a77718SDaniel Axtens /** 5007a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5008a4a77718SDaniel Axtens * 5009a4a77718SDaniel Axtens * @skb: GSO skb 5010a4a77718SDaniel Axtens * 5011a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5012a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5013a4a77718SDaniel Axtens * 5014a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5015a4a77718SDaniel Axtens */ 5016a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5017a4a77718SDaniel Axtens { 5018a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5019a4a77718SDaniel Axtens skb_network_header(skb); 5020a4a77718SDaniel Axtens 5021a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5022a4a77718SDaniel Axtens } 5023a4a77718SDaniel Axtens 5024a4a77718SDaniel Axtens /** 5025a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5026a4a77718SDaniel Axtens * 5027a4a77718SDaniel Axtens * @skb: GSO skb 5028a4a77718SDaniel Axtens * 5029a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5030a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5031a4a77718SDaniel Axtens * headers (TCP/UDP). 5032a4a77718SDaniel Axtens */ 5033a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5034a4a77718SDaniel Axtens { 5035a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5036a4a77718SDaniel Axtens 5037a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5038a4a77718SDaniel Axtens } 50390d5501c1SVlad Yasevich 5040ae7ef81eSMarcelo Ricardo Leitner /** 50412b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 50422b16f048SDaniel Axtens * 50432b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 50442b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 50452b16f048SDaniel Axtens * 50462b16f048SDaniel Axtens * We might want to check: 50472b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 50482b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 50492b16f048SDaniel Axtens * 50502b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 50512b16f048SDaniel Axtens * 505249682bfaSMathieu Malaterre * @skb: GSO skb 505349682bfaSMathieu Malaterre * 50542b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 50552b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 50562b16f048SDaniel Axtens * 50572b16f048SDaniel Axtens * @max_len: The maximum permissible length. 50582b16f048SDaniel Axtens * 50592b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 50602b16f048SDaniel Axtens */ 50612b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 50622b16f048SDaniel Axtens unsigned int seg_len, 50632b16f048SDaniel Axtens unsigned int max_len) { 50642b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 50652b16f048SDaniel Axtens const struct sk_buff *iter; 50662b16f048SDaniel Axtens 50672b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 50682b16f048SDaniel Axtens return seg_len <= max_len; 50692b16f048SDaniel Axtens 50702b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 50712b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 50722b16f048SDaniel Axtens 50732b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 50742b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 50752b16f048SDaniel Axtens return false; 50762b16f048SDaniel Axtens } 50772b16f048SDaniel Axtens 50782b16f048SDaniel Axtens return true; 50792b16f048SDaniel Axtens } 50802b16f048SDaniel Axtens 50812b16f048SDaniel Axtens /** 5082779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5083ae7ef81eSMarcelo Ricardo Leitner * 5084ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 508576f21b99SDavid S. Miller * @mtu: MTU to validate against 5086ae7ef81eSMarcelo Ricardo Leitner * 5087779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5088779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5089779b7931SDaniel Axtens * payload. 5090ae7ef81eSMarcelo Ricardo Leitner */ 5091779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5092ae7ef81eSMarcelo Ricardo Leitner { 50932b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5094ae7ef81eSMarcelo Ricardo Leitner } 5095779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5096ae7ef81eSMarcelo Ricardo Leitner 50972b16f048SDaniel Axtens /** 50982b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 50992b16f048SDaniel Axtens * 51002b16f048SDaniel Axtens * @skb: GSO skb 51012b16f048SDaniel Axtens * @len: length to validate against 51022b16f048SDaniel Axtens * 51032b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 51042b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 51052b16f048SDaniel Axtens */ 51062b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 51072b16f048SDaniel Axtens { 51082b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 51092b16f048SDaniel Axtens } 51102b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 51112b16f048SDaniel Axtens 51120d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 51130d5501c1SVlad Yasevich { 5114d85e8be2SYuya Kusakabe int mac_len, meta_len; 5115d85e8be2SYuya Kusakabe void *meta; 51164bbb3e0eSToshiaki Makita 51170d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 51180d5501c1SVlad Yasevich kfree_skb(skb); 51190d5501c1SVlad Yasevich return NULL; 51200d5501c1SVlad Yasevich } 51210d5501c1SVlad Yasevich 51224bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5123ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 51244bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 51254bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5126ae474573SToshiaki Makita } 5127d85e8be2SYuya Kusakabe 5128d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5129d85e8be2SYuya Kusakabe if (meta_len) { 5130d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5131d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5132d85e8be2SYuya Kusakabe } 5133d85e8be2SYuya Kusakabe 51340d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 51350d5501c1SVlad Yasevich return skb; 51360d5501c1SVlad Yasevich } 51370d5501c1SVlad Yasevich 51380d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 51390d5501c1SVlad Yasevich { 51400d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 51410d5501c1SVlad Yasevich u16 vlan_tci; 51420d5501c1SVlad Yasevich 5143df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 51440d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 51450d5501c1SVlad Yasevich return skb; 51460d5501c1SVlad Yasevich } 51470d5501c1SVlad Yasevich 51480d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 51490d5501c1SVlad Yasevich if (unlikely(!skb)) 51500d5501c1SVlad Yasevich goto err_free; 51510d5501c1SVlad Yasevich 51520d5501c1SVlad Yasevich if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 51530d5501c1SVlad Yasevich goto err_free; 51540d5501c1SVlad Yasevich 51550d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 51560d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 51570d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 51580d5501c1SVlad Yasevich 51590d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 51600d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 51610d5501c1SVlad Yasevich 51620d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 51630d5501c1SVlad Yasevich if (unlikely(!skb)) 51640d5501c1SVlad Yasevich goto err_free; 51650d5501c1SVlad Yasevich 51660d5501c1SVlad Yasevich skb_reset_network_header(skb); 51670d5501c1SVlad Yasevich skb_reset_transport_header(skb); 51680d5501c1SVlad Yasevich skb_reset_mac_len(skb); 51690d5501c1SVlad Yasevich 51700d5501c1SVlad Yasevich return skb; 51710d5501c1SVlad Yasevich 51720d5501c1SVlad Yasevich err_free: 51730d5501c1SVlad Yasevich kfree_skb(skb); 51740d5501c1SVlad Yasevich return NULL; 51750d5501c1SVlad Yasevich } 51760d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 51772e4e4410SEric Dumazet 5178e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5179e2195121SJiri Pirko { 5180e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5181e2195121SJiri Pirko return -ENOMEM; 5182e2195121SJiri Pirko 5183e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5184e2195121SJiri Pirko return 0; 5185e2195121SJiri Pirko 5186e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5187e2195121SJiri Pirko } 5188e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5189e2195121SJiri Pirko 5190bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5191bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5192bfca4c52SShmulik Ladkani */ 5193bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 519493515d53SJiri Pirko { 519593515d53SJiri Pirko struct vlan_hdr *vhdr; 5196b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 519793515d53SJiri Pirko int err; 519893515d53SJiri Pirko 5199b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5200b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5201b6a79208SShmulik Ladkani offset)) { 5202b6a79208SShmulik Ladkani return -EINVAL; 5203b6a79208SShmulik Ladkani } 5204b6a79208SShmulik Ladkani 520593515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 520693515d53SJiri Pirko if (unlikely(err)) 5207b6a79208SShmulik Ladkani return err; 520893515d53SJiri Pirko 520993515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 521093515d53SJiri Pirko 521193515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 521293515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 521393515d53SJiri Pirko 521493515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 521593515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 521693515d53SJiri Pirko 521793515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 521893515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 521993515d53SJiri Pirko 522093515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 522193515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 522293515d53SJiri Pirko 522393515d53SJiri Pirko skb_reset_mac_len(skb); 522493515d53SJiri Pirko 522593515d53SJiri Pirko return err; 522693515d53SJiri Pirko } 5227bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 522893515d53SJiri Pirko 5229b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5230b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5231b6a79208SShmulik Ladkani */ 523293515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 523393515d53SJiri Pirko { 523493515d53SJiri Pirko u16 vlan_tci; 523593515d53SJiri Pirko __be16 vlan_proto; 523693515d53SJiri Pirko int err; 523793515d53SJiri Pirko 5238df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5239b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 524093515d53SJiri Pirko } else { 5241ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 524293515d53SJiri Pirko return 0; 524393515d53SJiri Pirko 524493515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 524593515d53SJiri Pirko if (err) 524693515d53SJiri Pirko return err; 524793515d53SJiri Pirko } 524893515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5249ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 525093515d53SJiri Pirko return 0; 525193515d53SJiri Pirko 525293515d53SJiri Pirko vlan_proto = skb->protocol; 525393515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 525493515d53SJiri Pirko if (unlikely(err)) 525593515d53SJiri Pirko return err; 525693515d53SJiri Pirko 525793515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 525893515d53SJiri Pirko return 0; 525993515d53SJiri Pirko } 526093515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 526193515d53SJiri Pirko 5262b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5263b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5264b6a79208SShmulik Ladkani */ 526593515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 526693515d53SJiri Pirko { 5267df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5268b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 526993515d53SJiri Pirko int err; 527093515d53SJiri Pirko 5271b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5272b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5273b6a79208SShmulik Ladkani offset)) { 5274b6a79208SShmulik Ladkani return -EINVAL; 5275b6a79208SShmulik Ladkani } 5276b6a79208SShmulik Ladkani 527793515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5278df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5279b6a79208SShmulik Ladkani if (err) 528093515d53SJiri Pirko return err; 52819241e2dfSDaniel Borkmann 528293515d53SJiri Pirko skb->protocol = skb->vlan_proto; 528393515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 528493515d53SJiri Pirko 52856b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 528693515d53SJiri Pirko } 528793515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 528893515d53SJiri Pirko return 0; 528993515d53SJiri Pirko } 529093515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 529193515d53SJiri Pirko 52922e4e4410SEric Dumazet /** 52932e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 52942e4e4410SEric Dumazet * 5295de3f0d0eSMasanari Iida * @header_len: size of linear part 5296de3f0d0eSMasanari Iida * @data_len: needed length in frags 5297de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5298de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5299de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 53002e4e4410SEric Dumazet * 53012e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 53022e4e4410SEric Dumazet */ 53032e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 53042e4e4410SEric Dumazet unsigned long data_len, 53052e4e4410SEric Dumazet int max_page_order, 53062e4e4410SEric Dumazet int *errcode, 53072e4e4410SEric Dumazet gfp_t gfp_mask) 53082e4e4410SEric Dumazet { 53092e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 53102e4e4410SEric Dumazet unsigned long chunk; 53112e4e4410SEric Dumazet struct sk_buff *skb; 53122e4e4410SEric Dumazet struct page *page; 53132e4e4410SEric Dumazet int i; 53142e4e4410SEric Dumazet 53152e4e4410SEric Dumazet *errcode = -EMSGSIZE; 53162e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 53172e4e4410SEric Dumazet * high order pages... 53182e4e4410SEric Dumazet */ 53192e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 53202e4e4410SEric Dumazet return NULL; 53212e4e4410SEric Dumazet 53222e4e4410SEric Dumazet *errcode = -ENOBUFS; 5323f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 53242e4e4410SEric Dumazet if (!skb) 53252e4e4410SEric Dumazet return NULL; 53262e4e4410SEric Dumazet 53272e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 53282e4e4410SEric Dumazet 53292e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 53302e4e4410SEric Dumazet int order = max_page_order; 53312e4e4410SEric Dumazet 53322e4e4410SEric Dumazet while (order) { 53332e4e4410SEric Dumazet if (npages >= 1 << order) { 5334d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 53352e4e4410SEric Dumazet __GFP_COMP | 5336d14b56f5SMichal Hocko __GFP_NOWARN, 53372e4e4410SEric Dumazet order); 53382e4e4410SEric Dumazet if (page) 53392e4e4410SEric Dumazet goto fill_page; 53402e4e4410SEric Dumazet /* Do not retry other high order allocations */ 53412e4e4410SEric Dumazet order = 1; 53422e4e4410SEric Dumazet max_page_order = 0; 53432e4e4410SEric Dumazet } 53442e4e4410SEric Dumazet order--; 53452e4e4410SEric Dumazet } 53462e4e4410SEric Dumazet page = alloc_page(gfp_mask); 53472e4e4410SEric Dumazet if (!page) 53482e4e4410SEric Dumazet goto failure; 53492e4e4410SEric Dumazet fill_page: 53502e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 53512e4e4410SEric Dumazet PAGE_SIZE << order); 53522e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 53532e4e4410SEric Dumazet data_len -= chunk; 53542e4e4410SEric Dumazet npages -= 1 << order; 53552e4e4410SEric Dumazet } 53562e4e4410SEric Dumazet return skb; 53572e4e4410SEric Dumazet 53582e4e4410SEric Dumazet failure: 53592e4e4410SEric Dumazet kfree_skb(skb); 53602e4e4410SEric Dumazet return NULL; 53612e4e4410SEric Dumazet } 53622e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 53636fa01ccdSSowmini Varadhan 53646fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 53656fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 53666fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 53676fa01ccdSSowmini Varadhan { 53686fa01ccdSSowmini Varadhan int i; 53696fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 53706fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 53716fa01ccdSSowmini Varadhan u8 *data; 53726fa01ccdSSowmini Varadhan 53736fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 53746fa01ccdSSowmini Varadhan 53756fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 53766fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 53776fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 53786fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 53796fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 53806fa01ccdSSowmini Varadhan if (!data) 53816fa01ccdSSowmini Varadhan return -ENOMEM; 53826fa01ccdSSowmini Varadhan 53836fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 53846fa01ccdSSowmini Varadhan 53856fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 53866fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 53876fa01ccdSSowmini Varadhan skb->len -= off; 53886fa01ccdSSowmini Varadhan 53896fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 53906fa01ccdSSowmini Varadhan skb_shinfo(skb), 53916fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 53926fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 53936fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 53946fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 53956fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 53966fa01ccdSSowmini Varadhan kfree(data); 53976fa01ccdSSowmini Varadhan return -ENOMEM; 53986fa01ccdSSowmini Varadhan } 53996fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 54006fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 54016fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 54026fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 54036fa01ccdSSowmini Varadhan skb_release_data(skb); 54046fa01ccdSSowmini Varadhan } else { 54056fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 54066fa01ccdSSowmini Varadhan * relocate values 54076fa01ccdSSowmini Varadhan */ 54086fa01ccdSSowmini Varadhan skb_free_head(skb); 54096fa01ccdSSowmini Varadhan } 54106fa01ccdSSowmini Varadhan 54116fa01ccdSSowmini Varadhan skb->head = data; 54126fa01ccdSSowmini Varadhan skb->data = data; 54136fa01ccdSSowmini Varadhan skb->head_frag = 0; 54146fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 54156fa01ccdSSowmini Varadhan skb->end = size; 54166fa01ccdSSowmini Varadhan #else 54176fa01ccdSSowmini Varadhan skb->end = skb->head + size; 54186fa01ccdSSowmini Varadhan #endif 54196fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 54206fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 54216fa01ccdSSowmini Varadhan skb->cloned = 0; 54226fa01ccdSSowmini Varadhan skb->hdr_len = 0; 54236fa01ccdSSowmini Varadhan skb->nohdr = 0; 54246fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 54256fa01ccdSSowmini Varadhan 54266fa01ccdSSowmini Varadhan return 0; 54276fa01ccdSSowmini Varadhan } 54286fa01ccdSSowmini Varadhan 54296fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 54306fa01ccdSSowmini Varadhan 54316fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 54326fa01ccdSSowmini Varadhan * pskb_carve() 54336fa01ccdSSowmini Varadhan */ 54346fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 54356fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 54366fa01ccdSSowmini Varadhan gfp_t gfp_mask) 54376fa01ccdSSowmini Varadhan { 54386fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 54396fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 54406fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 54416fa01ccdSSowmini Varadhan 54426fa01ccdSSowmini Varadhan do { 54436fa01ccdSSowmini Varadhan if (!list) { 54446fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 54456fa01ccdSSowmini Varadhan return -EFAULT; 54466fa01ccdSSowmini Varadhan } 54476fa01ccdSSowmini Varadhan if (list->len <= eat) { 54486fa01ccdSSowmini Varadhan /* Eaten as whole. */ 54496fa01ccdSSowmini Varadhan eat -= list->len; 54506fa01ccdSSowmini Varadhan list = list->next; 54516fa01ccdSSowmini Varadhan insp = list; 54526fa01ccdSSowmini Varadhan } else { 54536fa01ccdSSowmini Varadhan /* Eaten partially. */ 54546fa01ccdSSowmini Varadhan if (skb_shared(list)) { 54556fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 54566fa01ccdSSowmini Varadhan if (!clone) 54576fa01ccdSSowmini Varadhan return -ENOMEM; 54586fa01ccdSSowmini Varadhan insp = list->next; 54596fa01ccdSSowmini Varadhan list = clone; 54606fa01ccdSSowmini Varadhan } else { 54616fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 54626fa01ccdSSowmini Varadhan insp = list; 54636fa01ccdSSowmini Varadhan } 54646fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 54656fa01ccdSSowmini Varadhan kfree_skb(clone); 54666fa01ccdSSowmini Varadhan return -ENOMEM; 54676fa01ccdSSowmini Varadhan } 54686fa01ccdSSowmini Varadhan break; 54696fa01ccdSSowmini Varadhan } 54706fa01ccdSSowmini Varadhan } while (eat); 54716fa01ccdSSowmini Varadhan 54726fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 54736fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 54746fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 54756fa01ccdSSowmini Varadhan kfree_skb(list); 54766fa01ccdSSowmini Varadhan } 54776fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 54786fa01ccdSSowmini Varadhan if (clone) { 54796fa01ccdSSowmini Varadhan clone->next = list; 54806fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 54816fa01ccdSSowmini Varadhan } 54826fa01ccdSSowmini Varadhan return 0; 54836fa01ccdSSowmini Varadhan } 54846fa01ccdSSowmini Varadhan 54856fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 54866fa01ccdSSowmini Varadhan * non-linear part of skb 54876fa01ccdSSowmini Varadhan */ 54886fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 54896fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 54906fa01ccdSSowmini Varadhan { 54916fa01ccdSSowmini Varadhan int i, k = 0; 54926fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 54936fa01ccdSSowmini Varadhan u8 *data; 54946fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 54956fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 54966fa01ccdSSowmini Varadhan 54976fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 54986fa01ccdSSowmini Varadhan 54996fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 55006fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 55016fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 55026fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 55036fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 55046fa01ccdSSowmini Varadhan if (!data) 55056fa01ccdSSowmini Varadhan return -ENOMEM; 55066fa01ccdSSowmini Varadhan 55076fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 55086fa01ccdSSowmini Varadhan 55096fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 55106fa01ccdSSowmini Varadhan skb_shinfo(skb), offsetof(struct skb_shared_info, 55116fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 55126fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 55136fa01ccdSSowmini Varadhan kfree(data); 55146fa01ccdSSowmini Varadhan return -ENOMEM; 55156fa01ccdSSowmini Varadhan } 55166fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 55176fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 55186fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 55196fa01ccdSSowmini Varadhan 55206fa01ccdSSowmini Varadhan if (pos + fsize > off) { 55216fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 55226fa01ccdSSowmini Varadhan 55236fa01ccdSSowmini Varadhan if (pos < off) { 55246fa01ccdSSowmini Varadhan /* Split frag. 55256fa01ccdSSowmini Varadhan * We have two variants in this case: 55266fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 55276fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 55286fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 55296fa01ccdSSowmini Varadhan * where splitting is expensive. 55306fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 55316fa01ccdSSowmini Varadhan */ 55326fa01ccdSSowmini Varadhan shinfo->frags[0].page_offset += off - pos; 55336fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 55346fa01ccdSSowmini Varadhan } 55356fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 55366fa01ccdSSowmini Varadhan k++; 55376fa01ccdSSowmini Varadhan } 55386fa01ccdSSowmini Varadhan pos += fsize; 55396fa01ccdSSowmini Varadhan } 55406fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 55416fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 55426fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 55436fa01ccdSSowmini Varadhan 55446fa01ccdSSowmini Varadhan if (k == 0) { 55456fa01ccdSSowmini Varadhan /* split line is in frag list */ 55466fa01ccdSSowmini Varadhan pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask); 55476fa01ccdSSowmini Varadhan } 55486fa01ccdSSowmini Varadhan skb_release_data(skb); 55496fa01ccdSSowmini Varadhan 55506fa01ccdSSowmini Varadhan skb->head = data; 55516fa01ccdSSowmini Varadhan skb->head_frag = 0; 55526fa01ccdSSowmini Varadhan skb->data = data; 55536fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 55546fa01ccdSSowmini Varadhan skb->end = size; 55556fa01ccdSSowmini Varadhan #else 55566fa01ccdSSowmini Varadhan skb->end = skb->head + size; 55576fa01ccdSSowmini Varadhan #endif 55586fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 55596fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 55606fa01ccdSSowmini Varadhan skb->cloned = 0; 55616fa01ccdSSowmini Varadhan skb->hdr_len = 0; 55626fa01ccdSSowmini Varadhan skb->nohdr = 0; 55636fa01ccdSSowmini Varadhan skb->len -= off; 55646fa01ccdSSowmini Varadhan skb->data_len = skb->len; 55656fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 55666fa01ccdSSowmini Varadhan return 0; 55676fa01ccdSSowmini Varadhan } 55686fa01ccdSSowmini Varadhan 55696fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 55706fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 55716fa01ccdSSowmini Varadhan { 55726fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 55736fa01ccdSSowmini Varadhan 55746fa01ccdSSowmini Varadhan if (len < headlen) 55756fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 55766fa01ccdSSowmini Varadhan else 55776fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 55786fa01ccdSSowmini Varadhan } 55796fa01ccdSSowmini Varadhan 55806fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 55816fa01ccdSSowmini Varadhan * a new skb 55826fa01ccdSSowmini Varadhan */ 55836fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 55846fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 55856fa01ccdSSowmini Varadhan { 55866fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 55876fa01ccdSSowmini Varadhan 55886fa01ccdSSowmini Varadhan if (!clone) 55896fa01ccdSSowmini Varadhan return NULL; 55906fa01ccdSSowmini Varadhan 55916fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 55926fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 55936fa01ccdSSowmini Varadhan kfree_skb(clone); 55946fa01ccdSSowmini Varadhan return NULL; 55956fa01ccdSSowmini Varadhan } 55966fa01ccdSSowmini Varadhan return clone; 55976fa01ccdSSowmini Varadhan } 55986fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 5599c8c8b127SEric Dumazet 5600c8c8b127SEric Dumazet /** 5601c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 5602c8c8b127SEric Dumazet * @skb: buffer 5603c8c8b127SEric Dumazet * 5604c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 5605c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 5606c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 5607c8c8b127SEric Dumazet * truesize. 5608c8c8b127SEric Dumazet * Notes: 5609c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 5610c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 5611c8c8b127SEric Dumazet */ 5612c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 5613c8c8b127SEric Dumazet { 56143174fed9SEric Dumazet if (skb->data_len) { 56153174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 5616c8c8b127SEric Dumazet skb_cloned(skb)) 5617c8c8b127SEric Dumazet return; 5618c8c8b127SEric Dumazet 5619c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 5620c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 56213174fed9SEric Dumazet } 56223174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 56233174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 56243174fed9SEric Dumazet * their truesize. 56253174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 56263174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 56273174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 5628c8c8b127SEric Dumazet */ 5629c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 5630c8c8b127SEric Dumazet } 5631df5042f4SFlorian Westphal 5632df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 5633df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 5634df5042f4SFlorian Westphal { 5635df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 5636df5042f4SFlorian Westphal } 5637df5042f4SFlorian Westphal 5638df5042f4SFlorian Westphal static struct skb_ext *skb_ext_alloc(void) 5639df5042f4SFlorian Westphal { 5640df5042f4SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 5641df5042f4SFlorian Westphal 5642df5042f4SFlorian Westphal if (new) { 5643df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 5644df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 5645df5042f4SFlorian Westphal } 5646df5042f4SFlorian Westphal 5647df5042f4SFlorian Westphal return new; 5648df5042f4SFlorian Westphal } 5649df5042f4SFlorian Westphal 56504165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 56514165079bSFlorian Westphal unsigned int old_active) 5652df5042f4SFlorian Westphal { 5653df5042f4SFlorian Westphal struct skb_ext *new; 5654df5042f4SFlorian Westphal 5655df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 5656df5042f4SFlorian Westphal return old; 5657df5042f4SFlorian Westphal 5658df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 5659df5042f4SFlorian Westphal if (!new) 5660df5042f4SFlorian Westphal return NULL; 5661df5042f4SFlorian Westphal 5662df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 5663df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 5664df5042f4SFlorian Westphal 56654165079bSFlorian Westphal #ifdef CONFIG_XFRM 56664165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 56674165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 56684165079bSFlorian Westphal unsigned int i; 56694165079bSFlorian Westphal 56704165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 56714165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 56724165079bSFlorian Westphal } 56734165079bSFlorian Westphal #endif 5674df5042f4SFlorian Westphal __skb_ext_put(old); 5675df5042f4SFlorian Westphal return new; 5676df5042f4SFlorian Westphal } 5677df5042f4SFlorian Westphal 5678df5042f4SFlorian Westphal /** 5679df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 5680df5042f4SFlorian Westphal * @skb: buffer 5681df5042f4SFlorian Westphal * @id: extension to allocate space for 5682df5042f4SFlorian Westphal * 5683df5042f4SFlorian Westphal * Allocates enough space for the given extension. 5684df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 5685df5042f4SFlorian Westphal * is returned. 5686df5042f4SFlorian Westphal * 5687df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 5688df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 5689df5042f4SFlorian Westphal * 5690df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 5691df5042f4SFlorian Westphal */ 5692df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 5693df5042f4SFlorian Westphal { 5694df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 5695df5042f4SFlorian Westphal unsigned int newlen, newoff; 5696df5042f4SFlorian Westphal 5697df5042f4SFlorian Westphal if (skb->active_extensions) { 5698df5042f4SFlorian Westphal old = skb->extensions; 5699df5042f4SFlorian Westphal 57004165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 5701df5042f4SFlorian Westphal if (!new) 5702df5042f4SFlorian Westphal return NULL; 5703df5042f4SFlorian Westphal 5704682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 5705df5042f4SFlorian Westphal goto set_active; 5706df5042f4SFlorian Westphal 5707e94e50bdSPaolo Abeni newoff = new->chunks; 5708df5042f4SFlorian Westphal } else { 5709df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 5710df5042f4SFlorian Westphal 5711df5042f4SFlorian Westphal new = skb_ext_alloc(); 5712df5042f4SFlorian Westphal if (!new) 5713df5042f4SFlorian Westphal return NULL; 5714df5042f4SFlorian Westphal } 5715df5042f4SFlorian Westphal 5716df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 5717df5042f4SFlorian Westphal new->chunks = newlen; 5718df5042f4SFlorian Westphal new->offset[id] = newoff; 5719df5042f4SFlorian Westphal set_active: 5720682ec859SPaolo Abeni skb->extensions = new; 5721df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 5722df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 5723df5042f4SFlorian Westphal } 5724df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 5725df5042f4SFlorian Westphal 57264165079bSFlorian Westphal #ifdef CONFIG_XFRM 57274165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 57284165079bSFlorian Westphal { 57294165079bSFlorian Westphal unsigned int i; 57304165079bSFlorian Westphal 57314165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 57324165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 57334165079bSFlorian Westphal } 57344165079bSFlorian Westphal #endif 57354165079bSFlorian Westphal 5736df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 5737df5042f4SFlorian Westphal { 5738df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 5739df5042f4SFlorian Westphal 5740df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 5741df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 5742df5042f4SFlorian Westphal skb->extensions = NULL; 5743df5042f4SFlorian Westphal __skb_ext_put(ext); 57444165079bSFlorian Westphal #ifdef CONFIG_XFRM 57454165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 57464165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 57474165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 57484165079bSFlorian Westphal 57494165079bSFlorian Westphal skb_ext_put_sp(sp); 57504165079bSFlorian Westphal sp->len = 0; 57514165079bSFlorian Westphal #endif 5752df5042f4SFlorian Westphal } 5753df5042f4SFlorian Westphal } 5754df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 5755df5042f4SFlorian Westphal 5756df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 5757df5042f4SFlorian Westphal { 5758df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 5759df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 5760df5042f4SFlorian Westphal */ 5761df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 5762df5042f4SFlorian Westphal goto free_now; 5763df5042f4SFlorian Westphal 5764df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 5765df5042f4SFlorian Westphal return; 5766df5042f4SFlorian Westphal free_now: 57674165079bSFlorian Westphal #ifdef CONFIG_XFRM 57684165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 57694165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 57704165079bSFlorian Westphal #endif 57714165079bSFlorian Westphal 5772df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 5773df5042f4SFlorian Westphal } 5774df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 5775df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 5776