12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 41da177e4SLinus Torvalds * 5113aa838SAlan Cox * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 61da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Fixes: 91da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 101da177e4SLinus Torvalds * balancer bugs. 111da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 121da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 131da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 141da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 151da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 161da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 171da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 181da177e4SLinus Torvalds * only put in the headers 191da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 201da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 211da177e4SLinus Torvalds * Andi Kleen : slabified it. 221da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 231da177e4SLinus Torvalds * 241da177e4SLinus Torvalds * NOTE: 251da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 261da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 271da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 281da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 331da177e4SLinus Torvalds */ 341da177e4SLinus Torvalds 35e005d193SJoe Perches #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 36e005d193SJoe Perches 371da177e4SLinus Torvalds #include <linux/module.h> 381da177e4SLinus Torvalds #include <linux/types.h> 391da177e4SLinus Torvalds #include <linux/kernel.h> 401da177e4SLinus Torvalds #include <linux/mm.h> 411da177e4SLinus Torvalds #include <linux/interrupt.h> 421da177e4SLinus Torvalds #include <linux/in.h> 431da177e4SLinus Torvalds #include <linux/inet.h> 441da177e4SLinus Torvalds #include <linux/slab.h> 45de960aa9SFlorian Westphal #include <linux/tcp.h> 46de960aa9SFlorian Westphal #include <linux/udp.h> 4790017accSMarcelo Ricardo Leitner #include <linux/sctp.h> 481da177e4SLinus Torvalds #include <linux/netdevice.h> 491da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 501da177e4SLinus Torvalds #include <net/pkt_sched.h> 511da177e4SLinus Torvalds #endif 521da177e4SLinus Torvalds #include <linux/string.h> 531da177e4SLinus Torvalds #include <linux/skbuff.h> 549c55e01cSJens Axboe #include <linux/splice.h> 551da177e4SLinus Torvalds #include <linux/cache.h> 561da177e4SLinus Torvalds #include <linux/rtnetlink.h> 571da177e4SLinus Torvalds #include <linux/init.h> 58716ea3a7SDavid Howells #include <linux/scatterlist.h> 59ac45f602SPatrick Ohly #include <linux/errqueue.h> 60268bb0ceSLinus Torvalds #include <linux/prefetch.h> 610d5501c1SVlad Yasevich #include <linux/if_vlan.h> 622a2ea508SJohn Hurley #include <linux/mpls.h> 63183f47fcSSebastian Andrzej Siewior #include <linux/kcov.h> 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds #include <net/protocol.h> 661da177e4SLinus Torvalds #include <net/dst.h> 671da177e4SLinus Torvalds #include <net/sock.h> 681da177e4SLinus Torvalds #include <net/checksum.h> 69ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 701da177e4SLinus Torvalds #include <net/xfrm.h> 718822e270SJohn Hurley #include <net/mpls.h> 723ee17bc7SMat Martineau #include <net/mptcp.h> 7378476d31SJeremy Kerr #include <net/mctp.h> 746a5bcd84SIlias Apalodimas #include <net/page_pool.h> 751da177e4SLinus Torvalds 767c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 77ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7851c56b00SEric Dumazet #include <linux/highmem.h> 79b245be1fSWillem de Bruijn #include <linux/capability.h> 80b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 812544af03SMatteo Croce #include <linux/indirect_call_wrapper.h> 82a1f8e7f7SAl Viro 837b7ed885SBart Van Assche #include "datagram.h" 847f678defSVasily Averin #include "sock_destructor.h" 857b7ed885SBart Van Assche 8608009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8708009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 88df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 89df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 90df5042f4SFlorian Westphal #endif 915f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 925f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /** 95f05de73bSJean Sacren * skb_panic - private function for out-of-line support 961da177e4SLinus Torvalds * @skb: buffer 971da177e4SLinus Torvalds * @sz: size 98f05de73bSJean Sacren * @addr: address 9999d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 1001da177e4SLinus Torvalds * 101f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 102f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 103f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 104f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1051da177e4SLinus Torvalds */ 106f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 10799d5851eSJames Hogan const char msg[]) 1081da177e4SLinus Torvalds { 10941a46913SJesper Dangaard Brouer pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", 11099d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1114305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 11226095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1131da177e4SLinus Torvalds BUG(); 1141da177e4SLinus Torvalds } 1151da177e4SLinus Torvalds 116f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1171da177e4SLinus Torvalds { 118f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1191da177e4SLinus Torvalds } 1201da177e4SLinus Torvalds 121f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 122f05de73bSJean Sacren { 123f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 124f05de73bSJean Sacren } 125c93bdd0eSMel Gorman 12650fad4b5SAlexander Lobakin #define NAPI_SKB_CACHE_SIZE 64 127f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_BULK 16 128f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) 12950fad4b5SAlexander Lobakin 13050fad4b5SAlexander Lobakin struct napi_alloc_cache { 13150fad4b5SAlexander Lobakin struct page_frag_cache page; 13250fad4b5SAlexander Lobakin unsigned int skb_count; 13350fad4b5SAlexander Lobakin void *skb_cache[NAPI_SKB_CACHE_SIZE]; 13450fad4b5SAlexander Lobakin }; 13550fad4b5SAlexander Lobakin 13650fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 13750fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 13850fad4b5SAlexander Lobakin 13932e3573fSYajun Deng void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 14050fad4b5SAlexander Lobakin { 14150fad4b5SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 14250fad4b5SAlexander Lobakin 14350fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 14450fad4b5SAlexander Lobakin 14532e3573fSYajun Deng return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 14650fad4b5SAlexander Lobakin } 14750fad4b5SAlexander Lobakin EXPORT_SYMBOL(__napi_alloc_frag_align); 14850fad4b5SAlexander Lobakin 14950fad4b5SAlexander Lobakin void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 15050fad4b5SAlexander Lobakin { 15150fad4b5SAlexander Lobakin void *data; 15250fad4b5SAlexander Lobakin 15350fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 154afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 15532e3573fSYajun Deng struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache); 15632e3573fSYajun Deng 15750fad4b5SAlexander Lobakin data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask); 15850fad4b5SAlexander Lobakin } else { 15932e3573fSYajun Deng struct napi_alloc_cache *nc; 16032e3573fSYajun Deng 16150fad4b5SAlexander Lobakin local_bh_disable(); 16232e3573fSYajun Deng nc = this_cpu_ptr(&napi_alloc_cache); 16332e3573fSYajun Deng data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 16450fad4b5SAlexander Lobakin local_bh_enable(); 16550fad4b5SAlexander Lobakin } 16650fad4b5SAlexander Lobakin return data; 16750fad4b5SAlexander Lobakin } 16850fad4b5SAlexander Lobakin EXPORT_SYMBOL(__netdev_alloc_frag_align); 16950fad4b5SAlexander Lobakin 170f450d539SAlexander Lobakin static struct sk_buff *napi_skb_cache_get(void) 171f450d539SAlexander Lobakin { 172f450d539SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 173f450d539SAlexander Lobakin struct sk_buff *skb; 174f450d539SAlexander Lobakin 175f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 176f450d539SAlexander Lobakin nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache, 177f450d539SAlexander Lobakin GFP_ATOMIC, 178f450d539SAlexander Lobakin NAPI_SKB_CACHE_BULK, 179f450d539SAlexander Lobakin nc->skb_cache); 180f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 181f450d539SAlexander Lobakin return NULL; 182f450d539SAlexander Lobakin 183f450d539SAlexander Lobakin skb = nc->skb_cache[--nc->skb_count]; 184f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, skb); 185f450d539SAlexander Lobakin 186f450d539SAlexander Lobakin return skb; 187f450d539SAlexander Lobakin } 188f450d539SAlexander Lobakin 189ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 190483126b3SAlexander Lobakin static void __build_skb_around(struct sk_buff *skb, void *data, 191483126b3SAlexander Lobakin unsigned int frag_size) 192ba0509b6SJesper Dangaard Brouer { 193ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 194ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 195ba0509b6SJesper Dangaard Brouer 196ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 197ba0509b6SJesper Dangaard Brouer 198ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 199ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 200ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 201ba0509b6SJesper Dangaard Brouer skb->head = data; 202ba0509b6SJesper Dangaard Brouer skb->data = data; 203ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 204ba0509b6SJesper Dangaard Brouer skb->end = skb->tail + size; 205ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 206ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 207ba0509b6SJesper Dangaard Brouer 208ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 209ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 210ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 211ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 212ba0509b6SJesper Dangaard Brouer 2136370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 214ba0509b6SJesper Dangaard Brouer } 215ba0509b6SJesper Dangaard Brouer 2161da177e4SLinus Torvalds /** 2172ea2f62cSEric Dumazet * __build_skb - build a network buffer 218b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2192ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 220b2b5ce9dSEric Dumazet * 221b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 222deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2232ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2242ea2f62cSEric Dumazet * or vmalloc() 225b2b5ce9dSEric Dumazet * The return is the new skb buffer. 226b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 227b2b5ce9dSEric Dumazet * Notes : 228b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 229b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 230b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 231b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 232b2b5ce9dSEric Dumazet * before giving packet to stack. 233b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 234b2b5ce9dSEric Dumazet */ 2352ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 236b2b5ce9dSEric Dumazet { 237b2b5ce9dSEric Dumazet struct sk_buff *skb; 238b2b5ce9dSEric Dumazet 239b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 240ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 241b2b5ce9dSEric Dumazet return NULL; 242b2b5ce9dSEric Dumazet 243b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 244483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 245b2b5ce9dSEric Dumazet 246483126b3SAlexander Lobakin return skb; 247b2b5ce9dSEric Dumazet } 2482ea2f62cSEric Dumazet 2492ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 2502ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 2512ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 2522ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 2532ea2f62cSEric Dumazet */ 2542ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 2552ea2f62cSEric Dumazet { 2562ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 2572ea2f62cSEric Dumazet 2582ea2f62cSEric Dumazet if (skb && frag_size) { 2592ea2f62cSEric Dumazet skb->head_frag = 1; 2602f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 2612ea2f62cSEric Dumazet skb->pfmemalloc = 1; 2622ea2f62cSEric Dumazet } 2632ea2f62cSEric Dumazet return skb; 2642ea2f62cSEric Dumazet } 265b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 266b2b5ce9dSEric Dumazet 267ba0509b6SJesper Dangaard Brouer /** 268ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 269ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 270ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 271ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 272ba0509b6SJesper Dangaard Brouer */ 273ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 274ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 275ba0509b6SJesper Dangaard Brouer { 276ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 277ba0509b6SJesper Dangaard Brouer return NULL; 278ba0509b6SJesper Dangaard Brouer 279483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 280ba0509b6SJesper Dangaard Brouer 281483126b3SAlexander Lobakin if (frag_size) { 282ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 283ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 284ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 285ba0509b6SJesper Dangaard Brouer } 286ba0509b6SJesper Dangaard Brouer return skb; 287ba0509b6SJesper Dangaard Brouer } 288ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 289ba0509b6SJesper Dangaard Brouer 290f450d539SAlexander Lobakin /** 291f450d539SAlexander Lobakin * __napi_build_skb - build a network buffer 292f450d539SAlexander Lobakin * @data: data buffer provided by caller 293f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 294f450d539SAlexander Lobakin * 295f450d539SAlexander Lobakin * Version of __build_skb() that uses NAPI percpu caches to obtain 296f450d539SAlexander Lobakin * skbuff_head instead of inplace allocation. 297f450d539SAlexander Lobakin * 298f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 299f450d539SAlexander Lobakin */ 300f450d539SAlexander Lobakin static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size) 301f450d539SAlexander Lobakin { 302f450d539SAlexander Lobakin struct sk_buff *skb; 303f450d539SAlexander Lobakin 304f450d539SAlexander Lobakin skb = napi_skb_cache_get(); 305f450d539SAlexander Lobakin if (unlikely(!skb)) 306f450d539SAlexander Lobakin return NULL; 307f450d539SAlexander Lobakin 308f450d539SAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 309f450d539SAlexander Lobakin __build_skb_around(skb, data, frag_size); 310f450d539SAlexander Lobakin 311f450d539SAlexander Lobakin return skb; 312f450d539SAlexander Lobakin } 313f450d539SAlexander Lobakin 314f450d539SAlexander Lobakin /** 315f450d539SAlexander Lobakin * napi_build_skb - build a network buffer 316f450d539SAlexander Lobakin * @data: data buffer provided by caller 317f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 318f450d539SAlexander Lobakin * 319f450d539SAlexander Lobakin * Version of __napi_build_skb() that takes care of skb->head_frag 320f450d539SAlexander Lobakin * and skb->pfmemalloc when the data is a page or page fragment. 321f450d539SAlexander Lobakin * 322f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 323f450d539SAlexander Lobakin */ 324f450d539SAlexander Lobakin struct sk_buff *napi_build_skb(void *data, unsigned int frag_size) 325f450d539SAlexander Lobakin { 326f450d539SAlexander Lobakin struct sk_buff *skb = __napi_build_skb(data, frag_size); 327f450d539SAlexander Lobakin 328f450d539SAlexander Lobakin if (likely(skb) && frag_size) { 329f450d539SAlexander Lobakin skb->head_frag = 1; 330f450d539SAlexander Lobakin skb_propagate_pfmemalloc(virt_to_head_page(data), skb); 331f450d539SAlexander Lobakin } 332f450d539SAlexander Lobakin 333f450d539SAlexander Lobakin return skb; 334f450d539SAlexander Lobakin } 335f450d539SAlexander Lobakin EXPORT_SYMBOL(napi_build_skb); 336f450d539SAlexander Lobakin 3375381b23dSAlexander Lobakin /* 3385381b23dSAlexander Lobakin * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 3395381b23dSAlexander Lobakin * the caller if emergency pfmemalloc reserves are being used. If it is and 3405381b23dSAlexander Lobakin * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 3415381b23dSAlexander Lobakin * may be used. Otherwise, the packet data may be discarded until enough 3425381b23dSAlexander Lobakin * memory is free 3435381b23dSAlexander Lobakin */ 344ef28095fSAlexander Lobakin static void *kmalloc_reserve(size_t size, gfp_t flags, int node, 345ef28095fSAlexander Lobakin bool *pfmemalloc) 3465381b23dSAlexander Lobakin { 3475381b23dSAlexander Lobakin void *obj; 3485381b23dSAlexander Lobakin bool ret_pfmemalloc = false; 3495381b23dSAlexander Lobakin 3505381b23dSAlexander Lobakin /* 3515381b23dSAlexander Lobakin * Try a regular allocation, when that fails and we're not entitled 3525381b23dSAlexander Lobakin * to the reserves, fail. 3535381b23dSAlexander Lobakin */ 3545381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, 3555381b23dSAlexander Lobakin flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 3565381b23dSAlexander Lobakin node); 3575381b23dSAlexander Lobakin if (obj || !(gfp_pfmemalloc_allowed(flags))) 3585381b23dSAlexander Lobakin goto out; 3595381b23dSAlexander Lobakin 3605381b23dSAlexander Lobakin /* Try again but now we are using pfmemalloc reserves */ 3615381b23dSAlexander Lobakin ret_pfmemalloc = true; 3625381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, flags, node); 3635381b23dSAlexander Lobakin 3645381b23dSAlexander Lobakin out: 3655381b23dSAlexander Lobakin if (pfmemalloc) 3665381b23dSAlexander Lobakin *pfmemalloc = ret_pfmemalloc; 3675381b23dSAlexander Lobakin 3685381b23dSAlexander Lobakin return obj; 3695381b23dSAlexander Lobakin } 3705381b23dSAlexander Lobakin 3715381b23dSAlexander Lobakin /* Allocate a new skbuff. We do this ourselves so we can fill in a few 3725381b23dSAlexander Lobakin * 'private' fields and also do memory statistics to find all the 3735381b23dSAlexander Lobakin * [BEEP] leaks. 3745381b23dSAlexander Lobakin * 3755381b23dSAlexander Lobakin */ 3765381b23dSAlexander Lobakin 3775381b23dSAlexander Lobakin /** 3785381b23dSAlexander Lobakin * __alloc_skb - allocate a network buffer 3795381b23dSAlexander Lobakin * @size: size to allocate 3805381b23dSAlexander Lobakin * @gfp_mask: allocation mask 3815381b23dSAlexander Lobakin * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 3825381b23dSAlexander Lobakin * instead of head cache and allocate a cloned (child) skb. 3835381b23dSAlexander Lobakin * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 3845381b23dSAlexander Lobakin * allocations in case the data is required for writeback 3855381b23dSAlexander Lobakin * @node: numa node to allocate memory on 3865381b23dSAlexander Lobakin * 3875381b23dSAlexander Lobakin * Allocate a new &sk_buff. The returned buffer has no headroom and a 3885381b23dSAlexander Lobakin * tail room of at least size bytes. The object has a reference count 3895381b23dSAlexander Lobakin * of one. The return is the buffer. On a failure the return is %NULL. 3905381b23dSAlexander Lobakin * 3915381b23dSAlexander Lobakin * Buffers may only be allocated from interrupts using a @gfp_mask of 3925381b23dSAlexander Lobakin * %GFP_ATOMIC. 3935381b23dSAlexander Lobakin */ 3945381b23dSAlexander Lobakin struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 3955381b23dSAlexander Lobakin int flags, int node) 3965381b23dSAlexander Lobakin { 3975381b23dSAlexander Lobakin struct kmem_cache *cache; 3985381b23dSAlexander Lobakin struct sk_buff *skb; 399a5df6333SLi RongQing unsigned int osize; 4005381b23dSAlexander Lobakin bool pfmemalloc; 401a5df6333SLi RongQing u8 *data; 4025381b23dSAlexander Lobakin 4035381b23dSAlexander Lobakin cache = (flags & SKB_ALLOC_FCLONE) 4045381b23dSAlexander Lobakin ? skbuff_fclone_cache : skbuff_head_cache; 4055381b23dSAlexander Lobakin 4065381b23dSAlexander Lobakin if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 4075381b23dSAlexander Lobakin gfp_mask |= __GFP_MEMALLOC; 4085381b23dSAlexander Lobakin 4095381b23dSAlexander Lobakin /* Get the HEAD */ 410d13612b5SAlexander Lobakin if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI && 411d13612b5SAlexander Lobakin likely(node == NUMA_NO_NODE || node == numa_mem_id())) 412d13612b5SAlexander Lobakin skb = napi_skb_cache_get(); 413d13612b5SAlexander Lobakin else 414d13612b5SAlexander Lobakin skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node); 415df1ae022SAlexander Lobakin if (unlikely(!skb)) 416df1ae022SAlexander Lobakin return NULL; 4175381b23dSAlexander Lobakin prefetchw(skb); 4185381b23dSAlexander Lobakin 4195381b23dSAlexander Lobakin /* We do our best to align skb_shared_info on a separate cache 4205381b23dSAlexander Lobakin * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 4215381b23dSAlexander Lobakin * aligned memory blocks, unless SLUB/SLAB debug is enabled. 4225381b23dSAlexander Lobakin * Both skb->head and skb_shared_info are cache line aligned. 4235381b23dSAlexander Lobakin */ 4245381b23dSAlexander Lobakin size = SKB_DATA_ALIGN(size); 4255381b23dSAlexander Lobakin size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4265381b23dSAlexander Lobakin data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 427df1ae022SAlexander Lobakin if (unlikely(!data)) 4285381b23dSAlexander Lobakin goto nodata; 4295381b23dSAlexander Lobakin /* kmalloc(size) might give us more room than requested. 4305381b23dSAlexander Lobakin * Put skb_shared_info exactly at the end of allocated zone, 4315381b23dSAlexander Lobakin * to allow max possible filling before reallocation. 4325381b23dSAlexander Lobakin */ 433a5df6333SLi RongQing osize = ksize(data); 434a5df6333SLi RongQing size = SKB_WITH_OVERHEAD(osize); 4355381b23dSAlexander Lobakin prefetchw(data + size); 4365381b23dSAlexander Lobakin 4375381b23dSAlexander Lobakin /* 4385381b23dSAlexander Lobakin * Only clear those fields we need to clear, not those that we will 4395381b23dSAlexander Lobakin * actually initialise below. Hence, don't put any more fields after 4405381b23dSAlexander Lobakin * the tail pointer in struct sk_buff! 4415381b23dSAlexander Lobakin */ 4425381b23dSAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 443a5df6333SLi RongQing __build_skb_around(skb, data, osize); 4445381b23dSAlexander Lobakin skb->pfmemalloc = pfmemalloc; 4455381b23dSAlexander Lobakin 4465381b23dSAlexander Lobakin if (flags & SKB_ALLOC_FCLONE) { 4475381b23dSAlexander Lobakin struct sk_buff_fclones *fclones; 4485381b23dSAlexander Lobakin 4495381b23dSAlexander Lobakin fclones = container_of(skb, struct sk_buff_fclones, skb1); 4505381b23dSAlexander Lobakin 4515381b23dSAlexander Lobakin skb->fclone = SKB_FCLONE_ORIG; 4525381b23dSAlexander Lobakin refcount_set(&fclones->fclone_ref, 1); 4535381b23dSAlexander Lobakin 4545381b23dSAlexander Lobakin fclones->skb2.fclone = SKB_FCLONE_CLONE; 4555381b23dSAlexander Lobakin } 4565381b23dSAlexander Lobakin 4575381b23dSAlexander Lobakin return skb; 458df1ae022SAlexander Lobakin 4595381b23dSAlexander Lobakin nodata: 4605381b23dSAlexander Lobakin kmem_cache_free(cache, skb); 461df1ae022SAlexander Lobakin return NULL; 4625381b23dSAlexander Lobakin } 4635381b23dSAlexander Lobakin EXPORT_SYMBOL(__alloc_skb); 4645381b23dSAlexander Lobakin 4657ba7aeabSSebastian Andrzej Siewior /** 466fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 467fd11a83dSAlexander Duyck * @dev: network device to receive on 468d7499160SMasanari Iida * @len: length to allocate 469fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 470fd11a83dSAlexander Duyck * 471fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 472fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 473fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 474fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 475fd11a83dSAlexander Duyck * 476fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 477fd11a83dSAlexander Duyck */ 4789451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4799451980aSAlexander Duyck gfp_t gfp_mask) 480fd11a83dSAlexander Duyck { 481b63ae8caSAlexander Duyck struct page_frag_cache *nc; 482fd11a83dSAlexander Duyck struct sk_buff *skb; 4839451980aSAlexander Duyck bool pfmemalloc; 4849451980aSAlexander Duyck void *data; 485fd11a83dSAlexander Duyck 4869451980aSAlexander Duyck len += NET_SKB_PAD; 487fd11a83dSAlexander Duyck 48866c55602SAlexander Lobakin /* If requested length is either too small or too big, 48966c55602SAlexander Lobakin * we use kmalloc() for skb->head allocation. 49066c55602SAlexander Lobakin */ 49166c55602SAlexander Lobakin if (len <= SKB_WITH_OVERHEAD(1024) || 49266c55602SAlexander Lobakin len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 493d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 494a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 495a080e7bdSAlexander Duyck if (!skb) 496a080e7bdSAlexander Duyck goto skb_fail; 497a080e7bdSAlexander Duyck goto skb_success; 498a080e7bdSAlexander Duyck } 4999451980aSAlexander Duyck 5009451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5019451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5029451980aSAlexander Duyck 5039451980aSAlexander Duyck if (sk_memalloc_socks()) 5049451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5059451980aSAlexander Duyck 506afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 5079451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 5088c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 5099451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 51092dcabd7SSebastian Andrzej Siewior } else { 51192dcabd7SSebastian Andrzej Siewior local_bh_disable(); 51292dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 51392dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 51492dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 51592dcabd7SSebastian Andrzej Siewior local_bh_enable(); 51692dcabd7SSebastian Andrzej Siewior } 5179451980aSAlexander Duyck 5189451980aSAlexander Duyck if (unlikely(!data)) 5199451980aSAlexander Duyck return NULL; 5209451980aSAlexander Duyck 5219451980aSAlexander Duyck skb = __build_skb(data, len); 5229451980aSAlexander Duyck if (unlikely(!skb)) { 523181edb2bSAlexander Duyck skb_free_frag(data); 5249451980aSAlexander Duyck return NULL; 5259451980aSAlexander Duyck } 5269451980aSAlexander Duyck 5279451980aSAlexander Duyck if (pfmemalloc) 5289451980aSAlexander Duyck skb->pfmemalloc = 1; 5299451980aSAlexander Duyck skb->head_frag = 1; 5309451980aSAlexander Duyck 531a080e7bdSAlexander Duyck skb_success: 5328af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 5337b2e497aSChristoph Hellwig skb->dev = dev; 534fd11a83dSAlexander Duyck 535a080e7bdSAlexander Duyck skb_fail: 5368af27456SChristoph Hellwig return skb; 5378af27456SChristoph Hellwig } 538b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 5391da177e4SLinus Torvalds 540fd11a83dSAlexander Duyck /** 541fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 542fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 543d7499160SMasanari Iida * @len: length to allocate 544fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 545fd11a83dSAlexander Duyck * 546fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 547fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 548fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 549fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 550fd11a83dSAlexander Duyck * 551fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 552fd11a83dSAlexander Duyck */ 5539451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 5549451980aSAlexander Duyck gfp_t gfp_mask) 555fd11a83dSAlexander Duyck { 5563226b158SEric Dumazet struct napi_alloc_cache *nc; 557fd11a83dSAlexander Duyck struct sk_buff *skb; 5589451980aSAlexander Duyck void *data; 559fd11a83dSAlexander Duyck 5609451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 561fd11a83dSAlexander Duyck 5623226b158SEric Dumazet /* If requested length is either too small or too big, 5633226b158SEric Dumazet * we use kmalloc() for skb->head allocation. 5643226b158SEric Dumazet */ 5653226b158SEric Dumazet if (len <= SKB_WITH_OVERHEAD(1024) || 5663226b158SEric Dumazet len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 567d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 568cfb8ec65SAlexander Lobakin skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, 569cfb8ec65SAlexander Lobakin NUMA_NO_NODE); 570a080e7bdSAlexander Duyck if (!skb) 571a080e7bdSAlexander Duyck goto skb_fail; 572a080e7bdSAlexander Duyck goto skb_success; 573a080e7bdSAlexander Duyck } 5749451980aSAlexander Duyck 5753226b158SEric Dumazet nc = this_cpu_ptr(&napi_alloc_cache); 5769451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5779451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5789451980aSAlexander Duyck 5799451980aSAlexander Duyck if (sk_memalloc_socks()) 5809451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5819451980aSAlexander Duyck 5828c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5839451980aSAlexander Duyck if (unlikely(!data)) 5849451980aSAlexander Duyck return NULL; 5859451980aSAlexander Duyck 586cfb8ec65SAlexander Lobakin skb = __napi_build_skb(data, len); 5879451980aSAlexander Duyck if (unlikely(!skb)) { 588181edb2bSAlexander Duyck skb_free_frag(data); 5899451980aSAlexander Duyck return NULL; 5909451980aSAlexander Duyck } 5919451980aSAlexander Duyck 592795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5939451980aSAlexander Duyck skb->pfmemalloc = 1; 5949451980aSAlexander Duyck skb->head_frag = 1; 5959451980aSAlexander Duyck 596a080e7bdSAlexander Duyck skb_success: 597fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 598fd11a83dSAlexander Duyck skb->dev = napi->dev; 599fd11a83dSAlexander Duyck 600a080e7bdSAlexander Duyck skb_fail: 601fd11a83dSAlexander Duyck return skb; 602fd11a83dSAlexander Duyck } 603fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 604fd11a83dSAlexander Duyck 605654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 60650269e19SEric Dumazet int size, unsigned int truesize) 607654bed16SPeter Zijlstra { 608654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 609654bed16SPeter Zijlstra skb->len += size; 610654bed16SPeter Zijlstra skb->data_len += size; 61150269e19SEric Dumazet skb->truesize += truesize; 612654bed16SPeter Zijlstra } 613654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 614654bed16SPeter Zijlstra 615f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 616f8e617e1SJason Wang unsigned int truesize) 617f8e617e1SJason Wang { 618f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 619f8e617e1SJason Wang 620f8e617e1SJason Wang skb_frag_size_add(frag, size); 621f8e617e1SJason Wang skb->len += size; 622f8e617e1SJason Wang skb->data_len += size; 623f8e617e1SJason Wang skb->truesize += truesize; 624f8e617e1SJason Wang } 625f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 626f8e617e1SJason Wang 62727b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 6281da177e4SLinus Torvalds { 629bd8a7036SEric Dumazet kfree_skb_list(*listp); 63027b437c8SHerbert Xu *listp = NULL; 6311da177e4SLinus Torvalds } 6321da177e4SLinus Torvalds 63327b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 63427b437c8SHerbert Xu { 63527b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 63627b437c8SHerbert Xu } 63727b437c8SHerbert Xu 6381da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 6391da177e4SLinus Torvalds { 6401da177e4SLinus Torvalds struct sk_buff *list; 6411da177e4SLinus Torvalds 642fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 6431da177e4SLinus Torvalds skb_get(list); 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds 646d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 647d3836f21SEric Dumazet { 648181edb2bSAlexander Duyck unsigned char *head = skb->head; 649181edb2bSAlexander Duyck 6506a5bcd84SIlias Apalodimas if (skb->head_frag) { 6516a5bcd84SIlias Apalodimas if (skb_pp_recycle(skb, head)) 6526a5bcd84SIlias Apalodimas return; 653181edb2bSAlexander Duyck skb_free_frag(head); 6546a5bcd84SIlias Apalodimas } else { 655181edb2bSAlexander Duyck kfree(head); 656d3836f21SEric Dumazet } 6576a5bcd84SIlias Apalodimas } 658d3836f21SEric Dumazet 6595bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 6601da177e4SLinus Torvalds { 661ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6621da177e4SLinus Torvalds int i; 663ff04a771SEric Dumazet 664ff04a771SEric Dumazet if (skb->cloned && 665ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 666ff04a771SEric Dumazet &shinfo->dataref)) 6672cc3aeb5SIlias Apalodimas goto exit; 668ff04a771SEric Dumazet 66970c43167SJonathan Lemon skb_zcopy_clear(skb, true); 67070c43167SJonathan Lemon 671ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 6726a5bcd84SIlias Apalodimas __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); 6731da177e4SLinus Torvalds 674ff04a771SEric Dumazet if (shinfo->frag_list) 675ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6761da177e4SLinus Torvalds 677d3836f21SEric Dumazet skb_free_head(skb); 6782cc3aeb5SIlias Apalodimas exit: 6792cc3aeb5SIlias Apalodimas /* When we clone an SKB we copy the reycling bit. The pp_recycle 6802cc3aeb5SIlias Apalodimas * bit is only set on the head though, so in order to avoid races 6812cc3aeb5SIlias Apalodimas * while trying to recycle fragments on __skb_frag_unref() we need 6822cc3aeb5SIlias Apalodimas * to make one SKB responsible for triggering the recycle path. 6832cc3aeb5SIlias Apalodimas * So disable the recycling bit if an SKB is cloned and we have 6842cc3aeb5SIlias Apalodimas * additional references to to the fragmented part of the SKB. 6852cc3aeb5SIlias Apalodimas * Eventually the last SKB will have the recycling bit set and it's 6862cc3aeb5SIlias Apalodimas * dataref set to 0, which will trigger the recycling 6872cc3aeb5SIlias Apalodimas */ 6882cc3aeb5SIlias Apalodimas skb->pp_recycle = 0; 6891da177e4SLinus Torvalds } 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds /* 6921da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6931da177e4SLinus Torvalds */ 6942d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6951da177e4SLinus Torvalds { 696d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 697d179cd12SDavid S. Miller 698d179cd12SDavid S. Miller switch (skb->fclone) { 699d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 7001da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 7016ffe75ebSEric Dumazet return; 702d179cd12SDavid S. Miller 703d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 704d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 7056ffe75ebSEric Dumazet 7066ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 7076ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 7086ffe75ebSEric Dumazet * while here, branch prediction will be good. 7096ffe75ebSEric Dumazet */ 7102638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 7116ffe75ebSEric Dumazet goto fastpath; 712d179cd12SDavid S. Miller break; 713d179cd12SDavid S. Miller 7146ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 715d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 716d179cd12SDavid S. Miller break; 7173ff50b79SStephen Hemminger } 7182638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 7196ffe75ebSEric Dumazet return; 7206ffe75ebSEric Dumazet fastpath: 7216ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds 7240a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 7251da177e4SLinus Torvalds { 726adf30907SEric Dumazet skb_dst_drop(skb); 7271da177e4SLinus Torvalds if (skb->destructor) { 728afa79d08SChangbin Du WARN_ON(in_hardirq()); 7291da177e4SLinus Torvalds skb->destructor(skb); 7301da177e4SLinus Torvalds } 731a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 732cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 7332fc72c7bSKOVACS Krisztian #endif 734df5042f4SFlorian Westphal skb_ext_put(skb); 73504a4bb55SLennert Buytenhek } 73604a4bb55SLennert Buytenhek 73704a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 73804a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 73904a4bb55SLennert Buytenhek { 74004a4bb55SLennert Buytenhek skb_release_head_state(skb); 741a28b1b90SFlorian Westphal if (likely(skb->head)) 7422d4baff8SHerbert Xu skb_release_data(skb); 7432d4baff8SHerbert Xu } 7441da177e4SLinus Torvalds 7452d4baff8SHerbert Xu /** 7462d4baff8SHerbert Xu * __kfree_skb - private function 7472d4baff8SHerbert Xu * @skb: buffer 7482d4baff8SHerbert Xu * 7492d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 7502d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 7512d4baff8SHerbert Xu * always call kfree_skb 7522d4baff8SHerbert Xu */ 7532d4baff8SHerbert Xu 7542d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 7552d4baff8SHerbert Xu { 7562d4baff8SHerbert Xu skb_release_all(skb); 7571da177e4SLinus Torvalds kfree_skbmem(skb); 7581da177e4SLinus Torvalds } 759b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 7601da177e4SLinus Torvalds 7611da177e4SLinus Torvalds /** 762231d06aeSJörn Engel * kfree_skb - free an sk_buff 763231d06aeSJörn Engel * @skb: buffer to free 764231d06aeSJörn Engel * 765231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 766231d06aeSJörn Engel * hit zero. 767231d06aeSJörn Engel */ 768231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 769231d06aeSJörn Engel { 7703889a803SPaolo Abeni if (!skb_unref(skb)) 771231d06aeSJörn Engel return; 7723889a803SPaolo Abeni 773ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 774231d06aeSJörn Engel __kfree_skb(skb); 775231d06aeSJörn Engel } 776b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 777231d06aeSJörn Engel 778bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 779bd8a7036SEric Dumazet { 780bd8a7036SEric Dumazet while (segs) { 781bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 782bd8a7036SEric Dumazet 783bd8a7036SEric Dumazet kfree_skb(segs); 784bd8a7036SEric Dumazet segs = next; 785bd8a7036SEric Dumazet } 786bd8a7036SEric Dumazet } 787bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 788bd8a7036SEric Dumazet 7896413139dSWillem de Bruijn /* Dump skb information and contents. 7906413139dSWillem de Bruijn * 7916413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 7926413139dSWillem de Bruijn * 793302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 7946413139dSWillem de Bruijn */ 7956413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 7966413139dSWillem de Bruijn { 7976413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 7986413139dSWillem de Bruijn struct net_device *dev = skb->dev; 7996413139dSWillem de Bruijn struct sock *sk = skb->sk; 8006413139dSWillem de Bruijn struct sk_buff *list_skb; 8016413139dSWillem de Bruijn bool has_mac, has_trans; 8026413139dSWillem de Bruijn int headroom, tailroom; 8036413139dSWillem de Bruijn int i, len, seg_len; 8046413139dSWillem de Bruijn 8056413139dSWillem de Bruijn if (full_pkt) 8066413139dSWillem de Bruijn len = skb->len; 8076413139dSWillem de Bruijn else 8086413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 8096413139dSWillem de Bruijn 8106413139dSWillem de Bruijn headroom = skb_headroom(skb); 8116413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 8126413139dSWillem de Bruijn 8136413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 8146413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 8156413139dSWillem de Bruijn 8166413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 8176413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 8186413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 8196413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 8206413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 8216413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 8226413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 8236413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 8246413139dSWillem de Bruijn skb->network_header, 8256413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 8266413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 8276413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 8286413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 8296413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 8306413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 8316413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 8326413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 8336413139dSWillem de Bruijn 8346413139dSWillem de Bruijn if (dev) 8356413139dSWillem de Bruijn printk("%sdev name=%s feat=0x%pNF\n", 8366413139dSWillem de Bruijn level, dev->name, &dev->features); 8376413139dSWillem de Bruijn if (sk) 838db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 8396413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 8406413139dSWillem de Bruijn 8416413139dSWillem de Bruijn if (full_pkt && headroom) 8426413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 8436413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 8446413139dSWillem de Bruijn 8456413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 8466413139dSWillem de Bruijn if (seg_len) 8476413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 8486413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 8496413139dSWillem de Bruijn len -= seg_len; 8506413139dSWillem de Bruijn 8516413139dSWillem de Bruijn if (full_pkt && tailroom) 8526413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 8536413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 8546413139dSWillem de Bruijn 8556413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 8566413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 8576413139dSWillem de Bruijn u32 p_off, p_len, copied; 8586413139dSWillem de Bruijn struct page *p; 8596413139dSWillem de Bruijn u8 *vaddr; 8606413139dSWillem de Bruijn 861b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 8626413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 8636413139dSWillem de Bruijn copied) { 8646413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 8656413139dSWillem de Bruijn vaddr = kmap_atomic(p); 8666413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 8676413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 8686413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 8696413139dSWillem de Bruijn kunmap_atomic(vaddr); 8706413139dSWillem de Bruijn len -= seg_len; 8716413139dSWillem de Bruijn if (!len) 8726413139dSWillem de Bruijn break; 8736413139dSWillem de Bruijn } 8746413139dSWillem de Bruijn } 8756413139dSWillem de Bruijn 8766413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8776413139dSWillem de Bruijn printk("skb fraglist:\n"); 8786413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8796413139dSWillem de Bruijn skb_dump(level, list_skb, true); 8806413139dSWillem de Bruijn } 8816413139dSWillem de Bruijn } 8826413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 8836413139dSWillem de Bruijn 884d1a203eaSStephen Hemminger /** 88525121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 88625121173SMichael S. Tsirkin * @skb: buffer that triggered an error 88725121173SMichael S. Tsirkin * 88825121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 88925121173SMichael S. Tsirkin * skb must be freed afterwards. 89025121173SMichael S. Tsirkin */ 89125121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 89225121173SMichael S. Tsirkin { 8931f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 89425121173SMichael S. Tsirkin } 89525121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 89625121173SMichael S. Tsirkin 897be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 89825121173SMichael S. Tsirkin /** 899ead2ceb0SNeil Horman * consume_skb - free an skbuff 900ead2ceb0SNeil Horman * @skb: buffer to free 901ead2ceb0SNeil Horman * 902ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 903ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 904ead2ceb0SNeil Horman * is being dropped after a failure and notes that 905ead2ceb0SNeil Horman */ 906ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 907ead2ceb0SNeil Horman { 9083889a803SPaolo Abeni if (!skb_unref(skb)) 909ead2ceb0SNeil Horman return; 9103889a803SPaolo Abeni 91107dc22e7SKoki Sanagi trace_consume_skb(skb); 912ead2ceb0SNeil Horman __kfree_skb(skb); 913ead2ceb0SNeil Horman } 914ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 915be769db2SHerbert Xu #endif 916ead2ceb0SNeil Horman 9170a463c78SPaolo Abeni /** 918c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 9190a463c78SPaolo Abeni * @skb: buffer to free 9200a463c78SPaolo Abeni * 921ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 922ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 9230a463c78SPaolo Abeni */ 924ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 9250a463c78SPaolo Abeni { 9260a463c78SPaolo Abeni trace_consume_skb(skb); 9270a463c78SPaolo Abeni skb_release_data(skb); 9280a463c78SPaolo Abeni kfree_skbmem(skb); 9290a463c78SPaolo Abeni } 9300a463c78SPaolo Abeni 931f450d539SAlexander Lobakin static void napi_skb_cache_put(struct sk_buff *skb) 932795bb1c0SJesper Dangaard Brouer { 933795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 934f450d539SAlexander Lobakin u32 i; 935795bb1c0SJesper Dangaard Brouer 936f450d539SAlexander Lobakin kasan_poison_object_data(skbuff_head_cache, skb); 937795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 938795bb1c0SJesper Dangaard Brouer 939795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 940f450d539SAlexander Lobakin for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++) 941f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, 942f450d539SAlexander Lobakin nc->skb_cache[i]); 943f450d539SAlexander Lobakin 944f450d539SAlexander Lobakin kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF, 945f450d539SAlexander Lobakin nc->skb_cache + NAPI_SKB_CACHE_HALF); 946f450d539SAlexander Lobakin nc->skb_count = NAPI_SKB_CACHE_HALF; 947795bb1c0SJesper Dangaard Brouer } 948795bb1c0SJesper Dangaard Brouer } 949f450d539SAlexander Lobakin 95015fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 95115fad714SJesper Dangaard Brouer { 9529243adfcSAlexander Lobakin skb_release_all(skb); 9539243adfcSAlexander Lobakin napi_skb_cache_put(skb); 9549243adfcSAlexander Lobakin } 9559243adfcSAlexander Lobakin 9569243adfcSAlexander Lobakin void napi_skb_free_stolen_head(struct sk_buff *skb) 9579243adfcSAlexander Lobakin { 9589efb4b5bSPaolo Abeni if (unlikely(skb->slow_gro)) { 9598550ff8dSPaul Blakey nf_reset_ct(skb); 9609243adfcSAlexander Lobakin skb_dst_drop(skb); 9619243adfcSAlexander Lobakin skb_ext_put(skb); 9625e10da53SPaolo Abeni skb_orphan(skb); 9639efb4b5bSPaolo Abeni skb->slow_gro = 0; 9649efb4b5bSPaolo Abeni } 965f450d539SAlexander Lobakin napi_skb_cache_put(skb); 96615fad714SJesper Dangaard Brouer } 967795bb1c0SJesper Dangaard Brouer 968795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 969795bb1c0SJesper Dangaard Brouer { 970885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 971795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 972885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 973795bb1c0SJesper Dangaard Brouer return; 974795bb1c0SJesper Dangaard Brouer } 975795bb1c0SJesper Dangaard Brouer 9766454eca8SYunsheng Lin lockdep_assert_in_softirq(); 9776454eca8SYunsheng Lin 9787608894eSPaolo Abeni if (!skb_unref(skb)) 979795bb1c0SJesper Dangaard Brouer return; 9807608894eSPaolo Abeni 981795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 982795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 983795bb1c0SJesper Dangaard Brouer 984795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 985abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 986795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 987795bb1c0SJesper Dangaard Brouer return; 988795bb1c0SJesper Dangaard Brouer } 989795bb1c0SJesper Dangaard Brouer 9909243adfcSAlexander Lobakin skb_release_all(skb); 991f450d539SAlexander Lobakin napi_skb_cache_put(skb); 992795bb1c0SJesper Dangaard Brouer } 993795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 994795bb1c0SJesper Dangaard Brouer 995b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 996b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 997b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 998b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 999b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 1000b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 1001b1937227SEric Dumazet 1002dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 1003dec18810SHerbert Xu { 1004dec18810SHerbert Xu new->tstamp = old->tstamp; 1005b1937227SEric Dumazet /* We do not copy old->sk */ 1006dec18810SHerbert Xu new->dev = old->dev; 1007b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 10087fee226aSEric Dumazet skb_dst_copy(new, old); 1009df5042f4SFlorian Westphal __skb_ext_copy(new, old); 1010b1937227SEric Dumazet __nf_copy(new, old, false); 10116aa895b0SPatrick McHardy 1012b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 1013b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 1014b1937227SEric Dumazet */ 1015b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 101606021292SEliezer Tamir 1017b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 1018b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 1019b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 1020b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 1021b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 1022b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 1023b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 1024b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 1025b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 1026b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 1027b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 1028b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 1029b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 1030b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 1031b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 1032b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 1033b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 1034b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 1035b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 1036b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 1037b1937227SEric Dumazet #endif 1038e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 1039b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 104006021292SEliezer Tamir #endif 10412bd82484SEric Dumazet #ifdef CONFIG_XPS 10422bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 10432bd82484SEric Dumazet #endif 1044b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 1045b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 1046b1937227SEric Dumazet #endif 1047b1937227SEric Dumazet 1048dec18810SHerbert Xu } 1049dec18810SHerbert Xu 105082c49a35SHerbert Xu /* 105182c49a35SHerbert Xu * You should not add any new code to this function. Add it to 105282c49a35SHerbert Xu * __copy_skb_header above instead. 105382c49a35SHerbert Xu */ 1054e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 10551da177e4SLinus Torvalds { 10561da177e4SLinus Torvalds #define C(x) n->x = skb->x 10571da177e4SLinus Torvalds 10581da177e4SLinus Torvalds n->next = n->prev = NULL; 10591da177e4SLinus Torvalds n->sk = NULL; 1060dec18810SHerbert Xu __copy_skb_header(n, skb); 1061dec18810SHerbert Xu 10621da177e4SLinus Torvalds C(len); 10631da177e4SLinus Torvalds C(data_len); 10643e6b3b2eSAlexey Dobriyan C(mac_len); 1065334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 106602f1c89dSPaul Moore n->cloned = 1; 10671da177e4SLinus Torvalds n->nohdr = 0; 1068b13dda9fSEric Dumazet n->peeked = 0; 1069e78bfb07SStefano Brivio C(pfmemalloc); 10706a5bcd84SIlias Apalodimas C(pp_recycle); 10711da177e4SLinus Torvalds n->destructor = NULL; 10721da177e4SLinus Torvalds C(tail); 10731da177e4SLinus Torvalds C(end); 107402f1c89dSPaul Moore C(head); 1075d3836f21SEric Dumazet C(head_frag); 107602f1c89dSPaul Moore C(data); 107702f1c89dSPaul Moore C(truesize); 107863354797SReshetova, Elena refcount_set(&n->users, 1); 10791da177e4SLinus Torvalds 10801da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 10811da177e4SLinus Torvalds skb->cloned = 1; 10821da177e4SLinus Torvalds 10831da177e4SLinus Torvalds return n; 1084e0053ec0SHerbert Xu #undef C 1085e0053ec0SHerbert Xu } 1086e0053ec0SHerbert Xu 1087e0053ec0SHerbert Xu /** 1088da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1089da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1090da29e4b4SJakub Kicinski */ 1091da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1092da29e4b4SJakub Kicinski { 1093da29e4b4SJakub Kicinski struct sk_buff *n; 1094da29e4b4SJakub Kicinski 1095da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1096da29e4b4SJakub Kicinski if (!n) 1097da29e4b4SJakub Kicinski return NULL; 1098da29e4b4SJakub Kicinski 1099da29e4b4SJakub Kicinski n->len = first->len; 1100da29e4b4SJakub Kicinski n->data_len = first->len; 1101da29e4b4SJakub Kicinski n->truesize = first->truesize; 1102da29e4b4SJakub Kicinski 1103da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1104da29e4b4SJakub Kicinski 1105da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1106da29e4b4SJakub Kicinski n->destructor = NULL; 1107da29e4b4SJakub Kicinski 1108da29e4b4SJakub Kicinski return n; 1109da29e4b4SJakub Kicinski } 1110da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1111da29e4b4SJakub Kicinski 1112da29e4b4SJakub Kicinski /** 1113e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1114e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1115e0053ec0SHerbert Xu * @src: the skb to supply the contents 1116e0053ec0SHerbert Xu * 1117e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1118e0053ec0SHerbert Xu * supplied by the user. 1119e0053ec0SHerbert Xu * 1120e0053ec0SHerbert Xu * The target skb is returned upon exit. 1121e0053ec0SHerbert Xu */ 1122e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1123e0053ec0SHerbert Xu { 11242d4baff8SHerbert Xu skb_release_all(dst); 1125e0053ec0SHerbert Xu return __skb_clone(dst, src); 1126e0053ec0SHerbert Xu } 1127e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1128e0053ec0SHerbert Xu 11296f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1130a91dbff5SWillem de Bruijn { 1131a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1132a91dbff5SWillem de Bruijn struct user_struct *user; 1133a91dbff5SWillem de Bruijn 1134a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1135a91dbff5SWillem de Bruijn return 0; 1136a91dbff5SWillem de Bruijn 1137a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1138a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1139a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1140a91dbff5SWillem de Bruijn 1141a91dbff5SWillem de Bruijn do { 1142a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1143a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1144a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1145a91dbff5SWillem de Bruijn return -ENOBUFS; 1146a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1147a91dbff5SWillem de Bruijn old_pg); 1148a91dbff5SWillem de Bruijn 1149a91dbff5SWillem de Bruijn if (!mmp->user) { 1150a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1151a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1152a91dbff5SWillem de Bruijn } else { 1153a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1154a91dbff5SWillem de Bruijn } 1155a91dbff5SWillem de Bruijn 1156a91dbff5SWillem de Bruijn return 0; 1157a91dbff5SWillem de Bruijn } 11586f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1159a91dbff5SWillem de Bruijn 11606f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1161a91dbff5SWillem de Bruijn { 1162a91dbff5SWillem de Bruijn if (mmp->user) { 1163a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1164a91dbff5SWillem de Bruijn free_uid(mmp->user); 1165a91dbff5SWillem de Bruijn } 1166a91dbff5SWillem de Bruijn } 11676f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1168a91dbff5SWillem de Bruijn 11698c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size) 117052267790SWillem de Bruijn { 117152267790SWillem de Bruijn struct ubuf_info *uarg; 117252267790SWillem de Bruijn struct sk_buff *skb; 117352267790SWillem de Bruijn 117452267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 117552267790SWillem de Bruijn 117652267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 117752267790SWillem de Bruijn if (!skb) 117852267790SWillem de Bruijn return NULL; 117952267790SWillem de Bruijn 118052267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 118152267790SWillem de Bruijn uarg = (void *)skb->cb; 1182a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1183a91dbff5SWillem de Bruijn 1184a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1185a91dbff5SWillem de Bruijn kfree_skb(skb); 1186a91dbff5SWillem de Bruijn return NULL; 1187a91dbff5SWillem de Bruijn } 118852267790SWillem de Bruijn 11898c793822SJonathan Lemon uarg->callback = msg_zerocopy_callback; 11904ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 11914ab6c99dSWillem de Bruijn uarg->len = 1; 11924ab6c99dSWillem de Bruijn uarg->bytelen = size; 119352267790SWillem de Bruijn uarg->zerocopy = 1; 119404c2d33eSJonathan Lemon uarg->flags = SKBFL_ZEROCOPY_FRAG; 1195c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 119652267790SWillem de Bruijn sock_hold(sk); 119752267790SWillem de Bruijn 119852267790SWillem de Bruijn return uarg; 119952267790SWillem de Bruijn } 12008c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_alloc); 120152267790SWillem de Bruijn 120252267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 120352267790SWillem de Bruijn { 120452267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 120552267790SWillem de Bruijn } 120652267790SWillem de Bruijn 12078c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 12084ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 12094ab6c99dSWillem de Bruijn { 12104ab6c99dSWillem de Bruijn if (uarg) { 12114ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 12124ab6c99dSWillem de Bruijn u32 bytelen, next; 12134ab6c99dSWillem de Bruijn 12144ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 12154ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 12164ab6c99dSWillem de Bruijn */ 12174ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 12184ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 12194ab6c99dSWillem de Bruijn return NULL; 12204ab6c99dSWillem de Bruijn } 12214ab6c99dSWillem de Bruijn 12224ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 12234ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 12244ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 12254ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12264ab6c99dSWillem de Bruijn goto new_alloc; 12274ab6c99dSWillem de Bruijn return NULL; 12284ab6c99dSWillem de Bruijn } 12294ab6c99dSWillem de Bruijn 12304ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 12314ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1232a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1233a91dbff5SWillem de Bruijn return NULL; 12344ab6c99dSWillem de Bruijn uarg->len++; 12354ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 12364ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1237100f6d8eSWillem de Bruijn 1238100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1239100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12408e044917SJonathan Lemon net_zcopy_get(uarg); 1241100f6d8eSWillem de Bruijn 12424ab6c99dSWillem de Bruijn return uarg; 12434ab6c99dSWillem de Bruijn } 12444ab6c99dSWillem de Bruijn } 12454ab6c99dSWillem de Bruijn 12464ab6c99dSWillem de Bruijn new_alloc: 12478c793822SJonathan Lemon return msg_zerocopy_alloc(sk, size); 12484ab6c99dSWillem de Bruijn } 12498c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_realloc); 12504ab6c99dSWillem de Bruijn 12514ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 12524ab6c99dSWillem de Bruijn { 12534ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 12544ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 12554ab6c99dSWillem de Bruijn u64 sum_len; 12564ab6c99dSWillem de Bruijn 12574ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 12584ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 12594ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 12604ab6c99dSWillem de Bruijn 12614ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 12624ab6c99dSWillem de Bruijn return false; 12634ab6c99dSWillem de Bruijn 12644ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 12654ab6c99dSWillem de Bruijn return false; 12664ab6c99dSWillem de Bruijn 12674ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 12684ab6c99dSWillem de Bruijn return true; 12694ab6c99dSWillem de Bruijn } 12704ab6c99dSWillem de Bruijn 12718c793822SJonathan Lemon static void __msg_zerocopy_callback(struct ubuf_info *uarg) 127252267790SWillem de Bruijn { 12734ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 127452267790SWillem de Bruijn struct sock_exterr_skb *serr; 127552267790SWillem de Bruijn struct sock *sk = skb->sk; 12764ab6c99dSWillem de Bruijn struct sk_buff_head *q; 12774ab6c99dSWillem de Bruijn unsigned long flags; 12783bdd5ee0SWillem de Bruijn bool is_zerocopy; 12794ab6c99dSWillem de Bruijn u32 lo, hi; 12804ab6c99dSWillem de Bruijn u16 len; 128152267790SWillem de Bruijn 1282ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1283ccaffff1SWillem de Bruijn 12844ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 12854ab6c99dSWillem de Bruijn * so do not queue a completion notification 12864ab6c99dSWillem de Bruijn */ 12874ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 128852267790SWillem de Bruijn goto release; 128952267790SWillem de Bruijn 12904ab6c99dSWillem de Bruijn len = uarg->len; 12914ab6c99dSWillem de Bruijn lo = uarg->id; 12924ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 12933bdd5ee0SWillem de Bruijn is_zerocopy = uarg->zerocopy; 12944ab6c99dSWillem de Bruijn 129552267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 129652267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 129752267790SWillem de Bruijn serr->ee.ee_errno = 0; 129852267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 12994ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 13004ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 13013bdd5ee0SWillem de Bruijn if (!is_zerocopy) 130252267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 130352267790SWillem de Bruijn 13044ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 13054ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 13064ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 13074ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 13084ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 13094ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 131052267790SWillem de Bruijn skb = NULL; 13114ab6c99dSWillem de Bruijn } 13124ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 131352267790SWillem de Bruijn 1314e3ae2365SAlexander Aring sk_error_report(sk); 131552267790SWillem de Bruijn 131652267790SWillem de Bruijn release: 131752267790SWillem de Bruijn consume_skb(skb); 131852267790SWillem de Bruijn sock_put(sk); 131952267790SWillem de Bruijn } 132075518851SJonathan Lemon 13218c793822SJonathan Lemon void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 132236177832SJonathan Lemon bool success) 132375518851SJonathan Lemon { 132475518851SJonathan Lemon uarg->zerocopy = uarg->zerocopy & success; 132575518851SJonathan Lemon 132675518851SJonathan Lemon if (refcount_dec_and_test(&uarg->refcnt)) 13278c793822SJonathan Lemon __msg_zerocopy_callback(uarg); 132875518851SJonathan Lemon } 13298c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_callback); 133052267790SWillem de Bruijn 13318c793822SJonathan Lemon void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 133252267790SWillem de Bruijn { 133352267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 133452267790SWillem de Bruijn 133552267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 13364ab6c99dSWillem de Bruijn uarg->len--; 133752267790SWillem de Bruijn 133852900d22SWillem de Bruijn if (have_uref) 13398c793822SJonathan Lemon msg_zerocopy_callback(NULL, uarg, true); 134052267790SWillem de Bruijn } 13418c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort); 134252267790SWillem de Bruijn 1343b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1344b5947e5dSWillem de Bruijn { 1345b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1346b5947e5dSWillem de Bruijn } 1347b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1348b5947e5dSWillem de Bruijn 134952267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 135052267790SWillem de Bruijn struct msghdr *msg, int len, 135152267790SWillem de Bruijn struct ubuf_info *uarg) 135252267790SWillem de Bruijn { 13534ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 135452267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 135552267790SWillem de Bruijn int err, orig_len = skb->len; 135652267790SWillem de Bruijn 13574ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 13584ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 13594ab6c99dSWillem de Bruijn */ 13604ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 13614ab6c99dSWillem de Bruijn return -EEXIST; 13624ab6c99dSWillem de Bruijn 136352267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 136452267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 136554d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 136654d43117SWillem de Bruijn 136752267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 136852267790SWillem de Bruijn msg->msg_iter = orig_iter; 136954d43117SWillem de Bruijn skb->sk = sk; 137052267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 137154d43117SWillem de Bruijn skb->sk = save_sk; 137252267790SWillem de Bruijn return err; 137352267790SWillem de Bruijn } 137452267790SWillem de Bruijn 137552900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 137652267790SWillem de Bruijn return skb->len - orig_len; 137752267790SWillem de Bruijn } 137852267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 137952267790SWillem de Bruijn 13801f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 138152267790SWillem de Bruijn gfp_t gfp_mask) 138252267790SWillem de Bruijn { 138352267790SWillem de Bruijn if (skb_zcopy(orig)) { 138452267790SWillem de Bruijn if (skb_zcopy(nskb)) { 138552267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 138652267790SWillem de Bruijn if (!gfp_mask) { 138752267790SWillem de Bruijn WARN_ON_ONCE(1); 138852267790SWillem de Bruijn return -ENOMEM; 138952267790SWillem de Bruijn } 139052267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 139152267790SWillem de Bruijn return 0; 139252267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 139352267790SWillem de Bruijn return -EIO; 139452267790SWillem de Bruijn } 139552900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 139652267790SWillem de Bruijn } 139752267790SWillem de Bruijn return 0; 139852267790SWillem de Bruijn } 139952267790SWillem de Bruijn 14002c53040fSBen Hutchings /** 14012c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 140248c83012SMichael S. Tsirkin * @skb: the skb to modify 140348c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 140448c83012SMichael S. Tsirkin * 140506b4feb3SJonathan Lemon * This must be called on skb with SKBFL_ZEROCOPY_ENABLE. 140648c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 140748c83012SMichael S. Tsirkin * to userspace pages. 140848c83012SMichael S. Tsirkin * 140948c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 141048c83012SMichael S. Tsirkin * %GFP_ATOMIC. 141148c83012SMichael S. Tsirkin * 141248c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 141348c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 141448c83012SMichael S. Tsirkin */ 141548c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1416a6686f2fSShirley Ma { 1417a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1418a6686f2fSShirley Ma struct page *page, *head = NULL; 14193ece7826SWillem de Bruijn int i, new_frags; 14203ece7826SWillem de Bruijn u32 d_off; 1421a6686f2fSShirley Ma 14223ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 14233ece7826SWillem de Bruijn return -EINVAL; 14243ece7826SWillem de Bruijn 1425f72c4ac6SWillem de Bruijn if (!num_frags) 1426f72c4ac6SWillem de Bruijn goto release; 1427f72c4ac6SWillem de Bruijn 14283ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 14293ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 143002756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1431a6686f2fSShirley Ma if (!page) { 1432a6686f2fSShirley Ma while (head) { 143340dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1434a6686f2fSShirley Ma put_page(head); 1435a6686f2fSShirley Ma head = next; 1436a6686f2fSShirley Ma } 1437a6686f2fSShirley Ma return -ENOMEM; 1438a6686f2fSShirley Ma } 14393ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 14403ece7826SWillem de Bruijn head = page; 14413ece7826SWillem de Bruijn } 14423ece7826SWillem de Bruijn 14433ece7826SWillem de Bruijn page = head; 14443ece7826SWillem de Bruijn d_off = 0; 14453ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 14463ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 14473ece7826SWillem de Bruijn u32 p_off, p_len, copied; 14483ece7826SWillem de Bruijn struct page *p; 14493ece7826SWillem de Bruijn u8 *vaddr; 1450c613c209SWillem de Bruijn 1451b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1452c613c209SWillem de Bruijn p, p_off, p_len, copied) { 14533ece7826SWillem de Bruijn u32 copy, done = 0; 1454c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 14553ece7826SWillem de Bruijn 14563ece7826SWillem de Bruijn while (done < p_len) { 14573ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 14583ece7826SWillem de Bruijn d_off = 0; 14593ece7826SWillem de Bruijn page = (struct page *)page_private(page); 14603ece7826SWillem de Bruijn } 14613ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 14623ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 14633ece7826SWillem de Bruijn vaddr + p_off + done, copy); 14643ece7826SWillem de Bruijn done += copy; 14653ece7826SWillem de Bruijn d_off += copy; 14663ece7826SWillem de Bruijn } 146751c56b00SEric Dumazet kunmap_atomic(vaddr); 1468c613c209SWillem de Bruijn } 1469a6686f2fSShirley Ma } 1470a6686f2fSShirley Ma 1471a6686f2fSShirley Ma /* skb frags release userspace buffers */ 147202756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1473a8605c60SIan Campbell skb_frag_unref(skb, i); 1474a6686f2fSShirley Ma 1475a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 14763ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 14773ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 147840dadff2SSunghan Suh head = (struct page *)page_private(head); 1479a6686f2fSShirley Ma } 14803ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 14813ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 148248c83012SMichael S. Tsirkin 1483b90ddd56SWillem de Bruijn release: 14841f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1485a6686f2fSShirley Ma return 0; 1486a6686f2fSShirley Ma } 1487dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1488a6686f2fSShirley Ma 1489e0053ec0SHerbert Xu /** 1490e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1491e0053ec0SHerbert Xu * @skb: buffer to clone 1492e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1493e0053ec0SHerbert Xu * 1494e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1495e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1496e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1497e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1498e0053ec0SHerbert Xu * 1499e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1500e0053ec0SHerbert Xu * %GFP_ATOMIC. 1501e0053ec0SHerbert Xu */ 1502e0053ec0SHerbert Xu 1503e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1504e0053ec0SHerbert Xu { 1505d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1506d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1507d0bf4a9eSEric Dumazet skb1); 15086ffe75ebSEric Dumazet struct sk_buff *n; 1509e0053ec0SHerbert Xu 151070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1511a6686f2fSShirley Ma return NULL; 1512a6686f2fSShirley Ma 1513e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 15142638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 15156ffe75ebSEric Dumazet n = &fclones->skb2; 15162638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1517e0053ec0SHerbert Xu } else { 1518c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1519c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1520c93bdd0eSMel Gorman 1521e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1522e0053ec0SHerbert Xu if (!n) 1523e0053ec0SHerbert Xu return NULL; 1524fe55f6d5SVegard Nossum 1525e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1526e0053ec0SHerbert Xu } 1527e0053ec0SHerbert Xu 1528e0053ec0SHerbert Xu return __skb_clone(n, skb); 15291da177e4SLinus Torvalds } 1530b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 15311da177e4SLinus Torvalds 1532b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1533f5b17294SPravin B Shelar { 1534030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1535030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1536030737bcSEric Dumazet skb->csum_start += off; 1537f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1538f5b17294SPravin B Shelar skb->transport_header += off; 1539f5b17294SPravin B Shelar skb->network_header += off; 1540f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1541f5b17294SPravin B Shelar skb->mac_header += off; 1542f5b17294SPravin B Shelar skb->inner_transport_header += off; 1543f5b17294SPravin B Shelar skb->inner_network_header += off; 1544aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1545f5b17294SPravin B Shelar } 1546b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1547f5b17294SPravin B Shelar 154808303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 15491da177e4SLinus Torvalds { 1550dec18810SHerbert Xu __copy_skb_header(new, old); 1551dec18810SHerbert Xu 15527967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 15537967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 15547967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 15551da177e4SLinus Torvalds } 155608303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 15571da177e4SLinus Torvalds 1558c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1559c93bdd0eSMel Gorman { 1560c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1561c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1562c93bdd0eSMel Gorman return 0; 1563c93bdd0eSMel Gorman } 1564c93bdd0eSMel Gorman 15651da177e4SLinus Torvalds /** 15661da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 15671da177e4SLinus Torvalds * @skb: buffer to copy 15681da177e4SLinus Torvalds * @gfp_mask: allocation priority 15691da177e4SLinus Torvalds * 15701da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 15711da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 15721da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 15731da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 15741da177e4SLinus Torvalds * 15751da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 15761da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 15771da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 15781da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 15791da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 15801da177e4SLinus Torvalds */ 15811da177e4SLinus Torvalds 1582dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 15831da177e4SLinus Torvalds { 15846602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1585ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1586c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1587c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 15886602cebbSEric Dumazet 15891da177e4SLinus Torvalds if (!n) 15901da177e4SLinus Torvalds return NULL; 15911da177e4SLinus Torvalds 15921da177e4SLinus Torvalds /* Set the data pointer */ 15931da177e4SLinus Torvalds skb_reserve(n, headerlen); 15941da177e4SLinus Torvalds /* Set the tail pointer and length */ 15951da177e4SLinus Torvalds skb_put(n, skb->len); 15961da177e4SLinus Torvalds 15979f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 15981da177e4SLinus Torvalds 159908303c18SIlya Lesokhin skb_copy_header(n, skb); 16001da177e4SLinus Torvalds return n; 16011da177e4SLinus Torvalds } 1602b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 16031da177e4SLinus Torvalds 16041da177e4SLinus Torvalds /** 1605bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 16061da177e4SLinus Torvalds * @skb: buffer to copy 1607117632e6SEric Dumazet * @headroom: headroom of new skb 16081da177e4SLinus Torvalds * @gfp_mask: allocation priority 1609bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1610bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1611bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 16121da177e4SLinus Torvalds * 16131da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 16141da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 16151da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 16161da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 16171da177e4SLinus Torvalds * or the pointer to the buffer on success. 16181da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 16191da177e4SLinus Torvalds */ 16201da177e4SLinus Torvalds 1621bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1622bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 16231da177e4SLinus Torvalds { 1624117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1625bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1626bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 16276602cebbSEric Dumazet 16281da177e4SLinus Torvalds if (!n) 16291da177e4SLinus Torvalds goto out; 16301da177e4SLinus Torvalds 16311da177e4SLinus Torvalds /* Set the data pointer */ 1632117632e6SEric Dumazet skb_reserve(n, headroom); 16331da177e4SLinus Torvalds /* Set the tail pointer and length */ 16341da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 16351da177e4SLinus Torvalds /* Copy the bytes */ 1636d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 16371da177e4SLinus Torvalds 163825f484a6SHerbert Xu n->truesize += skb->data_len; 16391da177e4SLinus Torvalds n->data_len = skb->data_len; 16401da177e4SLinus Torvalds n->len = skb->len; 16411da177e4SLinus Torvalds 16421da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 16431da177e4SLinus Torvalds int i; 16441da177e4SLinus Torvalds 16451f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 16461f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 16471511022cSDan Carpenter kfree_skb(n); 16481511022cSDan Carpenter n = NULL; 1649a6686f2fSShirley Ma goto out; 1650a6686f2fSShirley Ma } 16511da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16521da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1653ea2ab693SIan Campbell skb_frag_ref(skb, i); 16541da177e4SLinus Torvalds } 16551da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 16561da177e4SLinus Torvalds } 16571da177e4SLinus Torvalds 165821dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 16591da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 16601da177e4SLinus Torvalds skb_clone_fraglist(n); 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 166308303c18SIlya Lesokhin skb_copy_header(n, skb); 16641da177e4SLinus Torvalds out: 16651da177e4SLinus Torvalds return n; 16661da177e4SLinus Torvalds } 1667bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 16681da177e4SLinus Torvalds 16691da177e4SLinus Torvalds /** 16701da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 16711da177e4SLinus Torvalds * @skb: buffer to reallocate 16721da177e4SLinus Torvalds * @nhead: room to add at head 16731da177e4SLinus Torvalds * @ntail: room to add at tail 16741da177e4SLinus Torvalds * @gfp_mask: allocation priority 16751da177e4SLinus Torvalds * 1676bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1677bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 16781da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 16791da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 16801da177e4SLinus Torvalds * 16811da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 16821da177e4SLinus Torvalds * reloaded after call to this function. 16831da177e4SLinus Torvalds */ 16841da177e4SLinus Torvalds 168586a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1686dd0fc66fSAl Viro gfp_t gfp_mask) 16871da177e4SLinus Torvalds { 1688158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1689158f323bSEric Dumazet int size = osize + nhead + ntail; 16901da177e4SLinus Torvalds long off; 1691158f323bSEric Dumazet u8 *data; 16921da177e4SLinus Torvalds 16934edd87adSHerbert Xu BUG_ON(nhead < 0); 16944edd87adSHerbert Xu 16959f77fad3STim Hansen BUG_ON(skb_shared(skb)); 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 16981da177e4SLinus Torvalds 1699c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1700c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1701c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1702c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 17031da177e4SLinus Torvalds if (!data) 17041da177e4SLinus Torvalds goto nodata; 170587151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 17061da177e4SLinus Torvalds 17071da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 17086602cebbSEric Dumazet * optimized for the cases when header is void. 17096602cebbSEric Dumazet */ 17106602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 17116602cebbSEric Dumazet 17126602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 17136602cebbSEric Dumazet skb_shinfo(skb), 1714fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 17151da177e4SLinus Torvalds 17163e24591aSAlexander Duyck /* 17173e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 17183e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 17193e24591aSAlexander Duyck * be since all we did is relocate the values 17203e24591aSAlexander Duyck */ 17213e24591aSAlexander Duyck if (skb_cloned(skb)) { 172270008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1723a6686f2fSShirley Ma goto nofrags; 17241f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1725c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 17261da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1727ea2ab693SIan Campbell skb_frag_ref(skb, i); 17281da177e4SLinus Torvalds 172921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 17301da177e4SLinus Torvalds skb_clone_fraglist(skb); 17311da177e4SLinus Torvalds 17321da177e4SLinus Torvalds skb_release_data(skb); 17333e24591aSAlexander Duyck } else { 17343e24591aSAlexander Duyck skb_free_head(skb); 17351fd63041SEric Dumazet } 17361da177e4SLinus Torvalds off = (data + nhead) - skb->head; 17371da177e4SLinus Torvalds 17381da177e4SLinus Torvalds skb->head = data; 1739d3836f21SEric Dumazet skb->head_frag = 0; 17401da177e4SLinus Torvalds skb->data += off; 17414305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 17424305b541SArnaldo Carvalho de Melo skb->end = size; 174356eb8882SPatrick McHardy off = nhead; 17444305b541SArnaldo Carvalho de Melo #else 17454305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 174656eb8882SPatrick McHardy #endif 174727a884dcSArnaldo Carvalho de Melo skb->tail += off; 1748b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 17491da177e4SLinus Torvalds skb->cloned = 0; 1750334a8132SPatrick McHardy skb->hdr_len = 0; 17511da177e4SLinus Torvalds skb->nohdr = 0; 17521da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1753158f323bSEric Dumazet 1754de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1755de8f3a83SDaniel Borkmann 1756158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1757158f323bSEric Dumazet * For the moment, we really care of rx path, or 1758158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1759158f323bSEric Dumazet */ 1760158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1761158f323bSEric Dumazet skb->truesize += size - osize; 1762158f323bSEric Dumazet 17631da177e4SLinus Torvalds return 0; 17641da177e4SLinus Torvalds 1765a6686f2fSShirley Ma nofrags: 1766a6686f2fSShirley Ma kfree(data); 17671da177e4SLinus Torvalds nodata: 17681da177e4SLinus Torvalds return -ENOMEM; 17691da177e4SLinus Torvalds } 1770b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 17711da177e4SLinus Torvalds 17721da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 17731da177e4SLinus Torvalds 17741da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 17751da177e4SLinus Torvalds { 17761da177e4SLinus Torvalds struct sk_buff *skb2; 17771da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds if (delta <= 0) 17801da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 17811da177e4SLinus Torvalds else { 17821da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 17831da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 17841da177e4SLinus Torvalds GFP_ATOMIC)) { 17851da177e4SLinus Torvalds kfree_skb(skb2); 17861da177e4SLinus Torvalds skb2 = NULL; 17871da177e4SLinus Torvalds } 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds return skb2; 17901da177e4SLinus Torvalds } 1791b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds /** 1794f1260ff1SVasily Averin * skb_expand_head - reallocate header of &sk_buff 1795f1260ff1SVasily Averin * @skb: buffer to reallocate 1796f1260ff1SVasily Averin * @headroom: needed headroom 1797f1260ff1SVasily Averin * 1798f1260ff1SVasily Averin * Unlike skb_realloc_headroom, this one does not allocate a new skb 1799f1260ff1SVasily Averin * if possible; copies skb->sk to new skb as needed 1800f1260ff1SVasily Averin * and frees original skb in case of failures. 1801f1260ff1SVasily Averin * 1802f1260ff1SVasily Averin * It expect increased headroom and generates warning otherwise. 1803f1260ff1SVasily Averin */ 1804f1260ff1SVasily Averin 1805f1260ff1SVasily Averin struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom) 1806f1260ff1SVasily Averin { 1807f1260ff1SVasily Averin int delta = headroom - skb_headroom(skb); 18087f678defSVasily Averin int osize = skb_end_offset(skb); 18097f678defSVasily Averin struct sock *sk = skb->sk; 1810f1260ff1SVasily Averin 1811f1260ff1SVasily Averin if (WARN_ONCE(delta <= 0, 1812f1260ff1SVasily Averin "%s is expecting an increase in the headroom", __func__)) 1813f1260ff1SVasily Averin return skb; 1814f1260ff1SVasily Averin 18157f678defSVasily Averin delta = SKB_DATA_ALIGN(delta); 18167f678defSVasily Averin /* pskb_expand_head() might crash, if skb is shared. */ 18177f678defSVasily Averin if (skb_shared(skb) || !is_skb_wmem(skb)) { 1818f1260ff1SVasily Averin struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1819f1260ff1SVasily Averin 18207f678defSVasily Averin if (unlikely(!nskb)) 18217f678defSVasily Averin goto fail; 18227f678defSVasily Averin 18237f678defSVasily Averin if (sk) 18247f678defSVasily Averin skb_set_owner_w(nskb, sk); 1825f1260ff1SVasily Averin consume_skb(skb); 1826f1260ff1SVasily Averin skb = nskb; 1827f1260ff1SVasily Averin } 18287f678defSVasily Averin if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC)) 18297f678defSVasily Averin goto fail; 18307f678defSVasily Averin 18317f678defSVasily Averin if (sk && is_skb_wmem(skb)) { 18327f678defSVasily Averin delta = skb_end_offset(skb) - osize; 18337f678defSVasily Averin refcount_add(delta, &sk->sk_wmem_alloc); 18347f678defSVasily Averin skb->truesize += delta; 1835f1260ff1SVasily Averin } 1836f1260ff1SVasily Averin return skb; 18377f678defSVasily Averin 18387f678defSVasily Averin fail: 18397f678defSVasily Averin kfree_skb(skb); 18407f678defSVasily Averin return NULL; 1841f1260ff1SVasily Averin } 1842f1260ff1SVasily Averin EXPORT_SYMBOL(skb_expand_head); 1843f1260ff1SVasily Averin 1844f1260ff1SVasily Averin /** 18451da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 18461da177e4SLinus Torvalds * @skb: buffer to copy 18471da177e4SLinus Torvalds * @newheadroom: new free bytes at head 18481da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 18491da177e4SLinus Torvalds * @gfp_mask: allocation priority 18501da177e4SLinus Torvalds * 18511da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 18521da177e4SLinus Torvalds * allocate additional space. 18531da177e4SLinus Torvalds * 18541da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 18551da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 18561da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 18571da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 18581da177e4SLinus Torvalds * 18591da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 18601da177e4SLinus Torvalds * is called from an interrupt. 18611da177e4SLinus Torvalds */ 18621da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 186386a76cafSVictor Fusco int newheadroom, int newtailroom, 1864dd0fc66fSAl Viro gfp_t gfp_mask) 18651da177e4SLinus Torvalds { 18661da177e4SLinus Torvalds /* 18671da177e4SLinus Torvalds * Allocate the copy buffer 18681da177e4SLinus Torvalds */ 1869c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1870c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1871c93bdd0eSMel Gorman NUMA_NO_NODE); 1872efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 18731da177e4SLinus Torvalds int head_copy_len, head_copy_off; 18741da177e4SLinus Torvalds 18751da177e4SLinus Torvalds if (!n) 18761da177e4SLinus Torvalds return NULL; 18771da177e4SLinus Torvalds 18781da177e4SLinus Torvalds skb_reserve(n, newheadroom); 18791da177e4SLinus Torvalds 18801da177e4SLinus Torvalds /* Set the tail pointer and length */ 18811da177e4SLinus Torvalds skb_put(n, skb->len); 18821da177e4SLinus Torvalds 1883efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 18841da177e4SLinus Torvalds head_copy_off = 0; 18851da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 18861da177e4SLinus Torvalds head_copy_len = newheadroom; 18871da177e4SLinus Torvalds else 18881da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds /* Copy the linear header and data. */ 18919f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 18929f77fad3STim Hansen skb->len + head_copy_len)); 18931da177e4SLinus Torvalds 189408303c18SIlya Lesokhin skb_copy_header(n, skb); 18951da177e4SLinus Torvalds 1896030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1897efd1e8d5SPatrick McHardy 18981da177e4SLinus Torvalds return n; 18991da177e4SLinus Torvalds } 1900b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 19011da177e4SLinus Torvalds 19021da177e4SLinus Torvalds /** 1903cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 19041da177e4SLinus Torvalds * @skb: buffer to pad 19051da177e4SLinus Torvalds * @pad: space to pad 1906cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 19071da177e4SLinus Torvalds * 19081da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 19091da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 19101da177e4SLinus Torvalds * beyond the buffer end onto the wire. 19111da177e4SLinus Torvalds * 1912cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1913cd0a137aSFlorian Fainelli * if @free_on_error is true. 19141da177e4SLinus Torvalds */ 19151da177e4SLinus Torvalds 1916cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 19171da177e4SLinus Torvalds { 19185b057c6bSHerbert Xu int err; 19195b057c6bSHerbert Xu int ntail; 19201da177e4SLinus Torvalds 19211da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 19225b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 19231da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 19245b057c6bSHerbert Xu return 0; 19251da177e4SLinus Torvalds } 19261da177e4SLinus Torvalds 19274305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 19285b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 19295b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 19305b057c6bSHerbert Xu if (unlikely(err)) 19315b057c6bSHerbert Xu goto free_skb; 19325b057c6bSHerbert Xu } 19335b057c6bSHerbert Xu 19345b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 19355b057c6bSHerbert Xu * to be audited. 19365b057c6bSHerbert Xu */ 19375b057c6bSHerbert Xu err = skb_linearize(skb); 19385b057c6bSHerbert Xu if (unlikely(err)) 19395b057c6bSHerbert Xu goto free_skb; 19405b057c6bSHerbert Xu 19415b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 19425b057c6bSHerbert Xu return 0; 19435b057c6bSHerbert Xu 19445b057c6bSHerbert Xu free_skb: 1945cd0a137aSFlorian Fainelli if (free_on_error) 19461da177e4SLinus Torvalds kfree_skb(skb); 19475b057c6bSHerbert Xu return err; 19481da177e4SLinus Torvalds } 1949cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 19501da177e4SLinus Torvalds 19510dde3e16SIlpo Järvinen /** 19520c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 19530c7ddf36SMathias Krause * @skb: start of the buffer to use 19540c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 19550c7ddf36SMathias Krause * @len: amount of data to add 19560c7ddf36SMathias Krause * 19570c7ddf36SMathias Krause * This function extends the used data area of the potentially 19580c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 19590c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 19600c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 19610c7ddf36SMathias Krause * returned. 19620c7ddf36SMathias Krause */ 19630c7ddf36SMathias Krause 19644df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 19650c7ddf36SMathias Krause { 19660c7ddf36SMathias Krause if (tail != skb) { 19670c7ddf36SMathias Krause skb->data_len += len; 19680c7ddf36SMathias Krause skb->len += len; 19690c7ddf36SMathias Krause } 19700c7ddf36SMathias Krause return skb_put(tail, len); 19710c7ddf36SMathias Krause } 19720c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 19730c7ddf36SMathias Krause 19740c7ddf36SMathias Krause /** 19750dde3e16SIlpo Järvinen * skb_put - add data to a buffer 19760dde3e16SIlpo Järvinen * @skb: buffer to use 19770dde3e16SIlpo Järvinen * @len: amount of data to add 19780dde3e16SIlpo Järvinen * 19790dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 19800dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 19810dde3e16SIlpo Järvinen * first byte of the extra data is returned. 19820dde3e16SIlpo Järvinen */ 19834df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 19840dde3e16SIlpo Järvinen { 19854df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 19860dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 19870dde3e16SIlpo Järvinen skb->tail += len; 19880dde3e16SIlpo Järvinen skb->len += len; 19890dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 19900dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 19910dde3e16SIlpo Järvinen return tmp; 19920dde3e16SIlpo Järvinen } 19930dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 19940dde3e16SIlpo Järvinen 19956be8ac2fSIlpo Järvinen /** 1996c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1997c2aa270aSIlpo Järvinen * @skb: buffer to use 1998c2aa270aSIlpo Järvinen * @len: amount of data to add 1999c2aa270aSIlpo Järvinen * 2000c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 2001c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 2002c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 2003c2aa270aSIlpo Järvinen */ 2004d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 2005c2aa270aSIlpo Järvinen { 2006c2aa270aSIlpo Järvinen skb->data -= len; 2007c2aa270aSIlpo Järvinen skb->len += len; 2008c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 2009c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 2010c2aa270aSIlpo Järvinen return skb->data; 2011c2aa270aSIlpo Järvinen } 2012c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 2013c2aa270aSIlpo Järvinen 2014c2aa270aSIlpo Järvinen /** 20156be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 20166be8ac2fSIlpo Järvinen * @skb: buffer to use 20176be8ac2fSIlpo Järvinen * @len: amount of data to remove 20186be8ac2fSIlpo Järvinen * 20196be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 20206be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 20216be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 20226be8ac2fSIlpo Järvinen * the old data. 20236be8ac2fSIlpo Järvinen */ 2024af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 20256be8ac2fSIlpo Järvinen { 202647d29646SDavid S. Miller return skb_pull_inline(skb, len); 20276be8ac2fSIlpo Järvinen } 20286be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 20296be8ac2fSIlpo Järvinen 2030419ae74eSIlpo Järvinen /** 2031419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 2032419ae74eSIlpo Järvinen * @skb: buffer to alter 2033419ae74eSIlpo Järvinen * @len: new length 2034419ae74eSIlpo Järvinen * 2035419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 2036419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 2037419ae74eSIlpo Järvinen * The skb must be linear. 2038419ae74eSIlpo Järvinen */ 2039419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 2040419ae74eSIlpo Järvinen { 2041419ae74eSIlpo Järvinen if (skb->len > len) 2042419ae74eSIlpo Järvinen __skb_trim(skb, len); 2043419ae74eSIlpo Järvinen } 2044419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 2045419ae74eSIlpo Järvinen 20463cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 20471da177e4SLinus Torvalds */ 20481da177e4SLinus Torvalds 20493cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 20501da177e4SLinus Torvalds { 205127b437c8SHerbert Xu struct sk_buff **fragp; 205227b437c8SHerbert Xu struct sk_buff *frag; 20531da177e4SLinus Torvalds int offset = skb_headlen(skb); 20541da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 20551da177e4SLinus Torvalds int i; 205627b437c8SHerbert Xu int err; 205727b437c8SHerbert Xu 205827b437c8SHerbert Xu if (skb_cloned(skb) && 205927b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 206027b437c8SHerbert Xu return err; 20611da177e4SLinus Torvalds 2062f4d26fb3SHerbert Xu i = 0; 2063f4d26fb3SHerbert Xu if (offset >= len) 2064f4d26fb3SHerbert Xu goto drop_pages; 2065f4d26fb3SHerbert Xu 2066f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 20679e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 206827b437c8SHerbert Xu 206927b437c8SHerbert Xu if (end < len) { 20701da177e4SLinus Torvalds offset = end; 207127b437c8SHerbert Xu continue; 20721da177e4SLinus Torvalds } 20731da177e4SLinus Torvalds 20749e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 207527b437c8SHerbert Xu 2076f4d26fb3SHerbert Xu drop_pages: 207727b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 207827b437c8SHerbert Xu 207927b437c8SHerbert Xu for (; i < nfrags; i++) 2080ea2ab693SIan Campbell skb_frag_unref(skb, i); 208127b437c8SHerbert Xu 208221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 208327b437c8SHerbert Xu skb_drop_fraglist(skb); 2084f4d26fb3SHerbert Xu goto done; 208527b437c8SHerbert Xu } 208627b437c8SHerbert Xu 208727b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 208827b437c8SHerbert Xu fragp = &frag->next) { 208927b437c8SHerbert Xu int end = offset + frag->len; 209027b437c8SHerbert Xu 209127b437c8SHerbert Xu if (skb_shared(frag)) { 209227b437c8SHerbert Xu struct sk_buff *nfrag; 209327b437c8SHerbert Xu 209427b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 209527b437c8SHerbert Xu if (unlikely(!nfrag)) 209627b437c8SHerbert Xu return -ENOMEM; 209727b437c8SHerbert Xu 209827b437c8SHerbert Xu nfrag->next = frag->next; 209985bb2a60SEric Dumazet consume_skb(frag); 210027b437c8SHerbert Xu frag = nfrag; 210127b437c8SHerbert Xu *fragp = frag; 210227b437c8SHerbert Xu } 210327b437c8SHerbert Xu 210427b437c8SHerbert Xu if (end < len) { 210527b437c8SHerbert Xu offset = end; 210627b437c8SHerbert Xu continue; 210727b437c8SHerbert Xu } 210827b437c8SHerbert Xu 210927b437c8SHerbert Xu if (end > len && 211027b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 211127b437c8SHerbert Xu return err; 211227b437c8SHerbert Xu 211327b437c8SHerbert Xu if (frag->next) 211427b437c8SHerbert Xu skb_drop_list(&frag->next); 211527b437c8SHerbert Xu break; 211627b437c8SHerbert Xu } 211727b437c8SHerbert Xu 2118f4d26fb3SHerbert Xu done: 211927b437c8SHerbert Xu if (len > skb_headlen(skb)) { 21201da177e4SLinus Torvalds skb->data_len -= skb->len - len; 21211da177e4SLinus Torvalds skb->len = len; 21221da177e4SLinus Torvalds } else { 21231da177e4SLinus Torvalds skb->len = len; 21241da177e4SLinus Torvalds skb->data_len = 0; 212527a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 21261da177e4SLinus Torvalds } 21271da177e4SLinus Torvalds 2128c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2129c21b48ccSEric Dumazet skb_condense(skb); 21301da177e4SLinus Torvalds return 0; 21311da177e4SLinus Torvalds } 2132b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 21331da177e4SLinus Torvalds 213488078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 213588078d98SEric Dumazet */ 213688078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 213788078d98SEric Dumazet { 213888078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 213988078d98SEric Dumazet int delta = skb->len - len; 214088078d98SEric Dumazet 2141d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2142d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2143d55bef50SDimitris Michailidis len); 214454970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 214554970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 214654970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 214754970a2fSVasily Averin 214854970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 214954970a2fSVasily Averin return -EINVAL; 215088078d98SEric Dumazet } 215188078d98SEric Dumazet return __pskb_trim(skb, len); 215288078d98SEric Dumazet } 215388078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 215488078d98SEric Dumazet 21551da177e4SLinus Torvalds /** 21561da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 21571da177e4SLinus Torvalds * @skb: buffer to reallocate 21581da177e4SLinus Torvalds * @delta: number of bytes to advance tail 21591da177e4SLinus Torvalds * 21601da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 21611da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 21621da177e4SLinus Torvalds * data from fragmented part. 21631da177e4SLinus Torvalds * 21641da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 21651da177e4SLinus Torvalds * 21661da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 21671da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 21681da177e4SLinus Torvalds * 21691da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 21701da177e4SLinus Torvalds * reloaded after call to this function. 21711da177e4SLinus Torvalds */ 21721da177e4SLinus Torvalds 21731da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 21741da177e4SLinus Torvalds * when it is necessary. 21751da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 21761da177e4SLinus Torvalds * 2. It may change skb pointers. 21771da177e4SLinus Torvalds * 21781da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 21791da177e4SLinus Torvalds */ 2180af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 21811da177e4SLinus Torvalds { 21821da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 21831da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 21841da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 21851da177e4SLinus Torvalds */ 21864305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 21871da177e4SLinus Torvalds 21881da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 21891da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 21901da177e4SLinus Torvalds GFP_ATOMIC)) 21911da177e4SLinus Torvalds return NULL; 21921da177e4SLinus Torvalds } 21931da177e4SLinus Torvalds 21949f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 21959f77fad3STim Hansen skb_tail_pointer(skb), delta)); 21961da177e4SLinus Torvalds 21971da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 21981da177e4SLinus Torvalds * size of pulled pages. Superb. 21991da177e4SLinus Torvalds */ 220021dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 22011da177e4SLinus Torvalds goto pull_pages; 22021da177e4SLinus Torvalds 22031da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 22041da177e4SLinus Torvalds eat = delta; 22051da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22069e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22079e903e08SEric Dumazet 22089e903e08SEric Dumazet if (size >= eat) 22091da177e4SLinus Torvalds goto pull_pages; 22109e903e08SEric Dumazet eat -= size; 22111da177e4SLinus Torvalds } 22121da177e4SLinus Torvalds 22131da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 221409001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 22151da177e4SLinus Torvalds * but taking into account that pulling is expected to 22161da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 22171da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 22181da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 22191da177e4SLinus Torvalds */ 22201da177e4SLinus Torvalds if (eat) { 22211da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 22221da177e4SLinus Torvalds struct sk_buff *clone = NULL; 22231da177e4SLinus Torvalds struct sk_buff *insp = NULL; 22241da177e4SLinus Torvalds 22251da177e4SLinus Torvalds do { 22261da177e4SLinus Torvalds if (list->len <= eat) { 22271da177e4SLinus Torvalds /* Eaten as whole. */ 22281da177e4SLinus Torvalds eat -= list->len; 22291da177e4SLinus Torvalds list = list->next; 22301da177e4SLinus Torvalds insp = list; 22311da177e4SLinus Torvalds } else { 22321da177e4SLinus Torvalds /* Eaten partially. */ 22331da177e4SLinus Torvalds 22341da177e4SLinus Torvalds if (skb_shared(list)) { 22351da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 22361da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 22371da177e4SLinus Torvalds if (!clone) 22381da177e4SLinus Torvalds return NULL; 22391da177e4SLinus Torvalds insp = list->next; 22401da177e4SLinus Torvalds list = clone; 22411da177e4SLinus Torvalds } else { 22421da177e4SLinus Torvalds /* This may be pulled without 22431da177e4SLinus Torvalds * problems. */ 22441da177e4SLinus Torvalds insp = list; 22451da177e4SLinus Torvalds } 22461da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 22471da177e4SLinus Torvalds kfree_skb(clone); 22481da177e4SLinus Torvalds return NULL; 22491da177e4SLinus Torvalds } 22501da177e4SLinus Torvalds break; 22511da177e4SLinus Torvalds } 22521da177e4SLinus Torvalds } while (eat); 22531da177e4SLinus Torvalds 22541da177e4SLinus Torvalds /* Free pulled out fragments. */ 22551da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 22561da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 22571da177e4SLinus Torvalds kfree_skb(list); 22581da177e4SLinus Torvalds } 22591da177e4SLinus Torvalds /* And insert new clone at head. */ 22601da177e4SLinus Torvalds if (clone) { 22611da177e4SLinus Torvalds clone->next = list; 22621da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 22631da177e4SLinus Torvalds } 22641da177e4SLinus Torvalds } 22651da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 22661da177e4SLinus Torvalds 22671da177e4SLinus Torvalds pull_pages: 22681da177e4SLinus Torvalds eat = delta; 22691da177e4SLinus Torvalds k = 0; 22701da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22719e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22729e903e08SEric Dumazet 22739e903e08SEric Dumazet if (size <= eat) { 2274ea2ab693SIan Campbell skb_frag_unref(skb, i); 22759e903e08SEric Dumazet eat -= size; 22761da177e4SLinus Torvalds } else { 2277b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2278b54c9d5bSJonathan Lemon 2279b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 22801da177e4SLinus Torvalds if (eat) { 2281b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2282b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 22833ccc6c6fSlinzhang if (!i) 22843ccc6c6fSlinzhang goto end; 22851da177e4SLinus Torvalds eat = 0; 22861da177e4SLinus Torvalds } 22871da177e4SLinus Torvalds k++; 22881da177e4SLinus Torvalds } 22891da177e4SLinus Torvalds } 22901da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 22911da177e4SLinus Torvalds 22923ccc6c6fSlinzhang end: 22931da177e4SLinus Torvalds skb->tail += delta; 22941da177e4SLinus Torvalds skb->data_len -= delta; 22951da177e4SLinus Torvalds 22961f8b977aSWillem de Bruijn if (!skb->data_len) 22971f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 22981f8b977aSWillem de Bruijn 229927a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 23001da177e4SLinus Torvalds } 2301b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 23021da177e4SLinus Torvalds 230322019b17SEric Dumazet /** 230422019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 230522019b17SEric Dumazet * @skb: source skb 230622019b17SEric Dumazet * @offset: offset in source 230722019b17SEric Dumazet * @to: destination buffer 230822019b17SEric Dumazet * @len: number of bytes to copy 230922019b17SEric Dumazet * 231022019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 231122019b17SEric Dumazet * destination buffer. 231222019b17SEric Dumazet * 231322019b17SEric Dumazet * CAUTION ! : 231422019b17SEric Dumazet * If its prototype is ever changed, 231522019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 231622019b17SEric Dumazet * since it is called from BPF assembly code. 231722019b17SEric Dumazet */ 23181da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 23191da177e4SLinus Torvalds { 23201a028e50SDavid S. Miller int start = skb_headlen(skb); 2321fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2322fbb398a8SDavid S. Miller int i, copy; 23231da177e4SLinus Torvalds 23241da177e4SLinus Torvalds if (offset > (int)skb->len - len) 23251da177e4SLinus Torvalds goto fault; 23261da177e4SLinus Torvalds 23271da177e4SLinus Torvalds /* Copy header. */ 23281a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 23291da177e4SLinus Torvalds if (copy > len) 23301da177e4SLinus Torvalds copy = len; 2331d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 23321da177e4SLinus Torvalds if ((len -= copy) == 0) 23331da177e4SLinus Torvalds return 0; 23341da177e4SLinus Torvalds offset += copy; 23351da177e4SLinus Torvalds to += copy; 23361da177e4SLinus Torvalds } 23371da177e4SLinus Torvalds 23381da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 23391a028e50SDavid S. Miller int end; 234051c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 23411da177e4SLinus Torvalds 2342547b792cSIlpo Järvinen WARN_ON(start > offset + len); 23431a028e50SDavid S. Miller 234451c56b00SEric Dumazet end = start + skb_frag_size(f); 23451da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2346c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2347c613c209SWillem de Bruijn struct page *p; 23481da177e4SLinus Torvalds u8 *vaddr; 23491da177e4SLinus Torvalds 23501da177e4SLinus Torvalds if (copy > len) 23511da177e4SLinus Torvalds copy = len; 23521da177e4SLinus Torvalds 2353c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2354b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2355c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2356c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2357c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 235851c56b00SEric Dumazet kunmap_atomic(vaddr); 2359c613c209SWillem de Bruijn } 23601da177e4SLinus Torvalds 23611da177e4SLinus Torvalds if ((len -= copy) == 0) 23621da177e4SLinus Torvalds return 0; 23631da177e4SLinus Torvalds offset += copy; 23641da177e4SLinus Torvalds to += copy; 23651da177e4SLinus Torvalds } 23661a028e50SDavid S. Miller start = end; 23671da177e4SLinus Torvalds } 23681da177e4SLinus Torvalds 2369fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 23701a028e50SDavid S. Miller int end; 23711da177e4SLinus Torvalds 2372547b792cSIlpo Järvinen WARN_ON(start > offset + len); 23731a028e50SDavid S. Miller 2374fbb398a8SDavid S. Miller end = start + frag_iter->len; 23751da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 23761da177e4SLinus Torvalds if (copy > len) 23771da177e4SLinus Torvalds copy = len; 2378fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 23791da177e4SLinus Torvalds goto fault; 23801da177e4SLinus Torvalds if ((len -= copy) == 0) 23811da177e4SLinus Torvalds return 0; 23821da177e4SLinus Torvalds offset += copy; 23831da177e4SLinus Torvalds to += copy; 23841da177e4SLinus Torvalds } 23851a028e50SDavid S. Miller start = end; 23861da177e4SLinus Torvalds } 2387a6686f2fSShirley Ma 23881da177e4SLinus Torvalds if (!len) 23891da177e4SLinus Torvalds return 0; 23901da177e4SLinus Torvalds 23911da177e4SLinus Torvalds fault: 23921da177e4SLinus Torvalds return -EFAULT; 23931da177e4SLinus Torvalds } 2394b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 23951da177e4SLinus Torvalds 23969c55e01cSJens Axboe /* 23979c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 23989c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 23999c55e01cSJens Axboe */ 24009c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 24019c55e01cSJens Axboe { 24028b9d3728SJarek Poplawski put_page(spd->pages[i]); 24038b9d3728SJarek Poplawski } 24049c55e01cSJens Axboe 2405a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 24064fb66994SJarek Poplawski unsigned int *offset, 240718aafc62SEric Dumazet struct sock *sk) 24088b9d3728SJarek Poplawski { 24095640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 24108b9d3728SJarek Poplawski 24115640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 24128b9d3728SJarek Poplawski return NULL; 24134fb66994SJarek Poplawski 24145640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 24154fb66994SJarek Poplawski 24165640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 24175640f768SEric Dumazet page_address(page) + *offset, *len); 24185640f768SEric Dumazet *offset = pfrag->offset; 24195640f768SEric Dumazet pfrag->offset += *len; 24204fb66994SJarek Poplawski 24215640f768SEric Dumazet return pfrag->page; 24229c55e01cSJens Axboe } 24239c55e01cSJens Axboe 242441c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 242541c73a0dSEric Dumazet struct page *page, 242641c73a0dSEric Dumazet unsigned int offset) 242741c73a0dSEric Dumazet { 242841c73a0dSEric Dumazet return spd->nr_pages && 242941c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 243041c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 243141c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 243241c73a0dSEric Dumazet } 243341c73a0dSEric Dumazet 24349c55e01cSJens Axboe /* 24359c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 24369c55e01cSJens Axboe */ 2437a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 243835f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 24394fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 244018aafc62SEric Dumazet bool linear, 24417a67e56fSJarek Poplawski struct sock *sk) 24429c55e01cSJens Axboe { 244341c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2444a108d5f3SDavid S. Miller return true; 24459c55e01cSJens Axboe 24468b9d3728SJarek Poplawski if (linear) { 244718aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 24488b9d3728SJarek Poplawski if (!page) 2449a108d5f3SDavid S. Miller return true; 245041c73a0dSEric Dumazet } 245141c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 245241c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2453a108d5f3SDavid S. Miller return false; 245441c73a0dSEric Dumazet } 24558b9d3728SJarek Poplawski get_page(page); 24569c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 24574fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 24589c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 24599c55e01cSJens Axboe spd->nr_pages++; 24608b9d3728SJarek Poplawski 2461a108d5f3SDavid S. Miller return false; 24629c55e01cSJens Axboe } 24639c55e01cSJens Axboe 2464a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 24652870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 246618aafc62SEric Dumazet unsigned int *len, 2467d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 246835f3d14dSJens Axboe struct sock *sk, 246935f3d14dSJens Axboe struct pipe_inode_info *pipe) 24709c55e01cSJens Axboe { 24712870c43dSOctavian Purdila if (!*len) 2472a108d5f3SDavid S. Miller return true; 24739c55e01cSJens Axboe 24742870c43dSOctavian Purdila /* skip this segment if already processed */ 24752870c43dSOctavian Purdila if (*off >= plen) { 24762870c43dSOctavian Purdila *off -= plen; 2477a108d5f3SDavid S. Miller return false; 24782870c43dSOctavian Purdila } 24792870c43dSOctavian Purdila 24802870c43dSOctavian Purdila /* ignore any bits we already processed */ 24819ca1b22dSEric Dumazet poff += *off; 24829ca1b22dSEric Dumazet plen -= *off; 24832870c43dSOctavian Purdila *off = 0; 24842870c43dSOctavian Purdila 248518aafc62SEric Dumazet do { 248618aafc62SEric Dumazet unsigned int flen = min(*len, plen); 24872870c43dSOctavian Purdila 248818aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 248918aafc62SEric Dumazet linear, sk)) 2490a108d5f3SDavid S. Miller return true; 249118aafc62SEric Dumazet poff += flen; 249218aafc62SEric Dumazet plen -= flen; 24932870c43dSOctavian Purdila *len -= flen; 249418aafc62SEric Dumazet } while (*len && plen); 24952870c43dSOctavian Purdila 2496a108d5f3SDavid S. Miller return false; 2497db43a282SOctavian Purdila } 24989c55e01cSJens Axboe 24999c55e01cSJens Axboe /* 2500a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 25012870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 25029c55e01cSJens Axboe */ 2503a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 250435f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 250535f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 25062870c43dSOctavian Purdila { 25072870c43dSOctavian Purdila int seg; 2508fa9835e5STom Herbert struct sk_buff *iter; 25099c55e01cSJens Axboe 25101d0c0b32SEric Dumazet /* map the linear part : 25112996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 25122996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 25132996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 25149c55e01cSJens Axboe */ 25152870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 25162870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 25172870c43dSOctavian Purdila skb_headlen(skb), 251818aafc62SEric Dumazet offset, len, spd, 25193a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 25201d0c0b32SEric Dumazet sk, pipe)) 2521a108d5f3SDavid S. Miller return true; 25229c55e01cSJens Axboe 25239c55e01cSJens Axboe /* 25249c55e01cSJens Axboe * then map the fragments 25259c55e01cSJens Axboe */ 25269c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 25279c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 25289c55e01cSJens Axboe 2529ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2530b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 253118aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2532a108d5f3SDavid S. Miller return true; 25339c55e01cSJens Axboe } 25349c55e01cSJens Axboe 2535fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2536fa9835e5STom Herbert if (*offset >= iter->len) { 2537fa9835e5STom Herbert *offset -= iter->len; 2538fa9835e5STom Herbert continue; 2539fa9835e5STom Herbert } 2540fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2541fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2542fa9835e5STom Herbert * case. 2543fa9835e5STom Herbert */ 2544fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2545fa9835e5STom Herbert return true; 2546fa9835e5STom Herbert } 2547fa9835e5STom Herbert 2548a108d5f3SDavid S. Miller return false; 25499c55e01cSJens Axboe } 25509c55e01cSJens Axboe 25519c55e01cSJens Axboe /* 25529c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2553fa9835e5STom Herbert * the fragments, and the frag list. 25549c55e01cSJens Axboe */ 2555a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 25569c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 255725869262SAl Viro unsigned int flags) 25589c55e01cSJens Axboe { 255941c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 256041c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 25619c55e01cSJens Axboe struct splice_pipe_desc spd = { 25629c55e01cSJens Axboe .pages = pages, 25639c55e01cSJens Axboe .partial = partial, 2564047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 256528a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 25669c55e01cSJens Axboe .spd_release = sock_spd_release, 25679c55e01cSJens Axboe }; 256835f3d14dSJens Axboe int ret = 0; 256935f3d14dSJens Axboe 2570fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 25719c55e01cSJens Axboe 2572a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 257325869262SAl Viro ret = splice_to_pipe(pipe, &spd); 25749c55e01cSJens Axboe 257535f3d14dSJens Axboe return ret; 25769c55e01cSJens Axboe } 25772b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 25789c55e01cSJens Axboe 25790739cd28SCong Wang static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg, 25800739cd28SCong Wang struct kvec *vec, size_t num, size_t size) 25810739cd28SCong Wang { 25820739cd28SCong Wang struct socket *sock = sk->sk_socket; 25830739cd28SCong Wang 25840739cd28SCong Wang if (!sock) 25850739cd28SCong Wang return -EINVAL; 25860739cd28SCong Wang return kernel_sendmsg(sock, msg, vec, num, size); 25870739cd28SCong Wang } 25880739cd28SCong Wang 25890739cd28SCong Wang static int sendpage_unlocked(struct sock *sk, struct page *page, int offset, 25900739cd28SCong Wang size_t size, int flags) 25910739cd28SCong Wang { 25920739cd28SCong Wang struct socket *sock = sk->sk_socket; 25930739cd28SCong Wang 25940739cd28SCong Wang if (!sock) 25950739cd28SCong Wang return -EINVAL; 25960739cd28SCong Wang return kernel_sendpage(sock, page, offset, size, flags); 25970739cd28SCong Wang } 25980739cd28SCong Wang 25990739cd28SCong Wang typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg, 26000739cd28SCong Wang struct kvec *vec, size_t num, size_t size); 26010739cd28SCong Wang typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset, 26020739cd28SCong Wang size_t size, int flags); 26030739cd28SCong Wang static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, 26040739cd28SCong Wang int len, sendmsg_func sendmsg, sendpage_func sendpage) 260520bf50deSTom Herbert { 260620bf50deSTom Herbert unsigned int orig_len = len; 260720bf50deSTom Herbert struct sk_buff *head = skb; 260820bf50deSTom Herbert unsigned short fragidx; 260920bf50deSTom Herbert int slen, ret; 261020bf50deSTom Herbert 261120bf50deSTom Herbert do_frag_list: 261220bf50deSTom Herbert 261320bf50deSTom Herbert /* Deal with head data */ 261420bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 261520bf50deSTom Herbert struct kvec kv; 261620bf50deSTom Herbert struct msghdr msg; 261720bf50deSTom Herbert 261820bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 261920bf50deSTom Herbert kv.iov_base = skb->data + offset; 2620db5980d8SJohn Fastabend kv.iov_len = slen; 262120bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2622bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 262320bf50deSTom Herbert 26240739cd28SCong Wang ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked, 26250739cd28SCong Wang sendmsg_unlocked, sk, &msg, &kv, 1, slen); 262620bf50deSTom Herbert if (ret <= 0) 262720bf50deSTom Herbert goto error; 262820bf50deSTom Herbert 262920bf50deSTom Herbert offset += ret; 263020bf50deSTom Herbert len -= ret; 263120bf50deSTom Herbert } 263220bf50deSTom Herbert 263320bf50deSTom Herbert /* All the data was skb head? */ 263420bf50deSTom Herbert if (!len) 263520bf50deSTom Herbert goto out; 263620bf50deSTom Herbert 263720bf50deSTom Herbert /* Make offset relative to start of frags */ 263820bf50deSTom Herbert offset -= skb_headlen(skb); 263920bf50deSTom Herbert 264020bf50deSTom Herbert /* Find where we are in frag list */ 264120bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 264220bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 264320bf50deSTom Herbert 2644d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 264520bf50deSTom Herbert break; 264620bf50deSTom Herbert 2647d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 264820bf50deSTom Herbert } 264920bf50deSTom Herbert 265020bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 265120bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 265220bf50deSTom Herbert 2653d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 265420bf50deSTom Herbert 265520bf50deSTom Herbert while (slen) { 26560739cd28SCong Wang ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked, 26570739cd28SCong Wang sendpage_unlocked, sk, 26580739cd28SCong Wang skb_frag_page(frag), 2659b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 266020bf50deSTom Herbert slen, MSG_DONTWAIT); 266120bf50deSTom Herbert if (ret <= 0) 266220bf50deSTom Herbert goto error; 266320bf50deSTom Herbert 266420bf50deSTom Herbert len -= ret; 266520bf50deSTom Herbert offset += ret; 266620bf50deSTom Herbert slen -= ret; 266720bf50deSTom Herbert } 266820bf50deSTom Herbert 266920bf50deSTom Herbert offset = 0; 267020bf50deSTom Herbert } 267120bf50deSTom Herbert 267220bf50deSTom Herbert if (len) { 267320bf50deSTom Herbert /* Process any frag lists */ 267420bf50deSTom Herbert 267520bf50deSTom Herbert if (skb == head) { 267620bf50deSTom Herbert if (skb_has_frag_list(skb)) { 267720bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 267820bf50deSTom Herbert goto do_frag_list; 267920bf50deSTom Herbert } 268020bf50deSTom Herbert } else if (skb->next) { 268120bf50deSTom Herbert skb = skb->next; 268220bf50deSTom Herbert goto do_frag_list; 268320bf50deSTom Herbert } 268420bf50deSTom Herbert } 268520bf50deSTom Herbert 268620bf50deSTom Herbert out: 268720bf50deSTom Herbert return orig_len - len; 268820bf50deSTom Herbert 268920bf50deSTom Herbert error: 269020bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 269120bf50deSTom Herbert } 26920739cd28SCong Wang 26930739cd28SCong Wang /* Send skb data on a socket. Socket must be locked. */ 26940739cd28SCong Wang int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 26950739cd28SCong Wang int len) 26960739cd28SCong Wang { 26970739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked, 26980739cd28SCong Wang kernel_sendpage_locked); 26990739cd28SCong Wang } 270020bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 270120bf50deSTom Herbert 27020739cd28SCong Wang /* Send skb data on a socket. Socket must be unlocked. */ 27030739cd28SCong Wang int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 27040739cd28SCong Wang { 27050739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked, 27060739cd28SCong Wang sendpage_unlocked); 27070739cd28SCong Wang } 27080739cd28SCong Wang 2709357b40a1SHerbert Xu /** 2710357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2711357b40a1SHerbert Xu * @skb: destination buffer 2712357b40a1SHerbert Xu * @offset: offset in destination 2713357b40a1SHerbert Xu * @from: source buffer 2714357b40a1SHerbert Xu * @len: number of bytes to copy 2715357b40a1SHerbert Xu * 2716357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2717357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2718357b40a1SHerbert Xu * traversing fragment lists and such. 2719357b40a1SHerbert Xu */ 2720357b40a1SHerbert Xu 27210c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2722357b40a1SHerbert Xu { 27231a028e50SDavid S. Miller int start = skb_headlen(skb); 2724fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2725fbb398a8SDavid S. Miller int i, copy; 2726357b40a1SHerbert Xu 2727357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2728357b40a1SHerbert Xu goto fault; 2729357b40a1SHerbert Xu 27301a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2731357b40a1SHerbert Xu if (copy > len) 2732357b40a1SHerbert Xu copy = len; 273327d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2734357b40a1SHerbert Xu if ((len -= copy) == 0) 2735357b40a1SHerbert Xu return 0; 2736357b40a1SHerbert Xu offset += copy; 2737357b40a1SHerbert Xu from += copy; 2738357b40a1SHerbert Xu } 2739357b40a1SHerbert Xu 2740357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2741357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 27421a028e50SDavid S. Miller int end; 2743357b40a1SHerbert Xu 2744547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27451a028e50SDavid S. Miller 27469e903e08SEric Dumazet end = start + skb_frag_size(frag); 2747357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2748c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2749c613c209SWillem de Bruijn struct page *p; 2750357b40a1SHerbert Xu u8 *vaddr; 2751357b40a1SHerbert Xu 2752357b40a1SHerbert Xu if (copy > len) 2753357b40a1SHerbert Xu copy = len; 2754357b40a1SHerbert Xu 2755c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2756b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2757c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2758c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2759c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 276051c56b00SEric Dumazet kunmap_atomic(vaddr); 2761c613c209SWillem de Bruijn } 2762357b40a1SHerbert Xu 2763357b40a1SHerbert Xu if ((len -= copy) == 0) 2764357b40a1SHerbert Xu return 0; 2765357b40a1SHerbert Xu offset += copy; 2766357b40a1SHerbert Xu from += copy; 2767357b40a1SHerbert Xu } 27681a028e50SDavid S. Miller start = end; 2769357b40a1SHerbert Xu } 2770357b40a1SHerbert Xu 2771fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 27721a028e50SDavid S. Miller int end; 2773357b40a1SHerbert Xu 2774547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27751a028e50SDavid S. Miller 2776fbb398a8SDavid S. Miller end = start + frag_iter->len; 2777357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2778357b40a1SHerbert Xu if (copy > len) 2779357b40a1SHerbert Xu copy = len; 2780fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 27811a028e50SDavid S. Miller from, copy)) 2782357b40a1SHerbert Xu goto fault; 2783357b40a1SHerbert Xu if ((len -= copy) == 0) 2784357b40a1SHerbert Xu return 0; 2785357b40a1SHerbert Xu offset += copy; 2786357b40a1SHerbert Xu from += copy; 2787357b40a1SHerbert Xu } 27881a028e50SDavid S. Miller start = end; 2789357b40a1SHerbert Xu } 2790357b40a1SHerbert Xu if (!len) 2791357b40a1SHerbert Xu return 0; 2792357b40a1SHerbert Xu 2793357b40a1SHerbert Xu fault: 2794357b40a1SHerbert Xu return -EFAULT; 2795357b40a1SHerbert Xu } 2796357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2797357b40a1SHerbert Xu 27981da177e4SLinus Torvalds /* Checksum skb data. */ 27992817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 28002817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 28011da177e4SLinus Torvalds { 28021a028e50SDavid S. Miller int start = skb_headlen(skb); 28031a028e50SDavid S. Miller int i, copy = start - offset; 2804fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 28051da177e4SLinus Torvalds int pos = 0; 28061da177e4SLinus Torvalds 28071da177e4SLinus Torvalds /* Checksum header. */ 28081da177e4SLinus Torvalds if (copy > 0) { 28091da177e4SLinus Torvalds if (copy > len) 28101da177e4SLinus Torvalds copy = len; 28112544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 28122544af03SMatteo Croce skb->data + offset, copy, csum); 28131da177e4SLinus Torvalds if ((len -= copy) == 0) 28141da177e4SLinus Torvalds return csum; 28151da177e4SLinus Torvalds offset += copy; 28161da177e4SLinus Torvalds pos = copy; 28171da177e4SLinus Torvalds } 28181da177e4SLinus Torvalds 28191da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 28201a028e50SDavid S. Miller int end; 282151c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 28221da177e4SLinus Torvalds 2823547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28241a028e50SDavid S. Miller 282551c56b00SEric Dumazet end = start + skb_frag_size(frag); 28261da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2827c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2828c613c209SWillem de Bruijn struct page *p; 282944bb9363SAl Viro __wsum csum2; 28301da177e4SLinus Torvalds u8 *vaddr; 28311da177e4SLinus Torvalds 28321da177e4SLinus Torvalds if (copy > len) 28331da177e4SLinus Torvalds copy = len; 2834c613c209SWillem de Bruijn 2835c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2836b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2837c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2838c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 28392544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 28402544af03SMatteo Croce csum_partial_ext, 28412544af03SMatteo Croce vaddr + p_off, p_len, 0); 284251c56b00SEric Dumazet kunmap_atomic(vaddr); 28432544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 28442544af03SMatteo Croce csum_block_add_ext, csum, 28452544af03SMatteo Croce csum2, pos, p_len); 2846c613c209SWillem de Bruijn pos += p_len; 2847c613c209SWillem de Bruijn } 2848c613c209SWillem de Bruijn 28491da177e4SLinus Torvalds if (!(len -= copy)) 28501da177e4SLinus Torvalds return csum; 28511da177e4SLinus Torvalds offset += copy; 28521da177e4SLinus Torvalds } 28531a028e50SDavid S. Miller start = end; 28541da177e4SLinus Torvalds } 28551da177e4SLinus Torvalds 2856fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 28571a028e50SDavid S. Miller int end; 28581da177e4SLinus Torvalds 2859547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28601a028e50SDavid S. Miller 2861fbb398a8SDavid S. Miller end = start + frag_iter->len; 28621da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 28635f92a738SAl Viro __wsum csum2; 28641da177e4SLinus Torvalds if (copy > len) 28651da177e4SLinus Torvalds copy = len; 28662817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 28672817a336SDaniel Borkmann copy, 0, ops); 28682544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 28692544af03SMatteo Croce csum, csum2, pos, copy); 28701da177e4SLinus Torvalds if ((len -= copy) == 0) 28711da177e4SLinus Torvalds return csum; 28721da177e4SLinus Torvalds offset += copy; 28731da177e4SLinus Torvalds pos += copy; 28741da177e4SLinus Torvalds } 28751a028e50SDavid S. Miller start = end; 28761da177e4SLinus Torvalds } 287709a62660SKris Katterjohn BUG_ON(len); 28781da177e4SLinus Torvalds 28791da177e4SLinus Torvalds return csum; 28801da177e4SLinus Torvalds } 28812817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 28822817a336SDaniel Borkmann 28832817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 28842817a336SDaniel Borkmann int len, __wsum csum) 28852817a336SDaniel Borkmann { 28862817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2887cea80ea8SDaniel Borkmann .update = csum_partial_ext, 28882817a336SDaniel Borkmann .combine = csum_block_add_ext, 28892817a336SDaniel Borkmann }; 28902817a336SDaniel Borkmann 28912817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 28922817a336SDaniel Borkmann } 2893b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 28941da177e4SLinus Torvalds 28951da177e4SLinus Torvalds /* Both of above in one bottle. */ 28961da177e4SLinus Torvalds 289781d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 28988d5930dfSAl Viro u8 *to, int len) 28991da177e4SLinus Torvalds { 29001a028e50SDavid S. Miller int start = skb_headlen(skb); 29011a028e50SDavid S. Miller int i, copy = start - offset; 2902fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 29031da177e4SLinus Torvalds int pos = 0; 29048d5930dfSAl Viro __wsum csum = 0; 29051da177e4SLinus Torvalds 29061da177e4SLinus Torvalds /* Copy header. */ 29071da177e4SLinus Torvalds if (copy > 0) { 29081da177e4SLinus Torvalds if (copy > len) 29091da177e4SLinus Torvalds copy = len; 29101da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2911cc44c17bSAl Viro copy); 29121da177e4SLinus Torvalds if ((len -= copy) == 0) 29131da177e4SLinus Torvalds return csum; 29141da177e4SLinus Torvalds offset += copy; 29151da177e4SLinus Torvalds to += copy; 29161da177e4SLinus Torvalds pos = copy; 29171da177e4SLinus Torvalds } 29181da177e4SLinus Torvalds 29191da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 29201a028e50SDavid S. Miller int end; 29211da177e4SLinus Torvalds 2922547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29231a028e50SDavid S. Miller 29249e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 29251da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2926c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2927c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2928c613c209SWillem de Bruijn struct page *p; 29295084205fSAl Viro __wsum csum2; 29301da177e4SLinus Torvalds u8 *vaddr; 29311da177e4SLinus Torvalds 29321da177e4SLinus Torvalds if (copy > len) 29331da177e4SLinus Torvalds copy = len; 2934c613c209SWillem de Bruijn 2935c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2936b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2937c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2938c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2939c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2940c613c209SWillem de Bruijn to + copied, 2941cc44c17bSAl Viro p_len); 294251c56b00SEric Dumazet kunmap_atomic(vaddr); 29431da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2944c613c209SWillem de Bruijn pos += p_len; 2945c613c209SWillem de Bruijn } 2946c613c209SWillem de Bruijn 29471da177e4SLinus Torvalds if (!(len -= copy)) 29481da177e4SLinus Torvalds return csum; 29491da177e4SLinus Torvalds offset += copy; 29501da177e4SLinus Torvalds to += copy; 29511da177e4SLinus Torvalds } 29521a028e50SDavid S. Miller start = end; 29531da177e4SLinus Torvalds } 29541da177e4SLinus Torvalds 2955fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 295681d77662SAl Viro __wsum csum2; 29571a028e50SDavid S. Miller int end; 29581da177e4SLinus Torvalds 2959547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29601a028e50SDavid S. Miller 2961fbb398a8SDavid S. Miller end = start + frag_iter->len; 29621da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 29631da177e4SLinus Torvalds if (copy > len) 29641da177e4SLinus Torvalds copy = len; 2965fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 29661a028e50SDavid S. Miller offset - start, 29678d5930dfSAl Viro to, copy); 29681da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 29691da177e4SLinus Torvalds if ((len -= copy) == 0) 29701da177e4SLinus Torvalds return csum; 29711da177e4SLinus Torvalds offset += copy; 29721da177e4SLinus Torvalds to += copy; 29731da177e4SLinus Torvalds pos += copy; 29741da177e4SLinus Torvalds } 29751a028e50SDavid S. Miller start = end; 29761da177e4SLinus Torvalds } 297709a62660SKris Katterjohn BUG_ON(len); 29781da177e4SLinus Torvalds return csum; 29791da177e4SLinus Torvalds } 2980b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 29811da177e4SLinus Torvalds 298249f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 298349f8e832SCong Wang { 298449f8e832SCong Wang __sum16 sum; 298549f8e832SCong Wang 298649f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 298714641931SCong Wang /* See comments in __skb_checksum_complete(). */ 298849f8e832SCong Wang if (likely(!sum)) { 298949f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 299049f8e832SCong Wang !skb->csum_complete_sw) 29917fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 299249f8e832SCong Wang } 299349f8e832SCong Wang if (!skb_shared(skb)) 299449f8e832SCong Wang skb->csum_valid = !sum; 299549f8e832SCong Wang return sum; 299649f8e832SCong Wang } 299749f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 299849f8e832SCong Wang 299914641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 300014641931SCong Wang * which has been changed from the hardware checksum, for example, by 300114641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 300214641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 300314641931SCong Wang * 300414641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 300514641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 300614641931SCong Wang * shared. 300714641931SCong Wang */ 300849f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 300949f8e832SCong Wang { 301049f8e832SCong Wang __wsum csum; 301149f8e832SCong Wang __sum16 sum; 301249f8e832SCong Wang 301349f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 301449f8e832SCong Wang 301549f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 301614641931SCong Wang /* This check is inverted, because we already knew the hardware 301714641931SCong Wang * checksum is invalid before calling this function. So, if the 301814641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 301914641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 302014641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 302114641931SCong Wang * when moving skb->data around. 302214641931SCong Wang */ 302349f8e832SCong Wang if (likely(!sum)) { 302449f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 302549f8e832SCong Wang !skb->csum_complete_sw) 30267fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 302749f8e832SCong Wang } 302849f8e832SCong Wang 302949f8e832SCong Wang if (!skb_shared(skb)) { 303049f8e832SCong Wang /* Save full packet checksum */ 303149f8e832SCong Wang skb->csum = csum; 303249f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 303349f8e832SCong Wang skb->csum_complete_sw = 1; 303449f8e832SCong Wang skb->csum_valid = !sum; 303549f8e832SCong Wang } 303649f8e832SCong Wang 303749f8e832SCong Wang return sum; 303849f8e832SCong Wang } 303949f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 304049f8e832SCong Wang 30419617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 30429617813dSDavide Caratti { 30439617813dSDavide Caratti net_warn_ratelimited( 30449617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 30459617813dSDavide Caratti __func__); 30469617813dSDavide Caratti return 0; 30479617813dSDavide Caratti } 30489617813dSDavide Caratti 30499617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 30509617813dSDavide Caratti int offset, int len) 30519617813dSDavide Caratti { 30529617813dSDavide Caratti net_warn_ratelimited( 30539617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 30549617813dSDavide Caratti __func__); 30559617813dSDavide Caratti return 0; 30569617813dSDavide Caratti } 30579617813dSDavide Caratti 30589617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 30599617813dSDavide Caratti .update = warn_crc32c_csum_update, 30609617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 30619617813dSDavide Caratti }; 30629617813dSDavide Caratti 30639617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 30649617813dSDavide Caratti &default_crc32c_ops; 30659617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 30669617813dSDavide Caratti 3067af2806f8SThomas Graf /** 3068af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 3069af2806f8SThomas Graf * @from: source buffer 3070af2806f8SThomas Graf * 3071af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 3072af2806f8SThomas Graf * into skb_zerocopy(). 3073af2806f8SThomas Graf */ 3074af2806f8SThomas Graf unsigned int 3075af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 3076af2806f8SThomas Graf { 3077af2806f8SThomas Graf unsigned int hlen = 0; 3078af2806f8SThomas Graf 3079af2806f8SThomas Graf if (!from->head_frag || 3080af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 3081a17ad096SPravin B Shelar skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) { 3082af2806f8SThomas Graf hlen = skb_headlen(from); 3083a17ad096SPravin B Shelar if (!hlen) 3084a17ad096SPravin B Shelar hlen = from->len; 3085a17ad096SPravin B Shelar } 3086af2806f8SThomas Graf 3087af2806f8SThomas Graf if (skb_has_frag_list(from)) 3088af2806f8SThomas Graf hlen = from->len; 3089af2806f8SThomas Graf 3090af2806f8SThomas Graf return hlen; 3091af2806f8SThomas Graf } 3092af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 3093af2806f8SThomas Graf 3094af2806f8SThomas Graf /** 3095af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 3096af2806f8SThomas Graf * @to: destination buffer 30977fceb4deSMasanari Iida * @from: source buffer 3098af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 3099af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 3100af2806f8SThomas Graf * 3101af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 3102af2806f8SThomas Graf * to the frags in the source buffer. 3103af2806f8SThomas Graf * 3104af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 3105af2806f8SThomas Graf * headroom in the `to` buffer. 310636d5fe6aSZoltan Kiss * 310736d5fe6aSZoltan Kiss * Return value: 310836d5fe6aSZoltan Kiss * 0: everything is OK 310936d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 311036d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 3111af2806f8SThomas Graf */ 311236d5fe6aSZoltan Kiss int 311336d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 3114af2806f8SThomas Graf { 3115af2806f8SThomas Graf int i, j = 0; 3116af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 311736d5fe6aSZoltan Kiss int ret; 3118af2806f8SThomas Graf struct page *page; 3119af2806f8SThomas Graf unsigned int offset; 3120af2806f8SThomas Graf 3121af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 3122af2806f8SThomas Graf 3123af2806f8SThomas Graf /* dont bother with small payloads */ 312436d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 312536d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 3126af2806f8SThomas Graf 3127af2806f8SThomas Graf if (hlen) { 312836d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 312936d5fe6aSZoltan Kiss if (unlikely(ret)) 313036d5fe6aSZoltan Kiss return ret; 3131af2806f8SThomas Graf len -= hlen; 3132af2806f8SThomas Graf } else { 3133af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 3134af2806f8SThomas Graf if (plen) { 3135af2806f8SThomas Graf page = virt_to_head_page(from->head); 3136af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 3137af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 3138af2806f8SThomas Graf get_page(page); 3139af2806f8SThomas Graf j = 1; 3140af2806f8SThomas Graf len -= plen; 3141af2806f8SThomas Graf } 3142af2806f8SThomas Graf } 3143af2806f8SThomas Graf 3144af2806f8SThomas Graf to->truesize += len + plen; 3145af2806f8SThomas Graf to->len += len + plen; 3146af2806f8SThomas Graf to->data_len += len + plen; 3147af2806f8SThomas Graf 314836d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 314936d5fe6aSZoltan Kiss skb_tx_error(from); 315036d5fe6aSZoltan Kiss return -ENOMEM; 315136d5fe6aSZoltan Kiss } 31521f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 315336d5fe6aSZoltan Kiss 3154af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 3155d8e18a51SMatthew Wilcox (Oracle) int size; 3156d8e18a51SMatthew Wilcox (Oracle) 3157af2806f8SThomas Graf if (!len) 3158af2806f8SThomas Graf break; 3159af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 3160d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 3161d8e18a51SMatthew Wilcox (Oracle) len); 3162d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3163d8e18a51SMatthew Wilcox (Oracle) len -= size; 3164af2806f8SThomas Graf skb_frag_ref(to, j); 3165af2806f8SThomas Graf j++; 3166af2806f8SThomas Graf } 3167af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 316836d5fe6aSZoltan Kiss 316936d5fe6aSZoltan Kiss return 0; 3170af2806f8SThomas Graf } 3171af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3172af2806f8SThomas Graf 31731da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 31741da177e4SLinus Torvalds { 3175d3bc23e7SAl Viro __wsum csum; 31761da177e4SLinus Torvalds long csstart; 31771da177e4SLinus Torvalds 317884fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 317955508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 31801da177e4SLinus Torvalds else 31811da177e4SLinus Torvalds csstart = skb_headlen(skb); 31821da177e4SLinus Torvalds 318309a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 31841da177e4SLinus Torvalds 3185d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 31861da177e4SLinus Torvalds 31871da177e4SLinus Torvalds csum = 0; 31881da177e4SLinus Torvalds if (csstart != skb->len) 31891da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 31908d5930dfSAl Viro skb->len - csstart); 31911da177e4SLinus Torvalds 319284fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3193ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 31941da177e4SLinus Torvalds 3195d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 31961da177e4SLinus Torvalds } 31971da177e4SLinus Torvalds } 3198b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 31991da177e4SLinus Torvalds 32001da177e4SLinus Torvalds /** 32011da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 32021da177e4SLinus Torvalds * @list: list to dequeue from 32031da177e4SLinus Torvalds * 32041da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 32051da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 32061da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32071da177e4SLinus Torvalds */ 32081da177e4SLinus Torvalds 32091da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 32101da177e4SLinus Torvalds { 32111da177e4SLinus Torvalds unsigned long flags; 32121da177e4SLinus Torvalds struct sk_buff *result; 32131da177e4SLinus Torvalds 32141da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32151da177e4SLinus Torvalds result = __skb_dequeue(list); 32161da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32171da177e4SLinus Torvalds return result; 32181da177e4SLinus Torvalds } 3219b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 32201da177e4SLinus Torvalds 32211da177e4SLinus Torvalds /** 32221da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 32231da177e4SLinus Torvalds * @list: list to dequeue from 32241da177e4SLinus Torvalds * 32251da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 32261da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 32271da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32281da177e4SLinus Torvalds */ 32291da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 32301da177e4SLinus Torvalds { 32311da177e4SLinus Torvalds unsigned long flags; 32321da177e4SLinus Torvalds struct sk_buff *result; 32331da177e4SLinus Torvalds 32341da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32351da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 32361da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32371da177e4SLinus Torvalds return result; 32381da177e4SLinus Torvalds } 3239b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 32401da177e4SLinus Torvalds 32411da177e4SLinus Torvalds /** 32421da177e4SLinus Torvalds * skb_queue_purge - empty a list 32431da177e4SLinus Torvalds * @list: list to empty 32441da177e4SLinus Torvalds * 32451da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 32461da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 32471da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 32481da177e4SLinus Torvalds */ 32491da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 32501da177e4SLinus Torvalds { 32511da177e4SLinus Torvalds struct sk_buff *skb; 32521da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 32531da177e4SLinus Torvalds kfree_skb(skb); 32541da177e4SLinus Torvalds } 3255b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 32561da177e4SLinus Torvalds 32571da177e4SLinus Torvalds /** 32589f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 32599f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3260385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 32619f5afeaeSYaogong Wang * 32629f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 32639f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 32649f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 32659f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 32669f5afeaeSYaogong Wang */ 3267385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 32689f5afeaeSYaogong Wang { 32697c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3270385114deSPeter Oskolkov unsigned int sum = 0; 32719f5afeaeSYaogong Wang 32727c90584cSEric Dumazet while (p) { 32737c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 32747c90584cSEric Dumazet 32757c90584cSEric Dumazet p = rb_next(p); 32767c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3277385114deSPeter Oskolkov sum += skb->truesize; 32789f5afeaeSYaogong Wang kfree_skb(skb); 32797c90584cSEric Dumazet } 3280385114deSPeter Oskolkov return sum; 32819f5afeaeSYaogong Wang } 32829f5afeaeSYaogong Wang 32839f5afeaeSYaogong Wang /** 32841da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 32851da177e4SLinus Torvalds * @list: list to use 32861da177e4SLinus Torvalds * @newsk: buffer to queue 32871da177e4SLinus Torvalds * 32881da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 32891da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 32901da177e4SLinus Torvalds * safely. 32911da177e4SLinus Torvalds * 32921da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 32931da177e4SLinus Torvalds */ 32941da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 32951da177e4SLinus Torvalds { 32961da177e4SLinus Torvalds unsigned long flags; 32971da177e4SLinus Torvalds 32981da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32991da177e4SLinus Torvalds __skb_queue_head(list, newsk); 33001da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33011da177e4SLinus Torvalds } 3302b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 33031da177e4SLinus Torvalds 33041da177e4SLinus Torvalds /** 33051da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 33061da177e4SLinus Torvalds * @list: list to use 33071da177e4SLinus Torvalds * @newsk: buffer to queue 33081da177e4SLinus Torvalds * 33091da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 33101da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33111da177e4SLinus Torvalds * safely. 33121da177e4SLinus Torvalds * 33131da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33141da177e4SLinus Torvalds */ 33151da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 33161da177e4SLinus Torvalds { 33171da177e4SLinus Torvalds unsigned long flags; 33181da177e4SLinus Torvalds 33191da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33201da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 33211da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33221da177e4SLinus Torvalds } 3323b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 33248728b834SDavid S. Miller 33251da177e4SLinus Torvalds /** 33261da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 33271da177e4SLinus Torvalds * @skb: buffer to remove 33288728b834SDavid S. Miller * @list: list to use 33291da177e4SLinus Torvalds * 33308728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 33318728b834SDavid S. Miller * function is atomic with respect to other list locked calls 33321da177e4SLinus Torvalds * 33338728b834SDavid S. Miller * You must know what list the SKB is on. 33341da177e4SLinus Torvalds */ 33358728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 33361da177e4SLinus Torvalds { 33371da177e4SLinus Torvalds unsigned long flags; 33381da177e4SLinus Torvalds 33391da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33408728b834SDavid S. Miller __skb_unlink(skb, list); 33411da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33421da177e4SLinus Torvalds } 3343b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 33441da177e4SLinus Torvalds 33451da177e4SLinus Torvalds /** 33461da177e4SLinus Torvalds * skb_append - append a buffer 33471da177e4SLinus Torvalds * @old: buffer to insert after 33481da177e4SLinus Torvalds * @newsk: buffer to insert 33498728b834SDavid S. Miller * @list: list to use 33501da177e4SLinus Torvalds * 33511da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 33521da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 33531da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33541da177e4SLinus Torvalds */ 33558728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 33561da177e4SLinus Torvalds { 33571da177e4SLinus Torvalds unsigned long flags; 33581da177e4SLinus Torvalds 33598728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 33607de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 33618728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 33621da177e4SLinus Torvalds } 3363b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 33641da177e4SLinus Torvalds 33651da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 33661da177e4SLinus Torvalds struct sk_buff* skb1, 33671da177e4SLinus Torvalds const u32 len, const int pos) 33681da177e4SLinus Torvalds { 33691da177e4SLinus Torvalds int i; 33701da177e4SLinus Torvalds 3371d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3372d626f62bSArnaldo Carvalho de Melo pos - len); 33731da177e4SLinus Torvalds /* And move data appendix as is. */ 33741da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 33751da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 33761da177e4SLinus Torvalds 33771da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 33781da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 33791da177e4SLinus Torvalds skb1->data_len = skb->data_len; 33801da177e4SLinus Torvalds skb1->len += skb1->data_len; 33811da177e4SLinus Torvalds skb->data_len = 0; 33821da177e4SLinus Torvalds skb->len = len; 338327a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 33841da177e4SLinus Torvalds } 33851da177e4SLinus Torvalds 33861da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 33871da177e4SLinus Torvalds struct sk_buff* skb1, 33881da177e4SLinus Torvalds const u32 len, int pos) 33891da177e4SLinus Torvalds { 33901da177e4SLinus Torvalds int i, k = 0; 33911da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 33921da177e4SLinus Torvalds 33931da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 33941da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 33951da177e4SLinus Torvalds skb->len = len; 33961da177e4SLinus Torvalds skb->data_len = len - pos; 33971da177e4SLinus Torvalds 33981da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 33999e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 34001da177e4SLinus Torvalds 34011da177e4SLinus Torvalds if (pos + size > len) { 34021da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 34031da177e4SLinus Torvalds 34041da177e4SLinus Torvalds if (pos < len) { 34051da177e4SLinus Torvalds /* Split frag. 34061da177e4SLinus Torvalds * We have two variants in this case: 34071da177e4SLinus Torvalds * 1. Move all the frag to the second 34081da177e4SLinus Torvalds * part, if it is possible. F.e. 34091da177e4SLinus Torvalds * this approach is mandatory for TUX, 34101da177e4SLinus Torvalds * where splitting is expensive. 34111da177e4SLinus Torvalds * 2. Split is accurately. We make this. 34121da177e4SLinus Torvalds */ 3413ea2ab693SIan Campbell skb_frag_ref(skb, i); 3414b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 34159e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 34169e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 34171da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34181da177e4SLinus Torvalds } 34191da177e4SLinus Torvalds k++; 34201da177e4SLinus Torvalds } else 34211da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34221da177e4SLinus Torvalds pos += size; 34231da177e4SLinus Torvalds } 34241da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 34251da177e4SLinus Torvalds } 34261da177e4SLinus Torvalds 34271da177e4SLinus Torvalds /** 34281da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 34291da177e4SLinus Torvalds * @skb: the buffer to split 34301da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 34311da177e4SLinus Torvalds * @len: new length for skb 34321da177e4SLinus Torvalds */ 34331da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 34341da177e4SLinus Torvalds { 34351da177e4SLinus Torvalds int pos = skb_headlen(skb); 3436*9b65b17dSTalal Ahmad const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY; 34371da177e4SLinus Torvalds 3438*9b65b17dSTalal Ahmad skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags; 34391f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 34401da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 34411da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 34421da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 34431da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 34441da177e4SLinus Torvalds } 3445b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 34461da177e4SLinus Torvalds 34479f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 34489f782db3SIlpo Järvinen * 34499f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 34509f782db3SIlpo Järvinen */ 3451832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3452832d11c5SIlpo Järvinen { 3453c4777efaSEric Dumazet return skb_unclone_keeptruesize(skb, GFP_ATOMIC); 3454832d11c5SIlpo Järvinen } 3455832d11c5SIlpo Järvinen 3456832d11c5SIlpo Järvinen /** 3457832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3458832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3459832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3460832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3461832d11c5SIlpo Järvinen * 3462832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 346320e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3464832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3465832d11c5SIlpo Järvinen * 3466832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3467832d11c5SIlpo Järvinen * 3468832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3469832d11c5SIlpo Järvinen * to have non-paged data as well. 3470832d11c5SIlpo Järvinen * 3471832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3472832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3473832d11c5SIlpo Järvinen */ 3474832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3475832d11c5SIlpo Järvinen { 3476832d11c5SIlpo Järvinen int from, to, merge, todo; 3477d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3478832d11c5SIlpo Järvinen 3479832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3480f8071cdeSEric Dumazet 3481f8071cdeSEric Dumazet if (skb_headlen(skb)) 3482f8071cdeSEric Dumazet return 0; 34831f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 34841f8b977aSWillem de Bruijn return 0; 3485832d11c5SIlpo Järvinen 3486832d11c5SIlpo Järvinen todo = shiftlen; 3487832d11c5SIlpo Järvinen from = 0; 3488832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3489832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3490832d11c5SIlpo Järvinen 3491832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3492832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3493832d11c5SIlpo Järvinen */ 3494832d11c5SIlpo Järvinen if (!to || 3495ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3496b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3497832d11c5SIlpo Järvinen merge = -1; 3498832d11c5SIlpo Järvinen } else { 3499832d11c5SIlpo Järvinen merge = to - 1; 3500832d11c5SIlpo Järvinen 35019e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3502832d11c5SIlpo Järvinen if (todo < 0) { 3503832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3504832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3505832d11c5SIlpo Järvinen return 0; 3506832d11c5SIlpo Järvinen 35079f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 35089f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3509832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3510832d11c5SIlpo Järvinen 35119e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 35129e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3513b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3514832d11c5SIlpo Järvinen 3515832d11c5SIlpo Järvinen goto onlymerged; 3516832d11c5SIlpo Järvinen } 3517832d11c5SIlpo Järvinen 3518832d11c5SIlpo Järvinen from++; 3519832d11c5SIlpo Järvinen } 3520832d11c5SIlpo Järvinen 3521832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3522832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3523832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3524832d11c5SIlpo Järvinen return 0; 3525832d11c5SIlpo Järvinen 3526832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3527832d11c5SIlpo Järvinen return 0; 3528832d11c5SIlpo Järvinen 3529832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3530832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3531832d11c5SIlpo Järvinen return 0; 3532832d11c5SIlpo Järvinen 3533832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3534832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3535832d11c5SIlpo Järvinen 35369e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3537832d11c5SIlpo Järvinen *fragto = *fragfrom; 35389e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3539832d11c5SIlpo Järvinen from++; 3540832d11c5SIlpo Järvinen to++; 3541832d11c5SIlpo Järvinen 3542832d11c5SIlpo Järvinen } else { 3543ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3544b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3545b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 35469e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3547832d11c5SIlpo Järvinen 3548b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 35499e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3550832d11c5SIlpo Järvinen todo = 0; 3551832d11c5SIlpo Järvinen 3552832d11c5SIlpo Järvinen to++; 3553832d11c5SIlpo Järvinen break; 3554832d11c5SIlpo Järvinen } 3555832d11c5SIlpo Järvinen } 3556832d11c5SIlpo Järvinen 3557832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3558832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3559832d11c5SIlpo Järvinen 3560832d11c5SIlpo Järvinen if (merge >= 0) { 3561832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3562832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3563832d11c5SIlpo Järvinen 35649e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 35656a5bcd84SIlias Apalodimas __skb_frag_unref(fragfrom, skb->pp_recycle); 3566832d11c5SIlpo Järvinen } 3567832d11c5SIlpo Järvinen 3568832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3569832d11c5SIlpo Järvinen to = 0; 3570832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3571832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3572832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3573832d11c5SIlpo Järvinen 3574832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3575832d11c5SIlpo Järvinen 3576832d11c5SIlpo Järvinen onlymerged: 3577832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3578832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3579832d11c5SIlpo Järvinen */ 3580832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3581832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3582832d11c5SIlpo Järvinen 3583832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3584832d11c5SIlpo Järvinen skb->len -= shiftlen; 3585832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3586832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3587832d11c5SIlpo Järvinen tgt->len += shiftlen; 3588832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3589832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3590832d11c5SIlpo Järvinen 3591832d11c5SIlpo Järvinen return shiftlen; 3592832d11c5SIlpo Järvinen } 3593832d11c5SIlpo Järvinen 3594677e90edSThomas Graf /** 3595677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3596677e90edSThomas Graf * @skb: the buffer to read 3597677e90edSThomas Graf * @from: lower offset of data to be read 3598677e90edSThomas Graf * @to: upper offset of data to be read 3599677e90edSThomas Graf * @st: state variable 3600677e90edSThomas Graf * 3601677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3602677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3603677e90edSThomas Graf */ 3604677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3605677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3606677e90edSThomas Graf { 3607677e90edSThomas Graf st->lower_offset = from; 3608677e90edSThomas Graf st->upper_offset = to; 3609677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3610677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3611677e90edSThomas Graf st->frag_data = NULL; 361297550f6fSWillem de Bruijn st->frag_off = 0; 3613677e90edSThomas Graf } 3614b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3615677e90edSThomas Graf 3616677e90edSThomas Graf /** 3617677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3618677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3619677e90edSThomas Graf * @data: destination pointer for data to be returned 3620677e90edSThomas Graf * @st: state variable 3621677e90edSThomas Graf * 3622bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3623677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3624bc32383cSMathias Krause * the head of the data block to @data and returns the length 3625677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3626677e90edSThomas Graf * offset has been reached. 3627677e90edSThomas Graf * 3628677e90edSThomas Graf * The caller is not required to consume all of the data 3629bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3630677e90edSThomas Graf * of bytes already consumed and the next call to 3631677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3632677e90edSThomas Graf * 363325985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3634e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3635677e90edSThomas Graf * reads of potentially non linear data. 3636677e90edSThomas Graf * 3637bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3638677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3639677e90edSThomas Graf * a stack for this purpose. 3640677e90edSThomas Graf */ 3641677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3642677e90edSThomas Graf struct skb_seq_state *st) 3643677e90edSThomas Graf { 3644677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3645677e90edSThomas Graf skb_frag_t *frag; 3646677e90edSThomas Graf 3647aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3648aeb193eaSWedson Almeida Filho if (st->frag_data) { 3649aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3650aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3651aeb193eaSWedson Almeida Filho } 3652677e90edSThomas Graf return 0; 3653aeb193eaSWedson Almeida Filho } 3654677e90edSThomas Graf 3655677e90edSThomas Graf next_skb: 365695e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3657677e90edSThomas Graf 3658995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 365995e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3660677e90edSThomas Graf return block_limit - abs_offset; 3661677e90edSThomas Graf } 3662677e90edSThomas Graf 3663677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3664677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3665677e90edSThomas Graf 3666677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 366797550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3668677e90edSThomas Graf 366997550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 367097550f6fSWillem de Bruijn 367197550f6fSWillem de Bruijn pg_idx = 0; 367297550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 367397550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 367497550f6fSWillem de Bruijn 367597550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 367697550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 367797550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 367897550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 367997550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 368097550f6fSWillem de Bruijn } 368197550f6fSWillem de Bruijn 368297550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3683677e90edSThomas Graf if (abs_offset < block_limit) { 3684677e90edSThomas Graf if (!st->frag_data) 368597550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3686677e90edSThomas Graf 368797550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3688677e90edSThomas Graf (abs_offset - st->stepped_offset); 3689677e90edSThomas Graf 3690677e90edSThomas Graf return block_limit - abs_offset; 3691677e90edSThomas Graf } 3692677e90edSThomas Graf 3693677e90edSThomas Graf if (st->frag_data) { 369451c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3695677e90edSThomas Graf st->frag_data = NULL; 3696677e90edSThomas Graf } 3697677e90edSThomas Graf 369897550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 369997550f6fSWillem de Bruijn st->frag_off += pg_sz; 370097550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 370197550f6fSWillem de Bruijn st->frag_off = 0; 3702677e90edSThomas Graf st->frag_idx++; 370397550f6fSWillem de Bruijn } 3704677e90edSThomas Graf } 3705677e90edSThomas Graf 37065b5a60daSOlaf Kirch if (st->frag_data) { 370751c56b00SEric Dumazet kunmap_atomic(st->frag_data); 37085b5a60daSOlaf Kirch st->frag_data = NULL; 37095b5a60daSOlaf Kirch } 37105b5a60daSOlaf Kirch 371121dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3712677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 371395e3b24cSHerbert Xu st->frag_idx = 0; 3714677e90edSThomas Graf goto next_skb; 371571b3346dSShyam Iyer } else if (st->cur_skb->next) { 371671b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 371771b3346dSShyam Iyer st->frag_idx = 0; 3718677e90edSThomas Graf goto next_skb; 3719677e90edSThomas Graf } 3720677e90edSThomas Graf 3721677e90edSThomas Graf return 0; 3722677e90edSThomas Graf } 3723b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3724677e90edSThomas Graf 3725677e90edSThomas Graf /** 3726677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3727677e90edSThomas Graf * @st: state variable 3728677e90edSThomas Graf * 3729677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3730677e90edSThomas Graf * returned 0. 3731677e90edSThomas Graf */ 3732677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3733677e90edSThomas Graf { 3734677e90edSThomas Graf if (st->frag_data) 373551c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3736677e90edSThomas Graf } 3737b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3738677e90edSThomas Graf 37393fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 37403fc7e8a6SThomas Graf 37413fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 37423fc7e8a6SThomas Graf struct ts_config *conf, 37433fc7e8a6SThomas Graf struct ts_state *state) 37443fc7e8a6SThomas Graf { 37453fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 37463fc7e8a6SThomas Graf } 37473fc7e8a6SThomas Graf 37483fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 37493fc7e8a6SThomas Graf { 37503fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 37513fc7e8a6SThomas Graf } 37523fc7e8a6SThomas Graf 37533fc7e8a6SThomas Graf /** 37543fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 37553fc7e8a6SThomas Graf * @skb: the buffer to look in 37563fc7e8a6SThomas Graf * @from: search offset 37573fc7e8a6SThomas Graf * @to: search limit 37583fc7e8a6SThomas Graf * @config: textsearch configuration 37593fc7e8a6SThomas Graf * 37603fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 37613fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 37623fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 37633fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 37643fc7e8a6SThomas Graf */ 37653fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3766059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 37673fc7e8a6SThomas Graf { 3768059a2440SBojan Prtvar struct ts_state state; 3769f72b948dSPhil Oester unsigned int ret; 3770f72b948dSPhil Oester 3771b228c9b0SWillem de Bruijn BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb)); 3772b228c9b0SWillem de Bruijn 37733fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 37743fc7e8a6SThomas Graf config->finish = skb_ts_finish; 37753fc7e8a6SThomas Graf 3776059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 37773fc7e8a6SThomas Graf 3778059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3779f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 37803fc7e8a6SThomas Graf } 3781b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 37823fc7e8a6SThomas Graf 3783be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3784be12a1feSHannes Frederic Sowa int offset, size_t size) 3785be12a1feSHannes Frederic Sowa { 3786be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3787be12a1feSHannes Frederic Sowa 3788be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3789be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3790be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3791be12a1feSHannes Frederic Sowa get_page(page); 3792be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3793be12a1feSHannes Frederic Sowa } else { 3794be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3795be12a1feSHannes Frederic Sowa } 3796be12a1feSHannes Frederic Sowa 3797be12a1feSHannes Frederic Sowa return 0; 3798be12a1feSHannes Frederic Sowa } 3799be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3800be12a1feSHannes Frederic Sowa 3801cbb042f9SHerbert Xu /** 3802cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3803cbb042f9SHerbert Xu * @skb: buffer to update 3804cbb042f9SHerbert Xu * @len: length of data pulled 3805cbb042f9SHerbert Xu * 3806cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3807fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 380884fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 380984fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 381084fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3811cbb042f9SHerbert Xu */ 3812af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3813cbb042f9SHerbert Xu { 381431b33dfbSPravin B Shelar unsigned char *data = skb->data; 381531b33dfbSPravin B Shelar 3816cbb042f9SHerbert Xu BUG_ON(len > skb->len); 381731b33dfbSPravin B Shelar __skb_pull(skb, len); 381831b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 381931b33dfbSPravin B Shelar return skb->data; 3820cbb042f9SHerbert Xu } 3821f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3822f94691acSArnaldo Carvalho de Melo 382313acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 382413acc94eSYonghong Song { 382513acc94eSYonghong Song skb_frag_t head_frag; 382613acc94eSYonghong Song struct page *page; 382713acc94eSYonghong Song 382813acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3829d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3830b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3831b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3832d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 383313acc94eSYonghong Song return head_frag; 383413acc94eSYonghong Song } 383513acc94eSYonghong Song 38363a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 38373a1296a3SSteffen Klassert netdev_features_t features, 38383a1296a3SSteffen Klassert unsigned int offset) 38393a1296a3SSteffen Klassert { 38403a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 38413a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 38423a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 38433a1296a3SSteffen Klassert unsigned int delta_len = 0; 38443a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 384553475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 384653475c5dSDongseok Yi int err; 38473a1296a3SSteffen Klassert 38483a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 38493a1296a3SSteffen Klassert 38503a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 38513a1296a3SSteffen Klassert 38523a1296a3SSteffen Klassert do { 38533a1296a3SSteffen Klassert nskb = list_skb; 38543a1296a3SSteffen Klassert list_skb = list_skb->next; 38553a1296a3SSteffen Klassert 385653475c5dSDongseok Yi err = 0; 385753475c5dSDongseok Yi if (skb_shared(nskb)) { 385853475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 385953475c5dSDongseok Yi if (tmp) { 386053475c5dSDongseok Yi consume_skb(nskb); 386153475c5dSDongseok Yi nskb = tmp; 386253475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 386353475c5dSDongseok Yi } else { 386453475c5dSDongseok Yi err = -ENOMEM; 386553475c5dSDongseok Yi } 386653475c5dSDongseok Yi } 386753475c5dSDongseok Yi 38683a1296a3SSteffen Klassert if (!tail) 38693a1296a3SSteffen Klassert skb->next = nskb; 38703a1296a3SSteffen Klassert else 38713a1296a3SSteffen Klassert tail->next = nskb; 38723a1296a3SSteffen Klassert 387353475c5dSDongseok Yi if (unlikely(err)) { 387453475c5dSDongseok Yi nskb->next = list_skb; 387553475c5dSDongseok Yi goto err_linearize; 387653475c5dSDongseok Yi } 387753475c5dSDongseok Yi 38783a1296a3SSteffen Klassert tail = nskb; 38793a1296a3SSteffen Klassert 38803a1296a3SSteffen Klassert delta_len += nskb->len; 38813a1296a3SSteffen Klassert delta_truesize += nskb->truesize; 38823a1296a3SSteffen Klassert 38833a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 38843a1296a3SSteffen Klassert 3885cf673ed0SFlorian Westphal skb_release_head_state(nskb); 38863a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 38873a1296a3SSteffen Klassert 38883a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 38893a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 38903a1296a3SSteffen Klassert nskb->data - tnl_hlen, 38913a1296a3SSteffen Klassert offset + tnl_hlen); 38923a1296a3SSteffen Klassert 38933a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 38943a1296a3SSteffen Klassert __skb_linearize(nskb)) 38953a1296a3SSteffen Klassert goto err_linearize; 38963a1296a3SSteffen Klassert 38973a1296a3SSteffen Klassert } while (list_skb); 38983a1296a3SSteffen Klassert 38993a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 39003a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 39013a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 39023a1296a3SSteffen Klassert 39033a1296a3SSteffen Klassert skb_gso_reset(skb); 39043a1296a3SSteffen Klassert 39053a1296a3SSteffen Klassert skb->prev = tail; 39063a1296a3SSteffen Klassert 39073a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 39083a1296a3SSteffen Klassert __skb_linearize(skb)) 39093a1296a3SSteffen Klassert goto err_linearize; 39103a1296a3SSteffen Klassert 39113a1296a3SSteffen Klassert skb_get(skb); 39123a1296a3SSteffen Klassert 39133a1296a3SSteffen Klassert return skb; 39143a1296a3SSteffen Klassert 39153a1296a3SSteffen Klassert err_linearize: 39163a1296a3SSteffen Klassert kfree_skb_list(skb->next); 39173a1296a3SSteffen Klassert skb->next = NULL; 39183a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 39193a1296a3SSteffen Klassert } 39203a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 39213a1296a3SSteffen Klassert 39223a1296a3SSteffen Klassert int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb) 39233a1296a3SSteffen Klassert { 39243a1296a3SSteffen Klassert if (unlikely(p->len + skb->len >= 65536)) 39253a1296a3SSteffen Klassert return -E2BIG; 39263a1296a3SSteffen Klassert 39273a1296a3SSteffen Klassert if (NAPI_GRO_CB(p)->last == p) 39283a1296a3SSteffen Klassert skb_shinfo(p)->frag_list = skb; 39293a1296a3SSteffen Klassert else 39303a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last->next = skb; 39313a1296a3SSteffen Klassert 39323a1296a3SSteffen Klassert skb_pull(skb, skb_gro_offset(skb)); 39333a1296a3SSteffen Klassert 39343a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last = skb; 39353a1296a3SSteffen Klassert NAPI_GRO_CB(p)->count++; 39363a1296a3SSteffen Klassert p->data_len += skb->len; 39375e10da53SPaolo Abeni 39385e10da53SPaolo Abeni /* sk owenrship - if any - completely transferred to the aggregated packet */ 39395e10da53SPaolo Abeni skb->destructor = NULL; 39403a1296a3SSteffen Klassert p->truesize += skb->truesize; 39413a1296a3SSteffen Klassert p->len += skb->len; 39423a1296a3SSteffen Klassert 39433a1296a3SSteffen Klassert NAPI_GRO_CB(skb)->same_flow = 1; 39443a1296a3SSteffen Klassert 39453a1296a3SSteffen Klassert return 0; 39463a1296a3SSteffen Klassert } 39473a1296a3SSteffen Klassert 3948f4c50d99SHerbert Xu /** 3949f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3950df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3951576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3952f4c50d99SHerbert Xu * 3953f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 39544c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 39554c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3956f4c50d99SHerbert Xu */ 3957df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3958df5771ffSMichael S. Tsirkin netdev_features_t features) 3959f4c50d99SHerbert Xu { 3960f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3961f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 39621a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3963df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3964df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3965df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 39661fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3967f4c50d99SHerbert Xu unsigned int offset = doffset; 3968df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3969802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3970f4c50d99SHerbert Xu unsigned int headroom; 3971802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3972ec5f0615SPravin B Shelar __be16 proto; 397336c98382SAlexander Duyck bool csum, sg; 3974df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3975f4c50d99SHerbert Xu int err = -ENOMEM; 3976f4c50d99SHerbert Xu int i = 0; 3977f4c50d99SHerbert Xu int pos; 3978f4c50d99SHerbert Xu 39793dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 39803dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 39813dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 39823dcbdb13SShmulik Ladkani * non head_frag head. 39833dcbdb13SShmulik Ladkani * 39843dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 39853dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 39863dcbdb13SShmulik Ladkani * head, then the first one has too). 39873dcbdb13SShmulik Ladkani * 39883dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 39893dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 39903dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 39913dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 39923dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 39933dcbdb13SShmulik Ladkani */ 39943dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 39953dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 39963dcbdb13SShmulik Ladkani } 39973dcbdb13SShmulik Ladkani 39985882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 39992f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 4000ec5f0615SPravin B Shelar if (unlikely(!proto)) 4001ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 4002ec5f0615SPravin B Shelar 400336c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 4004f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 40057e2b10c1STom Herbert 400607b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 400707b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 400807b26c94SSteffen Klassert struct sk_buff *iter; 400943170c4eSIlan Tayari unsigned int frag_len; 401007b26c94SSteffen Klassert 401107b26c94SSteffen Klassert if (!list_skb || 401207b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 401307b26c94SSteffen Klassert goto normal; 401407b26c94SSteffen Klassert 401543170c4eSIlan Tayari /* If we get here then all the required 401643170c4eSIlan Tayari * GSO features except frag_list are supported. 401743170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 401843170c4eSIlan Tayari * with no frag_list. 401943170c4eSIlan Tayari * Currently we can do that only when the buffers don't 402043170c4eSIlan Tayari * have a linear part and all the buffers except 402143170c4eSIlan Tayari * the last are of the same length. 402207b26c94SSteffen Klassert */ 402343170c4eSIlan Tayari frag_len = list_skb->len; 402407b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 402543170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 402643170c4eSIlan Tayari goto normal; 4027eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 402807b26c94SSteffen Klassert goto normal; 402907b26c94SSteffen Klassert 403007b26c94SSteffen Klassert len -= iter->len; 403107b26c94SSteffen Klassert } 403243170c4eSIlan Tayari 403343170c4eSIlan Tayari if (len != frag_len) 403443170c4eSIlan Tayari goto normal; 403507b26c94SSteffen Klassert } 403607b26c94SSteffen Klassert 4037802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 4038802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 4039802ab55aSAlexander Duyck * now. 4040802ab55aSAlexander Duyck */ 4041802ab55aSAlexander Duyck partial_segs = len / mss; 4042d7fb5a80SAlexander Duyck if (partial_segs > 1) 4043802ab55aSAlexander Duyck mss *= partial_segs; 4044d7fb5a80SAlexander Duyck else 4045d7fb5a80SAlexander Duyck partial_segs = 0; 4046802ab55aSAlexander Duyck } 4047802ab55aSAlexander Duyck 404807b26c94SSteffen Klassert normal: 4049df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 4050df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 4051f4c50d99SHerbert Xu 4052f4c50d99SHerbert Xu do { 4053f4c50d99SHerbert Xu struct sk_buff *nskb; 40548cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 4055c8884eddSHerbert Xu int hsize; 4056f4c50d99SHerbert Xu int size; 4057f4c50d99SHerbert Xu 40583953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 40593953c46cSMarcelo Ricardo Leitner len = list_skb->len; 40603953c46cSMarcelo Ricardo Leitner } else { 4061df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 4062f4c50d99SHerbert Xu if (len > mss) 4063f4c50d99SHerbert Xu len = mss; 40643953c46cSMarcelo Ricardo Leitner } 4065f4c50d99SHerbert Xu 4066df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 4067f4c50d99SHerbert Xu 4068dbd50f23SXin Long if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) && 40691a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 40701a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 407189319d38SHerbert Xu 40729d8506ccSHerbert Xu i = 0; 40731a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 40741a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 40751fd819ecSMichael S. Tsirkin frag_skb = list_skb; 40761a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 40779d8506ccSHerbert Xu 40789d8506ccSHerbert Xu while (pos < offset + len) { 40799d8506ccSHerbert Xu BUG_ON(i >= nfrags); 40809d8506ccSHerbert Xu 40814e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 40829d8506ccSHerbert Xu if (pos + size > offset + len) 40839d8506ccSHerbert Xu break; 40849d8506ccSHerbert Xu 40859d8506ccSHerbert Xu i++; 40869d8506ccSHerbert Xu pos += size; 40874e1beba1SMichael S. Tsirkin frag++; 40889d8506ccSHerbert Xu } 40899d8506ccSHerbert Xu 40901a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 40911a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 409289319d38SHerbert Xu 4093f4c50d99SHerbert Xu if (unlikely(!nskb)) 4094f4c50d99SHerbert Xu goto err; 4095f4c50d99SHerbert Xu 40969d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 40979d8506ccSHerbert Xu kfree_skb(nskb); 40989d8506ccSHerbert Xu goto err; 40999d8506ccSHerbert Xu } 41009d8506ccSHerbert Xu 4101ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 410289319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 410389319d38SHerbert Xu kfree_skb(nskb); 410489319d38SHerbert Xu goto err; 410589319d38SHerbert Xu } 410689319d38SHerbert Xu 4107ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 410889319d38SHerbert Xu skb_release_head_state(nskb); 410989319d38SHerbert Xu __skb_push(nskb, doffset); 411089319d38SHerbert Xu } else { 411100b229f7SPaolo Abeni if (hsize < 0) 411200b229f7SPaolo Abeni hsize = 0; 4113dbd50f23SXin Long if (hsize > len || !sg) 4114dbd50f23SXin Long hsize = len; 4115dbd50f23SXin Long 4116c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 4117df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 4118c93bdd0eSMel Gorman NUMA_NO_NODE); 411989319d38SHerbert Xu 412089319d38SHerbert Xu if (unlikely(!nskb)) 412189319d38SHerbert Xu goto err; 412289319d38SHerbert Xu 412389319d38SHerbert Xu skb_reserve(nskb, headroom); 412489319d38SHerbert Xu __skb_put(nskb, doffset); 412589319d38SHerbert Xu } 412689319d38SHerbert Xu 4127f4c50d99SHerbert Xu if (segs) 4128f4c50d99SHerbert Xu tail->next = nskb; 4129f4c50d99SHerbert Xu else 4130f4c50d99SHerbert Xu segs = nskb; 4131f4c50d99SHerbert Xu tail = nskb; 4132f4c50d99SHerbert Xu 4133df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 4134f4c50d99SHerbert Xu 4135030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 4136fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 413768c33163SPravin B Shelar 4138df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 413968c33163SPravin B Shelar nskb->data - tnl_hlen, 414068c33163SPravin B Shelar doffset + tnl_hlen); 414189319d38SHerbert Xu 41429d8506ccSHerbert Xu if (nskb->len == len + doffset) 41431cdbcb79SSimon Horman goto perform_csum_check; 414489319d38SHerbert Xu 41457fbeffedSAlexander Duyck if (!sg) { 41461454c9faSYadu Kishore if (!csum) { 41477fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 41486f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 414976443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 415076443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 41511454c9faSYadu Kishore skb_put(nskb, 41521454c9faSYadu Kishore len), 41538d5930dfSAl Viro len); 41547e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 4155de843723STom Herbert skb_headroom(nskb) + doffset; 41561454c9faSYadu Kishore } else { 41571454c9faSYadu Kishore skb_copy_bits(head_skb, offset, 41581454c9faSYadu Kishore skb_put(nskb, len), 41591454c9faSYadu Kishore len); 41601454c9faSYadu Kishore } 4161f4c50d99SHerbert Xu continue; 4162f4c50d99SHerbert Xu } 4163f4c50d99SHerbert Xu 41648cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 4165f4c50d99SHerbert Xu 4166df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 4167d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4168f4c50d99SHerbert Xu 416906b4feb3SJonathan Lemon skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags & 417006b4feb3SJonathan Lemon SKBFL_SHARED_FRAG; 4171cef401deSEric Dumazet 4172bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4173bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4174bf5c25d6SWillem de Bruijn goto err; 4175bf5c25d6SWillem de Bruijn 41769d8506ccSHerbert Xu while (pos < offset + len) { 41779d8506ccSHerbert Xu if (i >= nfrags) { 41789d8506ccSHerbert Xu i = 0; 41791a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 41801a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 41811fd819ecSMichael S. Tsirkin frag_skb = list_skb; 418213acc94eSYonghong Song if (!skb_headlen(list_skb)) { 41839d8506ccSHerbert Xu BUG_ON(!nfrags); 418413acc94eSYonghong Song } else { 418513acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 41869d8506ccSHerbert Xu 418713acc94eSYonghong Song /* to make room for head_frag. */ 418813acc94eSYonghong Song i--; 418913acc94eSYonghong Song frag--; 419013acc94eSYonghong Song } 4191bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4192bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4193bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4194bf5c25d6SWillem de Bruijn goto err; 4195bf5c25d6SWillem de Bruijn 41961a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 41979d8506ccSHerbert Xu } 41989d8506ccSHerbert Xu 41999d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 42009d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 42019d8506ccSHerbert Xu net_warn_ratelimited( 42029d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 42039d8506ccSHerbert Xu pos, mss); 4204ff907a11SEric Dumazet err = -EINVAL; 42059d8506ccSHerbert Xu goto err; 42069d8506ccSHerbert Xu } 42079d8506ccSHerbert Xu 420813acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 42098cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 42108cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4211f4c50d99SHerbert Xu 4212f4c50d99SHerbert Xu if (pos < offset) { 4213b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 42148cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4215f4c50d99SHerbert Xu } 4216f4c50d99SHerbert Xu 421789319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4218f4c50d99SHerbert Xu 4219f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4220f4c50d99SHerbert Xu i++; 42214e1beba1SMichael S. Tsirkin frag++; 4222f4c50d99SHerbert Xu pos += size; 4223f4c50d99SHerbert Xu } else { 42248cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 422589319d38SHerbert Xu goto skip_fraglist; 4226f4c50d99SHerbert Xu } 4227f4c50d99SHerbert Xu 42288cb19905SMichael S. Tsirkin nskb_frag++; 4229f4c50d99SHerbert Xu } 4230f4c50d99SHerbert Xu 423189319d38SHerbert Xu skip_fraglist: 4232f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4233f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4234f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4235ec5f0615SPravin B Shelar 42361cdbcb79SSimon Horman perform_csum_check: 42377fbeffedSAlexander Duyck if (!csum) { 4238ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4239ff907a11SEric Dumazet __skb_linearize(nskb)) 4240ddff00d4SAlexander Duyck goto err; 4241ff907a11SEric Dumazet 42427fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4243ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 424476443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 424576443456SAlexander Duyck skb_checksum(nskb, doffset, 424676443456SAlexander Duyck nskb->len - doffset, 0); 42477e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 42487e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4249ec5f0615SPravin B Shelar } 4250df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4251f4c50d99SHerbert Xu 4252bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4253bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4254bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4255bec3cfdcSEric Dumazet */ 4256bec3cfdcSEric Dumazet segs->prev = tail; 4257432c856fSToshiaki Makita 4258802ab55aSAlexander Duyck if (partial_segs) { 425907b26c94SSteffen Klassert struct sk_buff *iter; 4260802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 426107b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4262802ab55aSAlexander Duyck 4263802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 426407b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4265802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4266802ab55aSAlexander Duyck 4267802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4268802ab55aSAlexander Duyck * our way back down the stack of protocols. 4269802ab55aSAlexander Duyck */ 427007b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 427107b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 427207b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 427307b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 427407b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 427507b26c94SSteffen Klassert } 427607b26c94SSteffen Klassert 427707b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 427807b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 427907b26c94SSteffen Klassert else if (tail != segs) 428007b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4281802ab55aSAlexander Duyck } 4282802ab55aSAlexander Duyck 4283432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4284432c856fSToshiaki Makita * using skb_set_owner_w(). 4285432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4286432c856fSToshiaki Makita */ 4287432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4288432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4289432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4290432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4291432c856fSToshiaki Makita } 4292f4c50d99SHerbert Xu return segs; 4293f4c50d99SHerbert Xu 4294f4c50d99SHerbert Xu err: 4295289dccbeSEric Dumazet kfree_skb_list(segs); 4296f4c50d99SHerbert Xu return ERR_PTR(err); 4297f4c50d99SHerbert Xu } 4298f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4299f4c50d99SHerbert Xu 4300d4546c25SDavid Miller int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb) 430171d93b39SHerbert Xu { 43028a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 430367147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 430467147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 43058a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 4306715dc1f3SEric Dumazet unsigned int delta_truesize; 43075e10da53SPaolo Abeni unsigned int new_truesize; 4308d4546c25SDavid Miller struct sk_buff *lp; 430971d93b39SHerbert Xu 43100ab03f35SSteffen Klassert if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush)) 431171d93b39SHerbert Xu return -E2BIG; 431271d93b39SHerbert Xu 431329e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 43148a29111cSEric Dumazet pinfo = skb_shinfo(lp); 43158a29111cSEric Dumazet 43168a29111cSEric Dumazet if (headlen <= offset) { 431742da6994SHerbert Xu skb_frag_t *frag; 431866e92fcfSHerbert Xu skb_frag_t *frag2; 43199aaa156cSHerbert Xu int i = skbinfo->nr_frags; 43209aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 432142da6994SHerbert Xu 432266e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 43238a29111cSEric Dumazet goto merge; 432481705ad1SHerbert Xu 43258a29111cSEric Dumazet offset -= headlen; 43269aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 43279aaa156cSHerbert Xu skbinfo->nr_frags = 0; 4328f5572068SHerbert Xu 43299aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 43309aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 433166e92fcfSHerbert Xu do { 433266e92fcfSHerbert Xu *--frag = *--frag2; 433366e92fcfSHerbert Xu } while (--i); 433466e92fcfSHerbert Xu 4335b54c9d5bSJonathan Lemon skb_frag_off_add(frag, offset); 43369e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 433766e92fcfSHerbert Xu 4338715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 43395e10da53SPaolo Abeni new_truesize = SKB_TRUESIZE(skb_end_offset(skb)); 43405e10da53SPaolo Abeni delta_truesize = skb->truesize - new_truesize; 4341715dc1f3SEric Dumazet 43425e10da53SPaolo Abeni skb->truesize = new_truesize; 4343f5572068SHerbert Xu skb->len -= skb->data_len; 4344f5572068SHerbert Xu skb->data_len = 0; 4345f5572068SHerbert Xu 4346715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 43475d38a079SHerbert Xu goto done; 4348d7e8883cSEric Dumazet } else if (skb->head_frag) { 4349d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 4350d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 4351d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 4352d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 4353d7e8883cSEric Dumazet unsigned int first_offset; 4354d7e8883cSEric Dumazet 4355d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 43568a29111cSEric Dumazet goto merge; 4357d7e8883cSEric Dumazet 4358d7e8883cSEric Dumazet first_offset = skb->data - 4359d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 4360d7e8883cSEric Dumazet offset; 4361d7e8883cSEric Dumazet 4362d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 4363d7e8883cSEric Dumazet 4364d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(frag, page); 4365b54c9d5bSJonathan Lemon skb_frag_off_set(frag, first_offset); 4366d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 4367d7e8883cSEric Dumazet 4368d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 4369d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 4370d7e8883cSEric Dumazet 4371af352460SPaolo Abeni new_truesize = SKB_DATA_ALIGN(sizeof(struct sk_buff)); 43725e10da53SPaolo Abeni delta_truesize = skb->truesize - new_truesize; 43735e10da53SPaolo Abeni skb->truesize = new_truesize; 4374d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 4375d7e8883cSEric Dumazet goto done; 43768a29111cSEric Dumazet } 437771d93b39SHerbert Xu 437871d93b39SHerbert Xu merge: 43795e10da53SPaolo Abeni /* sk owenrship - if any - completely transferred to the aggregated packet */ 43805e10da53SPaolo Abeni skb->destructor = NULL; 4381715dc1f3SEric Dumazet delta_truesize = skb->truesize; 438267147ba9SHerbert Xu if (offset > headlen) { 4383d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 4384d1dc7abfSMichal Schmidt 4385b54c9d5bSJonathan Lemon skb_frag_off_add(&skbinfo->frags[0], eat); 43869e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 4387d1dc7abfSMichal Schmidt skb->data_len -= eat; 4388d1dc7abfSMichal Schmidt skb->len -= eat; 438967147ba9SHerbert Xu offset = headlen; 439056035022SHerbert Xu } 439156035022SHerbert Xu 439267147ba9SHerbert Xu __skb_pull(skb, offset); 439356035022SHerbert Xu 439429e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 43958a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 43968a29111cSEric Dumazet else 4397c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 4398c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 4399f4a775d1SEric Dumazet __skb_header_release(skb); 44008a29111cSEric Dumazet lp = p; 440171d93b39SHerbert Xu 44025d38a079SHerbert Xu done: 44035d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 440437fe4732SHerbert Xu p->data_len += len; 4405715dc1f3SEric Dumazet p->truesize += delta_truesize; 440637fe4732SHerbert Xu p->len += len; 44078a29111cSEric Dumazet if (lp != p) { 44088a29111cSEric Dumazet lp->data_len += len; 44098a29111cSEric Dumazet lp->truesize += delta_truesize; 44108a29111cSEric Dumazet lp->len += len; 44118a29111cSEric Dumazet } 441271d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 441371d93b39SHerbert Xu return 0; 441471d93b39SHerbert Xu } 441571d93b39SHerbert Xu 4416df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4417df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4418df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4419df5042f4SFlorian Westphal 4420df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4421df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4422df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4423df5042f4SFlorian Westphal #endif 44244165079bSFlorian Westphal #ifdef CONFIG_XFRM 44254165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 44264165079bSFlorian Westphal #endif 442795a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 442895a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 442995a7233cSPaul Blakey #endif 44303ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 44313ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 44323ee17bc7SMat Martineau #endif 443378476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 443478476d31SJeremy Kerr [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow), 443578476d31SJeremy Kerr #endif 4436df5042f4SFlorian Westphal }; 4437df5042f4SFlorian Westphal 4438df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4439df5042f4SFlorian Westphal { 4440df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4441df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4442df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4443df5042f4SFlorian Westphal #endif 44444165079bSFlorian Westphal #ifdef CONFIG_XFRM 44454165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 44464165079bSFlorian Westphal #endif 444795a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 444895a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 444995a7233cSPaul Blakey #endif 44503ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 44513ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 44523ee17bc7SMat Martineau #endif 445378476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 445478476d31SJeremy Kerr skb_ext_type_len[SKB_EXT_MCTP] + 445578476d31SJeremy Kerr #endif 4456df5042f4SFlorian Westphal 0; 4457df5042f4SFlorian Westphal } 4458df5042f4SFlorian Westphal 4459df5042f4SFlorian Westphal static void skb_extensions_init(void) 4460df5042f4SFlorian Westphal { 4461df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4462df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4463df5042f4SFlorian Westphal 4464df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4465df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4466df5042f4SFlorian Westphal 0, 4467df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4468df5042f4SFlorian Westphal NULL); 4469df5042f4SFlorian Westphal } 4470df5042f4SFlorian Westphal #else 4471df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4472df5042f4SFlorian Westphal #endif 4473df5042f4SFlorian Westphal 44741da177e4SLinus Torvalds void __init skb_init(void) 44751da177e4SLinus Torvalds { 447679a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 44771da177e4SLinus Torvalds sizeof(struct sk_buff), 44781da177e4SLinus Torvalds 0, 4479e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 448079a8a642SKees Cook offsetof(struct sk_buff, cb), 448179a8a642SKees Cook sizeof_field(struct sk_buff, cb), 448220c2df83SPaul Mundt NULL); 4483d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4484d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4485d179cd12SDavid S. Miller 0, 4486e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 448720c2df83SPaul Mundt NULL); 4488df5042f4SFlorian Westphal skb_extensions_init(); 44891da177e4SLinus Torvalds } 44901da177e4SLinus Torvalds 449151c739d1SDavid S. Miller static int 449248a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 449348a1df65SJason A. Donenfeld unsigned int recursion_level) 4494716ea3a7SDavid Howells { 44951a028e50SDavid S. Miller int start = skb_headlen(skb); 44961a028e50SDavid S. Miller int i, copy = start - offset; 4497fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4498716ea3a7SDavid Howells int elt = 0; 4499716ea3a7SDavid Howells 450048a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 450148a1df65SJason A. Donenfeld return -EMSGSIZE; 450248a1df65SJason A. Donenfeld 4503716ea3a7SDavid Howells if (copy > 0) { 4504716ea3a7SDavid Howells if (copy > len) 4505716ea3a7SDavid Howells copy = len; 4506642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4507716ea3a7SDavid Howells elt++; 4508716ea3a7SDavid Howells if ((len -= copy) == 0) 4509716ea3a7SDavid Howells return elt; 4510716ea3a7SDavid Howells offset += copy; 4511716ea3a7SDavid Howells } 4512716ea3a7SDavid Howells 4513716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 45141a028e50SDavid S. Miller int end; 4515716ea3a7SDavid Howells 4516547b792cSIlpo Järvinen WARN_ON(start > offset + len); 45171a028e50SDavid S. Miller 45189e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4519716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4520716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 452148a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 452248a1df65SJason A. Donenfeld return -EMSGSIZE; 4523716ea3a7SDavid Howells 4524716ea3a7SDavid Howells if (copy > len) 4525716ea3a7SDavid Howells copy = len; 4526ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4527b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4528716ea3a7SDavid Howells elt++; 4529716ea3a7SDavid Howells if (!(len -= copy)) 4530716ea3a7SDavid Howells return elt; 4531716ea3a7SDavid Howells offset += copy; 4532716ea3a7SDavid Howells } 45331a028e50SDavid S. Miller start = end; 4534716ea3a7SDavid Howells } 4535716ea3a7SDavid Howells 4536fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 453748a1df65SJason A. Donenfeld int end, ret; 4538716ea3a7SDavid Howells 4539547b792cSIlpo Järvinen WARN_ON(start > offset + len); 45401a028e50SDavid S. Miller 4541fbb398a8SDavid S. Miller end = start + frag_iter->len; 4542716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 454348a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 454448a1df65SJason A. Donenfeld return -EMSGSIZE; 454548a1df65SJason A. Donenfeld 4546716ea3a7SDavid Howells if (copy > len) 4547716ea3a7SDavid Howells copy = len; 454848a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 454948a1df65SJason A. Donenfeld copy, recursion_level + 1); 455048a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 455148a1df65SJason A. Donenfeld return ret; 455248a1df65SJason A. Donenfeld elt += ret; 4553716ea3a7SDavid Howells if ((len -= copy) == 0) 4554716ea3a7SDavid Howells return elt; 4555716ea3a7SDavid Howells offset += copy; 4556716ea3a7SDavid Howells } 45571a028e50SDavid S. Miller start = end; 4558716ea3a7SDavid Howells } 4559716ea3a7SDavid Howells BUG_ON(len); 4560716ea3a7SDavid Howells return elt; 4561716ea3a7SDavid Howells } 4562716ea3a7SDavid Howells 456348a1df65SJason A. Donenfeld /** 456448a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 456548a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 456648a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 456748a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 456848a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 456948a1df65SJason A. Donenfeld * 457048a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 457148a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 457248a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 457348a1df65SJason A. Donenfeld * could not fit. 457448a1df65SJason A. Donenfeld */ 457548a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 457648a1df65SJason A. Donenfeld { 457748a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 457848a1df65SJason A. Donenfeld 457948a1df65SJason A. Donenfeld if (nsg <= 0) 458048a1df65SJason A. Donenfeld return nsg; 458148a1df65SJason A. Donenfeld 458248a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 458348a1df65SJason A. Donenfeld 458448a1df65SJason A. Donenfeld return nsg; 458548a1df65SJason A. Donenfeld } 458648a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 458748a1df65SJason A. Donenfeld 458825a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 458925a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 459025a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 459125a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 459225a91d8dSFan Du * 459325a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 459425a91d8dSFan Du * 1. sg_init_table 459525a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 459625a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 459725a91d8dSFan Du * 459825a91d8dSFan Du * This is equivalent to: 459925a91d8dSFan Du * 1. sg_init_table 460025a91d8dSFan Du * 2. skb_to_sgvec(payload1) 460125a91d8dSFan Du * 3. sg_unmark_end 460225a91d8dSFan Du * 4. skb_to_sgvec(payload2) 460325a91d8dSFan Du * 460425a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 460525a91d8dSFan Du * is more preferable. 460625a91d8dSFan Du */ 460725a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 460825a91d8dSFan Du int offset, int len) 460925a91d8dSFan Du { 461048a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 461125a91d8dSFan Du } 461225a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 461325a91d8dSFan Du 461451c739d1SDavid S. Miller 461551c739d1SDavid S. Miller 4616716ea3a7SDavid Howells /** 4617716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4618716ea3a7SDavid Howells * @skb: The socket buffer to check. 4619716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4620716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4621716ea3a7SDavid Howells * 4622716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4623716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4624716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4625716ea3a7SDavid Howells * 4626716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4627716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4628716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4629716ea3a7SDavid Howells * 4630716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4631716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4632716ea3a7SDavid Howells */ 4633716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4634716ea3a7SDavid Howells { 4635716ea3a7SDavid Howells int copyflag; 4636716ea3a7SDavid Howells int elt; 4637716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4638716ea3a7SDavid Howells 4639716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4640716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4641716ea3a7SDavid Howells * at the moment even if they are anonymous). 4642716ea3a7SDavid Howells */ 4643716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4644c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4645716ea3a7SDavid Howells return -ENOMEM; 4646716ea3a7SDavid Howells 4647716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 464821dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4649716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4650716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4651716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4652716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4653716ea3a7SDavid Howells 4654716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4655716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4656716ea3a7SDavid Howells return -ENOMEM; 4657716ea3a7SDavid Howells 4658716ea3a7SDavid Howells /* Voila! */ 4659716ea3a7SDavid Howells *trailer = skb; 4660716ea3a7SDavid Howells return 1; 4661716ea3a7SDavid Howells } 4662716ea3a7SDavid Howells 4663716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4664716ea3a7SDavid Howells 4665716ea3a7SDavid Howells elt = 1; 4666716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4667716ea3a7SDavid Howells copyflag = 0; 4668716ea3a7SDavid Howells 4669716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4670716ea3a7SDavid Howells int ntail = 0; 4671716ea3a7SDavid Howells 4672716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4673716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4674716ea3a7SDavid Howells * after it. */ 4675716ea3a7SDavid Howells 4676716ea3a7SDavid Howells if (skb_shared(skb1)) 4677716ea3a7SDavid Howells copyflag = 1; 4678716ea3a7SDavid Howells 4679716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4680716ea3a7SDavid Howells 4681716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4682716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 468321dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4684716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4685716ea3a7SDavid Howells ntail = tailbits + 128; 4686716ea3a7SDavid Howells } 4687716ea3a7SDavid Howells 4688716ea3a7SDavid Howells if (copyflag || 4689716ea3a7SDavid Howells skb_cloned(skb1) || 4690716ea3a7SDavid Howells ntail || 4691716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 469221dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4693716ea3a7SDavid Howells struct sk_buff *skb2; 4694716ea3a7SDavid Howells 4695716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4696716ea3a7SDavid Howells if (ntail == 0) 4697716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4698716ea3a7SDavid Howells else 4699716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4700716ea3a7SDavid Howells skb_headroom(skb1), 4701716ea3a7SDavid Howells ntail, 4702716ea3a7SDavid Howells GFP_ATOMIC); 4703716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4704716ea3a7SDavid Howells return -ENOMEM; 4705716ea3a7SDavid Howells 4706716ea3a7SDavid Howells if (skb1->sk) 4707716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4708716ea3a7SDavid Howells 4709716ea3a7SDavid Howells /* Looking around. Are we still alive? 4710716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4711716ea3a7SDavid Howells 4712716ea3a7SDavid Howells skb2->next = skb1->next; 4713716ea3a7SDavid Howells *skb_p = skb2; 4714716ea3a7SDavid Howells kfree_skb(skb1); 4715716ea3a7SDavid Howells skb1 = skb2; 4716716ea3a7SDavid Howells } 4717716ea3a7SDavid Howells elt++; 4718716ea3a7SDavid Howells *trailer = skb1; 4719716ea3a7SDavid Howells skb_p = &skb1->next; 4720716ea3a7SDavid Howells } 4721716ea3a7SDavid Howells 4722716ea3a7SDavid Howells return elt; 4723716ea3a7SDavid Howells } 4724b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4725716ea3a7SDavid Howells 4726b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4727b1faf566SEric Dumazet { 4728b1faf566SEric Dumazet struct sock *sk = skb->sk; 4729b1faf566SEric Dumazet 4730b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4731b1faf566SEric Dumazet } 4732b1faf566SEric Dumazet 47338605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 47348605330aSSoheil Hassas Yeganeh { 47358605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 47368605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 47378605330aSSoheil Hassas Yeganeh */ 47388605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 47398605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 47408605330aSSoheil Hassas Yeganeh } 47418605330aSSoheil Hassas Yeganeh 4742b1faf566SEric Dumazet /* 4743b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4744b1faf566SEric Dumazet */ 4745b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4746b1faf566SEric Dumazet { 4747b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4748ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4749b1faf566SEric Dumazet return -ENOMEM; 4750b1faf566SEric Dumazet 4751b1faf566SEric Dumazet skb_orphan(skb); 4752b1faf566SEric Dumazet skb->sk = sk; 4753b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4754b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 47558605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4756b1faf566SEric Dumazet 4757abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4758abb57ea4SEric Dumazet skb_dst_force(skb); 4759abb57ea4SEric Dumazet 4760b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4761b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 4762e3ae2365SAlexander Aring sk_error_report(sk); 4763b1faf566SEric Dumazet return 0; 4764b1faf566SEric Dumazet } 4765b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4766b1faf566SEric Dumazet 476783a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 476883a1a1a7SSoheil Hassas Yeganeh { 476983a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 477083a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 477183a1a1a7SSoheil Hassas Yeganeh } 477283a1a1a7SSoheil Hassas Yeganeh 4773364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4774364a9e93SWillem de Bruijn { 4775364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 477683a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 477783a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4778997d5c3fSEric Dumazet unsigned long flags; 4779364a9e93SWillem de Bruijn 4780997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4781364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 478238b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 478383a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 478438b25793SSoheil Hassas Yeganeh if (icmp_next) 4785985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 478638b25793SSoheil Hassas Yeganeh } 4787997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4788364a9e93SWillem de Bruijn 478983a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 479083a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 479183a1a1a7SSoheil Hassas Yeganeh 479283a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4793e3ae2365SAlexander Aring sk_error_report(sk); 4794364a9e93SWillem de Bruijn 4795364a9e93SWillem de Bruijn return skb; 4796364a9e93SWillem de Bruijn } 4797364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4798364a9e93SWillem de Bruijn 4799cab41c47SAlexander Duyck /** 4800cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4801cab41c47SAlexander Duyck * @skb: the skb to clone 4802cab41c47SAlexander Duyck * 4803cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4804cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4805cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4806cab41c47SAlexander Duyck * 4807cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4808cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4809cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4810cab41c47SAlexander Duyck * the sk_error_queue. 4811cab41c47SAlexander Duyck */ 481262bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 481362bccb8cSAlexander Duyck { 481462bccb8cSAlexander Duyck struct sock *sk = skb->sk; 481562bccb8cSAlexander Duyck struct sk_buff *clone; 481662bccb8cSAlexander Duyck 481741c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 481862bccb8cSAlexander Duyck return NULL; 481962bccb8cSAlexander Duyck 482062bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 482162bccb8cSAlexander Duyck if (!clone) { 482262bccb8cSAlexander Duyck sock_put(sk); 482362bccb8cSAlexander Duyck return NULL; 482462bccb8cSAlexander Duyck } 482562bccb8cSAlexander Duyck 482662bccb8cSAlexander Duyck clone->sk = sk; 482762bccb8cSAlexander Duyck clone->destructor = sock_efree; 482862bccb8cSAlexander Duyck 482962bccb8cSAlexander Duyck return clone; 483062bccb8cSAlexander Duyck } 483162bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 483262bccb8cSAlexander Duyck 483337846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 483437846ef0SAlexander Duyck struct sock *sk, 48354ef1b286SSoheil Hassas Yeganeh int tstype, 48364ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4837ac45f602SPatrick Ohly { 4838ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4839ac45f602SPatrick Ohly int err; 4840ac45f602SPatrick Ohly 48414ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 48424ef1b286SSoheil Hassas Yeganeh 4843ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4844ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4845ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4846ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4847e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 48484ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 48491862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 48504ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 485109c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 4852ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 4853ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 48544ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 48554ed2d765SWillem de Bruijn } 485629030374SEric Dumazet 4857ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 485829030374SEric Dumazet 4859ac45f602SPatrick Ohly if (err) 4860ac45f602SPatrick Ohly kfree_skb(skb); 4861ac45f602SPatrick Ohly } 486237846ef0SAlexander Duyck 4863b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4864b245be1fSWillem de Bruijn { 4865b245be1fSWillem de Bruijn bool ret; 4866b245be1fSWillem de Bruijn 4867b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4868b245be1fSWillem de Bruijn return true; 4869b245be1fSWillem de Bruijn 4870b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4871b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4872b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4873b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4874b245be1fSWillem de Bruijn return ret; 4875b245be1fSWillem de Bruijn } 4876b245be1fSWillem de Bruijn 487737846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 487837846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 487937846ef0SAlexander Duyck { 488037846ef0SAlexander Duyck struct sock *sk = skb->sk; 488137846ef0SAlexander Duyck 4882b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 488335b99dffSWillem de Bruijn goto err; 4884b245be1fSWillem de Bruijn 48859ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 48869ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 48879ac25fc0SEric Dumazet */ 488841c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 488937846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 48904ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 489137846ef0SAlexander Duyck sock_put(sk); 489235b99dffSWillem de Bruijn return; 489337846ef0SAlexander Duyck } 489435b99dffSWillem de Bruijn 489535b99dffSWillem de Bruijn err: 489635b99dffSWillem de Bruijn kfree_skb(skb); 48979ac25fc0SEric Dumazet } 489837846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 489937846ef0SAlexander Duyck 490037846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 4901e7ed11eeSYousuk Seung const struct sk_buff *ack_skb, 490237846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 490337846ef0SAlexander Duyck struct sock *sk, int tstype) 490437846ef0SAlexander Duyck { 490537846ef0SAlexander Duyck struct sk_buff *skb; 49064ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 490737846ef0SAlexander Duyck 49083a8dd971SWillem de Bruijn if (!sk) 49093a8dd971SWillem de Bruijn return; 49103a8dd971SWillem de Bruijn 4911b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4912b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4913b50a5c70SMiroslav Lichvar return; 4914b50a5c70SMiroslav Lichvar 49153a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 49163a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 491737846ef0SAlexander Duyck return; 491837846ef0SAlexander Duyck 49191c885808SFrancis Yan if (tsonly) { 49201c885808SFrancis Yan #ifdef CONFIG_INET 49211c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 49221c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 49234ef1b286SSoheil Hassas Yeganeh sk->sk_type == SOCK_STREAM) { 4924e7ed11eeSYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb, 4925e7ed11eeSYousuk Seung ack_skb); 49264ef1b286SSoheil Hassas Yeganeh opt_stats = true; 49274ef1b286SSoheil Hassas Yeganeh } else 49281c885808SFrancis Yan #endif 49291c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 49301c885808SFrancis Yan } else { 493137846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 49321c885808SFrancis Yan } 493337846ef0SAlexander Duyck if (!skb) 493437846ef0SAlexander Duyck return; 493537846ef0SAlexander Duyck 493649ca0d8bSWillem de Bruijn if (tsonly) { 4937fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4938fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 493949ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 494049ca0d8bSWillem de Bruijn } 494149ca0d8bSWillem de Bruijn 494249ca0d8bSWillem de Bruijn if (hwtstamps) 494349ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 494449ca0d8bSWillem de Bruijn else 494549ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 494649ca0d8bSWillem de Bruijn 49474ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 494837846ef0SAlexander Duyck } 4949e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4950e7fd2885SWillem de Bruijn 4951e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4952e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4953e7fd2885SWillem de Bruijn { 4954e7ed11eeSYousuk Seung return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk, 4955e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4956e7fd2885SWillem de Bruijn } 4957ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4958ac45f602SPatrick Ohly 49596e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 49606e3e939fSJohannes Berg { 49616e3e939fSJohannes Berg struct sock *sk = skb->sk; 49626e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4963dd4f1072SEric Dumazet int err = 1; 49646e3e939fSJohannes Berg 49656e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 49666e3e939fSJohannes Berg skb->wifi_acked = acked; 49676e3e939fSJohannes Berg 49686e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 49696e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 49706e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 49716e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 49726e3e939fSJohannes Berg 4973dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4974dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4975dd4f1072SEric Dumazet */ 497641c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 49776e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4978dd4f1072SEric Dumazet sock_put(sk); 4979dd4f1072SEric Dumazet } 49806e3e939fSJohannes Berg if (err) 49816e3e939fSJohannes Berg kfree_skb(skb); 49826e3e939fSJohannes Berg } 49836e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 49846e3e939fSJohannes Berg 4985f35d9d8aSRusty Russell /** 4986f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4987f35d9d8aSRusty Russell * @skb: the skb to set 4988f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4989f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4990f35d9d8aSRusty Russell * 4991f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4992f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4993f35d9d8aSRusty Russell * 4994f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4995f35d9d8aSRusty Russell * returns false you should drop the packet. 4996f35d9d8aSRusty Russell */ 4997f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4998f35d9d8aSRusty Russell { 499952b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 500052b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 500152b5d6f5SEric Dumazet 500252b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 500352b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 500452b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 5005f35d9d8aSRusty Russell return false; 5006f35d9d8aSRusty Russell } 5007f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 500852b5d6f5SEric Dumazet skb->csum_start = csum_start; 5009f35d9d8aSRusty Russell skb->csum_offset = off; 5010e5d5decaSJason Wang skb_set_transport_header(skb, start); 5011f35d9d8aSRusty Russell return true; 5012f35d9d8aSRusty Russell } 5013b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 5014f35d9d8aSRusty Russell 5015ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 5016ed1f50c3SPaul Durrant unsigned int max) 5017ed1f50c3SPaul Durrant { 5018ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 5019ed1f50c3SPaul Durrant return 0; 5020ed1f50c3SPaul Durrant 5021ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 5022ed1f50c3SPaul Durrant * won't need to do it again. 5023ed1f50c3SPaul Durrant */ 5024ed1f50c3SPaul Durrant if (max > skb->len) 5025ed1f50c3SPaul Durrant max = skb->len; 5026ed1f50c3SPaul Durrant 5027ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 5028ed1f50c3SPaul Durrant return -ENOMEM; 5029ed1f50c3SPaul Durrant 5030ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 5031ed1f50c3SPaul Durrant return -EPROTO; 5032ed1f50c3SPaul Durrant 5033ed1f50c3SPaul Durrant return 0; 5034ed1f50c3SPaul Durrant } 5035ed1f50c3SPaul Durrant 5036f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 5037f9708b43SJan Beulich 5038f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 5039f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 5040f9708b43SJan Beulich unsigned int off) 5041f9708b43SJan Beulich { 5042f9708b43SJan Beulich int err; 5043f9708b43SJan Beulich 5044161d1792SKees Cook switch (proto) { 5045f9708b43SJan Beulich case IPPROTO_TCP: 5046f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 5047f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 5048f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 5049f9708b43SJan Beulich offsetof(struct tcphdr, 5050f9708b43SJan Beulich check))) 5051f9708b43SJan Beulich err = -EPROTO; 5052f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 5053f9708b43SJan Beulich 5054f9708b43SJan Beulich case IPPROTO_UDP: 5055f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 5056f9708b43SJan Beulich off + sizeof(struct udphdr)); 5057f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 5058f9708b43SJan Beulich offsetof(struct udphdr, 5059f9708b43SJan Beulich check))) 5060f9708b43SJan Beulich err = -EPROTO; 5061f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 5062f9708b43SJan Beulich } 5063f9708b43SJan Beulich 5064f9708b43SJan Beulich return ERR_PTR(-EPROTO); 5065f9708b43SJan Beulich } 5066f9708b43SJan Beulich 5067ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5068ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 5069ed1f50c3SPaul Durrant */ 5070ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 5071ed1f50c3SPaul Durrant 5072f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 5073ed1f50c3SPaul Durrant { 5074ed1f50c3SPaul Durrant unsigned int off; 5075ed1f50c3SPaul Durrant bool fragment; 5076f9708b43SJan Beulich __sum16 *csum; 5077ed1f50c3SPaul Durrant int err; 5078ed1f50c3SPaul Durrant 5079ed1f50c3SPaul Durrant fragment = false; 5080ed1f50c3SPaul Durrant 5081ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5082ed1f50c3SPaul Durrant sizeof(struct iphdr), 5083ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 5084ed1f50c3SPaul Durrant if (err < 0) 5085ed1f50c3SPaul Durrant goto out; 5086ed1f50c3SPaul Durrant 508711f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 5088ed1f50c3SPaul Durrant fragment = true; 5089ed1f50c3SPaul Durrant 5090ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 5091ed1f50c3SPaul Durrant 5092ed1f50c3SPaul Durrant err = -EPROTO; 5093ed1f50c3SPaul Durrant 5094ed1f50c3SPaul Durrant if (fragment) 5095ed1f50c3SPaul Durrant goto out; 5096ed1f50c3SPaul Durrant 5097f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 5098f9708b43SJan Beulich if (IS_ERR(csum)) 5099f9708b43SJan Beulich return PTR_ERR(csum); 5100ed1f50c3SPaul Durrant 5101ed1f50c3SPaul Durrant if (recalculate) 5102f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 5103ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 5104ed1f50c3SPaul Durrant skb->len - off, 5105f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 5106ed1f50c3SPaul Durrant err = 0; 5107ed1f50c3SPaul Durrant 5108ed1f50c3SPaul Durrant out: 5109ed1f50c3SPaul Durrant return err; 5110ed1f50c3SPaul Durrant } 5111ed1f50c3SPaul Durrant 5112ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5113ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 5114ed1f50c3SPaul Durrant */ 5115ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 5116ed1f50c3SPaul Durrant 5117ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 5118ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 5119ed1f50c3SPaul Durrant 5120ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 5121ed1f50c3SPaul Durrant { 5122ed1f50c3SPaul Durrant int err; 5123ed1f50c3SPaul Durrant u8 nexthdr; 5124ed1f50c3SPaul Durrant unsigned int off; 5125ed1f50c3SPaul Durrant unsigned int len; 5126ed1f50c3SPaul Durrant bool fragment; 5127ed1f50c3SPaul Durrant bool done; 5128f9708b43SJan Beulich __sum16 *csum; 5129ed1f50c3SPaul Durrant 5130ed1f50c3SPaul Durrant fragment = false; 5131ed1f50c3SPaul Durrant done = false; 5132ed1f50c3SPaul Durrant 5133ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 5134ed1f50c3SPaul Durrant 5135ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 5136ed1f50c3SPaul Durrant if (err < 0) 5137ed1f50c3SPaul Durrant goto out; 5138ed1f50c3SPaul Durrant 5139ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 5140ed1f50c3SPaul Durrant 5141ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 5142ed1f50c3SPaul Durrant while (off <= len && !done) { 5143ed1f50c3SPaul Durrant switch (nexthdr) { 5144ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 5145ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 5146ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 5147ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 5148ed1f50c3SPaul Durrant 5149ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5150ed1f50c3SPaul Durrant off + 5151ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 5152ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5153ed1f50c3SPaul Durrant if (err < 0) 5154ed1f50c3SPaul Durrant goto out; 5155ed1f50c3SPaul Durrant 5156ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 5157ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5158ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 5159ed1f50c3SPaul Durrant break; 5160ed1f50c3SPaul Durrant } 5161ed1f50c3SPaul Durrant case IPPROTO_AH: { 5162ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 5163ed1f50c3SPaul Durrant 5164ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5165ed1f50c3SPaul Durrant off + 5166ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 5167ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5168ed1f50c3SPaul Durrant if (err < 0) 5169ed1f50c3SPaul Durrant goto out; 5170ed1f50c3SPaul Durrant 5171ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 5172ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5173ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 5174ed1f50c3SPaul Durrant break; 5175ed1f50c3SPaul Durrant } 5176ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 5177ed1f50c3SPaul Durrant struct frag_hdr *hp; 5178ed1f50c3SPaul Durrant 5179ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5180ed1f50c3SPaul Durrant off + 5181ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5182ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5183ed1f50c3SPaul Durrant if (err < 0) 5184ed1f50c3SPaul Durrant goto out; 5185ed1f50c3SPaul Durrant 5186ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5187ed1f50c3SPaul Durrant 5188ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5189ed1f50c3SPaul Durrant fragment = true; 5190ed1f50c3SPaul Durrant 5191ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5192ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5193ed1f50c3SPaul Durrant break; 5194ed1f50c3SPaul Durrant } 5195ed1f50c3SPaul Durrant default: 5196ed1f50c3SPaul Durrant done = true; 5197ed1f50c3SPaul Durrant break; 5198ed1f50c3SPaul Durrant } 5199ed1f50c3SPaul Durrant } 5200ed1f50c3SPaul Durrant 5201ed1f50c3SPaul Durrant err = -EPROTO; 5202ed1f50c3SPaul Durrant 5203ed1f50c3SPaul Durrant if (!done || fragment) 5204ed1f50c3SPaul Durrant goto out; 5205ed1f50c3SPaul Durrant 5206f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5207f9708b43SJan Beulich if (IS_ERR(csum)) 5208f9708b43SJan Beulich return PTR_ERR(csum); 5209ed1f50c3SPaul Durrant 5210ed1f50c3SPaul Durrant if (recalculate) 5211f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5212ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5213f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5214ed1f50c3SPaul Durrant err = 0; 5215ed1f50c3SPaul Durrant 5216ed1f50c3SPaul Durrant out: 5217ed1f50c3SPaul Durrant return err; 5218ed1f50c3SPaul Durrant } 5219ed1f50c3SPaul Durrant 5220ed1f50c3SPaul Durrant /** 5221ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5222ed1f50c3SPaul Durrant * @skb: the skb to set up 5223ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5224ed1f50c3SPaul Durrant */ 5225ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5226ed1f50c3SPaul Durrant { 5227ed1f50c3SPaul Durrant int err; 5228ed1f50c3SPaul Durrant 5229ed1f50c3SPaul Durrant switch (skb->protocol) { 5230ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5231f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5232ed1f50c3SPaul Durrant break; 5233ed1f50c3SPaul Durrant 5234ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5235ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5236ed1f50c3SPaul Durrant break; 5237ed1f50c3SPaul Durrant 5238ed1f50c3SPaul Durrant default: 5239ed1f50c3SPaul Durrant err = -EPROTO; 5240ed1f50c3SPaul Durrant break; 5241ed1f50c3SPaul Durrant } 5242ed1f50c3SPaul Durrant 5243ed1f50c3SPaul Durrant return err; 5244ed1f50c3SPaul Durrant } 5245ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5246ed1f50c3SPaul Durrant 52479afd85c9SLinus Lüssing /** 52489afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 52499afd85c9SLinus Lüssing * @skb: the skb to check 52509afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 52519afd85c9SLinus Lüssing * 52529afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 52539afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 52549afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 52559afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 52569afd85c9SLinus Lüssing * 5257a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5258a516993fSLinus Lüssing * differs from the provided skb. 52599afd85c9SLinus Lüssing */ 52609afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 52619afd85c9SLinus Lüssing unsigned int transport_len) 52629afd85c9SLinus Lüssing { 52639afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52649afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 52659afd85c9SLinus Lüssing int ret; 52669afd85c9SLinus Lüssing 5267a516993fSLinus Lüssing if (skb->len < len) 52689afd85c9SLinus Lüssing return NULL; 5269a516993fSLinus Lüssing else if (skb->len == len) 52709afd85c9SLinus Lüssing return skb; 52719afd85c9SLinus Lüssing 52729afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 52739afd85c9SLinus Lüssing if (!skb_chk) 52749afd85c9SLinus Lüssing return NULL; 52759afd85c9SLinus Lüssing 52769afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 52779afd85c9SLinus Lüssing if (ret) { 52789afd85c9SLinus Lüssing kfree_skb(skb_chk); 52799afd85c9SLinus Lüssing return NULL; 52809afd85c9SLinus Lüssing } 52819afd85c9SLinus Lüssing 52829afd85c9SLinus Lüssing return skb_chk; 52839afd85c9SLinus Lüssing } 52849afd85c9SLinus Lüssing 52859afd85c9SLinus Lüssing /** 52869afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 52879afd85c9SLinus Lüssing * @skb: the skb to check 52889afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 52899afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 52909afd85c9SLinus Lüssing * 52919afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 52929afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 52939afd85c9SLinus Lüssing * 52949afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 52959afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 52969afd85c9SLinus Lüssing * 5297a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5298a516993fSLinus Lüssing * differs from the provided skb. 52999afd85c9SLinus Lüssing */ 53009afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 53019afd85c9SLinus Lüssing unsigned int transport_len, 53029afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 53039afd85c9SLinus Lüssing { 53049afd85c9SLinus Lüssing struct sk_buff *skb_chk; 53059afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5306fcba67c9SLinus Lüssing __sum16 ret; 53079afd85c9SLinus Lüssing 53089afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 53099afd85c9SLinus Lüssing if (!skb_chk) 5310a516993fSLinus Lüssing goto err; 53119afd85c9SLinus Lüssing 5312a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5313a516993fSLinus Lüssing goto err; 53149afd85c9SLinus Lüssing 53159b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 53169afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 53179b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 53189afd85c9SLinus Lüssing 5319a516993fSLinus Lüssing if (ret) 5320a516993fSLinus Lüssing goto err; 53219afd85c9SLinus Lüssing 53229afd85c9SLinus Lüssing return skb_chk; 5323a516993fSLinus Lüssing 5324a516993fSLinus Lüssing err: 5325a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5326a516993fSLinus Lüssing kfree_skb(skb_chk); 5327a516993fSLinus Lüssing 5328a516993fSLinus Lüssing return NULL; 5329a516993fSLinus Lüssing 53309afd85c9SLinus Lüssing } 53319afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 53329afd85c9SLinus Lüssing 53334497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 53344497b076SBen Hutchings { 5335e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5336e87cc472SJoe Perches skb->dev->name); 53374497b076SBen Hutchings } 53384497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5339bad43ca8SEric Dumazet 5340bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5341bad43ca8SEric Dumazet { 53423d861f66SEric Dumazet if (head_stolen) { 53433d861f66SEric Dumazet skb_release_head_state(skb); 5344bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 53453d861f66SEric Dumazet } else { 5346bad43ca8SEric Dumazet __kfree_skb(skb); 5347bad43ca8SEric Dumazet } 53483d861f66SEric Dumazet } 5349bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5350bad43ca8SEric Dumazet 5351bad43ca8SEric Dumazet /** 5352bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5353bad43ca8SEric Dumazet * @to: prior buffer 5354bad43ca8SEric Dumazet * @from: buffer to add 5355bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5356c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5357bad43ca8SEric Dumazet */ 5358bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5359bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5360bad43ca8SEric Dumazet { 5361c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5362bad43ca8SEric Dumazet int i, delta, len = from->len; 5363bad43ca8SEric Dumazet 5364bad43ca8SEric Dumazet *fragstolen = false; 5365bad43ca8SEric Dumazet 5366bad43ca8SEric Dumazet if (skb_cloned(to)) 5367bad43ca8SEric Dumazet return false; 5368bad43ca8SEric Dumazet 53696a5bcd84SIlias Apalodimas /* The page pool signature of struct page will eventually figure out 53706a5bcd84SIlias Apalodimas * which pages can be recycled or not but for now let's prohibit slab 53716a5bcd84SIlias Apalodimas * allocated and page_pool allocated SKBs from being coalesced. 53726a5bcd84SIlias Apalodimas */ 53736a5bcd84SIlias Apalodimas if (to->pp_recycle != from->pp_recycle) 53746a5bcd84SIlias Apalodimas return false; 53756a5bcd84SIlias Apalodimas 5376bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5377e93a0435SEric Dumazet if (len) 5378bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5379bad43ca8SEric Dumazet *delta_truesize = 0; 5380bad43ca8SEric Dumazet return true; 5381bad43ca8SEric Dumazet } 5382bad43ca8SEric Dumazet 5383c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5384c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5385c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5386bad43ca8SEric Dumazet return false; 53871f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 53881f8b977aSWillem de Bruijn return false; 5389bad43ca8SEric Dumazet 5390bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5391bad43ca8SEric Dumazet struct page *page; 5392bad43ca8SEric Dumazet unsigned int offset; 5393bad43ca8SEric Dumazet 5394c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5395c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5396bad43ca8SEric Dumazet return false; 5397bad43ca8SEric Dumazet 5398bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5399bad43ca8SEric Dumazet return false; 5400bad43ca8SEric Dumazet 5401bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5402bad43ca8SEric Dumazet 5403bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5404bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5405bad43ca8SEric Dumazet 5406c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5407bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5408bad43ca8SEric Dumazet *fragstolen = true; 5409bad43ca8SEric Dumazet } else { 5410c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5411c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5412bad43ca8SEric Dumazet return false; 5413bad43ca8SEric Dumazet 5414f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5415bad43ca8SEric Dumazet } 5416bad43ca8SEric Dumazet 5417bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5418bad43ca8SEric Dumazet 5419c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5420c818fa9eSEric Dumazet from_shinfo->frags, 5421c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5422c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5423bad43ca8SEric Dumazet 5424bad43ca8SEric Dumazet if (!skb_cloned(from)) 5425c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5426bad43ca8SEric Dumazet 54278ea853fdSLi RongQing /* if the skb is not cloned this does nothing 54288ea853fdSLi RongQing * since we set nr_frags to 0. 54298ea853fdSLi RongQing */ 5430c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5431c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5432bad43ca8SEric Dumazet 5433bad43ca8SEric Dumazet to->truesize += delta; 5434bad43ca8SEric Dumazet to->len += len; 5435bad43ca8SEric Dumazet to->data_len += len; 5436bad43ca8SEric Dumazet 5437bad43ca8SEric Dumazet *delta_truesize = delta; 5438bad43ca8SEric Dumazet return true; 5439bad43ca8SEric Dumazet } 5440bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5441621e84d6SNicolas Dichtel 5442621e84d6SNicolas Dichtel /** 54438b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5444621e84d6SNicolas Dichtel * 5445621e84d6SNicolas Dichtel * @skb: buffer to clean 54468b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5447621e84d6SNicolas Dichtel * 54488b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 54498b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 54508b27f277SNicolas Dichtel * operations. 54518b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 54528b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 54538b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5454621e84d6SNicolas Dichtel */ 54558b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5456621e84d6SNicolas Dichtel { 5457621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5458621e84d6SNicolas Dichtel skb->skb_iif = 0; 545960ff7467SWANG Cong skb->ignore_df = 0; 5460621e84d6SNicolas Dichtel skb_dst_drop(skb); 5461174e2381SFlorian Westphal skb_ext_reset(skb); 5462895b5c9fSFlorian Westphal nf_reset_ct(skb); 5463621e84d6SNicolas Dichtel nf_reset_trace(skb); 5464213dd74aSHerbert Xu 54656f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 54666f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5467875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 54686f9a5069SPetr Machata #endif 54696f9a5069SPetr Machata 5470213dd74aSHerbert Xu if (!xnet) 5471213dd74aSHerbert Xu return; 5472213dd74aSHerbert Xu 54732b5ec1a5SYe Yin ipvs_reset(skb); 5474213dd74aSHerbert Xu skb->mark = 0; 5475c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 5476621e84d6SNicolas Dichtel } 5477621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5478de960aa9SFlorian Westphal 5479de960aa9SFlorian Westphal /** 5480de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5481de960aa9SFlorian Westphal * 5482de960aa9SFlorian Westphal * @skb: GSO skb 5483de960aa9SFlorian Westphal * 5484de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5485de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5486de960aa9SFlorian Westphal * 5487de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5488de960aa9SFlorian Westphal */ 5489a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5490de960aa9SFlorian Westphal { 5491de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5492f993bc25SFlorian Westphal unsigned int thlen = 0; 5493f993bc25SFlorian Westphal 5494f993bc25SFlorian Westphal if (skb->encapsulation) { 5495f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5496f993bc25SFlorian Westphal skb_transport_header(skb); 5497de960aa9SFlorian Westphal 5498de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5499f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5500f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5501f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 55021dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 550390017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5504ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5505ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5506f993bc25SFlorian Westphal } 55076d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 55086d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 55096d39d589SFlorian Westphal * accounted for. 55106d39d589SFlorian Westphal */ 5511f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5512de960aa9SFlorian Westphal } 5513a4a77718SDaniel Axtens 5514a4a77718SDaniel Axtens /** 5515a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5516a4a77718SDaniel Axtens * 5517a4a77718SDaniel Axtens * @skb: GSO skb 5518a4a77718SDaniel Axtens * 5519a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5520a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5521a4a77718SDaniel Axtens * 5522a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5523a4a77718SDaniel Axtens */ 5524a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5525a4a77718SDaniel Axtens { 5526a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5527a4a77718SDaniel Axtens skb_network_header(skb); 5528a4a77718SDaniel Axtens 5529a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5530a4a77718SDaniel Axtens } 5531a4a77718SDaniel Axtens 5532a4a77718SDaniel Axtens /** 5533a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5534a4a77718SDaniel Axtens * 5535a4a77718SDaniel Axtens * @skb: GSO skb 5536a4a77718SDaniel Axtens * 5537a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5538a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5539a4a77718SDaniel Axtens * headers (TCP/UDP). 5540a4a77718SDaniel Axtens */ 5541a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5542a4a77718SDaniel Axtens { 5543a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5544a4a77718SDaniel Axtens 5545a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5546a4a77718SDaniel Axtens } 55470d5501c1SVlad Yasevich 5548ae7ef81eSMarcelo Ricardo Leitner /** 55492b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 55502b16f048SDaniel Axtens * 55512b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 55522b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 55532b16f048SDaniel Axtens * 55542b16f048SDaniel Axtens * We might want to check: 55552b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 55562b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 55572b16f048SDaniel Axtens * 55582b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 55592b16f048SDaniel Axtens * 556049682bfaSMathieu Malaterre * @skb: GSO skb 556149682bfaSMathieu Malaterre * 55622b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 55632b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 55642b16f048SDaniel Axtens * 55652b16f048SDaniel Axtens * @max_len: The maximum permissible length. 55662b16f048SDaniel Axtens * 55672b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 55682b16f048SDaniel Axtens */ 55692b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 55702b16f048SDaniel Axtens unsigned int seg_len, 55712b16f048SDaniel Axtens unsigned int max_len) { 55722b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 55732b16f048SDaniel Axtens const struct sk_buff *iter; 55742b16f048SDaniel Axtens 55752b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 55762b16f048SDaniel Axtens return seg_len <= max_len; 55772b16f048SDaniel Axtens 55782b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 55792b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 55802b16f048SDaniel Axtens 55812b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 55822b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 55832b16f048SDaniel Axtens return false; 55842b16f048SDaniel Axtens } 55852b16f048SDaniel Axtens 55862b16f048SDaniel Axtens return true; 55872b16f048SDaniel Axtens } 55882b16f048SDaniel Axtens 55892b16f048SDaniel Axtens /** 5590779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5591ae7ef81eSMarcelo Ricardo Leitner * 5592ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 559376f21b99SDavid S. Miller * @mtu: MTU to validate against 5594ae7ef81eSMarcelo Ricardo Leitner * 5595779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5596779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5597779b7931SDaniel Axtens * payload. 5598ae7ef81eSMarcelo Ricardo Leitner */ 5599779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5600ae7ef81eSMarcelo Ricardo Leitner { 56012b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5602ae7ef81eSMarcelo Ricardo Leitner } 5603779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5604ae7ef81eSMarcelo Ricardo Leitner 56052b16f048SDaniel Axtens /** 56062b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 56072b16f048SDaniel Axtens * 56082b16f048SDaniel Axtens * @skb: GSO skb 56092b16f048SDaniel Axtens * @len: length to validate against 56102b16f048SDaniel Axtens * 56112b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 56122b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 56132b16f048SDaniel Axtens */ 56142b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 56152b16f048SDaniel Axtens { 56162b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 56172b16f048SDaniel Axtens } 56182b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 56192b16f048SDaniel Axtens 56200d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 56210d5501c1SVlad Yasevich { 5622d85e8be2SYuya Kusakabe int mac_len, meta_len; 5623d85e8be2SYuya Kusakabe void *meta; 56244bbb3e0eSToshiaki Makita 56250d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 56260d5501c1SVlad Yasevich kfree_skb(skb); 56270d5501c1SVlad Yasevich return NULL; 56280d5501c1SVlad Yasevich } 56290d5501c1SVlad Yasevich 56304bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5631ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 56324bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 56334bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5634ae474573SToshiaki Makita } 5635d85e8be2SYuya Kusakabe 5636d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5637d85e8be2SYuya Kusakabe if (meta_len) { 5638d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5639d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5640d85e8be2SYuya Kusakabe } 5641d85e8be2SYuya Kusakabe 56420d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 56430d5501c1SVlad Yasevich return skb; 56440d5501c1SVlad Yasevich } 56450d5501c1SVlad Yasevich 56460d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 56470d5501c1SVlad Yasevich { 56480d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 56490d5501c1SVlad Yasevich u16 vlan_tci; 56500d5501c1SVlad Yasevich 5651df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 56520d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 56530d5501c1SVlad Yasevich return skb; 56540d5501c1SVlad Yasevich } 56550d5501c1SVlad Yasevich 56560d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 56570d5501c1SVlad Yasevich if (unlikely(!skb)) 56580d5501c1SVlad Yasevich goto err_free; 565955eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 566055eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 56610d5501c1SVlad Yasevich goto err_free; 56620d5501c1SVlad Yasevich 56630d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 56640d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 56650d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 56660d5501c1SVlad Yasevich 56670d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 56680d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 56690d5501c1SVlad Yasevich 56700d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 56710d5501c1SVlad Yasevich if (unlikely(!skb)) 56720d5501c1SVlad Yasevich goto err_free; 56730d5501c1SVlad Yasevich 56740d5501c1SVlad Yasevich skb_reset_network_header(skb); 56758be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 56760d5501c1SVlad Yasevich skb_reset_transport_header(skb); 56770d5501c1SVlad Yasevich skb_reset_mac_len(skb); 56780d5501c1SVlad Yasevich 56790d5501c1SVlad Yasevich return skb; 56800d5501c1SVlad Yasevich 56810d5501c1SVlad Yasevich err_free: 56820d5501c1SVlad Yasevich kfree_skb(skb); 56830d5501c1SVlad Yasevich return NULL; 56840d5501c1SVlad Yasevich } 56850d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 56862e4e4410SEric Dumazet 5687e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5688e2195121SJiri Pirko { 5689e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5690e2195121SJiri Pirko return -ENOMEM; 5691e2195121SJiri Pirko 5692e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5693e2195121SJiri Pirko return 0; 5694e2195121SJiri Pirko 5695e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5696e2195121SJiri Pirko } 5697e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5698e2195121SJiri Pirko 5699bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5700bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5701bfca4c52SShmulik Ladkani */ 5702bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 570393515d53SJiri Pirko { 570493515d53SJiri Pirko struct vlan_hdr *vhdr; 5705b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 570693515d53SJiri Pirko int err; 570793515d53SJiri Pirko 5708b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5709b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5710b6a79208SShmulik Ladkani offset)) { 5711b6a79208SShmulik Ladkani return -EINVAL; 5712b6a79208SShmulik Ladkani } 5713b6a79208SShmulik Ladkani 571493515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 571593515d53SJiri Pirko if (unlikely(err)) 5716b6a79208SShmulik Ladkani return err; 571793515d53SJiri Pirko 571893515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 571993515d53SJiri Pirko 572093515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 572193515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 572293515d53SJiri Pirko 572393515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 572493515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 572593515d53SJiri Pirko 572693515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 572793515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 572893515d53SJiri Pirko 572993515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 573093515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 573193515d53SJiri Pirko 573293515d53SJiri Pirko skb_reset_mac_len(skb); 573393515d53SJiri Pirko 573493515d53SJiri Pirko return err; 573593515d53SJiri Pirko } 5736bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 573793515d53SJiri Pirko 5738b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5739b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5740b6a79208SShmulik Ladkani */ 574193515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 574293515d53SJiri Pirko { 574393515d53SJiri Pirko u16 vlan_tci; 574493515d53SJiri Pirko __be16 vlan_proto; 574593515d53SJiri Pirko int err; 574693515d53SJiri Pirko 5747df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5748b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 574993515d53SJiri Pirko } else { 5750ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 575193515d53SJiri Pirko return 0; 575293515d53SJiri Pirko 575393515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 575493515d53SJiri Pirko if (err) 575593515d53SJiri Pirko return err; 575693515d53SJiri Pirko } 575793515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5758ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 575993515d53SJiri Pirko return 0; 576093515d53SJiri Pirko 576193515d53SJiri Pirko vlan_proto = skb->protocol; 576293515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 576393515d53SJiri Pirko if (unlikely(err)) 576493515d53SJiri Pirko return err; 576593515d53SJiri Pirko 576693515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 576793515d53SJiri Pirko return 0; 576893515d53SJiri Pirko } 576993515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 577093515d53SJiri Pirko 5771b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5772b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5773b6a79208SShmulik Ladkani */ 577493515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 577593515d53SJiri Pirko { 5776df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5777b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 577893515d53SJiri Pirko int err; 577993515d53SJiri Pirko 5780b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5781b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5782b6a79208SShmulik Ladkani offset)) { 5783b6a79208SShmulik Ladkani return -EINVAL; 5784b6a79208SShmulik Ladkani } 5785b6a79208SShmulik Ladkani 578693515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5787df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5788b6a79208SShmulik Ladkani if (err) 578993515d53SJiri Pirko return err; 57909241e2dfSDaniel Borkmann 579193515d53SJiri Pirko skb->protocol = skb->vlan_proto; 579293515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 579393515d53SJiri Pirko 57946b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 579593515d53SJiri Pirko } 579693515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 579793515d53SJiri Pirko return 0; 579893515d53SJiri Pirko } 579993515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 580093515d53SJiri Pirko 580119fbcb36SGuillaume Nault /** 580219fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 580319fbcb36SGuillaume Nault * 580419fbcb36SGuillaume Nault * @skb: Socket buffer to modify 580519fbcb36SGuillaume Nault * 580619fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 580719fbcb36SGuillaume Nault * 580819fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 580919fbcb36SGuillaume Nault * present. 581019fbcb36SGuillaume Nault * 581119fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 581219fbcb36SGuillaume Nault */ 581319fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 581419fbcb36SGuillaume Nault { 581519fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 581619fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 581719fbcb36SGuillaume Nault return -EPROTO; 581819fbcb36SGuillaume Nault 581919fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 582019fbcb36SGuillaume Nault skb_reset_mac_header(skb); 582119fbcb36SGuillaume Nault skb_reset_mac_len(skb); 582219fbcb36SGuillaume Nault 582319fbcb36SGuillaume Nault return 0; 582419fbcb36SGuillaume Nault } 582519fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 582619fbcb36SGuillaume Nault 582719fbcb36SGuillaume Nault /** 582819fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 582919fbcb36SGuillaume Nault * 583019fbcb36SGuillaume Nault * @skb: Socket buffer to modify 583119fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 583219fbcb36SGuillaume Nault * @src: Source MAC address of the new header 583319fbcb36SGuillaume Nault * 583419fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 583519fbcb36SGuillaume Nault * 583619fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 583719fbcb36SGuillaume Nault * 583819fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 583919fbcb36SGuillaume Nault */ 584019fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 584119fbcb36SGuillaume Nault const unsigned char *src) 584219fbcb36SGuillaume Nault { 584319fbcb36SGuillaume Nault struct ethhdr *eth; 584419fbcb36SGuillaume Nault int err; 584519fbcb36SGuillaume Nault 584619fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 584719fbcb36SGuillaume Nault return -EPROTO; 584819fbcb36SGuillaume Nault 584919fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 585019fbcb36SGuillaume Nault if (err < 0) 585119fbcb36SGuillaume Nault return err; 585219fbcb36SGuillaume Nault 585319fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 585419fbcb36SGuillaume Nault skb_reset_mac_header(skb); 585519fbcb36SGuillaume Nault skb_reset_mac_len(skb); 585619fbcb36SGuillaume Nault 585719fbcb36SGuillaume Nault eth = eth_hdr(skb); 585819fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 585919fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 586019fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 586119fbcb36SGuillaume Nault 586219fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 586319fbcb36SGuillaume Nault 586419fbcb36SGuillaume Nault return 0; 586519fbcb36SGuillaume Nault } 586619fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 586719fbcb36SGuillaume Nault 58688822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 58698822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 58708822e270SJohn Hurley __be16 ethertype) 58718822e270SJohn Hurley { 58728822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 58738822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 58748822e270SJohn Hurley 58758822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 58768822e270SJohn Hurley } 58778822e270SJohn Hurley 58788822e270SJohn Hurley hdr->h_proto = ethertype; 58798822e270SJohn Hurley } 58808822e270SJohn Hurley 58818822e270SJohn Hurley /** 5882e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5883e7dbfed1SMartin Varghese * the packet 58848822e270SJohn Hurley * 58858822e270SJohn Hurley * @skb: buffer 58868822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 58878822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5888fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5889e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5890e7dbfed1SMartin Varghese * ethernet 58918822e270SJohn Hurley * 58928822e270SJohn Hurley * Expects skb->data at mac header. 58938822e270SJohn Hurley * 58948822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 58958822e270SJohn Hurley */ 5896fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5897d04ac224SMartin Varghese int mac_len, bool ethernet) 58988822e270SJohn Hurley { 58998822e270SJohn Hurley struct mpls_shim_hdr *lse; 59008822e270SJohn Hurley int err; 59018822e270SJohn Hurley 59028822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 59038822e270SJohn Hurley return -EINVAL; 59048822e270SJohn Hurley 59058822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 59068822e270SJohn Hurley if (skb->encapsulation) 59078822e270SJohn Hurley return -EINVAL; 59088822e270SJohn Hurley 59098822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 59108822e270SJohn Hurley if (unlikely(err)) 59118822e270SJohn Hurley return err; 59128822e270SJohn Hurley 59138822e270SJohn Hurley if (!skb->inner_protocol) { 5914e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 59158822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 59168822e270SJohn Hurley } 59178822e270SJohn Hurley 59188822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 59198822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5920fa4e0f88SDavide Caratti mac_len); 59218822e270SJohn Hurley skb_reset_mac_header(skb); 5922fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5923e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 59248822e270SJohn Hurley 59258822e270SJohn Hurley lse = mpls_hdr(skb); 59268822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 59278822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 59288822e270SJohn Hurley 59294296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 59308822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 59318822e270SJohn Hurley skb->protocol = mpls_proto; 59328822e270SJohn Hurley 59338822e270SJohn Hurley return 0; 59348822e270SJohn Hurley } 59358822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 59368822e270SJohn Hurley 59372e4e4410SEric Dumazet /** 5938ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5939ed246ceeSJohn Hurley * 5940ed246ceeSJohn Hurley * @skb: buffer 5941ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5942fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 594376f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5944ed246ceeSJohn Hurley * 5945ed246ceeSJohn Hurley * Expects skb->data at mac header. 5946ed246ceeSJohn Hurley * 5947ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5948ed246ceeSJohn Hurley */ 5949040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5950040b5cfbSMartin Varghese bool ethernet) 5951ed246ceeSJohn Hurley { 5952ed246ceeSJohn Hurley int err; 5953ed246ceeSJohn Hurley 5954ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5955dedc5a08SDavide Caratti return 0; 5956ed246ceeSJohn Hurley 5957fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5958ed246ceeSJohn Hurley if (unlikely(err)) 5959ed246ceeSJohn Hurley return err; 5960ed246ceeSJohn Hurley 5961ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5962ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5963fa4e0f88SDavide Caratti mac_len); 5964ed246ceeSJohn Hurley 5965ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5966ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5967fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5968ed246ceeSJohn Hurley 59694296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5970ed246ceeSJohn Hurley struct ethhdr *hdr; 5971ed246ceeSJohn Hurley 5972ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5973ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5974ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5975ed246ceeSJohn Hurley } 5976ed246ceeSJohn Hurley skb->protocol = next_proto; 5977ed246ceeSJohn Hurley 5978ed246ceeSJohn Hurley return 0; 5979ed246ceeSJohn Hurley } 5980ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5981ed246ceeSJohn Hurley 5982ed246ceeSJohn Hurley /** 5983d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5984d27cf5c5SJohn Hurley * 5985d27cf5c5SJohn Hurley * @skb: buffer 5986d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5987d27cf5c5SJohn Hurley * 5988d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5989d27cf5c5SJohn Hurley * 5990d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5991d27cf5c5SJohn Hurley */ 5992d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5993d27cf5c5SJohn Hurley { 5994d27cf5c5SJohn Hurley int err; 5995d27cf5c5SJohn Hurley 5996d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5997d27cf5c5SJohn Hurley return -EINVAL; 5998d27cf5c5SJohn Hurley 5999d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 6000d27cf5c5SJohn Hurley if (unlikely(err)) 6001d27cf5c5SJohn Hurley return err; 6002d27cf5c5SJohn Hurley 6003d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 6004d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 6005d27cf5c5SJohn Hurley 6006d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 6007d27cf5c5SJohn Hurley } 6008d27cf5c5SJohn Hurley 6009d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 6010d27cf5c5SJohn Hurley 6011d27cf5c5SJohn Hurley return 0; 6012d27cf5c5SJohn Hurley } 6013d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 6014d27cf5c5SJohn Hurley 6015d27cf5c5SJohn Hurley /** 60162a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 60172a2ea508SJohn Hurley * 60182a2ea508SJohn Hurley * @skb: buffer 60192a2ea508SJohn Hurley * 60202a2ea508SJohn Hurley * Expects skb->data at mac header. 60212a2ea508SJohn Hurley * 60222a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 60232a2ea508SJohn Hurley */ 60242a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 60252a2ea508SJohn Hurley { 60262a2ea508SJohn Hurley u32 lse; 60272a2ea508SJohn Hurley u8 ttl; 60282a2ea508SJohn Hurley 60292a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 60302a2ea508SJohn Hurley return -EINVAL; 60312a2ea508SJohn Hurley 603213de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 603313de4ed9SDavide Caratti return -ENOMEM; 603413de4ed9SDavide Caratti 60352a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 60362a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 60372a2ea508SJohn Hurley if (!--ttl) 60382a2ea508SJohn Hurley return -EINVAL; 60392a2ea508SJohn Hurley 60402a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 60412a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 60422a2ea508SJohn Hurley 60432a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 60442a2ea508SJohn Hurley } 60452a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 60462a2ea508SJohn Hurley 60472a2ea508SJohn Hurley /** 60482e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 60492e4e4410SEric Dumazet * 6050de3f0d0eSMasanari Iida * @header_len: size of linear part 6051de3f0d0eSMasanari Iida * @data_len: needed length in frags 6052de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 6053de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 6054de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 60552e4e4410SEric Dumazet * 60562e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 60572e4e4410SEric Dumazet */ 60582e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 60592e4e4410SEric Dumazet unsigned long data_len, 60602e4e4410SEric Dumazet int max_page_order, 60612e4e4410SEric Dumazet int *errcode, 60622e4e4410SEric Dumazet gfp_t gfp_mask) 60632e4e4410SEric Dumazet { 60642e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 60652e4e4410SEric Dumazet unsigned long chunk; 60662e4e4410SEric Dumazet struct sk_buff *skb; 60672e4e4410SEric Dumazet struct page *page; 60682e4e4410SEric Dumazet int i; 60692e4e4410SEric Dumazet 60702e4e4410SEric Dumazet *errcode = -EMSGSIZE; 60712e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 60722e4e4410SEric Dumazet * high order pages... 60732e4e4410SEric Dumazet */ 60742e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 60752e4e4410SEric Dumazet return NULL; 60762e4e4410SEric Dumazet 60772e4e4410SEric Dumazet *errcode = -ENOBUFS; 6078f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 60792e4e4410SEric Dumazet if (!skb) 60802e4e4410SEric Dumazet return NULL; 60812e4e4410SEric Dumazet 60822e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 60832e4e4410SEric Dumazet 60842e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 60852e4e4410SEric Dumazet int order = max_page_order; 60862e4e4410SEric Dumazet 60872e4e4410SEric Dumazet while (order) { 60882e4e4410SEric Dumazet if (npages >= 1 << order) { 6089d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 60902e4e4410SEric Dumazet __GFP_COMP | 6091d14b56f5SMichal Hocko __GFP_NOWARN, 60922e4e4410SEric Dumazet order); 60932e4e4410SEric Dumazet if (page) 60942e4e4410SEric Dumazet goto fill_page; 60952e4e4410SEric Dumazet /* Do not retry other high order allocations */ 60962e4e4410SEric Dumazet order = 1; 60972e4e4410SEric Dumazet max_page_order = 0; 60982e4e4410SEric Dumazet } 60992e4e4410SEric Dumazet order--; 61002e4e4410SEric Dumazet } 61012e4e4410SEric Dumazet page = alloc_page(gfp_mask); 61022e4e4410SEric Dumazet if (!page) 61032e4e4410SEric Dumazet goto failure; 61042e4e4410SEric Dumazet fill_page: 61052e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 61062e4e4410SEric Dumazet PAGE_SIZE << order); 61072e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 61082e4e4410SEric Dumazet data_len -= chunk; 61092e4e4410SEric Dumazet npages -= 1 << order; 61102e4e4410SEric Dumazet } 61112e4e4410SEric Dumazet return skb; 61122e4e4410SEric Dumazet 61132e4e4410SEric Dumazet failure: 61142e4e4410SEric Dumazet kfree_skb(skb); 61152e4e4410SEric Dumazet return NULL; 61162e4e4410SEric Dumazet } 61172e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 61186fa01ccdSSowmini Varadhan 61196fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 61206fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 61216fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 61226fa01ccdSSowmini Varadhan { 61236fa01ccdSSowmini Varadhan int i; 61246fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 61256fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 61266fa01ccdSSowmini Varadhan u8 *data; 61276fa01ccdSSowmini Varadhan 61286fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 61296fa01ccdSSowmini Varadhan 61306fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 61316fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 61326fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 61336fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 61346fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 61356fa01ccdSSowmini Varadhan if (!data) 61366fa01ccdSSowmini Varadhan return -ENOMEM; 61376fa01ccdSSowmini Varadhan 61386fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 61396fa01ccdSSowmini Varadhan 61406fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 61416fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 61426fa01ccdSSowmini Varadhan skb->len -= off; 61436fa01ccdSSowmini Varadhan 61446fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 61456fa01ccdSSowmini Varadhan skb_shinfo(skb), 61466fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 61476fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 61486fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 61496fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 61506fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 61516fa01ccdSSowmini Varadhan kfree(data); 61526fa01ccdSSowmini Varadhan return -ENOMEM; 61536fa01ccdSSowmini Varadhan } 61546fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 61556fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 61566fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 61576fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 61586fa01ccdSSowmini Varadhan skb_release_data(skb); 61596fa01ccdSSowmini Varadhan } else { 61606fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 61616fa01ccdSSowmini Varadhan * relocate values 61626fa01ccdSSowmini Varadhan */ 61636fa01ccdSSowmini Varadhan skb_free_head(skb); 61646fa01ccdSSowmini Varadhan } 61656fa01ccdSSowmini Varadhan 61666fa01ccdSSowmini Varadhan skb->head = data; 61676fa01ccdSSowmini Varadhan skb->data = data; 61686fa01ccdSSowmini Varadhan skb->head_frag = 0; 61696fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 61706fa01ccdSSowmini Varadhan skb->end = size; 61716fa01ccdSSowmini Varadhan #else 61726fa01ccdSSowmini Varadhan skb->end = skb->head + size; 61736fa01ccdSSowmini Varadhan #endif 61746fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 61756fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 61766fa01ccdSSowmini Varadhan skb->cloned = 0; 61776fa01ccdSSowmini Varadhan skb->hdr_len = 0; 61786fa01ccdSSowmini Varadhan skb->nohdr = 0; 61796fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 61806fa01ccdSSowmini Varadhan 61816fa01ccdSSowmini Varadhan return 0; 61826fa01ccdSSowmini Varadhan } 61836fa01ccdSSowmini Varadhan 61846fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 61856fa01ccdSSowmini Varadhan 61866fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 61876fa01ccdSSowmini Varadhan * pskb_carve() 61886fa01ccdSSowmini Varadhan */ 61896fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 61906fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 61916fa01ccdSSowmini Varadhan gfp_t gfp_mask) 61926fa01ccdSSowmini Varadhan { 61936fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 61946fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 61956fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 61966fa01ccdSSowmini Varadhan 61976fa01ccdSSowmini Varadhan do { 61986fa01ccdSSowmini Varadhan if (!list) { 61996fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 62006fa01ccdSSowmini Varadhan return -EFAULT; 62016fa01ccdSSowmini Varadhan } 62026fa01ccdSSowmini Varadhan if (list->len <= eat) { 62036fa01ccdSSowmini Varadhan /* Eaten as whole. */ 62046fa01ccdSSowmini Varadhan eat -= list->len; 62056fa01ccdSSowmini Varadhan list = list->next; 62066fa01ccdSSowmini Varadhan insp = list; 62076fa01ccdSSowmini Varadhan } else { 62086fa01ccdSSowmini Varadhan /* Eaten partially. */ 62096fa01ccdSSowmini Varadhan if (skb_shared(list)) { 62106fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 62116fa01ccdSSowmini Varadhan if (!clone) 62126fa01ccdSSowmini Varadhan return -ENOMEM; 62136fa01ccdSSowmini Varadhan insp = list->next; 62146fa01ccdSSowmini Varadhan list = clone; 62156fa01ccdSSowmini Varadhan } else { 62166fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 62176fa01ccdSSowmini Varadhan insp = list; 62186fa01ccdSSowmini Varadhan } 62196fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 62206fa01ccdSSowmini Varadhan kfree_skb(clone); 62216fa01ccdSSowmini Varadhan return -ENOMEM; 62226fa01ccdSSowmini Varadhan } 62236fa01ccdSSowmini Varadhan break; 62246fa01ccdSSowmini Varadhan } 62256fa01ccdSSowmini Varadhan } while (eat); 62266fa01ccdSSowmini Varadhan 62276fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 62286fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 62296fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 62306fa01ccdSSowmini Varadhan kfree_skb(list); 62316fa01ccdSSowmini Varadhan } 62326fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 62336fa01ccdSSowmini Varadhan if (clone) { 62346fa01ccdSSowmini Varadhan clone->next = list; 62356fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 62366fa01ccdSSowmini Varadhan } 62376fa01ccdSSowmini Varadhan return 0; 62386fa01ccdSSowmini Varadhan } 62396fa01ccdSSowmini Varadhan 62406fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 62416fa01ccdSSowmini Varadhan * non-linear part of skb 62426fa01ccdSSowmini Varadhan */ 62436fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 62446fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 62456fa01ccdSSowmini Varadhan { 62466fa01ccdSSowmini Varadhan int i, k = 0; 62476fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 62486fa01ccdSSowmini Varadhan u8 *data; 62496fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 62506fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 62516fa01ccdSSowmini Varadhan 62526fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 62536fa01ccdSSowmini Varadhan 62546fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 62556fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 62566fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 62576fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 62586fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 62596fa01ccdSSowmini Varadhan if (!data) 62606fa01ccdSSowmini Varadhan return -ENOMEM; 62616fa01ccdSSowmini Varadhan 62626fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 62636fa01ccdSSowmini Varadhan 62646fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6265e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 62666fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 62676fa01ccdSSowmini Varadhan kfree(data); 62686fa01ccdSSowmini Varadhan return -ENOMEM; 62696fa01ccdSSowmini Varadhan } 62706fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 62716fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 62726fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 62736fa01ccdSSowmini Varadhan 62746fa01ccdSSowmini Varadhan if (pos + fsize > off) { 62756fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 62766fa01ccdSSowmini Varadhan 62776fa01ccdSSowmini Varadhan if (pos < off) { 62786fa01ccdSSowmini Varadhan /* Split frag. 62796fa01ccdSSowmini Varadhan * We have two variants in this case: 62806fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 62816fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 62826fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 62836fa01ccdSSowmini Varadhan * where splitting is expensive. 62846fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 62856fa01ccdSSowmini Varadhan */ 6286b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 62876fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 62886fa01ccdSSowmini Varadhan } 62896fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 62906fa01ccdSSowmini Varadhan k++; 62916fa01ccdSSowmini Varadhan } 62926fa01ccdSSowmini Varadhan pos += fsize; 62936fa01ccdSSowmini Varadhan } 62946fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 62956fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 62966fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 62976fa01ccdSSowmini Varadhan 62986fa01ccdSSowmini Varadhan /* split line is in frag list */ 6299eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6300eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6301eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6302eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6303eabe8618SMiaohe Lin kfree(data); 6304eabe8618SMiaohe Lin return -ENOMEM; 63056fa01ccdSSowmini Varadhan } 63066fa01ccdSSowmini Varadhan skb_release_data(skb); 63076fa01ccdSSowmini Varadhan 63086fa01ccdSSowmini Varadhan skb->head = data; 63096fa01ccdSSowmini Varadhan skb->head_frag = 0; 63106fa01ccdSSowmini Varadhan skb->data = data; 63116fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 63126fa01ccdSSowmini Varadhan skb->end = size; 63136fa01ccdSSowmini Varadhan #else 63146fa01ccdSSowmini Varadhan skb->end = skb->head + size; 63156fa01ccdSSowmini Varadhan #endif 63166fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 63176fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 63186fa01ccdSSowmini Varadhan skb->cloned = 0; 63196fa01ccdSSowmini Varadhan skb->hdr_len = 0; 63206fa01ccdSSowmini Varadhan skb->nohdr = 0; 63216fa01ccdSSowmini Varadhan skb->len -= off; 63226fa01ccdSSowmini Varadhan skb->data_len = skb->len; 63236fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 63246fa01ccdSSowmini Varadhan return 0; 63256fa01ccdSSowmini Varadhan } 63266fa01ccdSSowmini Varadhan 63276fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 63286fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 63296fa01ccdSSowmini Varadhan { 63306fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 63316fa01ccdSSowmini Varadhan 63326fa01ccdSSowmini Varadhan if (len < headlen) 63336fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 63346fa01ccdSSowmini Varadhan else 63356fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 63366fa01ccdSSowmini Varadhan } 63376fa01ccdSSowmini Varadhan 63386fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 63396fa01ccdSSowmini Varadhan * a new skb 63406fa01ccdSSowmini Varadhan */ 63416fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 63426fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 63436fa01ccdSSowmini Varadhan { 63446fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 63456fa01ccdSSowmini Varadhan 63466fa01ccdSSowmini Varadhan if (!clone) 63476fa01ccdSSowmini Varadhan return NULL; 63486fa01ccdSSowmini Varadhan 63496fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 63506fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 63516fa01ccdSSowmini Varadhan kfree_skb(clone); 63526fa01ccdSSowmini Varadhan return NULL; 63536fa01ccdSSowmini Varadhan } 63546fa01ccdSSowmini Varadhan return clone; 63556fa01ccdSSowmini Varadhan } 63566fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6357c8c8b127SEric Dumazet 6358c8c8b127SEric Dumazet /** 6359c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6360c8c8b127SEric Dumazet * @skb: buffer 6361c8c8b127SEric Dumazet * 6362c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6363c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6364c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6365c8c8b127SEric Dumazet * truesize. 6366c8c8b127SEric Dumazet * Notes: 6367c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6368c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6369c8c8b127SEric Dumazet */ 6370c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6371c8c8b127SEric Dumazet { 63723174fed9SEric Dumazet if (skb->data_len) { 63733174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6374c8c8b127SEric Dumazet skb_cloned(skb)) 6375c8c8b127SEric Dumazet return; 6376c8c8b127SEric Dumazet 6377c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6378c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 63793174fed9SEric Dumazet } 63803174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 63813174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 63823174fed9SEric Dumazet * their truesize. 63833174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 63843174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 63853174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6386c8c8b127SEric Dumazet */ 6387c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6388c8c8b127SEric Dumazet } 6389df5042f4SFlorian Westphal 6390df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6391df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6392df5042f4SFlorian Westphal { 6393df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6394df5042f4SFlorian Westphal } 6395df5042f4SFlorian Westphal 63968b69a803SPaolo Abeni /** 63978b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 63988b69a803SPaolo Abeni * 63994930f483SFlorian Westphal * @flags: See kmalloc(). 64004930f483SFlorian Westphal * 64018b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 64028b69a803SPaolo Abeni * skb via __skb_ext_set(). 64038b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 64048b69a803SPaolo Abeni */ 64054930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6406df5042f4SFlorian Westphal { 64074930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6408df5042f4SFlorian Westphal 6409df5042f4SFlorian Westphal if (new) { 6410df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6411df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6412df5042f4SFlorian Westphal } 6413df5042f4SFlorian Westphal 6414df5042f4SFlorian Westphal return new; 6415df5042f4SFlorian Westphal } 6416df5042f4SFlorian Westphal 64174165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 64184165079bSFlorian Westphal unsigned int old_active) 6419df5042f4SFlorian Westphal { 6420df5042f4SFlorian Westphal struct skb_ext *new; 6421df5042f4SFlorian Westphal 6422df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6423df5042f4SFlorian Westphal return old; 6424df5042f4SFlorian Westphal 6425df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6426df5042f4SFlorian Westphal if (!new) 6427df5042f4SFlorian Westphal return NULL; 6428df5042f4SFlorian Westphal 6429df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6430df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6431df5042f4SFlorian Westphal 64324165079bSFlorian Westphal #ifdef CONFIG_XFRM 64334165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 64344165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 64354165079bSFlorian Westphal unsigned int i; 64364165079bSFlorian Westphal 64374165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 64384165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 64394165079bSFlorian Westphal } 64404165079bSFlorian Westphal #endif 6441df5042f4SFlorian Westphal __skb_ext_put(old); 6442df5042f4SFlorian Westphal return new; 6443df5042f4SFlorian Westphal } 6444df5042f4SFlorian Westphal 6445df5042f4SFlorian Westphal /** 64468b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 64478b69a803SPaolo Abeni * @skb: buffer 64488b69a803SPaolo Abeni * @id: extension id 64498b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 64508b69a803SPaolo Abeni * 64518b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 64528b69a803SPaolo Abeni * 64538b69a803SPaolo Abeni * Returns the pointer to the extension. 64548b69a803SPaolo Abeni */ 64558b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 64568b69a803SPaolo Abeni struct skb_ext *ext) 64578b69a803SPaolo Abeni { 64588b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 64598b69a803SPaolo Abeni 64608b69a803SPaolo Abeni skb_ext_put(skb); 64618b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 64628b69a803SPaolo Abeni ext->chunks = newlen; 64638b69a803SPaolo Abeni ext->offset[id] = newoff; 64648b69a803SPaolo Abeni skb->extensions = ext; 64658b69a803SPaolo Abeni skb->active_extensions = 1 << id; 64668b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 64678b69a803SPaolo Abeni } 64688b69a803SPaolo Abeni 64698b69a803SPaolo Abeni /** 6470df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6471df5042f4SFlorian Westphal * @skb: buffer 6472df5042f4SFlorian Westphal * @id: extension to allocate space for 6473df5042f4SFlorian Westphal * 6474df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6475df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6476df5042f4SFlorian Westphal * is returned. 6477df5042f4SFlorian Westphal * 6478df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6479df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6480df5042f4SFlorian Westphal * 6481df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6482df5042f4SFlorian Westphal */ 6483df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6484df5042f4SFlorian Westphal { 6485df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6486df5042f4SFlorian Westphal unsigned int newlen, newoff; 6487df5042f4SFlorian Westphal 6488df5042f4SFlorian Westphal if (skb->active_extensions) { 6489df5042f4SFlorian Westphal old = skb->extensions; 6490df5042f4SFlorian Westphal 64914165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6492df5042f4SFlorian Westphal if (!new) 6493df5042f4SFlorian Westphal return NULL; 6494df5042f4SFlorian Westphal 6495682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6496df5042f4SFlorian Westphal goto set_active; 6497df5042f4SFlorian Westphal 6498e94e50bdSPaolo Abeni newoff = new->chunks; 6499df5042f4SFlorian Westphal } else { 6500df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6501df5042f4SFlorian Westphal 65024930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6503df5042f4SFlorian Westphal if (!new) 6504df5042f4SFlorian Westphal return NULL; 6505df5042f4SFlorian Westphal } 6506df5042f4SFlorian Westphal 6507df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6508df5042f4SFlorian Westphal new->chunks = newlen; 6509df5042f4SFlorian Westphal new->offset[id] = newoff; 6510df5042f4SFlorian Westphal set_active: 6511b0999f38SPaolo Abeni skb->slow_gro = 1; 6512682ec859SPaolo Abeni skb->extensions = new; 6513df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6514df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6515df5042f4SFlorian Westphal } 6516df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6517df5042f4SFlorian Westphal 65184165079bSFlorian Westphal #ifdef CONFIG_XFRM 65194165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 65204165079bSFlorian Westphal { 65214165079bSFlorian Westphal unsigned int i; 65224165079bSFlorian Westphal 65234165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 65244165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 65254165079bSFlorian Westphal } 65264165079bSFlorian Westphal #endif 65274165079bSFlorian Westphal 652878476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 652978476d31SJeremy Kerr static void skb_ext_put_mctp(struct mctp_flow *flow) 653078476d31SJeremy Kerr { 653178476d31SJeremy Kerr if (flow->key) 653278476d31SJeremy Kerr mctp_key_unref(flow->key); 653378476d31SJeremy Kerr } 653478476d31SJeremy Kerr #endif 653578476d31SJeremy Kerr 6536df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6537df5042f4SFlorian Westphal { 6538df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6539df5042f4SFlorian Westphal 6540df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6541df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6542df5042f4SFlorian Westphal skb->extensions = NULL; 6543df5042f4SFlorian Westphal __skb_ext_put(ext); 65444165079bSFlorian Westphal #ifdef CONFIG_XFRM 65454165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 65464165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 65474165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 65484165079bSFlorian Westphal 65494165079bSFlorian Westphal skb_ext_put_sp(sp); 65504165079bSFlorian Westphal sp->len = 0; 65514165079bSFlorian Westphal #endif 6552df5042f4SFlorian Westphal } 6553df5042f4SFlorian Westphal } 6554df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6555df5042f4SFlorian Westphal 6556df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6557df5042f4SFlorian Westphal { 6558df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6559df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6560df5042f4SFlorian Westphal */ 6561df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6562df5042f4SFlorian Westphal goto free_now; 6563df5042f4SFlorian Westphal 6564df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6565df5042f4SFlorian Westphal return; 6566df5042f4SFlorian Westphal free_now: 65674165079bSFlorian Westphal #ifdef CONFIG_XFRM 65684165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 65694165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 65704165079bSFlorian Westphal #endif 657178476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 657278476d31SJeremy Kerr if (__skb_ext_exist(ext, SKB_EXT_MCTP)) 657378476d31SJeremy Kerr skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP)); 657478476d31SJeremy Kerr #endif 65754165079bSFlorian Westphal 6576df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6577df5042f4SFlorian Westphal } 6578df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6579df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 6580