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> 736a5bcd84SIlias Apalodimas #include <net/page_pool.h> 741da177e4SLinus Torvalds 757c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 76ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7751c56b00SEric Dumazet #include <linux/highmem.h> 78b245be1fSWillem de Bruijn #include <linux/capability.h> 79b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 802544af03SMatteo Croce #include <linux/indirect_call_wrapper.h> 81a1f8e7f7SAl Viro 827b7ed885SBart Van Assche #include "datagram.h" 837b7ed885SBart Van Assche 8408009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8508009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 86df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 87df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 88df5042f4SFlorian Westphal #endif 895f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 905f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /** 93f05de73bSJean Sacren * skb_panic - private function for out-of-line support 941da177e4SLinus Torvalds * @skb: buffer 951da177e4SLinus Torvalds * @sz: size 96f05de73bSJean Sacren * @addr: address 9799d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 981da177e4SLinus Torvalds * 99f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 100f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 101f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 102f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1031da177e4SLinus Torvalds */ 104f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 10599d5851eSJames Hogan const char msg[]) 1061da177e4SLinus Torvalds { 10741a46913SJesper Dangaard Brouer pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", 10899d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1094305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 11026095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1111da177e4SLinus Torvalds BUG(); 1121da177e4SLinus Torvalds } 1131da177e4SLinus Torvalds 114f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1151da177e4SLinus Torvalds { 116f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1171da177e4SLinus Torvalds } 1181da177e4SLinus Torvalds 119f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 120f05de73bSJean Sacren { 121f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 122f05de73bSJean Sacren } 123c93bdd0eSMel Gorman 12450fad4b5SAlexander Lobakin #define NAPI_SKB_CACHE_SIZE 64 125f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_BULK 16 126f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) 12750fad4b5SAlexander Lobakin 12850fad4b5SAlexander Lobakin struct napi_alloc_cache { 12950fad4b5SAlexander Lobakin struct page_frag_cache page; 13050fad4b5SAlexander Lobakin unsigned int skb_count; 13150fad4b5SAlexander Lobakin void *skb_cache[NAPI_SKB_CACHE_SIZE]; 13250fad4b5SAlexander Lobakin }; 13350fad4b5SAlexander Lobakin 13450fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 13550fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 13650fad4b5SAlexander Lobakin 137*32e3573fSYajun Deng void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 13850fad4b5SAlexander Lobakin { 13950fad4b5SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 14050fad4b5SAlexander Lobakin 14150fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 14250fad4b5SAlexander Lobakin 143*32e3573fSYajun Deng return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 14450fad4b5SAlexander Lobakin } 14550fad4b5SAlexander Lobakin EXPORT_SYMBOL(__napi_alloc_frag_align); 14650fad4b5SAlexander Lobakin 14750fad4b5SAlexander Lobakin void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 14850fad4b5SAlexander Lobakin { 14950fad4b5SAlexander Lobakin void *data; 15050fad4b5SAlexander Lobakin 15150fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 152afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 153*32e3573fSYajun Deng struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache); 154*32e3573fSYajun Deng 15550fad4b5SAlexander Lobakin data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask); 15650fad4b5SAlexander Lobakin } else { 157*32e3573fSYajun Deng struct napi_alloc_cache *nc; 158*32e3573fSYajun Deng 15950fad4b5SAlexander Lobakin local_bh_disable(); 160*32e3573fSYajun Deng nc = this_cpu_ptr(&napi_alloc_cache); 161*32e3573fSYajun Deng data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 16250fad4b5SAlexander Lobakin local_bh_enable(); 16350fad4b5SAlexander Lobakin } 16450fad4b5SAlexander Lobakin return data; 16550fad4b5SAlexander Lobakin } 16650fad4b5SAlexander Lobakin EXPORT_SYMBOL(__netdev_alloc_frag_align); 16750fad4b5SAlexander Lobakin 168f450d539SAlexander Lobakin static struct sk_buff *napi_skb_cache_get(void) 169f450d539SAlexander Lobakin { 170f450d539SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 171f450d539SAlexander Lobakin struct sk_buff *skb; 172f450d539SAlexander Lobakin 173f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 174f450d539SAlexander Lobakin nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache, 175f450d539SAlexander Lobakin GFP_ATOMIC, 176f450d539SAlexander Lobakin NAPI_SKB_CACHE_BULK, 177f450d539SAlexander Lobakin nc->skb_cache); 178f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 179f450d539SAlexander Lobakin return NULL; 180f450d539SAlexander Lobakin 181f450d539SAlexander Lobakin skb = nc->skb_cache[--nc->skb_count]; 182f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, skb); 183f450d539SAlexander Lobakin 184f450d539SAlexander Lobakin return skb; 185f450d539SAlexander Lobakin } 186f450d539SAlexander Lobakin 187ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 188483126b3SAlexander Lobakin static void __build_skb_around(struct sk_buff *skb, void *data, 189483126b3SAlexander Lobakin unsigned int frag_size) 190ba0509b6SJesper Dangaard Brouer { 191ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 192ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 193ba0509b6SJesper Dangaard Brouer 194ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 195ba0509b6SJesper Dangaard Brouer 196ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 197ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 198ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 199ba0509b6SJesper Dangaard Brouer skb->head = data; 200ba0509b6SJesper Dangaard Brouer skb->data = data; 201ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 202ba0509b6SJesper Dangaard Brouer skb->end = skb->tail + size; 203ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 204ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 205ba0509b6SJesper Dangaard Brouer 206ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 207ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 208ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 209ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 210ba0509b6SJesper Dangaard Brouer 2116370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 212ba0509b6SJesper Dangaard Brouer } 213ba0509b6SJesper Dangaard Brouer 2141da177e4SLinus Torvalds /** 2152ea2f62cSEric Dumazet * __build_skb - build a network buffer 216b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2172ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 218b2b5ce9dSEric Dumazet * 219b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 220deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2212ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2222ea2f62cSEric Dumazet * or vmalloc() 223b2b5ce9dSEric Dumazet * The return is the new skb buffer. 224b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 225b2b5ce9dSEric Dumazet * Notes : 226b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 227b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 228b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 229b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 230b2b5ce9dSEric Dumazet * before giving packet to stack. 231b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 232b2b5ce9dSEric Dumazet */ 2332ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 234b2b5ce9dSEric Dumazet { 235b2b5ce9dSEric Dumazet struct sk_buff *skb; 236b2b5ce9dSEric Dumazet 237b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 238ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 239b2b5ce9dSEric Dumazet return NULL; 240b2b5ce9dSEric Dumazet 241b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 242483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 243b2b5ce9dSEric Dumazet 244483126b3SAlexander Lobakin return skb; 245b2b5ce9dSEric Dumazet } 2462ea2f62cSEric Dumazet 2472ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 2482ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 2492ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 2502ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 2512ea2f62cSEric Dumazet */ 2522ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 2532ea2f62cSEric Dumazet { 2542ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 2552ea2f62cSEric Dumazet 2562ea2f62cSEric Dumazet if (skb && frag_size) { 2572ea2f62cSEric Dumazet skb->head_frag = 1; 2582f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 2592ea2f62cSEric Dumazet skb->pfmemalloc = 1; 2602ea2f62cSEric Dumazet } 2612ea2f62cSEric Dumazet return skb; 2622ea2f62cSEric Dumazet } 263b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 264b2b5ce9dSEric Dumazet 265ba0509b6SJesper Dangaard Brouer /** 266ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 267ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 268ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 269ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 270ba0509b6SJesper Dangaard Brouer */ 271ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 272ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 273ba0509b6SJesper Dangaard Brouer { 274ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 275ba0509b6SJesper Dangaard Brouer return NULL; 276ba0509b6SJesper Dangaard Brouer 277483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 278ba0509b6SJesper Dangaard Brouer 279483126b3SAlexander Lobakin if (frag_size) { 280ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 281ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 282ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 283ba0509b6SJesper Dangaard Brouer } 284ba0509b6SJesper Dangaard Brouer return skb; 285ba0509b6SJesper Dangaard Brouer } 286ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 287ba0509b6SJesper Dangaard Brouer 288f450d539SAlexander Lobakin /** 289f450d539SAlexander Lobakin * __napi_build_skb - build a network buffer 290f450d539SAlexander Lobakin * @data: data buffer provided by caller 291f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 292f450d539SAlexander Lobakin * 293f450d539SAlexander Lobakin * Version of __build_skb() that uses NAPI percpu caches to obtain 294f450d539SAlexander Lobakin * skbuff_head instead of inplace allocation. 295f450d539SAlexander Lobakin * 296f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 297f450d539SAlexander Lobakin */ 298f450d539SAlexander Lobakin static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size) 299f450d539SAlexander Lobakin { 300f450d539SAlexander Lobakin struct sk_buff *skb; 301f450d539SAlexander Lobakin 302f450d539SAlexander Lobakin skb = napi_skb_cache_get(); 303f450d539SAlexander Lobakin if (unlikely(!skb)) 304f450d539SAlexander Lobakin return NULL; 305f450d539SAlexander Lobakin 306f450d539SAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 307f450d539SAlexander Lobakin __build_skb_around(skb, data, frag_size); 308f450d539SAlexander Lobakin 309f450d539SAlexander Lobakin return skb; 310f450d539SAlexander Lobakin } 311f450d539SAlexander Lobakin 312f450d539SAlexander Lobakin /** 313f450d539SAlexander Lobakin * napi_build_skb - build a network buffer 314f450d539SAlexander Lobakin * @data: data buffer provided by caller 315f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 316f450d539SAlexander Lobakin * 317f450d539SAlexander Lobakin * Version of __napi_build_skb() that takes care of skb->head_frag 318f450d539SAlexander Lobakin * and skb->pfmemalloc when the data is a page or page fragment. 319f450d539SAlexander Lobakin * 320f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 321f450d539SAlexander Lobakin */ 322f450d539SAlexander Lobakin struct sk_buff *napi_build_skb(void *data, unsigned int frag_size) 323f450d539SAlexander Lobakin { 324f450d539SAlexander Lobakin struct sk_buff *skb = __napi_build_skb(data, frag_size); 325f450d539SAlexander Lobakin 326f450d539SAlexander Lobakin if (likely(skb) && frag_size) { 327f450d539SAlexander Lobakin skb->head_frag = 1; 328f450d539SAlexander Lobakin skb_propagate_pfmemalloc(virt_to_head_page(data), skb); 329f450d539SAlexander Lobakin } 330f450d539SAlexander Lobakin 331f450d539SAlexander Lobakin return skb; 332f450d539SAlexander Lobakin } 333f450d539SAlexander Lobakin EXPORT_SYMBOL(napi_build_skb); 334f450d539SAlexander Lobakin 3355381b23dSAlexander Lobakin /* 3365381b23dSAlexander Lobakin * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 3375381b23dSAlexander Lobakin * the caller if emergency pfmemalloc reserves are being used. If it is and 3385381b23dSAlexander Lobakin * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 3395381b23dSAlexander Lobakin * may be used. Otherwise, the packet data may be discarded until enough 3405381b23dSAlexander Lobakin * memory is free 3415381b23dSAlexander Lobakin */ 342ef28095fSAlexander Lobakin static void *kmalloc_reserve(size_t size, gfp_t flags, int node, 343ef28095fSAlexander Lobakin bool *pfmemalloc) 3445381b23dSAlexander Lobakin { 3455381b23dSAlexander Lobakin void *obj; 3465381b23dSAlexander Lobakin bool ret_pfmemalloc = false; 3475381b23dSAlexander Lobakin 3485381b23dSAlexander Lobakin /* 3495381b23dSAlexander Lobakin * Try a regular allocation, when that fails and we're not entitled 3505381b23dSAlexander Lobakin * to the reserves, fail. 3515381b23dSAlexander Lobakin */ 3525381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, 3535381b23dSAlexander Lobakin flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 3545381b23dSAlexander Lobakin node); 3555381b23dSAlexander Lobakin if (obj || !(gfp_pfmemalloc_allowed(flags))) 3565381b23dSAlexander Lobakin goto out; 3575381b23dSAlexander Lobakin 3585381b23dSAlexander Lobakin /* Try again but now we are using pfmemalloc reserves */ 3595381b23dSAlexander Lobakin ret_pfmemalloc = true; 3605381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, flags, node); 3615381b23dSAlexander Lobakin 3625381b23dSAlexander Lobakin out: 3635381b23dSAlexander Lobakin if (pfmemalloc) 3645381b23dSAlexander Lobakin *pfmemalloc = ret_pfmemalloc; 3655381b23dSAlexander Lobakin 3665381b23dSAlexander Lobakin return obj; 3675381b23dSAlexander Lobakin } 3685381b23dSAlexander Lobakin 3695381b23dSAlexander Lobakin /* Allocate a new skbuff. We do this ourselves so we can fill in a few 3705381b23dSAlexander Lobakin * 'private' fields and also do memory statistics to find all the 3715381b23dSAlexander Lobakin * [BEEP] leaks. 3725381b23dSAlexander Lobakin * 3735381b23dSAlexander Lobakin */ 3745381b23dSAlexander Lobakin 3755381b23dSAlexander Lobakin /** 3765381b23dSAlexander Lobakin * __alloc_skb - allocate a network buffer 3775381b23dSAlexander Lobakin * @size: size to allocate 3785381b23dSAlexander Lobakin * @gfp_mask: allocation mask 3795381b23dSAlexander Lobakin * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 3805381b23dSAlexander Lobakin * instead of head cache and allocate a cloned (child) skb. 3815381b23dSAlexander Lobakin * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 3825381b23dSAlexander Lobakin * allocations in case the data is required for writeback 3835381b23dSAlexander Lobakin * @node: numa node to allocate memory on 3845381b23dSAlexander Lobakin * 3855381b23dSAlexander Lobakin * Allocate a new &sk_buff. The returned buffer has no headroom and a 3865381b23dSAlexander Lobakin * tail room of at least size bytes. The object has a reference count 3875381b23dSAlexander Lobakin * of one. The return is the buffer. On a failure the return is %NULL. 3885381b23dSAlexander Lobakin * 3895381b23dSAlexander Lobakin * Buffers may only be allocated from interrupts using a @gfp_mask of 3905381b23dSAlexander Lobakin * %GFP_ATOMIC. 3915381b23dSAlexander Lobakin */ 3925381b23dSAlexander Lobakin struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 3935381b23dSAlexander Lobakin int flags, int node) 3945381b23dSAlexander Lobakin { 3955381b23dSAlexander Lobakin struct kmem_cache *cache; 3965381b23dSAlexander Lobakin struct sk_buff *skb; 3975381b23dSAlexander Lobakin u8 *data; 3985381b23dSAlexander Lobakin bool pfmemalloc; 3995381b23dSAlexander Lobakin 4005381b23dSAlexander Lobakin cache = (flags & SKB_ALLOC_FCLONE) 4015381b23dSAlexander Lobakin ? skbuff_fclone_cache : skbuff_head_cache; 4025381b23dSAlexander Lobakin 4035381b23dSAlexander Lobakin if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 4045381b23dSAlexander Lobakin gfp_mask |= __GFP_MEMALLOC; 4055381b23dSAlexander Lobakin 4065381b23dSAlexander Lobakin /* Get the HEAD */ 407d13612b5SAlexander Lobakin if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI && 408d13612b5SAlexander Lobakin likely(node == NUMA_NO_NODE || node == numa_mem_id())) 409d13612b5SAlexander Lobakin skb = napi_skb_cache_get(); 410d13612b5SAlexander Lobakin else 411d13612b5SAlexander Lobakin skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node); 412df1ae022SAlexander Lobakin if (unlikely(!skb)) 413df1ae022SAlexander Lobakin return NULL; 4145381b23dSAlexander Lobakin prefetchw(skb); 4155381b23dSAlexander Lobakin 4165381b23dSAlexander Lobakin /* We do our best to align skb_shared_info on a separate cache 4175381b23dSAlexander Lobakin * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 4185381b23dSAlexander Lobakin * aligned memory blocks, unless SLUB/SLAB debug is enabled. 4195381b23dSAlexander Lobakin * Both skb->head and skb_shared_info are cache line aligned. 4205381b23dSAlexander Lobakin */ 4215381b23dSAlexander Lobakin size = SKB_DATA_ALIGN(size); 4225381b23dSAlexander Lobakin size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4235381b23dSAlexander Lobakin data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 424df1ae022SAlexander Lobakin if (unlikely(!data)) 4255381b23dSAlexander Lobakin goto nodata; 4265381b23dSAlexander Lobakin /* kmalloc(size) might give us more room than requested. 4275381b23dSAlexander Lobakin * Put skb_shared_info exactly at the end of allocated zone, 4285381b23dSAlexander Lobakin * to allow max possible filling before reallocation. 4295381b23dSAlexander Lobakin */ 4305381b23dSAlexander Lobakin size = SKB_WITH_OVERHEAD(ksize(data)); 4315381b23dSAlexander Lobakin prefetchw(data + size); 4325381b23dSAlexander Lobakin 4335381b23dSAlexander Lobakin /* 4345381b23dSAlexander Lobakin * Only clear those fields we need to clear, not those that we will 4355381b23dSAlexander Lobakin * actually initialise below. Hence, don't put any more fields after 4365381b23dSAlexander Lobakin * the tail pointer in struct sk_buff! 4375381b23dSAlexander Lobakin */ 4385381b23dSAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 439f9d6725bSAlexander Lobakin __build_skb_around(skb, data, 0); 4405381b23dSAlexander Lobakin skb->pfmemalloc = pfmemalloc; 4415381b23dSAlexander Lobakin 4425381b23dSAlexander Lobakin if (flags & SKB_ALLOC_FCLONE) { 4435381b23dSAlexander Lobakin struct sk_buff_fclones *fclones; 4445381b23dSAlexander Lobakin 4455381b23dSAlexander Lobakin fclones = container_of(skb, struct sk_buff_fclones, skb1); 4465381b23dSAlexander Lobakin 4475381b23dSAlexander Lobakin skb->fclone = SKB_FCLONE_ORIG; 4485381b23dSAlexander Lobakin refcount_set(&fclones->fclone_ref, 1); 4495381b23dSAlexander Lobakin 4505381b23dSAlexander Lobakin fclones->skb2.fclone = SKB_FCLONE_CLONE; 4515381b23dSAlexander Lobakin } 4525381b23dSAlexander Lobakin 4535381b23dSAlexander Lobakin return skb; 454df1ae022SAlexander Lobakin 4555381b23dSAlexander Lobakin nodata: 4565381b23dSAlexander Lobakin kmem_cache_free(cache, skb); 457df1ae022SAlexander Lobakin return NULL; 4585381b23dSAlexander Lobakin } 4595381b23dSAlexander Lobakin EXPORT_SYMBOL(__alloc_skb); 4605381b23dSAlexander Lobakin 4617ba7aeabSSebastian Andrzej Siewior /** 462fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 463fd11a83dSAlexander Duyck * @dev: network device to receive on 464d7499160SMasanari Iida * @len: length to allocate 465fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 466fd11a83dSAlexander Duyck * 467fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 468fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 469fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 470fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 471fd11a83dSAlexander Duyck * 472fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 473fd11a83dSAlexander Duyck */ 4749451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4759451980aSAlexander Duyck gfp_t gfp_mask) 476fd11a83dSAlexander Duyck { 477b63ae8caSAlexander Duyck struct page_frag_cache *nc; 478fd11a83dSAlexander Duyck struct sk_buff *skb; 4799451980aSAlexander Duyck bool pfmemalloc; 4809451980aSAlexander Duyck void *data; 481fd11a83dSAlexander Duyck 4829451980aSAlexander Duyck len += NET_SKB_PAD; 483fd11a83dSAlexander Duyck 48466c55602SAlexander Lobakin /* If requested length is either too small or too big, 48566c55602SAlexander Lobakin * we use kmalloc() for skb->head allocation. 48666c55602SAlexander Lobakin */ 48766c55602SAlexander Lobakin if (len <= SKB_WITH_OVERHEAD(1024) || 48866c55602SAlexander Lobakin len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 489d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 490a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 491a080e7bdSAlexander Duyck if (!skb) 492a080e7bdSAlexander Duyck goto skb_fail; 493a080e7bdSAlexander Duyck goto skb_success; 494a080e7bdSAlexander Duyck } 4959451980aSAlexander Duyck 4969451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4979451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 4989451980aSAlexander Duyck 4999451980aSAlexander Duyck if (sk_memalloc_socks()) 5009451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5019451980aSAlexander Duyck 502afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 5039451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 5048c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 5059451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 50692dcabd7SSebastian Andrzej Siewior } else { 50792dcabd7SSebastian Andrzej Siewior local_bh_disable(); 50892dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 50992dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 51092dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 51192dcabd7SSebastian Andrzej Siewior local_bh_enable(); 51292dcabd7SSebastian Andrzej Siewior } 5139451980aSAlexander Duyck 5149451980aSAlexander Duyck if (unlikely(!data)) 5159451980aSAlexander Duyck return NULL; 5169451980aSAlexander Duyck 5179451980aSAlexander Duyck skb = __build_skb(data, len); 5189451980aSAlexander Duyck if (unlikely(!skb)) { 519181edb2bSAlexander Duyck skb_free_frag(data); 5209451980aSAlexander Duyck return NULL; 5219451980aSAlexander Duyck } 5229451980aSAlexander Duyck 5239451980aSAlexander Duyck if (pfmemalloc) 5249451980aSAlexander Duyck skb->pfmemalloc = 1; 5259451980aSAlexander Duyck skb->head_frag = 1; 5269451980aSAlexander Duyck 527a080e7bdSAlexander Duyck skb_success: 5288af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 5297b2e497aSChristoph Hellwig skb->dev = dev; 530fd11a83dSAlexander Duyck 531a080e7bdSAlexander Duyck skb_fail: 5328af27456SChristoph Hellwig return skb; 5338af27456SChristoph Hellwig } 534b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 5351da177e4SLinus Torvalds 536fd11a83dSAlexander Duyck /** 537fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 538fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 539d7499160SMasanari Iida * @len: length to allocate 540fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 541fd11a83dSAlexander Duyck * 542fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 543fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 544fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 545fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 546fd11a83dSAlexander Duyck * 547fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 548fd11a83dSAlexander Duyck */ 5499451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 5509451980aSAlexander Duyck gfp_t gfp_mask) 551fd11a83dSAlexander Duyck { 5523226b158SEric Dumazet struct napi_alloc_cache *nc; 553fd11a83dSAlexander Duyck struct sk_buff *skb; 5549451980aSAlexander Duyck void *data; 555fd11a83dSAlexander Duyck 5569451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 557fd11a83dSAlexander Duyck 5583226b158SEric Dumazet /* If requested length is either too small or too big, 5593226b158SEric Dumazet * we use kmalloc() for skb->head allocation. 5603226b158SEric Dumazet */ 5613226b158SEric Dumazet if (len <= SKB_WITH_OVERHEAD(1024) || 5623226b158SEric Dumazet len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 563d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 564cfb8ec65SAlexander Lobakin skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, 565cfb8ec65SAlexander Lobakin NUMA_NO_NODE); 566a080e7bdSAlexander Duyck if (!skb) 567a080e7bdSAlexander Duyck goto skb_fail; 568a080e7bdSAlexander Duyck goto skb_success; 569a080e7bdSAlexander Duyck } 5709451980aSAlexander Duyck 5713226b158SEric Dumazet nc = this_cpu_ptr(&napi_alloc_cache); 5729451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5739451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5749451980aSAlexander Duyck 5759451980aSAlexander Duyck if (sk_memalloc_socks()) 5769451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5779451980aSAlexander Duyck 5788c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5799451980aSAlexander Duyck if (unlikely(!data)) 5809451980aSAlexander Duyck return NULL; 5819451980aSAlexander Duyck 582cfb8ec65SAlexander Lobakin skb = __napi_build_skb(data, len); 5839451980aSAlexander Duyck if (unlikely(!skb)) { 584181edb2bSAlexander Duyck skb_free_frag(data); 5859451980aSAlexander Duyck return NULL; 5869451980aSAlexander Duyck } 5879451980aSAlexander Duyck 588795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5899451980aSAlexander Duyck skb->pfmemalloc = 1; 5909451980aSAlexander Duyck skb->head_frag = 1; 5919451980aSAlexander Duyck 592a080e7bdSAlexander Duyck skb_success: 593fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 594fd11a83dSAlexander Duyck skb->dev = napi->dev; 595fd11a83dSAlexander Duyck 596a080e7bdSAlexander Duyck skb_fail: 597fd11a83dSAlexander Duyck return skb; 598fd11a83dSAlexander Duyck } 599fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 600fd11a83dSAlexander Duyck 601654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 60250269e19SEric Dumazet int size, unsigned int truesize) 603654bed16SPeter Zijlstra { 604654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 605654bed16SPeter Zijlstra skb->len += size; 606654bed16SPeter Zijlstra skb->data_len += size; 60750269e19SEric Dumazet skb->truesize += truesize; 608654bed16SPeter Zijlstra } 609654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 610654bed16SPeter Zijlstra 611f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 612f8e617e1SJason Wang unsigned int truesize) 613f8e617e1SJason Wang { 614f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 615f8e617e1SJason Wang 616f8e617e1SJason Wang skb_frag_size_add(frag, size); 617f8e617e1SJason Wang skb->len += size; 618f8e617e1SJason Wang skb->data_len += size; 619f8e617e1SJason Wang skb->truesize += truesize; 620f8e617e1SJason Wang } 621f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 622f8e617e1SJason Wang 62327b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 6241da177e4SLinus Torvalds { 625bd8a7036SEric Dumazet kfree_skb_list(*listp); 62627b437c8SHerbert Xu *listp = NULL; 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds 62927b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 63027b437c8SHerbert Xu { 63127b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 63227b437c8SHerbert Xu } 63327b437c8SHerbert Xu 6341da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 6351da177e4SLinus Torvalds { 6361da177e4SLinus Torvalds struct sk_buff *list; 6371da177e4SLinus Torvalds 638fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 6391da177e4SLinus Torvalds skb_get(list); 6401da177e4SLinus Torvalds } 6411da177e4SLinus Torvalds 642d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 643d3836f21SEric Dumazet { 644181edb2bSAlexander Duyck unsigned char *head = skb->head; 645181edb2bSAlexander Duyck 6466a5bcd84SIlias Apalodimas if (skb->head_frag) { 6476a5bcd84SIlias Apalodimas if (skb_pp_recycle(skb, head)) 6486a5bcd84SIlias Apalodimas return; 649181edb2bSAlexander Duyck skb_free_frag(head); 6506a5bcd84SIlias Apalodimas } else { 651181edb2bSAlexander Duyck kfree(head); 652d3836f21SEric Dumazet } 6536a5bcd84SIlias Apalodimas } 654d3836f21SEric Dumazet 6555bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 6561da177e4SLinus Torvalds { 657ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6581da177e4SLinus Torvalds int i; 659ff04a771SEric Dumazet 660ff04a771SEric Dumazet if (skb->cloned && 661ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 662ff04a771SEric Dumazet &shinfo->dataref)) 6632cc3aeb5SIlias Apalodimas goto exit; 664ff04a771SEric Dumazet 66570c43167SJonathan Lemon skb_zcopy_clear(skb, true); 66670c43167SJonathan Lemon 667ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 6686a5bcd84SIlias Apalodimas __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); 6691da177e4SLinus Torvalds 670ff04a771SEric Dumazet if (shinfo->frag_list) 671ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6721da177e4SLinus Torvalds 673d3836f21SEric Dumazet skb_free_head(skb); 6742cc3aeb5SIlias Apalodimas exit: 6752cc3aeb5SIlias Apalodimas /* When we clone an SKB we copy the reycling bit. The pp_recycle 6762cc3aeb5SIlias Apalodimas * bit is only set on the head though, so in order to avoid races 6772cc3aeb5SIlias Apalodimas * while trying to recycle fragments on __skb_frag_unref() we need 6782cc3aeb5SIlias Apalodimas * to make one SKB responsible for triggering the recycle path. 6792cc3aeb5SIlias Apalodimas * So disable the recycling bit if an SKB is cloned and we have 6802cc3aeb5SIlias Apalodimas * additional references to to the fragmented part of the SKB. 6812cc3aeb5SIlias Apalodimas * Eventually the last SKB will have the recycling bit set and it's 6822cc3aeb5SIlias Apalodimas * dataref set to 0, which will trigger the recycling 6832cc3aeb5SIlias Apalodimas */ 6842cc3aeb5SIlias Apalodimas skb->pp_recycle = 0; 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds 6871da177e4SLinus Torvalds /* 6881da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6891da177e4SLinus Torvalds */ 6902d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6911da177e4SLinus Torvalds { 692d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 693d179cd12SDavid S. Miller 694d179cd12SDavid S. Miller switch (skb->fclone) { 695d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 6961da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 6976ffe75ebSEric Dumazet return; 698d179cd12SDavid S. Miller 699d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 700d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 7016ffe75ebSEric Dumazet 7026ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 7036ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 7046ffe75ebSEric Dumazet * while here, branch prediction will be good. 7056ffe75ebSEric Dumazet */ 7062638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 7076ffe75ebSEric Dumazet goto fastpath; 708d179cd12SDavid S. Miller break; 709d179cd12SDavid S. Miller 7106ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 711d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 712d179cd12SDavid S. Miller break; 7133ff50b79SStephen Hemminger } 7142638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 7156ffe75ebSEric Dumazet return; 7166ffe75ebSEric Dumazet fastpath: 7176ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 7181da177e4SLinus Torvalds } 7191da177e4SLinus Torvalds 7200a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 7211da177e4SLinus Torvalds { 722adf30907SEric Dumazet skb_dst_drop(skb); 7231da177e4SLinus Torvalds if (skb->destructor) { 724afa79d08SChangbin Du WARN_ON(in_hardirq()); 7251da177e4SLinus Torvalds skb->destructor(skb); 7261da177e4SLinus Torvalds } 727a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 728cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 7292fc72c7bSKOVACS Krisztian #endif 730df5042f4SFlorian Westphal skb_ext_put(skb); 73104a4bb55SLennert Buytenhek } 73204a4bb55SLennert Buytenhek 73304a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 73404a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 73504a4bb55SLennert Buytenhek { 73604a4bb55SLennert Buytenhek skb_release_head_state(skb); 737a28b1b90SFlorian Westphal if (likely(skb->head)) 7382d4baff8SHerbert Xu skb_release_data(skb); 7392d4baff8SHerbert Xu } 7401da177e4SLinus Torvalds 7412d4baff8SHerbert Xu /** 7422d4baff8SHerbert Xu * __kfree_skb - private function 7432d4baff8SHerbert Xu * @skb: buffer 7442d4baff8SHerbert Xu * 7452d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 7462d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 7472d4baff8SHerbert Xu * always call kfree_skb 7482d4baff8SHerbert Xu */ 7492d4baff8SHerbert Xu 7502d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 7512d4baff8SHerbert Xu { 7522d4baff8SHerbert Xu skb_release_all(skb); 7531da177e4SLinus Torvalds kfree_skbmem(skb); 7541da177e4SLinus Torvalds } 755b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 7561da177e4SLinus Torvalds 7571da177e4SLinus Torvalds /** 758231d06aeSJörn Engel * kfree_skb - free an sk_buff 759231d06aeSJörn Engel * @skb: buffer to free 760231d06aeSJörn Engel * 761231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 762231d06aeSJörn Engel * hit zero. 763231d06aeSJörn Engel */ 764231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 765231d06aeSJörn Engel { 7663889a803SPaolo Abeni if (!skb_unref(skb)) 767231d06aeSJörn Engel return; 7683889a803SPaolo Abeni 769ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 770231d06aeSJörn Engel __kfree_skb(skb); 771231d06aeSJörn Engel } 772b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 773231d06aeSJörn Engel 774bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 775bd8a7036SEric Dumazet { 776bd8a7036SEric Dumazet while (segs) { 777bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 778bd8a7036SEric Dumazet 779bd8a7036SEric Dumazet kfree_skb(segs); 780bd8a7036SEric Dumazet segs = next; 781bd8a7036SEric Dumazet } 782bd8a7036SEric Dumazet } 783bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 784bd8a7036SEric Dumazet 7856413139dSWillem de Bruijn /* Dump skb information and contents. 7866413139dSWillem de Bruijn * 7876413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 7886413139dSWillem de Bruijn * 789302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 7906413139dSWillem de Bruijn */ 7916413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 7926413139dSWillem de Bruijn { 7936413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 7946413139dSWillem de Bruijn struct net_device *dev = skb->dev; 7956413139dSWillem de Bruijn struct sock *sk = skb->sk; 7966413139dSWillem de Bruijn struct sk_buff *list_skb; 7976413139dSWillem de Bruijn bool has_mac, has_trans; 7986413139dSWillem de Bruijn int headroom, tailroom; 7996413139dSWillem de Bruijn int i, len, seg_len; 8006413139dSWillem de Bruijn 8016413139dSWillem de Bruijn if (full_pkt) 8026413139dSWillem de Bruijn len = skb->len; 8036413139dSWillem de Bruijn else 8046413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 8056413139dSWillem de Bruijn 8066413139dSWillem de Bruijn headroom = skb_headroom(skb); 8076413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 8086413139dSWillem de Bruijn 8096413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 8106413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 8116413139dSWillem de Bruijn 8126413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 8136413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 8146413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 8156413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 8166413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 8176413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 8186413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 8196413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 8206413139dSWillem de Bruijn skb->network_header, 8216413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 8226413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 8236413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 8246413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 8256413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 8266413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 8276413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 8286413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 8296413139dSWillem de Bruijn 8306413139dSWillem de Bruijn if (dev) 8316413139dSWillem de Bruijn printk("%sdev name=%s feat=0x%pNF\n", 8326413139dSWillem de Bruijn level, dev->name, &dev->features); 8336413139dSWillem de Bruijn if (sk) 834db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 8356413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 8366413139dSWillem de Bruijn 8376413139dSWillem de Bruijn if (full_pkt && headroom) 8386413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 8396413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 8406413139dSWillem de Bruijn 8416413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 8426413139dSWillem de Bruijn if (seg_len) 8436413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 8446413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 8456413139dSWillem de Bruijn len -= seg_len; 8466413139dSWillem de Bruijn 8476413139dSWillem de Bruijn if (full_pkt && tailroom) 8486413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 8496413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 8506413139dSWillem de Bruijn 8516413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 8526413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 8536413139dSWillem de Bruijn u32 p_off, p_len, copied; 8546413139dSWillem de Bruijn struct page *p; 8556413139dSWillem de Bruijn u8 *vaddr; 8566413139dSWillem de Bruijn 857b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 8586413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 8596413139dSWillem de Bruijn copied) { 8606413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 8616413139dSWillem de Bruijn vaddr = kmap_atomic(p); 8626413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 8636413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 8646413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 8656413139dSWillem de Bruijn kunmap_atomic(vaddr); 8666413139dSWillem de Bruijn len -= seg_len; 8676413139dSWillem de Bruijn if (!len) 8686413139dSWillem de Bruijn break; 8696413139dSWillem de Bruijn } 8706413139dSWillem de Bruijn } 8716413139dSWillem de Bruijn 8726413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8736413139dSWillem de Bruijn printk("skb fraglist:\n"); 8746413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8756413139dSWillem de Bruijn skb_dump(level, list_skb, true); 8766413139dSWillem de Bruijn } 8776413139dSWillem de Bruijn } 8786413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 8796413139dSWillem de Bruijn 880d1a203eaSStephen Hemminger /** 88125121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 88225121173SMichael S. Tsirkin * @skb: buffer that triggered an error 88325121173SMichael S. Tsirkin * 88425121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 88525121173SMichael S. Tsirkin * skb must be freed afterwards. 88625121173SMichael S. Tsirkin */ 88725121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 88825121173SMichael S. Tsirkin { 8891f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 89025121173SMichael S. Tsirkin } 89125121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 89225121173SMichael S. Tsirkin 893be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 89425121173SMichael S. Tsirkin /** 895ead2ceb0SNeil Horman * consume_skb - free an skbuff 896ead2ceb0SNeil Horman * @skb: buffer to free 897ead2ceb0SNeil Horman * 898ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 899ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 900ead2ceb0SNeil Horman * is being dropped after a failure and notes that 901ead2ceb0SNeil Horman */ 902ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 903ead2ceb0SNeil Horman { 9043889a803SPaolo Abeni if (!skb_unref(skb)) 905ead2ceb0SNeil Horman return; 9063889a803SPaolo Abeni 90707dc22e7SKoki Sanagi trace_consume_skb(skb); 908ead2ceb0SNeil Horman __kfree_skb(skb); 909ead2ceb0SNeil Horman } 910ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 911be769db2SHerbert Xu #endif 912ead2ceb0SNeil Horman 9130a463c78SPaolo Abeni /** 914c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 9150a463c78SPaolo Abeni * @skb: buffer to free 9160a463c78SPaolo Abeni * 917ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 918ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 9190a463c78SPaolo Abeni */ 920ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 9210a463c78SPaolo Abeni { 9220a463c78SPaolo Abeni trace_consume_skb(skb); 9230a463c78SPaolo Abeni skb_release_data(skb); 9240a463c78SPaolo Abeni kfree_skbmem(skb); 9250a463c78SPaolo Abeni } 9260a463c78SPaolo Abeni 927f450d539SAlexander Lobakin static void napi_skb_cache_put(struct sk_buff *skb) 928795bb1c0SJesper Dangaard Brouer { 929795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 930f450d539SAlexander Lobakin u32 i; 931795bb1c0SJesper Dangaard Brouer 932f450d539SAlexander Lobakin kasan_poison_object_data(skbuff_head_cache, skb); 933795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 934795bb1c0SJesper Dangaard Brouer 935795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 936f450d539SAlexander Lobakin for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++) 937f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, 938f450d539SAlexander Lobakin nc->skb_cache[i]); 939f450d539SAlexander Lobakin 940f450d539SAlexander Lobakin kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF, 941f450d539SAlexander Lobakin nc->skb_cache + NAPI_SKB_CACHE_HALF); 942f450d539SAlexander Lobakin nc->skb_count = NAPI_SKB_CACHE_HALF; 943795bb1c0SJesper Dangaard Brouer } 944795bb1c0SJesper Dangaard Brouer } 945f450d539SAlexander Lobakin 94615fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 94715fad714SJesper Dangaard Brouer { 9489243adfcSAlexander Lobakin skb_release_all(skb); 9499243adfcSAlexander Lobakin napi_skb_cache_put(skb); 9509243adfcSAlexander Lobakin } 9519243adfcSAlexander Lobakin 9529243adfcSAlexander Lobakin void napi_skb_free_stolen_head(struct sk_buff *skb) 9539243adfcSAlexander Lobakin { 9549efb4b5bSPaolo Abeni if (unlikely(skb->slow_gro)) { 9558550ff8dSPaul Blakey nf_reset_ct(skb); 9569243adfcSAlexander Lobakin skb_dst_drop(skb); 9579243adfcSAlexander Lobakin skb_ext_put(skb); 9585e10da53SPaolo Abeni skb_orphan(skb); 9599efb4b5bSPaolo Abeni skb->slow_gro = 0; 9609efb4b5bSPaolo Abeni } 961f450d539SAlexander Lobakin napi_skb_cache_put(skb); 96215fad714SJesper Dangaard Brouer } 963795bb1c0SJesper Dangaard Brouer 964795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 965795bb1c0SJesper Dangaard Brouer { 966885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 967795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 968885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 969795bb1c0SJesper Dangaard Brouer return; 970795bb1c0SJesper Dangaard Brouer } 971795bb1c0SJesper Dangaard Brouer 9726454eca8SYunsheng Lin lockdep_assert_in_softirq(); 9736454eca8SYunsheng Lin 9747608894eSPaolo Abeni if (!skb_unref(skb)) 975795bb1c0SJesper Dangaard Brouer return; 9767608894eSPaolo Abeni 977795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 978795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 979795bb1c0SJesper Dangaard Brouer 980795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 981abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 982795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 983795bb1c0SJesper Dangaard Brouer return; 984795bb1c0SJesper Dangaard Brouer } 985795bb1c0SJesper Dangaard Brouer 9869243adfcSAlexander Lobakin skb_release_all(skb); 987f450d539SAlexander Lobakin napi_skb_cache_put(skb); 988795bb1c0SJesper Dangaard Brouer } 989795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 990795bb1c0SJesper Dangaard Brouer 991b1937227SEric Dumazet /* Make sure a field is enclosed inside headers_start/headers_end section */ 992b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 993b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 994b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); \ 995b1937227SEric Dumazet BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 996b1937227SEric Dumazet offsetof(struct sk_buff, headers_end)); \ 997b1937227SEric Dumazet 998dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 999dec18810SHerbert Xu { 1000dec18810SHerbert Xu new->tstamp = old->tstamp; 1001b1937227SEric Dumazet /* We do not copy old->sk */ 1002dec18810SHerbert Xu new->dev = old->dev; 1003b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 10047fee226aSEric Dumazet skb_dst_copy(new, old); 1005df5042f4SFlorian Westphal __skb_ext_copy(new, old); 1006b1937227SEric Dumazet __nf_copy(new, old, false); 10076aa895b0SPatrick McHardy 1008b1937227SEric Dumazet /* Note : this field could be in headers_start/headers_end section 1009b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 1010b1937227SEric Dumazet */ 1011b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 101206021292SEliezer Tamir 1013b1937227SEric Dumazet memcpy(&new->headers_start, &old->headers_start, 1014b1937227SEric Dumazet offsetof(struct sk_buff, headers_end) - 1015b1937227SEric Dumazet offsetof(struct sk_buff, headers_start)); 1016b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 1017b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 1018b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 1019b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 1020b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 1021b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 1022b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 1023b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 1024b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 1025b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 1026b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 1027b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 1028b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 1029b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 1030b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 1031b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 1032b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 1033b1937227SEric Dumazet #endif 1034e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 1035b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 103606021292SEliezer Tamir #endif 10372bd82484SEric Dumazet #ifdef CONFIG_XPS 10382bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 10392bd82484SEric Dumazet #endif 1040b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 1041b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 1042b1937227SEric Dumazet #endif 1043b1937227SEric Dumazet 1044dec18810SHerbert Xu } 1045dec18810SHerbert Xu 104682c49a35SHerbert Xu /* 104782c49a35SHerbert Xu * You should not add any new code to this function. Add it to 104882c49a35SHerbert Xu * __copy_skb_header above instead. 104982c49a35SHerbert Xu */ 1050e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 10511da177e4SLinus Torvalds { 10521da177e4SLinus Torvalds #define C(x) n->x = skb->x 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds n->next = n->prev = NULL; 10551da177e4SLinus Torvalds n->sk = NULL; 1056dec18810SHerbert Xu __copy_skb_header(n, skb); 1057dec18810SHerbert Xu 10581da177e4SLinus Torvalds C(len); 10591da177e4SLinus Torvalds C(data_len); 10603e6b3b2eSAlexey Dobriyan C(mac_len); 1061334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 106202f1c89dSPaul Moore n->cloned = 1; 10631da177e4SLinus Torvalds n->nohdr = 0; 1064b13dda9fSEric Dumazet n->peeked = 0; 1065e78bfb07SStefano Brivio C(pfmemalloc); 10666a5bcd84SIlias Apalodimas C(pp_recycle); 10671da177e4SLinus Torvalds n->destructor = NULL; 10681da177e4SLinus Torvalds C(tail); 10691da177e4SLinus Torvalds C(end); 107002f1c89dSPaul Moore C(head); 1071d3836f21SEric Dumazet C(head_frag); 107202f1c89dSPaul Moore C(data); 107302f1c89dSPaul Moore C(truesize); 107463354797SReshetova, Elena refcount_set(&n->users, 1); 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 10771da177e4SLinus Torvalds skb->cloned = 1; 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds return n; 1080e0053ec0SHerbert Xu #undef C 1081e0053ec0SHerbert Xu } 1082e0053ec0SHerbert Xu 1083e0053ec0SHerbert Xu /** 1084da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1085da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1086da29e4b4SJakub Kicinski */ 1087da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1088da29e4b4SJakub Kicinski { 1089da29e4b4SJakub Kicinski struct sk_buff *n; 1090da29e4b4SJakub Kicinski 1091da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1092da29e4b4SJakub Kicinski if (!n) 1093da29e4b4SJakub Kicinski return NULL; 1094da29e4b4SJakub Kicinski 1095da29e4b4SJakub Kicinski n->len = first->len; 1096da29e4b4SJakub Kicinski n->data_len = first->len; 1097da29e4b4SJakub Kicinski n->truesize = first->truesize; 1098da29e4b4SJakub Kicinski 1099da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1100da29e4b4SJakub Kicinski 1101da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1102da29e4b4SJakub Kicinski n->destructor = NULL; 1103da29e4b4SJakub Kicinski 1104da29e4b4SJakub Kicinski return n; 1105da29e4b4SJakub Kicinski } 1106da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1107da29e4b4SJakub Kicinski 1108da29e4b4SJakub Kicinski /** 1109e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1110e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1111e0053ec0SHerbert Xu * @src: the skb to supply the contents 1112e0053ec0SHerbert Xu * 1113e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1114e0053ec0SHerbert Xu * supplied by the user. 1115e0053ec0SHerbert Xu * 1116e0053ec0SHerbert Xu * The target skb is returned upon exit. 1117e0053ec0SHerbert Xu */ 1118e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1119e0053ec0SHerbert Xu { 11202d4baff8SHerbert Xu skb_release_all(dst); 1121e0053ec0SHerbert Xu return __skb_clone(dst, src); 1122e0053ec0SHerbert Xu } 1123e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1124e0053ec0SHerbert Xu 11256f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1126a91dbff5SWillem de Bruijn { 1127a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1128a91dbff5SWillem de Bruijn struct user_struct *user; 1129a91dbff5SWillem de Bruijn 1130a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1131a91dbff5SWillem de Bruijn return 0; 1132a91dbff5SWillem de Bruijn 1133a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1134a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1135a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1136a91dbff5SWillem de Bruijn 1137a91dbff5SWillem de Bruijn do { 1138a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1139a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1140a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1141a91dbff5SWillem de Bruijn return -ENOBUFS; 1142a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1143a91dbff5SWillem de Bruijn old_pg); 1144a91dbff5SWillem de Bruijn 1145a91dbff5SWillem de Bruijn if (!mmp->user) { 1146a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1147a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1148a91dbff5SWillem de Bruijn } else { 1149a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1150a91dbff5SWillem de Bruijn } 1151a91dbff5SWillem de Bruijn 1152a91dbff5SWillem de Bruijn return 0; 1153a91dbff5SWillem de Bruijn } 11546f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1155a91dbff5SWillem de Bruijn 11566f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1157a91dbff5SWillem de Bruijn { 1158a91dbff5SWillem de Bruijn if (mmp->user) { 1159a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1160a91dbff5SWillem de Bruijn free_uid(mmp->user); 1161a91dbff5SWillem de Bruijn } 1162a91dbff5SWillem de Bruijn } 11636f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1164a91dbff5SWillem de Bruijn 11658c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size) 116652267790SWillem de Bruijn { 116752267790SWillem de Bruijn struct ubuf_info *uarg; 116852267790SWillem de Bruijn struct sk_buff *skb; 116952267790SWillem de Bruijn 117052267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 117152267790SWillem de Bruijn 117252267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 117352267790SWillem de Bruijn if (!skb) 117452267790SWillem de Bruijn return NULL; 117552267790SWillem de Bruijn 117652267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 117752267790SWillem de Bruijn uarg = (void *)skb->cb; 1178a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1179a91dbff5SWillem de Bruijn 1180a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1181a91dbff5SWillem de Bruijn kfree_skb(skb); 1182a91dbff5SWillem de Bruijn return NULL; 1183a91dbff5SWillem de Bruijn } 118452267790SWillem de Bruijn 11858c793822SJonathan Lemon uarg->callback = msg_zerocopy_callback; 11864ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 11874ab6c99dSWillem de Bruijn uarg->len = 1; 11884ab6c99dSWillem de Bruijn uarg->bytelen = size; 118952267790SWillem de Bruijn uarg->zerocopy = 1; 119004c2d33eSJonathan Lemon uarg->flags = SKBFL_ZEROCOPY_FRAG; 1191c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 119252267790SWillem de Bruijn sock_hold(sk); 119352267790SWillem de Bruijn 119452267790SWillem de Bruijn return uarg; 119552267790SWillem de Bruijn } 11968c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_alloc); 119752267790SWillem de Bruijn 119852267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 119952267790SWillem de Bruijn { 120052267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 120152267790SWillem de Bruijn } 120252267790SWillem de Bruijn 12038c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 12044ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 12054ab6c99dSWillem de Bruijn { 12064ab6c99dSWillem de Bruijn if (uarg) { 12074ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 12084ab6c99dSWillem de Bruijn u32 bytelen, next; 12094ab6c99dSWillem de Bruijn 12104ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 12114ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 12124ab6c99dSWillem de Bruijn */ 12134ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 12144ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 12154ab6c99dSWillem de Bruijn return NULL; 12164ab6c99dSWillem de Bruijn } 12174ab6c99dSWillem de Bruijn 12184ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 12194ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 12204ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 12214ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12224ab6c99dSWillem de Bruijn goto new_alloc; 12234ab6c99dSWillem de Bruijn return NULL; 12244ab6c99dSWillem de Bruijn } 12254ab6c99dSWillem de Bruijn 12264ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 12274ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1228a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1229a91dbff5SWillem de Bruijn return NULL; 12304ab6c99dSWillem de Bruijn uarg->len++; 12314ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 12324ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1233100f6d8eSWillem de Bruijn 1234100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1235100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12368e044917SJonathan Lemon net_zcopy_get(uarg); 1237100f6d8eSWillem de Bruijn 12384ab6c99dSWillem de Bruijn return uarg; 12394ab6c99dSWillem de Bruijn } 12404ab6c99dSWillem de Bruijn } 12414ab6c99dSWillem de Bruijn 12424ab6c99dSWillem de Bruijn new_alloc: 12438c793822SJonathan Lemon return msg_zerocopy_alloc(sk, size); 12444ab6c99dSWillem de Bruijn } 12458c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_realloc); 12464ab6c99dSWillem de Bruijn 12474ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 12484ab6c99dSWillem de Bruijn { 12494ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 12504ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 12514ab6c99dSWillem de Bruijn u64 sum_len; 12524ab6c99dSWillem de Bruijn 12534ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 12544ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 12554ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 12564ab6c99dSWillem de Bruijn 12574ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 12584ab6c99dSWillem de Bruijn return false; 12594ab6c99dSWillem de Bruijn 12604ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 12614ab6c99dSWillem de Bruijn return false; 12624ab6c99dSWillem de Bruijn 12634ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 12644ab6c99dSWillem de Bruijn return true; 12654ab6c99dSWillem de Bruijn } 12664ab6c99dSWillem de Bruijn 12678c793822SJonathan Lemon static void __msg_zerocopy_callback(struct ubuf_info *uarg) 126852267790SWillem de Bruijn { 12694ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 127052267790SWillem de Bruijn struct sock_exterr_skb *serr; 127152267790SWillem de Bruijn struct sock *sk = skb->sk; 12724ab6c99dSWillem de Bruijn struct sk_buff_head *q; 12734ab6c99dSWillem de Bruijn unsigned long flags; 12743bdd5ee0SWillem de Bruijn bool is_zerocopy; 12754ab6c99dSWillem de Bruijn u32 lo, hi; 12764ab6c99dSWillem de Bruijn u16 len; 127752267790SWillem de Bruijn 1278ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1279ccaffff1SWillem de Bruijn 12804ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 12814ab6c99dSWillem de Bruijn * so do not queue a completion notification 12824ab6c99dSWillem de Bruijn */ 12834ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 128452267790SWillem de Bruijn goto release; 128552267790SWillem de Bruijn 12864ab6c99dSWillem de Bruijn len = uarg->len; 12874ab6c99dSWillem de Bruijn lo = uarg->id; 12884ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 12893bdd5ee0SWillem de Bruijn is_zerocopy = uarg->zerocopy; 12904ab6c99dSWillem de Bruijn 129152267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 129252267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 129352267790SWillem de Bruijn serr->ee.ee_errno = 0; 129452267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 12954ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 12964ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 12973bdd5ee0SWillem de Bruijn if (!is_zerocopy) 129852267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 129952267790SWillem de Bruijn 13004ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 13014ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 13024ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 13034ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 13044ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 13054ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 130652267790SWillem de Bruijn skb = NULL; 13074ab6c99dSWillem de Bruijn } 13084ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 130952267790SWillem de Bruijn 1310e3ae2365SAlexander Aring sk_error_report(sk); 131152267790SWillem de Bruijn 131252267790SWillem de Bruijn release: 131352267790SWillem de Bruijn consume_skb(skb); 131452267790SWillem de Bruijn sock_put(sk); 131552267790SWillem de Bruijn } 131675518851SJonathan Lemon 13178c793822SJonathan Lemon void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 131836177832SJonathan Lemon bool success) 131975518851SJonathan Lemon { 132075518851SJonathan Lemon uarg->zerocopy = uarg->zerocopy & success; 132175518851SJonathan Lemon 132275518851SJonathan Lemon if (refcount_dec_and_test(&uarg->refcnt)) 13238c793822SJonathan Lemon __msg_zerocopy_callback(uarg); 132475518851SJonathan Lemon } 13258c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_callback); 132652267790SWillem de Bruijn 13278c793822SJonathan Lemon void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 132852267790SWillem de Bruijn { 132952267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 133052267790SWillem de Bruijn 133152267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 13324ab6c99dSWillem de Bruijn uarg->len--; 133352267790SWillem de Bruijn 133452900d22SWillem de Bruijn if (have_uref) 13358c793822SJonathan Lemon msg_zerocopy_callback(NULL, uarg, true); 133652267790SWillem de Bruijn } 13378c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort); 133852267790SWillem de Bruijn 1339b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1340b5947e5dSWillem de Bruijn { 1341b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1342b5947e5dSWillem de Bruijn } 1343b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1344b5947e5dSWillem de Bruijn 134552267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 134652267790SWillem de Bruijn struct msghdr *msg, int len, 134752267790SWillem de Bruijn struct ubuf_info *uarg) 134852267790SWillem de Bruijn { 13494ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 135052267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 135152267790SWillem de Bruijn int err, orig_len = skb->len; 135252267790SWillem de Bruijn 13534ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 13544ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 13554ab6c99dSWillem de Bruijn */ 13564ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 13574ab6c99dSWillem de Bruijn return -EEXIST; 13584ab6c99dSWillem de Bruijn 135952267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 136052267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 136154d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 136254d43117SWillem de Bruijn 136352267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 136452267790SWillem de Bruijn msg->msg_iter = orig_iter; 136554d43117SWillem de Bruijn skb->sk = sk; 136652267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 136754d43117SWillem de Bruijn skb->sk = save_sk; 136852267790SWillem de Bruijn return err; 136952267790SWillem de Bruijn } 137052267790SWillem de Bruijn 137152900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 137252267790SWillem de Bruijn return skb->len - orig_len; 137352267790SWillem de Bruijn } 137452267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 137552267790SWillem de Bruijn 13761f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 137752267790SWillem de Bruijn gfp_t gfp_mask) 137852267790SWillem de Bruijn { 137952267790SWillem de Bruijn if (skb_zcopy(orig)) { 138052267790SWillem de Bruijn if (skb_zcopy(nskb)) { 138152267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 138252267790SWillem de Bruijn if (!gfp_mask) { 138352267790SWillem de Bruijn WARN_ON_ONCE(1); 138452267790SWillem de Bruijn return -ENOMEM; 138552267790SWillem de Bruijn } 138652267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 138752267790SWillem de Bruijn return 0; 138852267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 138952267790SWillem de Bruijn return -EIO; 139052267790SWillem de Bruijn } 139152900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 139252267790SWillem de Bruijn } 139352267790SWillem de Bruijn return 0; 139452267790SWillem de Bruijn } 139552267790SWillem de Bruijn 13962c53040fSBen Hutchings /** 13972c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 139848c83012SMichael S. Tsirkin * @skb: the skb to modify 139948c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 140048c83012SMichael S. Tsirkin * 140106b4feb3SJonathan Lemon * This must be called on skb with SKBFL_ZEROCOPY_ENABLE. 140248c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 140348c83012SMichael S. Tsirkin * to userspace pages. 140448c83012SMichael S. Tsirkin * 140548c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 140648c83012SMichael S. Tsirkin * %GFP_ATOMIC. 140748c83012SMichael S. Tsirkin * 140848c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 140948c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 141048c83012SMichael S. Tsirkin */ 141148c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1412a6686f2fSShirley Ma { 1413a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1414a6686f2fSShirley Ma struct page *page, *head = NULL; 14153ece7826SWillem de Bruijn int i, new_frags; 14163ece7826SWillem de Bruijn u32 d_off; 1417a6686f2fSShirley Ma 14183ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 14193ece7826SWillem de Bruijn return -EINVAL; 14203ece7826SWillem de Bruijn 1421f72c4ac6SWillem de Bruijn if (!num_frags) 1422f72c4ac6SWillem de Bruijn goto release; 1423f72c4ac6SWillem de Bruijn 14243ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 14253ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 142602756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1427a6686f2fSShirley Ma if (!page) { 1428a6686f2fSShirley Ma while (head) { 142940dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1430a6686f2fSShirley Ma put_page(head); 1431a6686f2fSShirley Ma head = next; 1432a6686f2fSShirley Ma } 1433a6686f2fSShirley Ma return -ENOMEM; 1434a6686f2fSShirley Ma } 14353ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 14363ece7826SWillem de Bruijn head = page; 14373ece7826SWillem de Bruijn } 14383ece7826SWillem de Bruijn 14393ece7826SWillem de Bruijn page = head; 14403ece7826SWillem de Bruijn d_off = 0; 14413ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 14423ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 14433ece7826SWillem de Bruijn u32 p_off, p_len, copied; 14443ece7826SWillem de Bruijn struct page *p; 14453ece7826SWillem de Bruijn u8 *vaddr; 1446c613c209SWillem de Bruijn 1447b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1448c613c209SWillem de Bruijn p, p_off, p_len, copied) { 14493ece7826SWillem de Bruijn u32 copy, done = 0; 1450c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 14513ece7826SWillem de Bruijn 14523ece7826SWillem de Bruijn while (done < p_len) { 14533ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 14543ece7826SWillem de Bruijn d_off = 0; 14553ece7826SWillem de Bruijn page = (struct page *)page_private(page); 14563ece7826SWillem de Bruijn } 14573ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 14583ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 14593ece7826SWillem de Bruijn vaddr + p_off + done, copy); 14603ece7826SWillem de Bruijn done += copy; 14613ece7826SWillem de Bruijn d_off += copy; 14623ece7826SWillem de Bruijn } 146351c56b00SEric Dumazet kunmap_atomic(vaddr); 1464c613c209SWillem de Bruijn } 1465a6686f2fSShirley Ma } 1466a6686f2fSShirley Ma 1467a6686f2fSShirley Ma /* skb frags release userspace buffers */ 146802756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1469a8605c60SIan Campbell skb_frag_unref(skb, i); 1470a6686f2fSShirley Ma 1471a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 14723ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 14733ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 147440dadff2SSunghan Suh head = (struct page *)page_private(head); 1475a6686f2fSShirley Ma } 14763ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 14773ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 147848c83012SMichael S. Tsirkin 1479b90ddd56SWillem de Bruijn release: 14801f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1481a6686f2fSShirley Ma return 0; 1482a6686f2fSShirley Ma } 1483dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1484a6686f2fSShirley Ma 1485e0053ec0SHerbert Xu /** 1486e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1487e0053ec0SHerbert Xu * @skb: buffer to clone 1488e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1489e0053ec0SHerbert Xu * 1490e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1491e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1492e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1493e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1494e0053ec0SHerbert Xu * 1495e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1496e0053ec0SHerbert Xu * %GFP_ATOMIC. 1497e0053ec0SHerbert Xu */ 1498e0053ec0SHerbert Xu 1499e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1500e0053ec0SHerbert Xu { 1501d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1502d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1503d0bf4a9eSEric Dumazet skb1); 15046ffe75ebSEric Dumazet struct sk_buff *n; 1505e0053ec0SHerbert Xu 150670008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1507a6686f2fSShirley Ma return NULL; 1508a6686f2fSShirley Ma 1509e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 15102638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 15116ffe75ebSEric Dumazet n = &fclones->skb2; 15122638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1513e0053ec0SHerbert Xu } else { 1514c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1515c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1516c93bdd0eSMel Gorman 1517e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1518e0053ec0SHerbert Xu if (!n) 1519e0053ec0SHerbert Xu return NULL; 1520fe55f6d5SVegard Nossum 1521e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1522e0053ec0SHerbert Xu } 1523e0053ec0SHerbert Xu 1524e0053ec0SHerbert Xu return __skb_clone(n, skb); 15251da177e4SLinus Torvalds } 1526b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 15271da177e4SLinus Torvalds 1528b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1529f5b17294SPravin B Shelar { 1530030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1531030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1532030737bcSEric Dumazet skb->csum_start += off; 1533f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1534f5b17294SPravin B Shelar skb->transport_header += off; 1535f5b17294SPravin B Shelar skb->network_header += off; 1536f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1537f5b17294SPravin B Shelar skb->mac_header += off; 1538f5b17294SPravin B Shelar skb->inner_transport_header += off; 1539f5b17294SPravin B Shelar skb->inner_network_header += off; 1540aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1541f5b17294SPravin B Shelar } 1542b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1543f5b17294SPravin B Shelar 154408303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 15451da177e4SLinus Torvalds { 1546dec18810SHerbert Xu __copy_skb_header(new, old); 1547dec18810SHerbert Xu 15487967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 15497967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 15507967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 15511da177e4SLinus Torvalds } 155208303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 15531da177e4SLinus Torvalds 1554c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1555c93bdd0eSMel Gorman { 1556c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1557c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1558c93bdd0eSMel Gorman return 0; 1559c93bdd0eSMel Gorman } 1560c93bdd0eSMel Gorman 15611da177e4SLinus Torvalds /** 15621da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 15631da177e4SLinus Torvalds * @skb: buffer to copy 15641da177e4SLinus Torvalds * @gfp_mask: allocation priority 15651da177e4SLinus Torvalds * 15661da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 15671da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 15681da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 15691da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 15701da177e4SLinus Torvalds * 15711da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 15721da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 15731da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 15741da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 15751da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 15761da177e4SLinus Torvalds */ 15771da177e4SLinus Torvalds 1578dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 15791da177e4SLinus Torvalds { 15806602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1581ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1582c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1583c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 15846602cebbSEric Dumazet 15851da177e4SLinus Torvalds if (!n) 15861da177e4SLinus Torvalds return NULL; 15871da177e4SLinus Torvalds 15881da177e4SLinus Torvalds /* Set the data pointer */ 15891da177e4SLinus Torvalds skb_reserve(n, headerlen); 15901da177e4SLinus Torvalds /* Set the tail pointer and length */ 15911da177e4SLinus Torvalds skb_put(n, skb->len); 15921da177e4SLinus Torvalds 15939f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 15941da177e4SLinus Torvalds 159508303c18SIlya Lesokhin skb_copy_header(n, skb); 15961da177e4SLinus Torvalds return n; 15971da177e4SLinus Torvalds } 1598b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 15991da177e4SLinus Torvalds 16001da177e4SLinus Torvalds /** 1601bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 16021da177e4SLinus Torvalds * @skb: buffer to copy 1603117632e6SEric Dumazet * @headroom: headroom of new skb 16041da177e4SLinus Torvalds * @gfp_mask: allocation priority 1605bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1606bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1607bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 16081da177e4SLinus Torvalds * 16091da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 16101da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 16111da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 16121da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 16131da177e4SLinus Torvalds * or the pointer to the buffer on success. 16141da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 16151da177e4SLinus Torvalds */ 16161da177e4SLinus Torvalds 1617bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1618bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 16191da177e4SLinus Torvalds { 1620117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1621bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1622bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 16236602cebbSEric Dumazet 16241da177e4SLinus Torvalds if (!n) 16251da177e4SLinus Torvalds goto out; 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds /* Set the data pointer */ 1628117632e6SEric Dumazet skb_reserve(n, headroom); 16291da177e4SLinus Torvalds /* Set the tail pointer and length */ 16301da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 16311da177e4SLinus Torvalds /* Copy the bytes */ 1632d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 16331da177e4SLinus Torvalds 163425f484a6SHerbert Xu n->truesize += skb->data_len; 16351da177e4SLinus Torvalds n->data_len = skb->data_len; 16361da177e4SLinus Torvalds n->len = skb->len; 16371da177e4SLinus Torvalds 16381da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 16391da177e4SLinus Torvalds int i; 16401da177e4SLinus Torvalds 16411f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 16421f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 16431511022cSDan Carpenter kfree_skb(n); 16441511022cSDan Carpenter n = NULL; 1645a6686f2fSShirley Ma goto out; 1646a6686f2fSShirley Ma } 16471da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16481da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1649ea2ab693SIan Campbell skb_frag_ref(skb, i); 16501da177e4SLinus Torvalds } 16511da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 16521da177e4SLinus Torvalds } 16531da177e4SLinus Torvalds 165421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 16551da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 16561da177e4SLinus Torvalds skb_clone_fraglist(n); 16571da177e4SLinus Torvalds } 16581da177e4SLinus Torvalds 165908303c18SIlya Lesokhin skb_copy_header(n, skb); 16601da177e4SLinus Torvalds out: 16611da177e4SLinus Torvalds return n; 16621da177e4SLinus Torvalds } 1663bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds /** 16661da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 16671da177e4SLinus Torvalds * @skb: buffer to reallocate 16681da177e4SLinus Torvalds * @nhead: room to add at head 16691da177e4SLinus Torvalds * @ntail: room to add at tail 16701da177e4SLinus Torvalds * @gfp_mask: allocation priority 16711da177e4SLinus Torvalds * 1672bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1673bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 16741da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 16751da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 16761da177e4SLinus Torvalds * 16771da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 16781da177e4SLinus Torvalds * reloaded after call to this function. 16791da177e4SLinus Torvalds */ 16801da177e4SLinus Torvalds 168186a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1682dd0fc66fSAl Viro gfp_t gfp_mask) 16831da177e4SLinus Torvalds { 1684158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1685158f323bSEric Dumazet int size = osize + nhead + ntail; 16861da177e4SLinus Torvalds long off; 1687158f323bSEric Dumazet u8 *data; 16881da177e4SLinus Torvalds 16894edd87adSHerbert Xu BUG_ON(nhead < 0); 16904edd87adSHerbert Xu 16919f77fad3STim Hansen BUG_ON(skb_shared(skb)); 16921da177e4SLinus Torvalds 16931da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 16941da177e4SLinus Torvalds 1695c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1696c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1697c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1698c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 16991da177e4SLinus Torvalds if (!data) 17001da177e4SLinus Torvalds goto nodata; 170187151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 17021da177e4SLinus Torvalds 17031da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 17046602cebbSEric Dumazet * optimized for the cases when header is void. 17056602cebbSEric Dumazet */ 17066602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 17076602cebbSEric Dumazet 17086602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 17096602cebbSEric Dumazet skb_shinfo(skb), 1710fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 17111da177e4SLinus Torvalds 17123e24591aSAlexander Duyck /* 17133e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 17143e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 17153e24591aSAlexander Duyck * be since all we did is relocate the values 17163e24591aSAlexander Duyck */ 17173e24591aSAlexander Duyck if (skb_cloned(skb)) { 171870008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1719a6686f2fSShirley Ma goto nofrags; 17201f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1721c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 17221da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1723ea2ab693SIan Campbell skb_frag_ref(skb, i); 17241da177e4SLinus Torvalds 172521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 17261da177e4SLinus Torvalds skb_clone_fraglist(skb); 17271da177e4SLinus Torvalds 17281da177e4SLinus Torvalds skb_release_data(skb); 17293e24591aSAlexander Duyck } else { 17303e24591aSAlexander Duyck skb_free_head(skb); 17311fd63041SEric Dumazet } 17321da177e4SLinus Torvalds off = (data + nhead) - skb->head; 17331da177e4SLinus Torvalds 17341da177e4SLinus Torvalds skb->head = data; 1735d3836f21SEric Dumazet skb->head_frag = 0; 17361da177e4SLinus Torvalds skb->data += off; 17374305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 17384305b541SArnaldo Carvalho de Melo skb->end = size; 173956eb8882SPatrick McHardy off = nhead; 17404305b541SArnaldo Carvalho de Melo #else 17414305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 174256eb8882SPatrick McHardy #endif 174327a884dcSArnaldo Carvalho de Melo skb->tail += off; 1744b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 17451da177e4SLinus Torvalds skb->cloned = 0; 1746334a8132SPatrick McHardy skb->hdr_len = 0; 17471da177e4SLinus Torvalds skb->nohdr = 0; 17481da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1749158f323bSEric Dumazet 1750de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1751de8f3a83SDaniel Borkmann 1752158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1753158f323bSEric Dumazet * For the moment, we really care of rx path, or 1754158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1755158f323bSEric Dumazet */ 1756158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1757158f323bSEric Dumazet skb->truesize += size - osize; 1758158f323bSEric Dumazet 17591da177e4SLinus Torvalds return 0; 17601da177e4SLinus Torvalds 1761a6686f2fSShirley Ma nofrags: 1762a6686f2fSShirley Ma kfree(data); 17631da177e4SLinus Torvalds nodata: 17641da177e4SLinus Torvalds return -ENOMEM; 17651da177e4SLinus Torvalds } 1766b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 17671da177e4SLinus Torvalds 17681da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 17691da177e4SLinus Torvalds 17701da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 17711da177e4SLinus Torvalds { 17721da177e4SLinus Torvalds struct sk_buff *skb2; 17731da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds if (delta <= 0) 17761da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 17771da177e4SLinus Torvalds else { 17781da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 17791da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 17801da177e4SLinus Torvalds GFP_ATOMIC)) { 17811da177e4SLinus Torvalds kfree_skb(skb2); 17821da177e4SLinus Torvalds skb2 = NULL; 17831da177e4SLinus Torvalds } 17841da177e4SLinus Torvalds } 17851da177e4SLinus Torvalds return skb2; 17861da177e4SLinus Torvalds } 1787b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 17881da177e4SLinus Torvalds 17891da177e4SLinus Torvalds /** 1790f1260ff1SVasily Averin * skb_expand_head - reallocate header of &sk_buff 1791f1260ff1SVasily Averin * @skb: buffer to reallocate 1792f1260ff1SVasily Averin * @headroom: needed headroom 1793f1260ff1SVasily Averin * 1794f1260ff1SVasily Averin * Unlike skb_realloc_headroom, this one does not allocate a new skb 1795f1260ff1SVasily Averin * if possible; copies skb->sk to new skb as needed 1796f1260ff1SVasily Averin * and frees original skb in case of failures. 1797f1260ff1SVasily Averin * 1798f1260ff1SVasily Averin * It expect increased headroom and generates warning otherwise. 1799f1260ff1SVasily Averin */ 1800f1260ff1SVasily Averin 1801f1260ff1SVasily Averin struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom) 1802f1260ff1SVasily Averin { 1803f1260ff1SVasily Averin int delta = headroom - skb_headroom(skb); 1804f1260ff1SVasily Averin 1805f1260ff1SVasily Averin if (WARN_ONCE(delta <= 0, 1806f1260ff1SVasily Averin "%s is expecting an increase in the headroom", __func__)) 1807f1260ff1SVasily Averin return skb; 1808f1260ff1SVasily Averin 1809f1260ff1SVasily Averin /* pskb_expand_head() might crash, if skb is shared */ 1810f1260ff1SVasily Averin if (skb_shared(skb)) { 1811f1260ff1SVasily Averin struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1812f1260ff1SVasily Averin 1813f1260ff1SVasily Averin if (likely(nskb)) { 1814f1260ff1SVasily Averin if (skb->sk) 1815f1260ff1SVasily Averin skb_set_owner_w(nskb, skb->sk); 1816f1260ff1SVasily Averin consume_skb(skb); 1817f1260ff1SVasily Averin } else { 1818f1260ff1SVasily Averin kfree_skb(skb); 1819f1260ff1SVasily Averin } 1820f1260ff1SVasily Averin skb = nskb; 1821f1260ff1SVasily Averin } 1822f1260ff1SVasily Averin if (skb && 1823f1260ff1SVasily Averin pskb_expand_head(skb, SKB_DATA_ALIGN(delta), 0, GFP_ATOMIC)) { 1824f1260ff1SVasily Averin kfree_skb(skb); 1825f1260ff1SVasily Averin skb = NULL; 1826f1260ff1SVasily Averin } 1827f1260ff1SVasily Averin return skb; 1828f1260ff1SVasily Averin } 1829f1260ff1SVasily Averin EXPORT_SYMBOL(skb_expand_head); 1830f1260ff1SVasily Averin 1831f1260ff1SVasily Averin /** 18321da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 18331da177e4SLinus Torvalds * @skb: buffer to copy 18341da177e4SLinus Torvalds * @newheadroom: new free bytes at head 18351da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 18361da177e4SLinus Torvalds * @gfp_mask: allocation priority 18371da177e4SLinus Torvalds * 18381da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 18391da177e4SLinus Torvalds * allocate additional space. 18401da177e4SLinus Torvalds * 18411da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 18421da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 18431da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 18441da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 18451da177e4SLinus Torvalds * 18461da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 18471da177e4SLinus Torvalds * is called from an interrupt. 18481da177e4SLinus Torvalds */ 18491da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 185086a76cafSVictor Fusco int newheadroom, int newtailroom, 1851dd0fc66fSAl Viro gfp_t gfp_mask) 18521da177e4SLinus Torvalds { 18531da177e4SLinus Torvalds /* 18541da177e4SLinus Torvalds * Allocate the copy buffer 18551da177e4SLinus Torvalds */ 1856c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1857c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1858c93bdd0eSMel Gorman NUMA_NO_NODE); 1859efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 18601da177e4SLinus Torvalds int head_copy_len, head_copy_off; 18611da177e4SLinus Torvalds 18621da177e4SLinus Torvalds if (!n) 18631da177e4SLinus Torvalds return NULL; 18641da177e4SLinus Torvalds 18651da177e4SLinus Torvalds skb_reserve(n, newheadroom); 18661da177e4SLinus Torvalds 18671da177e4SLinus Torvalds /* Set the tail pointer and length */ 18681da177e4SLinus Torvalds skb_put(n, skb->len); 18691da177e4SLinus Torvalds 1870efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 18711da177e4SLinus Torvalds head_copy_off = 0; 18721da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 18731da177e4SLinus Torvalds head_copy_len = newheadroom; 18741da177e4SLinus Torvalds else 18751da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 18761da177e4SLinus Torvalds 18771da177e4SLinus Torvalds /* Copy the linear header and data. */ 18789f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 18799f77fad3STim Hansen skb->len + head_copy_len)); 18801da177e4SLinus Torvalds 188108303c18SIlya Lesokhin skb_copy_header(n, skb); 18821da177e4SLinus Torvalds 1883030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1884efd1e8d5SPatrick McHardy 18851da177e4SLinus Torvalds return n; 18861da177e4SLinus Torvalds } 1887b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds /** 1890cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 18911da177e4SLinus Torvalds * @skb: buffer to pad 18921da177e4SLinus Torvalds * @pad: space to pad 1893cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 18941da177e4SLinus Torvalds * 18951da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 18961da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 18971da177e4SLinus Torvalds * beyond the buffer end onto the wire. 18981da177e4SLinus Torvalds * 1899cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1900cd0a137aSFlorian Fainelli * if @free_on_error is true. 19011da177e4SLinus Torvalds */ 19021da177e4SLinus Torvalds 1903cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 19041da177e4SLinus Torvalds { 19055b057c6bSHerbert Xu int err; 19065b057c6bSHerbert Xu int ntail; 19071da177e4SLinus Torvalds 19081da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 19095b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 19101da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 19115b057c6bSHerbert Xu return 0; 19121da177e4SLinus Torvalds } 19131da177e4SLinus Torvalds 19144305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 19155b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 19165b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 19175b057c6bSHerbert Xu if (unlikely(err)) 19185b057c6bSHerbert Xu goto free_skb; 19195b057c6bSHerbert Xu } 19205b057c6bSHerbert Xu 19215b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 19225b057c6bSHerbert Xu * to be audited. 19235b057c6bSHerbert Xu */ 19245b057c6bSHerbert Xu err = skb_linearize(skb); 19255b057c6bSHerbert Xu if (unlikely(err)) 19265b057c6bSHerbert Xu goto free_skb; 19275b057c6bSHerbert Xu 19285b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 19295b057c6bSHerbert Xu return 0; 19305b057c6bSHerbert Xu 19315b057c6bSHerbert Xu free_skb: 1932cd0a137aSFlorian Fainelli if (free_on_error) 19331da177e4SLinus Torvalds kfree_skb(skb); 19345b057c6bSHerbert Xu return err; 19351da177e4SLinus Torvalds } 1936cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 19371da177e4SLinus Torvalds 19380dde3e16SIlpo Järvinen /** 19390c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 19400c7ddf36SMathias Krause * @skb: start of the buffer to use 19410c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 19420c7ddf36SMathias Krause * @len: amount of data to add 19430c7ddf36SMathias Krause * 19440c7ddf36SMathias Krause * This function extends the used data area of the potentially 19450c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 19460c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 19470c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 19480c7ddf36SMathias Krause * returned. 19490c7ddf36SMathias Krause */ 19500c7ddf36SMathias Krause 19514df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 19520c7ddf36SMathias Krause { 19530c7ddf36SMathias Krause if (tail != skb) { 19540c7ddf36SMathias Krause skb->data_len += len; 19550c7ddf36SMathias Krause skb->len += len; 19560c7ddf36SMathias Krause } 19570c7ddf36SMathias Krause return skb_put(tail, len); 19580c7ddf36SMathias Krause } 19590c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 19600c7ddf36SMathias Krause 19610c7ddf36SMathias Krause /** 19620dde3e16SIlpo Järvinen * skb_put - add data to a buffer 19630dde3e16SIlpo Järvinen * @skb: buffer to use 19640dde3e16SIlpo Järvinen * @len: amount of data to add 19650dde3e16SIlpo Järvinen * 19660dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 19670dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 19680dde3e16SIlpo Järvinen * first byte of the extra data is returned. 19690dde3e16SIlpo Järvinen */ 19704df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 19710dde3e16SIlpo Järvinen { 19724df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 19730dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 19740dde3e16SIlpo Järvinen skb->tail += len; 19750dde3e16SIlpo Järvinen skb->len += len; 19760dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 19770dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 19780dde3e16SIlpo Järvinen return tmp; 19790dde3e16SIlpo Järvinen } 19800dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 19810dde3e16SIlpo Järvinen 19826be8ac2fSIlpo Järvinen /** 1983c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1984c2aa270aSIlpo Järvinen * @skb: buffer to use 1985c2aa270aSIlpo Järvinen * @len: amount of data to add 1986c2aa270aSIlpo Järvinen * 1987c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1988c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1989c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1990c2aa270aSIlpo Järvinen */ 1991d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 1992c2aa270aSIlpo Järvinen { 1993c2aa270aSIlpo Järvinen skb->data -= len; 1994c2aa270aSIlpo Järvinen skb->len += len; 1995c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 1996c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1997c2aa270aSIlpo Järvinen return skb->data; 1998c2aa270aSIlpo Järvinen } 1999c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 2000c2aa270aSIlpo Järvinen 2001c2aa270aSIlpo Järvinen /** 20026be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 20036be8ac2fSIlpo Järvinen * @skb: buffer to use 20046be8ac2fSIlpo Järvinen * @len: amount of data to remove 20056be8ac2fSIlpo Järvinen * 20066be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 20076be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 20086be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 20096be8ac2fSIlpo Järvinen * the old data. 20106be8ac2fSIlpo Järvinen */ 2011af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 20126be8ac2fSIlpo Järvinen { 201347d29646SDavid S. Miller return skb_pull_inline(skb, len); 20146be8ac2fSIlpo Järvinen } 20156be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 20166be8ac2fSIlpo Järvinen 2017419ae74eSIlpo Järvinen /** 2018419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 2019419ae74eSIlpo Järvinen * @skb: buffer to alter 2020419ae74eSIlpo Järvinen * @len: new length 2021419ae74eSIlpo Järvinen * 2022419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 2023419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 2024419ae74eSIlpo Järvinen * The skb must be linear. 2025419ae74eSIlpo Järvinen */ 2026419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 2027419ae74eSIlpo Järvinen { 2028419ae74eSIlpo Järvinen if (skb->len > len) 2029419ae74eSIlpo Järvinen __skb_trim(skb, len); 2030419ae74eSIlpo Järvinen } 2031419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 2032419ae74eSIlpo Järvinen 20333cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 20341da177e4SLinus Torvalds */ 20351da177e4SLinus Torvalds 20363cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 20371da177e4SLinus Torvalds { 203827b437c8SHerbert Xu struct sk_buff **fragp; 203927b437c8SHerbert Xu struct sk_buff *frag; 20401da177e4SLinus Torvalds int offset = skb_headlen(skb); 20411da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 20421da177e4SLinus Torvalds int i; 204327b437c8SHerbert Xu int err; 204427b437c8SHerbert Xu 204527b437c8SHerbert Xu if (skb_cloned(skb) && 204627b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 204727b437c8SHerbert Xu return err; 20481da177e4SLinus Torvalds 2049f4d26fb3SHerbert Xu i = 0; 2050f4d26fb3SHerbert Xu if (offset >= len) 2051f4d26fb3SHerbert Xu goto drop_pages; 2052f4d26fb3SHerbert Xu 2053f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 20549e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 205527b437c8SHerbert Xu 205627b437c8SHerbert Xu if (end < len) { 20571da177e4SLinus Torvalds offset = end; 205827b437c8SHerbert Xu continue; 20591da177e4SLinus Torvalds } 20601da177e4SLinus Torvalds 20619e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 206227b437c8SHerbert Xu 2063f4d26fb3SHerbert Xu drop_pages: 206427b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 206527b437c8SHerbert Xu 206627b437c8SHerbert Xu for (; i < nfrags; i++) 2067ea2ab693SIan Campbell skb_frag_unref(skb, i); 206827b437c8SHerbert Xu 206921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 207027b437c8SHerbert Xu skb_drop_fraglist(skb); 2071f4d26fb3SHerbert Xu goto done; 207227b437c8SHerbert Xu } 207327b437c8SHerbert Xu 207427b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 207527b437c8SHerbert Xu fragp = &frag->next) { 207627b437c8SHerbert Xu int end = offset + frag->len; 207727b437c8SHerbert Xu 207827b437c8SHerbert Xu if (skb_shared(frag)) { 207927b437c8SHerbert Xu struct sk_buff *nfrag; 208027b437c8SHerbert Xu 208127b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 208227b437c8SHerbert Xu if (unlikely(!nfrag)) 208327b437c8SHerbert Xu return -ENOMEM; 208427b437c8SHerbert Xu 208527b437c8SHerbert Xu nfrag->next = frag->next; 208685bb2a60SEric Dumazet consume_skb(frag); 208727b437c8SHerbert Xu frag = nfrag; 208827b437c8SHerbert Xu *fragp = frag; 208927b437c8SHerbert Xu } 209027b437c8SHerbert Xu 209127b437c8SHerbert Xu if (end < len) { 209227b437c8SHerbert Xu offset = end; 209327b437c8SHerbert Xu continue; 209427b437c8SHerbert Xu } 209527b437c8SHerbert Xu 209627b437c8SHerbert Xu if (end > len && 209727b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 209827b437c8SHerbert Xu return err; 209927b437c8SHerbert Xu 210027b437c8SHerbert Xu if (frag->next) 210127b437c8SHerbert Xu skb_drop_list(&frag->next); 210227b437c8SHerbert Xu break; 210327b437c8SHerbert Xu } 210427b437c8SHerbert Xu 2105f4d26fb3SHerbert Xu done: 210627b437c8SHerbert Xu if (len > skb_headlen(skb)) { 21071da177e4SLinus Torvalds skb->data_len -= skb->len - len; 21081da177e4SLinus Torvalds skb->len = len; 21091da177e4SLinus Torvalds } else { 21101da177e4SLinus Torvalds skb->len = len; 21111da177e4SLinus Torvalds skb->data_len = 0; 211227a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 21131da177e4SLinus Torvalds } 21141da177e4SLinus Torvalds 2115c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2116c21b48ccSEric Dumazet skb_condense(skb); 21171da177e4SLinus Torvalds return 0; 21181da177e4SLinus Torvalds } 2119b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 21201da177e4SLinus Torvalds 212188078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 212288078d98SEric Dumazet */ 212388078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 212488078d98SEric Dumazet { 212588078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 212688078d98SEric Dumazet int delta = skb->len - len; 212788078d98SEric Dumazet 2128d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2129d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2130d55bef50SDimitris Michailidis len); 213154970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 213254970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 213354970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 213454970a2fSVasily Averin 213554970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 213654970a2fSVasily Averin return -EINVAL; 213788078d98SEric Dumazet } 213888078d98SEric Dumazet return __pskb_trim(skb, len); 213988078d98SEric Dumazet } 214088078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 214188078d98SEric Dumazet 21421da177e4SLinus Torvalds /** 21431da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 21441da177e4SLinus Torvalds * @skb: buffer to reallocate 21451da177e4SLinus Torvalds * @delta: number of bytes to advance tail 21461da177e4SLinus Torvalds * 21471da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 21481da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 21491da177e4SLinus Torvalds * data from fragmented part. 21501da177e4SLinus Torvalds * 21511da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 21521da177e4SLinus Torvalds * 21531da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 21541da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 21551da177e4SLinus Torvalds * 21561da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 21571da177e4SLinus Torvalds * reloaded after call to this function. 21581da177e4SLinus Torvalds */ 21591da177e4SLinus Torvalds 21601da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 21611da177e4SLinus Torvalds * when it is necessary. 21621da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 21631da177e4SLinus Torvalds * 2. It may change skb pointers. 21641da177e4SLinus Torvalds * 21651da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 21661da177e4SLinus Torvalds */ 2167af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 21681da177e4SLinus Torvalds { 21691da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 21701da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 21711da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 21721da177e4SLinus Torvalds */ 21734305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 21741da177e4SLinus Torvalds 21751da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 21761da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 21771da177e4SLinus Torvalds GFP_ATOMIC)) 21781da177e4SLinus Torvalds return NULL; 21791da177e4SLinus Torvalds } 21801da177e4SLinus Torvalds 21819f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 21829f77fad3STim Hansen skb_tail_pointer(skb), delta)); 21831da177e4SLinus Torvalds 21841da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 21851da177e4SLinus Torvalds * size of pulled pages. Superb. 21861da177e4SLinus Torvalds */ 218721dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 21881da177e4SLinus Torvalds goto pull_pages; 21891da177e4SLinus Torvalds 21901da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 21911da177e4SLinus Torvalds eat = delta; 21921da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 21939e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 21949e903e08SEric Dumazet 21959e903e08SEric Dumazet if (size >= eat) 21961da177e4SLinus Torvalds goto pull_pages; 21979e903e08SEric Dumazet eat -= size; 21981da177e4SLinus Torvalds } 21991da177e4SLinus Torvalds 22001da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 220109001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 22021da177e4SLinus Torvalds * but taking into account that pulling is expected to 22031da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 22041da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 22051da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 22061da177e4SLinus Torvalds */ 22071da177e4SLinus Torvalds if (eat) { 22081da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 22091da177e4SLinus Torvalds struct sk_buff *clone = NULL; 22101da177e4SLinus Torvalds struct sk_buff *insp = NULL; 22111da177e4SLinus Torvalds 22121da177e4SLinus Torvalds do { 22131da177e4SLinus Torvalds if (list->len <= eat) { 22141da177e4SLinus Torvalds /* Eaten as whole. */ 22151da177e4SLinus Torvalds eat -= list->len; 22161da177e4SLinus Torvalds list = list->next; 22171da177e4SLinus Torvalds insp = list; 22181da177e4SLinus Torvalds } else { 22191da177e4SLinus Torvalds /* Eaten partially. */ 22201da177e4SLinus Torvalds 22211da177e4SLinus Torvalds if (skb_shared(list)) { 22221da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 22231da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 22241da177e4SLinus Torvalds if (!clone) 22251da177e4SLinus Torvalds return NULL; 22261da177e4SLinus Torvalds insp = list->next; 22271da177e4SLinus Torvalds list = clone; 22281da177e4SLinus Torvalds } else { 22291da177e4SLinus Torvalds /* This may be pulled without 22301da177e4SLinus Torvalds * problems. */ 22311da177e4SLinus Torvalds insp = list; 22321da177e4SLinus Torvalds } 22331da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 22341da177e4SLinus Torvalds kfree_skb(clone); 22351da177e4SLinus Torvalds return NULL; 22361da177e4SLinus Torvalds } 22371da177e4SLinus Torvalds break; 22381da177e4SLinus Torvalds } 22391da177e4SLinus Torvalds } while (eat); 22401da177e4SLinus Torvalds 22411da177e4SLinus Torvalds /* Free pulled out fragments. */ 22421da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 22431da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 22441da177e4SLinus Torvalds kfree_skb(list); 22451da177e4SLinus Torvalds } 22461da177e4SLinus Torvalds /* And insert new clone at head. */ 22471da177e4SLinus Torvalds if (clone) { 22481da177e4SLinus Torvalds clone->next = list; 22491da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 22501da177e4SLinus Torvalds } 22511da177e4SLinus Torvalds } 22521da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 22531da177e4SLinus Torvalds 22541da177e4SLinus Torvalds pull_pages: 22551da177e4SLinus Torvalds eat = delta; 22561da177e4SLinus Torvalds k = 0; 22571da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22589e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22599e903e08SEric Dumazet 22609e903e08SEric Dumazet if (size <= eat) { 2261ea2ab693SIan Campbell skb_frag_unref(skb, i); 22629e903e08SEric Dumazet eat -= size; 22631da177e4SLinus Torvalds } else { 2264b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2265b54c9d5bSJonathan Lemon 2266b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 22671da177e4SLinus Torvalds if (eat) { 2268b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2269b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 22703ccc6c6fSlinzhang if (!i) 22713ccc6c6fSlinzhang goto end; 22721da177e4SLinus Torvalds eat = 0; 22731da177e4SLinus Torvalds } 22741da177e4SLinus Torvalds k++; 22751da177e4SLinus Torvalds } 22761da177e4SLinus Torvalds } 22771da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 22781da177e4SLinus Torvalds 22793ccc6c6fSlinzhang end: 22801da177e4SLinus Torvalds skb->tail += delta; 22811da177e4SLinus Torvalds skb->data_len -= delta; 22821da177e4SLinus Torvalds 22831f8b977aSWillem de Bruijn if (!skb->data_len) 22841f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 22851f8b977aSWillem de Bruijn 228627a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 22871da177e4SLinus Torvalds } 2288b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 22891da177e4SLinus Torvalds 229022019b17SEric Dumazet /** 229122019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 229222019b17SEric Dumazet * @skb: source skb 229322019b17SEric Dumazet * @offset: offset in source 229422019b17SEric Dumazet * @to: destination buffer 229522019b17SEric Dumazet * @len: number of bytes to copy 229622019b17SEric Dumazet * 229722019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 229822019b17SEric Dumazet * destination buffer. 229922019b17SEric Dumazet * 230022019b17SEric Dumazet * CAUTION ! : 230122019b17SEric Dumazet * If its prototype is ever changed, 230222019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 230322019b17SEric Dumazet * since it is called from BPF assembly code. 230422019b17SEric Dumazet */ 23051da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 23061da177e4SLinus Torvalds { 23071a028e50SDavid S. Miller int start = skb_headlen(skb); 2308fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2309fbb398a8SDavid S. Miller int i, copy; 23101da177e4SLinus Torvalds 23111da177e4SLinus Torvalds if (offset > (int)skb->len - len) 23121da177e4SLinus Torvalds goto fault; 23131da177e4SLinus Torvalds 23141da177e4SLinus Torvalds /* Copy header. */ 23151a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 23161da177e4SLinus Torvalds if (copy > len) 23171da177e4SLinus Torvalds copy = len; 2318d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 23191da177e4SLinus Torvalds if ((len -= copy) == 0) 23201da177e4SLinus Torvalds return 0; 23211da177e4SLinus Torvalds offset += copy; 23221da177e4SLinus Torvalds to += copy; 23231da177e4SLinus Torvalds } 23241da177e4SLinus Torvalds 23251da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 23261a028e50SDavid S. Miller int end; 232751c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 23281da177e4SLinus Torvalds 2329547b792cSIlpo Järvinen WARN_ON(start > offset + len); 23301a028e50SDavid S. Miller 233151c56b00SEric Dumazet end = start + skb_frag_size(f); 23321da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2333c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2334c613c209SWillem de Bruijn struct page *p; 23351da177e4SLinus Torvalds u8 *vaddr; 23361da177e4SLinus Torvalds 23371da177e4SLinus Torvalds if (copy > len) 23381da177e4SLinus Torvalds copy = len; 23391da177e4SLinus Torvalds 2340c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2341b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2342c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2343c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2344c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 234551c56b00SEric Dumazet kunmap_atomic(vaddr); 2346c613c209SWillem de Bruijn } 23471da177e4SLinus Torvalds 23481da177e4SLinus Torvalds if ((len -= copy) == 0) 23491da177e4SLinus Torvalds return 0; 23501da177e4SLinus Torvalds offset += copy; 23511da177e4SLinus Torvalds to += copy; 23521da177e4SLinus Torvalds } 23531a028e50SDavid S. Miller start = end; 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds 2356fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 23571a028e50SDavid S. Miller int end; 23581da177e4SLinus Torvalds 2359547b792cSIlpo Järvinen WARN_ON(start > offset + len); 23601a028e50SDavid S. Miller 2361fbb398a8SDavid S. Miller end = start + frag_iter->len; 23621da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 23631da177e4SLinus Torvalds if (copy > len) 23641da177e4SLinus Torvalds copy = len; 2365fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 23661da177e4SLinus Torvalds goto fault; 23671da177e4SLinus Torvalds if ((len -= copy) == 0) 23681da177e4SLinus Torvalds return 0; 23691da177e4SLinus Torvalds offset += copy; 23701da177e4SLinus Torvalds to += copy; 23711da177e4SLinus Torvalds } 23721a028e50SDavid S. Miller start = end; 23731da177e4SLinus Torvalds } 2374a6686f2fSShirley Ma 23751da177e4SLinus Torvalds if (!len) 23761da177e4SLinus Torvalds return 0; 23771da177e4SLinus Torvalds 23781da177e4SLinus Torvalds fault: 23791da177e4SLinus Torvalds return -EFAULT; 23801da177e4SLinus Torvalds } 2381b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 23821da177e4SLinus Torvalds 23839c55e01cSJens Axboe /* 23849c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 23859c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 23869c55e01cSJens Axboe */ 23879c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 23889c55e01cSJens Axboe { 23898b9d3728SJarek Poplawski put_page(spd->pages[i]); 23908b9d3728SJarek Poplawski } 23919c55e01cSJens Axboe 2392a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 23934fb66994SJarek Poplawski unsigned int *offset, 239418aafc62SEric Dumazet struct sock *sk) 23958b9d3728SJarek Poplawski { 23965640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 23978b9d3728SJarek Poplawski 23985640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 23998b9d3728SJarek Poplawski return NULL; 24004fb66994SJarek Poplawski 24015640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 24024fb66994SJarek Poplawski 24035640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 24045640f768SEric Dumazet page_address(page) + *offset, *len); 24055640f768SEric Dumazet *offset = pfrag->offset; 24065640f768SEric Dumazet pfrag->offset += *len; 24074fb66994SJarek Poplawski 24085640f768SEric Dumazet return pfrag->page; 24099c55e01cSJens Axboe } 24109c55e01cSJens Axboe 241141c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 241241c73a0dSEric Dumazet struct page *page, 241341c73a0dSEric Dumazet unsigned int offset) 241441c73a0dSEric Dumazet { 241541c73a0dSEric Dumazet return spd->nr_pages && 241641c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 241741c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 241841c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 241941c73a0dSEric Dumazet } 242041c73a0dSEric Dumazet 24219c55e01cSJens Axboe /* 24229c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 24239c55e01cSJens Axboe */ 2424a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 242535f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 24264fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 242718aafc62SEric Dumazet bool linear, 24287a67e56fSJarek Poplawski struct sock *sk) 24299c55e01cSJens Axboe { 243041c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2431a108d5f3SDavid S. Miller return true; 24329c55e01cSJens Axboe 24338b9d3728SJarek Poplawski if (linear) { 243418aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 24358b9d3728SJarek Poplawski if (!page) 2436a108d5f3SDavid S. Miller return true; 243741c73a0dSEric Dumazet } 243841c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 243941c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2440a108d5f3SDavid S. Miller return false; 244141c73a0dSEric Dumazet } 24428b9d3728SJarek Poplawski get_page(page); 24439c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 24444fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 24459c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 24469c55e01cSJens Axboe spd->nr_pages++; 24478b9d3728SJarek Poplawski 2448a108d5f3SDavid S. Miller return false; 24499c55e01cSJens Axboe } 24509c55e01cSJens Axboe 2451a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 24522870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 245318aafc62SEric Dumazet unsigned int *len, 2454d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 245535f3d14dSJens Axboe struct sock *sk, 245635f3d14dSJens Axboe struct pipe_inode_info *pipe) 24579c55e01cSJens Axboe { 24582870c43dSOctavian Purdila if (!*len) 2459a108d5f3SDavid S. Miller return true; 24609c55e01cSJens Axboe 24612870c43dSOctavian Purdila /* skip this segment if already processed */ 24622870c43dSOctavian Purdila if (*off >= plen) { 24632870c43dSOctavian Purdila *off -= plen; 2464a108d5f3SDavid S. Miller return false; 24652870c43dSOctavian Purdila } 24662870c43dSOctavian Purdila 24672870c43dSOctavian Purdila /* ignore any bits we already processed */ 24689ca1b22dSEric Dumazet poff += *off; 24699ca1b22dSEric Dumazet plen -= *off; 24702870c43dSOctavian Purdila *off = 0; 24712870c43dSOctavian Purdila 247218aafc62SEric Dumazet do { 247318aafc62SEric Dumazet unsigned int flen = min(*len, plen); 24742870c43dSOctavian Purdila 247518aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 247618aafc62SEric Dumazet linear, sk)) 2477a108d5f3SDavid S. Miller return true; 247818aafc62SEric Dumazet poff += flen; 247918aafc62SEric Dumazet plen -= flen; 24802870c43dSOctavian Purdila *len -= flen; 248118aafc62SEric Dumazet } while (*len && plen); 24822870c43dSOctavian Purdila 2483a108d5f3SDavid S. Miller return false; 2484db43a282SOctavian Purdila } 24859c55e01cSJens Axboe 24869c55e01cSJens Axboe /* 2487a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 24882870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 24899c55e01cSJens Axboe */ 2490a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 249135f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 249235f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 24932870c43dSOctavian Purdila { 24942870c43dSOctavian Purdila int seg; 2495fa9835e5STom Herbert struct sk_buff *iter; 24969c55e01cSJens Axboe 24971d0c0b32SEric Dumazet /* map the linear part : 24982996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 24992996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 25002996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 25019c55e01cSJens Axboe */ 25022870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 25032870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 25042870c43dSOctavian Purdila skb_headlen(skb), 250518aafc62SEric Dumazet offset, len, spd, 25063a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 25071d0c0b32SEric Dumazet sk, pipe)) 2508a108d5f3SDavid S. Miller return true; 25099c55e01cSJens Axboe 25109c55e01cSJens Axboe /* 25119c55e01cSJens Axboe * then map the fragments 25129c55e01cSJens Axboe */ 25139c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 25149c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 25159c55e01cSJens Axboe 2516ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2517b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 251818aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2519a108d5f3SDavid S. Miller return true; 25209c55e01cSJens Axboe } 25219c55e01cSJens Axboe 2522fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2523fa9835e5STom Herbert if (*offset >= iter->len) { 2524fa9835e5STom Herbert *offset -= iter->len; 2525fa9835e5STom Herbert continue; 2526fa9835e5STom Herbert } 2527fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2528fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2529fa9835e5STom Herbert * case. 2530fa9835e5STom Herbert */ 2531fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2532fa9835e5STom Herbert return true; 2533fa9835e5STom Herbert } 2534fa9835e5STom Herbert 2535a108d5f3SDavid S. Miller return false; 25369c55e01cSJens Axboe } 25379c55e01cSJens Axboe 25389c55e01cSJens Axboe /* 25399c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2540fa9835e5STom Herbert * the fragments, and the frag list. 25419c55e01cSJens Axboe */ 2542a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 25439c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 254425869262SAl Viro unsigned int flags) 25459c55e01cSJens Axboe { 254641c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 254741c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 25489c55e01cSJens Axboe struct splice_pipe_desc spd = { 25499c55e01cSJens Axboe .pages = pages, 25509c55e01cSJens Axboe .partial = partial, 2551047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 255228a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 25539c55e01cSJens Axboe .spd_release = sock_spd_release, 25549c55e01cSJens Axboe }; 255535f3d14dSJens Axboe int ret = 0; 255635f3d14dSJens Axboe 2557fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 25589c55e01cSJens Axboe 2559a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 256025869262SAl Viro ret = splice_to_pipe(pipe, &spd); 25619c55e01cSJens Axboe 256235f3d14dSJens Axboe return ret; 25639c55e01cSJens Axboe } 25642b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 25659c55e01cSJens Axboe 25660739cd28SCong Wang static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg, 25670739cd28SCong Wang struct kvec *vec, size_t num, size_t size) 25680739cd28SCong Wang { 25690739cd28SCong Wang struct socket *sock = sk->sk_socket; 25700739cd28SCong Wang 25710739cd28SCong Wang if (!sock) 25720739cd28SCong Wang return -EINVAL; 25730739cd28SCong Wang return kernel_sendmsg(sock, msg, vec, num, size); 25740739cd28SCong Wang } 25750739cd28SCong Wang 25760739cd28SCong Wang static int sendpage_unlocked(struct sock *sk, struct page *page, int offset, 25770739cd28SCong Wang size_t size, int flags) 25780739cd28SCong Wang { 25790739cd28SCong Wang struct socket *sock = sk->sk_socket; 25800739cd28SCong Wang 25810739cd28SCong Wang if (!sock) 25820739cd28SCong Wang return -EINVAL; 25830739cd28SCong Wang return kernel_sendpage(sock, page, offset, size, flags); 25840739cd28SCong Wang } 25850739cd28SCong Wang 25860739cd28SCong Wang typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg, 25870739cd28SCong Wang struct kvec *vec, size_t num, size_t size); 25880739cd28SCong Wang typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset, 25890739cd28SCong Wang size_t size, int flags); 25900739cd28SCong Wang static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, 25910739cd28SCong Wang int len, sendmsg_func sendmsg, sendpage_func sendpage) 259220bf50deSTom Herbert { 259320bf50deSTom Herbert unsigned int orig_len = len; 259420bf50deSTom Herbert struct sk_buff *head = skb; 259520bf50deSTom Herbert unsigned short fragidx; 259620bf50deSTom Herbert int slen, ret; 259720bf50deSTom Herbert 259820bf50deSTom Herbert do_frag_list: 259920bf50deSTom Herbert 260020bf50deSTom Herbert /* Deal with head data */ 260120bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 260220bf50deSTom Herbert struct kvec kv; 260320bf50deSTom Herbert struct msghdr msg; 260420bf50deSTom Herbert 260520bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 260620bf50deSTom Herbert kv.iov_base = skb->data + offset; 2607db5980d8SJohn Fastabend kv.iov_len = slen; 260820bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2609bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 261020bf50deSTom Herbert 26110739cd28SCong Wang ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked, 26120739cd28SCong Wang sendmsg_unlocked, sk, &msg, &kv, 1, slen); 261320bf50deSTom Herbert if (ret <= 0) 261420bf50deSTom Herbert goto error; 261520bf50deSTom Herbert 261620bf50deSTom Herbert offset += ret; 261720bf50deSTom Herbert len -= ret; 261820bf50deSTom Herbert } 261920bf50deSTom Herbert 262020bf50deSTom Herbert /* All the data was skb head? */ 262120bf50deSTom Herbert if (!len) 262220bf50deSTom Herbert goto out; 262320bf50deSTom Herbert 262420bf50deSTom Herbert /* Make offset relative to start of frags */ 262520bf50deSTom Herbert offset -= skb_headlen(skb); 262620bf50deSTom Herbert 262720bf50deSTom Herbert /* Find where we are in frag list */ 262820bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 262920bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 263020bf50deSTom Herbert 2631d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 263220bf50deSTom Herbert break; 263320bf50deSTom Herbert 2634d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 263520bf50deSTom Herbert } 263620bf50deSTom Herbert 263720bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 263820bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 263920bf50deSTom Herbert 2640d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 264120bf50deSTom Herbert 264220bf50deSTom Herbert while (slen) { 26430739cd28SCong Wang ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked, 26440739cd28SCong Wang sendpage_unlocked, sk, 26450739cd28SCong Wang skb_frag_page(frag), 2646b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 264720bf50deSTom Herbert slen, MSG_DONTWAIT); 264820bf50deSTom Herbert if (ret <= 0) 264920bf50deSTom Herbert goto error; 265020bf50deSTom Herbert 265120bf50deSTom Herbert len -= ret; 265220bf50deSTom Herbert offset += ret; 265320bf50deSTom Herbert slen -= ret; 265420bf50deSTom Herbert } 265520bf50deSTom Herbert 265620bf50deSTom Herbert offset = 0; 265720bf50deSTom Herbert } 265820bf50deSTom Herbert 265920bf50deSTom Herbert if (len) { 266020bf50deSTom Herbert /* Process any frag lists */ 266120bf50deSTom Herbert 266220bf50deSTom Herbert if (skb == head) { 266320bf50deSTom Herbert if (skb_has_frag_list(skb)) { 266420bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 266520bf50deSTom Herbert goto do_frag_list; 266620bf50deSTom Herbert } 266720bf50deSTom Herbert } else if (skb->next) { 266820bf50deSTom Herbert skb = skb->next; 266920bf50deSTom Herbert goto do_frag_list; 267020bf50deSTom Herbert } 267120bf50deSTom Herbert } 267220bf50deSTom Herbert 267320bf50deSTom Herbert out: 267420bf50deSTom Herbert return orig_len - len; 267520bf50deSTom Herbert 267620bf50deSTom Herbert error: 267720bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 267820bf50deSTom Herbert } 26790739cd28SCong Wang 26800739cd28SCong Wang /* Send skb data on a socket. Socket must be locked. */ 26810739cd28SCong Wang int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 26820739cd28SCong Wang int len) 26830739cd28SCong Wang { 26840739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked, 26850739cd28SCong Wang kernel_sendpage_locked); 26860739cd28SCong Wang } 268720bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 268820bf50deSTom Herbert 26890739cd28SCong Wang /* Send skb data on a socket. Socket must be unlocked. */ 26900739cd28SCong Wang int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 26910739cd28SCong Wang { 26920739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked, 26930739cd28SCong Wang sendpage_unlocked); 26940739cd28SCong Wang } 26950739cd28SCong Wang 2696357b40a1SHerbert Xu /** 2697357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2698357b40a1SHerbert Xu * @skb: destination buffer 2699357b40a1SHerbert Xu * @offset: offset in destination 2700357b40a1SHerbert Xu * @from: source buffer 2701357b40a1SHerbert Xu * @len: number of bytes to copy 2702357b40a1SHerbert Xu * 2703357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2704357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2705357b40a1SHerbert Xu * traversing fragment lists and such. 2706357b40a1SHerbert Xu */ 2707357b40a1SHerbert Xu 27080c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2709357b40a1SHerbert Xu { 27101a028e50SDavid S. Miller int start = skb_headlen(skb); 2711fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2712fbb398a8SDavid S. Miller int i, copy; 2713357b40a1SHerbert Xu 2714357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2715357b40a1SHerbert Xu goto fault; 2716357b40a1SHerbert Xu 27171a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2718357b40a1SHerbert Xu if (copy > len) 2719357b40a1SHerbert Xu copy = len; 272027d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2721357b40a1SHerbert Xu if ((len -= copy) == 0) 2722357b40a1SHerbert Xu return 0; 2723357b40a1SHerbert Xu offset += copy; 2724357b40a1SHerbert Xu from += copy; 2725357b40a1SHerbert Xu } 2726357b40a1SHerbert Xu 2727357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2728357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 27291a028e50SDavid S. Miller int end; 2730357b40a1SHerbert Xu 2731547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27321a028e50SDavid S. Miller 27339e903e08SEric Dumazet end = start + skb_frag_size(frag); 2734357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2735c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2736c613c209SWillem de Bruijn struct page *p; 2737357b40a1SHerbert Xu u8 *vaddr; 2738357b40a1SHerbert Xu 2739357b40a1SHerbert Xu if (copy > len) 2740357b40a1SHerbert Xu copy = len; 2741357b40a1SHerbert Xu 2742c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2743b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2744c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2745c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2746c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 274751c56b00SEric Dumazet kunmap_atomic(vaddr); 2748c613c209SWillem de Bruijn } 2749357b40a1SHerbert Xu 2750357b40a1SHerbert Xu if ((len -= copy) == 0) 2751357b40a1SHerbert Xu return 0; 2752357b40a1SHerbert Xu offset += copy; 2753357b40a1SHerbert Xu from += copy; 2754357b40a1SHerbert Xu } 27551a028e50SDavid S. Miller start = end; 2756357b40a1SHerbert Xu } 2757357b40a1SHerbert Xu 2758fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 27591a028e50SDavid S. Miller int end; 2760357b40a1SHerbert Xu 2761547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27621a028e50SDavid S. Miller 2763fbb398a8SDavid S. Miller end = start + frag_iter->len; 2764357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2765357b40a1SHerbert Xu if (copy > len) 2766357b40a1SHerbert Xu copy = len; 2767fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 27681a028e50SDavid S. Miller from, copy)) 2769357b40a1SHerbert Xu goto fault; 2770357b40a1SHerbert Xu if ((len -= copy) == 0) 2771357b40a1SHerbert Xu return 0; 2772357b40a1SHerbert Xu offset += copy; 2773357b40a1SHerbert Xu from += copy; 2774357b40a1SHerbert Xu } 27751a028e50SDavid S. Miller start = end; 2776357b40a1SHerbert Xu } 2777357b40a1SHerbert Xu if (!len) 2778357b40a1SHerbert Xu return 0; 2779357b40a1SHerbert Xu 2780357b40a1SHerbert Xu fault: 2781357b40a1SHerbert Xu return -EFAULT; 2782357b40a1SHerbert Xu } 2783357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2784357b40a1SHerbert Xu 27851da177e4SLinus Torvalds /* Checksum skb data. */ 27862817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 27872817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 27881da177e4SLinus Torvalds { 27891a028e50SDavid S. Miller int start = skb_headlen(skb); 27901a028e50SDavid S. Miller int i, copy = start - offset; 2791fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 27921da177e4SLinus Torvalds int pos = 0; 27931da177e4SLinus Torvalds 27941da177e4SLinus Torvalds /* Checksum header. */ 27951da177e4SLinus Torvalds if (copy > 0) { 27961da177e4SLinus Torvalds if (copy > len) 27971da177e4SLinus Torvalds copy = len; 27982544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 27992544af03SMatteo Croce skb->data + offset, copy, csum); 28001da177e4SLinus Torvalds if ((len -= copy) == 0) 28011da177e4SLinus Torvalds return csum; 28021da177e4SLinus Torvalds offset += copy; 28031da177e4SLinus Torvalds pos = copy; 28041da177e4SLinus Torvalds } 28051da177e4SLinus Torvalds 28061da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 28071a028e50SDavid S. Miller int end; 280851c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 28091da177e4SLinus Torvalds 2810547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28111a028e50SDavid S. Miller 281251c56b00SEric Dumazet end = start + skb_frag_size(frag); 28131da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2814c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2815c613c209SWillem de Bruijn struct page *p; 281644bb9363SAl Viro __wsum csum2; 28171da177e4SLinus Torvalds u8 *vaddr; 28181da177e4SLinus Torvalds 28191da177e4SLinus Torvalds if (copy > len) 28201da177e4SLinus Torvalds copy = len; 2821c613c209SWillem de Bruijn 2822c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2823b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2824c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2825c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 28262544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 28272544af03SMatteo Croce csum_partial_ext, 28282544af03SMatteo Croce vaddr + p_off, p_len, 0); 282951c56b00SEric Dumazet kunmap_atomic(vaddr); 28302544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 28312544af03SMatteo Croce csum_block_add_ext, csum, 28322544af03SMatteo Croce csum2, pos, p_len); 2833c613c209SWillem de Bruijn pos += p_len; 2834c613c209SWillem de Bruijn } 2835c613c209SWillem de Bruijn 28361da177e4SLinus Torvalds if (!(len -= copy)) 28371da177e4SLinus Torvalds return csum; 28381da177e4SLinus Torvalds offset += copy; 28391da177e4SLinus Torvalds } 28401a028e50SDavid S. Miller start = end; 28411da177e4SLinus Torvalds } 28421da177e4SLinus Torvalds 2843fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 28441a028e50SDavid S. Miller int end; 28451da177e4SLinus Torvalds 2846547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28471a028e50SDavid S. Miller 2848fbb398a8SDavid S. Miller end = start + frag_iter->len; 28491da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 28505f92a738SAl Viro __wsum csum2; 28511da177e4SLinus Torvalds if (copy > len) 28521da177e4SLinus Torvalds copy = len; 28532817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 28542817a336SDaniel Borkmann copy, 0, ops); 28552544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 28562544af03SMatteo Croce csum, csum2, pos, copy); 28571da177e4SLinus Torvalds if ((len -= copy) == 0) 28581da177e4SLinus Torvalds return csum; 28591da177e4SLinus Torvalds offset += copy; 28601da177e4SLinus Torvalds pos += copy; 28611da177e4SLinus Torvalds } 28621a028e50SDavid S. Miller start = end; 28631da177e4SLinus Torvalds } 286409a62660SKris Katterjohn BUG_ON(len); 28651da177e4SLinus Torvalds 28661da177e4SLinus Torvalds return csum; 28671da177e4SLinus Torvalds } 28682817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 28692817a336SDaniel Borkmann 28702817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 28712817a336SDaniel Borkmann int len, __wsum csum) 28722817a336SDaniel Borkmann { 28732817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2874cea80ea8SDaniel Borkmann .update = csum_partial_ext, 28752817a336SDaniel Borkmann .combine = csum_block_add_ext, 28762817a336SDaniel Borkmann }; 28772817a336SDaniel Borkmann 28782817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 28792817a336SDaniel Borkmann } 2880b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 28811da177e4SLinus Torvalds 28821da177e4SLinus Torvalds /* Both of above in one bottle. */ 28831da177e4SLinus Torvalds 288481d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 28858d5930dfSAl Viro u8 *to, int len) 28861da177e4SLinus Torvalds { 28871a028e50SDavid S. Miller int start = skb_headlen(skb); 28881a028e50SDavid S. Miller int i, copy = start - offset; 2889fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 28901da177e4SLinus Torvalds int pos = 0; 28918d5930dfSAl Viro __wsum csum = 0; 28921da177e4SLinus Torvalds 28931da177e4SLinus Torvalds /* Copy header. */ 28941da177e4SLinus Torvalds if (copy > 0) { 28951da177e4SLinus Torvalds if (copy > len) 28961da177e4SLinus Torvalds copy = len; 28971da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2898cc44c17bSAl Viro copy); 28991da177e4SLinus Torvalds if ((len -= copy) == 0) 29001da177e4SLinus Torvalds return csum; 29011da177e4SLinus Torvalds offset += copy; 29021da177e4SLinus Torvalds to += copy; 29031da177e4SLinus Torvalds pos = copy; 29041da177e4SLinus Torvalds } 29051da177e4SLinus Torvalds 29061da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 29071a028e50SDavid S. Miller int end; 29081da177e4SLinus Torvalds 2909547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29101a028e50SDavid S. Miller 29119e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 29121da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2913c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2914c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2915c613c209SWillem de Bruijn struct page *p; 29165084205fSAl Viro __wsum csum2; 29171da177e4SLinus Torvalds u8 *vaddr; 29181da177e4SLinus Torvalds 29191da177e4SLinus Torvalds if (copy > len) 29201da177e4SLinus Torvalds copy = len; 2921c613c209SWillem de Bruijn 2922c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2923b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2924c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2925c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2926c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2927c613c209SWillem de Bruijn to + copied, 2928cc44c17bSAl Viro p_len); 292951c56b00SEric Dumazet kunmap_atomic(vaddr); 29301da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2931c613c209SWillem de Bruijn pos += p_len; 2932c613c209SWillem de Bruijn } 2933c613c209SWillem de Bruijn 29341da177e4SLinus Torvalds if (!(len -= copy)) 29351da177e4SLinus Torvalds return csum; 29361da177e4SLinus Torvalds offset += copy; 29371da177e4SLinus Torvalds to += copy; 29381da177e4SLinus Torvalds } 29391a028e50SDavid S. Miller start = end; 29401da177e4SLinus Torvalds } 29411da177e4SLinus Torvalds 2942fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 294381d77662SAl Viro __wsum csum2; 29441a028e50SDavid S. Miller int end; 29451da177e4SLinus Torvalds 2946547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29471a028e50SDavid S. Miller 2948fbb398a8SDavid S. Miller end = start + frag_iter->len; 29491da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 29501da177e4SLinus Torvalds if (copy > len) 29511da177e4SLinus Torvalds copy = len; 2952fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 29531a028e50SDavid S. Miller offset - start, 29548d5930dfSAl Viro to, copy); 29551da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 29561da177e4SLinus Torvalds if ((len -= copy) == 0) 29571da177e4SLinus Torvalds return csum; 29581da177e4SLinus Torvalds offset += copy; 29591da177e4SLinus Torvalds to += copy; 29601da177e4SLinus Torvalds pos += copy; 29611da177e4SLinus Torvalds } 29621a028e50SDavid S. Miller start = end; 29631da177e4SLinus Torvalds } 296409a62660SKris Katterjohn BUG_ON(len); 29651da177e4SLinus Torvalds return csum; 29661da177e4SLinus Torvalds } 2967b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 29681da177e4SLinus Torvalds 296949f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 297049f8e832SCong Wang { 297149f8e832SCong Wang __sum16 sum; 297249f8e832SCong Wang 297349f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 297414641931SCong Wang /* See comments in __skb_checksum_complete(). */ 297549f8e832SCong Wang if (likely(!sum)) { 297649f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 297749f8e832SCong Wang !skb->csum_complete_sw) 29787fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 297949f8e832SCong Wang } 298049f8e832SCong Wang if (!skb_shared(skb)) 298149f8e832SCong Wang skb->csum_valid = !sum; 298249f8e832SCong Wang return sum; 298349f8e832SCong Wang } 298449f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 298549f8e832SCong Wang 298614641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 298714641931SCong Wang * which has been changed from the hardware checksum, for example, by 298814641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 298914641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 299014641931SCong Wang * 299114641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 299214641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 299314641931SCong Wang * shared. 299414641931SCong Wang */ 299549f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 299649f8e832SCong Wang { 299749f8e832SCong Wang __wsum csum; 299849f8e832SCong Wang __sum16 sum; 299949f8e832SCong Wang 300049f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 300149f8e832SCong Wang 300249f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 300314641931SCong Wang /* This check is inverted, because we already knew the hardware 300414641931SCong Wang * checksum is invalid before calling this function. So, if the 300514641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 300614641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 300714641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 300814641931SCong Wang * when moving skb->data around. 300914641931SCong Wang */ 301049f8e832SCong Wang if (likely(!sum)) { 301149f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 301249f8e832SCong Wang !skb->csum_complete_sw) 30137fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 301449f8e832SCong Wang } 301549f8e832SCong Wang 301649f8e832SCong Wang if (!skb_shared(skb)) { 301749f8e832SCong Wang /* Save full packet checksum */ 301849f8e832SCong Wang skb->csum = csum; 301949f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 302049f8e832SCong Wang skb->csum_complete_sw = 1; 302149f8e832SCong Wang skb->csum_valid = !sum; 302249f8e832SCong Wang } 302349f8e832SCong Wang 302449f8e832SCong Wang return sum; 302549f8e832SCong Wang } 302649f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 302749f8e832SCong Wang 30289617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 30299617813dSDavide Caratti { 30309617813dSDavide Caratti net_warn_ratelimited( 30319617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 30329617813dSDavide Caratti __func__); 30339617813dSDavide Caratti return 0; 30349617813dSDavide Caratti } 30359617813dSDavide Caratti 30369617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 30379617813dSDavide Caratti int offset, int len) 30389617813dSDavide Caratti { 30399617813dSDavide Caratti net_warn_ratelimited( 30409617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 30419617813dSDavide Caratti __func__); 30429617813dSDavide Caratti return 0; 30439617813dSDavide Caratti } 30449617813dSDavide Caratti 30459617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 30469617813dSDavide Caratti .update = warn_crc32c_csum_update, 30479617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 30489617813dSDavide Caratti }; 30499617813dSDavide Caratti 30509617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 30519617813dSDavide Caratti &default_crc32c_ops; 30529617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 30539617813dSDavide Caratti 3054af2806f8SThomas Graf /** 3055af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 3056af2806f8SThomas Graf * @from: source buffer 3057af2806f8SThomas Graf * 3058af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 3059af2806f8SThomas Graf * into skb_zerocopy(). 3060af2806f8SThomas Graf */ 3061af2806f8SThomas Graf unsigned int 3062af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 3063af2806f8SThomas Graf { 3064af2806f8SThomas Graf unsigned int hlen = 0; 3065af2806f8SThomas Graf 3066af2806f8SThomas Graf if (!from->head_frag || 3067af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 3068a17ad096SPravin B Shelar skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) { 3069af2806f8SThomas Graf hlen = skb_headlen(from); 3070a17ad096SPravin B Shelar if (!hlen) 3071a17ad096SPravin B Shelar hlen = from->len; 3072a17ad096SPravin B Shelar } 3073af2806f8SThomas Graf 3074af2806f8SThomas Graf if (skb_has_frag_list(from)) 3075af2806f8SThomas Graf hlen = from->len; 3076af2806f8SThomas Graf 3077af2806f8SThomas Graf return hlen; 3078af2806f8SThomas Graf } 3079af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 3080af2806f8SThomas Graf 3081af2806f8SThomas Graf /** 3082af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 3083af2806f8SThomas Graf * @to: destination buffer 30847fceb4deSMasanari Iida * @from: source buffer 3085af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 3086af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 3087af2806f8SThomas Graf * 3088af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 3089af2806f8SThomas Graf * to the frags in the source buffer. 3090af2806f8SThomas Graf * 3091af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 3092af2806f8SThomas Graf * headroom in the `to` buffer. 309336d5fe6aSZoltan Kiss * 309436d5fe6aSZoltan Kiss * Return value: 309536d5fe6aSZoltan Kiss * 0: everything is OK 309636d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 309736d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 3098af2806f8SThomas Graf */ 309936d5fe6aSZoltan Kiss int 310036d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 3101af2806f8SThomas Graf { 3102af2806f8SThomas Graf int i, j = 0; 3103af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 310436d5fe6aSZoltan Kiss int ret; 3105af2806f8SThomas Graf struct page *page; 3106af2806f8SThomas Graf unsigned int offset; 3107af2806f8SThomas Graf 3108af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 3109af2806f8SThomas Graf 3110af2806f8SThomas Graf /* dont bother with small payloads */ 311136d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 311236d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 3113af2806f8SThomas Graf 3114af2806f8SThomas Graf if (hlen) { 311536d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 311636d5fe6aSZoltan Kiss if (unlikely(ret)) 311736d5fe6aSZoltan Kiss return ret; 3118af2806f8SThomas Graf len -= hlen; 3119af2806f8SThomas Graf } else { 3120af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 3121af2806f8SThomas Graf if (plen) { 3122af2806f8SThomas Graf page = virt_to_head_page(from->head); 3123af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 3124af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 3125af2806f8SThomas Graf get_page(page); 3126af2806f8SThomas Graf j = 1; 3127af2806f8SThomas Graf len -= plen; 3128af2806f8SThomas Graf } 3129af2806f8SThomas Graf } 3130af2806f8SThomas Graf 3131af2806f8SThomas Graf to->truesize += len + plen; 3132af2806f8SThomas Graf to->len += len + plen; 3133af2806f8SThomas Graf to->data_len += len + plen; 3134af2806f8SThomas Graf 313536d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 313636d5fe6aSZoltan Kiss skb_tx_error(from); 313736d5fe6aSZoltan Kiss return -ENOMEM; 313836d5fe6aSZoltan Kiss } 31391f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 314036d5fe6aSZoltan Kiss 3141af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 3142d8e18a51SMatthew Wilcox (Oracle) int size; 3143d8e18a51SMatthew Wilcox (Oracle) 3144af2806f8SThomas Graf if (!len) 3145af2806f8SThomas Graf break; 3146af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 3147d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 3148d8e18a51SMatthew Wilcox (Oracle) len); 3149d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3150d8e18a51SMatthew Wilcox (Oracle) len -= size; 3151af2806f8SThomas Graf skb_frag_ref(to, j); 3152af2806f8SThomas Graf j++; 3153af2806f8SThomas Graf } 3154af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 315536d5fe6aSZoltan Kiss 315636d5fe6aSZoltan Kiss return 0; 3157af2806f8SThomas Graf } 3158af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3159af2806f8SThomas Graf 31601da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 31611da177e4SLinus Torvalds { 3162d3bc23e7SAl Viro __wsum csum; 31631da177e4SLinus Torvalds long csstart; 31641da177e4SLinus Torvalds 316584fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 316655508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 31671da177e4SLinus Torvalds else 31681da177e4SLinus Torvalds csstart = skb_headlen(skb); 31691da177e4SLinus Torvalds 317009a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 31711da177e4SLinus Torvalds 3172d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 31731da177e4SLinus Torvalds 31741da177e4SLinus Torvalds csum = 0; 31751da177e4SLinus Torvalds if (csstart != skb->len) 31761da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 31778d5930dfSAl Viro skb->len - csstart); 31781da177e4SLinus Torvalds 317984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3180ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 31811da177e4SLinus Torvalds 3182d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 31831da177e4SLinus Torvalds } 31841da177e4SLinus Torvalds } 3185b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 31861da177e4SLinus Torvalds 31871da177e4SLinus Torvalds /** 31881da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 31891da177e4SLinus Torvalds * @list: list to dequeue from 31901da177e4SLinus Torvalds * 31911da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 31921da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 31931da177e4SLinus Torvalds * returned or %NULL if the list is empty. 31941da177e4SLinus Torvalds */ 31951da177e4SLinus Torvalds 31961da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 31971da177e4SLinus Torvalds { 31981da177e4SLinus Torvalds unsigned long flags; 31991da177e4SLinus Torvalds struct sk_buff *result; 32001da177e4SLinus Torvalds 32011da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32021da177e4SLinus Torvalds result = __skb_dequeue(list); 32031da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32041da177e4SLinus Torvalds return result; 32051da177e4SLinus Torvalds } 3206b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 32071da177e4SLinus Torvalds 32081da177e4SLinus Torvalds /** 32091da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 32101da177e4SLinus Torvalds * @list: list to dequeue from 32111da177e4SLinus Torvalds * 32121da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 32131da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 32141da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32151da177e4SLinus Torvalds */ 32161da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 32171da177e4SLinus Torvalds { 32181da177e4SLinus Torvalds unsigned long flags; 32191da177e4SLinus Torvalds struct sk_buff *result; 32201da177e4SLinus Torvalds 32211da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32221da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 32231da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32241da177e4SLinus Torvalds return result; 32251da177e4SLinus Torvalds } 3226b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 32271da177e4SLinus Torvalds 32281da177e4SLinus Torvalds /** 32291da177e4SLinus Torvalds * skb_queue_purge - empty a list 32301da177e4SLinus Torvalds * @list: list to empty 32311da177e4SLinus Torvalds * 32321da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 32331da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 32341da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 32351da177e4SLinus Torvalds */ 32361da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 32371da177e4SLinus Torvalds { 32381da177e4SLinus Torvalds struct sk_buff *skb; 32391da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 32401da177e4SLinus Torvalds kfree_skb(skb); 32411da177e4SLinus Torvalds } 3242b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 32431da177e4SLinus Torvalds 32441da177e4SLinus Torvalds /** 32459f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 32469f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3247385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 32489f5afeaeSYaogong Wang * 32499f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 32509f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 32519f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 32529f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 32539f5afeaeSYaogong Wang */ 3254385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 32559f5afeaeSYaogong Wang { 32567c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3257385114deSPeter Oskolkov unsigned int sum = 0; 32589f5afeaeSYaogong Wang 32597c90584cSEric Dumazet while (p) { 32607c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 32617c90584cSEric Dumazet 32627c90584cSEric Dumazet p = rb_next(p); 32637c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3264385114deSPeter Oskolkov sum += skb->truesize; 32659f5afeaeSYaogong Wang kfree_skb(skb); 32667c90584cSEric Dumazet } 3267385114deSPeter Oskolkov return sum; 32689f5afeaeSYaogong Wang } 32699f5afeaeSYaogong Wang 32709f5afeaeSYaogong Wang /** 32711da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 32721da177e4SLinus Torvalds * @list: list to use 32731da177e4SLinus Torvalds * @newsk: buffer to queue 32741da177e4SLinus Torvalds * 32751da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 32761da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 32771da177e4SLinus Torvalds * safely. 32781da177e4SLinus Torvalds * 32791da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 32801da177e4SLinus Torvalds */ 32811da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 32821da177e4SLinus Torvalds { 32831da177e4SLinus Torvalds unsigned long flags; 32841da177e4SLinus Torvalds 32851da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32861da177e4SLinus Torvalds __skb_queue_head(list, newsk); 32871da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32881da177e4SLinus Torvalds } 3289b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 32901da177e4SLinus Torvalds 32911da177e4SLinus Torvalds /** 32921da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 32931da177e4SLinus Torvalds * @list: list to use 32941da177e4SLinus Torvalds * @newsk: buffer to queue 32951da177e4SLinus Torvalds * 32961da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 32971da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 32981da177e4SLinus Torvalds * safely. 32991da177e4SLinus Torvalds * 33001da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33011da177e4SLinus Torvalds */ 33021da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 33031da177e4SLinus Torvalds { 33041da177e4SLinus Torvalds unsigned long flags; 33051da177e4SLinus Torvalds 33061da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33071da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 33081da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33091da177e4SLinus Torvalds } 3310b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 33118728b834SDavid S. Miller 33121da177e4SLinus Torvalds /** 33131da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 33141da177e4SLinus Torvalds * @skb: buffer to remove 33158728b834SDavid S. Miller * @list: list to use 33161da177e4SLinus Torvalds * 33178728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 33188728b834SDavid S. Miller * function is atomic with respect to other list locked calls 33191da177e4SLinus Torvalds * 33208728b834SDavid S. Miller * You must know what list the SKB is on. 33211da177e4SLinus Torvalds */ 33228728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 33231da177e4SLinus Torvalds { 33241da177e4SLinus Torvalds unsigned long flags; 33251da177e4SLinus Torvalds 33261da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33278728b834SDavid S. Miller __skb_unlink(skb, list); 33281da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33291da177e4SLinus Torvalds } 3330b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 33311da177e4SLinus Torvalds 33321da177e4SLinus Torvalds /** 33331da177e4SLinus Torvalds * skb_append - append a buffer 33341da177e4SLinus Torvalds * @old: buffer to insert after 33351da177e4SLinus Torvalds * @newsk: buffer to insert 33368728b834SDavid S. Miller * @list: list to use 33371da177e4SLinus Torvalds * 33381da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 33391da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 33401da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33411da177e4SLinus Torvalds */ 33428728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 33431da177e4SLinus Torvalds { 33441da177e4SLinus Torvalds unsigned long flags; 33451da177e4SLinus Torvalds 33468728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 33477de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 33488728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 33491da177e4SLinus Torvalds } 3350b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 33511da177e4SLinus Torvalds 33521da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 33531da177e4SLinus Torvalds struct sk_buff* skb1, 33541da177e4SLinus Torvalds const u32 len, const int pos) 33551da177e4SLinus Torvalds { 33561da177e4SLinus Torvalds int i; 33571da177e4SLinus Torvalds 3358d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3359d626f62bSArnaldo Carvalho de Melo pos - len); 33601da177e4SLinus Torvalds /* And move data appendix as is. */ 33611da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 33621da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 33631da177e4SLinus Torvalds 33641da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 33651da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 33661da177e4SLinus Torvalds skb1->data_len = skb->data_len; 33671da177e4SLinus Torvalds skb1->len += skb1->data_len; 33681da177e4SLinus Torvalds skb->data_len = 0; 33691da177e4SLinus Torvalds skb->len = len; 337027a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 33711da177e4SLinus Torvalds } 33721da177e4SLinus Torvalds 33731da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 33741da177e4SLinus Torvalds struct sk_buff* skb1, 33751da177e4SLinus Torvalds const u32 len, int pos) 33761da177e4SLinus Torvalds { 33771da177e4SLinus Torvalds int i, k = 0; 33781da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 33791da177e4SLinus Torvalds 33801da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 33811da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 33821da177e4SLinus Torvalds skb->len = len; 33831da177e4SLinus Torvalds skb->data_len = len - pos; 33841da177e4SLinus Torvalds 33851da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 33869e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 33871da177e4SLinus Torvalds 33881da177e4SLinus Torvalds if (pos + size > len) { 33891da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 33901da177e4SLinus Torvalds 33911da177e4SLinus Torvalds if (pos < len) { 33921da177e4SLinus Torvalds /* Split frag. 33931da177e4SLinus Torvalds * We have two variants in this case: 33941da177e4SLinus Torvalds * 1. Move all the frag to the second 33951da177e4SLinus Torvalds * part, if it is possible. F.e. 33961da177e4SLinus Torvalds * this approach is mandatory for TUX, 33971da177e4SLinus Torvalds * where splitting is expensive. 33981da177e4SLinus Torvalds * 2. Split is accurately. We make this. 33991da177e4SLinus Torvalds */ 3400ea2ab693SIan Campbell skb_frag_ref(skb, i); 3401b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 34029e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 34039e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 34041da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34051da177e4SLinus Torvalds } 34061da177e4SLinus Torvalds k++; 34071da177e4SLinus Torvalds } else 34081da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34091da177e4SLinus Torvalds pos += size; 34101da177e4SLinus Torvalds } 34111da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 34121da177e4SLinus Torvalds } 34131da177e4SLinus Torvalds 34141da177e4SLinus Torvalds /** 34151da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 34161da177e4SLinus Torvalds * @skb: the buffer to split 34171da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 34181da177e4SLinus Torvalds * @len: new length for skb 34191da177e4SLinus Torvalds */ 34201da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 34211da177e4SLinus Torvalds { 34221da177e4SLinus Torvalds int pos = skb_headlen(skb); 34231da177e4SLinus Torvalds 342406b4feb3SJonathan Lemon skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 34251f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 34261da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 34271da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 34281da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 34291da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 34301da177e4SLinus Torvalds } 3431b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 34321da177e4SLinus Torvalds 34339f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 34349f782db3SIlpo Järvinen * 34359f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 34369f782db3SIlpo Järvinen */ 3437832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3438832d11c5SIlpo Järvinen { 3439097b9146SMarco Elver int ret = 0; 3440097b9146SMarco Elver 3441097b9146SMarco Elver if (skb_cloned(skb)) { 3442097b9146SMarco Elver /* Save and restore truesize: pskb_expand_head() may reallocate 3443097b9146SMarco Elver * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we 3444097b9146SMarco Elver * cannot change truesize at this point. 3445097b9146SMarco Elver */ 3446097b9146SMarco Elver unsigned int save_truesize = skb->truesize; 3447097b9146SMarco Elver 3448097b9146SMarco Elver ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3449097b9146SMarco Elver skb->truesize = save_truesize; 3450097b9146SMarco Elver } 3451097b9146SMarco Elver return ret; 3452832d11c5SIlpo Järvinen } 3453832d11c5SIlpo Järvinen 3454832d11c5SIlpo Järvinen /** 3455832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3456832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3457832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3458832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3459832d11c5SIlpo Järvinen * 3460832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 346120e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3462832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3463832d11c5SIlpo Järvinen * 3464832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3465832d11c5SIlpo Järvinen * 3466832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3467832d11c5SIlpo Järvinen * to have non-paged data as well. 3468832d11c5SIlpo Järvinen * 3469832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3470832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3471832d11c5SIlpo Järvinen */ 3472832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3473832d11c5SIlpo Järvinen { 3474832d11c5SIlpo Järvinen int from, to, merge, todo; 3475d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3476832d11c5SIlpo Järvinen 3477832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3478f8071cdeSEric Dumazet 3479f8071cdeSEric Dumazet if (skb_headlen(skb)) 3480f8071cdeSEric Dumazet return 0; 34811f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 34821f8b977aSWillem de Bruijn return 0; 3483832d11c5SIlpo Järvinen 3484832d11c5SIlpo Järvinen todo = shiftlen; 3485832d11c5SIlpo Järvinen from = 0; 3486832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3487832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3488832d11c5SIlpo Järvinen 3489832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3490832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3491832d11c5SIlpo Järvinen */ 3492832d11c5SIlpo Järvinen if (!to || 3493ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3494b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3495832d11c5SIlpo Järvinen merge = -1; 3496832d11c5SIlpo Järvinen } else { 3497832d11c5SIlpo Järvinen merge = to - 1; 3498832d11c5SIlpo Järvinen 34999e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3500832d11c5SIlpo Järvinen if (todo < 0) { 3501832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3502832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3503832d11c5SIlpo Järvinen return 0; 3504832d11c5SIlpo Järvinen 35059f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 35069f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3507832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3508832d11c5SIlpo Järvinen 35099e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 35109e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3511b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3512832d11c5SIlpo Järvinen 3513832d11c5SIlpo Järvinen goto onlymerged; 3514832d11c5SIlpo Järvinen } 3515832d11c5SIlpo Järvinen 3516832d11c5SIlpo Järvinen from++; 3517832d11c5SIlpo Järvinen } 3518832d11c5SIlpo Järvinen 3519832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3520832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3521832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3522832d11c5SIlpo Järvinen return 0; 3523832d11c5SIlpo Järvinen 3524832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3525832d11c5SIlpo Järvinen return 0; 3526832d11c5SIlpo Järvinen 3527832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3528832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3529832d11c5SIlpo Järvinen return 0; 3530832d11c5SIlpo Järvinen 3531832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3532832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3533832d11c5SIlpo Järvinen 35349e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3535832d11c5SIlpo Järvinen *fragto = *fragfrom; 35369e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3537832d11c5SIlpo Järvinen from++; 3538832d11c5SIlpo Järvinen to++; 3539832d11c5SIlpo Järvinen 3540832d11c5SIlpo Järvinen } else { 3541ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3542b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3543b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 35449e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3545832d11c5SIlpo Järvinen 3546b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 35479e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3548832d11c5SIlpo Järvinen todo = 0; 3549832d11c5SIlpo Järvinen 3550832d11c5SIlpo Järvinen to++; 3551832d11c5SIlpo Järvinen break; 3552832d11c5SIlpo Järvinen } 3553832d11c5SIlpo Järvinen } 3554832d11c5SIlpo Järvinen 3555832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3556832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3557832d11c5SIlpo Järvinen 3558832d11c5SIlpo Järvinen if (merge >= 0) { 3559832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3560832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3561832d11c5SIlpo Järvinen 35629e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 35636a5bcd84SIlias Apalodimas __skb_frag_unref(fragfrom, skb->pp_recycle); 3564832d11c5SIlpo Järvinen } 3565832d11c5SIlpo Järvinen 3566832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3567832d11c5SIlpo Järvinen to = 0; 3568832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3569832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3570832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3571832d11c5SIlpo Järvinen 3572832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3573832d11c5SIlpo Järvinen 3574832d11c5SIlpo Järvinen onlymerged: 3575832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3576832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3577832d11c5SIlpo Järvinen */ 3578832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3579832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3580832d11c5SIlpo Järvinen 3581832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3582832d11c5SIlpo Järvinen skb->len -= shiftlen; 3583832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3584832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3585832d11c5SIlpo Järvinen tgt->len += shiftlen; 3586832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3587832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3588832d11c5SIlpo Järvinen 3589832d11c5SIlpo Järvinen return shiftlen; 3590832d11c5SIlpo Järvinen } 3591832d11c5SIlpo Järvinen 3592677e90edSThomas Graf /** 3593677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3594677e90edSThomas Graf * @skb: the buffer to read 3595677e90edSThomas Graf * @from: lower offset of data to be read 3596677e90edSThomas Graf * @to: upper offset of data to be read 3597677e90edSThomas Graf * @st: state variable 3598677e90edSThomas Graf * 3599677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3600677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3601677e90edSThomas Graf */ 3602677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3603677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3604677e90edSThomas Graf { 3605677e90edSThomas Graf st->lower_offset = from; 3606677e90edSThomas Graf st->upper_offset = to; 3607677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3608677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3609677e90edSThomas Graf st->frag_data = NULL; 361097550f6fSWillem de Bruijn st->frag_off = 0; 3611677e90edSThomas Graf } 3612b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3613677e90edSThomas Graf 3614677e90edSThomas Graf /** 3615677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3616677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3617677e90edSThomas Graf * @data: destination pointer for data to be returned 3618677e90edSThomas Graf * @st: state variable 3619677e90edSThomas Graf * 3620bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3621677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3622bc32383cSMathias Krause * the head of the data block to @data and returns the length 3623677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3624677e90edSThomas Graf * offset has been reached. 3625677e90edSThomas Graf * 3626677e90edSThomas Graf * The caller is not required to consume all of the data 3627bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3628677e90edSThomas Graf * of bytes already consumed and the next call to 3629677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3630677e90edSThomas Graf * 363125985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3632e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3633677e90edSThomas Graf * reads of potentially non linear data. 3634677e90edSThomas Graf * 3635bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3636677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3637677e90edSThomas Graf * a stack for this purpose. 3638677e90edSThomas Graf */ 3639677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3640677e90edSThomas Graf struct skb_seq_state *st) 3641677e90edSThomas Graf { 3642677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3643677e90edSThomas Graf skb_frag_t *frag; 3644677e90edSThomas Graf 3645aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3646aeb193eaSWedson Almeida Filho if (st->frag_data) { 3647aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3648aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3649aeb193eaSWedson Almeida Filho } 3650677e90edSThomas Graf return 0; 3651aeb193eaSWedson Almeida Filho } 3652677e90edSThomas Graf 3653677e90edSThomas Graf next_skb: 365495e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3655677e90edSThomas Graf 3656995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 365795e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3658677e90edSThomas Graf return block_limit - abs_offset; 3659677e90edSThomas Graf } 3660677e90edSThomas Graf 3661677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3662677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3663677e90edSThomas Graf 3664677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 366597550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3666677e90edSThomas Graf 366797550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 366897550f6fSWillem de Bruijn 366997550f6fSWillem de Bruijn pg_idx = 0; 367097550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 367197550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 367297550f6fSWillem de Bruijn 367397550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 367497550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 367597550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 367697550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 367797550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 367897550f6fSWillem de Bruijn } 367997550f6fSWillem de Bruijn 368097550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3681677e90edSThomas Graf if (abs_offset < block_limit) { 3682677e90edSThomas Graf if (!st->frag_data) 368397550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3684677e90edSThomas Graf 368597550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3686677e90edSThomas Graf (abs_offset - st->stepped_offset); 3687677e90edSThomas Graf 3688677e90edSThomas Graf return block_limit - abs_offset; 3689677e90edSThomas Graf } 3690677e90edSThomas Graf 3691677e90edSThomas Graf if (st->frag_data) { 369251c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3693677e90edSThomas Graf st->frag_data = NULL; 3694677e90edSThomas Graf } 3695677e90edSThomas Graf 369697550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 369797550f6fSWillem de Bruijn st->frag_off += pg_sz; 369897550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 369997550f6fSWillem de Bruijn st->frag_off = 0; 3700677e90edSThomas Graf st->frag_idx++; 370197550f6fSWillem de Bruijn } 3702677e90edSThomas Graf } 3703677e90edSThomas Graf 37045b5a60daSOlaf Kirch if (st->frag_data) { 370551c56b00SEric Dumazet kunmap_atomic(st->frag_data); 37065b5a60daSOlaf Kirch st->frag_data = NULL; 37075b5a60daSOlaf Kirch } 37085b5a60daSOlaf Kirch 370921dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3710677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 371195e3b24cSHerbert Xu st->frag_idx = 0; 3712677e90edSThomas Graf goto next_skb; 371371b3346dSShyam Iyer } else if (st->cur_skb->next) { 371471b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 371571b3346dSShyam Iyer st->frag_idx = 0; 3716677e90edSThomas Graf goto next_skb; 3717677e90edSThomas Graf } 3718677e90edSThomas Graf 3719677e90edSThomas Graf return 0; 3720677e90edSThomas Graf } 3721b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3722677e90edSThomas Graf 3723677e90edSThomas Graf /** 3724677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3725677e90edSThomas Graf * @st: state variable 3726677e90edSThomas Graf * 3727677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3728677e90edSThomas Graf * returned 0. 3729677e90edSThomas Graf */ 3730677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3731677e90edSThomas Graf { 3732677e90edSThomas Graf if (st->frag_data) 373351c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3734677e90edSThomas Graf } 3735b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3736677e90edSThomas Graf 37373fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 37383fc7e8a6SThomas Graf 37393fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 37403fc7e8a6SThomas Graf struct ts_config *conf, 37413fc7e8a6SThomas Graf struct ts_state *state) 37423fc7e8a6SThomas Graf { 37433fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 37443fc7e8a6SThomas Graf } 37453fc7e8a6SThomas Graf 37463fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 37473fc7e8a6SThomas Graf { 37483fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 37493fc7e8a6SThomas Graf } 37503fc7e8a6SThomas Graf 37513fc7e8a6SThomas Graf /** 37523fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 37533fc7e8a6SThomas Graf * @skb: the buffer to look in 37543fc7e8a6SThomas Graf * @from: search offset 37553fc7e8a6SThomas Graf * @to: search limit 37563fc7e8a6SThomas Graf * @config: textsearch configuration 37573fc7e8a6SThomas Graf * 37583fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 37593fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 37603fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 37613fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 37623fc7e8a6SThomas Graf */ 37633fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3764059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 37653fc7e8a6SThomas Graf { 3766059a2440SBojan Prtvar struct ts_state state; 3767f72b948dSPhil Oester unsigned int ret; 3768f72b948dSPhil Oester 3769b228c9b0SWillem de Bruijn BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb)); 3770b228c9b0SWillem de Bruijn 37713fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 37723fc7e8a6SThomas Graf config->finish = skb_ts_finish; 37733fc7e8a6SThomas Graf 3774059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 37753fc7e8a6SThomas Graf 3776059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3777f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 37783fc7e8a6SThomas Graf } 3779b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 37803fc7e8a6SThomas Graf 3781be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3782be12a1feSHannes Frederic Sowa int offset, size_t size) 3783be12a1feSHannes Frederic Sowa { 3784be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3785be12a1feSHannes Frederic Sowa 3786be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3787be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3788be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3789be12a1feSHannes Frederic Sowa get_page(page); 3790be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3791be12a1feSHannes Frederic Sowa } else { 3792be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3793be12a1feSHannes Frederic Sowa } 3794be12a1feSHannes Frederic Sowa 3795be12a1feSHannes Frederic Sowa return 0; 3796be12a1feSHannes Frederic Sowa } 3797be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3798be12a1feSHannes Frederic Sowa 3799cbb042f9SHerbert Xu /** 3800cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3801cbb042f9SHerbert Xu * @skb: buffer to update 3802cbb042f9SHerbert Xu * @len: length of data pulled 3803cbb042f9SHerbert Xu * 3804cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3805fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 380684fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 380784fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 380884fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3809cbb042f9SHerbert Xu */ 3810af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3811cbb042f9SHerbert Xu { 381231b33dfbSPravin B Shelar unsigned char *data = skb->data; 381331b33dfbSPravin B Shelar 3814cbb042f9SHerbert Xu BUG_ON(len > skb->len); 381531b33dfbSPravin B Shelar __skb_pull(skb, len); 381631b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 381731b33dfbSPravin B Shelar return skb->data; 3818cbb042f9SHerbert Xu } 3819f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3820f94691acSArnaldo Carvalho de Melo 382113acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 382213acc94eSYonghong Song { 382313acc94eSYonghong Song skb_frag_t head_frag; 382413acc94eSYonghong Song struct page *page; 382513acc94eSYonghong Song 382613acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3827d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3828b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3829b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3830d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 383113acc94eSYonghong Song return head_frag; 383213acc94eSYonghong Song } 383313acc94eSYonghong Song 38343a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 38353a1296a3SSteffen Klassert netdev_features_t features, 38363a1296a3SSteffen Klassert unsigned int offset) 38373a1296a3SSteffen Klassert { 38383a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 38393a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 38403a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 38413a1296a3SSteffen Klassert unsigned int delta_len = 0; 38423a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 384353475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 384453475c5dSDongseok Yi int err; 38453a1296a3SSteffen Klassert 38463a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 38473a1296a3SSteffen Klassert 38483a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 38493a1296a3SSteffen Klassert 38503a1296a3SSteffen Klassert do { 38513a1296a3SSteffen Klassert nskb = list_skb; 38523a1296a3SSteffen Klassert list_skb = list_skb->next; 38533a1296a3SSteffen Klassert 385453475c5dSDongseok Yi err = 0; 385553475c5dSDongseok Yi if (skb_shared(nskb)) { 385653475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 385753475c5dSDongseok Yi if (tmp) { 385853475c5dSDongseok Yi consume_skb(nskb); 385953475c5dSDongseok Yi nskb = tmp; 386053475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 386153475c5dSDongseok Yi } else { 386253475c5dSDongseok Yi err = -ENOMEM; 386353475c5dSDongseok Yi } 386453475c5dSDongseok Yi } 386553475c5dSDongseok Yi 38663a1296a3SSteffen Klassert if (!tail) 38673a1296a3SSteffen Klassert skb->next = nskb; 38683a1296a3SSteffen Klassert else 38693a1296a3SSteffen Klassert tail->next = nskb; 38703a1296a3SSteffen Klassert 387153475c5dSDongseok Yi if (unlikely(err)) { 387253475c5dSDongseok Yi nskb->next = list_skb; 387353475c5dSDongseok Yi goto err_linearize; 387453475c5dSDongseok Yi } 387553475c5dSDongseok Yi 38763a1296a3SSteffen Klassert tail = nskb; 38773a1296a3SSteffen Klassert 38783a1296a3SSteffen Klassert delta_len += nskb->len; 38793a1296a3SSteffen Klassert delta_truesize += nskb->truesize; 38803a1296a3SSteffen Klassert 38813a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 38823a1296a3SSteffen Klassert 3883cf673ed0SFlorian Westphal skb_release_head_state(nskb); 38843a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 38853a1296a3SSteffen Klassert 38863a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 38873a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 38883a1296a3SSteffen Klassert nskb->data - tnl_hlen, 38893a1296a3SSteffen Klassert offset + tnl_hlen); 38903a1296a3SSteffen Klassert 38913a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 38923a1296a3SSteffen Klassert __skb_linearize(nskb)) 38933a1296a3SSteffen Klassert goto err_linearize; 38943a1296a3SSteffen Klassert 38953a1296a3SSteffen Klassert } while (list_skb); 38963a1296a3SSteffen Klassert 38973a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 38983a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 38993a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 39003a1296a3SSteffen Klassert 39013a1296a3SSteffen Klassert skb_gso_reset(skb); 39023a1296a3SSteffen Klassert 39033a1296a3SSteffen Klassert skb->prev = tail; 39043a1296a3SSteffen Klassert 39053a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 39063a1296a3SSteffen Klassert __skb_linearize(skb)) 39073a1296a3SSteffen Klassert goto err_linearize; 39083a1296a3SSteffen Klassert 39093a1296a3SSteffen Klassert skb_get(skb); 39103a1296a3SSteffen Klassert 39113a1296a3SSteffen Klassert return skb; 39123a1296a3SSteffen Klassert 39133a1296a3SSteffen Klassert err_linearize: 39143a1296a3SSteffen Klassert kfree_skb_list(skb->next); 39153a1296a3SSteffen Klassert skb->next = NULL; 39163a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 39173a1296a3SSteffen Klassert } 39183a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 39193a1296a3SSteffen Klassert 39203a1296a3SSteffen Klassert int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb) 39213a1296a3SSteffen Klassert { 39223a1296a3SSteffen Klassert if (unlikely(p->len + skb->len >= 65536)) 39233a1296a3SSteffen Klassert return -E2BIG; 39243a1296a3SSteffen Klassert 39253a1296a3SSteffen Klassert if (NAPI_GRO_CB(p)->last == p) 39263a1296a3SSteffen Klassert skb_shinfo(p)->frag_list = skb; 39273a1296a3SSteffen Klassert else 39283a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last->next = skb; 39293a1296a3SSteffen Klassert 39303a1296a3SSteffen Klassert skb_pull(skb, skb_gro_offset(skb)); 39313a1296a3SSteffen Klassert 39323a1296a3SSteffen Klassert NAPI_GRO_CB(p)->last = skb; 39333a1296a3SSteffen Klassert NAPI_GRO_CB(p)->count++; 39343a1296a3SSteffen Klassert p->data_len += skb->len; 39355e10da53SPaolo Abeni 39365e10da53SPaolo Abeni /* sk owenrship - if any - completely transferred to the aggregated packet */ 39375e10da53SPaolo Abeni skb->destructor = NULL; 39383a1296a3SSteffen Klassert p->truesize += skb->truesize; 39393a1296a3SSteffen Klassert p->len += skb->len; 39403a1296a3SSteffen Klassert 39413a1296a3SSteffen Klassert NAPI_GRO_CB(skb)->same_flow = 1; 39423a1296a3SSteffen Klassert 39433a1296a3SSteffen Klassert return 0; 39443a1296a3SSteffen Klassert } 39453a1296a3SSteffen Klassert 3946f4c50d99SHerbert Xu /** 3947f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3948df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3949576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3950f4c50d99SHerbert Xu * 3951f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 39524c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 39534c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3954f4c50d99SHerbert Xu */ 3955df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3956df5771ffSMichael S. Tsirkin netdev_features_t features) 3957f4c50d99SHerbert Xu { 3958f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3959f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 39601a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3961df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3962df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3963df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 39641fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3965f4c50d99SHerbert Xu unsigned int offset = doffset; 3966df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3967802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3968f4c50d99SHerbert Xu unsigned int headroom; 3969802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3970ec5f0615SPravin B Shelar __be16 proto; 397136c98382SAlexander Duyck bool csum, sg; 3972df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3973f4c50d99SHerbert Xu int err = -ENOMEM; 3974f4c50d99SHerbert Xu int i = 0; 3975f4c50d99SHerbert Xu int pos; 3976f4c50d99SHerbert Xu 39773dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 39783dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 39793dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 39803dcbdb13SShmulik Ladkani * non head_frag head. 39813dcbdb13SShmulik Ladkani * 39823dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 39833dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 39843dcbdb13SShmulik Ladkani * head, then the first one has too). 39853dcbdb13SShmulik Ladkani * 39863dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 39873dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 39883dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 39893dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 39903dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 39913dcbdb13SShmulik Ladkani */ 39923dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 39933dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 39943dcbdb13SShmulik Ladkani } 39953dcbdb13SShmulik Ladkani 39965882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 39972f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 3998ec5f0615SPravin B Shelar if (unlikely(!proto)) 3999ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 4000ec5f0615SPravin B Shelar 400136c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 4002f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 40037e2b10c1STom Herbert 400407b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 400507b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 400607b26c94SSteffen Klassert struct sk_buff *iter; 400743170c4eSIlan Tayari unsigned int frag_len; 400807b26c94SSteffen Klassert 400907b26c94SSteffen Klassert if (!list_skb || 401007b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 401107b26c94SSteffen Klassert goto normal; 401207b26c94SSteffen Klassert 401343170c4eSIlan Tayari /* If we get here then all the required 401443170c4eSIlan Tayari * GSO features except frag_list are supported. 401543170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 401643170c4eSIlan Tayari * with no frag_list. 401743170c4eSIlan Tayari * Currently we can do that only when the buffers don't 401843170c4eSIlan Tayari * have a linear part and all the buffers except 401943170c4eSIlan Tayari * the last are of the same length. 402007b26c94SSteffen Klassert */ 402143170c4eSIlan Tayari frag_len = list_skb->len; 402207b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 402343170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 402443170c4eSIlan Tayari goto normal; 4025eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 402607b26c94SSteffen Klassert goto normal; 402707b26c94SSteffen Klassert 402807b26c94SSteffen Klassert len -= iter->len; 402907b26c94SSteffen Klassert } 403043170c4eSIlan Tayari 403143170c4eSIlan Tayari if (len != frag_len) 403243170c4eSIlan Tayari goto normal; 403307b26c94SSteffen Klassert } 403407b26c94SSteffen Klassert 4035802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 4036802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 4037802ab55aSAlexander Duyck * now. 4038802ab55aSAlexander Duyck */ 4039802ab55aSAlexander Duyck partial_segs = len / mss; 4040d7fb5a80SAlexander Duyck if (partial_segs > 1) 4041802ab55aSAlexander Duyck mss *= partial_segs; 4042d7fb5a80SAlexander Duyck else 4043d7fb5a80SAlexander Duyck partial_segs = 0; 4044802ab55aSAlexander Duyck } 4045802ab55aSAlexander Duyck 404607b26c94SSteffen Klassert normal: 4047df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 4048df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 4049f4c50d99SHerbert Xu 4050f4c50d99SHerbert Xu do { 4051f4c50d99SHerbert Xu struct sk_buff *nskb; 40528cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 4053c8884eddSHerbert Xu int hsize; 4054f4c50d99SHerbert Xu int size; 4055f4c50d99SHerbert Xu 40563953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 40573953c46cSMarcelo Ricardo Leitner len = list_skb->len; 40583953c46cSMarcelo Ricardo Leitner } else { 4059df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 4060f4c50d99SHerbert Xu if (len > mss) 4061f4c50d99SHerbert Xu len = mss; 40623953c46cSMarcelo Ricardo Leitner } 4063f4c50d99SHerbert Xu 4064df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 4065f4c50d99SHerbert Xu 4066dbd50f23SXin Long if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) && 40671a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 40681a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 406989319d38SHerbert Xu 40709d8506ccSHerbert Xu i = 0; 40711a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 40721a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 40731fd819ecSMichael S. Tsirkin frag_skb = list_skb; 40741a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 40759d8506ccSHerbert Xu 40769d8506ccSHerbert Xu while (pos < offset + len) { 40779d8506ccSHerbert Xu BUG_ON(i >= nfrags); 40789d8506ccSHerbert Xu 40794e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 40809d8506ccSHerbert Xu if (pos + size > offset + len) 40819d8506ccSHerbert Xu break; 40829d8506ccSHerbert Xu 40839d8506ccSHerbert Xu i++; 40849d8506ccSHerbert Xu pos += size; 40854e1beba1SMichael S. Tsirkin frag++; 40869d8506ccSHerbert Xu } 40879d8506ccSHerbert Xu 40881a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 40891a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 409089319d38SHerbert Xu 4091f4c50d99SHerbert Xu if (unlikely(!nskb)) 4092f4c50d99SHerbert Xu goto err; 4093f4c50d99SHerbert Xu 40949d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 40959d8506ccSHerbert Xu kfree_skb(nskb); 40969d8506ccSHerbert Xu goto err; 40979d8506ccSHerbert Xu } 40989d8506ccSHerbert Xu 4099ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 410089319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 410189319d38SHerbert Xu kfree_skb(nskb); 410289319d38SHerbert Xu goto err; 410389319d38SHerbert Xu } 410489319d38SHerbert Xu 4105ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 410689319d38SHerbert Xu skb_release_head_state(nskb); 410789319d38SHerbert Xu __skb_push(nskb, doffset); 410889319d38SHerbert Xu } else { 410900b229f7SPaolo Abeni if (hsize < 0) 411000b229f7SPaolo Abeni hsize = 0; 4111dbd50f23SXin Long if (hsize > len || !sg) 4112dbd50f23SXin Long hsize = len; 4113dbd50f23SXin Long 4114c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 4115df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 4116c93bdd0eSMel Gorman NUMA_NO_NODE); 411789319d38SHerbert Xu 411889319d38SHerbert Xu if (unlikely(!nskb)) 411989319d38SHerbert Xu goto err; 412089319d38SHerbert Xu 412189319d38SHerbert Xu skb_reserve(nskb, headroom); 412289319d38SHerbert Xu __skb_put(nskb, doffset); 412389319d38SHerbert Xu } 412489319d38SHerbert Xu 4125f4c50d99SHerbert Xu if (segs) 4126f4c50d99SHerbert Xu tail->next = nskb; 4127f4c50d99SHerbert Xu else 4128f4c50d99SHerbert Xu segs = nskb; 4129f4c50d99SHerbert Xu tail = nskb; 4130f4c50d99SHerbert Xu 4131df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 4132f4c50d99SHerbert Xu 4133030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 4134fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 413568c33163SPravin B Shelar 4136df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 413768c33163SPravin B Shelar nskb->data - tnl_hlen, 413868c33163SPravin B Shelar doffset + tnl_hlen); 413989319d38SHerbert Xu 41409d8506ccSHerbert Xu if (nskb->len == len + doffset) 41411cdbcb79SSimon Horman goto perform_csum_check; 414289319d38SHerbert Xu 41437fbeffedSAlexander Duyck if (!sg) { 41441454c9faSYadu Kishore if (!csum) { 41457fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 41466f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 414776443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 414876443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 41491454c9faSYadu Kishore skb_put(nskb, 41501454c9faSYadu Kishore len), 41518d5930dfSAl Viro len); 41527e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 4153de843723STom Herbert skb_headroom(nskb) + doffset; 41541454c9faSYadu Kishore } else { 41551454c9faSYadu Kishore skb_copy_bits(head_skb, offset, 41561454c9faSYadu Kishore skb_put(nskb, len), 41571454c9faSYadu Kishore len); 41581454c9faSYadu Kishore } 4159f4c50d99SHerbert Xu continue; 4160f4c50d99SHerbert Xu } 4161f4c50d99SHerbert Xu 41628cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 4163f4c50d99SHerbert Xu 4164df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 4165d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4166f4c50d99SHerbert Xu 416706b4feb3SJonathan Lemon skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags & 416806b4feb3SJonathan Lemon SKBFL_SHARED_FRAG; 4169cef401deSEric Dumazet 4170bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4171bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4172bf5c25d6SWillem de Bruijn goto err; 4173bf5c25d6SWillem de Bruijn 41749d8506ccSHerbert Xu while (pos < offset + len) { 41759d8506ccSHerbert Xu if (i >= nfrags) { 41769d8506ccSHerbert Xu i = 0; 41771a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 41781a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 41791fd819ecSMichael S. Tsirkin frag_skb = list_skb; 418013acc94eSYonghong Song if (!skb_headlen(list_skb)) { 41819d8506ccSHerbert Xu BUG_ON(!nfrags); 418213acc94eSYonghong Song } else { 418313acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 41849d8506ccSHerbert Xu 418513acc94eSYonghong Song /* to make room for head_frag. */ 418613acc94eSYonghong Song i--; 418713acc94eSYonghong Song frag--; 418813acc94eSYonghong Song } 4189bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4190bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4191bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4192bf5c25d6SWillem de Bruijn goto err; 4193bf5c25d6SWillem de Bruijn 41941a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 41959d8506ccSHerbert Xu } 41969d8506ccSHerbert Xu 41979d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 41989d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 41999d8506ccSHerbert Xu net_warn_ratelimited( 42009d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 42019d8506ccSHerbert Xu pos, mss); 4202ff907a11SEric Dumazet err = -EINVAL; 42039d8506ccSHerbert Xu goto err; 42049d8506ccSHerbert Xu } 42059d8506ccSHerbert Xu 420613acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 42078cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 42088cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4209f4c50d99SHerbert Xu 4210f4c50d99SHerbert Xu if (pos < offset) { 4211b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 42128cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4213f4c50d99SHerbert Xu } 4214f4c50d99SHerbert Xu 421589319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4216f4c50d99SHerbert Xu 4217f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4218f4c50d99SHerbert Xu i++; 42194e1beba1SMichael S. Tsirkin frag++; 4220f4c50d99SHerbert Xu pos += size; 4221f4c50d99SHerbert Xu } else { 42228cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 422389319d38SHerbert Xu goto skip_fraglist; 4224f4c50d99SHerbert Xu } 4225f4c50d99SHerbert Xu 42268cb19905SMichael S. Tsirkin nskb_frag++; 4227f4c50d99SHerbert Xu } 4228f4c50d99SHerbert Xu 422989319d38SHerbert Xu skip_fraglist: 4230f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4231f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4232f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4233ec5f0615SPravin B Shelar 42341cdbcb79SSimon Horman perform_csum_check: 42357fbeffedSAlexander Duyck if (!csum) { 4236ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4237ff907a11SEric Dumazet __skb_linearize(nskb)) 4238ddff00d4SAlexander Duyck goto err; 4239ff907a11SEric Dumazet 42407fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4241ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 424276443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 424376443456SAlexander Duyck skb_checksum(nskb, doffset, 424476443456SAlexander Duyck nskb->len - doffset, 0); 42457e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 42467e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4247ec5f0615SPravin B Shelar } 4248df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4249f4c50d99SHerbert Xu 4250bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4251bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4252bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4253bec3cfdcSEric Dumazet */ 4254bec3cfdcSEric Dumazet segs->prev = tail; 4255432c856fSToshiaki Makita 4256802ab55aSAlexander Duyck if (partial_segs) { 425707b26c94SSteffen Klassert struct sk_buff *iter; 4258802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 425907b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4260802ab55aSAlexander Duyck 4261802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 426207b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4263802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4264802ab55aSAlexander Duyck 4265802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4266802ab55aSAlexander Duyck * our way back down the stack of protocols. 4267802ab55aSAlexander Duyck */ 426807b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 426907b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 427007b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 427107b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 427207b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 427307b26c94SSteffen Klassert } 427407b26c94SSteffen Klassert 427507b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 427607b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 427707b26c94SSteffen Klassert else if (tail != segs) 427807b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4279802ab55aSAlexander Duyck } 4280802ab55aSAlexander Duyck 4281432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4282432c856fSToshiaki Makita * using skb_set_owner_w(). 4283432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4284432c856fSToshiaki Makita */ 4285432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4286432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4287432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4288432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4289432c856fSToshiaki Makita } 4290f4c50d99SHerbert Xu return segs; 4291f4c50d99SHerbert Xu 4292f4c50d99SHerbert Xu err: 4293289dccbeSEric Dumazet kfree_skb_list(segs); 4294f4c50d99SHerbert Xu return ERR_PTR(err); 4295f4c50d99SHerbert Xu } 4296f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4297f4c50d99SHerbert Xu 4298d4546c25SDavid Miller int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb) 429971d93b39SHerbert Xu { 43008a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 430167147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 430267147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 43038a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 4304715dc1f3SEric Dumazet unsigned int delta_truesize; 43055e10da53SPaolo Abeni unsigned int new_truesize; 4306d4546c25SDavid Miller struct sk_buff *lp; 430771d93b39SHerbert Xu 43080ab03f35SSteffen Klassert if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush)) 430971d93b39SHerbert Xu return -E2BIG; 431071d93b39SHerbert Xu 431129e98242SEric Dumazet lp = NAPI_GRO_CB(p)->last; 43128a29111cSEric Dumazet pinfo = skb_shinfo(lp); 43138a29111cSEric Dumazet 43148a29111cSEric Dumazet if (headlen <= offset) { 431542da6994SHerbert Xu skb_frag_t *frag; 431666e92fcfSHerbert Xu skb_frag_t *frag2; 43179aaa156cSHerbert Xu int i = skbinfo->nr_frags; 43189aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 431942da6994SHerbert Xu 432066e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 43218a29111cSEric Dumazet goto merge; 432281705ad1SHerbert Xu 43238a29111cSEric Dumazet offset -= headlen; 43249aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 43259aaa156cSHerbert Xu skbinfo->nr_frags = 0; 4326f5572068SHerbert Xu 43279aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 43289aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 432966e92fcfSHerbert Xu do { 433066e92fcfSHerbert Xu *--frag = *--frag2; 433166e92fcfSHerbert Xu } while (--i); 433266e92fcfSHerbert Xu 4333b54c9d5bSJonathan Lemon skb_frag_off_add(frag, offset); 43349e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 433566e92fcfSHerbert Xu 4336715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 43375e10da53SPaolo Abeni new_truesize = SKB_TRUESIZE(skb_end_offset(skb)); 43385e10da53SPaolo Abeni delta_truesize = skb->truesize - new_truesize; 4339715dc1f3SEric Dumazet 43405e10da53SPaolo Abeni skb->truesize = new_truesize; 4341f5572068SHerbert Xu skb->len -= skb->data_len; 4342f5572068SHerbert Xu skb->data_len = 0; 4343f5572068SHerbert Xu 4344715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 43455d38a079SHerbert Xu goto done; 4346d7e8883cSEric Dumazet } else if (skb->head_frag) { 4347d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 4348d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 4349d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 4350d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 4351d7e8883cSEric Dumazet unsigned int first_offset; 4352d7e8883cSEric Dumazet 4353d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 43548a29111cSEric Dumazet goto merge; 4355d7e8883cSEric Dumazet 4356d7e8883cSEric Dumazet first_offset = skb->data - 4357d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 4358d7e8883cSEric Dumazet offset; 4359d7e8883cSEric Dumazet 4360d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 4361d7e8883cSEric Dumazet 4362d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(frag, page); 4363b54c9d5bSJonathan Lemon skb_frag_off_set(frag, first_offset); 4364d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 4365d7e8883cSEric Dumazet 4366d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 4367d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 4368d7e8883cSEric Dumazet 4369af352460SPaolo Abeni new_truesize = SKB_DATA_ALIGN(sizeof(struct sk_buff)); 43705e10da53SPaolo Abeni delta_truesize = skb->truesize - new_truesize; 43715e10da53SPaolo Abeni skb->truesize = new_truesize; 4372d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 4373d7e8883cSEric Dumazet goto done; 43748a29111cSEric Dumazet } 437571d93b39SHerbert Xu 437671d93b39SHerbert Xu merge: 43775e10da53SPaolo Abeni /* sk owenrship - if any - completely transferred to the aggregated packet */ 43785e10da53SPaolo Abeni skb->destructor = NULL; 4379715dc1f3SEric Dumazet delta_truesize = skb->truesize; 438067147ba9SHerbert Xu if (offset > headlen) { 4381d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 4382d1dc7abfSMichal Schmidt 4383b54c9d5bSJonathan Lemon skb_frag_off_add(&skbinfo->frags[0], eat); 43849e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 4385d1dc7abfSMichal Schmidt skb->data_len -= eat; 4386d1dc7abfSMichal Schmidt skb->len -= eat; 438767147ba9SHerbert Xu offset = headlen; 438856035022SHerbert Xu } 438956035022SHerbert Xu 439067147ba9SHerbert Xu __skb_pull(skb, offset); 439156035022SHerbert Xu 439229e98242SEric Dumazet if (NAPI_GRO_CB(p)->last == p) 43938a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 43948a29111cSEric Dumazet else 4395c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 4396c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 4397f4a775d1SEric Dumazet __skb_header_release(skb); 43988a29111cSEric Dumazet lp = p; 439971d93b39SHerbert Xu 44005d38a079SHerbert Xu done: 44015d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 440237fe4732SHerbert Xu p->data_len += len; 4403715dc1f3SEric Dumazet p->truesize += delta_truesize; 440437fe4732SHerbert Xu p->len += len; 44058a29111cSEric Dumazet if (lp != p) { 44068a29111cSEric Dumazet lp->data_len += len; 44078a29111cSEric Dumazet lp->truesize += delta_truesize; 44088a29111cSEric Dumazet lp->len += len; 44098a29111cSEric Dumazet } 441071d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 441171d93b39SHerbert Xu return 0; 441271d93b39SHerbert Xu } 441371d93b39SHerbert Xu 4414df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4415df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4416df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4417df5042f4SFlorian Westphal 4418df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4419df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4420df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4421df5042f4SFlorian Westphal #endif 44224165079bSFlorian Westphal #ifdef CONFIG_XFRM 44234165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 44244165079bSFlorian Westphal #endif 442595a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 442695a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 442795a7233cSPaul Blakey #endif 44283ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 44293ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 44303ee17bc7SMat Martineau #endif 4431df5042f4SFlorian Westphal }; 4432df5042f4SFlorian Westphal 4433df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4434df5042f4SFlorian Westphal { 4435df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4436df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4437df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4438df5042f4SFlorian Westphal #endif 44394165079bSFlorian Westphal #ifdef CONFIG_XFRM 44404165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 44414165079bSFlorian Westphal #endif 444295a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 444395a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 444495a7233cSPaul Blakey #endif 44453ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 44463ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 44473ee17bc7SMat Martineau #endif 4448df5042f4SFlorian Westphal 0; 4449df5042f4SFlorian Westphal } 4450df5042f4SFlorian Westphal 4451df5042f4SFlorian Westphal static void skb_extensions_init(void) 4452df5042f4SFlorian Westphal { 4453df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4454df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4455df5042f4SFlorian Westphal 4456df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4457df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4458df5042f4SFlorian Westphal 0, 4459df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4460df5042f4SFlorian Westphal NULL); 4461df5042f4SFlorian Westphal } 4462df5042f4SFlorian Westphal #else 4463df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4464df5042f4SFlorian Westphal #endif 4465df5042f4SFlorian Westphal 44661da177e4SLinus Torvalds void __init skb_init(void) 44671da177e4SLinus Torvalds { 446879a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 44691da177e4SLinus Torvalds sizeof(struct sk_buff), 44701da177e4SLinus Torvalds 0, 4471e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 447279a8a642SKees Cook offsetof(struct sk_buff, cb), 447379a8a642SKees Cook sizeof_field(struct sk_buff, cb), 447420c2df83SPaul Mundt NULL); 4475d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4476d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4477d179cd12SDavid S. Miller 0, 4478e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 447920c2df83SPaul Mundt NULL); 4480df5042f4SFlorian Westphal skb_extensions_init(); 44811da177e4SLinus Torvalds } 44821da177e4SLinus Torvalds 448351c739d1SDavid S. Miller static int 448448a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 448548a1df65SJason A. Donenfeld unsigned int recursion_level) 4486716ea3a7SDavid Howells { 44871a028e50SDavid S. Miller int start = skb_headlen(skb); 44881a028e50SDavid S. Miller int i, copy = start - offset; 4489fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4490716ea3a7SDavid Howells int elt = 0; 4491716ea3a7SDavid Howells 449248a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 449348a1df65SJason A. Donenfeld return -EMSGSIZE; 449448a1df65SJason A. Donenfeld 4495716ea3a7SDavid Howells if (copy > 0) { 4496716ea3a7SDavid Howells if (copy > len) 4497716ea3a7SDavid Howells copy = len; 4498642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4499716ea3a7SDavid Howells elt++; 4500716ea3a7SDavid Howells if ((len -= copy) == 0) 4501716ea3a7SDavid Howells return elt; 4502716ea3a7SDavid Howells offset += copy; 4503716ea3a7SDavid Howells } 4504716ea3a7SDavid Howells 4505716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 45061a028e50SDavid S. Miller int end; 4507716ea3a7SDavid Howells 4508547b792cSIlpo Järvinen WARN_ON(start > offset + len); 45091a028e50SDavid S. Miller 45109e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4511716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4512716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 451348a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 451448a1df65SJason A. Donenfeld return -EMSGSIZE; 4515716ea3a7SDavid Howells 4516716ea3a7SDavid Howells if (copy > len) 4517716ea3a7SDavid Howells copy = len; 4518ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4519b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4520716ea3a7SDavid Howells elt++; 4521716ea3a7SDavid Howells if (!(len -= copy)) 4522716ea3a7SDavid Howells return elt; 4523716ea3a7SDavid Howells offset += copy; 4524716ea3a7SDavid Howells } 45251a028e50SDavid S. Miller start = end; 4526716ea3a7SDavid Howells } 4527716ea3a7SDavid Howells 4528fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 452948a1df65SJason A. Donenfeld int end, ret; 4530716ea3a7SDavid Howells 4531547b792cSIlpo Järvinen WARN_ON(start > offset + len); 45321a028e50SDavid S. Miller 4533fbb398a8SDavid S. Miller end = start + frag_iter->len; 4534716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 453548a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 453648a1df65SJason A. Donenfeld return -EMSGSIZE; 453748a1df65SJason A. Donenfeld 4538716ea3a7SDavid Howells if (copy > len) 4539716ea3a7SDavid Howells copy = len; 454048a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 454148a1df65SJason A. Donenfeld copy, recursion_level + 1); 454248a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 454348a1df65SJason A. Donenfeld return ret; 454448a1df65SJason A. Donenfeld elt += ret; 4545716ea3a7SDavid Howells if ((len -= copy) == 0) 4546716ea3a7SDavid Howells return elt; 4547716ea3a7SDavid Howells offset += copy; 4548716ea3a7SDavid Howells } 45491a028e50SDavid S. Miller start = end; 4550716ea3a7SDavid Howells } 4551716ea3a7SDavid Howells BUG_ON(len); 4552716ea3a7SDavid Howells return elt; 4553716ea3a7SDavid Howells } 4554716ea3a7SDavid Howells 455548a1df65SJason A. Donenfeld /** 455648a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 455748a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 455848a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 455948a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 456048a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 456148a1df65SJason A. Donenfeld * 456248a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 456348a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 456448a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 456548a1df65SJason A. Donenfeld * could not fit. 456648a1df65SJason A. Donenfeld */ 456748a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 456848a1df65SJason A. Donenfeld { 456948a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 457048a1df65SJason A. Donenfeld 457148a1df65SJason A. Donenfeld if (nsg <= 0) 457248a1df65SJason A. Donenfeld return nsg; 457348a1df65SJason A. Donenfeld 457448a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 457548a1df65SJason A. Donenfeld 457648a1df65SJason A. Donenfeld return nsg; 457748a1df65SJason A. Donenfeld } 457848a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 457948a1df65SJason A. Donenfeld 458025a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 458125a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 458225a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 458325a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 458425a91d8dSFan Du * 458525a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 458625a91d8dSFan Du * 1. sg_init_table 458725a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 458825a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 458925a91d8dSFan Du * 459025a91d8dSFan Du * This is equivalent to: 459125a91d8dSFan Du * 1. sg_init_table 459225a91d8dSFan Du * 2. skb_to_sgvec(payload1) 459325a91d8dSFan Du * 3. sg_unmark_end 459425a91d8dSFan Du * 4. skb_to_sgvec(payload2) 459525a91d8dSFan Du * 459625a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 459725a91d8dSFan Du * is more preferable. 459825a91d8dSFan Du */ 459925a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 460025a91d8dSFan Du int offset, int len) 460125a91d8dSFan Du { 460248a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 460325a91d8dSFan Du } 460425a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 460525a91d8dSFan Du 460651c739d1SDavid S. Miller 460751c739d1SDavid S. Miller 4608716ea3a7SDavid Howells /** 4609716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4610716ea3a7SDavid Howells * @skb: The socket buffer to check. 4611716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4612716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4613716ea3a7SDavid Howells * 4614716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4615716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4616716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4617716ea3a7SDavid Howells * 4618716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4619716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4620716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4621716ea3a7SDavid Howells * 4622716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4623716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4624716ea3a7SDavid Howells */ 4625716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4626716ea3a7SDavid Howells { 4627716ea3a7SDavid Howells int copyflag; 4628716ea3a7SDavid Howells int elt; 4629716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4630716ea3a7SDavid Howells 4631716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4632716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4633716ea3a7SDavid Howells * at the moment even if they are anonymous). 4634716ea3a7SDavid Howells */ 4635716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4636c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4637716ea3a7SDavid Howells return -ENOMEM; 4638716ea3a7SDavid Howells 4639716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 464021dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4641716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4642716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4643716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4644716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4645716ea3a7SDavid Howells 4646716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4647716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4648716ea3a7SDavid Howells return -ENOMEM; 4649716ea3a7SDavid Howells 4650716ea3a7SDavid Howells /* Voila! */ 4651716ea3a7SDavid Howells *trailer = skb; 4652716ea3a7SDavid Howells return 1; 4653716ea3a7SDavid Howells } 4654716ea3a7SDavid Howells 4655716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4656716ea3a7SDavid Howells 4657716ea3a7SDavid Howells elt = 1; 4658716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4659716ea3a7SDavid Howells copyflag = 0; 4660716ea3a7SDavid Howells 4661716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4662716ea3a7SDavid Howells int ntail = 0; 4663716ea3a7SDavid Howells 4664716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4665716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4666716ea3a7SDavid Howells * after it. */ 4667716ea3a7SDavid Howells 4668716ea3a7SDavid Howells if (skb_shared(skb1)) 4669716ea3a7SDavid Howells copyflag = 1; 4670716ea3a7SDavid Howells 4671716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4672716ea3a7SDavid Howells 4673716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4674716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 467521dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4676716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4677716ea3a7SDavid Howells ntail = tailbits + 128; 4678716ea3a7SDavid Howells } 4679716ea3a7SDavid Howells 4680716ea3a7SDavid Howells if (copyflag || 4681716ea3a7SDavid Howells skb_cloned(skb1) || 4682716ea3a7SDavid Howells ntail || 4683716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 468421dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4685716ea3a7SDavid Howells struct sk_buff *skb2; 4686716ea3a7SDavid Howells 4687716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4688716ea3a7SDavid Howells if (ntail == 0) 4689716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4690716ea3a7SDavid Howells else 4691716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4692716ea3a7SDavid Howells skb_headroom(skb1), 4693716ea3a7SDavid Howells ntail, 4694716ea3a7SDavid Howells GFP_ATOMIC); 4695716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4696716ea3a7SDavid Howells return -ENOMEM; 4697716ea3a7SDavid Howells 4698716ea3a7SDavid Howells if (skb1->sk) 4699716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4700716ea3a7SDavid Howells 4701716ea3a7SDavid Howells /* Looking around. Are we still alive? 4702716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4703716ea3a7SDavid Howells 4704716ea3a7SDavid Howells skb2->next = skb1->next; 4705716ea3a7SDavid Howells *skb_p = skb2; 4706716ea3a7SDavid Howells kfree_skb(skb1); 4707716ea3a7SDavid Howells skb1 = skb2; 4708716ea3a7SDavid Howells } 4709716ea3a7SDavid Howells elt++; 4710716ea3a7SDavid Howells *trailer = skb1; 4711716ea3a7SDavid Howells skb_p = &skb1->next; 4712716ea3a7SDavid Howells } 4713716ea3a7SDavid Howells 4714716ea3a7SDavid Howells return elt; 4715716ea3a7SDavid Howells } 4716b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4717716ea3a7SDavid Howells 4718b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4719b1faf566SEric Dumazet { 4720b1faf566SEric Dumazet struct sock *sk = skb->sk; 4721b1faf566SEric Dumazet 4722b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4723b1faf566SEric Dumazet } 4724b1faf566SEric Dumazet 47258605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 47268605330aSSoheil Hassas Yeganeh { 47278605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 47288605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 47298605330aSSoheil Hassas Yeganeh */ 47308605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 47318605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 47328605330aSSoheil Hassas Yeganeh } 47338605330aSSoheil Hassas Yeganeh 4734b1faf566SEric Dumazet /* 4735b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4736b1faf566SEric Dumazet */ 4737b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4738b1faf566SEric Dumazet { 4739b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4740ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4741b1faf566SEric Dumazet return -ENOMEM; 4742b1faf566SEric Dumazet 4743b1faf566SEric Dumazet skb_orphan(skb); 4744b1faf566SEric Dumazet skb->sk = sk; 4745b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4746b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 47478605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4748b1faf566SEric Dumazet 4749abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4750abb57ea4SEric Dumazet skb_dst_force(skb); 4751abb57ea4SEric Dumazet 4752b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4753b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 4754e3ae2365SAlexander Aring sk_error_report(sk); 4755b1faf566SEric Dumazet return 0; 4756b1faf566SEric Dumazet } 4757b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4758b1faf566SEric Dumazet 475983a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 476083a1a1a7SSoheil Hassas Yeganeh { 476183a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 476283a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 476383a1a1a7SSoheil Hassas Yeganeh } 476483a1a1a7SSoheil Hassas Yeganeh 4765364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4766364a9e93SWillem de Bruijn { 4767364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 476883a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 476983a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4770997d5c3fSEric Dumazet unsigned long flags; 4771364a9e93SWillem de Bruijn 4772997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4773364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 477438b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 477583a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 477638b25793SSoheil Hassas Yeganeh if (icmp_next) 4777985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 477838b25793SSoheil Hassas Yeganeh } 4779997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4780364a9e93SWillem de Bruijn 478183a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 478283a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 478383a1a1a7SSoheil Hassas Yeganeh 478483a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4785e3ae2365SAlexander Aring sk_error_report(sk); 4786364a9e93SWillem de Bruijn 4787364a9e93SWillem de Bruijn return skb; 4788364a9e93SWillem de Bruijn } 4789364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4790364a9e93SWillem de Bruijn 4791cab41c47SAlexander Duyck /** 4792cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4793cab41c47SAlexander Duyck * @skb: the skb to clone 4794cab41c47SAlexander Duyck * 4795cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4796cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4797cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4798cab41c47SAlexander Duyck * 4799cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4800cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4801cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4802cab41c47SAlexander Duyck * the sk_error_queue. 4803cab41c47SAlexander Duyck */ 480462bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 480562bccb8cSAlexander Duyck { 480662bccb8cSAlexander Duyck struct sock *sk = skb->sk; 480762bccb8cSAlexander Duyck struct sk_buff *clone; 480862bccb8cSAlexander Duyck 480941c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 481062bccb8cSAlexander Duyck return NULL; 481162bccb8cSAlexander Duyck 481262bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 481362bccb8cSAlexander Duyck if (!clone) { 481462bccb8cSAlexander Duyck sock_put(sk); 481562bccb8cSAlexander Duyck return NULL; 481662bccb8cSAlexander Duyck } 481762bccb8cSAlexander Duyck 481862bccb8cSAlexander Duyck clone->sk = sk; 481962bccb8cSAlexander Duyck clone->destructor = sock_efree; 482062bccb8cSAlexander Duyck 482162bccb8cSAlexander Duyck return clone; 482262bccb8cSAlexander Duyck } 482362bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 482462bccb8cSAlexander Duyck 482537846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 482637846ef0SAlexander Duyck struct sock *sk, 48274ef1b286SSoheil Hassas Yeganeh int tstype, 48284ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4829ac45f602SPatrick Ohly { 4830ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4831ac45f602SPatrick Ohly int err; 4832ac45f602SPatrick Ohly 48334ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 48344ef1b286SSoheil Hassas Yeganeh 4835ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4836ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4837ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4838ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4839e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 48404ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 48411862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 48424ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 484309c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 4844ac5cc977SWANG Cong if (sk->sk_protocol == IPPROTO_TCP && 4845ac5cc977SWANG Cong sk->sk_type == SOCK_STREAM) 48464ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 48474ed2d765SWillem de Bruijn } 484829030374SEric Dumazet 4849ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 485029030374SEric Dumazet 4851ac45f602SPatrick Ohly if (err) 4852ac45f602SPatrick Ohly kfree_skb(skb); 4853ac45f602SPatrick Ohly } 485437846ef0SAlexander Duyck 4855b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4856b245be1fSWillem de Bruijn { 4857b245be1fSWillem de Bruijn bool ret; 4858b245be1fSWillem de Bruijn 4859b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4860b245be1fSWillem de Bruijn return true; 4861b245be1fSWillem de Bruijn 4862b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4863b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4864b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4865b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4866b245be1fSWillem de Bruijn return ret; 4867b245be1fSWillem de Bruijn } 4868b245be1fSWillem de Bruijn 486937846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 487037846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 487137846ef0SAlexander Duyck { 487237846ef0SAlexander Duyck struct sock *sk = skb->sk; 487337846ef0SAlexander Duyck 4874b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 487535b99dffSWillem de Bruijn goto err; 4876b245be1fSWillem de Bruijn 48779ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 48789ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 48799ac25fc0SEric Dumazet */ 488041c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 488137846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 48824ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 488337846ef0SAlexander Duyck sock_put(sk); 488435b99dffSWillem de Bruijn return; 488537846ef0SAlexander Duyck } 488635b99dffSWillem de Bruijn 488735b99dffSWillem de Bruijn err: 488835b99dffSWillem de Bruijn kfree_skb(skb); 48899ac25fc0SEric Dumazet } 489037846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 489137846ef0SAlexander Duyck 489237846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 4893e7ed11eeSYousuk Seung const struct sk_buff *ack_skb, 489437846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 489537846ef0SAlexander Duyck struct sock *sk, int tstype) 489637846ef0SAlexander Duyck { 489737846ef0SAlexander Duyck struct sk_buff *skb; 48984ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 489937846ef0SAlexander Duyck 49003a8dd971SWillem de Bruijn if (!sk) 49013a8dd971SWillem de Bruijn return; 49023a8dd971SWillem de Bruijn 4903b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4904b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4905b50a5c70SMiroslav Lichvar return; 4906b50a5c70SMiroslav Lichvar 49073a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 49083a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 490937846ef0SAlexander Duyck return; 491037846ef0SAlexander Duyck 49111c885808SFrancis Yan if (tsonly) { 49121c885808SFrancis Yan #ifdef CONFIG_INET 49131c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 49141c885808SFrancis Yan sk->sk_protocol == IPPROTO_TCP && 49154ef1b286SSoheil Hassas Yeganeh sk->sk_type == SOCK_STREAM) { 4916e7ed11eeSYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb, 4917e7ed11eeSYousuk Seung ack_skb); 49184ef1b286SSoheil Hassas Yeganeh opt_stats = true; 49194ef1b286SSoheil Hassas Yeganeh } else 49201c885808SFrancis Yan #endif 49211c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 49221c885808SFrancis Yan } else { 492337846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 49241c885808SFrancis Yan } 492537846ef0SAlexander Duyck if (!skb) 492637846ef0SAlexander Duyck return; 492737846ef0SAlexander Duyck 492849ca0d8bSWillem de Bruijn if (tsonly) { 4929fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4930fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 493149ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 493249ca0d8bSWillem de Bruijn } 493349ca0d8bSWillem de Bruijn 493449ca0d8bSWillem de Bruijn if (hwtstamps) 493549ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 493649ca0d8bSWillem de Bruijn else 493749ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 493849ca0d8bSWillem de Bruijn 49394ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 494037846ef0SAlexander Duyck } 4941e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4942e7fd2885SWillem de Bruijn 4943e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4944e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4945e7fd2885SWillem de Bruijn { 4946e7ed11eeSYousuk Seung return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk, 4947e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4948e7fd2885SWillem de Bruijn } 4949ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4950ac45f602SPatrick Ohly 49516e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 49526e3e939fSJohannes Berg { 49536e3e939fSJohannes Berg struct sock *sk = skb->sk; 49546e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4955dd4f1072SEric Dumazet int err = 1; 49566e3e939fSJohannes Berg 49576e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 49586e3e939fSJohannes Berg skb->wifi_acked = acked; 49596e3e939fSJohannes Berg 49606e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 49616e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 49626e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 49636e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 49646e3e939fSJohannes Berg 4965dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4966dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4967dd4f1072SEric Dumazet */ 496841c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 49696e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4970dd4f1072SEric Dumazet sock_put(sk); 4971dd4f1072SEric Dumazet } 49726e3e939fSJohannes Berg if (err) 49736e3e939fSJohannes Berg kfree_skb(skb); 49746e3e939fSJohannes Berg } 49756e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 49766e3e939fSJohannes Berg 4977f35d9d8aSRusty Russell /** 4978f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4979f35d9d8aSRusty Russell * @skb: the skb to set 4980f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4981f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4982f35d9d8aSRusty Russell * 4983f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4984f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4985f35d9d8aSRusty Russell * 4986f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4987f35d9d8aSRusty Russell * returns false you should drop the packet. 4988f35d9d8aSRusty Russell */ 4989f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4990f35d9d8aSRusty Russell { 499152b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 499252b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 499352b5d6f5SEric Dumazet 499452b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 499552b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 499652b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4997f35d9d8aSRusty Russell return false; 4998f35d9d8aSRusty Russell } 4999f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 500052b5d6f5SEric Dumazet skb->csum_start = csum_start; 5001f35d9d8aSRusty Russell skb->csum_offset = off; 5002e5d5decaSJason Wang skb_set_transport_header(skb, start); 5003f35d9d8aSRusty Russell return true; 5004f35d9d8aSRusty Russell } 5005b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 5006f35d9d8aSRusty Russell 5007ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 5008ed1f50c3SPaul Durrant unsigned int max) 5009ed1f50c3SPaul Durrant { 5010ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 5011ed1f50c3SPaul Durrant return 0; 5012ed1f50c3SPaul Durrant 5013ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 5014ed1f50c3SPaul Durrant * won't need to do it again. 5015ed1f50c3SPaul Durrant */ 5016ed1f50c3SPaul Durrant if (max > skb->len) 5017ed1f50c3SPaul Durrant max = skb->len; 5018ed1f50c3SPaul Durrant 5019ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 5020ed1f50c3SPaul Durrant return -ENOMEM; 5021ed1f50c3SPaul Durrant 5022ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 5023ed1f50c3SPaul Durrant return -EPROTO; 5024ed1f50c3SPaul Durrant 5025ed1f50c3SPaul Durrant return 0; 5026ed1f50c3SPaul Durrant } 5027ed1f50c3SPaul Durrant 5028f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 5029f9708b43SJan Beulich 5030f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 5031f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 5032f9708b43SJan Beulich unsigned int off) 5033f9708b43SJan Beulich { 5034f9708b43SJan Beulich int err; 5035f9708b43SJan Beulich 5036161d1792SKees Cook switch (proto) { 5037f9708b43SJan Beulich case IPPROTO_TCP: 5038f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 5039f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 5040f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 5041f9708b43SJan Beulich offsetof(struct tcphdr, 5042f9708b43SJan Beulich check))) 5043f9708b43SJan Beulich err = -EPROTO; 5044f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 5045f9708b43SJan Beulich 5046f9708b43SJan Beulich case IPPROTO_UDP: 5047f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 5048f9708b43SJan Beulich off + sizeof(struct udphdr)); 5049f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 5050f9708b43SJan Beulich offsetof(struct udphdr, 5051f9708b43SJan Beulich check))) 5052f9708b43SJan Beulich err = -EPROTO; 5053f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 5054f9708b43SJan Beulich } 5055f9708b43SJan Beulich 5056f9708b43SJan Beulich return ERR_PTR(-EPROTO); 5057f9708b43SJan Beulich } 5058f9708b43SJan Beulich 5059ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5060ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 5061ed1f50c3SPaul Durrant */ 5062ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 5063ed1f50c3SPaul Durrant 5064f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 5065ed1f50c3SPaul Durrant { 5066ed1f50c3SPaul Durrant unsigned int off; 5067ed1f50c3SPaul Durrant bool fragment; 5068f9708b43SJan Beulich __sum16 *csum; 5069ed1f50c3SPaul Durrant int err; 5070ed1f50c3SPaul Durrant 5071ed1f50c3SPaul Durrant fragment = false; 5072ed1f50c3SPaul Durrant 5073ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5074ed1f50c3SPaul Durrant sizeof(struct iphdr), 5075ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 5076ed1f50c3SPaul Durrant if (err < 0) 5077ed1f50c3SPaul Durrant goto out; 5078ed1f50c3SPaul Durrant 507911f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 5080ed1f50c3SPaul Durrant fragment = true; 5081ed1f50c3SPaul Durrant 5082ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 5083ed1f50c3SPaul Durrant 5084ed1f50c3SPaul Durrant err = -EPROTO; 5085ed1f50c3SPaul Durrant 5086ed1f50c3SPaul Durrant if (fragment) 5087ed1f50c3SPaul Durrant goto out; 5088ed1f50c3SPaul Durrant 5089f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 5090f9708b43SJan Beulich if (IS_ERR(csum)) 5091f9708b43SJan Beulich return PTR_ERR(csum); 5092ed1f50c3SPaul Durrant 5093ed1f50c3SPaul Durrant if (recalculate) 5094f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 5095ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 5096ed1f50c3SPaul Durrant skb->len - off, 5097f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 5098ed1f50c3SPaul Durrant err = 0; 5099ed1f50c3SPaul Durrant 5100ed1f50c3SPaul Durrant out: 5101ed1f50c3SPaul Durrant return err; 5102ed1f50c3SPaul Durrant } 5103ed1f50c3SPaul Durrant 5104ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5105ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 5106ed1f50c3SPaul Durrant */ 5107ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 5108ed1f50c3SPaul Durrant 5109ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 5110ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 5111ed1f50c3SPaul Durrant 5112ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 5113ed1f50c3SPaul Durrant { 5114ed1f50c3SPaul Durrant int err; 5115ed1f50c3SPaul Durrant u8 nexthdr; 5116ed1f50c3SPaul Durrant unsigned int off; 5117ed1f50c3SPaul Durrant unsigned int len; 5118ed1f50c3SPaul Durrant bool fragment; 5119ed1f50c3SPaul Durrant bool done; 5120f9708b43SJan Beulich __sum16 *csum; 5121ed1f50c3SPaul Durrant 5122ed1f50c3SPaul Durrant fragment = false; 5123ed1f50c3SPaul Durrant done = false; 5124ed1f50c3SPaul Durrant 5125ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 5126ed1f50c3SPaul Durrant 5127ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 5128ed1f50c3SPaul Durrant if (err < 0) 5129ed1f50c3SPaul Durrant goto out; 5130ed1f50c3SPaul Durrant 5131ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 5132ed1f50c3SPaul Durrant 5133ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 5134ed1f50c3SPaul Durrant while (off <= len && !done) { 5135ed1f50c3SPaul Durrant switch (nexthdr) { 5136ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 5137ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 5138ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 5139ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 5140ed1f50c3SPaul Durrant 5141ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5142ed1f50c3SPaul Durrant off + 5143ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 5144ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5145ed1f50c3SPaul Durrant if (err < 0) 5146ed1f50c3SPaul Durrant goto out; 5147ed1f50c3SPaul Durrant 5148ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 5149ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5150ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 5151ed1f50c3SPaul Durrant break; 5152ed1f50c3SPaul Durrant } 5153ed1f50c3SPaul Durrant case IPPROTO_AH: { 5154ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 5155ed1f50c3SPaul Durrant 5156ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5157ed1f50c3SPaul Durrant off + 5158ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 5159ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5160ed1f50c3SPaul Durrant if (err < 0) 5161ed1f50c3SPaul Durrant goto out; 5162ed1f50c3SPaul Durrant 5163ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 5164ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5165ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 5166ed1f50c3SPaul Durrant break; 5167ed1f50c3SPaul Durrant } 5168ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 5169ed1f50c3SPaul Durrant struct frag_hdr *hp; 5170ed1f50c3SPaul Durrant 5171ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5172ed1f50c3SPaul Durrant off + 5173ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5174ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5175ed1f50c3SPaul Durrant if (err < 0) 5176ed1f50c3SPaul Durrant goto out; 5177ed1f50c3SPaul Durrant 5178ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5179ed1f50c3SPaul Durrant 5180ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5181ed1f50c3SPaul Durrant fragment = true; 5182ed1f50c3SPaul Durrant 5183ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5184ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5185ed1f50c3SPaul Durrant break; 5186ed1f50c3SPaul Durrant } 5187ed1f50c3SPaul Durrant default: 5188ed1f50c3SPaul Durrant done = true; 5189ed1f50c3SPaul Durrant break; 5190ed1f50c3SPaul Durrant } 5191ed1f50c3SPaul Durrant } 5192ed1f50c3SPaul Durrant 5193ed1f50c3SPaul Durrant err = -EPROTO; 5194ed1f50c3SPaul Durrant 5195ed1f50c3SPaul Durrant if (!done || fragment) 5196ed1f50c3SPaul Durrant goto out; 5197ed1f50c3SPaul Durrant 5198f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5199f9708b43SJan Beulich if (IS_ERR(csum)) 5200f9708b43SJan Beulich return PTR_ERR(csum); 5201ed1f50c3SPaul Durrant 5202ed1f50c3SPaul Durrant if (recalculate) 5203f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5204ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5205f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5206ed1f50c3SPaul Durrant err = 0; 5207ed1f50c3SPaul Durrant 5208ed1f50c3SPaul Durrant out: 5209ed1f50c3SPaul Durrant return err; 5210ed1f50c3SPaul Durrant } 5211ed1f50c3SPaul Durrant 5212ed1f50c3SPaul Durrant /** 5213ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5214ed1f50c3SPaul Durrant * @skb: the skb to set up 5215ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5216ed1f50c3SPaul Durrant */ 5217ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5218ed1f50c3SPaul Durrant { 5219ed1f50c3SPaul Durrant int err; 5220ed1f50c3SPaul Durrant 5221ed1f50c3SPaul Durrant switch (skb->protocol) { 5222ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5223f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5224ed1f50c3SPaul Durrant break; 5225ed1f50c3SPaul Durrant 5226ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5227ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5228ed1f50c3SPaul Durrant break; 5229ed1f50c3SPaul Durrant 5230ed1f50c3SPaul Durrant default: 5231ed1f50c3SPaul Durrant err = -EPROTO; 5232ed1f50c3SPaul Durrant break; 5233ed1f50c3SPaul Durrant } 5234ed1f50c3SPaul Durrant 5235ed1f50c3SPaul Durrant return err; 5236ed1f50c3SPaul Durrant } 5237ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5238ed1f50c3SPaul Durrant 52399afd85c9SLinus Lüssing /** 52409afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 52419afd85c9SLinus Lüssing * @skb: the skb to check 52429afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 52439afd85c9SLinus Lüssing * 52449afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 52459afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 52469afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 52479afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 52489afd85c9SLinus Lüssing * 5249a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5250a516993fSLinus Lüssing * differs from the provided skb. 52519afd85c9SLinus Lüssing */ 52529afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 52539afd85c9SLinus Lüssing unsigned int transport_len) 52549afd85c9SLinus Lüssing { 52559afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52569afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 52579afd85c9SLinus Lüssing int ret; 52589afd85c9SLinus Lüssing 5259a516993fSLinus Lüssing if (skb->len < len) 52609afd85c9SLinus Lüssing return NULL; 5261a516993fSLinus Lüssing else if (skb->len == len) 52629afd85c9SLinus Lüssing return skb; 52639afd85c9SLinus Lüssing 52649afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 52659afd85c9SLinus Lüssing if (!skb_chk) 52669afd85c9SLinus Lüssing return NULL; 52679afd85c9SLinus Lüssing 52689afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 52699afd85c9SLinus Lüssing if (ret) { 52709afd85c9SLinus Lüssing kfree_skb(skb_chk); 52719afd85c9SLinus Lüssing return NULL; 52729afd85c9SLinus Lüssing } 52739afd85c9SLinus Lüssing 52749afd85c9SLinus Lüssing return skb_chk; 52759afd85c9SLinus Lüssing } 52769afd85c9SLinus Lüssing 52779afd85c9SLinus Lüssing /** 52789afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 52799afd85c9SLinus Lüssing * @skb: the skb to check 52809afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 52819afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 52829afd85c9SLinus Lüssing * 52839afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 52849afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 52859afd85c9SLinus Lüssing * 52869afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 52879afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 52889afd85c9SLinus Lüssing * 5289a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5290a516993fSLinus Lüssing * differs from the provided skb. 52919afd85c9SLinus Lüssing */ 52929afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 52939afd85c9SLinus Lüssing unsigned int transport_len, 52949afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 52959afd85c9SLinus Lüssing { 52969afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52979afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5298fcba67c9SLinus Lüssing __sum16 ret; 52999afd85c9SLinus Lüssing 53009afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 53019afd85c9SLinus Lüssing if (!skb_chk) 5302a516993fSLinus Lüssing goto err; 53039afd85c9SLinus Lüssing 5304a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5305a516993fSLinus Lüssing goto err; 53069afd85c9SLinus Lüssing 53079b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 53089afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 53099b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 53109afd85c9SLinus Lüssing 5311a516993fSLinus Lüssing if (ret) 5312a516993fSLinus Lüssing goto err; 53139afd85c9SLinus Lüssing 53149afd85c9SLinus Lüssing return skb_chk; 5315a516993fSLinus Lüssing 5316a516993fSLinus Lüssing err: 5317a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5318a516993fSLinus Lüssing kfree_skb(skb_chk); 5319a516993fSLinus Lüssing 5320a516993fSLinus Lüssing return NULL; 5321a516993fSLinus Lüssing 53229afd85c9SLinus Lüssing } 53239afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 53249afd85c9SLinus Lüssing 53254497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 53264497b076SBen Hutchings { 5327e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5328e87cc472SJoe Perches skb->dev->name); 53294497b076SBen Hutchings } 53304497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5331bad43ca8SEric Dumazet 5332bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5333bad43ca8SEric Dumazet { 53343d861f66SEric Dumazet if (head_stolen) { 53353d861f66SEric Dumazet skb_release_head_state(skb); 5336bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 53373d861f66SEric Dumazet } else { 5338bad43ca8SEric Dumazet __kfree_skb(skb); 5339bad43ca8SEric Dumazet } 53403d861f66SEric Dumazet } 5341bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5342bad43ca8SEric Dumazet 5343bad43ca8SEric Dumazet /** 5344bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5345bad43ca8SEric Dumazet * @to: prior buffer 5346bad43ca8SEric Dumazet * @from: buffer to add 5347bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5348c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5349bad43ca8SEric Dumazet */ 5350bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5351bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5352bad43ca8SEric Dumazet { 5353c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5354bad43ca8SEric Dumazet int i, delta, len = from->len; 5355bad43ca8SEric Dumazet 5356bad43ca8SEric Dumazet *fragstolen = false; 5357bad43ca8SEric Dumazet 5358bad43ca8SEric Dumazet if (skb_cloned(to)) 5359bad43ca8SEric Dumazet return false; 5360bad43ca8SEric Dumazet 53616a5bcd84SIlias Apalodimas /* The page pool signature of struct page will eventually figure out 53626a5bcd84SIlias Apalodimas * which pages can be recycled or not but for now let's prohibit slab 53636a5bcd84SIlias Apalodimas * allocated and page_pool allocated SKBs from being coalesced. 53646a5bcd84SIlias Apalodimas */ 53656a5bcd84SIlias Apalodimas if (to->pp_recycle != from->pp_recycle) 53666a5bcd84SIlias Apalodimas return false; 53676a5bcd84SIlias Apalodimas 5368bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5369e93a0435SEric Dumazet if (len) 5370bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5371bad43ca8SEric Dumazet *delta_truesize = 0; 5372bad43ca8SEric Dumazet return true; 5373bad43ca8SEric Dumazet } 5374bad43ca8SEric Dumazet 5375c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5376c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5377c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5378bad43ca8SEric Dumazet return false; 53791f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 53801f8b977aSWillem de Bruijn return false; 5381bad43ca8SEric Dumazet 5382bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5383bad43ca8SEric Dumazet struct page *page; 5384bad43ca8SEric Dumazet unsigned int offset; 5385bad43ca8SEric Dumazet 5386c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5387c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5388bad43ca8SEric Dumazet return false; 5389bad43ca8SEric Dumazet 5390bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5391bad43ca8SEric Dumazet return false; 5392bad43ca8SEric Dumazet 5393bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5394bad43ca8SEric Dumazet 5395bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5396bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5397bad43ca8SEric Dumazet 5398c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5399bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5400bad43ca8SEric Dumazet *fragstolen = true; 5401bad43ca8SEric Dumazet } else { 5402c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5403c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5404bad43ca8SEric Dumazet return false; 5405bad43ca8SEric Dumazet 5406f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5407bad43ca8SEric Dumazet } 5408bad43ca8SEric Dumazet 5409bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5410bad43ca8SEric Dumazet 5411c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5412c818fa9eSEric Dumazet from_shinfo->frags, 5413c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5414c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5415bad43ca8SEric Dumazet 5416bad43ca8SEric Dumazet if (!skb_cloned(from)) 5417c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5418bad43ca8SEric Dumazet 54198ea853fdSLi RongQing /* if the skb is not cloned this does nothing 54208ea853fdSLi RongQing * since we set nr_frags to 0. 54218ea853fdSLi RongQing */ 5422c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5423c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5424bad43ca8SEric Dumazet 5425bad43ca8SEric Dumazet to->truesize += delta; 5426bad43ca8SEric Dumazet to->len += len; 5427bad43ca8SEric Dumazet to->data_len += len; 5428bad43ca8SEric Dumazet 5429bad43ca8SEric Dumazet *delta_truesize = delta; 5430bad43ca8SEric Dumazet return true; 5431bad43ca8SEric Dumazet } 5432bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5433621e84d6SNicolas Dichtel 5434621e84d6SNicolas Dichtel /** 54358b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5436621e84d6SNicolas Dichtel * 5437621e84d6SNicolas Dichtel * @skb: buffer to clean 54388b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5439621e84d6SNicolas Dichtel * 54408b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 54418b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 54428b27f277SNicolas Dichtel * operations. 54438b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 54448b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 54458b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5446621e84d6SNicolas Dichtel */ 54478b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5448621e84d6SNicolas Dichtel { 5449621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5450621e84d6SNicolas Dichtel skb->skb_iif = 0; 545160ff7467SWANG Cong skb->ignore_df = 0; 5452621e84d6SNicolas Dichtel skb_dst_drop(skb); 5453174e2381SFlorian Westphal skb_ext_reset(skb); 5454895b5c9fSFlorian Westphal nf_reset_ct(skb); 5455621e84d6SNicolas Dichtel nf_reset_trace(skb); 5456213dd74aSHerbert Xu 54576f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 54586f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5459875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 54606f9a5069SPetr Machata #endif 54616f9a5069SPetr Machata 5462213dd74aSHerbert Xu if (!xnet) 5463213dd74aSHerbert Xu return; 5464213dd74aSHerbert Xu 54652b5ec1a5SYe Yin ipvs_reset(skb); 5466213dd74aSHerbert Xu skb->mark = 0; 5467c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 5468621e84d6SNicolas Dichtel } 5469621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5470de960aa9SFlorian Westphal 5471de960aa9SFlorian Westphal /** 5472de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5473de960aa9SFlorian Westphal * 5474de960aa9SFlorian Westphal * @skb: GSO skb 5475de960aa9SFlorian Westphal * 5476de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5477de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5478de960aa9SFlorian Westphal * 5479de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5480de960aa9SFlorian Westphal */ 5481a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5482de960aa9SFlorian Westphal { 5483de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5484f993bc25SFlorian Westphal unsigned int thlen = 0; 5485f993bc25SFlorian Westphal 5486f993bc25SFlorian Westphal if (skb->encapsulation) { 5487f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5488f993bc25SFlorian Westphal skb_transport_header(skb); 5489de960aa9SFlorian Westphal 5490de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5491f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5492f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5493f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 54941dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 549590017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5496ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5497ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5498f993bc25SFlorian Westphal } 54996d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 55006d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 55016d39d589SFlorian Westphal * accounted for. 55026d39d589SFlorian Westphal */ 5503f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5504de960aa9SFlorian Westphal } 5505a4a77718SDaniel Axtens 5506a4a77718SDaniel Axtens /** 5507a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5508a4a77718SDaniel Axtens * 5509a4a77718SDaniel Axtens * @skb: GSO skb 5510a4a77718SDaniel Axtens * 5511a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5512a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5513a4a77718SDaniel Axtens * 5514a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5515a4a77718SDaniel Axtens */ 5516a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5517a4a77718SDaniel Axtens { 5518a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5519a4a77718SDaniel Axtens skb_network_header(skb); 5520a4a77718SDaniel Axtens 5521a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5522a4a77718SDaniel Axtens } 5523a4a77718SDaniel Axtens 5524a4a77718SDaniel Axtens /** 5525a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5526a4a77718SDaniel Axtens * 5527a4a77718SDaniel Axtens * @skb: GSO skb 5528a4a77718SDaniel Axtens * 5529a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5530a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5531a4a77718SDaniel Axtens * headers (TCP/UDP). 5532a4a77718SDaniel Axtens */ 5533a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5534a4a77718SDaniel Axtens { 5535a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5536a4a77718SDaniel Axtens 5537a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5538a4a77718SDaniel Axtens } 55390d5501c1SVlad Yasevich 5540ae7ef81eSMarcelo Ricardo Leitner /** 55412b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 55422b16f048SDaniel Axtens * 55432b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 55442b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 55452b16f048SDaniel Axtens * 55462b16f048SDaniel Axtens * We might want to check: 55472b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 55482b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 55492b16f048SDaniel Axtens * 55502b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 55512b16f048SDaniel Axtens * 555249682bfaSMathieu Malaterre * @skb: GSO skb 555349682bfaSMathieu Malaterre * 55542b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 55552b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 55562b16f048SDaniel Axtens * 55572b16f048SDaniel Axtens * @max_len: The maximum permissible length. 55582b16f048SDaniel Axtens * 55592b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 55602b16f048SDaniel Axtens */ 55612b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 55622b16f048SDaniel Axtens unsigned int seg_len, 55632b16f048SDaniel Axtens unsigned int max_len) { 55642b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 55652b16f048SDaniel Axtens const struct sk_buff *iter; 55662b16f048SDaniel Axtens 55672b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 55682b16f048SDaniel Axtens return seg_len <= max_len; 55692b16f048SDaniel Axtens 55702b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 55712b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 55722b16f048SDaniel Axtens 55732b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 55742b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 55752b16f048SDaniel Axtens return false; 55762b16f048SDaniel Axtens } 55772b16f048SDaniel Axtens 55782b16f048SDaniel Axtens return true; 55792b16f048SDaniel Axtens } 55802b16f048SDaniel Axtens 55812b16f048SDaniel Axtens /** 5582779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5583ae7ef81eSMarcelo Ricardo Leitner * 5584ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 558576f21b99SDavid S. Miller * @mtu: MTU to validate against 5586ae7ef81eSMarcelo Ricardo Leitner * 5587779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5588779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5589779b7931SDaniel Axtens * payload. 5590ae7ef81eSMarcelo Ricardo Leitner */ 5591779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5592ae7ef81eSMarcelo Ricardo Leitner { 55932b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5594ae7ef81eSMarcelo Ricardo Leitner } 5595779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5596ae7ef81eSMarcelo Ricardo Leitner 55972b16f048SDaniel Axtens /** 55982b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 55992b16f048SDaniel Axtens * 56002b16f048SDaniel Axtens * @skb: GSO skb 56012b16f048SDaniel Axtens * @len: length to validate against 56022b16f048SDaniel Axtens * 56032b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 56042b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 56052b16f048SDaniel Axtens */ 56062b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 56072b16f048SDaniel Axtens { 56082b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 56092b16f048SDaniel Axtens } 56102b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 56112b16f048SDaniel Axtens 56120d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 56130d5501c1SVlad Yasevich { 5614d85e8be2SYuya Kusakabe int mac_len, meta_len; 5615d85e8be2SYuya Kusakabe void *meta; 56164bbb3e0eSToshiaki Makita 56170d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 56180d5501c1SVlad Yasevich kfree_skb(skb); 56190d5501c1SVlad Yasevich return NULL; 56200d5501c1SVlad Yasevich } 56210d5501c1SVlad Yasevich 56224bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5623ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 56244bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 56254bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5626ae474573SToshiaki Makita } 5627d85e8be2SYuya Kusakabe 5628d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5629d85e8be2SYuya Kusakabe if (meta_len) { 5630d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5631d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5632d85e8be2SYuya Kusakabe } 5633d85e8be2SYuya Kusakabe 56340d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 56350d5501c1SVlad Yasevich return skb; 56360d5501c1SVlad Yasevich } 56370d5501c1SVlad Yasevich 56380d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 56390d5501c1SVlad Yasevich { 56400d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 56410d5501c1SVlad Yasevich u16 vlan_tci; 56420d5501c1SVlad Yasevich 5643df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 56440d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 56450d5501c1SVlad Yasevich return skb; 56460d5501c1SVlad Yasevich } 56470d5501c1SVlad Yasevich 56480d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 56490d5501c1SVlad Yasevich if (unlikely(!skb)) 56500d5501c1SVlad Yasevich goto err_free; 565155eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 565255eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 56530d5501c1SVlad Yasevich goto err_free; 56540d5501c1SVlad Yasevich 56550d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 56560d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 56570d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 56580d5501c1SVlad Yasevich 56590d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 56600d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 56610d5501c1SVlad Yasevich 56620d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 56630d5501c1SVlad Yasevich if (unlikely(!skb)) 56640d5501c1SVlad Yasevich goto err_free; 56650d5501c1SVlad Yasevich 56660d5501c1SVlad Yasevich skb_reset_network_header(skb); 56678be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 56680d5501c1SVlad Yasevich skb_reset_transport_header(skb); 56690d5501c1SVlad Yasevich skb_reset_mac_len(skb); 56700d5501c1SVlad Yasevich 56710d5501c1SVlad Yasevich return skb; 56720d5501c1SVlad Yasevich 56730d5501c1SVlad Yasevich err_free: 56740d5501c1SVlad Yasevich kfree_skb(skb); 56750d5501c1SVlad Yasevich return NULL; 56760d5501c1SVlad Yasevich } 56770d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 56782e4e4410SEric Dumazet 5679e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5680e2195121SJiri Pirko { 5681e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5682e2195121SJiri Pirko return -ENOMEM; 5683e2195121SJiri Pirko 5684e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5685e2195121SJiri Pirko return 0; 5686e2195121SJiri Pirko 5687e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5688e2195121SJiri Pirko } 5689e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5690e2195121SJiri Pirko 5691bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5692bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5693bfca4c52SShmulik Ladkani */ 5694bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 569593515d53SJiri Pirko { 569693515d53SJiri Pirko struct vlan_hdr *vhdr; 5697b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 569893515d53SJiri Pirko int err; 569993515d53SJiri Pirko 5700b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5701b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5702b6a79208SShmulik Ladkani offset)) { 5703b6a79208SShmulik Ladkani return -EINVAL; 5704b6a79208SShmulik Ladkani } 5705b6a79208SShmulik Ladkani 570693515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 570793515d53SJiri Pirko if (unlikely(err)) 5708b6a79208SShmulik Ladkani return err; 570993515d53SJiri Pirko 571093515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 571193515d53SJiri Pirko 571293515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 571393515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 571493515d53SJiri Pirko 571593515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 571693515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 571793515d53SJiri Pirko 571893515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 571993515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 572093515d53SJiri Pirko 572193515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 572293515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 572393515d53SJiri Pirko 572493515d53SJiri Pirko skb_reset_mac_len(skb); 572593515d53SJiri Pirko 572693515d53SJiri Pirko return err; 572793515d53SJiri Pirko } 5728bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 572993515d53SJiri Pirko 5730b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5731b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5732b6a79208SShmulik Ladkani */ 573393515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 573493515d53SJiri Pirko { 573593515d53SJiri Pirko u16 vlan_tci; 573693515d53SJiri Pirko __be16 vlan_proto; 573793515d53SJiri Pirko int err; 573893515d53SJiri Pirko 5739df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5740b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 574193515d53SJiri Pirko } else { 5742ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 574393515d53SJiri Pirko return 0; 574493515d53SJiri Pirko 574593515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 574693515d53SJiri Pirko if (err) 574793515d53SJiri Pirko return err; 574893515d53SJiri Pirko } 574993515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5750ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 575193515d53SJiri Pirko return 0; 575293515d53SJiri Pirko 575393515d53SJiri Pirko vlan_proto = skb->protocol; 575493515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 575593515d53SJiri Pirko if (unlikely(err)) 575693515d53SJiri Pirko return err; 575793515d53SJiri Pirko 575893515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 575993515d53SJiri Pirko return 0; 576093515d53SJiri Pirko } 576193515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 576293515d53SJiri Pirko 5763b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5764b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5765b6a79208SShmulik Ladkani */ 576693515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 576793515d53SJiri Pirko { 5768df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5769b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 577093515d53SJiri Pirko int err; 577193515d53SJiri Pirko 5772b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5773b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5774b6a79208SShmulik Ladkani offset)) { 5775b6a79208SShmulik Ladkani return -EINVAL; 5776b6a79208SShmulik Ladkani } 5777b6a79208SShmulik Ladkani 577893515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5779df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5780b6a79208SShmulik Ladkani if (err) 578193515d53SJiri Pirko return err; 57829241e2dfSDaniel Borkmann 578393515d53SJiri Pirko skb->protocol = skb->vlan_proto; 578493515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 578593515d53SJiri Pirko 57866b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 578793515d53SJiri Pirko } 578893515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 578993515d53SJiri Pirko return 0; 579093515d53SJiri Pirko } 579193515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 579293515d53SJiri Pirko 579319fbcb36SGuillaume Nault /** 579419fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 579519fbcb36SGuillaume Nault * 579619fbcb36SGuillaume Nault * @skb: Socket buffer to modify 579719fbcb36SGuillaume Nault * 579819fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 579919fbcb36SGuillaume Nault * 580019fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 580119fbcb36SGuillaume Nault * present. 580219fbcb36SGuillaume Nault * 580319fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 580419fbcb36SGuillaume Nault */ 580519fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 580619fbcb36SGuillaume Nault { 580719fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 580819fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 580919fbcb36SGuillaume Nault return -EPROTO; 581019fbcb36SGuillaume Nault 581119fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 581219fbcb36SGuillaume Nault skb_reset_mac_header(skb); 581319fbcb36SGuillaume Nault skb_reset_mac_len(skb); 581419fbcb36SGuillaume Nault 581519fbcb36SGuillaume Nault return 0; 581619fbcb36SGuillaume Nault } 581719fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 581819fbcb36SGuillaume Nault 581919fbcb36SGuillaume Nault /** 582019fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 582119fbcb36SGuillaume Nault * 582219fbcb36SGuillaume Nault * @skb: Socket buffer to modify 582319fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 582419fbcb36SGuillaume Nault * @src: Source MAC address of the new header 582519fbcb36SGuillaume Nault * 582619fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 582719fbcb36SGuillaume Nault * 582819fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 582919fbcb36SGuillaume Nault * 583019fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 583119fbcb36SGuillaume Nault */ 583219fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 583319fbcb36SGuillaume Nault const unsigned char *src) 583419fbcb36SGuillaume Nault { 583519fbcb36SGuillaume Nault struct ethhdr *eth; 583619fbcb36SGuillaume Nault int err; 583719fbcb36SGuillaume Nault 583819fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 583919fbcb36SGuillaume Nault return -EPROTO; 584019fbcb36SGuillaume Nault 584119fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 584219fbcb36SGuillaume Nault if (err < 0) 584319fbcb36SGuillaume Nault return err; 584419fbcb36SGuillaume Nault 584519fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 584619fbcb36SGuillaume Nault skb_reset_mac_header(skb); 584719fbcb36SGuillaume Nault skb_reset_mac_len(skb); 584819fbcb36SGuillaume Nault 584919fbcb36SGuillaume Nault eth = eth_hdr(skb); 585019fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 585119fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 585219fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 585319fbcb36SGuillaume Nault 585419fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 585519fbcb36SGuillaume Nault 585619fbcb36SGuillaume Nault return 0; 585719fbcb36SGuillaume Nault } 585819fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 585919fbcb36SGuillaume Nault 58608822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 58618822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 58628822e270SJohn Hurley __be16 ethertype) 58638822e270SJohn Hurley { 58648822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 58658822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 58668822e270SJohn Hurley 58678822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 58688822e270SJohn Hurley } 58698822e270SJohn Hurley 58708822e270SJohn Hurley hdr->h_proto = ethertype; 58718822e270SJohn Hurley } 58728822e270SJohn Hurley 58738822e270SJohn Hurley /** 5874e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5875e7dbfed1SMartin Varghese * the packet 58768822e270SJohn Hurley * 58778822e270SJohn Hurley * @skb: buffer 58788822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 58798822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5880fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5881e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5882e7dbfed1SMartin Varghese * ethernet 58838822e270SJohn Hurley * 58848822e270SJohn Hurley * Expects skb->data at mac header. 58858822e270SJohn Hurley * 58868822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 58878822e270SJohn Hurley */ 5888fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5889d04ac224SMartin Varghese int mac_len, bool ethernet) 58908822e270SJohn Hurley { 58918822e270SJohn Hurley struct mpls_shim_hdr *lse; 58928822e270SJohn Hurley int err; 58938822e270SJohn Hurley 58948822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 58958822e270SJohn Hurley return -EINVAL; 58968822e270SJohn Hurley 58978822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 58988822e270SJohn Hurley if (skb->encapsulation) 58998822e270SJohn Hurley return -EINVAL; 59008822e270SJohn Hurley 59018822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 59028822e270SJohn Hurley if (unlikely(err)) 59038822e270SJohn Hurley return err; 59048822e270SJohn Hurley 59058822e270SJohn Hurley if (!skb->inner_protocol) { 5906e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 59078822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 59088822e270SJohn Hurley } 59098822e270SJohn Hurley 59108822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 59118822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5912fa4e0f88SDavide Caratti mac_len); 59138822e270SJohn Hurley skb_reset_mac_header(skb); 5914fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5915e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 59168822e270SJohn Hurley 59178822e270SJohn Hurley lse = mpls_hdr(skb); 59188822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 59198822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 59208822e270SJohn Hurley 59214296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 59228822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 59238822e270SJohn Hurley skb->protocol = mpls_proto; 59248822e270SJohn Hurley 59258822e270SJohn Hurley return 0; 59268822e270SJohn Hurley } 59278822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 59288822e270SJohn Hurley 59292e4e4410SEric Dumazet /** 5930ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5931ed246ceeSJohn Hurley * 5932ed246ceeSJohn Hurley * @skb: buffer 5933ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5934fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 593576f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5936ed246ceeSJohn Hurley * 5937ed246ceeSJohn Hurley * Expects skb->data at mac header. 5938ed246ceeSJohn Hurley * 5939ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5940ed246ceeSJohn Hurley */ 5941040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5942040b5cfbSMartin Varghese bool ethernet) 5943ed246ceeSJohn Hurley { 5944ed246ceeSJohn Hurley int err; 5945ed246ceeSJohn Hurley 5946ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5947dedc5a08SDavide Caratti return 0; 5948ed246ceeSJohn Hurley 5949fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5950ed246ceeSJohn Hurley if (unlikely(err)) 5951ed246ceeSJohn Hurley return err; 5952ed246ceeSJohn Hurley 5953ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5954ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5955fa4e0f88SDavide Caratti mac_len); 5956ed246ceeSJohn Hurley 5957ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5958ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5959fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5960ed246ceeSJohn Hurley 59614296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5962ed246ceeSJohn Hurley struct ethhdr *hdr; 5963ed246ceeSJohn Hurley 5964ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5965ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5966ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5967ed246ceeSJohn Hurley } 5968ed246ceeSJohn Hurley skb->protocol = next_proto; 5969ed246ceeSJohn Hurley 5970ed246ceeSJohn Hurley return 0; 5971ed246ceeSJohn Hurley } 5972ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5973ed246ceeSJohn Hurley 5974ed246ceeSJohn Hurley /** 5975d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5976d27cf5c5SJohn Hurley * 5977d27cf5c5SJohn Hurley * @skb: buffer 5978d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5979d27cf5c5SJohn Hurley * 5980d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5981d27cf5c5SJohn Hurley * 5982d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5983d27cf5c5SJohn Hurley */ 5984d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5985d27cf5c5SJohn Hurley { 5986d27cf5c5SJohn Hurley int err; 5987d27cf5c5SJohn Hurley 5988d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5989d27cf5c5SJohn Hurley return -EINVAL; 5990d27cf5c5SJohn Hurley 5991d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5992d27cf5c5SJohn Hurley if (unlikely(err)) 5993d27cf5c5SJohn Hurley return err; 5994d27cf5c5SJohn Hurley 5995d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5996d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5997d27cf5c5SJohn Hurley 5998d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5999d27cf5c5SJohn Hurley } 6000d27cf5c5SJohn Hurley 6001d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 6002d27cf5c5SJohn Hurley 6003d27cf5c5SJohn Hurley return 0; 6004d27cf5c5SJohn Hurley } 6005d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 6006d27cf5c5SJohn Hurley 6007d27cf5c5SJohn Hurley /** 60082a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 60092a2ea508SJohn Hurley * 60102a2ea508SJohn Hurley * @skb: buffer 60112a2ea508SJohn Hurley * 60122a2ea508SJohn Hurley * Expects skb->data at mac header. 60132a2ea508SJohn Hurley * 60142a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 60152a2ea508SJohn Hurley */ 60162a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 60172a2ea508SJohn Hurley { 60182a2ea508SJohn Hurley u32 lse; 60192a2ea508SJohn Hurley u8 ttl; 60202a2ea508SJohn Hurley 60212a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 60222a2ea508SJohn Hurley return -EINVAL; 60232a2ea508SJohn Hurley 602413de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 602513de4ed9SDavide Caratti return -ENOMEM; 602613de4ed9SDavide Caratti 60272a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 60282a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 60292a2ea508SJohn Hurley if (!--ttl) 60302a2ea508SJohn Hurley return -EINVAL; 60312a2ea508SJohn Hurley 60322a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 60332a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 60342a2ea508SJohn Hurley 60352a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 60362a2ea508SJohn Hurley } 60372a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 60382a2ea508SJohn Hurley 60392a2ea508SJohn Hurley /** 60402e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 60412e4e4410SEric Dumazet * 6042de3f0d0eSMasanari Iida * @header_len: size of linear part 6043de3f0d0eSMasanari Iida * @data_len: needed length in frags 6044de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 6045de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 6046de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 60472e4e4410SEric Dumazet * 60482e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 60492e4e4410SEric Dumazet */ 60502e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 60512e4e4410SEric Dumazet unsigned long data_len, 60522e4e4410SEric Dumazet int max_page_order, 60532e4e4410SEric Dumazet int *errcode, 60542e4e4410SEric Dumazet gfp_t gfp_mask) 60552e4e4410SEric Dumazet { 60562e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 60572e4e4410SEric Dumazet unsigned long chunk; 60582e4e4410SEric Dumazet struct sk_buff *skb; 60592e4e4410SEric Dumazet struct page *page; 60602e4e4410SEric Dumazet int i; 60612e4e4410SEric Dumazet 60622e4e4410SEric Dumazet *errcode = -EMSGSIZE; 60632e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 60642e4e4410SEric Dumazet * high order pages... 60652e4e4410SEric Dumazet */ 60662e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 60672e4e4410SEric Dumazet return NULL; 60682e4e4410SEric Dumazet 60692e4e4410SEric Dumazet *errcode = -ENOBUFS; 6070f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 60712e4e4410SEric Dumazet if (!skb) 60722e4e4410SEric Dumazet return NULL; 60732e4e4410SEric Dumazet 60742e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 60752e4e4410SEric Dumazet 60762e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 60772e4e4410SEric Dumazet int order = max_page_order; 60782e4e4410SEric Dumazet 60792e4e4410SEric Dumazet while (order) { 60802e4e4410SEric Dumazet if (npages >= 1 << order) { 6081d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 60822e4e4410SEric Dumazet __GFP_COMP | 6083d14b56f5SMichal Hocko __GFP_NOWARN, 60842e4e4410SEric Dumazet order); 60852e4e4410SEric Dumazet if (page) 60862e4e4410SEric Dumazet goto fill_page; 60872e4e4410SEric Dumazet /* Do not retry other high order allocations */ 60882e4e4410SEric Dumazet order = 1; 60892e4e4410SEric Dumazet max_page_order = 0; 60902e4e4410SEric Dumazet } 60912e4e4410SEric Dumazet order--; 60922e4e4410SEric Dumazet } 60932e4e4410SEric Dumazet page = alloc_page(gfp_mask); 60942e4e4410SEric Dumazet if (!page) 60952e4e4410SEric Dumazet goto failure; 60962e4e4410SEric Dumazet fill_page: 60972e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 60982e4e4410SEric Dumazet PAGE_SIZE << order); 60992e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 61002e4e4410SEric Dumazet data_len -= chunk; 61012e4e4410SEric Dumazet npages -= 1 << order; 61022e4e4410SEric Dumazet } 61032e4e4410SEric Dumazet return skb; 61042e4e4410SEric Dumazet 61052e4e4410SEric Dumazet failure: 61062e4e4410SEric Dumazet kfree_skb(skb); 61072e4e4410SEric Dumazet return NULL; 61082e4e4410SEric Dumazet } 61092e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 61106fa01ccdSSowmini Varadhan 61116fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 61126fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 61136fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 61146fa01ccdSSowmini Varadhan { 61156fa01ccdSSowmini Varadhan int i; 61166fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 61176fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 61186fa01ccdSSowmini Varadhan u8 *data; 61196fa01ccdSSowmini Varadhan 61206fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 61216fa01ccdSSowmini Varadhan 61226fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 61236fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 61246fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 61256fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 61266fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 61276fa01ccdSSowmini Varadhan if (!data) 61286fa01ccdSSowmini Varadhan return -ENOMEM; 61296fa01ccdSSowmini Varadhan 61306fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 61316fa01ccdSSowmini Varadhan 61326fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 61336fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 61346fa01ccdSSowmini Varadhan skb->len -= off; 61356fa01ccdSSowmini Varadhan 61366fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 61376fa01ccdSSowmini Varadhan skb_shinfo(skb), 61386fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 61396fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 61406fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 61416fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 61426fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 61436fa01ccdSSowmini Varadhan kfree(data); 61446fa01ccdSSowmini Varadhan return -ENOMEM; 61456fa01ccdSSowmini Varadhan } 61466fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 61476fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 61486fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 61496fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 61506fa01ccdSSowmini Varadhan skb_release_data(skb); 61516fa01ccdSSowmini Varadhan } else { 61526fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 61536fa01ccdSSowmini Varadhan * relocate values 61546fa01ccdSSowmini Varadhan */ 61556fa01ccdSSowmini Varadhan skb_free_head(skb); 61566fa01ccdSSowmini Varadhan } 61576fa01ccdSSowmini Varadhan 61586fa01ccdSSowmini Varadhan skb->head = data; 61596fa01ccdSSowmini Varadhan skb->data = data; 61606fa01ccdSSowmini Varadhan skb->head_frag = 0; 61616fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 61626fa01ccdSSowmini Varadhan skb->end = size; 61636fa01ccdSSowmini Varadhan #else 61646fa01ccdSSowmini Varadhan skb->end = skb->head + size; 61656fa01ccdSSowmini Varadhan #endif 61666fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 61676fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 61686fa01ccdSSowmini Varadhan skb->cloned = 0; 61696fa01ccdSSowmini Varadhan skb->hdr_len = 0; 61706fa01ccdSSowmini Varadhan skb->nohdr = 0; 61716fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 61726fa01ccdSSowmini Varadhan 61736fa01ccdSSowmini Varadhan return 0; 61746fa01ccdSSowmini Varadhan } 61756fa01ccdSSowmini Varadhan 61766fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 61776fa01ccdSSowmini Varadhan 61786fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 61796fa01ccdSSowmini Varadhan * pskb_carve() 61806fa01ccdSSowmini Varadhan */ 61816fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 61826fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 61836fa01ccdSSowmini Varadhan gfp_t gfp_mask) 61846fa01ccdSSowmini Varadhan { 61856fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 61866fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 61876fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 61886fa01ccdSSowmini Varadhan 61896fa01ccdSSowmini Varadhan do { 61906fa01ccdSSowmini Varadhan if (!list) { 61916fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 61926fa01ccdSSowmini Varadhan return -EFAULT; 61936fa01ccdSSowmini Varadhan } 61946fa01ccdSSowmini Varadhan if (list->len <= eat) { 61956fa01ccdSSowmini Varadhan /* Eaten as whole. */ 61966fa01ccdSSowmini Varadhan eat -= list->len; 61976fa01ccdSSowmini Varadhan list = list->next; 61986fa01ccdSSowmini Varadhan insp = list; 61996fa01ccdSSowmini Varadhan } else { 62006fa01ccdSSowmini Varadhan /* Eaten partially. */ 62016fa01ccdSSowmini Varadhan if (skb_shared(list)) { 62026fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 62036fa01ccdSSowmini Varadhan if (!clone) 62046fa01ccdSSowmini Varadhan return -ENOMEM; 62056fa01ccdSSowmini Varadhan insp = list->next; 62066fa01ccdSSowmini Varadhan list = clone; 62076fa01ccdSSowmini Varadhan } else { 62086fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 62096fa01ccdSSowmini Varadhan insp = list; 62106fa01ccdSSowmini Varadhan } 62116fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 62126fa01ccdSSowmini Varadhan kfree_skb(clone); 62136fa01ccdSSowmini Varadhan return -ENOMEM; 62146fa01ccdSSowmini Varadhan } 62156fa01ccdSSowmini Varadhan break; 62166fa01ccdSSowmini Varadhan } 62176fa01ccdSSowmini Varadhan } while (eat); 62186fa01ccdSSowmini Varadhan 62196fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 62206fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 62216fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 62226fa01ccdSSowmini Varadhan kfree_skb(list); 62236fa01ccdSSowmini Varadhan } 62246fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 62256fa01ccdSSowmini Varadhan if (clone) { 62266fa01ccdSSowmini Varadhan clone->next = list; 62276fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 62286fa01ccdSSowmini Varadhan } 62296fa01ccdSSowmini Varadhan return 0; 62306fa01ccdSSowmini Varadhan } 62316fa01ccdSSowmini Varadhan 62326fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 62336fa01ccdSSowmini Varadhan * non-linear part of skb 62346fa01ccdSSowmini Varadhan */ 62356fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 62366fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 62376fa01ccdSSowmini Varadhan { 62386fa01ccdSSowmini Varadhan int i, k = 0; 62396fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 62406fa01ccdSSowmini Varadhan u8 *data; 62416fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 62426fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 62436fa01ccdSSowmini Varadhan 62446fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 62456fa01ccdSSowmini Varadhan 62466fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 62476fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 62486fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 62496fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 62506fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 62516fa01ccdSSowmini Varadhan if (!data) 62526fa01ccdSSowmini Varadhan return -ENOMEM; 62536fa01ccdSSowmini Varadhan 62546fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 62556fa01ccdSSowmini Varadhan 62566fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6257e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 62586fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 62596fa01ccdSSowmini Varadhan kfree(data); 62606fa01ccdSSowmini Varadhan return -ENOMEM; 62616fa01ccdSSowmini Varadhan } 62626fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 62636fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 62646fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 62656fa01ccdSSowmini Varadhan 62666fa01ccdSSowmini Varadhan if (pos + fsize > off) { 62676fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 62686fa01ccdSSowmini Varadhan 62696fa01ccdSSowmini Varadhan if (pos < off) { 62706fa01ccdSSowmini Varadhan /* Split frag. 62716fa01ccdSSowmini Varadhan * We have two variants in this case: 62726fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 62736fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 62746fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 62756fa01ccdSSowmini Varadhan * where splitting is expensive. 62766fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 62776fa01ccdSSowmini Varadhan */ 6278b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 62796fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 62806fa01ccdSSowmini Varadhan } 62816fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 62826fa01ccdSSowmini Varadhan k++; 62836fa01ccdSSowmini Varadhan } 62846fa01ccdSSowmini Varadhan pos += fsize; 62856fa01ccdSSowmini Varadhan } 62866fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 62876fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 62886fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 62896fa01ccdSSowmini Varadhan 62906fa01ccdSSowmini Varadhan /* split line is in frag list */ 6291eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6292eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6293eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6294eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6295eabe8618SMiaohe Lin kfree(data); 6296eabe8618SMiaohe Lin return -ENOMEM; 62976fa01ccdSSowmini Varadhan } 62986fa01ccdSSowmini Varadhan skb_release_data(skb); 62996fa01ccdSSowmini Varadhan 63006fa01ccdSSowmini Varadhan skb->head = data; 63016fa01ccdSSowmini Varadhan skb->head_frag = 0; 63026fa01ccdSSowmini Varadhan skb->data = data; 63036fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 63046fa01ccdSSowmini Varadhan skb->end = size; 63056fa01ccdSSowmini Varadhan #else 63066fa01ccdSSowmini Varadhan skb->end = skb->head + size; 63076fa01ccdSSowmini Varadhan #endif 63086fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 63096fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 63106fa01ccdSSowmini Varadhan skb->cloned = 0; 63116fa01ccdSSowmini Varadhan skb->hdr_len = 0; 63126fa01ccdSSowmini Varadhan skb->nohdr = 0; 63136fa01ccdSSowmini Varadhan skb->len -= off; 63146fa01ccdSSowmini Varadhan skb->data_len = skb->len; 63156fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 63166fa01ccdSSowmini Varadhan return 0; 63176fa01ccdSSowmini Varadhan } 63186fa01ccdSSowmini Varadhan 63196fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 63206fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 63216fa01ccdSSowmini Varadhan { 63226fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 63236fa01ccdSSowmini Varadhan 63246fa01ccdSSowmini Varadhan if (len < headlen) 63256fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 63266fa01ccdSSowmini Varadhan else 63276fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 63286fa01ccdSSowmini Varadhan } 63296fa01ccdSSowmini Varadhan 63306fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 63316fa01ccdSSowmini Varadhan * a new skb 63326fa01ccdSSowmini Varadhan */ 63336fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 63346fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 63356fa01ccdSSowmini Varadhan { 63366fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 63376fa01ccdSSowmini Varadhan 63386fa01ccdSSowmini Varadhan if (!clone) 63396fa01ccdSSowmini Varadhan return NULL; 63406fa01ccdSSowmini Varadhan 63416fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 63426fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 63436fa01ccdSSowmini Varadhan kfree_skb(clone); 63446fa01ccdSSowmini Varadhan return NULL; 63456fa01ccdSSowmini Varadhan } 63466fa01ccdSSowmini Varadhan return clone; 63476fa01ccdSSowmini Varadhan } 63486fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6349c8c8b127SEric Dumazet 6350c8c8b127SEric Dumazet /** 6351c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6352c8c8b127SEric Dumazet * @skb: buffer 6353c8c8b127SEric Dumazet * 6354c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6355c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6356c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6357c8c8b127SEric Dumazet * truesize. 6358c8c8b127SEric Dumazet * Notes: 6359c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6360c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6361c8c8b127SEric Dumazet */ 6362c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6363c8c8b127SEric Dumazet { 63643174fed9SEric Dumazet if (skb->data_len) { 63653174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6366c8c8b127SEric Dumazet skb_cloned(skb)) 6367c8c8b127SEric Dumazet return; 6368c8c8b127SEric Dumazet 6369c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6370c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 63713174fed9SEric Dumazet } 63723174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 63733174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 63743174fed9SEric Dumazet * their truesize. 63753174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 63763174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 63773174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6378c8c8b127SEric Dumazet */ 6379c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6380c8c8b127SEric Dumazet } 6381df5042f4SFlorian Westphal 6382df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6383df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6384df5042f4SFlorian Westphal { 6385df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6386df5042f4SFlorian Westphal } 6387df5042f4SFlorian Westphal 63888b69a803SPaolo Abeni /** 63898b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 63908b69a803SPaolo Abeni * 63914930f483SFlorian Westphal * @flags: See kmalloc(). 63924930f483SFlorian Westphal * 63938b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 63948b69a803SPaolo Abeni * skb via __skb_ext_set(). 63958b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 63968b69a803SPaolo Abeni */ 63974930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6398df5042f4SFlorian Westphal { 63994930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6400df5042f4SFlorian Westphal 6401df5042f4SFlorian Westphal if (new) { 6402df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6403df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6404df5042f4SFlorian Westphal } 6405df5042f4SFlorian Westphal 6406df5042f4SFlorian Westphal return new; 6407df5042f4SFlorian Westphal } 6408df5042f4SFlorian Westphal 64094165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 64104165079bSFlorian Westphal unsigned int old_active) 6411df5042f4SFlorian Westphal { 6412df5042f4SFlorian Westphal struct skb_ext *new; 6413df5042f4SFlorian Westphal 6414df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6415df5042f4SFlorian Westphal return old; 6416df5042f4SFlorian Westphal 6417df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6418df5042f4SFlorian Westphal if (!new) 6419df5042f4SFlorian Westphal return NULL; 6420df5042f4SFlorian Westphal 6421df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6422df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6423df5042f4SFlorian Westphal 64244165079bSFlorian Westphal #ifdef CONFIG_XFRM 64254165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 64264165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 64274165079bSFlorian Westphal unsigned int i; 64284165079bSFlorian Westphal 64294165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 64304165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 64314165079bSFlorian Westphal } 64324165079bSFlorian Westphal #endif 6433df5042f4SFlorian Westphal __skb_ext_put(old); 6434df5042f4SFlorian Westphal return new; 6435df5042f4SFlorian Westphal } 6436df5042f4SFlorian Westphal 6437df5042f4SFlorian Westphal /** 64388b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 64398b69a803SPaolo Abeni * @skb: buffer 64408b69a803SPaolo Abeni * @id: extension id 64418b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 64428b69a803SPaolo Abeni * 64438b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 64448b69a803SPaolo Abeni * 64458b69a803SPaolo Abeni * Returns the pointer to the extension. 64468b69a803SPaolo Abeni */ 64478b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 64488b69a803SPaolo Abeni struct skb_ext *ext) 64498b69a803SPaolo Abeni { 64508b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 64518b69a803SPaolo Abeni 64528b69a803SPaolo Abeni skb_ext_put(skb); 64538b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 64548b69a803SPaolo Abeni ext->chunks = newlen; 64558b69a803SPaolo Abeni ext->offset[id] = newoff; 64568b69a803SPaolo Abeni skb->extensions = ext; 64578b69a803SPaolo Abeni skb->active_extensions = 1 << id; 64588b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 64598b69a803SPaolo Abeni } 64608b69a803SPaolo Abeni 64618b69a803SPaolo Abeni /** 6462df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6463df5042f4SFlorian Westphal * @skb: buffer 6464df5042f4SFlorian Westphal * @id: extension to allocate space for 6465df5042f4SFlorian Westphal * 6466df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6467df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6468df5042f4SFlorian Westphal * is returned. 6469df5042f4SFlorian Westphal * 6470df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6471df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6472df5042f4SFlorian Westphal * 6473df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6474df5042f4SFlorian Westphal */ 6475df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6476df5042f4SFlorian Westphal { 6477df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6478df5042f4SFlorian Westphal unsigned int newlen, newoff; 6479df5042f4SFlorian Westphal 6480df5042f4SFlorian Westphal if (skb->active_extensions) { 6481df5042f4SFlorian Westphal old = skb->extensions; 6482df5042f4SFlorian Westphal 64834165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6484df5042f4SFlorian Westphal if (!new) 6485df5042f4SFlorian Westphal return NULL; 6486df5042f4SFlorian Westphal 6487682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6488df5042f4SFlorian Westphal goto set_active; 6489df5042f4SFlorian Westphal 6490e94e50bdSPaolo Abeni newoff = new->chunks; 6491df5042f4SFlorian Westphal } else { 6492df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6493df5042f4SFlorian Westphal 64944930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6495df5042f4SFlorian Westphal if (!new) 6496df5042f4SFlorian Westphal return NULL; 6497df5042f4SFlorian Westphal } 6498df5042f4SFlorian Westphal 6499df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6500df5042f4SFlorian Westphal new->chunks = newlen; 6501df5042f4SFlorian Westphal new->offset[id] = newoff; 6502df5042f4SFlorian Westphal set_active: 6503b0999f38SPaolo Abeni skb->slow_gro = 1; 6504682ec859SPaolo Abeni skb->extensions = new; 6505df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6506df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6507df5042f4SFlorian Westphal } 6508df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6509df5042f4SFlorian Westphal 65104165079bSFlorian Westphal #ifdef CONFIG_XFRM 65114165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 65124165079bSFlorian Westphal { 65134165079bSFlorian Westphal unsigned int i; 65144165079bSFlorian Westphal 65154165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 65164165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 65174165079bSFlorian Westphal } 65184165079bSFlorian Westphal #endif 65194165079bSFlorian Westphal 6520df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6521df5042f4SFlorian Westphal { 6522df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6523df5042f4SFlorian Westphal 6524df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6525df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6526df5042f4SFlorian Westphal skb->extensions = NULL; 6527df5042f4SFlorian Westphal __skb_ext_put(ext); 65284165079bSFlorian Westphal #ifdef CONFIG_XFRM 65294165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 65304165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 65314165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 65324165079bSFlorian Westphal 65334165079bSFlorian Westphal skb_ext_put_sp(sp); 65344165079bSFlorian Westphal sp->len = 0; 65354165079bSFlorian Westphal #endif 6536df5042f4SFlorian Westphal } 6537df5042f4SFlorian Westphal } 6538df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6539df5042f4SFlorian Westphal 6540df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6541df5042f4SFlorian Westphal { 6542df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6543df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6544df5042f4SFlorian Westphal */ 6545df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6546df5042f4SFlorian Westphal goto free_now; 6547df5042f4SFlorian Westphal 6548df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6549df5042f4SFlorian Westphal return; 6550df5042f4SFlorian Westphal free_now: 65514165079bSFlorian Westphal #ifdef CONFIG_XFRM 65524165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 65534165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 65544165079bSFlorian Westphal #endif 65554165079bSFlorian Westphal 6556df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6557df5042f4SFlorian Westphal } 6558df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6559df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 6560