12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 41da177e4SLinus Torvalds * 5113aa838SAlan Cox * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 61da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Fixes: 91da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 101da177e4SLinus Torvalds * balancer bugs. 111da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 121da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 131da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 141da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 151da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 161da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 171da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 181da177e4SLinus Torvalds * only put in the headers 191da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 201da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 211da177e4SLinus Torvalds * Andi Kleen : slabified it. 221da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 231da177e4SLinus Torvalds * 241da177e4SLinus Torvalds * NOTE: 251da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 261da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 271da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 281da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 331da177e4SLinus Torvalds */ 341da177e4SLinus Torvalds 35e005d193SJoe Perches #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 36e005d193SJoe Perches 371da177e4SLinus Torvalds #include <linux/module.h> 381da177e4SLinus Torvalds #include <linux/types.h> 391da177e4SLinus Torvalds #include <linux/kernel.h> 401da177e4SLinus Torvalds #include <linux/mm.h> 411da177e4SLinus Torvalds #include <linux/interrupt.h> 421da177e4SLinus Torvalds #include <linux/in.h> 431da177e4SLinus Torvalds #include <linux/inet.h> 441da177e4SLinus Torvalds #include <linux/slab.h> 45de960aa9SFlorian Westphal #include <linux/tcp.h> 46de960aa9SFlorian Westphal #include <linux/udp.h> 4790017accSMarcelo Ricardo Leitner #include <linux/sctp.h> 481da177e4SLinus Torvalds #include <linux/netdevice.h> 491da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 501da177e4SLinus Torvalds #include <net/pkt_sched.h> 511da177e4SLinus Torvalds #endif 521da177e4SLinus Torvalds #include <linux/string.h> 531da177e4SLinus Torvalds #include <linux/skbuff.h> 549c55e01cSJens Axboe #include <linux/splice.h> 551da177e4SLinus Torvalds #include <linux/cache.h> 561da177e4SLinus Torvalds #include <linux/rtnetlink.h> 571da177e4SLinus Torvalds #include <linux/init.h> 58716ea3a7SDavid Howells #include <linux/scatterlist.h> 59ac45f602SPatrick Ohly #include <linux/errqueue.h> 60268bb0ceSLinus Torvalds #include <linux/prefetch.h> 610d5501c1SVlad Yasevich #include <linux/if_vlan.h> 622a2ea508SJohn Hurley #include <linux/mpls.h> 63183f47fcSSebastian Andrzej Siewior #include <linux/kcov.h> 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds #include <net/protocol.h> 661da177e4SLinus Torvalds #include <net/dst.h> 671da177e4SLinus Torvalds #include <net/sock.h> 681da177e4SLinus Torvalds #include <net/checksum.h> 69ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 701da177e4SLinus Torvalds #include <net/xfrm.h> 718822e270SJohn Hurley #include <net/mpls.h> 723ee17bc7SMat Martineau #include <net/mptcp.h> 7378476d31SJeremy Kerr #include <net/mctp.h> 746a5bcd84SIlias Apalodimas #include <net/page_pool.h> 751da177e4SLinus Torvalds 767c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 77ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7851c56b00SEric Dumazet #include <linux/highmem.h> 79b245be1fSWillem de Bruijn #include <linux/capability.h> 80b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 812544af03SMatteo Croce #include <linux/indirect_call_wrapper.h> 82a1f8e7f7SAl Viro 837b7ed885SBart Van Assche #include "datagram.h" 847f678defSVasily Averin #include "sock_destructor.h" 857b7ed885SBart Van Assche 8608009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8708009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 88df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 89df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 90df5042f4SFlorian Westphal #endif 915f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 925f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /** 95f05de73bSJean Sacren * skb_panic - private function for out-of-line support 961da177e4SLinus Torvalds * @skb: buffer 971da177e4SLinus Torvalds * @sz: size 98f05de73bSJean Sacren * @addr: address 9999d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 1001da177e4SLinus Torvalds * 101f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 102f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 103f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 104f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1051da177e4SLinus Torvalds */ 106f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 10799d5851eSJames Hogan const char msg[]) 1081da177e4SLinus Torvalds { 10941a46913SJesper Dangaard Brouer pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", 11099d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1114305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 11226095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1131da177e4SLinus Torvalds BUG(); 1141da177e4SLinus Torvalds } 1151da177e4SLinus Torvalds 116f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1171da177e4SLinus Torvalds { 118f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1191da177e4SLinus Torvalds } 1201da177e4SLinus Torvalds 121f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 122f05de73bSJean Sacren { 123f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 124f05de73bSJean Sacren } 125c93bdd0eSMel Gorman 12650fad4b5SAlexander Lobakin #define NAPI_SKB_CACHE_SIZE 64 127f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_BULK 16 128f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) 12950fad4b5SAlexander Lobakin 13050fad4b5SAlexander Lobakin struct napi_alloc_cache { 13150fad4b5SAlexander Lobakin struct page_frag_cache page; 13250fad4b5SAlexander Lobakin unsigned int skb_count; 13350fad4b5SAlexander Lobakin void *skb_cache[NAPI_SKB_CACHE_SIZE]; 13450fad4b5SAlexander Lobakin }; 13550fad4b5SAlexander Lobakin 13650fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 13750fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 13850fad4b5SAlexander Lobakin 13932e3573fSYajun Deng void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 14050fad4b5SAlexander Lobakin { 14150fad4b5SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 14250fad4b5SAlexander Lobakin 14350fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 14450fad4b5SAlexander Lobakin 14532e3573fSYajun Deng return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 14650fad4b5SAlexander Lobakin } 14750fad4b5SAlexander Lobakin EXPORT_SYMBOL(__napi_alloc_frag_align); 14850fad4b5SAlexander Lobakin 14950fad4b5SAlexander Lobakin void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 15050fad4b5SAlexander Lobakin { 15150fad4b5SAlexander Lobakin void *data; 15250fad4b5SAlexander Lobakin 15350fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 154afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 15532e3573fSYajun Deng struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache); 15632e3573fSYajun Deng 15750fad4b5SAlexander Lobakin data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask); 15850fad4b5SAlexander Lobakin } else { 15932e3573fSYajun Deng struct napi_alloc_cache *nc; 16032e3573fSYajun Deng 16150fad4b5SAlexander Lobakin local_bh_disable(); 16232e3573fSYajun Deng nc = this_cpu_ptr(&napi_alloc_cache); 16332e3573fSYajun Deng data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 16450fad4b5SAlexander Lobakin local_bh_enable(); 16550fad4b5SAlexander Lobakin } 16650fad4b5SAlexander Lobakin return data; 16750fad4b5SAlexander Lobakin } 16850fad4b5SAlexander Lobakin EXPORT_SYMBOL(__netdev_alloc_frag_align); 16950fad4b5SAlexander Lobakin 170f450d539SAlexander Lobakin static struct sk_buff *napi_skb_cache_get(void) 171f450d539SAlexander Lobakin { 172f450d539SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 173f450d539SAlexander Lobakin struct sk_buff *skb; 174f450d539SAlexander Lobakin 175f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 176f450d539SAlexander Lobakin nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache, 177f450d539SAlexander Lobakin GFP_ATOMIC, 178f450d539SAlexander Lobakin NAPI_SKB_CACHE_BULK, 179f450d539SAlexander Lobakin nc->skb_cache); 180f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 181f450d539SAlexander Lobakin return NULL; 182f450d539SAlexander Lobakin 183f450d539SAlexander Lobakin skb = nc->skb_cache[--nc->skb_count]; 184f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, skb); 185f450d539SAlexander Lobakin 186f450d539SAlexander Lobakin return skb; 187f450d539SAlexander Lobakin } 188f450d539SAlexander Lobakin 189ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 190483126b3SAlexander Lobakin static void __build_skb_around(struct sk_buff *skb, void *data, 191483126b3SAlexander Lobakin unsigned int frag_size) 192ba0509b6SJesper Dangaard Brouer { 193ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 194ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 195ba0509b6SJesper Dangaard Brouer 196ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 197ba0509b6SJesper Dangaard Brouer 198ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 199ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 200ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 201ba0509b6SJesper Dangaard Brouer skb->head = data; 202ba0509b6SJesper Dangaard Brouer skb->data = data; 203ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 204ba0509b6SJesper Dangaard Brouer skb->end = skb->tail + size; 205ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 206ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 207ba0509b6SJesper Dangaard Brouer 208ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 209ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 210ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 211ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 212ba0509b6SJesper Dangaard Brouer 2136370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 214ba0509b6SJesper Dangaard Brouer } 215ba0509b6SJesper Dangaard Brouer 2161da177e4SLinus Torvalds /** 2172ea2f62cSEric Dumazet * __build_skb - build a network buffer 218b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2192ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 220b2b5ce9dSEric Dumazet * 221b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 222deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2232ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2242ea2f62cSEric Dumazet * or vmalloc() 225b2b5ce9dSEric Dumazet * The return is the new skb buffer. 226b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 227b2b5ce9dSEric Dumazet * Notes : 228b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 229b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 230b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 231b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 232b2b5ce9dSEric Dumazet * before giving packet to stack. 233b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 234b2b5ce9dSEric Dumazet */ 2352ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 236b2b5ce9dSEric Dumazet { 237b2b5ce9dSEric Dumazet struct sk_buff *skb; 238b2b5ce9dSEric Dumazet 239b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 240ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 241b2b5ce9dSEric Dumazet return NULL; 242b2b5ce9dSEric Dumazet 243b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 244483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 245b2b5ce9dSEric Dumazet 246483126b3SAlexander Lobakin return skb; 247b2b5ce9dSEric Dumazet } 2482ea2f62cSEric Dumazet 2492ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 2502ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 2512ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 2522ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 2532ea2f62cSEric Dumazet */ 2542ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 2552ea2f62cSEric Dumazet { 2562ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 2572ea2f62cSEric Dumazet 2582ea2f62cSEric Dumazet if (skb && frag_size) { 2592ea2f62cSEric Dumazet skb->head_frag = 1; 2602f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 2612ea2f62cSEric Dumazet skb->pfmemalloc = 1; 2622ea2f62cSEric Dumazet } 2632ea2f62cSEric Dumazet return skb; 2642ea2f62cSEric Dumazet } 265b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 266b2b5ce9dSEric Dumazet 267ba0509b6SJesper Dangaard Brouer /** 268ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 269ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 270ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 271ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 272ba0509b6SJesper Dangaard Brouer */ 273ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 274ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 275ba0509b6SJesper Dangaard Brouer { 276ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 277ba0509b6SJesper Dangaard Brouer return NULL; 278ba0509b6SJesper Dangaard Brouer 279483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 280ba0509b6SJesper Dangaard Brouer 281483126b3SAlexander Lobakin if (frag_size) { 282ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 283ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 284ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 285ba0509b6SJesper Dangaard Brouer } 286ba0509b6SJesper Dangaard Brouer return skb; 287ba0509b6SJesper Dangaard Brouer } 288ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 289ba0509b6SJesper Dangaard Brouer 290f450d539SAlexander Lobakin /** 291f450d539SAlexander Lobakin * __napi_build_skb - build a network buffer 292f450d539SAlexander Lobakin * @data: data buffer provided by caller 293f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 294f450d539SAlexander Lobakin * 295f450d539SAlexander Lobakin * Version of __build_skb() that uses NAPI percpu caches to obtain 296f450d539SAlexander Lobakin * skbuff_head instead of inplace allocation. 297f450d539SAlexander Lobakin * 298f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 299f450d539SAlexander Lobakin */ 300f450d539SAlexander Lobakin static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size) 301f450d539SAlexander Lobakin { 302f450d539SAlexander Lobakin struct sk_buff *skb; 303f450d539SAlexander Lobakin 304f450d539SAlexander Lobakin skb = napi_skb_cache_get(); 305f450d539SAlexander Lobakin if (unlikely(!skb)) 306f450d539SAlexander Lobakin return NULL; 307f450d539SAlexander Lobakin 308f450d539SAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 309f450d539SAlexander Lobakin __build_skb_around(skb, data, frag_size); 310f450d539SAlexander Lobakin 311f450d539SAlexander Lobakin return skb; 312f450d539SAlexander Lobakin } 313f450d539SAlexander Lobakin 314f450d539SAlexander Lobakin /** 315f450d539SAlexander Lobakin * napi_build_skb - build a network buffer 316f450d539SAlexander Lobakin * @data: data buffer provided by caller 317f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 318f450d539SAlexander Lobakin * 319f450d539SAlexander Lobakin * Version of __napi_build_skb() that takes care of skb->head_frag 320f450d539SAlexander Lobakin * and skb->pfmemalloc when the data is a page or page fragment. 321f450d539SAlexander Lobakin * 322f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 323f450d539SAlexander Lobakin */ 324f450d539SAlexander Lobakin struct sk_buff *napi_build_skb(void *data, unsigned int frag_size) 325f450d539SAlexander Lobakin { 326f450d539SAlexander Lobakin struct sk_buff *skb = __napi_build_skb(data, frag_size); 327f450d539SAlexander Lobakin 328f450d539SAlexander Lobakin if (likely(skb) && frag_size) { 329f450d539SAlexander Lobakin skb->head_frag = 1; 330f450d539SAlexander Lobakin skb_propagate_pfmemalloc(virt_to_head_page(data), skb); 331f450d539SAlexander Lobakin } 332f450d539SAlexander Lobakin 333f450d539SAlexander Lobakin return skb; 334f450d539SAlexander Lobakin } 335f450d539SAlexander Lobakin EXPORT_SYMBOL(napi_build_skb); 336f450d539SAlexander Lobakin 3375381b23dSAlexander Lobakin /* 3385381b23dSAlexander Lobakin * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 3395381b23dSAlexander Lobakin * the caller if emergency pfmemalloc reserves are being used. If it is and 3405381b23dSAlexander Lobakin * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 3415381b23dSAlexander Lobakin * may be used. Otherwise, the packet data may be discarded until enough 3425381b23dSAlexander Lobakin * memory is free 3435381b23dSAlexander Lobakin */ 344ef28095fSAlexander Lobakin static void *kmalloc_reserve(size_t size, gfp_t flags, int node, 345ef28095fSAlexander Lobakin bool *pfmemalloc) 3465381b23dSAlexander Lobakin { 3475381b23dSAlexander Lobakin void *obj; 3485381b23dSAlexander Lobakin bool ret_pfmemalloc = false; 3495381b23dSAlexander Lobakin 3505381b23dSAlexander Lobakin /* 3515381b23dSAlexander Lobakin * Try a regular allocation, when that fails and we're not entitled 3525381b23dSAlexander Lobakin * to the reserves, fail. 3535381b23dSAlexander Lobakin */ 3545381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, 3555381b23dSAlexander Lobakin flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 3565381b23dSAlexander Lobakin node); 3575381b23dSAlexander Lobakin if (obj || !(gfp_pfmemalloc_allowed(flags))) 3585381b23dSAlexander Lobakin goto out; 3595381b23dSAlexander Lobakin 3605381b23dSAlexander Lobakin /* Try again but now we are using pfmemalloc reserves */ 3615381b23dSAlexander Lobakin ret_pfmemalloc = true; 3625381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, flags, node); 3635381b23dSAlexander Lobakin 3645381b23dSAlexander Lobakin out: 3655381b23dSAlexander Lobakin if (pfmemalloc) 3665381b23dSAlexander Lobakin *pfmemalloc = ret_pfmemalloc; 3675381b23dSAlexander Lobakin 3685381b23dSAlexander Lobakin return obj; 3695381b23dSAlexander Lobakin } 3705381b23dSAlexander Lobakin 3715381b23dSAlexander Lobakin /* Allocate a new skbuff. We do this ourselves so we can fill in a few 3725381b23dSAlexander Lobakin * 'private' fields and also do memory statistics to find all the 3735381b23dSAlexander Lobakin * [BEEP] leaks. 3745381b23dSAlexander Lobakin * 3755381b23dSAlexander Lobakin */ 3765381b23dSAlexander Lobakin 3775381b23dSAlexander Lobakin /** 3785381b23dSAlexander Lobakin * __alloc_skb - allocate a network buffer 3795381b23dSAlexander Lobakin * @size: size to allocate 3805381b23dSAlexander Lobakin * @gfp_mask: allocation mask 3815381b23dSAlexander Lobakin * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 3825381b23dSAlexander Lobakin * instead of head cache and allocate a cloned (child) skb. 3835381b23dSAlexander Lobakin * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 3845381b23dSAlexander Lobakin * allocations in case the data is required for writeback 3855381b23dSAlexander Lobakin * @node: numa node to allocate memory on 3865381b23dSAlexander Lobakin * 3875381b23dSAlexander Lobakin * Allocate a new &sk_buff. The returned buffer has no headroom and a 3885381b23dSAlexander Lobakin * tail room of at least size bytes. The object has a reference count 3895381b23dSAlexander Lobakin * of one. The return is the buffer. On a failure the return is %NULL. 3905381b23dSAlexander Lobakin * 3915381b23dSAlexander Lobakin * Buffers may only be allocated from interrupts using a @gfp_mask of 3925381b23dSAlexander Lobakin * %GFP_ATOMIC. 3935381b23dSAlexander Lobakin */ 3945381b23dSAlexander Lobakin struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 3955381b23dSAlexander Lobakin int flags, int node) 3965381b23dSAlexander Lobakin { 3975381b23dSAlexander Lobakin struct kmem_cache *cache; 3985381b23dSAlexander Lobakin struct sk_buff *skb; 399a5df6333SLi RongQing unsigned int osize; 4005381b23dSAlexander Lobakin bool pfmemalloc; 401a5df6333SLi RongQing u8 *data; 4025381b23dSAlexander Lobakin 4035381b23dSAlexander Lobakin cache = (flags & SKB_ALLOC_FCLONE) 4045381b23dSAlexander Lobakin ? skbuff_fclone_cache : skbuff_head_cache; 4055381b23dSAlexander Lobakin 4065381b23dSAlexander Lobakin if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 4075381b23dSAlexander Lobakin gfp_mask |= __GFP_MEMALLOC; 4085381b23dSAlexander Lobakin 4095381b23dSAlexander Lobakin /* Get the HEAD */ 410d13612b5SAlexander Lobakin if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI && 411d13612b5SAlexander Lobakin likely(node == NUMA_NO_NODE || node == numa_mem_id())) 412d13612b5SAlexander Lobakin skb = napi_skb_cache_get(); 413d13612b5SAlexander Lobakin else 414d13612b5SAlexander Lobakin skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node); 415df1ae022SAlexander Lobakin if (unlikely(!skb)) 416df1ae022SAlexander Lobakin return NULL; 4175381b23dSAlexander Lobakin prefetchw(skb); 4185381b23dSAlexander Lobakin 4195381b23dSAlexander Lobakin /* We do our best to align skb_shared_info on a separate cache 4205381b23dSAlexander Lobakin * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 4215381b23dSAlexander Lobakin * aligned memory blocks, unless SLUB/SLAB debug is enabled. 4225381b23dSAlexander Lobakin * Both skb->head and skb_shared_info are cache line aligned. 4235381b23dSAlexander Lobakin */ 4245381b23dSAlexander Lobakin size = SKB_DATA_ALIGN(size); 4255381b23dSAlexander Lobakin size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4265381b23dSAlexander Lobakin data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 427df1ae022SAlexander Lobakin if (unlikely(!data)) 4285381b23dSAlexander Lobakin goto nodata; 4295381b23dSAlexander Lobakin /* kmalloc(size) might give us more room than requested. 4305381b23dSAlexander Lobakin * Put skb_shared_info exactly at the end of allocated zone, 4315381b23dSAlexander Lobakin * to allow max possible filling before reallocation. 4325381b23dSAlexander Lobakin */ 433a5df6333SLi RongQing osize = ksize(data); 434a5df6333SLi RongQing size = SKB_WITH_OVERHEAD(osize); 4355381b23dSAlexander Lobakin prefetchw(data + size); 4365381b23dSAlexander Lobakin 4375381b23dSAlexander Lobakin /* 4385381b23dSAlexander Lobakin * Only clear those fields we need to clear, not those that we will 4395381b23dSAlexander Lobakin * actually initialise below. Hence, don't put any more fields after 4405381b23dSAlexander Lobakin * the tail pointer in struct sk_buff! 4415381b23dSAlexander Lobakin */ 4425381b23dSAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 443a5df6333SLi RongQing __build_skb_around(skb, data, osize); 4445381b23dSAlexander Lobakin skb->pfmemalloc = pfmemalloc; 4455381b23dSAlexander Lobakin 4465381b23dSAlexander Lobakin if (flags & SKB_ALLOC_FCLONE) { 4475381b23dSAlexander Lobakin struct sk_buff_fclones *fclones; 4485381b23dSAlexander Lobakin 4495381b23dSAlexander Lobakin fclones = container_of(skb, struct sk_buff_fclones, skb1); 4505381b23dSAlexander Lobakin 4515381b23dSAlexander Lobakin skb->fclone = SKB_FCLONE_ORIG; 4525381b23dSAlexander Lobakin refcount_set(&fclones->fclone_ref, 1); 4535381b23dSAlexander Lobakin 4545381b23dSAlexander Lobakin fclones->skb2.fclone = SKB_FCLONE_CLONE; 4555381b23dSAlexander Lobakin } 4565381b23dSAlexander Lobakin 4575381b23dSAlexander Lobakin return skb; 458df1ae022SAlexander Lobakin 4595381b23dSAlexander Lobakin nodata: 4605381b23dSAlexander Lobakin kmem_cache_free(cache, skb); 461df1ae022SAlexander Lobakin return NULL; 4625381b23dSAlexander Lobakin } 4635381b23dSAlexander Lobakin EXPORT_SYMBOL(__alloc_skb); 4645381b23dSAlexander Lobakin 4657ba7aeabSSebastian Andrzej Siewior /** 466fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 467fd11a83dSAlexander Duyck * @dev: network device to receive on 468d7499160SMasanari Iida * @len: length to allocate 469fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 470fd11a83dSAlexander Duyck * 471fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 472fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 473fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 474fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 475fd11a83dSAlexander Duyck * 476fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 477fd11a83dSAlexander Duyck */ 4789451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4799451980aSAlexander Duyck gfp_t gfp_mask) 480fd11a83dSAlexander Duyck { 481b63ae8caSAlexander Duyck struct page_frag_cache *nc; 482fd11a83dSAlexander Duyck struct sk_buff *skb; 4839451980aSAlexander Duyck bool pfmemalloc; 4849451980aSAlexander Duyck void *data; 485fd11a83dSAlexander Duyck 4869451980aSAlexander Duyck len += NET_SKB_PAD; 487fd11a83dSAlexander Duyck 48866c55602SAlexander Lobakin /* If requested length is either too small or too big, 48966c55602SAlexander Lobakin * we use kmalloc() for skb->head allocation. 49066c55602SAlexander Lobakin */ 49166c55602SAlexander Lobakin if (len <= SKB_WITH_OVERHEAD(1024) || 49266c55602SAlexander Lobakin len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 493d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 494a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 495a080e7bdSAlexander Duyck if (!skb) 496a080e7bdSAlexander Duyck goto skb_fail; 497a080e7bdSAlexander Duyck goto skb_success; 498a080e7bdSAlexander Duyck } 4999451980aSAlexander Duyck 5009451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5019451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5029451980aSAlexander Duyck 5039451980aSAlexander Duyck if (sk_memalloc_socks()) 5049451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5059451980aSAlexander Duyck 506afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 5079451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 5088c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 5099451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 51092dcabd7SSebastian Andrzej Siewior } else { 51192dcabd7SSebastian Andrzej Siewior local_bh_disable(); 51292dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 51392dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 51492dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 51592dcabd7SSebastian Andrzej Siewior local_bh_enable(); 51692dcabd7SSebastian Andrzej Siewior } 5179451980aSAlexander Duyck 5189451980aSAlexander Duyck if (unlikely(!data)) 5199451980aSAlexander Duyck return NULL; 5209451980aSAlexander Duyck 5219451980aSAlexander Duyck skb = __build_skb(data, len); 5229451980aSAlexander Duyck if (unlikely(!skb)) { 523181edb2bSAlexander Duyck skb_free_frag(data); 5249451980aSAlexander Duyck return NULL; 5259451980aSAlexander Duyck } 5269451980aSAlexander Duyck 5279451980aSAlexander Duyck if (pfmemalloc) 5289451980aSAlexander Duyck skb->pfmemalloc = 1; 5299451980aSAlexander Duyck skb->head_frag = 1; 5309451980aSAlexander Duyck 531a080e7bdSAlexander Duyck skb_success: 5328af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 5337b2e497aSChristoph Hellwig skb->dev = dev; 534fd11a83dSAlexander Duyck 535a080e7bdSAlexander Duyck skb_fail: 5368af27456SChristoph Hellwig return skb; 5378af27456SChristoph Hellwig } 538b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 5391da177e4SLinus Torvalds 540fd11a83dSAlexander Duyck /** 541fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 542fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 543d7499160SMasanari Iida * @len: length to allocate 544fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 545fd11a83dSAlexander Duyck * 546fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 547fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 548fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 549fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 550fd11a83dSAlexander Duyck * 551fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 552fd11a83dSAlexander Duyck */ 5539451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 5549451980aSAlexander Duyck gfp_t gfp_mask) 555fd11a83dSAlexander Duyck { 5563226b158SEric Dumazet struct napi_alloc_cache *nc; 557fd11a83dSAlexander Duyck struct sk_buff *skb; 5589451980aSAlexander Duyck void *data; 559fd11a83dSAlexander Duyck 5609451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 561fd11a83dSAlexander Duyck 5623226b158SEric Dumazet /* If requested length is either too small or too big, 5633226b158SEric Dumazet * we use kmalloc() for skb->head allocation. 5643226b158SEric Dumazet */ 5653226b158SEric Dumazet if (len <= SKB_WITH_OVERHEAD(1024) || 5663226b158SEric Dumazet len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 567d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 568cfb8ec65SAlexander Lobakin skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, 569cfb8ec65SAlexander Lobakin NUMA_NO_NODE); 570a080e7bdSAlexander Duyck if (!skb) 571a080e7bdSAlexander Duyck goto skb_fail; 572a080e7bdSAlexander Duyck goto skb_success; 573a080e7bdSAlexander Duyck } 5749451980aSAlexander Duyck 5753226b158SEric Dumazet nc = this_cpu_ptr(&napi_alloc_cache); 5769451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5779451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5789451980aSAlexander Duyck 5799451980aSAlexander Duyck if (sk_memalloc_socks()) 5809451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5819451980aSAlexander Duyck 5828c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5839451980aSAlexander Duyck if (unlikely(!data)) 5849451980aSAlexander Duyck return NULL; 5859451980aSAlexander Duyck 586cfb8ec65SAlexander Lobakin skb = __napi_build_skb(data, len); 5879451980aSAlexander Duyck if (unlikely(!skb)) { 588181edb2bSAlexander Duyck skb_free_frag(data); 5899451980aSAlexander Duyck return NULL; 5909451980aSAlexander Duyck } 5919451980aSAlexander Duyck 592795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5939451980aSAlexander Duyck skb->pfmemalloc = 1; 5949451980aSAlexander Duyck skb->head_frag = 1; 5959451980aSAlexander Duyck 596a080e7bdSAlexander Duyck skb_success: 597fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 598fd11a83dSAlexander Duyck skb->dev = napi->dev; 599fd11a83dSAlexander Duyck 600a080e7bdSAlexander Duyck skb_fail: 601fd11a83dSAlexander Duyck return skb; 602fd11a83dSAlexander Duyck } 603fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 604fd11a83dSAlexander Duyck 605654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 60650269e19SEric Dumazet int size, unsigned int truesize) 607654bed16SPeter Zijlstra { 608654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 609654bed16SPeter Zijlstra skb->len += size; 610654bed16SPeter Zijlstra skb->data_len += size; 61150269e19SEric Dumazet skb->truesize += truesize; 612654bed16SPeter Zijlstra } 613654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 614654bed16SPeter Zijlstra 615f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 616f8e617e1SJason Wang unsigned int truesize) 617f8e617e1SJason Wang { 618f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 619f8e617e1SJason Wang 620f8e617e1SJason Wang skb_frag_size_add(frag, size); 621f8e617e1SJason Wang skb->len += size; 622f8e617e1SJason Wang skb->data_len += size; 623f8e617e1SJason Wang skb->truesize += truesize; 624f8e617e1SJason Wang } 625f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 626f8e617e1SJason Wang 62727b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 6281da177e4SLinus Torvalds { 629bd8a7036SEric Dumazet kfree_skb_list(*listp); 63027b437c8SHerbert Xu *listp = NULL; 6311da177e4SLinus Torvalds } 6321da177e4SLinus Torvalds 63327b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 63427b437c8SHerbert Xu { 63527b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 63627b437c8SHerbert Xu } 63727b437c8SHerbert Xu 6381da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 6391da177e4SLinus Torvalds { 6401da177e4SLinus Torvalds struct sk_buff *list; 6411da177e4SLinus Torvalds 642fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 6431da177e4SLinus Torvalds skb_get(list); 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds 646d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 647d3836f21SEric Dumazet { 648181edb2bSAlexander Duyck unsigned char *head = skb->head; 649181edb2bSAlexander Duyck 6506a5bcd84SIlias Apalodimas if (skb->head_frag) { 6516a5bcd84SIlias Apalodimas if (skb_pp_recycle(skb, head)) 6526a5bcd84SIlias Apalodimas return; 653181edb2bSAlexander Duyck skb_free_frag(head); 6546a5bcd84SIlias Apalodimas } else { 655181edb2bSAlexander Duyck kfree(head); 656d3836f21SEric Dumazet } 6576a5bcd84SIlias Apalodimas } 658d3836f21SEric Dumazet 6595bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 6601da177e4SLinus Torvalds { 661ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6621da177e4SLinus Torvalds int i; 663ff04a771SEric Dumazet 664ff04a771SEric Dumazet if (skb->cloned && 665ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 666ff04a771SEric Dumazet &shinfo->dataref)) 6672cc3aeb5SIlias Apalodimas goto exit; 668ff04a771SEric Dumazet 66970c43167SJonathan Lemon skb_zcopy_clear(skb, true); 67070c43167SJonathan Lemon 671ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 6726a5bcd84SIlias Apalodimas __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); 6731da177e4SLinus Torvalds 674ff04a771SEric Dumazet if (shinfo->frag_list) 675ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6761da177e4SLinus Torvalds 677d3836f21SEric Dumazet skb_free_head(skb); 6782cc3aeb5SIlias Apalodimas exit: 6792cc3aeb5SIlias Apalodimas /* When we clone an SKB we copy the reycling bit. The pp_recycle 6802cc3aeb5SIlias Apalodimas * bit is only set on the head though, so in order to avoid races 6812cc3aeb5SIlias Apalodimas * while trying to recycle fragments on __skb_frag_unref() we need 6822cc3aeb5SIlias Apalodimas * to make one SKB responsible for triggering the recycle path. 6832cc3aeb5SIlias Apalodimas * So disable the recycling bit if an SKB is cloned and we have 6842cc3aeb5SIlias Apalodimas * additional references to to the fragmented part of the SKB. 6852cc3aeb5SIlias Apalodimas * Eventually the last SKB will have the recycling bit set and it's 6862cc3aeb5SIlias Apalodimas * dataref set to 0, which will trigger the recycling 6872cc3aeb5SIlias Apalodimas */ 6882cc3aeb5SIlias Apalodimas skb->pp_recycle = 0; 6891da177e4SLinus Torvalds } 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds /* 6921da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6931da177e4SLinus Torvalds */ 6942d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6951da177e4SLinus Torvalds { 696d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 697d179cd12SDavid S. Miller 698d179cd12SDavid S. Miller switch (skb->fclone) { 699d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 7001da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 7016ffe75ebSEric Dumazet return; 702d179cd12SDavid S. Miller 703d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 704d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 7056ffe75ebSEric Dumazet 7066ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 7076ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 7086ffe75ebSEric Dumazet * while here, branch prediction will be good. 7096ffe75ebSEric Dumazet */ 7102638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 7116ffe75ebSEric Dumazet goto fastpath; 712d179cd12SDavid S. Miller break; 713d179cd12SDavid S. Miller 7146ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 715d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 716d179cd12SDavid S. Miller break; 7173ff50b79SStephen Hemminger } 7182638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 7196ffe75ebSEric Dumazet return; 7206ffe75ebSEric Dumazet fastpath: 7216ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds 7240a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 7251da177e4SLinus Torvalds { 726adf30907SEric Dumazet skb_dst_drop(skb); 7271da177e4SLinus Torvalds if (skb->destructor) { 728afa79d08SChangbin Du WARN_ON(in_hardirq()); 7291da177e4SLinus Torvalds skb->destructor(skb); 7301da177e4SLinus Torvalds } 731a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 732cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 7332fc72c7bSKOVACS Krisztian #endif 734df5042f4SFlorian Westphal skb_ext_put(skb); 73504a4bb55SLennert Buytenhek } 73604a4bb55SLennert Buytenhek 73704a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 73804a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 73904a4bb55SLennert Buytenhek { 74004a4bb55SLennert Buytenhek skb_release_head_state(skb); 741a28b1b90SFlorian Westphal if (likely(skb->head)) 7422d4baff8SHerbert Xu skb_release_data(skb); 7432d4baff8SHerbert Xu } 7441da177e4SLinus Torvalds 7452d4baff8SHerbert Xu /** 7462d4baff8SHerbert Xu * __kfree_skb - private function 7472d4baff8SHerbert Xu * @skb: buffer 7482d4baff8SHerbert Xu * 7492d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 7502d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 7512d4baff8SHerbert Xu * always call kfree_skb 7522d4baff8SHerbert Xu */ 7532d4baff8SHerbert Xu 7542d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 7552d4baff8SHerbert Xu { 7562d4baff8SHerbert Xu skb_release_all(skb); 7571da177e4SLinus Torvalds kfree_skbmem(skb); 7581da177e4SLinus Torvalds } 759b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 7601da177e4SLinus Torvalds 7611da177e4SLinus Torvalds /** 762231d06aeSJörn Engel * kfree_skb - free an sk_buff 763231d06aeSJörn Engel * @skb: buffer to free 764231d06aeSJörn Engel * 765231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 766231d06aeSJörn Engel * hit zero. 767231d06aeSJörn Engel */ 768231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 769231d06aeSJörn Engel { 7703889a803SPaolo Abeni if (!skb_unref(skb)) 771231d06aeSJörn Engel return; 7723889a803SPaolo Abeni 773ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 774231d06aeSJörn Engel __kfree_skb(skb); 775231d06aeSJörn Engel } 776b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 777231d06aeSJörn Engel 778bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 779bd8a7036SEric Dumazet { 780bd8a7036SEric Dumazet while (segs) { 781bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 782bd8a7036SEric Dumazet 783bd8a7036SEric Dumazet kfree_skb(segs); 784bd8a7036SEric Dumazet segs = next; 785bd8a7036SEric Dumazet } 786bd8a7036SEric Dumazet } 787bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 788bd8a7036SEric Dumazet 7896413139dSWillem de Bruijn /* Dump skb information and contents. 7906413139dSWillem de Bruijn * 7916413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 7926413139dSWillem de Bruijn * 793302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 7946413139dSWillem de Bruijn */ 7956413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 7966413139dSWillem de Bruijn { 7976413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 7986413139dSWillem de Bruijn struct net_device *dev = skb->dev; 7996413139dSWillem de Bruijn struct sock *sk = skb->sk; 8006413139dSWillem de Bruijn struct sk_buff *list_skb; 8016413139dSWillem de Bruijn bool has_mac, has_trans; 8026413139dSWillem de Bruijn int headroom, tailroom; 8036413139dSWillem de Bruijn int i, len, seg_len; 8046413139dSWillem de Bruijn 8056413139dSWillem de Bruijn if (full_pkt) 8066413139dSWillem de Bruijn len = skb->len; 8076413139dSWillem de Bruijn else 8086413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 8096413139dSWillem de Bruijn 8106413139dSWillem de Bruijn headroom = skb_headroom(skb); 8116413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 8126413139dSWillem de Bruijn 8136413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 8146413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 8156413139dSWillem de Bruijn 8166413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 8176413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 8186413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 8196413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 8206413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 8216413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 8226413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 8236413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 8246413139dSWillem de Bruijn skb->network_header, 8256413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 8266413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 8276413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 8286413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 8296413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 8306413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 8316413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 8326413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 8336413139dSWillem de Bruijn 8346413139dSWillem de Bruijn if (dev) 8356413139dSWillem de Bruijn printk("%sdev name=%s feat=0x%pNF\n", 8366413139dSWillem de Bruijn level, dev->name, &dev->features); 8376413139dSWillem de Bruijn if (sk) 838db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 8396413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 8406413139dSWillem de Bruijn 8416413139dSWillem de Bruijn if (full_pkt && headroom) 8426413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 8436413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 8446413139dSWillem de Bruijn 8456413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 8466413139dSWillem de Bruijn if (seg_len) 8476413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 8486413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 8496413139dSWillem de Bruijn len -= seg_len; 8506413139dSWillem de Bruijn 8516413139dSWillem de Bruijn if (full_pkt && tailroom) 8526413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 8536413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 8546413139dSWillem de Bruijn 8556413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 8566413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 8576413139dSWillem de Bruijn u32 p_off, p_len, copied; 8586413139dSWillem de Bruijn struct page *p; 8596413139dSWillem de Bruijn u8 *vaddr; 8606413139dSWillem de Bruijn 861b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 8626413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 8636413139dSWillem de Bruijn copied) { 8646413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 8656413139dSWillem de Bruijn vaddr = kmap_atomic(p); 8666413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 8676413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 8686413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 8696413139dSWillem de Bruijn kunmap_atomic(vaddr); 8706413139dSWillem de Bruijn len -= seg_len; 8716413139dSWillem de Bruijn if (!len) 8726413139dSWillem de Bruijn break; 8736413139dSWillem de Bruijn } 8746413139dSWillem de Bruijn } 8756413139dSWillem de Bruijn 8766413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8776413139dSWillem de Bruijn printk("skb fraglist:\n"); 8786413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8796413139dSWillem de Bruijn skb_dump(level, list_skb, true); 8806413139dSWillem de Bruijn } 8816413139dSWillem de Bruijn } 8826413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 8836413139dSWillem de Bruijn 884d1a203eaSStephen Hemminger /** 88525121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 88625121173SMichael S. Tsirkin * @skb: buffer that triggered an error 88725121173SMichael S. Tsirkin * 88825121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 88925121173SMichael S. Tsirkin * skb must be freed afterwards. 89025121173SMichael S. Tsirkin */ 89125121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 89225121173SMichael S. Tsirkin { 8931f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 89425121173SMichael S. Tsirkin } 89525121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 89625121173SMichael S. Tsirkin 897be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 89825121173SMichael S. Tsirkin /** 899ead2ceb0SNeil Horman * consume_skb - free an skbuff 900ead2ceb0SNeil Horman * @skb: buffer to free 901ead2ceb0SNeil Horman * 902ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 903ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 904ead2ceb0SNeil Horman * is being dropped after a failure and notes that 905ead2ceb0SNeil Horman */ 906ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 907ead2ceb0SNeil Horman { 9083889a803SPaolo Abeni if (!skb_unref(skb)) 909ead2ceb0SNeil Horman return; 9103889a803SPaolo Abeni 91107dc22e7SKoki Sanagi trace_consume_skb(skb); 912ead2ceb0SNeil Horman __kfree_skb(skb); 913ead2ceb0SNeil Horman } 914ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 915be769db2SHerbert Xu #endif 916ead2ceb0SNeil Horman 9170a463c78SPaolo Abeni /** 918c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 9190a463c78SPaolo Abeni * @skb: buffer to free 9200a463c78SPaolo Abeni * 921ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 922ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 9230a463c78SPaolo Abeni */ 924ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 9250a463c78SPaolo Abeni { 9260a463c78SPaolo Abeni trace_consume_skb(skb); 9270a463c78SPaolo Abeni skb_release_data(skb); 9280a463c78SPaolo Abeni kfree_skbmem(skb); 9290a463c78SPaolo Abeni } 9300a463c78SPaolo Abeni 931f450d539SAlexander Lobakin static void napi_skb_cache_put(struct sk_buff *skb) 932795bb1c0SJesper Dangaard Brouer { 933795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 934f450d539SAlexander Lobakin u32 i; 935795bb1c0SJesper Dangaard Brouer 936f450d539SAlexander Lobakin kasan_poison_object_data(skbuff_head_cache, skb); 937795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 938795bb1c0SJesper Dangaard Brouer 939795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 940f450d539SAlexander Lobakin for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++) 941f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, 942f450d539SAlexander Lobakin nc->skb_cache[i]); 943f450d539SAlexander Lobakin 944f450d539SAlexander Lobakin kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF, 945f450d539SAlexander Lobakin nc->skb_cache + NAPI_SKB_CACHE_HALF); 946f450d539SAlexander Lobakin nc->skb_count = NAPI_SKB_CACHE_HALF; 947795bb1c0SJesper Dangaard Brouer } 948795bb1c0SJesper Dangaard Brouer } 949f450d539SAlexander Lobakin 95015fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 95115fad714SJesper Dangaard Brouer { 9529243adfcSAlexander Lobakin skb_release_all(skb); 9539243adfcSAlexander Lobakin napi_skb_cache_put(skb); 9549243adfcSAlexander Lobakin } 9559243adfcSAlexander Lobakin 9569243adfcSAlexander Lobakin void napi_skb_free_stolen_head(struct sk_buff *skb) 9579243adfcSAlexander Lobakin { 9589efb4b5bSPaolo Abeni if (unlikely(skb->slow_gro)) { 9598550ff8dSPaul Blakey nf_reset_ct(skb); 9609243adfcSAlexander Lobakin skb_dst_drop(skb); 9619243adfcSAlexander Lobakin skb_ext_put(skb); 9625e10da53SPaolo Abeni skb_orphan(skb); 9639efb4b5bSPaolo Abeni skb->slow_gro = 0; 9649efb4b5bSPaolo Abeni } 965f450d539SAlexander Lobakin napi_skb_cache_put(skb); 96615fad714SJesper Dangaard Brouer } 967795bb1c0SJesper Dangaard Brouer 968795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 969795bb1c0SJesper Dangaard Brouer { 970885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 971795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 972885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 973795bb1c0SJesper Dangaard Brouer return; 974795bb1c0SJesper Dangaard Brouer } 975795bb1c0SJesper Dangaard Brouer 9766454eca8SYunsheng Lin lockdep_assert_in_softirq(); 9776454eca8SYunsheng Lin 9787608894eSPaolo Abeni if (!skb_unref(skb)) 979795bb1c0SJesper Dangaard Brouer return; 9807608894eSPaolo Abeni 981795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 982795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 983795bb1c0SJesper Dangaard Brouer 984795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 985abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 986795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 987795bb1c0SJesper Dangaard Brouer return; 988795bb1c0SJesper Dangaard Brouer } 989795bb1c0SJesper Dangaard Brouer 9909243adfcSAlexander Lobakin skb_release_all(skb); 991f450d539SAlexander Lobakin napi_skb_cache_put(skb); 992795bb1c0SJesper Dangaard Brouer } 993795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 994795bb1c0SJesper Dangaard Brouer 995*03f61041SKees Cook /* Make sure a field is contained by headers group */ 996b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 997*03f61041SKees Cook BUILD_BUG_ON(offsetof(struct sk_buff, field) != \ 998*03f61041SKees Cook offsetof(struct sk_buff, headers.field)); \ 999b1937227SEric Dumazet 1000dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 1001dec18810SHerbert Xu { 1002dec18810SHerbert Xu new->tstamp = old->tstamp; 1003b1937227SEric Dumazet /* We do not copy old->sk */ 1004dec18810SHerbert Xu new->dev = old->dev; 1005b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 10067fee226aSEric Dumazet skb_dst_copy(new, old); 1007df5042f4SFlorian Westphal __skb_ext_copy(new, old); 1008b1937227SEric Dumazet __nf_copy(new, old, false); 10096aa895b0SPatrick McHardy 1010*03f61041SKees Cook /* Note : this field could be in the headers group. 1011b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 1012b1937227SEric Dumazet */ 1013b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 101406021292SEliezer Tamir 1015*03f61041SKees Cook memcpy(&new->headers, &old->headers, sizeof(new->headers)); 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); 18047f678defSVasily Averin int osize = skb_end_offset(skb); 18057f678defSVasily Averin struct sock *sk = skb->sk; 1806f1260ff1SVasily Averin 1807f1260ff1SVasily Averin if (WARN_ONCE(delta <= 0, 1808f1260ff1SVasily Averin "%s is expecting an increase in the headroom", __func__)) 1809f1260ff1SVasily Averin return skb; 1810f1260ff1SVasily Averin 18117f678defSVasily Averin delta = SKB_DATA_ALIGN(delta); 18127f678defSVasily Averin /* pskb_expand_head() might crash, if skb is shared. */ 18137f678defSVasily Averin if (skb_shared(skb) || !is_skb_wmem(skb)) { 1814f1260ff1SVasily Averin struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1815f1260ff1SVasily Averin 18167f678defSVasily Averin if (unlikely(!nskb)) 18177f678defSVasily Averin goto fail; 18187f678defSVasily Averin 18197f678defSVasily Averin if (sk) 18207f678defSVasily Averin skb_set_owner_w(nskb, sk); 1821f1260ff1SVasily Averin consume_skb(skb); 1822f1260ff1SVasily Averin skb = nskb; 1823f1260ff1SVasily Averin } 18247f678defSVasily Averin if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC)) 18257f678defSVasily Averin goto fail; 18267f678defSVasily Averin 18277f678defSVasily Averin if (sk && is_skb_wmem(skb)) { 18287f678defSVasily Averin delta = skb_end_offset(skb) - osize; 18297f678defSVasily Averin refcount_add(delta, &sk->sk_wmem_alloc); 18307f678defSVasily Averin skb->truesize += delta; 1831f1260ff1SVasily Averin } 1832f1260ff1SVasily Averin return skb; 18337f678defSVasily Averin 18347f678defSVasily Averin fail: 18357f678defSVasily Averin kfree_skb(skb); 18367f678defSVasily Averin return NULL; 1837f1260ff1SVasily Averin } 1838f1260ff1SVasily Averin EXPORT_SYMBOL(skb_expand_head); 1839f1260ff1SVasily Averin 1840f1260ff1SVasily Averin /** 18411da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 18421da177e4SLinus Torvalds * @skb: buffer to copy 18431da177e4SLinus Torvalds * @newheadroom: new free bytes at head 18441da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 18451da177e4SLinus Torvalds * @gfp_mask: allocation priority 18461da177e4SLinus Torvalds * 18471da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 18481da177e4SLinus Torvalds * allocate additional space. 18491da177e4SLinus Torvalds * 18501da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 18511da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 18521da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 18531da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 18541da177e4SLinus Torvalds * 18551da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 18561da177e4SLinus Torvalds * is called from an interrupt. 18571da177e4SLinus Torvalds */ 18581da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 185986a76cafSVictor Fusco int newheadroom, int newtailroom, 1860dd0fc66fSAl Viro gfp_t gfp_mask) 18611da177e4SLinus Torvalds { 18621da177e4SLinus Torvalds /* 18631da177e4SLinus Torvalds * Allocate the copy buffer 18641da177e4SLinus Torvalds */ 1865c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1866c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1867c93bdd0eSMel Gorman NUMA_NO_NODE); 1868efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 18691da177e4SLinus Torvalds int head_copy_len, head_copy_off; 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds if (!n) 18721da177e4SLinus Torvalds return NULL; 18731da177e4SLinus Torvalds 18741da177e4SLinus Torvalds skb_reserve(n, newheadroom); 18751da177e4SLinus Torvalds 18761da177e4SLinus Torvalds /* Set the tail pointer and length */ 18771da177e4SLinus Torvalds skb_put(n, skb->len); 18781da177e4SLinus Torvalds 1879efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 18801da177e4SLinus Torvalds head_copy_off = 0; 18811da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 18821da177e4SLinus Torvalds head_copy_len = newheadroom; 18831da177e4SLinus Torvalds else 18841da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 18851da177e4SLinus Torvalds 18861da177e4SLinus Torvalds /* Copy the linear header and data. */ 18879f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 18889f77fad3STim Hansen skb->len + head_copy_len)); 18891da177e4SLinus Torvalds 189008303c18SIlya Lesokhin skb_copy_header(n, skb); 18911da177e4SLinus Torvalds 1892030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1893efd1e8d5SPatrick McHardy 18941da177e4SLinus Torvalds return n; 18951da177e4SLinus Torvalds } 1896b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 18971da177e4SLinus Torvalds 18981da177e4SLinus Torvalds /** 1899cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 19001da177e4SLinus Torvalds * @skb: buffer to pad 19011da177e4SLinus Torvalds * @pad: space to pad 1902cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 19031da177e4SLinus Torvalds * 19041da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 19051da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 19061da177e4SLinus Torvalds * beyond the buffer end onto the wire. 19071da177e4SLinus Torvalds * 1908cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1909cd0a137aSFlorian Fainelli * if @free_on_error is true. 19101da177e4SLinus Torvalds */ 19111da177e4SLinus Torvalds 1912cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 19131da177e4SLinus Torvalds { 19145b057c6bSHerbert Xu int err; 19155b057c6bSHerbert Xu int ntail; 19161da177e4SLinus Torvalds 19171da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 19185b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 19191da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 19205b057c6bSHerbert Xu return 0; 19211da177e4SLinus Torvalds } 19221da177e4SLinus Torvalds 19234305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 19245b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 19255b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 19265b057c6bSHerbert Xu if (unlikely(err)) 19275b057c6bSHerbert Xu goto free_skb; 19285b057c6bSHerbert Xu } 19295b057c6bSHerbert Xu 19305b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 19315b057c6bSHerbert Xu * to be audited. 19325b057c6bSHerbert Xu */ 19335b057c6bSHerbert Xu err = skb_linearize(skb); 19345b057c6bSHerbert Xu if (unlikely(err)) 19355b057c6bSHerbert Xu goto free_skb; 19365b057c6bSHerbert Xu 19375b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 19385b057c6bSHerbert Xu return 0; 19395b057c6bSHerbert Xu 19405b057c6bSHerbert Xu free_skb: 1941cd0a137aSFlorian Fainelli if (free_on_error) 19421da177e4SLinus Torvalds kfree_skb(skb); 19435b057c6bSHerbert Xu return err; 19441da177e4SLinus Torvalds } 1945cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 19461da177e4SLinus Torvalds 19470dde3e16SIlpo Järvinen /** 19480c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 19490c7ddf36SMathias Krause * @skb: start of the buffer to use 19500c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 19510c7ddf36SMathias Krause * @len: amount of data to add 19520c7ddf36SMathias Krause * 19530c7ddf36SMathias Krause * This function extends the used data area of the potentially 19540c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 19550c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 19560c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 19570c7ddf36SMathias Krause * returned. 19580c7ddf36SMathias Krause */ 19590c7ddf36SMathias Krause 19604df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 19610c7ddf36SMathias Krause { 19620c7ddf36SMathias Krause if (tail != skb) { 19630c7ddf36SMathias Krause skb->data_len += len; 19640c7ddf36SMathias Krause skb->len += len; 19650c7ddf36SMathias Krause } 19660c7ddf36SMathias Krause return skb_put(tail, len); 19670c7ddf36SMathias Krause } 19680c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 19690c7ddf36SMathias Krause 19700c7ddf36SMathias Krause /** 19710dde3e16SIlpo Järvinen * skb_put - add data to a buffer 19720dde3e16SIlpo Järvinen * @skb: buffer to use 19730dde3e16SIlpo Järvinen * @len: amount of data to add 19740dde3e16SIlpo Järvinen * 19750dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 19760dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 19770dde3e16SIlpo Järvinen * first byte of the extra data is returned. 19780dde3e16SIlpo Järvinen */ 19794df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 19800dde3e16SIlpo Järvinen { 19814df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 19820dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 19830dde3e16SIlpo Järvinen skb->tail += len; 19840dde3e16SIlpo Järvinen skb->len += len; 19850dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 19860dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 19870dde3e16SIlpo Järvinen return tmp; 19880dde3e16SIlpo Järvinen } 19890dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 19900dde3e16SIlpo Järvinen 19916be8ac2fSIlpo Järvinen /** 1992c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1993c2aa270aSIlpo Järvinen * @skb: buffer to use 1994c2aa270aSIlpo Järvinen * @len: amount of data to add 1995c2aa270aSIlpo Järvinen * 1996c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1997c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1998c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1999c2aa270aSIlpo Järvinen */ 2000d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 2001c2aa270aSIlpo Järvinen { 2002c2aa270aSIlpo Järvinen skb->data -= len; 2003c2aa270aSIlpo Järvinen skb->len += len; 2004c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 2005c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 2006c2aa270aSIlpo Järvinen return skb->data; 2007c2aa270aSIlpo Järvinen } 2008c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 2009c2aa270aSIlpo Järvinen 2010c2aa270aSIlpo Järvinen /** 20116be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 20126be8ac2fSIlpo Järvinen * @skb: buffer to use 20136be8ac2fSIlpo Järvinen * @len: amount of data to remove 20146be8ac2fSIlpo Järvinen * 20156be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 20166be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 20176be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 20186be8ac2fSIlpo Järvinen * the old data. 20196be8ac2fSIlpo Järvinen */ 2020af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 20216be8ac2fSIlpo Järvinen { 202247d29646SDavid S. Miller return skb_pull_inline(skb, len); 20236be8ac2fSIlpo Järvinen } 20246be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 20256be8ac2fSIlpo Järvinen 2026419ae74eSIlpo Järvinen /** 2027419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 2028419ae74eSIlpo Järvinen * @skb: buffer to alter 2029419ae74eSIlpo Järvinen * @len: new length 2030419ae74eSIlpo Järvinen * 2031419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 2032419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 2033419ae74eSIlpo Järvinen * The skb must be linear. 2034419ae74eSIlpo Järvinen */ 2035419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 2036419ae74eSIlpo Järvinen { 2037419ae74eSIlpo Järvinen if (skb->len > len) 2038419ae74eSIlpo Järvinen __skb_trim(skb, len); 2039419ae74eSIlpo Järvinen } 2040419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 2041419ae74eSIlpo Järvinen 20423cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 20431da177e4SLinus Torvalds */ 20441da177e4SLinus Torvalds 20453cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 20461da177e4SLinus Torvalds { 204727b437c8SHerbert Xu struct sk_buff **fragp; 204827b437c8SHerbert Xu struct sk_buff *frag; 20491da177e4SLinus Torvalds int offset = skb_headlen(skb); 20501da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 20511da177e4SLinus Torvalds int i; 205227b437c8SHerbert Xu int err; 205327b437c8SHerbert Xu 205427b437c8SHerbert Xu if (skb_cloned(skb) && 205527b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 205627b437c8SHerbert Xu return err; 20571da177e4SLinus Torvalds 2058f4d26fb3SHerbert Xu i = 0; 2059f4d26fb3SHerbert Xu if (offset >= len) 2060f4d26fb3SHerbert Xu goto drop_pages; 2061f4d26fb3SHerbert Xu 2062f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 20639e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 206427b437c8SHerbert Xu 206527b437c8SHerbert Xu if (end < len) { 20661da177e4SLinus Torvalds offset = end; 206727b437c8SHerbert Xu continue; 20681da177e4SLinus Torvalds } 20691da177e4SLinus Torvalds 20709e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 207127b437c8SHerbert Xu 2072f4d26fb3SHerbert Xu drop_pages: 207327b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 207427b437c8SHerbert Xu 207527b437c8SHerbert Xu for (; i < nfrags; i++) 2076ea2ab693SIan Campbell skb_frag_unref(skb, i); 207727b437c8SHerbert Xu 207821dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 207927b437c8SHerbert Xu skb_drop_fraglist(skb); 2080f4d26fb3SHerbert Xu goto done; 208127b437c8SHerbert Xu } 208227b437c8SHerbert Xu 208327b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 208427b437c8SHerbert Xu fragp = &frag->next) { 208527b437c8SHerbert Xu int end = offset + frag->len; 208627b437c8SHerbert Xu 208727b437c8SHerbert Xu if (skb_shared(frag)) { 208827b437c8SHerbert Xu struct sk_buff *nfrag; 208927b437c8SHerbert Xu 209027b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 209127b437c8SHerbert Xu if (unlikely(!nfrag)) 209227b437c8SHerbert Xu return -ENOMEM; 209327b437c8SHerbert Xu 209427b437c8SHerbert Xu nfrag->next = frag->next; 209585bb2a60SEric Dumazet consume_skb(frag); 209627b437c8SHerbert Xu frag = nfrag; 209727b437c8SHerbert Xu *fragp = frag; 209827b437c8SHerbert Xu } 209927b437c8SHerbert Xu 210027b437c8SHerbert Xu if (end < len) { 210127b437c8SHerbert Xu offset = end; 210227b437c8SHerbert Xu continue; 210327b437c8SHerbert Xu } 210427b437c8SHerbert Xu 210527b437c8SHerbert Xu if (end > len && 210627b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 210727b437c8SHerbert Xu return err; 210827b437c8SHerbert Xu 210927b437c8SHerbert Xu if (frag->next) 211027b437c8SHerbert Xu skb_drop_list(&frag->next); 211127b437c8SHerbert Xu break; 211227b437c8SHerbert Xu } 211327b437c8SHerbert Xu 2114f4d26fb3SHerbert Xu done: 211527b437c8SHerbert Xu if (len > skb_headlen(skb)) { 21161da177e4SLinus Torvalds skb->data_len -= skb->len - len; 21171da177e4SLinus Torvalds skb->len = len; 21181da177e4SLinus Torvalds } else { 21191da177e4SLinus Torvalds skb->len = len; 21201da177e4SLinus Torvalds skb->data_len = 0; 212127a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 21221da177e4SLinus Torvalds } 21231da177e4SLinus Torvalds 2124c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2125c21b48ccSEric Dumazet skb_condense(skb); 21261da177e4SLinus Torvalds return 0; 21271da177e4SLinus Torvalds } 2128b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 21291da177e4SLinus Torvalds 213088078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 213188078d98SEric Dumazet */ 213288078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 213388078d98SEric Dumazet { 213488078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 213588078d98SEric Dumazet int delta = skb->len - len; 213688078d98SEric Dumazet 2137d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2138d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2139d55bef50SDimitris Michailidis len); 214054970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 214154970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 214254970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 214354970a2fSVasily Averin 214454970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 214554970a2fSVasily Averin return -EINVAL; 214688078d98SEric Dumazet } 214788078d98SEric Dumazet return __pskb_trim(skb, len); 214888078d98SEric Dumazet } 214988078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 215088078d98SEric Dumazet 21511da177e4SLinus Torvalds /** 21521da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 21531da177e4SLinus Torvalds * @skb: buffer to reallocate 21541da177e4SLinus Torvalds * @delta: number of bytes to advance tail 21551da177e4SLinus Torvalds * 21561da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 21571da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 21581da177e4SLinus Torvalds * data from fragmented part. 21591da177e4SLinus Torvalds * 21601da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 21611da177e4SLinus Torvalds * 21621da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 21631da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 21641da177e4SLinus Torvalds * 21651da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 21661da177e4SLinus Torvalds * reloaded after call to this function. 21671da177e4SLinus Torvalds */ 21681da177e4SLinus Torvalds 21691da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 21701da177e4SLinus Torvalds * when it is necessary. 21711da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 21721da177e4SLinus Torvalds * 2. It may change skb pointers. 21731da177e4SLinus Torvalds * 21741da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 21751da177e4SLinus Torvalds */ 2176af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 21771da177e4SLinus Torvalds { 21781da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 21791da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 21801da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 21811da177e4SLinus Torvalds */ 21824305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 21831da177e4SLinus Torvalds 21841da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 21851da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 21861da177e4SLinus Torvalds GFP_ATOMIC)) 21871da177e4SLinus Torvalds return NULL; 21881da177e4SLinus Torvalds } 21891da177e4SLinus Torvalds 21909f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 21919f77fad3STim Hansen skb_tail_pointer(skb), delta)); 21921da177e4SLinus Torvalds 21931da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 21941da177e4SLinus Torvalds * size of pulled pages. Superb. 21951da177e4SLinus Torvalds */ 219621dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 21971da177e4SLinus Torvalds goto pull_pages; 21981da177e4SLinus Torvalds 21991da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 22001da177e4SLinus Torvalds eat = delta; 22011da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22029e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22039e903e08SEric Dumazet 22049e903e08SEric Dumazet if (size >= eat) 22051da177e4SLinus Torvalds goto pull_pages; 22069e903e08SEric Dumazet eat -= size; 22071da177e4SLinus Torvalds } 22081da177e4SLinus Torvalds 22091da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 221009001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 22111da177e4SLinus Torvalds * but taking into account that pulling is expected to 22121da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 22131da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 22141da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 22151da177e4SLinus Torvalds */ 22161da177e4SLinus Torvalds if (eat) { 22171da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 22181da177e4SLinus Torvalds struct sk_buff *clone = NULL; 22191da177e4SLinus Torvalds struct sk_buff *insp = NULL; 22201da177e4SLinus Torvalds 22211da177e4SLinus Torvalds do { 22221da177e4SLinus Torvalds if (list->len <= eat) { 22231da177e4SLinus Torvalds /* Eaten as whole. */ 22241da177e4SLinus Torvalds eat -= list->len; 22251da177e4SLinus Torvalds list = list->next; 22261da177e4SLinus Torvalds insp = list; 22271da177e4SLinus Torvalds } else { 22281da177e4SLinus Torvalds /* Eaten partially. */ 22291da177e4SLinus Torvalds 22301da177e4SLinus Torvalds if (skb_shared(list)) { 22311da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 22321da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 22331da177e4SLinus Torvalds if (!clone) 22341da177e4SLinus Torvalds return NULL; 22351da177e4SLinus Torvalds insp = list->next; 22361da177e4SLinus Torvalds list = clone; 22371da177e4SLinus Torvalds } else { 22381da177e4SLinus Torvalds /* This may be pulled without 22391da177e4SLinus Torvalds * problems. */ 22401da177e4SLinus Torvalds insp = list; 22411da177e4SLinus Torvalds } 22421da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 22431da177e4SLinus Torvalds kfree_skb(clone); 22441da177e4SLinus Torvalds return NULL; 22451da177e4SLinus Torvalds } 22461da177e4SLinus Torvalds break; 22471da177e4SLinus Torvalds } 22481da177e4SLinus Torvalds } while (eat); 22491da177e4SLinus Torvalds 22501da177e4SLinus Torvalds /* Free pulled out fragments. */ 22511da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 22521da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 22531da177e4SLinus Torvalds kfree_skb(list); 22541da177e4SLinus Torvalds } 22551da177e4SLinus Torvalds /* And insert new clone at head. */ 22561da177e4SLinus Torvalds if (clone) { 22571da177e4SLinus Torvalds clone->next = list; 22581da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 22591da177e4SLinus Torvalds } 22601da177e4SLinus Torvalds } 22611da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 22621da177e4SLinus Torvalds 22631da177e4SLinus Torvalds pull_pages: 22641da177e4SLinus Torvalds eat = delta; 22651da177e4SLinus Torvalds k = 0; 22661da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22679e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22689e903e08SEric Dumazet 22699e903e08SEric Dumazet if (size <= eat) { 2270ea2ab693SIan Campbell skb_frag_unref(skb, i); 22719e903e08SEric Dumazet eat -= size; 22721da177e4SLinus Torvalds } else { 2273b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2274b54c9d5bSJonathan Lemon 2275b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 22761da177e4SLinus Torvalds if (eat) { 2277b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2278b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 22793ccc6c6fSlinzhang if (!i) 22803ccc6c6fSlinzhang goto end; 22811da177e4SLinus Torvalds eat = 0; 22821da177e4SLinus Torvalds } 22831da177e4SLinus Torvalds k++; 22841da177e4SLinus Torvalds } 22851da177e4SLinus Torvalds } 22861da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 22871da177e4SLinus Torvalds 22883ccc6c6fSlinzhang end: 22891da177e4SLinus Torvalds skb->tail += delta; 22901da177e4SLinus Torvalds skb->data_len -= delta; 22911da177e4SLinus Torvalds 22921f8b977aSWillem de Bruijn if (!skb->data_len) 22931f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 22941f8b977aSWillem de Bruijn 229527a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 22961da177e4SLinus Torvalds } 2297b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 22981da177e4SLinus Torvalds 229922019b17SEric Dumazet /** 230022019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 230122019b17SEric Dumazet * @skb: source skb 230222019b17SEric Dumazet * @offset: offset in source 230322019b17SEric Dumazet * @to: destination buffer 230422019b17SEric Dumazet * @len: number of bytes to copy 230522019b17SEric Dumazet * 230622019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 230722019b17SEric Dumazet * destination buffer. 230822019b17SEric Dumazet * 230922019b17SEric Dumazet * CAUTION ! : 231022019b17SEric Dumazet * If its prototype is ever changed, 231122019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 231222019b17SEric Dumazet * since it is called from BPF assembly code. 231322019b17SEric Dumazet */ 23141da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 23151da177e4SLinus Torvalds { 23161a028e50SDavid S. Miller int start = skb_headlen(skb); 2317fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2318fbb398a8SDavid S. Miller int i, copy; 23191da177e4SLinus Torvalds 23201da177e4SLinus Torvalds if (offset > (int)skb->len - len) 23211da177e4SLinus Torvalds goto fault; 23221da177e4SLinus Torvalds 23231da177e4SLinus Torvalds /* Copy header. */ 23241a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 23251da177e4SLinus Torvalds if (copy > len) 23261da177e4SLinus Torvalds copy = len; 2327d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 23281da177e4SLinus Torvalds if ((len -= copy) == 0) 23291da177e4SLinus Torvalds return 0; 23301da177e4SLinus Torvalds offset += copy; 23311da177e4SLinus Torvalds to += copy; 23321da177e4SLinus Torvalds } 23331da177e4SLinus Torvalds 23341da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 23351a028e50SDavid S. Miller int end; 233651c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 23371da177e4SLinus Torvalds 2338547b792cSIlpo Järvinen WARN_ON(start > offset + len); 23391a028e50SDavid S. Miller 234051c56b00SEric Dumazet end = start + skb_frag_size(f); 23411da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2342c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2343c613c209SWillem de Bruijn struct page *p; 23441da177e4SLinus Torvalds u8 *vaddr; 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds if (copy > len) 23471da177e4SLinus Torvalds copy = len; 23481da177e4SLinus Torvalds 2349c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2350b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2351c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2352c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2353c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 235451c56b00SEric Dumazet kunmap_atomic(vaddr); 2355c613c209SWillem de Bruijn } 23561da177e4SLinus Torvalds 23571da177e4SLinus Torvalds if ((len -= copy) == 0) 23581da177e4SLinus Torvalds return 0; 23591da177e4SLinus Torvalds offset += copy; 23601da177e4SLinus Torvalds to += copy; 23611da177e4SLinus Torvalds } 23621a028e50SDavid S. Miller start = end; 23631da177e4SLinus Torvalds } 23641da177e4SLinus Torvalds 2365fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 23661a028e50SDavid S. Miller int end; 23671da177e4SLinus Torvalds 2368547b792cSIlpo Järvinen WARN_ON(start > offset + len); 23691a028e50SDavid S. Miller 2370fbb398a8SDavid S. Miller end = start + frag_iter->len; 23711da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 23721da177e4SLinus Torvalds if (copy > len) 23731da177e4SLinus Torvalds copy = len; 2374fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 23751da177e4SLinus Torvalds goto fault; 23761da177e4SLinus Torvalds if ((len -= copy) == 0) 23771da177e4SLinus Torvalds return 0; 23781da177e4SLinus Torvalds offset += copy; 23791da177e4SLinus Torvalds to += copy; 23801da177e4SLinus Torvalds } 23811a028e50SDavid S. Miller start = end; 23821da177e4SLinus Torvalds } 2383a6686f2fSShirley Ma 23841da177e4SLinus Torvalds if (!len) 23851da177e4SLinus Torvalds return 0; 23861da177e4SLinus Torvalds 23871da177e4SLinus Torvalds fault: 23881da177e4SLinus Torvalds return -EFAULT; 23891da177e4SLinus Torvalds } 2390b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 23911da177e4SLinus Torvalds 23929c55e01cSJens Axboe /* 23939c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 23949c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 23959c55e01cSJens Axboe */ 23969c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 23979c55e01cSJens Axboe { 23988b9d3728SJarek Poplawski put_page(spd->pages[i]); 23998b9d3728SJarek Poplawski } 24009c55e01cSJens Axboe 2401a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 24024fb66994SJarek Poplawski unsigned int *offset, 240318aafc62SEric Dumazet struct sock *sk) 24048b9d3728SJarek Poplawski { 24055640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 24068b9d3728SJarek Poplawski 24075640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 24088b9d3728SJarek Poplawski return NULL; 24094fb66994SJarek Poplawski 24105640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 24114fb66994SJarek Poplawski 24125640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 24135640f768SEric Dumazet page_address(page) + *offset, *len); 24145640f768SEric Dumazet *offset = pfrag->offset; 24155640f768SEric Dumazet pfrag->offset += *len; 24164fb66994SJarek Poplawski 24175640f768SEric Dumazet return pfrag->page; 24189c55e01cSJens Axboe } 24199c55e01cSJens Axboe 242041c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 242141c73a0dSEric Dumazet struct page *page, 242241c73a0dSEric Dumazet unsigned int offset) 242341c73a0dSEric Dumazet { 242441c73a0dSEric Dumazet return spd->nr_pages && 242541c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 242641c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 242741c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 242841c73a0dSEric Dumazet } 242941c73a0dSEric Dumazet 24309c55e01cSJens Axboe /* 24319c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 24329c55e01cSJens Axboe */ 2433a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 243435f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 24354fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 243618aafc62SEric Dumazet bool linear, 24377a67e56fSJarek Poplawski struct sock *sk) 24389c55e01cSJens Axboe { 243941c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2440a108d5f3SDavid S. Miller return true; 24419c55e01cSJens Axboe 24428b9d3728SJarek Poplawski if (linear) { 244318aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 24448b9d3728SJarek Poplawski if (!page) 2445a108d5f3SDavid S. Miller return true; 244641c73a0dSEric Dumazet } 244741c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 244841c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2449a108d5f3SDavid S. Miller return false; 245041c73a0dSEric Dumazet } 24518b9d3728SJarek Poplawski get_page(page); 24529c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 24534fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 24549c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 24559c55e01cSJens Axboe spd->nr_pages++; 24568b9d3728SJarek Poplawski 2457a108d5f3SDavid S. Miller return false; 24589c55e01cSJens Axboe } 24599c55e01cSJens Axboe 2460a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 24612870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 246218aafc62SEric Dumazet unsigned int *len, 2463d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 246435f3d14dSJens Axboe struct sock *sk, 246535f3d14dSJens Axboe struct pipe_inode_info *pipe) 24669c55e01cSJens Axboe { 24672870c43dSOctavian Purdila if (!*len) 2468a108d5f3SDavid S. Miller return true; 24699c55e01cSJens Axboe 24702870c43dSOctavian Purdila /* skip this segment if already processed */ 24712870c43dSOctavian Purdila if (*off >= plen) { 24722870c43dSOctavian Purdila *off -= plen; 2473a108d5f3SDavid S. Miller return false; 24742870c43dSOctavian Purdila } 24752870c43dSOctavian Purdila 24762870c43dSOctavian Purdila /* ignore any bits we already processed */ 24779ca1b22dSEric Dumazet poff += *off; 24789ca1b22dSEric Dumazet plen -= *off; 24792870c43dSOctavian Purdila *off = 0; 24802870c43dSOctavian Purdila 248118aafc62SEric Dumazet do { 248218aafc62SEric Dumazet unsigned int flen = min(*len, plen); 24832870c43dSOctavian Purdila 248418aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 248518aafc62SEric Dumazet linear, sk)) 2486a108d5f3SDavid S. Miller return true; 248718aafc62SEric Dumazet poff += flen; 248818aafc62SEric Dumazet plen -= flen; 24892870c43dSOctavian Purdila *len -= flen; 249018aafc62SEric Dumazet } while (*len && plen); 24912870c43dSOctavian Purdila 2492a108d5f3SDavid S. Miller return false; 2493db43a282SOctavian Purdila } 24949c55e01cSJens Axboe 24959c55e01cSJens Axboe /* 2496a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 24972870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 24989c55e01cSJens Axboe */ 2499a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 250035f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 250135f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 25022870c43dSOctavian Purdila { 25032870c43dSOctavian Purdila int seg; 2504fa9835e5STom Herbert struct sk_buff *iter; 25059c55e01cSJens Axboe 25061d0c0b32SEric Dumazet /* map the linear part : 25072996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 25082996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 25092996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 25109c55e01cSJens Axboe */ 25112870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 25122870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 25132870c43dSOctavian Purdila skb_headlen(skb), 251418aafc62SEric Dumazet offset, len, spd, 25153a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 25161d0c0b32SEric Dumazet sk, pipe)) 2517a108d5f3SDavid S. Miller return true; 25189c55e01cSJens Axboe 25199c55e01cSJens Axboe /* 25209c55e01cSJens Axboe * then map the fragments 25219c55e01cSJens Axboe */ 25229c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 25239c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 25249c55e01cSJens Axboe 2525ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2526b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 252718aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2528a108d5f3SDavid S. Miller return true; 25299c55e01cSJens Axboe } 25309c55e01cSJens Axboe 2531fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2532fa9835e5STom Herbert if (*offset >= iter->len) { 2533fa9835e5STom Herbert *offset -= iter->len; 2534fa9835e5STom Herbert continue; 2535fa9835e5STom Herbert } 2536fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2537fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2538fa9835e5STom Herbert * case. 2539fa9835e5STom Herbert */ 2540fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2541fa9835e5STom Herbert return true; 2542fa9835e5STom Herbert } 2543fa9835e5STom Herbert 2544a108d5f3SDavid S. Miller return false; 25459c55e01cSJens Axboe } 25469c55e01cSJens Axboe 25479c55e01cSJens Axboe /* 25489c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2549fa9835e5STom Herbert * the fragments, and the frag list. 25509c55e01cSJens Axboe */ 2551a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 25529c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 255325869262SAl Viro unsigned int flags) 25549c55e01cSJens Axboe { 255541c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 255641c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 25579c55e01cSJens Axboe struct splice_pipe_desc spd = { 25589c55e01cSJens Axboe .pages = pages, 25599c55e01cSJens Axboe .partial = partial, 2560047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 256128a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 25629c55e01cSJens Axboe .spd_release = sock_spd_release, 25639c55e01cSJens Axboe }; 256435f3d14dSJens Axboe int ret = 0; 256535f3d14dSJens Axboe 2566fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 25679c55e01cSJens Axboe 2568a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 256925869262SAl Viro ret = splice_to_pipe(pipe, &spd); 25709c55e01cSJens Axboe 257135f3d14dSJens Axboe return ret; 25729c55e01cSJens Axboe } 25732b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 25749c55e01cSJens Axboe 25750739cd28SCong Wang static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg, 25760739cd28SCong Wang struct kvec *vec, size_t num, size_t size) 25770739cd28SCong Wang { 25780739cd28SCong Wang struct socket *sock = sk->sk_socket; 25790739cd28SCong Wang 25800739cd28SCong Wang if (!sock) 25810739cd28SCong Wang return -EINVAL; 25820739cd28SCong Wang return kernel_sendmsg(sock, msg, vec, num, size); 25830739cd28SCong Wang } 25840739cd28SCong Wang 25850739cd28SCong Wang static int sendpage_unlocked(struct sock *sk, struct page *page, int offset, 25860739cd28SCong Wang size_t size, int flags) 25870739cd28SCong Wang { 25880739cd28SCong Wang struct socket *sock = sk->sk_socket; 25890739cd28SCong Wang 25900739cd28SCong Wang if (!sock) 25910739cd28SCong Wang return -EINVAL; 25920739cd28SCong Wang return kernel_sendpage(sock, page, offset, size, flags); 25930739cd28SCong Wang } 25940739cd28SCong Wang 25950739cd28SCong Wang typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg, 25960739cd28SCong Wang struct kvec *vec, size_t num, size_t size); 25970739cd28SCong Wang typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset, 25980739cd28SCong Wang size_t size, int flags); 25990739cd28SCong Wang static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, 26000739cd28SCong Wang int len, sendmsg_func sendmsg, sendpage_func sendpage) 260120bf50deSTom Herbert { 260220bf50deSTom Herbert unsigned int orig_len = len; 260320bf50deSTom Herbert struct sk_buff *head = skb; 260420bf50deSTom Herbert unsigned short fragidx; 260520bf50deSTom Herbert int slen, ret; 260620bf50deSTom Herbert 260720bf50deSTom Herbert do_frag_list: 260820bf50deSTom Herbert 260920bf50deSTom Herbert /* Deal with head data */ 261020bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 261120bf50deSTom Herbert struct kvec kv; 261220bf50deSTom Herbert struct msghdr msg; 261320bf50deSTom Herbert 261420bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 261520bf50deSTom Herbert kv.iov_base = skb->data + offset; 2616db5980d8SJohn Fastabend kv.iov_len = slen; 261720bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2618bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 261920bf50deSTom Herbert 26200739cd28SCong Wang ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked, 26210739cd28SCong Wang sendmsg_unlocked, sk, &msg, &kv, 1, slen); 262220bf50deSTom Herbert if (ret <= 0) 262320bf50deSTom Herbert goto error; 262420bf50deSTom Herbert 262520bf50deSTom Herbert offset += ret; 262620bf50deSTom Herbert len -= ret; 262720bf50deSTom Herbert } 262820bf50deSTom Herbert 262920bf50deSTom Herbert /* All the data was skb head? */ 263020bf50deSTom Herbert if (!len) 263120bf50deSTom Herbert goto out; 263220bf50deSTom Herbert 263320bf50deSTom Herbert /* Make offset relative to start of frags */ 263420bf50deSTom Herbert offset -= skb_headlen(skb); 263520bf50deSTom Herbert 263620bf50deSTom Herbert /* Find where we are in frag list */ 263720bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 263820bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 263920bf50deSTom Herbert 2640d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 264120bf50deSTom Herbert break; 264220bf50deSTom Herbert 2643d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 264420bf50deSTom Herbert } 264520bf50deSTom Herbert 264620bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 264720bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 264820bf50deSTom Herbert 2649d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 265020bf50deSTom Herbert 265120bf50deSTom Herbert while (slen) { 26520739cd28SCong Wang ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked, 26530739cd28SCong Wang sendpage_unlocked, sk, 26540739cd28SCong Wang skb_frag_page(frag), 2655b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 265620bf50deSTom Herbert slen, MSG_DONTWAIT); 265720bf50deSTom Herbert if (ret <= 0) 265820bf50deSTom Herbert goto error; 265920bf50deSTom Herbert 266020bf50deSTom Herbert len -= ret; 266120bf50deSTom Herbert offset += ret; 266220bf50deSTom Herbert slen -= ret; 266320bf50deSTom Herbert } 266420bf50deSTom Herbert 266520bf50deSTom Herbert offset = 0; 266620bf50deSTom Herbert } 266720bf50deSTom Herbert 266820bf50deSTom Herbert if (len) { 266920bf50deSTom Herbert /* Process any frag lists */ 267020bf50deSTom Herbert 267120bf50deSTom Herbert if (skb == head) { 267220bf50deSTom Herbert if (skb_has_frag_list(skb)) { 267320bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 267420bf50deSTom Herbert goto do_frag_list; 267520bf50deSTom Herbert } 267620bf50deSTom Herbert } else if (skb->next) { 267720bf50deSTom Herbert skb = skb->next; 267820bf50deSTom Herbert goto do_frag_list; 267920bf50deSTom Herbert } 268020bf50deSTom Herbert } 268120bf50deSTom Herbert 268220bf50deSTom Herbert out: 268320bf50deSTom Herbert return orig_len - len; 268420bf50deSTom Herbert 268520bf50deSTom Herbert error: 268620bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 268720bf50deSTom Herbert } 26880739cd28SCong Wang 26890739cd28SCong Wang /* Send skb data on a socket. Socket must be locked. */ 26900739cd28SCong Wang int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 26910739cd28SCong Wang int len) 26920739cd28SCong Wang { 26930739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked, 26940739cd28SCong Wang kernel_sendpage_locked); 26950739cd28SCong Wang } 269620bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 269720bf50deSTom Herbert 26980739cd28SCong Wang /* Send skb data on a socket. Socket must be unlocked. */ 26990739cd28SCong Wang int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 27000739cd28SCong Wang { 27010739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked, 27020739cd28SCong Wang sendpage_unlocked); 27030739cd28SCong Wang } 27040739cd28SCong Wang 2705357b40a1SHerbert Xu /** 2706357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2707357b40a1SHerbert Xu * @skb: destination buffer 2708357b40a1SHerbert Xu * @offset: offset in destination 2709357b40a1SHerbert Xu * @from: source buffer 2710357b40a1SHerbert Xu * @len: number of bytes to copy 2711357b40a1SHerbert Xu * 2712357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2713357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2714357b40a1SHerbert Xu * traversing fragment lists and such. 2715357b40a1SHerbert Xu */ 2716357b40a1SHerbert Xu 27170c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2718357b40a1SHerbert Xu { 27191a028e50SDavid S. Miller int start = skb_headlen(skb); 2720fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2721fbb398a8SDavid S. Miller int i, copy; 2722357b40a1SHerbert Xu 2723357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2724357b40a1SHerbert Xu goto fault; 2725357b40a1SHerbert Xu 27261a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2727357b40a1SHerbert Xu if (copy > len) 2728357b40a1SHerbert Xu copy = len; 272927d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2730357b40a1SHerbert Xu if ((len -= copy) == 0) 2731357b40a1SHerbert Xu return 0; 2732357b40a1SHerbert Xu offset += copy; 2733357b40a1SHerbert Xu from += copy; 2734357b40a1SHerbert Xu } 2735357b40a1SHerbert Xu 2736357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2737357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 27381a028e50SDavid S. Miller int end; 2739357b40a1SHerbert Xu 2740547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27411a028e50SDavid S. Miller 27429e903e08SEric Dumazet end = start + skb_frag_size(frag); 2743357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2744c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2745c613c209SWillem de Bruijn struct page *p; 2746357b40a1SHerbert Xu u8 *vaddr; 2747357b40a1SHerbert Xu 2748357b40a1SHerbert Xu if (copy > len) 2749357b40a1SHerbert Xu copy = len; 2750357b40a1SHerbert Xu 2751c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2752b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2753c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2754c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2755c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 275651c56b00SEric Dumazet kunmap_atomic(vaddr); 2757c613c209SWillem de Bruijn } 2758357b40a1SHerbert Xu 2759357b40a1SHerbert Xu if ((len -= copy) == 0) 2760357b40a1SHerbert Xu return 0; 2761357b40a1SHerbert Xu offset += copy; 2762357b40a1SHerbert Xu from += copy; 2763357b40a1SHerbert Xu } 27641a028e50SDavid S. Miller start = end; 2765357b40a1SHerbert Xu } 2766357b40a1SHerbert Xu 2767fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 27681a028e50SDavid S. Miller int end; 2769357b40a1SHerbert Xu 2770547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27711a028e50SDavid S. Miller 2772fbb398a8SDavid S. Miller end = start + frag_iter->len; 2773357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2774357b40a1SHerbert Xu if (copy > len) 2775357b40a1SHerbert Xu copy = len; 2776fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 27771a028e50SDavid S. Miller from, copy)) 2778357b40a1SHerbert Xu goto fault; 2779357b40a1SHerbert Xu if ((len -= copy) == 0) 2780357b40a1SHerbert Xu return 0; 2781357b40a1SHerbert Xu offset += copy; 2782357b40a1SHerbert Xu from += copy; 2783357b40a1SHerbert Xu } 27841a028e50SDavid S. Miller start = end; 2785357b40a1SHerbert Xu } 2786357b40a1SHerbert Xu if (!len) 2787357b40a1SHerbert Xu return 0; 2788357b40a1SHerbert Xu 2789357b40a1SHerbert Xu fault: 2790357b40a1SHerbert Xu return -EFAULT; 2791357b40a1SHerbert Xu } 2792357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2793357b40a1SHerbert Xu 27941da177e4SLinus Torvalds /* Checksum skb data. */ 27952817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 27962817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 27971da177e4SLinus Torvalds { 27981a028e50SDavid S. Miller int start = skb_headlen(skb); 27991a028e50SDavid S. Miller int i, copy = start - offset; 2800fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 28011da177e4SLinus Torvalds int pos = 0; 28021da177e4SLinus Torvalds 28031da177e4SLinus Torvalds /* Checksum header. */ 28041da177e4SLinus Torvalds if (copy > 0) { 28051da177e4SLinus Torvalds if (copy > len) 28061da177e4SLinus Torvalds copy = len; 28072544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 28082544af03SMatteo Croce skb->data + offset, copy, csum); 28091da177e4SLinus Torvalds if ((len -= copy) == 0) 28101da177e4SLinus Torvalds return csum; 28111da177e4SLinus Torvalds offset += copy; 28121da177e4SLinus Torvalds pos = copy; 28131da177e4SLinus Torvalds } 28141da177e4SLinus Torvalds 28151da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 28161a028e50SDavid S. Miller int end; 281751c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 28181da177e4SLinus Torvalds 2819547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28201a028e50SDavid S. Miller 282151c56b00SEric Dumazet end = start + skb_frag_size(frag); 28221da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2823c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2824c613c209SWillem de Bruijn struct page *p; 282544bb9363SAl Viro __wsum csum2; 28261da177e4SLinus Torvalds u8 *vaddr; 28271da177e4SLinus Torvalds 28281da177e4SLinus Torvalds if (copy > len) 28291da177e4SLinus Torvalds copy = len; 2830c613c209SWillem de Bruijn 2831c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2832b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2833c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2834c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 28352544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 28362544af03SMatteo Croce csum_partial_ext, 28372544af03SMatteo Croce vaddr + p_off, p_len, 0); 283851c56b00SEric Dumazet kunmap_atomic(vaddr); 28392544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 28402544af03SMatteo Croce csum_block_add_ext, csum, 28412544af03SMatteo Croce csum2, pos, p_len); 2842c613c209SWillem de Bruijn pos += p_len; 2843c613c209SWillem de Bruijn } 2844c613c209SWillem de Bruijn 28451da177e4SLinus Torvalds if (!(len -= copy)) 28461da177e4SLinus Torvalds return csum; 28471da177e4SLinus Torvalds offset += copy; 28481da177e4SLinus Torvalds } 28491a028e50SDavid S. Miller start = end; 28501da177e4SLinus Torvalds } 28511da177e4SLinus Torvalds 2852fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 28531a028e50SDavid S. Miller int end; 28541da177e4SLinus Torvalds 2855547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28561a028e50SDavid S. Miller 2857fbb398a8SDavid S. Miller end = start + frag_iter->len; 28581da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 28595f92a738SAl Viro __wsum csum2; 28601da177e4SLinus Torvalds if (copy > len) 28611da177e4SLinus Torvalds copy = len; 28622817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 28632817a336SDaniel Borkmann copy, 0, ops); 28642544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 28652544af03SMatteo Croce csum, csum2, pos, copy); 28661da177e4SLinus Torvalds if ((len -= copy) == 0) 28671da177e4SLinus Torvalds return csum; 28681da177e4SLinus Torvalds offset += copy; 28691da177e4SLinus Torvalds pos += copy; 28701da177e4SLinus Torvalds } 28711a028e50SDavid S. Miller start = end; 28721da177e4SLinus Torvalds } 287309a62660SKris Katterjohn BUG_ON(len); 28741da177e4SLinus Torvalds 28751da177e4SLinus Torvalds return csum; 28761da177e4SLinus Torvalds } 28772817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 28782817a336SDaniel Borkmann 28792817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 28802817a336SDaniel Borkmann int len, __wsum csum) 28812817a336SDaniel Borkmann { 28822817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2883cea80ea8SDaniel Borkmann .update = csum_partial_ext, 28842817a336SDaniel Borkmann .combine = csum_block_add_ext, 28852817a336SDaniel Borkmann }; 28862817a336SDaniel Borkmann 28872817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 28882817a336SDaniel Borkmann } 2889b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 28901da177e4SLinus Torvalds 28911da177e4SLinus Torvalds /* Both of above in one bottle. */ 28921da177e4SLinus Torvalds 289381d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 28948d5930dfSAl Viro u8 *to, int len) 28951da177e4SLinus Torvalds { 28961a028e50SDavid S. Miller int start = skb_headlen(skb); 28971a028e50SDavid S. Miller int i, copy = start - offset; 2898fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 28991da177e4SLinus Torvalds int pos = 0; 29008d5930dfSAl Viro __wsum csum = 0; 29011da177e4SLinus Torvalds 29021da177e4SLinus Torvalds /* Copy header. */ 29031da177e4SLinus Torvalds if (copy > 0) { 29041da177e4SLinus Torvalds if (copy > len) 29051da177e4SLinus Torvalds copy = len; 29061da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2907cc44c17bSAl Viro copy); 29081da177e4SLinus Torvalds if ((len -= copy) == 0) 29091da177e4SLinus Torvalds return csum; 29101da177e4SLinus Torvalds offset += copy; 29111da177e4SLinus Torvalds to += copy; 29121da177e4SLinus Torvalds pos = copy; 29131da177e4SLinus Torvalds } 29141da177e4SLinus Torvalds 29151da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 29161a028e50SDavid S. Miller int end; 29171da177e4SLinus Torvalds 2918547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29191a028e50SDavid S. Miller 29209e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 29211da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2922c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2923c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2924c613c209SWillem de Bruijn struct page *p; 29255084205fSAl Viro __wsum csum2; 29261da177e4SLinus Torvalds u8 *vaddr; 29271da177e4SLinus Torvalds 29281da177e4SLinus Torvalds if (copy > len) 29291da177e4SLinus Torvalds copy = len; 2930c613c209SWillem de Bruijn 2931c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2932b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2933c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2934c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2935c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2936c613c209SWillem de Bruijn to + copied, 2937cc44c17bSAl Viro p_len); 293851c56b00SEric Dumazet kunmap_atomic(vaddr); 29391da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2940c613c209SWillem de Bruijn pos += p_len; 2941c613c209SWillem de Bruijn } 2942c613c209SWillem de Bruijn 29431da177e4SLinus Torvalds if (!(len -= copy)) 29441da177e4SLinus Torvalds return csum; 29451da177e4SLinus Torvalds offset += copy; 29461da177e4SLinus Torvalds to += copy; 29471da177e4SLinus Torvalds } 29481a028e50SDavid S. Miller start = end; 29491da177e4SLinus Torvalds } 29501da177e4SLinus Torvalds 2951fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 295281d77662SAl Viro __wsum csum2; 29531a028e50SDavid S. Miller int end; 29541da177e4SLinus Torvalds 2955547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29561a028e50SDavid S. Miller 2957fbb398a8SDavid S. Miller end = start + frag_iter->len; 29581da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 29591da177e4SLinus Torvalds if (copy > len) 29601da177e4SLinus Torvalds copy = len; 2961fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 29621a028e50SDavid S. Miller offset - start, 29638d5930dfSAl Viro to, copy); 29641da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 29651da177e4SLinus Torvalds if ((len -= copy) == 0) 29661da177e4SLinus Torvalds return csum; 29671da177e4SLinus Torvalds offset += copy; 29681da177e4SLinus Torvalds to += copy; 29691da177e4SLinus Torvalds pos += copy; 29701da177e4SLinus Torvalds } 29711a028e50SDavid S. Miller start = end; 29721da177e4SLinus Torvalds } 297309a62660SKris Katterjohn BUG_ON(len); 29741da177e4SLinus Torvalds return csum; 29751da177e4SLinus Torvalds } 2976b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 29771da177e4SLinus Torvalds 297849f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 297949f8e832SCong Wang { 298049f8e832SCong Wang __sum16 sum; 298149f8e832SCong Wang 298249f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 298314641931SCong Wang /* See comments in __skb_checksum_complete(). */ 298449f8e832SCong Wang if (likely(!sum)) { 298549f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 298649f8e832SCong Wang !skb->csum_complete_sw) 29877fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 298849f8e832SCong Wang } 298949f8e832SCong Wang if (!skb_shared(skb)) 299049f8e832SCong Wang skb->csum_valid = !sum; 299149f8e832SCong Wang return sum; 299249f8e832SCong Wang } 299349f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 299449f8e832SCong Wang 299514641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 299614641931SCong Wang * which has been changed from the hardware checksum, for example, by 299714641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 299814641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 299914641931SCong Wang * 300014641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 300114641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 300214641931SCong Wang * shared. 300314641931SCong Wang */ 300449f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 300549f8e832SCong Wang { 300649f8e832SCong Wang __wsum csum; 300749f8e832SCong Wang __sum16 sum; 300849f8e832SCong Wang 300949f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 301049f8e832SCong Wang 301149f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 301214641931SCong Wang /* This check is inverted, because we already knew the hardware 301314641931SCong Wang * checksum is invalid before calling this function. So, if the 301414641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 301514641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 301614641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 301714641931SCong Wang * when moving skb->data around. 301814641931SCong Wang */ 301949f8e832SCong Wang if (likely(!sum)) { 302049f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 302149f8e832SCong Wang !skb->csum_complete_sw) 30227fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 302349f8e832SCong Wang } 302449f8e832SCong Wang 302549f8e832SCong Wang if (!skb_shared(skb)) { 302649f8e832SCong Wang /* Save full packet checksum */ 302749f8e832SCong Wang skb->csum = csum; 302849f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 302949f8e832SCong Wang skb->csum_complete_sw = 1; 303049f8e832SCong Wang skb->csum_valid = !sum; 303149f8e832SCong Wang } 303249f8e832SCong Wang 303349f8e832SCong Wang return sum; 303449f8e832SCong Wang } 303549f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 303649f8e832SCong Wang 30379617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 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 __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 30469617813dSDavide Caratti int offset, int len) 30479617813dSDavide Caratti { 30489617813dSDavide Caratti net_warn_ratelimited( 30499617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 30509617813dSDavide Caratti __func__); 30519617813dSDavide Caratti return 0; 30529617813dSDavide Caratti } 30539617813dSDavide Caratti 30549617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 30559617813dSDavide Caratti .update = warn_crc32c_csum_update, 30569617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 30579617813dSDavide Caratti }; 30589617813dSDavide Caratti 30599617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 30609617813dSDavide Caratti &default_crc32c_ops; 30619617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 30629617813dSDavide Caratti 3063af2806f8SThomas Graf /** 3064af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 3065af2806f8SThomas Graf * @from: source buffer 3066af2806f8SThomas Graf * 3067af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 3068af2806f8SThomas Graf * into skb_zerocopy(). 3069af2806f8SThomas Graf */ 3070af2806f8SThomas Graf unsigned int 3071af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 3072af2806f8SThomas Graf { 3073af2806f8SThomas Graf unsigned int hlen = 0; 3074af2806f8SThomas Graf 3075af2806f8SThomas Graf if (!from->head_frag || 3076af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 3077a17ad096SPravin B Shelar skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) { 3078af2806f8SThomas Graf hlen = skb_headlen(from); 3079a17ad096SPravin B Shelar if (!hlen) 3080a17ad096SPravin B Shelar hlen = from->len; 3081a17ad096SPravin B Shelar } 3082af2806f8SThomas Graf 3083af2806f8SThomas Graf if (skb_has_frag_list(from)) 3084af2806f8SThomas Graf hlen = from->len; 3085af2806f8SThomas Graf 3086af2806f8SThomas Graf return hlen; 3087af2806f8SThomas Graf } 3088af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 3089af2806f8SThomas Graf 3090af2806f8SThomas Graf /** 3091af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 3092af2806f8SThomas Graf * @to: destination buffer 30937fceb4deSMasanari Iida * @from: source buffer 3094af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 3095af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 3096af2806f8SThomas Graf * 3097af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 3098af2806f8SThomas Graf * to the frags in the source buffer. 3099af2806f8SThomas Graf * 3100af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 3101af2806f8SThomas Graf * headroom in the `to` buffer. 310236d5fe6aSZoltan Kiss * 310336d5fe6aSZoltan Kiss * Return value: 310436d5fe6aSZoltan Kiss * 0: everything is OK 310536d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 310636d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 3107af2806f8SThomas Graf */ 310836d5fe6aSZoltan Kiss int 310936d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 3110af2806f8SThomas Graf { 3111af2806f8SThomas Graf int i, j = 0; 3112af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 311336d5fe6aSZoltan Kiss int ret; 3114af2806f8SThomas Graf struct page *page; 3115af2806f8SThomas Graf unsigned int offset; 3116af2806f8SThomas Graf 3117af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 3118af2806f8SThomas Graf 3119af2806f8SThomas Graf /* dont bother with small payloads */ 312036d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 312136d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 3122af2806f8SThomas Graf 3123af2806f8SThomas Graf if (hlen) { 312436d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 312536d5fe6aSZoltan Kiss if (unlikely(ret)) 312636d5fe6aSZoltan Kiss return ret; 3127af2806f8SThomas Graf len -= hlen; 3128af2806f8SThomas Graf } else { 3129af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 3130af2806f8SThomas Graf if (plen) { 3131af2806f8SThomas Graf page = virt_to_head_page(from->head); 3132af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 3133af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 3134af2806f8SThomas Graf get_page(page); 3135af2806f8SThomas Graf j = 1; 3136af2806f8SThomas Graf len -= plen; 3137af2806f8SThomas Graf } 3138af2806f8SThomas Graf } 3139af2806f8SThomas Graf 3140af2806f8SThomas Graf to->truesize += len + plen; 3141af2806f8SThomas Graf to->len += len + plen; 3142af2806f8SThomas Graf to->data_len += len + plen; 3143af2806f8SThomas Graf 314436d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 314536d5fe6aSZoltan Kiss skb_tx_error(from); 314636d5fe6aSZoltan Kiss return -ENOMEM; 314736d5fe6aSZoltan Kiss } 31481f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 314936d5fe6aSZoltan Kiss 3150af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 3151d8e18a51SMatthew Wilcox (Oracle) int size; 3152d8e18a51SMatthew Wilcox (Oracle) 3153af2806f8SThomas Graf if (!len) 3154af2806f8SThomas Graf break; 3155af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 3156d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 3157d8e18a51SMatthew Wilcox (Oracle) len); 3158d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3159d8e18a51SMatthew Wilcox (Oracle) len -= size; 3160af2806f8SThomas Graf skb_frag_ref(to, j); 3161af2806f8SThomas Graf j++; 3162af2806f8SThomas Graf } 3163af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 316436d5fe6aSZoltan Kiss 316536d5fe6aSZoltan Kiss return 0; 3166af2806f8SThomas Graf } 3167af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3168af2806f8SThomas Graf 31691da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 31701da177e4SLinus Torvalds { 3171d3bc23e7SAl Viro __wsum csum; 31721da177e4SLinus Torvalds long csstart; 31731da177e4SLinus Torvalds 317484fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 317555508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 31761da177e4SLinus Torvalds else 31771da177e4SLinus Torvalds csstart = skb_headlen(skb); 31781da177e4SLinus Torvalds 317909a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 31801da177e4SLinus Torvalds 3181d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 31821da177e4SLinus Torvalds 31831da177e4SLinus Torvalds csum = 0; 31841da177e4SLinus Torvalds if (csstart != skb->len) 31851da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 31868d5930dfSAl Viro skb->len - csstart); 31871da177e4SLinus Torvalds 318884fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3189ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 31901da177e4SLinus Torvalds 3191d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 31921da177e4SLinus Torvalds } 31931da177e4SLinus Torvalds } 3194b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 31951da177e4SLinus Torvalds 31961da177e4SLinus Torvalds /** 31971da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 31981da177e4SLinus Torvalds * @list: list to dequeue from 31991da177e4SLinus Torvalds * 32001da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 32011da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 32021da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32031da177e4SLinus Torvalds */ 32041da177e4SLinus Torvalds 32051da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 32061da177e4SLinus Torvalds { 32071da177e4SLinus Torvalds unsigned long flags; 32081da177e4SLinus Torvalds struct sk_buff *result; 32091da177e4SLinus Torvalds 32101da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32111da177e4SLinus Torvalds result = __skb_dequeue(list); 32121da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32131da177e4SLinus Torvalds return result; 32141da177e4SLinus Torvalds } 3215b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 32161da177e4SLinus Torvalds 32171da177e4SLinus Torvalds /** 32181da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 32191da177e4SLinus Torvalds * @list: list to dequeue from 32201da177e4SLinus Torvalds * 32211da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 32221da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 32231da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32241da177e4SLinus Torvalds */ 32251da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 32261da177e4SLinus Torvalds { 32271da177e4SLinus Torvalds unsigned long flags; 32281da177e4SLinus Torvalds struct sk_buff *result; 32291da177e4SLinus Torvalds 32301da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32311da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 32321da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32331da177e4SLinus Torvalds return result; 32341da177e4SLinus Torvalds } 3235b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 32361da177e4SLinus Torvalds 32371da177e4SLinus Torvalds /** 32381da177e4SLinus Torvalds * skb_queue_purge - empty a list 32391da177e4SLinus Torvalds * @list: list to empty 32401da177e4SLinus Torvalds * 32411da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 32421da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 32431da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 32441da177e4SLinus Torvalds */ 32451da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 32461da177e4SLinus Torvalds { 32471da177e4SLinus Torvalds struct sk_buff *skb; 32481da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 32491da177e4SLinus Torvalds kfree_skb(skb); 32501da177e4SLinus Torvalds } 3251b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 32521da177e4SLinus Torvalds 32531da177e4SLinus Torvalds /** 32549f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 32559f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3256385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 32579f5afeaeSYaogong Wang * 32589f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 32599f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 32609f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 32619f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 32629f5afeaeSYaogong Wang */ 3263385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 32649f5afeaeSYaogong Wang { 32657c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3266385114deSPeter Oskolkov unsigned int sum = 0; 32679f5afeaeSYaogong Wang 32687c90584cSEric Dumazet while (p) { 32697c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 32707c90584cSEric Dumazet 32717c90584cSEric Dumazet p = rb_next(p); 32727c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3273385114deSPeter Oskolkov sum += skb->truesize; 32749f5afeaeSYaogong Wang kfree_skb(skb); 32757c90584cSEric Dumazet } 3276385114deSPeter Oskolkov return sum; 32779f5afeaeSYaogong Wang } 32789f5afeaeSYaogong Wang 32799f5afeaeSYaogong Wang /** 32801da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 32811da177e4SLinus Torvalds * @list: list to use 32821da177e4SLinus Torvalds * @newsk: buffer to queue 32831da177e4SLinus Torvalds * 32841da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 32851da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 32861da177e4SLinus Torvalds * safely. 32871da177e4SLinus Torvalds * 32881da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 32891da177e4SLinus Torvalds */ 32901da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 32911da177e4SLinus Torvalds { 32921da177e4SLinus Torvalds unsigned long flags; 32931da177e4SLinus Torvalds 32941da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32951da177e4SLinus Torvalds __skb_queue_head(list, newsk); 32961da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32971da177e4SLinus Torvalds } 3298b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 32991da177e4SLinus Torvalds 33001da177e4SLinus Torvalds /** 33011da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 33021da177e4SLinus Torvalds * @list: list to use 33031da177e4SLinus Torvalds * @newsk: buffer to queue 33041da177e4SLinus Torvalds * 33051da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 33061da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33071da177e4SLinus Torvalds * safely. 33081da177e4SLinus Torvalds * 33091da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33101da177e4SLinus Torvalds */ 33111da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 33121da177e4SLinus Torvalds { 33131da177e4SLinus Torvalds unsigned long flags; 33141da177e4SLinus Torvalds 33151da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33161da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 33171da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33181da177e4SLinus Torvalds } 3319b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 33208728b834SDavid S. Miller 33211da177e4SLinus Torvalds /** 33221da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 33231da177e4SLinus Torvalds * @skb: buffer to remove 33248728b834SDavid S. Miller * @list: list to use 33251da177e4SLinus Torvalds * 33268728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 33278728b834SDavid S. Miller * function is atomic with respect to other list locked calls 33281da177e4SLinus Torvalds * 33298728b834SDavid S. Miller * You must know what list the SKB is on. 33301da177e4SLinus Torvalds */ 33318728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 33321da177e4SLinus Torvalds { 33331da177e4SLinus Torvalds unsigned long flags; 33341da177e4SLinus Torvalds 33351da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33368728b834SDavid S. Miller __skb_unlink(skb, list); 33371da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33381da177e4SLinus Torvalds } 3339b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 33401da177e4SLinus Torvalds 33411da177e4SLinus Torvalds /** 33421da177e4SLinus Torvalds * skb_append - append a buffer 33431da177e4SLinus Torvalds * @old: buffer to insert after 33441da177e4SLinus Torvalds * @newsk: buffer to insert 33458728b834SDavid S. Miller * @list: list to use 33461da177e4SLinus Torvalds * 33471da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 33481da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 33491da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33501da177e4SLinus Torvalds */ 33518728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 33521da177e4SLinus Torvalds { 33531da177e4SLinus Torvalds unsigned long flags; 33541da177e4SLinus Torvalds 33558728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 33567de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 33578728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 33581da177e4SLinus Torvalds } 3359b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 33601da177e4SLinus Torvalds 33611da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 33621da177e4SLinus Torvalds struct sk_buff* skb1, 33631da177e4SLinus Torvalds const u32 len, const int pos) 33641da177e4SLinus Torvalds { 33651da177e4SLinus Torvalds int i; 33661da177e4SLinus Torvalds 3367d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3368d626f62bSArnaldo Carvalho de Melo pos - len); 33691da177e4SLinus Torvalds /* And move data appendix as is. */ 33701da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 33711da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 33721da177e4SLinus Torvalds 33731da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 33741da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 33751da177e4SLinus Torvalds skb1->data_len = skb->data_len; 33761da177e4SLinus Torvalds skb1->len += skb1->data_len; 33771da177e4SLinus Torvalds skb->data_len = 0; 33781da177e4SLinus Torvalds skb->len = len; 337927a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 33801da177e4SLinus Torvalds } 33811da177e4SLinus Torvalds 33821da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 33831da177e4SLinus Torvalds struct sk_buff* skb1, 33841da177e4SLinus Torvalds const u32 len, int pos) 33851da177e4SLinus Torvalds { 33861da177e4SLinus Torvalds int i, k = 0; 33871da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 33881da177e4SLinus Torvalds 33891da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 33901da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 33911da177e4SLinus Torvalds skb->len = len; 33921da177e4SLinus Torvalds skb->data_len = len - pos; 33931da177e4SLinus Torvalds 33941da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 33959e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 33961da177e4SLinus Torvalds 33971da177e4SLinus Torvalds if (pos + size > len) { 33981da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 33991da177e4SLinus Torvalds 34001da177e4SLinus Torvalds if (pos < len) { 34011da177e4SLinus Torvalds /* Split frag. 34021da177e4SLinus Torvalds * We have two variants in this case: 34031da177e4SLinus Torvalds * 1. Move all the frag to the second 34041da177e4SLinus Torvalds * part, if it is possible. F.e. 34051da177e4SLinus Torvalds * this approach is mandatory for TUX, 34061da177e4SLinus Torvalds * where splitting is expensive. 34071da177e4SLinus Torvalds * 2. Split is accurately. We make this. 34081da177e4SLinus Torvalds */ 3409ea2ab693SIan Campbell skb_frag_ref(skb, i); 3410b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 34119e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 34129e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 34131da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34141da177e4SLinus Torvalds } 34151da177e4SLinus Torvalds k++; 34161da177e4SLinus Torvalds } else 34171da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34181da177e4SLinus Torvalds pos += size; 34191da177e4SLinus Torvalds } 34201da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 34211da177e4SLinus Torvalds } 34221da177e4SLinus Torvalds 34231da177e4SLinus Torvalds /** 34241da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 34251da177e4SLinus Torvalds * @skb: the buffer to split 34261da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 34271da177e4SLinus Torvalds * @len: new length for skb 34281da177e4SLinus Torvalds */ 34291da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 34301da177e4SLinus Torvalds { 34311da177e4SLinus Torvalds int pos = skb_headlen(skb); 34329b65b17dSTalal Ahmad const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY; 34331da177e4SLinus Torvalds 34349b65b17dSTalal Ahmad skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags; 34351f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 34361da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 34371da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 34381da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 34391da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 34401da177e4SLinus Torvalds } 3441b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 34421da177e4SLinus Torvalds 34439f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 34449f782db3SIlpo Järvinen * 34459f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 34469f782db3SIlpo Järvinen */ 3447832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3448832d11c5SIlpo Järvinen { 3449c4777efaSEric Dumazet return skb_unclone_keeptruesize(skb, GFP_ATOMIC); 3450832d11c5SIlpo Järvinen } 3451832d11c5SIlpo Järvinen 3452832d11c5SIlpo Järvinen /** 3453832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3454832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3455832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3456832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3457832d11c5SIlpo Järvinen * 3458832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 345920e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3460832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3461832d11c5SIlpo Järvinen * 3462832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3463832d11c5SIlpo Järvinen * 3464832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3465832d11c5SIlpo Järvinen * to have non-paged data as well. 3466832d11c5SIlpo Järvinen * 3467832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3468832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3469832d11c5SIlpo Järvinen */ 3470832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3471832d11c5SIlpo Järvinen { 3472832d11c5SIlpo Järvinen int from, to, merge, todo; 3473d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3474832d11c5SIlpo Järvinen 3475832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3476f8071cdeSEric Dumazet 3477f8071cdeSEric Dumazet if (skb_headlen(skb)) 3478f8071cdeSEric Dumazet return 0; 34791f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 34801f8b977aSWillem de Bruijn return 0; 3481832d11c5SIlpo Järvinen 3482832d11c5SIlpo Järvinen todo = shiftlen; 3483832d11c5SIlpo Järvinen from = 0; 3484832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3485832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3486832d11c5SIlpo Järvinen 3487832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3488832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3489832d11c5SIlpo Järvinen */ 3490832d11c5SIlpo Järvinen if (!to || 3491ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3492b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3493832d11c5SIlpo Järvinen merge = -1; 3494832d11c5SIlpo Järvinen } else { 3495832d11c5SIlpo Järvinen merge = to - 1; 3496832d11c5SIlpo Järvinen 34979e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3498832d11c5SIlpo Järvinen if (todo < 0) { 3499832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3500832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3501832d11c5SIlpo Järvinen return 0; 3502832d11c5SIlpo Järvinen 35039f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 35049f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3505832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3506832d11c5SIlpo Järvinen 35079e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 35089e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3509b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3510832d11c5SIlpo Järvinen 3511832d11c5SIlpo Järvinen goto onlymerged; 3512832d11c5SIlpo Järvinen } 3513832d11c5SIlpo Järvinen 3514832d11c5SIlpo Järvinen from++; 3515832d11c5SIlpo Järvinen } 3516832d11c5SIlpo Järvinen 3517832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3518832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3519832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3520832d11c5SIlpo Järvinen return 0; 3521832d11c5SIlpo Järvinen 3522832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3523832d11c5SIlpo Järvinen return 0; 3524832d11c5SIlpo Järvinen 3525832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3526832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3527832d11c5SIlpo Järvinen return 0; 3528832d11c5SIlpo Järvinen 3529832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3530832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3531832d11c5SIlpo Järvinen 35329e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3533832d11c5SIlpo Järvinen *fragto = *fragfrom; 35349e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3535832d11c5SIlpo Järvinen from++; 3536832d11c5SIlpo Järvinen to++; 3537832d11c5SIlpo Järvinen 3538832d11c5SIlpo Järvinen } else { 3539ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3540b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3541b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 35429e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3543832d11c5SIlpo Järvinen 3544b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 35459e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3546832d11c5SIlpo Järvinen todo = 0; 3547832d11c5SIlpo Järvinen 3548832d11c5SIlpo Järvinen to++; 3549832d11c5SIlpo Järvinen break; 3550832d11c5SIlpo Järvinen } 3551832d11c5SIlpo Järvinen } 3552832d11c5SIlpo Järvinen 3553832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3554832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3555832d11c5SIlpo Järvinen 3556832d11c5SIlpo Järvinen if (merge >= 0) { 3557832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3558832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3559832d11c5SIlpo Järvinen 35609e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 35616a5bcd84SIlias Apalodimas __skb_frag_unref(fragfrom, skb->pp_recycle); 3562832d11c5SIlpo Järvinen } 3563832d11c5SIlpo Järvinen 3564832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3565832d11c5SIlpo Järvinen to = 0; 3566832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3567832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3568832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3569832d11c5SIlpo Järvinen 3570832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3571832d11c5SIlpo Järvinen 3572832d11c5SIlpo Järvinen onlymerged: 3573832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3574832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3575832d11c5SIlpo Järvinen */ 3576832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3577832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3578832d11c5SIlpo Järvinen 3579832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3580832d11c5SIlpo Järvinen skb->len -= shiftlen; 3581832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3582832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3583832d11c5SIlpo Järvinen tgt->len += shiftlen; 3584832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3585832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3586832d11c5SIlpo Järvinen 3587832d11c5SIlpo Järvinen return shiftlen; 3588832d11c5SIlpo Järvinen } 3589832d11c5SIlpo Järvinen 3590677e90edSThomas Graf /** 3591677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3592677e90edSThomas Graf * @skb: the buffer to read 3593677e90edSThomas Graf * @from: lower offset of data to be read 3594677e90edSThomas Graf * @to: upper offset of data to be read 3595677e90edSThomas Graf * @st: state variable 3596677e90edSThomas Graf * 3597677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3598677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3599677e90edSThomas Graf */ 3600677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3601677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3602677e90edSThomas Graf { 3603677e90edSThomas Graf st->lower_offset = from; 3604677e90edSThomas Graf st->upper_offset = to; 3605677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3606677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3607677e90edSThomas Graf st->frag_data = NULL; 360897550f6fSWillem de Bruijn st->frag_off = 0; 3609677e90edSThomas Graf } 3610b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3611677e90edSThomas Graf 3612677e90edSThomas Graf /** 3613677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3614677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3615677e90edSThomas Graf * @data: destination pointer for data to be returned 3616677e90edSThomas Graf * @st: state variable 3617677e90edSThomas Graf * 3618bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3619677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3620bc32383cSMathias Krause * the head of the data block to @data and returns the length 3621677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3622677e90edSThomas Graf * offset has been reached. 3623677e90edSThomas Graf * 3624677e90edSThomas Graf * The caller is not required to consume all of the data 3625bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3626677e90edSThomas Graf * of bytes already consumed and the next call to 3627677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3628677e90edSThomas Graf * 362925985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3630e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3631677e90edSThomas Graf * reads of potentially non linear data. 3632677e90edSThomas Graf * 3633bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3634677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3635677e90edSThomas Graf * a stack for this purpose. 3636677e90edSThomas Graf */ 3637677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3638677e90edSThomas Graf struct skb_seq_state *st) 3639677e90edSThomas Graf { 3640677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3641677e90edSThomas Graf skb_frag_t *frag; 3642677e90edSThomas Graf 3643aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3644aeb193eaSWedson Almeida Filho if (st->frag_data) { 3645aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3646aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3647aeb193eaSWedson Almeida Filho } 3648677e90edSThomas Graf return 0; 3649aeb193eaSWedson Almeida Filho } 3650677e90edSThomas Graf 3651677e90edSThomas Graf next_skb: 365295e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3653677e90edSThomas Graf 3654995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 365595e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3656677e90edSThomas Graf return block_limit - abs_offset; 3657677e90edSThomas Graf } 3658677e90edSThomas Graf 3659677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3660677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3661677e90edSThomas Graf 3662677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 366397550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3664677e90edSThomas Graf 366597550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 366697550f6fSWillem de Bruijn 366797550f6fSWillem de Bruijn pg_idx = 0; 366897550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 366997550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 367097550f6fSWillem de Bruijn 367197550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 367297550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 367397550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 367497550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 367597550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 367697550f6fSWillem de Bruijn } 367797550f6fSWillem de Bruijn 367897550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3679677e90edSThomas Graf if (abs_offset < block_limit) { 3680677e90edSThomas Graf if (!st->frag_data) 368197550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3682677e90edSThomas Graf 368397550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3684677e90edSThomas Graf (abs_offset - st->stepped_offset); 3685677e90edSThomas Graf 3686677e90edSThomas Graf return block_limit - abs_offset; 3687677e90edSThomas Graf } 3688677e90edSThomas Graf 3689677e90edSThomas Graf if (st->frag_data) { 369051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3691677e90edSThomas Graf st->frag_data = NULL; 3692677e90edSThomas Graf } 3693677e90edSThomas Graf 369497550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 369597550f6fSWillem de Bruijn st->frag_off += pg_sz; 369697550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 369797550f6fSWillem de Bruijn st->frag_off = 0; 3698677e90edSThomas Graf st->frag_idx++; 369997550f6fSWillem de Bruijn } 3700677e90edSThomas Graf } 3701677e90edSThomas Graf 37025b5a60daSOlaf Kirch if (st->frag_data) { 370351c56b00SEric Dumazet kunmap_atomic(st->frag_data); 37045b5a60daSOlaf Kirch st->frag_data = NULL; 37055b5a60daSOlaf Kirch } 37065b5a60daSOlaf Kirch 370721dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3708677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 370995e3b24cSHerbert Xu st->frag_idx = 0; 3710677e90edSThomas Graf goto next_skb; 371171b3346dSShyam Iyer } else if (st->cur_skb->next) { 371271b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 371371b3346dSShyam Iyer st->frag_idx = 0; 3714677e90edSThomas Graf goto next_skb; 3715677e90edSThomas Graf } 3716677e90edSThomas Graf 3717677e90edSThomas Graf return 0; 3718677e90edSThomas Graf } 3719b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3720677e90edSThomas Graf 3721677e90edSThomas Graf /** 3722677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3723677e90edSThomas Graf * @st: state variable 3724677e90edSThomas Graf * 3725677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3726677e90edSThomas Graf * returned 0. 3727677e90edSThomas Graf */ 3728677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3729677e90edSThomas Graf { 3730677e90edSThomas Graf if (st->frag_data) 373151c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3732677e90edSThomas Graf } 3733b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3734677e90edSThomas Graf 37353fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 37363fc7e8a6SThomas Graf 37373fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 37383fc7e8a6SThomas Graf struct ts_config *conf, 37393fc7e8a6SThomas Graf struct ts_state *state) 37403fc7e8a6SThomas Graf { 37413fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 37423fc7e8a6SThomas Graf } 37433fc7e8a6SThomas Graf 37443fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 37453fc7e8a6SThomas Graf { 37463fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 37473fc7e8a6SThomas Graf } 37483fc7e8a6SThomas Graf 37493fc7e8a6SThomas Graf /** 37503fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 37513fc7e8a6SThomas Graf * @skb: the buffer to look in 37523fc7e8a6SThomas Graf * @from: search offset 37533fc7e8a6SThomas Graf * @to: search limit 37543fc7e8a6SThomas Graf * @config: textsearch configuration 37553fc7e8a6SThomas Graf * 37563fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 37573fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 37583fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 37593fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 37603fc7e8a6SThomas Graf */ 37613fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3762059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 37633fc7e8a6SThomas Graf { 3764059a2440SBojan Prtvar struct ts_state state; 3765f72b948dSPhil Oester unsigned int ret; 3766f72b948dSPhil Oester 3767b228c9b0SWillem de Bruijn BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb)); 3768b228c9b0SWillem de Bruijn 37693fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 37703fc7e8a6SThomas Graf config->finish = skb_ts_finish; 37713fc7e8a6SThomas Graf 3772059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 37733fc7e8a6SThomas Graf 3774059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3775f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 37763fc7e8a6SThomas Graf } 3777b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 37783fc7e8a6SThomas Graf 3779be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3780be12a1feSHannes Frederic Sowa int offset, size_t size) 3781be12a1feSHannes Frederic Sowa { 3782be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3783be12a1feSHannes Frederic Sowa 3784be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3785be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3786be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3787be12a1feSHannes Frederic Sowa get_page(page); 3788be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3789be12a1feSHannes Frederic Sowa } else { 3790be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3791be12a1feSHannes Frederic Sowa } 3792be12a1feSHannes Frederic Sowa 3793be12a1feSHannes Frederic Sowa return 0; 3794be12a1feSHannes Frederic Sowa } 3795be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3796be12a1feSHannes Frederic Sowa 3797cbb042f9SHerbert Xu /** 3798cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3799cbb042f9SHerbert Xu * @skb: buffer to update 3800cbb042f9SHerbert Xu * @len: length of data pulled 3801cbb042f9SHerbert Xu * 3802cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3803fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 380484fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 380584fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 380684fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3807cbb042f9SHerbert Xu */ 3808af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3809cbb042f9SHerbert Xu { 381031b33dfbSPravin B Shelar unsigned char *data = skb->data; 381131b33dfbSPravin B Shelar 3812cbb042f9SHerbert Xu BUG_ON(len > skb->len); 381331b33dfbSPravin B Shelar __skb_pull(skb, len); 381431b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 381531b33dfbSPravin B Shelar return skb->data; 3816cbb042f9SHerbert Xu } 3817f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3818f94691acSArnaldo Carvalho de Melo 381913acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 382013acc94eSYonghong Song { 382113acc94eSYonghong Song skb_frag_t head_frag; 382213acc94eSYonghong Song struct page *page; 382313acc94eSYonghong Song 382413acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3825d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3826b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3827b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3828d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 382913acc94eSYonghong Song return head_frag; 383013acc94eSYonghong Song } 383113acc94eSYonghong Song 38323a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 38333a1296a3SSteffen Klassert netdev_features_t features, 38343a1296a3SSteffen Klassert unsigned int offset) 38353a1296a3SSteffen Klassert { 38363a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 38373a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 38383a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 38393a1296a3SSteffen Klassert unsigned int delta_len = 0; 38403a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 384153475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 384253475c5dSDongseok Yi int err; 38433a1296a3SSteffen Klassert 38443a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 38453a1296a3SSteffen Klassert 38463a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 38473a1296a3SSteffen Klassert 38483a1296a3SSteffen Klassert do { 38493a1296a3SSteffen Klassert nskb = list_skb; 38503a1296a3SSteffen Klassert list_skb = list_skb->next; 38513a1296a3SSteffen Klassert 385253475c5dSDongseok Yi err = 0; 385353475c5dSDongseok Yi if (skb_shared(nskb)) { 385453475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 385553475c5dSDongseok Yi if (tmp) { 385653475c5dSDongseok Yi consume_skb(nskb); 385753475c5dSDongseok Yi nskb = tmp; 385853475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 385953475c5dSDongseok Yi } else { 386053475c5dSDongseok Yi err = -ENOMEM; 386153475c5dSDongseok Yi } 386253475c5dSDongseok Yi } 386353475c5dSDongseok Yi 38643a1296a3SSteffen Klassert if (!tail) 38653a1296a3SSteffen Klassert skb->next = nskb; 38663a1296a3SSteffen Klassert else 38673a1296a3SSteffen Klassert tail->next = nskb; 38683a1296a3SSteffen Klassert 386953475c5dSDongseok Yi if (unlikely(err)) { 387053475c5dSDongseok Yi nskb->next = list_skb; 387153475c5dSDongseok Yi goto err_linearize; 387253475c5dSDongseok Yi } 387353475c5dSDongseok Yi 38743a1296a3SSteffen Klassert tail = nskb; 38753a1296a3SSteffen Klassert 38763a1296a3SSteffen Klassert delta_len += nskb->len; 38773a1296a3SSteffen Klassert delta_truesize += nskb->truesize; 38783a1296a3SSteffen Klassert 38793a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 38803a1296a3SSteffen Klassert 3881cf673ed0SFlorian Westphal skb_release_head_state(nskb); 38823a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 38833a1296a3SSteffen Klassert 38843a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 38853a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 38863a1296a3SSteffen Klassert nskb->data - tnl_hlen, 38873a1296a3SSteffen Klassert offset + tnl_hlen); 38883a1296a3SSteffen Klassert 38893a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 38903a1296a3SSteffen Klassert __skb_linearize(nskb)) 38913a1296a3SSteffen Klassert goto err_linearize; 38923a1296a3SSteffen Klassert 38933a1296a3SSteffen Klassert } while (list_skb); 38943a1296a3SSteffen Klassert 38953a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 38963a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 38973a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 38983a1296a3SSteffen Klassert 38993a1296a3SSteffen Klassert skb_gso_reset(skb); 39003a1296a3SSteffen Klassert 39013a1296a3SSteffen Klassert skb->prev = tail; 39023a1296a3SSteffen Klassert 39033a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 39043a1296a3SSteffen Klassert __skb_linearize(skb)) 39053a1296a3SSteffen Klassert goto err_linearize; 39063a1296a3SSteffen Klassert 39073a1296a3SSteffen Klassert skb_get(skb); 39083a1296a3SSteffen Klassert 39093a1296a3SSteffen Klassert return skb; 39103a1296a3SSteffen Klassert 39113a1296a3SSteffen Klassert err_linearize: 39123a1296a3SSteffen Klassert kfree_skb_list(skb->next); 39133a1296a3SSteffen Klassert skb->next = NULL; 39143a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 39153a1296a3SSteffen Klassert } 39163a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 39173a1296a3SSteffen Klassert 3918f4c50d99SHerbert Xu /** 3919f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3920df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3921576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3922f4c50d99SHerbert Xu * 3923f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 39244c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 39254c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3926f4c50d99SHerbert Xu */ 3927df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3928df5771ffSMichael S. Tsirkin netdev_features_t features) 3929f4c50d99SHerbert Xu { 3930f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3931f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 39321a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3933df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3934df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3935df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 39361fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3937f4c50d99SHerbert Xu unsigned int offset = doffset; 3938df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3939802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3940f4c50d99SHerbert Xu unsigned int headroom; 3941802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3942ec5f0615SPravin B Shelar __be16 proto; 394336c98382SAlexander Duyck bool csum, sg; 3944df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 3945f4c50d99SHerbert Xu int err = -ENOMEM; 3946f4c50d99SHerbert Xu int i = 0; 3947f4c50d99SHerbert Xu int pos; 3948f4c50d99SHerbert Xu 39493dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 39503dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 39513dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 39523dcbdb13SShmulik Ladkani * non head_frag head. 39533dcbdb13SShmulik Ladkani * 39543dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 39553dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 39563dcbdb13SShmulik Ladkani * head, then the first one has too). 39573dcbdb13SShmulik Ladkani * 39583dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 39593dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 39603dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 39613dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 39623dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 39633dcbdb13SShmulik Ladkani */ 39643dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 39653dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 39663dcbdb13SShmulik Ladkani } 39673dcbdb13SShmulik Ladkani 39685882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 39692f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 3970ec5f0615SPravin B Shelar if (unlikely(!proto)) 3971ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 3972ec5f0615SPravin B Shelar 397336c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 3974f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 39757e2b10c1STom Herbert 397607b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 397707b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 397807b26c94SSteffen Klassert struct sk_buff *iter; 397943170c4eSIlan Tayari unsigned int frag_len; 398007b26c94SSteffen Klassert 398107b26c94SSteffen Klassert if (!list_skb || 398207b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 398307b26c94SSteffen Klassert goto normal; 398407b26c94SSteffen Klassert 398543170c4eSIlan Tayari /* If we get here then all the required 398643170c4eSIlan Tayari * GSO features except frag_list are supported. 398743170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 398843170c4eSIlan Tayari * with no frag_list. 398943170c4eSIlan Tayari * Currently we can do that only when the buffers don't 399043170c4eSIlan Tayari * have a linear part and all the buffers except 399143170c4eSIlan Tayari * the last are of the same length. 399207b26c94SSteffen Klassert */ 399343170c4eSIlan Tayari frag_len = list_skb->len; 399407b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 399543170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 399643170c4eSIlan Tayari goto normal; 3997eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 399807b26c94SSteffen Klassert goto normal; 399907b26c94SSteffen Klassert 400007b26c94SSteffen Klassert len -= iter->len; 400107b26c94SSteffen Klassert } 400243170c4eSIlan Tayari 400343170c4eSIlan Tayari if (len != frag_len) 400443170c4eSIlan Tayari goto normal; 400507b26c94SSteffen Klassert } 400607b26c94SSteffen Klassert 4007802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 4008802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 4009802ab55aSAlexander Duyck * now. 4010802ab55aSAlexander Duyck */ 4011802ab55aSAlexander Duyck partial_segs = len / mss; 4012d7fb5a80SAlexander Duyck if (partial_segs > 1) 4013802ab55aSAlexander Duyck mss *= partial_segs; 4014d7fb5a80SAlexander Duyck else 4015d7fb5a80SAlexander Duyck partial_segs = 0; 4016802ab55aSAlexander Duyck } 4017802ab55aSAlexander Duyck 401807b26c94SSteffen Klassert normal: 4019df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 4020df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 4021f4c50d99SHerbert Xu 4022f4c50d99SHerbert Xu do { 4023f4c50d99SHerbert Xu struct sk_buff *nskb; 40248cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 4025c8884eddSHerbert Xu int hsize; 4026f4c50d99SHerbert Xu int size; 4027f4c50d99SHerbert Xu 40283953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 40293953c46cSMarcelo Ricardo Leitner len = list_skb->len; 40303953c46cSMarcelo Ricardo Leitner } else { 4031df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 4032f4c50d99SHerbert Xu if (len > mss) 4033f4c50d99SHerbert Xu len = mss; 40343953c46cSMarcelo Ricardo Leitner } 4035f4c50d99SHerbert Xu 4036df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 4037f4c50d99SHerbert Xu 4038dbd50f23SXin Long if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) && 40391a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 40401a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 404189319d38SHerbert Xu 40429d8506ccSHerbert Xu i = 0; 40431a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 40441a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 40451fd819ecSMichael S. Tsirkin frag_skb = list_skb; 40461a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 40479d8506ccSHerbert Xu 40489d8506ccSHerbert Xu while (pos < offset + len) { 40499d8506ccSHerbert Xu BUG_ON(i >= nfrags); 40509d8506ccSHerbert Xu 40514e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 40529d8506ccSHerbert Xu if (pos + size > offset + len) 40539d8506ccSHerbert Xu break; 40549d8506ccSHerbert Xu 40559d8506ccSHerbert Xu i++; 40569d8506ccSHerbert Xu pos += size; 40574e1beba1SMichael S. Tsirkin frag++; 40589d8506ccSHerbert Xu } 40599d8506ccSHerbert Xu 40601a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 40611a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 406289319d38SHerbert Xu 4063f4c50d99SHerbert Xu if (unlikely(!nskb)) 4064f4c50d99SHerbert Xu goto err; 4065f4c50d99SHerbert Xu 40669d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 40679d8506ccSHerbert Xu kfree_skb(nskb); 40689d8506ccSHerbert Xu goto err; 40699d8506ccSHerbert Xu } 40709d8506ccSHerbert Xu 4071ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 407289319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 407389319d38SHerbert Xu kfree_skb(nskb); 407489319d38SHerbert Xu goto err; 407589319d38SHerbert Xu } 407689319d38SHerbert Xu 4077ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 407889319d38SHerbert Xu skb_release_head_state(nskb); 407989319d38SHerbert Xu __skb_push(nskb, doffset); 408089319d38SHerbert Xu } else { 408100b229f7SPaolo Abeni if (hsize < 0) 408200b229f7SPaolo Abeni hsize = 0; 4083dbd50f23SXin Long if (hsize > len || !sg) 4084dbd50f23SXin Long hsize = len; 4085dbd50f23SXin Long 4086c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 4087df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 4088c93bdd0eSMel Gorman NUMA_NO_NODE); 408989319d38SHerbert Xu 409089319d38SHerbert Xu if (unlikely(!nskb)) 409189319d38SHerbert Xu goto err; 409289319d38SHerbert Xu 409389319d38SHerbert Xu skb_reserve(nskb, headroom); 409489319d38SHerbert Xu __skb_put(nskb, doffset); 409589319d38SHerbert Xu } 409689319d38SHerbert Xu 4097f4c50d99SHerbert Xu if (segs) 4098f4c50d99SHerbert Xu tail->next = nskb; 4099f4c50d99SHerbert Xu else 4100f4c50d99SHerbert Xu segs = nskb; 4101f4c50d99SHerbert Xu tail = nskb; 4102f4c50d99SHerbert Xu 4103df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 4104f4c50d99SHerbert Xu 4105030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 4106fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 410768c33163SPravin B Shelar 4108df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 410968c33163SPravin B Shelar nskb->data - tnl_hlen, 411068c33163SPravin B Shelar doffset + tnl_hlen); 411189319d38SHerbert Xu 41129d8506ccSHerbert Xu if (nskb->len == len + doffset) 41131cdbcb79SSimon Horman goto perform_csum_check; 411489319d38SHerbert Xu 41157fbeffedSAlexander Duyck if (!sg) { 41161454c9faSYadu Kishore if (!csum) { 41177fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 41186f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 411976443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 412076443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 41211454c9faSYadu Kishore skb_put(nskb, 41221454c9faSYadu Kishore len), 41238d5930dfSAl Viro len); 41247e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 4125de843723STom Herbert skb_headroom(nskb) + doffset; 41261454c9faSYadu Kishore } else { 41271454c9faSYadu Kishore skb_copy_bits(head_skb, offset, 41281454c9faSYadu Kishore skb_put(nskb, len), 41291454c9faSYadu Kishore len); 41301454c9faSYadu Kishore } 4131f4c50d99SHerbert Xu continue; 4132f4c50d99SHerbert Xu } 4133f4c50d99SHerbert Xu 41348cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 4135f4c50d99SHerbert Xu 4136df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 4137d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4138f4c50d99SHerbert Xu 413906b4feb3SJonathan Lemon skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags & 414006b4feb3SJonathan Lemon SKBFL_SHARED_FRAG; 4141cef401deSEric Dumazet 4142bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4143bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4144bf5c25d6SWillem de Bruijn goto err; 4145bf5c25d6SWillem de Bruijn 41469d8506ccSHerbert Xu while (pos < offset + len) { 41479d8506ccSHerbert Xu if (i >= nfrags) { 41489d8506ccSHerbert Xu i = 0; 41491a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 41501a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 41511fd819ecSMichael S. Tsirkin frag_skb = list_skb; 415213acc94eSYonghong Song if (!skb_headlen(list_skb)) { 41539d8506ccSHerbert Xu BUG_ON(!nfrags); 415413acc94eSYonghong Song } else { 415513acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 41569d8506ccSHerbert Xu 415713acc94eSYonghong Song /* to make room for head_frag. */ 415813acc94eSYonghong Song i--; 415913acc94eSYonghong Song frag--; 416013acc94eSYonghong Song } 4161bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4162bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4163bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4164bf5c25d6SWillem de Bruijn goto err; 4165bf5c25d6SWillem de Bruijn 41661a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 41679d8506ccSHerbert Xu } 41689d8506ccSHerbert Xu 41699d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 41709d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 41719d8506ccSHerbert Xu net_warn_ratelimited( 41729d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 41739d8506ccSHerbert Xu pos, mss); 4174ff907a11SEric Dumazet err = -EINVAL; 41759d8506ccSHerbert Xu goto err; 41769d8506ccSHerbert Xu } 41779d8506ccSHerbert Xu 417813acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 41798cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 41808cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4181f4c50d99SHerbert Xu 4182f4c50d99SHerbert Xu if (pos < offset) { 4183b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 41848cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4185f4c50d99SHerbert Xu } 4186f4c50d99SHerbert Xu 418789319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4188f4c50d99SHerbert Xu 4189f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4190f4c50d99SHerbert Xu i++; 41914e1beba1SMichael S. Tsirkin frag++; 4192f4c50d99SHerbert Xu pos += size; 4193f4c50d99SHerbert Xu } else { 41948cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 419589319d38SHerbert Xu goto skip_fraglist; 4196f4c50d99SHerbert Xu } 4197f4c50d99SHerbert Xu 41988cb19905SMichael S. Tsirkin nskb_frag++; 4199f4c50d99SHerbert Xu } 4200f4c50d99SHerbert Xu 420189319d38SHerbert Xu skip_fraglist: 4202f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4203f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4204f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4205ec5f0615SPravin B Shelar 42061cdbcb79SSimon Horman perform_csum_check: 42077fbeffedSAlexander Duyck if (!csum) { 4208ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4209ff907a11SEric Dumazet __skb_linearize(nskb)) 4210ddff00d4SAlexander Duyck goto err; 4211ff907a11SEric Dumazet 42127fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4213ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 421476443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 421576443456SAlexander Duyck skb_checksum(nskb, doffset, 421676443456SAlexander Duyck nskb->len - doffset, 0); 42177e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 42187e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4219ec5f0615SPravin B Shelar } 4220df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4221f4c50d99SHerbert Xu 4222bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4223bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4224bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4225bec3cfdcSEric Dumazet */ 4226bec3cfdcSEric Dumazet segs->prev = tail; 4227432c856fSToshiaki Makita 4228802ab55aSAlexander Duyck if (partial_segs) { 422907b26c94SSteffen Klassert struct sk_buff *iter; 4230802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 423107b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4232802ab55aSAlexander Duyck 4233802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 423407b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4235802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4236802ab55aSAlexander Duyck 4237802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4238802ab55aSAlexander Duyck * our way back down the stack of protocols. 4239802ab55aSAlexander Duyck */ 424007b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 424107b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 424207b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 424307b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 424407b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 424507b26c94SSteffen Klassert } 424607b26c94SSteffen Klassert 424707b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 424807b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 424907b26c94SSteffen Klassert else if (tail != segs) 425007b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4251802ab55aSAlexander Duyck } 4252802ab55aSAlexander Duyck 4253432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4254432c856fSToshiaki Makita * using skb_set_owner_w(). 4255432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4256432c856fSToshiaki Makita */ 4257432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4258432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4259432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4260432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4261432c856fSToshiaki Makita } 4262f4c50d99SHerbert Xu return segs; 4263f4c50d99SHerbert Xu 4264f4c50d99SHerbert Xu err: 4265289dccbeSEric Dumazet kfree_skb_list(segs); 4266f4c50d99SHerbert Xu return ERR_PTR(err); 4267f4c50d99SHerbert Xu } 4268f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4269f4c50d99SHerbert Xu 4270df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4271df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4272df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4273df5042f4SFlorian Westphal 4274df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4275df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4276df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4277df5042f4SFlorian Westphal #endif 42784165079bSFlorian Westphal #ifdef CONFIG_XFRM 42794165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 42804165079bSFlorian Westphal #endif 428195a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 428295a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 428395a7233cSPaul Blakey #endif 42843ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 42853ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 42863ee17bc7SMat Martineau #endif 428778476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 428878476d31SJeremy Kerr [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow), 428978476d31SJeremy Kerr #endif 4290df5042f4SFlorian Westphal }; 4291df5042f4SFlorian Westphal 4292df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4293df5042f4SFlorian Westphal { 4294df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4295df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4296df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4297df5042f4SFlorian Westphal #endif 42984165079bSFlorian Westphal #ifdef CONFIG_XFRM 42994165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 43004165079bSFlorian Westphal #endif 430195a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 430295a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 430395a7233cSPaul Blakey #endif 43043ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 43053ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 43063ee17bc7SMat Martineau #endif 430778476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 430878476d31SJeremy Kerr skb_ext_type_len[SKB_EXT_MCTP] + 430978476d31SJeremy Kerr #endif 4310df5042f4SFlorian Westphal 0; 4311df5042f4SFlorian Westphal } 4312df5042f4SFlorian Westphal 4313df5042f4SFlorian Westphal static void skb_extensions_init(void) 4314df5042f4SFlorian Westphal { 4315df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4316df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4317df5042f4SFlorian Westphal 4318df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4319df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4320df5042f4SFlorian Westphal 0, 4321df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4322df5042f4SFlorian Westphal NULL); 4323df5042f4SFlorian Westphal } 4324df5042f4SFlorian Westphal #else 4325df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4326df5042f4SFlorian Westphal #endif 4327df5042f4SFlorian Westphal 43281da177e4SLinus Torvalds void __init skb_init(void) 43291da177e4SLinus Torvalds { 433079a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 43311da177e4SLinus Torvalds sizeof(struct sk_buff), 43321da177e4SLinus Torvalds 0, 4333e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 433479a8a642SKees Cook offsetof(struct sk_buff, cb), 433579a8a642SKees Cook sizeof_field(struct sk_buff, cb), 433620c2df83SPaul Mundt NULL); 4337d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4338d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4339d179cd12SDavid S. Miller 0, 4340e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 434120c2df83SPaul Mundt NULL); 4342df5042f4SFlorian Westphal skb_extensions_init(); 43431da177e4SLinus Torvalds } 43441da177e4SLinus Torvalds 434551c739d1SDavid S. Miller static int 434648a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 434748a1df65SJason A. Donenfeld unsigned int recursion_level) 4348716ea3a7SDavid Howells { 43491a028e50SDavid S. Miller int start = skb_headlen(skb); 43501a028e50SDavid S. Miller int i, copy = start - offset; 4351fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4352716ea3a7SDavid Howells int elt = 0; 4353716ea3a7SDavid Howells 435448a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 435548a1df65SJason A. Donenfeld return -EMSGSIZE; 435648a1df65SJason A. Donenfeld 4357716ea3a7SDavid Howells if (copy > 0) { 4358716ea3a7SDavid Howells if (copy > len) 4359716ea3a7SDavid Howells copy = len; 4360642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4361716ea3a7SDavid Howells elt++; 4362716ea3a7SDavid Howells if ((len -= copy) == 0) 4363716ea3a7SDavid Howells return elt; 4364716ea3a7SDavid Howells offset += copy; 4365716ea3a7SDavid Howells } 4366716ea3a7SDavid Howells 4367716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 43681a028e50SDavid S. Miller int end; 4369716ea3a7SDavid Howells 4370547b792cSIlpo Järvinen WARN_ON(start > offset + len); 43711a028e50SDavid S. Miller 43729e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4373716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4374716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 437548a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 437648a1df65SJason A. Donenfeld return -EMSGSIZE; 4377716ea3a7SDavid Howells 4378716ea3a7SDavid Howells if (copy > len) 4379716ea3a7SDavid Howells copy = len; 4380ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4381b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4382716ea3a7SDavid Howells elt++; 4383716ea3a7SDavid Howells if (!(len -= copy)) 4384716ea3a7SDavid Howells return elt; 4385716ea3a7SDavid Howells offset += copy; 4386716ea3a7SDavid Howells } 43871a028e50SDavid S. Miller start = end; 4388716ea3a7SDavid Howells } 4389716ea3a7SDavid Howells 4390fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 439148a1df65SJason A. Donenfeld int end, ret; 4392716ea3a7SDavid Howells 4393547b792cSIlpo Järvinen WARN_ON(start > offset + len); 43941a028e50SDavid S. Miller 4395fbb398a8SDavid S. Miller end = start + frag_iter->len; 4396716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 439748a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 439848a1df65SJason A. Donenfeld return -EMSGSIZE; 439948a1df65SJason A. Donenfeld 4400716ea3a7SDavid Howells if (copy > len) 4401716ea3a7SDavid Howells copy = len; 440248a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 440348a1df65SJason A. Donenfeld copy, recursion_level + 1); 440448a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 440548a1df65SJason A. Donenfeld return ret; 440648a1df65SJason A. Donenfeld elt += ret; 4407716ea3a7SDavid Howells if ((len -= copy) == 0) 4408716ea3a7SDavid Howells return elt; 4409716ea3a7SDavid Howells offset += copy; 4410716ea3a7SDavid Howells } 44111a028e50SDavid S. Miller start = end; 4412716ea3a7SDavid Howells } 4413716ea3a7SDavid Howells BUG_ON(len); 4414716ea3a7SDavid Howells return elt; 4415716ea3a7SDavid Howells } 4416716ea3a7SDavid Howells 441748a1df65SJason A. Donenfeld /** 441848a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 441948a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 442048a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 442148a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 442248a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 442348a1df65SJason A. Donenfeld * 442448a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 442548a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 442648a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 442748a1df65SJason A. Donenfeld * could not fit. 442848a1df65SJason A. Donenfeld */ 442948a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 443048a1df65SJason A. Donenfeld { 443148a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 443248a1df65SJason A. Donenfeld 443348a1df65SJason A. Donenfeld if (nsg <= 0) 443448a1df65SJason A. Donenfeld return nsg; 443548a1df65SJason A. Donenfeld 443648a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 443748a1df65SJason A. Donenfeld 443848a1df65SJason A. Donenfeld return nsg; 443948a1df65SJason A. Donenfeld } 444048a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 444148a1df65SJason A. Donenfeld 444225a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 444325a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 444425a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 444525a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 444625a91d8dSFan Du * 444725a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 444825a91d8dSFan Du * 1. sg_init_table 444925a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 445025a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 445125a91d8dSFan Du * 445225a91d8dSFan Du * This is equivalent to: 445325a91d8dSFan Du * 1. sg_init_table 445425a91d8dSFan Du * 2. skb_to_sgvec(payload1) 445525a91d8dSFan Du * 3. sg_unmark_end 445625a91d8dSFan Du * 4. skb_to_sgvec(payload2) 445725a91d8dSFan Du * 445825a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 445925a91d8dSFan Du * is more preferable. 446025a91d8dSFan Du */ 446125a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 446225a91d8dSFan Du int offset, int len) 446325a91d8dSFan Du { 446448a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 446525a91d8dSFan Du } 446625a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 446725a91d8dSFan Du 446851c739d1SDavid S. Miller 446951c739d1SDavid S. Miller 4470716ea3a7SDavid Howells /** 4471716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4472716ea3a7SDavid Howells * @skb: The socket buffer to check. 4473716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4474716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4475716ea3a7SDavid Howells * 4476716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4477716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4478716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4479716ea3a7SDavid Howells * 4480716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4481716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4482716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4483716ea3a7SDavid Howells * 4484716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4485716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4486716ea3a7SDavid Howells */ 4487716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4488716ea3a7SDavid Howells { 4489716ea3a7SDavid Howells int copyflag; 4490716ea3a7SDavid Howells int elt; 4491716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4492716ea3a7SDavid Howells 4493716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4494716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4495716ea3a7SDavid Howells * at the moment even if they are anonymous). 4496716ea3a7SDavid Howells */ 4497716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4498c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4499716ea3a7SDavid Howells return -ENOMEM; 4500716ea3a7SDavid Howells 4501716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 450221dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4503716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4504716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4505716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4506716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4507716ea3a7SDavid Howells 4508716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4509716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4510716ea3a7SDavid Howells return -ENOMEM; 4511716ea3a7SDavid Howells 4512716ea3a7SDavid Howells /* Voila! */ 4513716ea3a7SDavid Howells *trailer = skb; 4514716ea3a7SDavid Howells return 1; 4515716ea3a7SDavid Howells } 4516716ea3a7SDavid Howells 4517716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4518716ea3a7SDavid Howells 4519716ea3a7SDavid Howells elt = 1; 4520716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4521716ea3a7SDavid Howells copyflag = 0; 4522716ea3a7SDavid Howells 4523716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4524716ea3a7SDavid Howells int ntail = 0; 4525716ea3a7SDavid Howells 4526716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4527716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4528716ea3a7SDavid Howells * after it. */ 4529716ea3a7SDavid Howells 4530716ea3a7SDavid Howells if (skb_shared(skb1)) 4531716ea3a7SDavid Howells copyflag = 1; 4532716ea3a7SDavid Howells 4533716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4534716ea3a7SDavid Howells 4535716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4536716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 453721dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4538716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4539716ea3a7SDavid Howells ntail = tailbits + 128; 4540716ea3a7SDavid Howells } 4541716ea3a7SDavid Howells 4542716ea3a7SDavid Howells if (copyflag || 4543716ea3a7SDavid Howells skb_cloned(skb1) || 4544716ea3a7SDavid Howells ntail || 4545716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 454621dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4547716ea3a7SDavid Howells struct sk_buff *skb2; 4548716ea3a7SDavid Howells 4549716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4550716ea3a7SDavid Howells if (ntail == 0) 4551716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4552716ea3a7SDavid Howells else 4553716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4554716ea3a7SDavid Howells skb_headroom(skb1), 4555716ea3a7SDavid Howells ntail, 4556716ea3a7SDavid Howells GFP_ATOMIC); 4557716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4558716ea3a7SDavid Howells return -ENOMEM; 4559716ea3a7SDavid Howells 4560716ea3a7SDavid Howells if (skb1->sk) 4561716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4562716ea3a7SDavid Howells 4563716ea3a7SDavid Howells /* Looking around. Are we still alive? 4564716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4565716ea3a7SDavid Howells 4566716ea3a7SDavid Howells skb2->next = skb1->next; 4567716ea3a7SDavid Howells *skb_p = skb2; 4568716ea3a7SDavid Howells kfree_skb(skb1); 4569716ea3a7SDavid Howells skb1 = skb2; 4570716ea3a7SDavid Howells } 4571716ea3a7SDavid Howells elt++; 4572716ea3a7SDavid Howells *trailer = skb1; 4573716ea3a7SDavid Howells skb_p = &skb1->next; 4574716ea3a7SDavid Howells } 4575716ea3a7SDavid Howells 4576716ea3a7SDavid Howells return elt; 4577716ea3a7SDavid Howells } 4578b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4579716ea3a7SDavid Howells 4580b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4581b1faf566SEric Dumazet { 4582b1faf566SEric Dumazet struct sock *sk = skb->sk; 4583b1faf566SEric Dumazet 4584b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4585b1faf566SEric Dumazet } 4586b1faf566SEric Dumazet 45878605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 45888605330aSSoheil Hassas Yeganeh { 45898605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 45908605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 45918605330aSSoheil Hassas Yeganeh */ 45928605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 45938605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 45948605330aSSoheil Hassas Yeganeh } 45958605330aSSoheil Hassas Yeganeh 4596b1faf566SEric Dumazet /* 4597b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4598b1faf566SEric Dumazet */ 4599b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4600b1faf566SEric Dumazet { 4601b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4602ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4603b1faf566SEric Dumazet return -ENOMEM; 4604b1faf566SEric Dumazet 4605b1faf566SEric Dumazet skb_orphan(skb); 4606b1faf566SEric Dumazet skb->sk = sk; 4607b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4608b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 46098605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4610b1faf566SEric Dumazet 4611abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4612abb57ea4SEric Dumazet skb_dst_force(skb); 4613abb57ea4SEric Dumazet 4614b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4615b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 4616e3ae2365SAlexander Aring sk_error_report(sk); 4617b1faf566SEric Dumazet return 0; 4618b1faf566SEric Dumazet } 4619b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4620b1faf566SEric Dumazet 462183a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 462283a1a1a7SSoheil Hassas Yeganeh { 462383a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 462483a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 462583a1a1a7SSoheil Hassas Yeganeh } 462683a1a1a7SSoheil Hassas Yeganeh 4627364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4628364a9e93SWillem de Bruijn { 4629364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 463083a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 463183a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4632997d5c3fSEric Dumazet unsigned long flags; 4633364a9e93SWillem de Bruijn 4634997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4635364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 463638b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 463783a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 463838b25793SSoheil Hassas Yeganeh if (icmp_next) 4639985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 464038b25793SSoheil Hassas Yeganeh } 4641997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4642364a9e93SWillem de Bruijn 464383a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 464483a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 464583a1a1a7SSoheil Hassas Yeganeh 464683a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4647e3ae2365SAlexander Aring sk_error_report(sk); 4648364a9e93SWillem de Bruijn 4649364a9e93SWillem de Bruijn return skb; 4650364a9e93SWillem de Bruijn } 4651364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4652364a9e93SWillem de Bruijn 4653cab41c47SAlexander Duyck /** 4654cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4655cab41c47SAlexander Duyck * @skb: the skb to clone 4656cab41c47SAlexander Duyck * 4657cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4658cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4659cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4660cab41c47SAlexander Duyck * 4661cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4662cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4663cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4664cab41c47SAlexander Duyck * the sk_error_queue. 4665cab41c47SAlexander Duyck */ 466662bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 466762bccb8cSAlexander Duyck { 466862bccb8cSAlexander Duyck struct sock *sk = skb->sk; 466962bccb8cSAlexander Duyck struct sk_buff *clone; 467062bccb8cSAlexander Duyck 467141c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 467262bccb8cSAlexander Duyck return NULL; 467362bccb8cSAlexander Duyck 467462bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 467562bccb8cSAlexander Duyck if (!clone) { 467662bccb8cSAlexander Duyck sock_put(sk); 467762bccb8cSAlexander Duyck return NULL; 467862bccb8cSAlexander Duyck } 467962bccb8cSAlexander Duyck 468062bccb8cSAlexander Duyck clone->sk = sk; 468162bccb8cSAlexander Duyck clone->destructor = sock_efree; 468262bccb8cSAlexander Duyck 468362bccb8cSAlexander Duyck return clone; 468462bccb8cSAlexander Duyck } 468562bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 468662bccb8cSAlexander Duyck 468737846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 468837846ef0SAlexander Duyck struct sock *sk, 46894ef1b286SSoheil Hassas Yeganeh int tstype, 46904ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4691ac45f602SPatrick Ohly { 4692ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4693ac45f602SPatrick Ohly int err; 4694ac45f602SPatrick Ohly 46954ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 46964ef1b286SSoheil Hassas Yeganeh 4697ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4698ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4699ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4700ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4701e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 47024ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 47031862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 47044ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 470509c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 470642f67eeaSEric Dumazet if (sk_is_tcp(sk)) 47074ed2d765SWillem de Bruijn serr->ee.ee_data -= sk->sk_tskey; 47084ed2d765SWillem de Bruijn } 470929030374SEric Dumazet 4710ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 471129030374SEric Dumazet 4712ac45f602SPatrick Ohly if (err) 4713ac45f602SPatrick Ohly kfree_skb(skb); 4714ac45f602SPatrick Ohly } 471537846ef0SAlexander Duyck 4716b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4717b245be1fSWillem de Bruijn { 4718b245be1fSWillem de Bruijn bool ret; 4719b245be1fSWillem de Bruijn 4720b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4721b245be1fSWillem de Bruijn return true; 4722b245be1fSWillem de Bruijn 4723b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4724b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4725b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4726b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4727b245be1fSWillem de Bruijn return ret; 4728b245be1fSWillem de Bruijn } 4729b245be1fSWillem de Bruijn 473037846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 473137846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 473237846ef0SAlexander Duyck { 473337846ef0SAlexander Duyck struct sock *sk = skb->sk; 473437846ef0SAlexander Duyck 4735b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 473635b99dffSWillem de Bruijn goto err; 4737b245be1fSWillem de Bruijn 47389ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 47399ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 47409ac25fc0SEric Dumazet */ 474141c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 474237846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 47434ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 474437846ef0SAlexander Duyck sock_put(sk); 474535b99dffSWillem de Bruijn return; 474637846ef0SAlexander Duyck } 474735b99dffSWillem de Bruijn 474835b99dffSWillem de Bruijn err: 474935b99dffSWillem de Bruijn kfree_skb(skb); 47509ac25fc0SEric Dumazet } 475137846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 475237846ef0SAlexander Duyck 475337846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 4754e7ed11eeSYousuk Seung const struct sk_buff *ack_skb, 475537846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 475637846ef0SAlexander Duyck struct sock *sk, int tstype) 475737846ef0SAlexander Duyck { 475837846ef0SAlexander Duyck struct sk_buff *skb; 47594ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 476037846ef0SAlexander Duyck 47613a8dd971SWillem de Bruijn if (!sk) 47623a8dd971SWillem de Bruijn return; 47633a8dd971SWillem de Bruijn 4764b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4765b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4766b50a5c70SMiroslav Lichvar return; 4767b50a5c70SMiroslav Lichvar 47683a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 47693a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 477037846ef0SAlexander Duyck return; 477137846ef0SAlexander Duyck 47721c885808SFrancis Yan if (tsonly) { 47731c885808SFrancis Yan #ifdef CONFIG_INET 47741c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 477542f67eeaSEric Dumazet sk_is_tcp(sk)) { 4776e7ed11eeSYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb, 4777e7ed11eeSYousuk Seung ack_skb); 47784ef1b286SSoheil Hassas Yeganeh opt_stats = true; 47794ef1b286SSoheil Hassas Yeganeh } else 47801c885808SFrancis Yan #endif 47811c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 47821c885808SFrancis Yan } else { 478337846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 47841c885808SFrancis Yan } 478537846ef0SAlexander Duyck if (!skb) 478637846ef0SAlexander Duyck return; 478737846ef0SAlexander Duyck 478849ca0d8bSWillem de Bruijn if (tsonly) { 4789fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4790fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 479149ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 479249ca0d8bSWillem de Bruijn } 479349ca0d8bSWillem de Bruijn 479449ca0d8bSWillem de Bruijn if (hwtstamps) 479549ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 479649ca0d8bSWillem de Bruijn else 479749ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 479849ca0d8bSWillem de Bruijn 47994ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 480037846ef0SAlexander Duyck } 4801e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4802e7fd2885SWillem de Bruijn 4803e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4804e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4805e7fd2885SWillem de Bruijn { 4806e7ed11eeSYousuk Seung return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk, 4807e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4808e7fd2885SWillem de Bruijn } 4809ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4810ac45f602SPatrick Ohly 48116e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 48126e3e939fSJohannes Berg { 48136e3e939fSJohannes Berg struct sock *sk = skb->sk; 48146e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4815dd4f1072SEric Dumazet int err = 1; 48166e3e939fSJohannes Berg 48176e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 48186e3e939fSJohannes Berg skb->wifi_acked = acked; 48196e3e939fSJohannes Berg 48206e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 48216e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 48226e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 48236e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 48246e3e939fSJohannes Berg 4825dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4826dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4827dd4f1072SEric Dumazet */ 482841c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 48296e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4830dd4f1072SEric Dumazet sock_put(sk); 4831dd4f1072SEric Dumazet } 48326e3e939fSJohannes Berg if (err) 48336e3e939fSJohannes Berg kfree_skb(skb); 48346e3e939fSJohannes Berg } 48356e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 48366e3e939fSJohannes Berg 4837f35d9d8aSRusty Russell /** 4838f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4839f35d9d8aSRusty Russell * @skb: the skb to set 4840f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4841f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4842f35d9d8aSRusty Russell * 4843f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4844f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4845f35d9d8aSRusty Russell * 4846f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4847f35d9d8aSRusty Russell * returns false you should drop the packet. 4848f35d9d8aSRusty Russell */ 4849f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4850f35d9d8aSRusty Russell { 485152b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 485252b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 485352b5d6f5SEric Dumazet 485452b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 485552b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 485652b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4857f35d9d8aSRusty Russell return false; 4858f35d9d8aSRusty Russell } 4859f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 486052b5d6f5SEric Dumazet skb->csum_start = csum_start; 4861f35d9d8aSRusty Russell skb->csum_offset = off; 4862e5d5decaSJason Wang skb_set_transport_header(skb, start); 4863f35d9d8aSRusty Russell return true; 4864f35d9d8aSRusty Russell } 4865b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4866f35d9d8aSRusty Russell 4867ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4868ed1f50c3SPaul Durrant unsigned int max) 4869ed1f50c3SPaul Durrant { 4870ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4871ed1f50c3SPaul Durrant return 0; 4872ed1f50c3SPaul Durrant 4873ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4874ed1f50c3SPaul Durrant * won't need to do it again. 4875ed1f50c3SPaul Durrant */ 4876ed1f50c3SPaul Durrant if (max > skb->len) 4877ed1f50c3SPaul Durrant max = skb->len; 4878ed1f50c3SPaul Durrant 4879ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4880ed1f50c3SPaul Durrant return -ENOMEM; 4881ed1f50c3SPaul Durrant 4882ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4883ed1f50c3SPaul Durrant return -EPROTO; 4884ed1f50c3SPaul Durrant 4885ed1f50c3SPaul Durrant return 0; 4886ed1f50c3SPaul Durrant } 4887ed1f50c3SPaul Durrant 4888f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4889f9708b43SJan Beulich 4890f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4891f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4892f9708b43SJan Beulich unsigned int off) 4893f9708b43SJan Beulich { 4894f9708b43SJan Beulich int err; 4895f9708b43SJan Beulich 4896161d1792SKees Cook switch (proto) { 4897f9708b43SJan Beulich case IPPROTO_TCP: 4898f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4899f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4900f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4901f9708b43SJan Beulich offsetof(struct tcphdr, 4902f9708b43SJan Beulich check))) 4903f9708b43SJan Beulich err = -EPROTO; 4904f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4905f9708b43SJan Beulich 4906f9708b43SJan Beulich case IPPROTO_UDP: 4907f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4908f9708b43SJan Beulich off + sizeof(struct udphdr)); 4909f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4910f9708b43SJan Beulich offsetof(struct udphdr, 4911f9708b43SJan Beulich check))) 4912f9708b43SJan Beulich err = -EPROTO; 4913f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4914f9708b43SJan Beulich } 4915f9708b43SJan Beulich 4916f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4917f9708b43SJan Beulich } 4918f9708b43SJan Beulich 4919ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4920ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4921ed1f50c3SPaul Durrant */ 4922ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4923ed1f50c3SPaul Durrant 4924f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4925ed1f50c3SPaul Durrant { 4926ed1f50c3SPaul Durrant unsigned int off; 4927ed1f50c3SPaul Durrant bool fragment; 4928f9708b43SJan Beulich __sum16 *csum; 4929ed1f50c3SPaul Durrant int err; 4930ed1f50c3SPaul Durrant 4931ed1f50c3SPaul Durrant fragment = false; 4932ed1f50c3SPaul Durrant 4933ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4934ed1f50c3SPaul Durrant sizeof(struct iphdr), 4935ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4936ed1f50c3SPaul Durrant if (err < 0) 4937ed1f50c3SPaul Durrant goto out; 4938ed1f50c3SPaul Durrant 493911f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 4940ed1f50c3SPaul Durrant fragment = true; 4941ed1f50c3SPaul Durrant 4942ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 4943ed1f50c3SPaul Durrant 4944ed1f50c3SPaul Durrant err = -EPROTO; 4945ed1f50c3SPaul Durrant 4946ed1f50c3SPaul Durrant if (fragment) 4947ed1f50c3SPaul Durrant goto out; 4948ed1f50c3SPaul Durrant 4949f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4950f9708b43SJan Beulich if (IS_ERR(csum)) 4951f9708b43SJan Beulich return PTR_ERR(csum); 4952ed1f50c3SPaul Durrant 4953ed1f50c3SPaul Durrant if (recalculate) 4954f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4955ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 4956ed1f50c3SPaul Durrant skb->len - off, 4957f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 4958ed1f50c3SPaul Durrant err = 0; 4959ed1f50c3SPaul Durrant 4960ed1f50c3SPaul Durrant out: 4961ed1f50c3SPaul Durrant return err; 4962ed1f50c3SPaul Durrant } 4963ed1f50c3SPaul Durrant 4964ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4965ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 4966ed1f50c3SPaul Durrant */ 4967ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 4968ed1f50c3SPaul Durrant 4969ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 4970ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 4971ed1f50c3SPaul Durrant 4972ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4973ed1f50c3SPaul Durrant { 4974ed1f50c3SPaul Durrant int err; 4975ed1f50c3SPaul Durrant u8 nexthdr; 4976ed1f50c3SPaul Durrant unsigned int off; 4977ed1f50c3SPaul Durrant unsigned int len; 4978ed1f50c3SPaul Durrant bool fragment; 4979ed1f50c3SPaul Durrant bool done; 4980f9708b43SJan Beulich __sum16 *csum; 4981ed1f50c3SPaul Durrant 4982ed1f50c3SPaul Durrant fragment = false; 4983ed1f50c3SPaul Durrant done = false; 4984ed1f50c3SPaul Durrant 4985ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 4986ed1f50c3SPaul Durrant 4987ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4988ed1f50c3SPaul Durrant if (err < 0) 4989ed1f50c3SPaul Durrant goto out; 4990ed1f50c3SPaul Durrant 4991ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 4992ed1f50c3SPaul Durrant 4993ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4994ed1f50c3SPaul Durrant while (off <= len && !done) { 4995ed1f50c3SPaul Durrant switch (nexthdr) { 4996ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 4997ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 4998ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 4999ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 5000ed1f50c3SPaul Durrant 5001ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5002ed1f50c3SPaul Durrant off + 5003ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 5004ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5005ed1f50c3SPaul Durrant if (err < 0) 5006ed1f50c3SPaul Durrant goto out; 5007ed1f50c3SPaul Durrant 5008ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 5009ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5010ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 5011ed1f50c3SPaul Durrant break; 5012ed1f50c3SPaul Durrant } 5013ed1f50c3SPaul Durrant case IPPROTO_AH: { 5014ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 5015ed1f50c3SPaul Durrant 5016ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5017ed1f50c3SPaul Durrant off + 5018ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 5019ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5020ed1f50c3SPaul Durrant if (err < 0) 5021ed1f50c3SPaul Durrant goto out; 5022ed1f50c3SPaul Durrant 5023ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 5024ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5025ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 5026ed1f50c3SPaul Durrant break; 5027ed1f50c3SPaul Durrant } 5028ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 5029ed1f50c3SPaul Durrant struct frag_hdr *hp; 5030ed1f50c3SPaul Durrant 5031ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5032ed1f50c3SPaul Durrant off + 5033ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5034ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5035ed1f50c3SPaul Durrant if (err < 0) 5036ed1f50c3SPaul Durrant goto out; 5037ed1f50c3SPaul Durrant 5038ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5039ed1f50c3SPaul Durrant 5040ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5041ed1f50c3SPaul Durrant fragment = true; 5042ed1f50c3SPaul Durrant 5043ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5044ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5045ed1f50c3SPaul Durrant break; 5046ed1f50c3SPaul Durrant } 5047ed1f50c3SPaul Durrant default: 5048ed1f50c3SPaul Durrant done = true; 5049ed1f50c3SPaul Durrant break; 5050ed1f50c3SPaul Durrant } 5051ed1f50c3SPaul Durrant } 5052ed1f50c3SPaul Durrant 5053ed1f50c3SPaul Durrant err = -EPROTO; 5054ed1f50c3SPaul Durrant 5055ed1f50c3SPaul Durrant if (!done || fragment) 5056ed1f50c3SPaul Durrant goto out; 5057ed1f50c3SPaul Durrant 5058f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5059f9708b43SJan Beulich if (IS_ERR(csum)) 5060f9708b43SJan Beulich return PTR_ERR(csum); 5061ed1f50c3SPaul Durrant 5062ed1f50c3SPaul Durrant if (recalculate) 5063f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5064ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5065f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5066ed1f50c3SPaul Durrant err = 0; 5067ed1f50c3SPaul Durrant 5068ed1f50c3SPaul Durrant out: 5069ed1f50c3SPaul Durrant return err; 5070ed1f50c3SPaul Durrant } 5071ed1f50c3SPaul Durrant 5072ed1f50c3SPaul Durrant /** 5073ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5074ed1f50c3SPaul Durrant * @skb: the skb to set up 5075ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5076ed1f50c3SPaul Durrant */ 5077ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5078ed1f50c3SPaul Durrant { 5079ed1f50c3SPaul Durrant int err; 5080ed1f50c3SPaul Durrant 5081ed1f50c3SPaul Durrant switch (skb->protocol) { 5082ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5083f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5084ed1f50c3SPaul Durrant break; 5085ed1f50c3SPaul Durrant 5086ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5087ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5088ed1f50c3SPaul Durrant break; 5089ed1f50c3SPaul Durrant 5090ed1f50c3SPaul Durrant default: 5091ed1f50c3SPaul Durrant err = -EPROTO; 5092ed1f50c3SPaul Durrant break; 5093ed1f50c3SPaul Durrant } 5094ed1f50c3SPaul Durrant 5095ed1f50c3SPaul Durrant return err; 5096ed1f50c3SPaul Durrant } 5097ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5098ed1f50c3SPaul Durrant 50999afd85c9SLinus Lüssing /** 51009afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 51019afd85c9SLinus Lüssing * @skb: the skb to check 51029afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51039afd85c9SLinus Lüssing * 51049afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 51059afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 51069afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 51079afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 51089afd85c9SLinus Lüssing * 5109a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5110a516993fSLinus Lüssing * differs from the provided skb. 51119afd85c9SLinus Lüssing */ 51129afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 51139afd85c9SLinus Lüssing unsigned int transport_len) 51149afd85c9SLinus Lüssing { 51159afd85c9SLinus Lüssing struct sk_buff *skb_chk; 51169afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 51179afd85c9SLinus Lüssing int ret; 51189afd85c9SLinus Lüssing 5119a516993fSLinus Lüssing if (skb->len < len) 51209afd85c9SLinus Lüssing return NULL; 5121a516993fSLinus Lüssing else if (skb->len == len) 51229afd85c9SLinus Lüssing return skb; 51239afd85c9SLinus Lüssing 51249afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 51259afd85c9SLinus Lüssing if (!skb_chk) 51269afd85c9SLinus Lüssing return NULL; 51279afd85c9SLinus Lüssing 51289afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 51299afd85c9SLinus Lüssing if (ret) { 51309afd85c9SLinus Lüssing kfree_skb(skb_chk); 51319afd85c9SLinus Lüssing return NULL; 51329afd85c9SLinus Lüssing } 51339afd85c9SLinus Lüssing 51349afd85c9SLinus Lüssing return skb_chk; 51359afd85c9SLinus Lüssing } 51369afd85c9SLinus Lüssing 51379afd85c9SLinus Lüssing /** 51389afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 51399afd85c9SLinus Lüssing * @skb: the skb to check 51409afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51419afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 51429afd85c9SLinus Lüssing * 51439afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 51449afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 51459afd85c9SLinus Lüssing * 51469afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 51479afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 51489afd85c9SLinus Lüssing * 5149a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5150a516993fSLinus Lüssing * differs from the provided skb. 51519afd85c9SLinus Lüssing */ 51529afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 51539afd85c9SLinus Lüssing unsigned int transport_len, 51549afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 51559afd85c9SLinus Lüssing { 51569afd85c9SLinus Lüssing struct sk_buff *skb_chk; 51579afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5158fcba67c9SLinus Lüssing __sum16 ret; 51599afd85c9SLinus Lüssing 51609afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 51619afd85c9SLinus Lüssing if (!skb_chk) 5162a516993fSLinus Lüssing goto err; 51639afd85c9SLinus Lüssing 5164a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5165a516993fSLinus Lüssing goto err; 51669afd85c9SLinus Lüssing 51679b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 51689afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 51699b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 51709afd85c9SLinus Lüssing 5171a516993fSLinus Lüssing if (ret) 5172a516993fSLinus Lüssing goto err; 51739afd85c9SLinus Lüssing 51749afd85c9SLinus Lüssing return skb_chk; 5175a516993fSLinus Lüssing 5176a516993fSLinus Lüssing err: 5177a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5178a516993fSLinus Lüssing kfree_skb(skb_chk); 5179a516993fSLinus Lüssing 5180a516993fSLinus Lüssing return NULL; 5181a516993fSLinus Lüssing 51829afd85c9SLinus Lüssing } 51839afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 51849afd85c9SLinus Lüssing 51854497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 51864497b076SBen Hutchings { 5187e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5188e87cc472SJoe Perches skb->dev->name); 51894497b076SBen Hutchings } 51904497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5191bad43ca8SEric Dumazet 5192bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5193bad43ca8SEric Dumazet { 51943d861f66SEric Dumazet if (head_stolen) { 51953d861f66SEric Dumazet skb_release_head_state(skb); 5196bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 51973d861f66SEric Dumazet } else { 5198bad43ca8SEric Dumazet __kfree_skb(skb); 5199bad43ca8SEric Dumazet } 52003d861f66SEric Dumazet } 5201bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5202bad43ca8SEric Dumazet 5203bad43ca8SEric Dumazet /** 5204bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5205bad43ca8SEric Dumazet * @to: prior buffer 5206bad43ca8SEric Dumazet * @from: buffer to add 5207bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5208c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5209bad43ca8SEric Dumazet */ 5210bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5211bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5212bad43ca8SEric Dumazet { 5213c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5214bad43ca8SEric Dumazet int i, delta, len = from->len; 5215bad43ca8SEric Dumazet 5216bad43ca8SEric Dumazet *fragstolen = false; 5217bad43ca8SEric Dumazet 5218bad43ca8SEric Dumazet if (skb_cloned(to)) 5219bad43ca8SEric Dumazet return false; 5220bad43ca8SEric Dumazet 52216a5bcd84SIlias Apalodimas /* The page pool signature of struct page will eventually figure out 52226a5bcd84SIlias Apalodimas * which pages can be recycled or not but for now let's prohibit slab 52236a5bcd84SIlias Apalodimas * allocated and page_pool allocated SKBs from being coalesced. 52246a5bcd84SIlias Apalodimas */ 52256a5bcd84SIlias Apalodimas if (to->pp_recycle != from->pp_recycle) 52266a5bcd84SIlias Apalodimas return false; 52276a5bcd84SIlias Apalodimas 5228bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5229e93a0435SEric Dumazet if (len) 5230bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5231bad43ca8SEric Dumazet *delta_truesize = 0; 5232bad43ca8SEric Dumazet return true; 5233bad43ca8SEric Dumazet } 5234bad43ca8SEric Dumazet 5235c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5236c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5237c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5238bad43ca8SEric Dumazet return false; 52391f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 52401f8b977aSWillem de Bruijn return false; 5241bad43ca8SEric Dumazet 5242bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5243bad43ca8SEric Dumazet struct page *page; 5244bad43ca8SEric Dumazet unsigned int offset; 5245bad43ca8SEric Dumazet 5246c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5247c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5248bad43ca8SEric Dumazet return false; 5249bad43ca8SEric Dumazet 5250bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5251bad43ca8SEric Dumazet return false; 5252bad43ca8SEric Dumazet 5253bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5254bad43ca8SEric Dumazet 5255bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5256bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5257bad43ca8SEric Dumazet 5258c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5259bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5260bad43ca8SEric Dumazet *fragstolen = true; 5261bad43ca8SEric Dumazet } else { 5262c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5263c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5264bad43ca8SEric Dumazet return false; 5265bad43ca8SEric Dumazet 5266f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5267bad43ca8SEric Dumazet } 5268bad43ca8SEric Dumazet 5269bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5270bad43ca8SEric Dumazet 5271c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5272c818fa9eSEric Dumazet from_shinfo->frags, 5273c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5274c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5275bad43ca8SEric Dumazet 5276bad43ca8SEric Dumazet if (!skb_cloned(from)) 5277c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5278bad43ca8SEric Dumazet 52798ea853fdSLi RongQing /* if the skb is not cloned this does nothing 52808ea853fdSLi RongQing * since we set nr_frags to 0. 52818ea853fdSLi RongQing */ 5282c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5283c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5284bad43ca8SEric Dumazet 5285bad43ca8SEric Dumazet to->truesize += delta; 5286bad43ca8SEric Dumazet to->len += len; 5287bad43ca8SEric Dumazet to->data_len += len; 5288bad43ca8SEric Dumazet 5289bad43ca8SEric Dumazet *delta_truesize = delta; 5290bad43ca8SEric Dumazet return true; 5291bad43ca8SEric Dumazet } 5292bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5293621e84d6SNicolas Dichtel 5294621e84d6SNicolas Dichtel /** 52958b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5296621e84d6SNicolas Dichtel * 5297621e84d6SNicolas Dichtel * @skb: buffer to clean 52988b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5299621e84d6SNicolas Dichtel * 53008b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 53018b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 53028b27f277SNicolas Dichtel * operations. 53038b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 53048b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 53058b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5306621e84d6SNicolas Dichtel */ 53078b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5308621e84d6SNicolas Dichtel { 5309621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5310621e84d6SNicolas Dichtel skb->skb_iif = 0; 531160ff7467SWANG Cong skb->ignore_df = 0; 5312621e84d6SNicolas Dichtel skb_dst_drop(skb); 5313174e2381SFlorian Westphal skb_ext_reset(skb); 5314895b5c9fSFlorian Westphal nf_reset_ct(skb); 5315621e84d6SNicolas Dichtel nf_reset_trace(skb); 5316213dd74aSHerbert Xu 53176f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 53186f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5319875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 53206f9a5069SPetr Machata #endif 53216f9a5069SPetr Machata 5322213dd74aSHerbert Xu if (!xnet) 5323213dd74aSHerbert Xu return; 5324213dd74aSHerbert Xu 53252b5ec1a5SYe Yin ipvs_reset(skb); 5326213dd74aSHerbert Xu skb->mark = 0; 5327c47d8c2fSJesus Sanchez-Palencia skb->tstamp = 0; 5328621e84d6SNicolas Dichtel } 5329621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5330de960aa9SFlorian Westphal 5331de960aa9SFlorian Westphal /** 5332de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5333de960aa9SFlorian Westphal * 5334de960aa9SFlorian Westphal * @skb: GSO skb 5335de960aa9SFlorian Westphal * 5336de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5337de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5338de960aa9SFlorian Westphal * 5339de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5340de960aa9SFlorian Westphal */ 5341a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5342de960aa9SFlorian Westphal { 5343de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5344f993bc25SFlorian Westphal unsigned int thlen = 0; 5345f993bc25SFlorian Westphal 5346f993bc25SFlorian Westphal if (skb->encapsulation) { 5347f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5348f993bc25SFlorian Westphal skb_transport_header(skb); 5349de960aa9SFlorian Westphal 5350de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5351f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5352f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5353f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 53541dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 535590017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5356ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5357ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5358f993bc25SFlorian Westphal } 53596d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 53606d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 53616d39d589SFlorian Westphal * accounted for. 53626d39d589SFlorian Westphal */ 5363f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5364de960aa9SFlorian Westphal } 5365a4a77718SDaniel Axtens 5366a4a77718SDaniel Axtens /** 5367a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5368a4a77718SDaniel Axtens * 5369a4a77718SDaniel Axtens * @skb: GSO skb 5370a4a77718SDaniel Axtens * 5371a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5372a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5373a4a77718SDaniel Axtens * 5374a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5375a4a77718SDaniel Axtens */ 5376a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5377a4a77718SDaniel Axtens { 5378a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5379a4a77718SDaniel Axtens skb_network_header(skb); 5380a4a77718SDaniel Axtens 5381a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5382a4a77718SDaniel Axtens } 5383a4a77718SDaniel Axtens 5384a4a77718SDaniel Axtens /** 5385a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5386a4a77718SDaniel Axtens * 5387a4a77718SDaniel Axtens * @skb: GSO skb 5388a4a77718SDaniel Axtens * 5389a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5390a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5391a4a77718SDaniel Axtens * headers (TCP/UDP). 5392a4a77718SDaniel Axtens */ 5393a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5394a4a77718SDaniel Axtens { 5395a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5396a4a77718SDaniel Axtens 5397a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5398a4a77718SDaniel Axtens } 53990d5501c1SVlad Yasevich 5400ae7ef81eSMarcelo Ricardo Leitner /** 54012b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 54022b16f048SDaniel Axtens * 54032b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 54042b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 54052b16f048SDaniel Axtens * 54062b16f048SDaniel Axtens * We might want to check: 54072b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 54082b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 54092b16f048SDaniel Axtens * 54102b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 54112b16f048SDaniel Axtens * 541249682bfaSMathieu Malaterre * @skb: GSO skb 541349682bfaSMathieu Malaterre * 54142b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 54152b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 54162b16f048SDaniel Axtens * 54172b16f048SDaniel Axtens * @max_len: The maximum permissible length. 54182b16f048SDaniel Axtens * 54192b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 54202b16f048SDaniel Axtens */ 54212b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 54222b16f048SDaniel Axtens unsigned int seg_len, 54232b16f048SDaniel Axtens unsigned int max_len) { 54242b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 54252b16f048SDaniel Axtens const struct sk_buff *iter; 54262b16f048SDaniel Axtens 54272b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 54282b16f048SDaniel Axtens return seg_len <= max_len; 54292b16f048SDaniel Axtens 54302b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 54312b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 54322b16f048SDaniel Axtens 54332b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 54342b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 54352b16f048SDaniel Axtens return false; 54362b16f048SDaniel Axtens } 54372b16f048SDaniel Axtens 54382b16f048SDaniel Axtens return true; 54392b16f048SDaniel Axtens } 54402b16f048SDaniel Axtens 54412b16f048SDaniel Axtens /** 5442779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5443ae7ef81eSMarcelo Ricardo Leitner * 5444ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 544576f21b99SDavid S. Miller * @mtu: MTU to validate against 5446ae7ef81eSMarcelo Ricardo Leitner * 5447779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5448779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5449779b7931SDaniel Axtens * payload. 5450ae7ef81eSMarcelo Ricardo Leitner */ 5451779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5452ae7ef81eSMarcelo Ricardo Leitner { 54532b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5454ae7ef81eSMarcelo Ricardo Leitner } 5455779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5456ae7ef81eSMarcelo Ricardo Leitner 54572b16f048SDaniel Axtens /** 54582b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 54592b16f048SDaniel Axtens * 54602b16f048SDaniel Axtens * @skb: GSO skb 54612b16f048SDaniel Axtens * @len: length to validate against 54622b16f048SDaniel Axtens * 54632b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 54642b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 54652b16f048SDaniel Axtens */ 54662b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 54672b16f048SDaniel Axtens { 54682b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 54692b16f048SDaniel Axtens } 54702b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 54712b16f048SDaniel Axtens 54720d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 54730d5501c1SVlad Yasevich { 5474d85e8be2SYuya Kusakabe int mac_len, meta_len; 5475d85e8be2SYuya Kusakabe void *meta; 54764bbb3e0eSToshiaki Makita 54770d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 54780d5501c1SVlad Yasevich kfree_skb(skb); 54790d5501c1SVlad Yasevich return NULL; 54800d5501c1SVlad Yasevich } 54810d5501c1SVlad Yasevich 54824bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5483ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 54844bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 54854bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5486ae474573SToshiaki Makita } 5487d85e8be2SYuya Kusakabe 5488d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5489d85e8be2SYuya Kusakabe if (meta_len) { 5490d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5491d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5492d85e8be2SYuya Kusakabe } 5493d85e8be2SYuya Kusakabe 54940d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 54950d5501c1SVlad Yasevich return skb; 54960d5501c1SVlad Yasevich } 54970d5501c1SVlad Yasevich 54980d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 54990d5501c1SVlad Yasevich { 55000d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 55010d5501c1SVlad Yasevich u16 vlan_tci; 55020d5501c1SVlad Yasevich 5503df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 55040d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 55050d5501c1SVlad Yasevich return skb; 55060d5501c1SVlad Yasevich } 55070d5501c1SVlad Yasevich 55080d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 55090d5501c1SVlad Yasevich if (unlikely(!skb)) 55100d5501c1SVlad Yasevich goto err_free; 551155eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 551255eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 55130d5501c1SVlad Yasevich goto err_free; 55140d5501c1SVlad Yasevich 55150d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 55160d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 55170d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 55180d5501c1SVlad Yasevich 55190d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 55200d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 55210d5501c1SVlad Yasevich 55220d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 55230d5501c1SVlad Yasevich if (unlikely(!skb)) 55240d5501c1SVlad Yasevich goto err_free; 55250d5501c1SVlad Yasevich 55260d5501c1SVlad Yasevich skb_reset_network_header(skb); 55278be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 55280d5501c1SVlad Yasevich skb_reset_transport_header(skb); 55290d5501c1SVlad Yasevich skb_reset_mac_len(skb); 55300d5501c1SVlad Yasevich 55310d5501c1SVlad Yasevich return skb; 55320d5501c1SVlad Yasevich 55330d5501c1SVlad Yasevich err_free: 55340d5501c1SVlad Yasevich kfree_skb(skb); 55350d5501c1SVlad Yasevich return NULL; 55360d5501c1SVlad Yasevich } 55370d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 55382e4e4410SEric Dumazet 5539e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5540e2195121SJiri Pirko { 5541e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5542e2195121SJiri Pirko return -ENOMEM; 5543e2195121SJiri Pirko 5544e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5545e2195121SJiri Pirko return 0; 5546e2195121SJiri Pirko 5547e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5548e2195121SJiri Pirko } 5549e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5550e2195121SJiri Pirko 5551bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5552bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5553bfca4c52SShmulik Ladkani */ 5554bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 555593515d53SJiri Pirko { 555693515d53SJiri Pirko struct vlan_hdr *vhdr; 5557b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 555893515d53SJiri Pirko int err; 555993515d53SJiri Pirko 5560b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5561b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5562b6a79208SShmulik Ladkani offset)) { 5563b6a79208SShmulik Ladkani return -EINVAL; 5564b6a79208SShmulik Ladkani } 5565b6a79208SShmulik Ladkani 556693515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 556793515d53SJiri Pirko if (unlikely(err)) 5568b6a79208SShmulik Ladkani return err; 556993515d53SJiri Pirko 557093515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 557193515d53SJiri Pirko 557293515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 557393515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 557493515d53SJiri Pirko 557593515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 557693515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 557793515d53SJiri Pirko 557893515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 557993515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 558093515d53SJiri Pirko 558193515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 558293515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 558393515d53SJiri Pirko 558493515d53SJiri Pirko skb_reset_mac_len(skb); 558593515d53SJiri Pirko 558693515d53SJiri Pirko return err; 558793515d53SJiri Pirko } 5588bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 558993515d53SJiri Pirko 5590b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5591b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5592b6a79208SShmulik Ladkani */ 559393515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 559493515d53SJiri Pirko { 559593515d53SJiri Pirko u16 vlan_tci; 559693515d53SJiri Pirko __be16 vlan_proto; 559793515d53SJiri Pirko int err; 559893515d53SJiri Pirko 5599df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5600b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 560193515d53SJiri Pirko } else { 5602ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 560393515d53SJiri Pirko return 0; 560493515d53SJiri Pirko 560593515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 560693515d53SJiri Pirko if (err) 560793515d53SJiri Pirko return err; 560893515d53SJiri Pirko } 560993515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5610ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 561193515d53SJiri Pirko return 0; 561293515d53SJiri Pirko 561393515d53SJiri Pirko vlan_proto = skb->protocol; 561493515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 561593515d53SJiri Pirko if (unlikely(err)) 561693515d53SJiri Pirko return err; 561793515d53SJiri Pirko 561893515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 561993515d53SJiri Pirko return 0; 562093515d53SJiri Pirko } 562193515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 562293515d53SJiri Pirko 5623b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5624b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5625b6a79208SShmulik Ladkani */ 562693515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 562793515d53SJiri Pirko { 5628df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5629b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 563093515d53SJiri Pirko int err; 563193515d53SJiri Pirko 5632b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5633b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5634b6a79208SShmulik Ladkani offset)) { 5635b6a79208SShmulik Ladkani return -EINVAL; 5636b6a79208SShmulik Ladkani } 5637b6a79208SShmulik Ladkani 563893515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5639df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5640b6a79208SShmulik Ladkani if (err) 564193515d53SJiri Pirko return err; 56429241e2dfSDaniel Borkmann 564393515d53SJiri Pirko skb->protocol = skb->vlan_proto; 564493515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 564593515d53SJiri Pirko 56466b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 564793515d53SJiri Pirko } 564893515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 564993515d53SJiri Pirko return 0; 565093515d53SJiri Pirko } 565193515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 565293515d53SJiri Pirko 565319fbcb36SGuillaume Nault /** 565419fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 565519fbcb36SGuillaume Nault * 565619fbcb36SGuillaume Nault * @skb: Socket buffer to modify 565719fbcb36SGuillaume Nault * 565819fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 565919fbcb36SGuillaume Nault * 566019fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 566119fbcb36SGuillaume Nault * present. 566219fbcb36SGuillaume Nault * 566319fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 566419fbcb36SGuillaume Nault */ 566519fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 566619fbcb36SGuillaume Nault { 566719fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 566819fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 566919fbcb36SGuillaume Nault return -EPROTO; 567019fbcb36SGuillaume Nault 567119fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 567219fbcb36SGuillaume Nault skb_reset_mac_header(skb); 567319fbcb36SGuillaume Nault skb_reset_mac_len(skb); 567419fbcb36SGuillaume Nault 567519fbcb36SGuillaume Nault return 0; 567619fbcb36SGuillaume Nault } 567719fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 567819fbcb36SGuillaume Nault 567919fbcb36SGuillaume Nault /** 568019fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 568119fbcb36SGuillaume Nault * 568219fbcb36SGuillaume Nault * @skb: Socket buffer to modify 568319fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 568419fbcb36SGuillaume Nault * @src: Source MAC address of the new header 568519fbcb36SGuillaume Nault * 568619fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 568719fbcb36SGuillaume Nault * 568819fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 568919fbcb36SGuillaume Nault * 569019fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 569119fbcb36SGuillaume Nault */ 569219fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 569319fbcb36SGuillaume Nault const unsigned char *src) 569419fbcb36SGuillaume Nault { 569519fbcb36SGuillaume Nault struct ethhdr *eth; 569619fbcb36SGuillaume Nault int err; 569719fbcb36SGuillaume Nault 569819fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 569919fbcb36SGuillaume Nault return -EPROTO; 570019fbcb36SGuillaume Nault 570119fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 570219fbcb36SGuillaume Nault if (err < 0) 570319fbcb36SGuillaume Nault return err; 570419fbcb36SGuillaume Nault 570519fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 570619fbcb36SGuillaume Nault skb_reset_mac_header(skb); 570719fbcb36SGuillaume Nault skb_reset_mac_len(skb); 570819fbcb36SGuillaume Nault 570919fbcb36SGuillaume Nault eth = eth_hdr(skb); 571019fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 571119fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 571219fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 571319fbcb36SGuillaume Nault 571419fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 571519fbcb36SGuillaume Nault 571619fbcb36SGuillaume Nault return 0; 571719fbcb36SGuillaume Nault } 571819fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 571919fbcb36SGuillaume Nault 57208822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 57218822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 57228822e270SJohn Hurley __be16 ethertype) 57238822e270SJohn Hurley { 57248822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 57258822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 57268822e270SJohn Hurley 57278822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 57288822e270SJohn Hurley } 57298822e270SJohn Hurley 57308822e270SJohn Hurley hdr->h_proto = ethertype; 57318822e270SJohn Hurley } 57328822e270SJohn Hurley 57338822e270SJohn Hurley /** 5734e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5735e7dbfed1SMartin Varghese * the packet 57368822e270SJohn Hurley * 57378822e270SJohn Hurley * @skb: buffer 57388822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 57398822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5740fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5741e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5742e7dbfed1SMartin Varghese * ethernet 57438822e270SJohn Hurley * 57448822e270SJohn Hurley * Expects skb->data at mac header. 57458822e270SJohn Hurley * 57468822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 57478822e270SJohn Hurley */ 5748fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5749d04ac224SMartin Varghese int mac_len, bool ethernet) 57508822e270SJohn Hurley { 57518822e270SJohn Hurley struct mpls_shim_hdr *lse; 57528822e270SJohn Hurley int err; 57538822e270SJohn Hurley 57548822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 57558822e270SJohn Hurley return -EINVAL; 57568822e270SJohn Hurley 57578822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 57588822e270SJohn Hurley if (skb->encapsulation) 57598822e270SJohn Hurley return -EINVAL; 57608822e270SJohn Hurley 57618822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 57628822e270SJohn Hurley if (unlikely(err)) 57638822e270SJohn Hurley return err; 57648822e270SJohn Hurley 57658822e270SJohn Hurley if (!skb->inner_protocol) { 5766e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 57678822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 57688822e270SJohn Hurley } 57698822e270SJohn Hurley 57708822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 57718822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5772fa4e0f88SDavide Caratti mac_len); 57738822e270SJohn Hurley skb_reset_mac_header(skb); 5774fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5775e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 57768822e270SJohn Hurley 57778822e270SJohn Hurley lse = mpls_hdr(skb); 57788822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 57798822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 57808822e270SJohn Hurley 57814296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 57828822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 57838822e270SJohn Hurley skb->protocol = mpls_proto; 57848822e270SJohn Hurley 57858822e270SJohn Hurley return 0; 57868822e270SJohn Hurley } 57878822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 57888822e270SJohn Hurley 57892e4e4410SEric Dumazet /** 5790ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5791ed246ceeSJohn Hurley * 5792ed246ceeSJohn Hurley * @skb: buffer 5793ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5794fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 579576f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5796ed246ceeSJohn Hurley * 5797ed246ceeSJohn Hurley * Expects skb->data at mac header. 5798ed246ceeSJohn Hurley * 5799ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5800ed246ceeSJohn Hurley */ 5801040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5802040b5cfbSMartin Varghese bool ethernet) 5803ed246ceeSJohn Hurley { 5804ed246ceeSJohn Hurley int err; 5805ed246ceeSJohn Hurley 5806ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5807dedc5a08SDavide Caratti return 0; 5808ed246ceeSJohn Hurley 5809fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5810ed246ceeSJohn Hurley if (unlikely(err)) 5811ed246ceeSJohn Hurley return err; 5812ed246ceeSJohn Hurley 5813ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5814ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5815fa4e0f88SDavide Caratti mac_len); 5816ed246ceeSJohn Hurley 5817ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5818ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5819fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5820ed246ceeSJohn Hurley 58214296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5822ed246ceeSJohn Hurley struct ethhdr *hdr; 5823ed246ceeSJohn Hurley 5824ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5825ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5826ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5827ed246ceeSJohn Hurley } 5828ed246ceeSJohn Hurley skb->protocol = next_proto; 5829ed246ceeSJohn Hurley 5830ed246ceeSJohn Hurley return 0; 5831ed246ceeSJohn Hurley } 5832ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5833ed246ceeSJohn Hurley 5834ed246ceeSJohn Hurley /** 5835d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5836d27cf5c5SJohn Hurley * 5837d27cf5c5SJohn Hurley * @skb: buffer 5838d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5839d27cf5c5SJohn Hurley * 5840d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5841d27cf5c5SJohn Hurley * 5842d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5843d27cf5c5SJohn Hurley */ 5844d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5845d27cf5c5SJohn Hurley { 5846d27cf5c5SJohn Hurley int err; 5847d27cf5c5SJohn Hurley 5848d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5849d27cf5c5SJohn Hurley return -EINVAL; 5850d27cf5c5SJohn Hurley 5851d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5852d27cf5c5SJohn Hurley if (unlikely(err)) 5853d27cf5c5SJohn Hurley return err; 5854d27cf5c5SJohn Hurley 5855d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5856d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5857d27cf5c5SJohn Hurley 5858d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5859d27cf5c5SJohn Hurley } 5860d27cf5c5SJohn Hurley 5861d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 5862d27cf5c5SJohn Hurley 5863d27cf5c5SJohn Hurley return 0; 5864d27cf5c5SJohn Hurley } 5865d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 5866d27cf5c5SJohn Hurley 5867d27cf5c5SJohn Hurley /** 58682a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 58692a2ea508SJohn Hurley * 58702a2ea508SJohn Hurley * @skb: buffer 58712a2ea508SJohn Hurley * 58722a2ea508SJohn Hurley * Expects skb->data at mac header. 58732a2ea508SJohn Hurley * 58742a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 58752a2ea508SJohn Hurley */ 58762a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 58772a2ea508SJohn Hurley { 58782a2ea508SJohn Hurley u32 lse; 58792a2ea508SJohn Hurley u8 ttl; 58802a2ea508SJohn Hurley 58812a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 58822a2ea508SJohn Hurley return -EINVAL; 58832a2ea508SJohn Hurley 588413de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 588513de4ed9SDavide Caratti return -ENOMEM; 588613de4ed9SDavide Caratti 58872a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 58882a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 58892a2ea508SJohn Hurley if (!--ttl) 58902a2ea508SJohn Hurley return -EINVAL; 58912a2ea508SJohn Hurley 58922a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 58932a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 58942a2ea508SJohn Hurley 58952a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 58962a2ea508SJohn Hurley } 58972a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 58982a2ea508SJohn Hurley 58992a2ea508SJohn Hurley /** 59002e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 59012e4e4410SEric Dumazet * 5902de3f0d0eSMasanari Iida * @header_len: size of linear part 5903de3f0d0eSMasanari Iida * @data_len: needed length in frags 5904de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5905de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5906de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 59072e4e4410SEric Dumazet * 59082e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 59092e4e4410SEric Dumazet */ 59102e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 59112e4e4410SEric Dumazet unsigned long data_len, 59122e4e4410SEric Dumazet int max_page_order, 59132e4e4410SEric Dumazet int *errcode, 59142e4e4410SEric Dumazet gfp_t gfp_mask) 59152e4e4410SEric Dumazet { 59162e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 59172e4e4410SEric Dumazet unsigned long chunk; 59182e4e4410SEric Dumazet struct sk_buff *skb; 59192e4e4410SEric Dumazet struct page *page; 59202e4e4410SEric Dumazet int i; 59212e4e4410SEric Dumazet 59222e4e4410SEric Dumazet *errcode = -EMSGSIZE; 59232e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 59242e4e4410SEric Dumazet * high order pages... 59252e4e4410SEric Dumazet */ 59262e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 59272e4e4410SEric Dumazet return NULL; 59282e4e4410SEric Dumazet 59292e4e4410SEric Dumazet *errcode = -ENOBUFS; 5930f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 59312e4e4410SEric Dumazet if (!skb) 59322e4e4410SEric Dumazet return NULL; 59332e4e4410SEric Dumazet 59342e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 59352e4e4410SEric Dumazet 59362e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 59372e4e4410SEric Dumazet int order = max_page_order; 59382e4e4410SEric Dumazet 59392e4e4410SEric Dumazet while (order) { 59402e4e4410SEric Dumazet if (npages >= 1 << order) { 5941d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 59422e4e4410SEric Dumazet __GFP_COMP | 5943d14b56f5SMichal Hocko __GFP_NOWARN, 59442e4e4410SEric Dumazet order); 59452e4e4410SEric Dumazet if (page) 59462e4e4410SEric Dumazet goto fill_page; 59472e4e4410SEric Dumazet /* Do not retry other high order allocations */ 59482e4e4410SEric Dumazet order = 1; 59492e4e4410SEric Dumazet max_page_order = 0; 59502e4e4410SEric Dumazet } 59512e4e4410SEric Dumazet order--; 59522e4e4410SEric Dumazet } 59532e4e4410SEric Dumazet page = alloc_page(gfp_mask); 59542e4e4410SEric Dumazet if (!page) 59552e4e4410SEric Dumazet goto failure; 59562e4e4410SEric Dumazet fill_page: 59572e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 59582e4e4410SEric Dumazet PAGE_SIZE << order); 59592e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 59602e4e4410SEric Dumazet data_len -= chunk; 59612e4e4410SEric Dumazet npages -= 1 << order; 59622e4e4410SEric Dumazet } 59632e4e4410SEric Dumazet return skb; 59642e4e4410SEric Dumazet 59652e4e4410SEric Dumazet failure: 59662e4e4410SEric Dumazet kfree_skb(skb); 59672e4e4410SEric Dumazet return NULL; 59682e4e4410SEric Dumazet } 59692e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 59706fa01ccdSSowmini Varadhan 59716fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 59726fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 59736fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 59746fa01ccdSSowmini Varadhan { 59756fa01ccdSSowmini Varadhan int i; 59766fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 59776fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 59786fa01ccdSSowmini Varadhan u8 *data; 59796fa01ccdSSowmini Varadhan 59806fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 59816fa01ccdSSowmini Varadhan 59826fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 59836fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 59846fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 59856fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 59866fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 59876fa01ccdSSowmini Varadhan if (!data) 59886fa01ccdSSowmini Varadhan return -ENOMEM; 59896fa01ccdSSowmini Varadhan 59906fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 59916fa01ccdSSowmini Varadhan 59926fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 59936fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 59946fa01ccdSSowmini Varadhan skb->len -= off; 59956fa01ccdSSowmini Varadhan 59966fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 59976fa01ccdSSowmini Varadhan skb_shinfo(skb), 59986fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 59996fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 60006fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 60016fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 60026fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 60036fa01ccdSSowmini Varadhan kfree(data); 60046fa01ccdSSowmini Varadhan return -ENOMEM; 60056fa01ccdSSowmini Varadhan } 60066fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 60076fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 60086fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 60096fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 60106fa01ccdSSowmini Varadhan skb_release_data(skb); 60116fa01ccdSSowmini Varadhan } else { 60126fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 60136fa01ccdSSowmini Varadhan * relocate values 60146fa01ccdSSowmini Varadhan */ 60156fa01ccdSSowmini Varadhan skb_free_head(skb); 60166fa01ccdSSowmini Varadhan } 60176fa01ccdSSowmini Varadhan 60186fa01ccdSSowmini Varadhan skb->head = data; 60196fa01ccdSSowmini Varadhan skb->data = data; 60206fa01ccdSSowmini Varadhan skb->head_frag = 0; 60216fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 60226fa01ccdSSowmini Varadhan skb->end = size; 60236fa01ccdSSowmini Varadhan #else 60246fa01ccdSSowmini Varadhan skb->end = skb->head + size; 60256fa01ccdSSowmini Varadhan #endif 60266fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 60276fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 60286fa01ccdSSowmini Varadhan skb->cloned = 0; 60296fa01ccdSSowmini Varadhan skb->hdr_len = 0; 60306fa01ccdSSowmini Varadhan skb->nohdr = 0; 60316fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 60326fa01ccdSSowmini Varadhan 60336fa01ccdSSowmini Varadhan return 0; 60346fa01ccdSSowmini Varadhan } 60356fa01ccdSSowmini Varadhan 60366fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 60376fa01ccdSSowmini Varadhan 60386fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 60396fa01ccdSSowmini Varadhan * pskb_carve() 60406fa01ccdSSowmini Varadhan */ 60416fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 60426fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 60436fa01ccdSSowmini Varadhan gfp_t gfp_mask) 60446fa01ccdSSowmini Varadhan { 60456fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 60466fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 60476fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 60486fa01ccdSSowmini Varadhan 60496fa01ccdSSowmini Varadhan do { 60506fa01ccdSSowmini Varadhan if (!list) { 60516fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 60526fa01ccdSSowmini Varadhan return -EFAULT; 60536fa01ccdSSowmini Varadhan } 60546fa01ccdSSowmini Varadhan if (list->len <= eat) { 60556fa01ccdSSowmini Varadhan /* Eaten as whole. */ 60566fa01ccdSSowmini Varadhan eat -= list->len; 60576fa01ccdSSowmini Varadhan list = list->next; 60586fa01ccdSSowmini Varadhan insp = list; 60596fa01ccdSSowmini Varadhan } else { 60606fa01ccdSSowmini Varadhan /* Eaten partially. */ 60616fa01ccdSSowmini Varadhan if (skb_shared(list)) { 60626fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 60636fa01ccdSSowmini Varadhan if (!clone) 60646fa01ccdSSowmini Varadhan return -ENOMEM; 60656fa01ccdSSowmini Varadhan insp = list->next; 60666fa01ccdSSowmini Varadhan list = clone; 60676fa01ccdSSowmini Varadhan } else { 60686fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 60696fa01ccdSSowmini Varadhan insp = list; 60706fa01ccdSSowmini Varadhan } 60716fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 60726fa01ccdSSowmini Varadhan kfree_skb(clone); 60736fa01ccdSSowmini Varadhan return -ENOMEM; 60746fa01ccdSSowmini Varadhan } 60756fa01ccdSSowmini Varadhan break; 60766fa01ccdSSowmini Varadhan } 60776fa01ccdSSowmini Varadhan } while (eat); 60786fa01ccdSSowmini Varadhan 60796fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 60806fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 60816fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 60826fa01ccdSSowmini Varadhan kfree_skb(list); 60836fa01ccdSSowmini Varadhan } 60846fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 60856fa01ccdSSowmini Varadhan if (clone) { 60866fa01ccdSSowmini Varadhan clone->next = list; 60876fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 60886fa01ccdSSowmini Varadhan } 60896fa01ccdSSowmini Varadhan return 0; 60906fa01ccdSSowmini Varadhan } 60916fa01ccdSSowmini Varadhan 60926fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 60936fa01ccdSSowmini Varadhan * non-linear part of skb 60946fa01ccdSSowmini Varadhan */ 60956fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 60966fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 60976fa01ccdSSowmini Varadhan { 60986fa01ccdSSowmini Varadhan int i, k = 0; 60996fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 61006fa01ccdSSowmini Varadhan u8 *data; 61016fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 61026fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 61036fa01ccdSSowmini Varadhan 61046fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 61056fa01ccdSSowmini Varadhan 61066fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 61076fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 61086fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 61096fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 61106fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 61116fa01ccdSSowmini Varadhan if (!data) 61126fa01ccdSSowmini Varadhan return -ENOMEM; 61136fa01ccdSSowmini Varadhan 61146fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 61156fa01ccdSSowmini Varadhan 61166fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6117e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 61186fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 61196fa01ccdSSowmini Varadhan kfree(data); 61206fa01ccdSSowmini Varadhan return -ENOMEM; 61216fa01ccdSSowmini Varadhan } 61226fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 61236fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 61246fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 61256fa01ccdSSowmini Varadhan 61266fa01ccdSSowmini Varadhan if (pos + fsize > off) { 61276fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 61286fa01ccdSSowmini Varadhan 61296fa01ccdSSowmini Varadhan if (pos < off) { 61306fa01ccdSSowmini Varadhan /* Split frag. 61316fa01ccdSSowmini Varadhan * We have two variants in this case: 61326fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 61336fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 61346fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 61356fa01ccdSSowmini Varadhan * where splitting is expensive. 61366fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 61376fa01ccdSSowmini Varadhan */ 6138b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 61396fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 61406fa01ccdSSowmini Varadhan } 61416fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 61426fa01ccdSSowmini Varadhan k++; 61436fa01ccdSSowmini Varadhan } 61446fa01ccdSSowmini Varadhan pos += fsize; 61456fa01ccdSSowmini Varadhan } 61466fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 61476fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 61486fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 61496fa01ccdSSowmini Varadhan 61506fa01ccdSSowmini Varadhan /* split line is in frag list */ 6151eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6152eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6153eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6154eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6155eabe8618SMiaohe Lin kfree(data); 6156eabe8618SMiaohe Lin return -ENOMEM; 61576fa01ccdSSowmini Varadhan } 61586fa01ccdSSowmini Varadhan skb_release_data(skb); 61596fa01ccdSSowmini Varadhan 61606fa01ccdSSowmini Varadhan skb->head = data; 61616fa01ccdSSowmini Varadhan skb->head_frag = 0; 61626fa01ccdSSowmini Varadhan skb->data = data; 61636fa01ccdSSowmini Varadhan #ifdef NET_SKBUFF_DATA_USES_OFFSET 61646fa01ccdSSowmini Varadhan skb->end = size; 61656fa01ccdSSowmini Varadhan #else 61666fa01ccdSSowmini Varadhan skb->end = skb->head + size; 61676fa01ccdSSowmini Varadhan #endif 61686fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 61696fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 61706fa01ccdSSowmini Varadhan skb->cloned = 0; 61716fa01ccdSSowmini Varadhan skb->hdr_len = 0; 61726fa01ccdSSowmini Varadhan skb->nohdr = 0; 61736fa01ccdSSowmini Varadhan skb->len -= off; 61746fa01ccdSSowmini Varadhan skb->data_len = skb->len; 61756fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 61766fa01ccdSSowmini Varadhan return 0; 61776fa01ccdSSowmini Varadhan } 61786fa01ccdSSowmini Varadhan 61796fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 61806fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 61816fa01ccdSSowmini Varadhan { 61826fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 61836fa01ccdSSowmini Varadhan 61846fa01ccdSSowmini Varadhan if (len < headlen) 61856fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 61866fa01ccdSSowmini Varadhan else 61876fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 61886fa01ccdSSowmini Varadhan } 61896fa01ccdSSowmini Varadhan 61906fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 61916fa01ccdSSowmini Varadhan * a new skb 61926fa01ccdSSowmini Varadhan */ 61936fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 61946fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 61956fa01ccdSSowmini Varadhan { 61966fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 61976fa01ccdSSowmini Varadhan 61986fa01ccdSSowmini Varadhan if (!clone) 61996fa01ccdSSowmini Varadhan return NULL; 62006fa01ccdSSowmini Varadhan 62016fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 62026fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 62036fa01ccdSSowmini Varadhan kfree_skb(clone); 62046fa01ccdSSowmini Varadhan return NULL; 62056fa01ccdSSowmini Varadhan } 62066fa01ccdSSowmini Varadhan return clone; 62076fa01ccdSSowmini Varadhan } 62086fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6209c8c8b127SEric Dumazet 6210c8c8b127SEric Dumazet /** 6211c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6212c8c8b127SEric Dumazet * @skb: buffer 6213c8c8b127SEric Dumazet * 6214c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6215c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6216c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6217c8c8b127SEric Dumazet * truesize. 6218c8c8b127SEric Dumazet * Notes: 6219c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6220c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6221c8c8b127SEric Dumazet */ 6222c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6223c8c8b127SEric Dumazet { 62243174fed9SEric Dumazet if (skb->data_len) { 62253174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6226c8c8b127SEric Dumazet skb_cloned(skb)) 6227c8c8b127SEric Dumazet return; 6228c8c8b127SEric Dumazet 6229c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6230c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 62313174fed9SEric Dumazet } 62323174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 62333174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 62343174fed9SEric Dumazet * their truesize. 62353174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 62363174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 62373174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6238c8c8b127SEric Dumazet */ 6239c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6240c8c8b127SEric Dumazet } 6241df5042f4SFlorian Westphal 6242df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6243df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6244df5042f4SFlorian Westphal { 6245df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6246df5042f4SFlorian Westphal } 6247df5042f4SFlorian Westphal 62488b69a803SPaolo Abeni /** 62498b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 62508b69a803SPaolo Abeni * 62514930f483SFlorian Westphal * @flags: See kmalloc(). 62524930f483SFlorian Westphal * 62538b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 62548b69a803SPaolo Abeni * skb via __skb_ext_set(). 62558b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 62568b69a803SPaolo Abeni */ 62574930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6258df5042f4SFlorian Westphal { 62594930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6260df5042f4SFlorian Westphal 6261df5042f4SFlorian Westphal if (new) { 6262df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6263df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6264df5042f4SFlorian Westphal } 6265df5042f4SFlorian Westphal 6266df5042f4SFlorian Westphal return new; 6267df5042f4SFlorian Westphal } 6268df5042f4SFlorian Westphal 62694165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 62704165079bSFlorian Westphal unsigned int old_active) 6271df5042f4SFlorian Westphal { 6272df5042f4SFlorian Westphal struct skb_ext *new; 6273df5042f4SFlorian Westphal 6274df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6275df5042f4SFlorian Westphal return old; 6276df5042f4SFlorian Westphal 6277df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6278df5042f4SFlorian Westphal if (!new) 6279df5042f4SFlorian Westphal return NULL; 6280df5042f4SFlorian Westphal 6281df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6282df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6283df5042f4SFlorian Westphal 62844165079bSFlorian Westphal #ifdef CONFIG_XFRM 62854165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 62864165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 62874165079bSFlorian Westphal unsigned int i; 62884165079bSFlorian Westphal 62894165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 62904165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 62914165079bSFlorian Westphal } 62924165079bSFlorian Westphal #endif 6293df5042f4SFlorian Westphal __skb_ext_put(old); 6294df5042f4SFlorian Westphal return new; 6295df5042f4SFlorian Westphal } 6296df5042f4SFlorian Westphal 6297df5042f4SFlorian Westphal /** 62988b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 62998b69a803SPaolo Abeni * @skb: buffer 63008b69a803SPaolo Abeni * @id: extension id 63018b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 63028b69a803SPaolo Abeni * 63038b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 63048b69a803SPaolo Abeni * 63058b69a803SPaolo Abeni * Returns the pointer to the extension. 63068b69a803SPaolo Abeni */ 63078b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 63088b69a803SPaolo Abeni struct skb_ext *ext) 63098b69a803SPaolo Abeni { 63108b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 63118b69a803SPaolo Abeni 63128b69a803SPaolo Abeni skb_ext_put(skb); 63138b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 63148b69a803SPaolo Abeni ext->chunks = newlen; 63158b69a803SPaolo Abeni ext->offset[id] = newoff; 63168b69a803SPaolo Abeni skb->extensions = ext; 63178b69a803SPaolo Abeni skb->active_extensions = 1 << id; 63188b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 63198b69a803SPaolo Abeni } 63208b69a803SPaolo Abeni 63218b69a803SPaolo Abeni /** 6322df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6323df5042f4SFlorian Westphal * @skb: buffer 6324df5042f4SFlorian Westphal * @id: extension to allocate space for 6325df5042f4SFlorian Westphal * 6326df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6327df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6328df5042f4SFlorian Westphal * is returned. 6329df5042f4SFlorian Westphal * 6330df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6331df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6332df5042f4SFlorian Westphal * 6333df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6334df5042f4SFlorian Westphal */ 6335df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6336df5042f4SFlorian Westphal { 6337df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6338df5042f4SFlorian Westphal unsigned int newlen, newoff; 6339df5042f4SFlorian Westphal 6340df5042f4SFlorian Westphal if (skb->active_extensions) { 6341df5042f4SFlorian Westphal old = skb->extensions; 6342df5042f4SFlorian Westphal 63434165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6344df5042f4SFlorian Westphal if (!new) 6345df5042f4SFlorian Westphal return NULL; 6346df5042f4SFlorian Westphal 6347682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6348df5042f4SFlorian Westphal goto set_active; 6349df5042f4SFlorian Westphal 6350e94e50bdSPaolo Abeni newoff = new->chunks; 6351df5042f4SFlorian Westphal } else { 6352df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6353df5042f4SFlorian Westphal 63544930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6355df5042f4SFlorian Westphal if (!new) 6356df5042f4SFlorian Westphal return NULL; 6357df5042f4SFlorian Westphal } 6358df5042f4SFlorian Westphal 6359df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6360df5042f4SFlorian Westphal new->chunks = newlen; 6361df5042f4SFlorian Westphal new->offset[id] = newoff; 6362df5042f4SFlorian Westphal set_active: 6363b0999f38SPaolo Abeni skb->slow_gro = 1; 6364682ec859SPaolo Abeni skb->extensions = new; 6365df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6366df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6367df5042f4SFlorian Westphal } 6368df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6369df5042f4SFlorian Westphal 63704165079bSFlorian Westphal #ifdef CONFIG_XFRM 63714165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 63724165079bSFlorian Westphal { 63734165079bSFlorian Westphal unsigned int i; 63744165079bSFlorian Westphal 63754165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 63764165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 63774165079bSFlorian Westphal } 63784165079bSFlorian Westphal #endif 63794165079bSFlorian Westphal 638078476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 638178476d31SJeremy Kerr static void skb_ext_put_mctp(struct mctp_flow *flow) 638278476d31SJeremy Kerr { 638378476d31SJeremy Kerr if (flow->key) 638478476d31SJeremy Kerr mctp_key_unref(flow->key); 638578476d31SJeremy Kerr } 638678476d31SJeremy Kerr #endif 638778476d31SJeremy Kerr 6388df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6389df5042f4SFlorian Westphal { 6390df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6391df5042f4SFlorian Westphal 6392df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6393df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6394df5042f4SFlorian Westphal skb->extensions = NULL; 6395df5042f4SFlorian Westphal __skb_ext_put(ext); 63964165079bSFlorian Westphal #ifdef CONFIG_XFRM 63974165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 63984165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 63994165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 64004165079bSFlorian Westphal 64014165079bSFlorian Westphal skb_ext_put_sp(sp); 64024165079bSFlorian Westphal sp->len = 0; 64034165079bSFlorian Westphal #endif 6404df5042f4SFlorian Westphal } 6405df5042f4SFlorian Westphal } 6406df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6407df5042f4SFlorian Westphal 6408df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6409df5042f4SFlorian Westphal { 6410df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6411df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6412df5042f4SFlorian Westphal */ 6413df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6414df5042f4SFlorian Westphal goto free_now; 6415df5042f4SFlorian Westphal 6416df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6417df5042f4SFlorian Westphal return; 6418df5042f4SFlorian Westphal free_now: 64194165079bSFlorian Westphal #ifdef CONFIG_XFRM 64204165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 64214165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 64224165079bSFlorian Westphal #endif 642378476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 642478476d31SJeremy Kerr if (__skb_ext_exist(ext, SKB_EXT_MCTP)) 642578476d31SJeremy Kerr skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP)); 642678476d31SJeremy Kerr #endif 64274165079bSFlorian Westphal 6428df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6429df5042f4SFlorian Westphal } 6430df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6431df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 6432