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 8339564c3fSEric Dumazet #include "dev.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); 204763087daSEric Dumazet skb_set_end_offset(skb, size); 205ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 206ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 20768822bdfSEric Dumazet skb->alloc_cpu = raw_smp_processor_id(); 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 669*753f1ca4SPavel Begunkov if (skb_zcopy(skb)) { 670*753f1ca4SPavel Begunkov bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS; 671*753f1ca4SPavel Begunkov 67270c43167SJonathan Lemon skb_zcopy_clear(skb, true); 673*753f1ca4SPavel Begunkov if (skip_unref) 674*753f1ca4SPavel Begunkov goto free_head; 675*753f1ca4SPavel Begunkov } 67670c43167SJonathan Lemon 677ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 6786a5bcd84SIlias Apalodimas __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); 6791da177e4SLinus Torvalds 680*753f1ca4SPavel Begunkov free_head: 681ff04a771SEric Dumazet if (shinfo->frag_list) 682ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6831da177e4SLinus Torvalds 684d3836f21SEric Dumazet skb_free_head(skb); 6852cc3aeb5SIlias Apalodimas exit: 6862cc3aeb5SIlias Apalodimas /* When we clone an SKB we copy the reycling bit. The pp_recycle 6872cc3aeb5SIlias Apalodimas * bit is only set on the head though, so in order to avoid races 6882cc3aeb5SIlias Apalodimas * while trying to recycle fragments on __skb_frag_unref() we need 6892cc3aeb5SIlias Apalodimas * to make one SKB responsible for triggering the recycle path. 6902cc3aeb5SIlias Apalodimas * So disable the recycling bit if an SKB is cloned and we have 69158e61e41STom Rix * additional references to the fragmented part of the SKB. 6922cc3aeb5SIlias Apalodimas * Eventually the last SKB will have the recycling bit set and it's 6932cc3aeb5SIlias Apalodimas * dataref set to 0, which will trigger the recycling 6942cc3aeb5SIlias Apalodimas */ 6952cc3aeb5SIlias Apalodimas skb->pp_recycle = 0; 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds /* 6991da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 7001da177e4SLinus Torvalds */ 7012d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 7021da177e4SLinus Torvalds { 703d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 704d179cd12SDavid S. Miller 705d179cd12SDavid S. Miller switch (skb->fclone) { 706d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 7071da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 7086ffe75ebSEric Dumazet return; 709d179cd12SDavid S. Miller 710d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 711d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 7126ffe75ebSEric Dumazet 7136ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 7146ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 7156ffe75ebSEric Dumazet * while here, branch prediction will be good. 7166ffe75ebSEric Dumazet */ 7172638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 7186ffe75ebSEric Dumazet goto fastpath; 719d179cd12SDavid S. Miller break; 720d179cd12SDavid S. Miller 7216ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 722d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 723d179cd12SDavid S. Miller break; 7243ff50b79SStephen Hemminger } 7252638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 7266ffe75ebSEric Dumazet return; 7276ffe75ebSEric Dumazet fastpath: 7286ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 7291da177e4SLinus Torvalds } 7301da177e4SLinus Torvalds 7310a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 7321da177e4SLinus Torvalds { 733adf30907SEric Dumazet skb_dst_drop(skb); 7341da177e4SLinus Torvalds if (skb->destructor) { 735afa79d08SChangbin Du WARN_ON(in_hardirq()); 7361da177e4SLinus Torvalds skb->destructor(skb); 7371da177e4SLinus Torvalds } 738a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 739cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 7402fc72c7bSKOVACS Krisztian #endif 741df5042f4SFlorian Westphal skb_ext_put(skb); 74204a4bb55SLennert Buytenhek } 74304a4bb55SLennert Buytenhek 74404a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 74504a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 74604a4bb55SLennert Buytenhek { 74704a4bb55SLennert Buytenhek skb_release_head_state(skb); 748a28b1b90SFlorian Westphal if (likely(skb->head)) 7492d4baff8SHerbert Xu skb_release_data(skb); 7502d4baff8SHerbert Xu } 7511da177e4SLinus Torvalds 7522d4baff8SHerbert Xu /** 7532d4baff8SHerbert Xu * __kfree_skb - private function 7542d4baff8SHerbert Xu * @skb: buffer 7552d4baff8SHerbert Xu * 7562d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 7572d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 7582d4baff8SHerbert Xu * always call kfree_skb 7592d4baff8SHerbert Xu */ 7602d4baff8SHerbert Xu 7612d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 7622d4baff8SHerbert Xu { 7632d4baff8SHerbert Xu skb_release_all(skb); 7641da177e4SLinus Torvalds kfree_skbmem(skb); 7651da177e4SLinus Torvalds } 766b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 7671da177e4SLinus Torvalds 7681da177e4SLinus Torvalds /** 769c504e5c2SMenglong Dong * kfree_skb_reason - free an sk_buff with special reason 770231d06aeSJörn Engel * @skb: buffer to free 771c504e5c2SMenglong Dong * @reason: reason why this skb is dropped 772231d06aeSJörn Engel * 773231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 774c504e5c2SMenglong Dong * hit zero. Meanwhile, pass the drop reason to 'kfree_skb' 775c504e5c2SMenglong Dong * tracepoint. 776231d06aeSJörn Engel */ 777c504e5c2SMenglong Dong void kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 778231d06aeSJörn Engel { 7793889a803SPaolo Abeni if (!skb_unref(skb)) 780231d06aeSJörn Engel return; 7813889a803SPaolo Abeni 78220bbcd0aSMenglong Dong DEBUG_NET_WARN_ON_ONCE(reason <= 0 || reason >= SKB_DROP_REASON_MAX); 78320bbcd0aSMenglong Dong 784c504e5c2SMenglong Dong trace_kfree_skb(skb, __builtin_return_address(0), reason); 785231d06aeSJörn Engel __kfree_skb(skb); 786231d06aeSJörn Engel } 787c504e5c2SMenglong Dong EXPORT_SYMBOL(kfree_skb_reason); 788231d06aeSJörn Engel 789215b0f19SMenglong Dong void kfree_skb_list_reason(struct sk_buff *segs, 790215b0f19SMenglong Dong enum skb_drop_reason reason) 791bd8a7036SEric Dumazet { 792bd8a7036SEric Dumazet while (segs) { 793bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 794bd8a7036SEric Dumazet 795215b0f19SMenglong Dong kfree_skb_reason(segs, reason); 796bd8a7036SEric Dumazet segs = next; 797bd8a7036SEric Dumazet } 798bd8a7036SEric Dumazet } 799215b0f19SMenglong Dong EXPORT_SYMBOL(kfree_skb_list_reason); 800bd8a7036SEric Dumazet 8016413139dSWillem de Bruijn /* Dump skb information and contents. 8026413139dSWillem de Bruijn * 8036413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 8046413139dSWillem de Bruijn * 805302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 8066413139dSWillem de Bruijn */ 8076413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 8086413139dSWillem de Bruijn { 8096413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 8106413139dSWillem de Bruijn struct net_device *dev = skb->dev; 8116413139dSWillem de Bruijn struct sock *sk = skb->sk; 8126413139dSWillem de Bruijn struct sk_buff *list_skb; 8136413139dSWillem de Bruijn bool has_mac, has_trans; 8146413139dSWillem de Bruijn int headroom, tailroom; 8156413139dSWillem de Bruijn int i, len, seg_len; 8166413139dSWillem de Bruijn 8176413139dSWillem de Bruijn if (full_pkt) 8186413139dSWillem de Bruijn len = skb->len; 8196413139dSWillem de Bruijn else 8206413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 8216413139dSWillem de Bruijn 8226413139dSWillem de Bruijn headroom = skb_headroom(skb); 8236413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 8246413139dSWillem de Bruijn 8256413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 8266413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 8276413139dSWillem de Bruijn 8286413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 8296413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 8306413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 8316413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 8326413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 8336413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 8346413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 8356413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 8366413139dSWillem de Bruijn skb->network_header, 8376413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 8386413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 8396413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 8406413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 8416413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 8426413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 8436413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 8446413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 8456413139dSWillem de Bruijn 8466413139dSWillem de Bruijn if (dev) 8478a03ef67SGal Pressman printk("%sdev name=%s feat=%pNF\n", 8486413139dSWillem de Bruijn level, dev->name, &dev->features); 8496413139dSWillem de Bruijn if (sk) 850db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 8516413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 8526413139dSWillem de Bruijn 8536413139dSWillem de Bruijn if (full_pkt && headroom) 8546413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 8556413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 8566413139dSWillem de Bruijn 8576413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 8586413139dSWillem de Bruijn if (seg_len) 8596413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 8606413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 8616413139dSWillem de Bruijn len -= seg_len; 8626413139dSWillem de Bruijn 8636413139dSWillem de Bruijn if (full_pkt && tailroom) 8646413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 8656413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 8666413139dSWillem de Bruijn 8676413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 8686413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 8696413139dSWillem de Bruijn u32 p_off, p_len, copied; 8706413139dSWillem de Bruijn struct page *p; 8716413139dSWillem de Bruijn u8 *vaddr; 8726413139dSWillem de Bruijn 873b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 8746413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 8756413139dSWillem de Bruijn copied) { 8766413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 8776413139dSWillem de Bruijn vaddr = kmap_atomic(p); 8786413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 8796413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 8806413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 8816413139dSWillem de Bruijn kunmap_atomic(vaddr); 8826413139dSWillem de Bruijn len -= seg_len; 8836413139dSWillem de Bruijn if (!len) 8846413139dSWillem de Bruijn break; 8856413139dSWillem de Bruijn } 8866413139dSWillem de Bruijn } 8876413139dSWillem de Bruijn 8886413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8896413139dSWillem de Bruijn printk("skb fraglist:\n"); 8906413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8916413139dSWillem de Bruijn skb_dump(level, list_skb, true); 8926413139dSWillem de Bruijn } 8936413139dSWillem de Bruijn } 8946413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 8956413139dSWillem de Bruijn 896d1a203eaSStephen Hemminger /** 89725121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 89825121173SMichael S. Tsirkin * @skb: buffer that triggered an error 89925121173SMichael S. Tsirkin * 90025121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 90125121173SMichael S. Tsirkin * skb must be freed afterwards. 90225121173SMichael S. Tsirkin */ 90325121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 90425121173SMichael S. Tsirkin { 905*753f1ca4SPavel Begunkov if (skb) { 906*753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 9071f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 90825121173SMichael S. Tsirkin } 909*753f1ca4SPavel Begunkov } 91025121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 91125121173SMichael S. Tsirkin 912be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 91325121173SMichael S. Tsirkin /** 914ead2ceb0SNeil Horman * consume_skb - free an skbuff 915ead2ceb0SNeil Horman * @skb: buffer to free 916ead2ceb0SNeil Horman * 917ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 918ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 919ead2ceb0SNeil Horman * is being dropped after a failure and notes that 920ead2ceb0SNeil Horman */ 921ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 922ead2ceb0SNeil Horman { 9233889a803SPaolo Abeni if (!skb_unref(skb)) 924ead2ceb0SNeil Horman return; 9253889a803SPaolo Abeni 92607dc22e7SKoki Sanagi trace_consume_skb(skb); 927ead2ceb0SNeil Horman __kfree_skb(skb); 928ead2ceb0SNeil Horman } 929ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 930be769db2SHerbert Xu #endif 931ead2ceb0SNeil Horman 9320a463c78SPaolo Abeni /** 933c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 9340a463c78SPaolo Abeni * @skb: buffer to free 9350a463c78SPaolo Abeni * 936ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 937ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 9380a463c78SPaolo Abeni */ 939ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 9400a463c78SPaolo Abeni { 9410a463c78SPaolo Abeni trace_consume_skb(skb); 9420a463c78SPaolo Abeni skb_release_data(skb); 9430a463c78SPaolo Abeni kfree_skbmem(skb); 9440a463c78SPaolo Abeni } 9450a463c78SPaolo Abeni 946f450d539SAlexander Lobakin static void napi_skb_cache_put(struct sk_buff *skb) 947795bb1c0SJesper Dangaard Brouer { 948795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 949f450d539SAlexander Lobakin u32 i; 950795bb1c0SJesper Dangaard Brouer 951f450d539SAlexander Lobakin kasan_poison_object_data(skbuff_head_cache, skb); 952795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 953795bb1c0SJesper Dangaard Brouer 954795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 955f450d539SAlexander Lobakin for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++) 956f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, 957f450d539SAlexander Lobakin nc->skb_cache[i]); 958f450d539SAlexander Lobakin 959f450d539SAlexander Lobakin kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF, 960f450d539SAlexander Lobakin nc->skb_cache + NAPI_SKB_CACHE_HALF); 961f450d539SAlexander Lobakin nc->skb_count = NAPI_SKB_CACHE_HALF; 962795bb1c0SJesper Dangaard Brouer } 963795bb1c0SJesper Dangaard Brouer } 964f450d539SAlexander Lobakin 96515fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 96615fad714SJesper Dangaard Brouer { 9679243adfcSAlexander Lobakin skb_release_all(skb); 9689243adfcSAlexander Lobakin napi_skb_cache_put(skb); 9699243adfcSAlexander Lobakin } 9709243adfcSAlexander Lobakin 9719243adfcSAlexander Lobakin void napi_skb_free_stolen_head(struct sk_buff *skb) 9729243adfcSAlexander Lobakin { 9739efb4b5bSPaolo Abeni if (unlikely(skb->slow_gro)) { 9748550ff8dSPaul Blakey nf_reset_ct(skb); 9759243adfcSAlexander Lobakin skb_dst_drop(skb); 9769243adfcSAlexander Lobakin skb_ext_put(skb); 9775e10da53SPaolo Abeni skb_orphan(skb); 9789efb4b5bSPaolo Abeni skb->slow_gro = 0; 9799efb4b5bSPaolo Abeni } 980f450d539SAlexander Lobakin napi_skb_cache_put(skb); 98115fad714SJesper Dangaard Brouer } 982795bb1c0SJesper Dangaard Brouer 983795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 984795bb1c0SJesper Dangaard Brouer { 985885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 986795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 987885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 988795bb1c0SJesper Dangaard Brouer return; 989795bb1c0SJesper Dangaard Brouer } 990795bb1c0SJesper Dangaard Brouer 9916454eca8SYunsheng Lin lockdep_assert_in_softirq(); 9926454eca8SYunsheng Lin 9937608894eSPaolo Abeni if (!skb_unref(skb)) 994795bb1c0SJesper Dangaard Brouer return; 9957608894eSPaolo Abeni 996795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 997795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 998795bb1c0SJesper Dangaard Brouer 999795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 1000abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 1001795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 1002795bb1c0SJesper Dangaard Brouer return; 1003795bb1c0SJesper Dangaard Brouer } 1004795bb1c0SJesper Dangaard Brouer 10059243adfcSAlexander Lobakin skb_release_all(skb); 1006f450d539SAlexander Lobakin napi_skb_cache_put(skb); 1007795bb1c0SJesper Dangaard Brouer } 1008795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 1009795bb1c0SJesper Dangaard Brouer 101003f61041SKees Cook /* Make sure a field is contained by headers group */ 1011b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 101203f61041SKees Cook BUILD_BUG_ON(offsetof(struct sk_buff, field) != \ 101303f61041SKees Cook offsetof(struct sk_buff, headers.field)); \ 1014b1937227SEric Dumazet 1015dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 1016dec18810SHerbert Xu { 1017dec18810SHerbert Xu new->tstamp = old->tstamp; 1018b1937227SEric Dumazet /* We do not copy old->sk */ 1019dec18810SHerbert Xu new->dev = old->dev; 1020b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 10217fee226aSEric Dumazet skb_dst_copy(new, old); 1022df5042f4SFlorian Westphal __skb_ext_copy(new, old); 1023b1937227SEric Dumazet __nf_copy(new, old, false); 10246aa895b0SPatrick McHardy 102503f61041SKees Cook /* Note : this field could be in the headers group. 1026b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 1027b1937227SEric Dumazet */ 1028b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 102906021292SEliezer Tamir 103003f61041SKees Cook memcpy(&new->headers, &old->headers, sizeof(new->headers)); 1031b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 1032b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 1033b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 1034b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 1035b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 1036b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 1037b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 1038b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 1039b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 1040b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 1041b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 1042b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 1043b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 1044b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 1045b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 1046b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 1047b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 1048b1937227SEric Dumazet #endif 1049e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 1050b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 105106021292SEliezer Tamir #endif 105268822bdfSEric Dumazet CHECK_SKB_FIELD(alloc_cpu); 10532bd82484SEric Dumazet #ifdef CONFIG_XPS 10542bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 10552bd82484SEric Dumazet #endif 1056b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 1057b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 1058b1937227SEric Dumazet #endif 1059b1937227SEric Dumazet 1060dec18810SHerbert Xu } 1061dec18810SHerbert Xu 106282c49a35SHerbert Xu /* 106382c49a35SHerbert Xu * You should not add any new code to this function. Add it to 106482c49a35SHerbert Xu * __copy_skb_header above instead. 106582c49a35SHerbert Xu */ 1066e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 10671da177e4SLinus Torvalds { 10681da177e4SLinus Torvalds #define C(x) n->x = skb->x 10691da177e4SLinus Torvalds 10701da177e4SLinus Torvalds n->next = n->prev = NULL; 10711da177e4SLinus Torvalds n->sk = NULL; 1072dec18810SHerbert Xu __copy_skb_header(n, skb); 1073dec18810SHerbert Xu 10741da177e4SLinus Torvalds C(len); 10751da177e4SLinus Torvalds C(data_len); 10763e6b3b2eSAlexey Dobriyan C(mac_len); 1077334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 107802f1c89dSPaul Moore n->cloned = 1; 10791da177e4SLinus Torvalds n->nohdr = 0; 1080b13dda9fSEric Dumazet n->peeked = 0; 1081e78bfb07SStefano Brivio C(pfmemalloc); 10826a5bcd84SIlias Apalodimas C(pp_recycle); 10831da177e4SLinus Torvalds n->destructor = NULL; 10841da177e4SLinus Torvalds C(tail); 10851da177e4SLinus Torvalds C(end); 108602f1c89dSPaul Moore C(head); 1087d3836f21SEric Dumazet C(head_frag); 108802f1c89dSPaul Moore C(data); 108902f1c89dSPaul Moore C(truesize); 109063354797SReshetova, Elena refcount_set(&n->users, 1); 10911da177e4SLinus Torvalds 10921da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 10931da177e4SLinus Torvalds skb->cloned = 1; 10941da177e4SLinus Torvalds 10951da177e4SLinus Torvalds return n; 1096e0053ec0SHerbert Xu #undef C 1097e0053ec0SHerbert Xu } 1098e0053ec0SHerbert Xu 1099e0053ec0SHerbert Xu /** 1100da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1101da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1102da29e4b4SJakub Kicinski */ 1103da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1104da29e4b4SJakub Kicinski { 1105da29e4b4SJakub Kicinski struct sk_buff *n; 1106da29e4b4SJakub Kicinski 1107da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1108da29e4b4SJakub Kicinski if (!n) 1109da29e4b4SJakub Kicinski return NULL; 1110da29e4b4SJakub Kicinski 1111da29e4b4SJakub Kicinski n->len = first->len; 1112da29e4b4SJakub Kicinski n->data_len = first->len; 1113da29e4b4SJakub Kicinski n->truesize = first->truesize; 1114da29e4b4SJakub Kicinski 1115da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1116da29e4b4SJakub Kicinski 1117da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1118da29e4b4SJakub Kicinski n->destructor = NULL; 1119da29e4b4SJakub Kicinski 1120da29e4b4SJakub Kicinski return n; 1121da29e4b4SJakub Kicinski } 1122da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1123da29e4b4SJakub Kicinski 1124da29e4b4SJakub Kicinski /** 1125e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1126e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1127e0053ec0SHerbert Xu * @src: the skb to supply the contents 1128e0053ec0SHerbert Xu * 1129e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1130e0053ec0SHerbert Xu * supplied by the user. 1131e0053ec0SHerbert Xu * 1132e0053ec0SHerbert Xu * The target skb is returned upon exit. 1133e0053ec0SHerbert Xu */ 1134e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1135e0053ec0SHerbert Xu { 11362d4baff8SHerbert Xu skb_release_all(dst); 1137e0053ec0SHerbert Xu return __skb_clone(dst, src); 1138e0053ec0SHerbert Xu } 1139e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1140e0053ec0SHerbert Xu 11416f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1142a91dbff5SWillem de Bruijn { 1143a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1144a91dbff5SWillem de Bruijn struct user_struct *user; 1145a91dbff5SWillem de Bruijn 1146a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1147a91dbff5SWillem de Bruijn return 0; 1148a91dbff5SWillem de Bruijn 1149a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1150a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1151a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1152a91dbff5SWillem de Bruijn 1153a91dbff5SWillem de Bruijn do { 1154a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1155a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1156a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1157a91dbff5SWillem de Bruijn return -ENOBUFS; 1158a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1159a91dbff5SWillem de Bruijn old_pg); 1160a91dbff5SWillem de Bruijn 1161a91dbff5SWillem de Bruijn if (!mmp->user) { 1162a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1163a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1164a91dbff5SWillem de Bruijn } else { 1165a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1166a91dbff5SWillem de Bruijn } 1167a91dbff5SWillem de Bruijn 1168a91dbff5SWillem de Bruijn return 0; 1169a91dbff5SWillem de Bruijn } 11706f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1171a91dbff5SWillem de Bruijn 11726f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1173a91dbff5SWillem de Bruijn { 1174a91dbff5SWillem de Bruijn if (mmp->user) { 1175a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1176a91dbff5SWillem de Bruijn free_uid(mmp->user); 1177a91dbff5SWillem de Bruijn } 1178a91dbff5SWillem de Bruijn } 11796f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1180a91dbff5SWillem de Bruijn 1181c67b627eSDavid Ahern static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size) 118252267790SWillem de Bruijn { 118352267790SWillem de Bruijn struct ubuf_info *uarg; 118452267790SWillem de Bruijn struct sk_buff *skb; 118552267790SWillem de Bruijn 118652267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 118752267790SWillem de Bruijn 118852267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 118952267790SWillem de Bruijn if (!skb) 119052267790SWillem de Bruijn return NULL; 119152267790SWillem de Bruijn 119252267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 119352267790SWillem de Bruijn uarg = (void *)skb->cb; 1194a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1195a91dbff5SWillem de Bruijn 1196a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1197a91dbff5SWillem de Bruijn kfree_skb(skb); 1198a91dbff5SWillem de Bruijn return NULL; 1199a91dbff5SWillem de Bruijn } 120052267790SWillem de Bruijn 12018c793822SJonathan Lemon uarg->callback = msg_zerocopy_callback; 12024ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 12034ab6c99dSWillem de Bruijn uarg->len = 1; 12044ab6c99dSWillem de Bruijn uarg->bytelen = size; 120552267790SWillem de Bruijn uarg->zerocopy = 1; 12062e07a521SPavel Begunkov uarg->flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN; 1207c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 120852267790SWillem de Bruijn sock_hold(sk); 120952267790SWillem de Bruijn 121052267790SWillem de Bruijn return uarg; 121152267790SWillem de Bruijn } 121252267790SWillem de Bruijn 121352267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 121452267790SWillem de Bruijn { 121552267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 121652267790SWillem de Bruijn } 121752267790SWillem de Bruijn 12188c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 12194ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 12204ab6c99dSWillem de Bruijn { 12214ab6c99dSWillem de Bruijn if (uarg) { 12224ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 12234ab6c99dSWillem de Bruijn u32 bytelen, next; 12244ab6c99dSWillem de Bruijn 12251b4b2b09SPavel Begunkov /* there might be non MSG_ZEROCOPY users */ 12261b4b2b09SPavel Begunkov if (uarg->callback != msg_zerocopy_callback) 12271b4b2b09SPavel Begunkov return NULL; 12281b4b2b09SPavel Begunkov 12294ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 12304ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 12314ab6c99dSWillem de Bruijn */ 12324ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 12334ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 12344ab6c99dSWillem de Bruijn return NULL; 12354ab6c99dSWillem de Bruijn } 12364ab6c99dSWillem de Bruijn 12374ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 12384ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 12394ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 12404ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12414ab6c99dSWillem de Bruijn goto new_alloc; 12424ab6c99dSWillem de Bruijn return NULL; 12434ab6c99dSWillem de Bruijn } 12444ab6c99dSWillem de Bruijn 12454ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 12464ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1247a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1248a91dbff5SWillem de Bruijn return NULL; 12494ab6c99dSWillem de Bruijn uarg->len++; 12504ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 12514ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1252100f6d8eSWillem de Bruijn 1253100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1254100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12558e044917SJonathan Lemon net_zcopy_get(uarg); 1256100f6d8eSWillem de Bruijn 12574ab6c99dSWillem de Bruijn return uarg; 12584ab6c99dSWillem de Bruijn } 12594ab6c99dSWillem de Bruijn } 12604ab6c99dSWillem de Bruijn 12614ab6c99dSWillem de Bruijn new_alloc: 12628c793822SJonathan Lemon return msg_zerocopy_alloc(sk, size); 12634ab6c99dSWillem de Bruijn } 12648c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_realloc); 12654ab6c99dSWillem de Bruijn 12664ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 12674ab6c99dSWillem de Bruijn { 12684ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 12694ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 12704ab6c99dSWillem de Bruijn u64 sum_len; 12714ab6c99dSWillem de Bruijn 12724ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 12734ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 12744ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 12754ab6c99dSWillem de Bruijn 12764ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 12774ab6c99dSWillem de Bruijn return false; 12784ab6c99dSWillem de Bruijn 12794ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 12804ab6c99dSWillem de Bruijn return false; 12814ab6c99dSWillem de Bruijn 12824ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 12834ab6c99dSWillem de Bruijn return true; 12844ab6c99dSWillem de Bruijn } 12854ab6c99dSWillem de Bruijn 12868c793822SJonathan Lemon static void __msg_zerocopy_callback(struct ubuf_info *uarg) 128752267790SWillem de Bruijn { 12884ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 128952267790SWillem de Bruijn struct sock_exterr_skb *serr; 129052267790SWillem de Bruijn struct sock *sk = skb->sk; 12914ab6c99dSWillem de Bruijn struct sk_buff_head *q; 12924ab6c99dSWillem de Bruijn unsigned long flags; 12933bdd5ee0SWillem de Bruijn bool is_zerocopy; 12944ab6c99dSWillem de Bruijn u32 lo, hi; 12954ab6c99dSWillem de Bruijn u16 len; 129652267790SWillem de Bruijn 1297ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1298ccaffff1SWillem de Bruijn 12994ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 13004ab6c99dSWillem de Bruijn * so do not queue a completion notification 13014ab6c99dSWillem de Bruijn */ 13024ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 130352267790SWillem de Bruijn goto release; 130452267790SWillem de Bruijn 13054ab6c99dSWillem de Bruijn len = uarg->len; 13064ab6c99dSWillem de Bruijn lo = uarg->id; 13074ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 13083bdd5ee0SWillem de Bruijn is_zerocopy = uarg->zerocopy; 13094ab6c99dSWillem de Bruijn 131052267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 131152267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 131252267790SWillem de Bruijn serr->ee.ee_errno = 0; 131352267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 13144ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 13154ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 13163bdd5ee0SWillem de Bruijn if (!is_zerocopy) 131752267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 131852267790SWillem de Bruijn 13194ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 13204ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 13214ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 13224ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 13234ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 13244ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 132552267790SWillem de Bruijn skb = NULL; 13264ab6c99dSWillem de Bruijn } 13274ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 132852267790SWillem de Bruijn 1329e3ae2365SAlexander Aring sk_error_report(sk); 133052267790SWillem de Bruijn 133152267790SWillem de Bruijn release: 133252267790SWillem de Bruijn consume_skb(skb); 133352267790SWillem de Bruijn sock_put(sk); 133452267790SWillem de Bruijn } 133575518851SJonathan Lemon 13368c793822SJonathan Lemon void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 133736177832SJonathan Lemon bool success) 133875518851SJonathan Lemon { 133975518851SJonathan Lemon uarg->zerocopy = uarg->zerocopy & success; 134075518851SJonathan Lemon 134175518851SJonathan Lemon if (refcount_dec_and_test(&uarg->refcnt)) 13428c793822SJonathan Lemon __msg_zerocopy_callback(uarg); 134375518851SJonathan Lemon } 13448c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_callback); 134552267790SWillem de Bruijn 13468c793822SJonathan Lemon void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 134752267790SWillem de Bruijn { 134852267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 134952267790SWillem de Bruijn 135052267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 13514ab6c99dSWillem de Bruijn uarg->len--; 135252267790SWillem de Bruijn 135352900d22SWillem de Bruijn if (have_uref) 13548c793822SJonathan Lemon msg_zerocopy_callback(NULL, uarg, true); 135552267790SWillem de Bruijn } 13568c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort); 135752267790SWillem de Bruijn 135852267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 135952267790SWillem de Bruijn struct msghdr *msg, int len, 136052267790SWillem de Bruijn struct ubuf_info *uarg) 136152267790SWillem de Bruijn { 13624ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 136352267790SWillem de Bruijn int err, orig_len = skb->len; 136452267790SWillem de Bruijn 13654ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 13664ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 13674ab6c99dSWillem de Bruijn */ 13684ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 13694ab6c99dSWillem de Bruijn return -EEXIST; 13704ab6c99dSWillem de Bruijn 1371ebe73a28SDavid Ahern err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len); 137252267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 137354d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 137454d43117SWillem de Bruijn 137552267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 1376829b7bddSPavel Begunkov iov_iter_revert(&msg->msg_iter, skb->len - orig_len); 137754d43117SWillem de Bruijn skb->sk = sk; 137852267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 137954d43117SWillem de Bruijn skb->sk = save_sk; 138052267790SWillem de Bruijn return err; 138152267790SWillem de Bruijn } 138252267790SWillem de Bruijn 138352900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 138452267790SWillem de Bruijn return skb->len - orig_len; 138552267790SWillem de Bruijn } 138652267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 138752267790SWillem de Bruijn 1388*753f1ca4SPavel Begunkov void __skb_zcopy_downgrade_managed(struct sk_buff *skb) 1389*753f1ca4SPavel Begunkov { 1390*753f1ca4SPavel Begunkov int i; 1391*753f1ca4SPavel Begunkov 1392*753f1ca4SPavel Begunkov skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS; 1393*753f1ca4SPavel Begunkov for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1394*753f1ca4SPavel Begunkov skb_frag_ref(skb, i); 1395*753f1ca4SPavel Begunkov } 1396*753f1ca4SPavel Begunkov EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed); 1397*753f1ca4SPavel Begunkov 13981f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 139952267790SWillem de Bruijn gfp_t gfp_mask) 140052267790SWillem de Bruijn { 140152267790SWillem de Bruijn if (skb_zcopy(orig)) { 140252267790SWillem de Bruijn if (skb_zcopy(nskb)) { 140352267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 140452267790SWillem de Bruijn if (!gfp_mask) { 140552267790SWillem de Bruijn WARN_ON_ONCE(1); 140652267790SWillem de Bruijn return -ENOMEM; 140752267790SWillem de Bruijn } 140852267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 140952267790SWillem de Bruijn return 0; 141052267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 141152267790SWillem de Bruijn return -EIO; 141252267790SWillem de Bruijn } 141352900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 141452267790SWillem de Bruijn } 141552267790SWillem de Bruijn return 0; 141652267790SWillem de Bruijn } 141752267790SWillem de Bruijn 14182c53040fSBen Hutchings /** 14192c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 142048c83012SMichael S. Tsirkin * @skb: the skb to modify 142148c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 142248c83012SMichael S. Tsirkin * 142306b4feb3SJonathan Lemon * This must be called on skb with SKBFL_ZEROCOPY_ENABLE. 142448c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 142548c83012SMichael S. Tsirkin * to userspace pages. 142648c83012SMichael S. Tsirkin * 142748c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 142848c83012SMichael S. Tsirkin * %GFP_ATOMIC. 142948c83012SMichael S. Tsirkin * 143048c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 143148c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 143248c83012SMichael S. Tsirkin */ 143348c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1434a6686f2fSShirley Ma { 1435a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1436a6686f2fSShirley Ma struct page *page, *head = NULL; 14373ece7826SWillem de Bruijn int i, new_frags; 14383ece7826SWillem de Bruijn u32 d_off; 1439a6686f2fSShirley Ma 14403ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 14413ece7826SWillem de Bruijn return -EINVAL; 14423ece7826SWillem de Bruijn 1443f72c4ac6SWillem de Bruijn if (!num_frags) 1444f72c4ac6SWillem de Bruijn goto release; 1445f72c4ac6SWillem de Bruijn 14463ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 14473ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 144802756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1449a6686f2fSShirley Ma if (!page) { 1450a6686f2fSShirley Ma while (head) { 145140dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1452a6686f2fSShirley Ma put_page(head); 1453a6686f2fSShirley Ma head = next; 1454a6686f2fSShirley Ma } 1455a6686f2fSShirley Ma return -ENOMEM; 1456a6686f2fSShirley Ma } 14573ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 14583ece7826SWillem de Bruijn head = page; 14593ece7826SWillem de Bruijn } 14603ece7826SWillem de Bruijn 14613ece7826SWillem de Bruijn page = head; 14623ece7826SWillem de Bruijn d_off = 0; 14633ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 14643ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 14653ece7826SWillem de Bruijn u32 p_off, p_len, copied; 14663ece7826SWillem de Bruijn struct page *p; 14673ece7826SWillem de Bruijn u8 *vaddr; 1468c613c209SWillem de Bruijn 1469b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1470c613c209SWillem de Bruijn p, p_off, p_len, copied) { 14713ece7826SWillem de Bruijn u32 copy, done = 0; 1472c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 14733ece7826SWillem de Bruijn 14743ece7826SWillem de Bruijn while (done < p_len) { 14753ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 14763ece7826SWillem de Bruijn d_off = 0; 14773ece7826SWillem de Bruijn page = (struct page *)page_private(page); 14783ece7826SWillem de Bruijn } 14793ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 14803ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 14813ece7826SWillem de Bruijn vaddr + p_off + done, copy); 14823ece7826SWillem de Bruijn done += copy; 14833ece7826SWillem de Bruijn d_off += copy; 14843ece7826SWillem de Bruijn } 148551c56b00SEric Dumazet kunmap_atomic(vaddr); 1486c613c209SWillem de Bruijn } 1487a6686f2fSShirley Ma } 1488a6686f2fSShirley Ma 1489a6686f2fSShirley Ma /* skb frags release userspace buffers */ 149002756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1491a8605c60SIan Campbell skb_frag_unref(skb, i); 1492a6686f2fSShirley Ma 1493a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 14943ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 14953ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 149640dadff2SSunghan Suh head = (struct page *)page_private(head); 1497a6686f2fSShirley Ma } 14983ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 14993ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 150048c83012SMichael S. Tsirkin 1501b90ddd56SWillem de Bruijn release: 15021f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1503a6686f2fSShirley Ma return 0; 1504a6686f2fSShirley Ma } 1505dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1506a6686f2fSShirley Ma 1507e0053ec0SHerbert Xu /** 1508e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1509e0053ec0SHerbert Xu * @skb: buffer to clone 1510e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1511e0053ec0SHerbert Xu * 1512e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1513e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1514e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1515e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1516e0053ec0SHerbert Xu * 1517e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1518e0053ec0SHerbert Xu * %GFP_ATOMIC. 1519e0053ec0SHerbert Xu */ 1520e0053ec0SHerbert Xu 1521e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1522e0053ec0SHerbert Xu { 1523d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1524d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1525d0bf4a9eSEric Dumazet skb1); 15266ffe75ebSEric Dumazet struct sk_buff *n; 1527e0053ec0SHerbert Xu 152870008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1529a6686f2fSShirley Ma return NULL; 1530a6686f2fSShirley Ma 1531e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 15322638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 15336ffe75ebSEric Dumazet n = &fclones->skb2; 15342638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1535e0053ec0SHerbert Xu } else { 1536c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1537c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1538c93bdd0eSMel Gorman 1539e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1540e0053ec0SHerbert Xu if (!n) 1541e0053ec0SHerbert Xu return NULL; 1542fe55f6d5SVegard Nossum 1543e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1544e0053ec0SHerbert Xu } 1545e0053ec0SHerbert Xu 1546e0053ec0SHerbert Xu return __skb_clone(n, skb); 15471da177e4SLinus Torvalds } 1548b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 15491da177e4SLinus Torvalds 1550b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1551f5b17294SPravin B Shelar { 1552030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1553030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1554030737bcSEric Dumazet skb->csum_start += off; 1555f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1556f5b17294SPravin B Shelar skb->transport_header += off; 1557f5b17294SPravin B Shelar skb->network_header += off; 1558f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1559f5b17294SPravin B Shelar skb->mac_header += off; 1560f5b17294SPravin B Shelar skb->inner_transport_header += off; 1561f5b17294SPravin B Shelar skb->inner_network_header += off; 1562aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1563f5b17294SPravin B Shelar } 1564b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1565f5b17294SPravin B Shelar 156608303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 15671da177e4SLinus Torvalds { 1568dec18810SHerbert Xu __copy_skb_header(new, old); 1569dec18810SHerbert Xu 15707967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 15717967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 15727967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 15731da177e4SLinus Torvalds } 157408303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 15751da177e4SLinus Torvalds 1576c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1577c93bdd0eSMel Gorman { 1578c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1579c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1580c93bdd0eSMel Gorman return 0; 1581c93bdd0eSMel Gorman } 1582c93bdd0eSMel Gorman 15831da177e4SLinus Torvalds /** 15841da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 15851da177e4SLinus Torvalds * @skb: buffer to copy 15861da177e4SLinus Torvalds * @gfp_mask: allocation priority 15871da177e4SLinus Torvalds * 15881da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 15891da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 15901da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 15911da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 15921da177e4SLinus Torvalds * 15931da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 15941da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 15951da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 15961da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 15971da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 15981da177e4SLinus Torvalds */ 15991da177e4SLinus Torvalds 1600dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 16011da177e4SLinus Torvalds { 16026602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1603ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1604c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1605c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 16066602cebbSEric Dumazet 16071da177e4SLinus Torvalds if (!n) 16081da177e4SLinus Torvalds return NULL; 16091da177e4SLinus Torvalds 16101da177e4SLinus Torvalds /* Set the data pointer */ 16111da177e4SLinus Torvalds skb_reserve(n, headerlen); 16121da177e4SLinus Torvalds /* Set the tail pointer and length */ 16131da177e4SLinus Torvalds skb_put(n, skb->len); 16141da177e4SLinus Torvalds 16159f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 16161da177e4SLinus Torvalds 161708303c18SIlya Lesokhin skb_copy_header(n, skb); 16181da177e4SLinus Torvalds return n; 16191da177e4SLinus Torvalds } 1620b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 16211da177e4SLinus Torvalds 16221da177e4SLinus Torvalds /** 1623bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 16241da177e4SLinus Torvalds * @skb: buffer to copy 1625117632e6SEric Dumazet * @headroom: headroom of new skb 16261da177e4SLinus Torvalds * @gfp_mask: allocation priority 1627bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1628bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1629bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 16301da177e4SLinus Torvalds * 16311da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 16321da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 16331da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 16341da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 16351da177e4SLinus Torvalds * or the pointer to the buffer on success. 16361da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 16371da177e4SLinus Torvalds */ 16381da177e4SLinus Torvalds 1639bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1640bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 16411da177e4SLinus Torvalds { 1642117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1643bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1644bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 16456602cebbSEric Dumazet 16461da177e4SLinus Torvalds if (!n) 16471da177e4SLinus Torvalds goto out; 16481da177e4SLinus Torvalds 16491da177e4SLinus Torvalds /* Set the data pointer */ 1650117632e6SEric Dumazet skb_reserve(n, headroom); 16511da177e4SLinus Torvalds /* Set the tail pointer and length */ 16521da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 16531da177e4SLinus Torvalds /* Copy the bytes */ 1654d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 16551da177e4SLinus Torvalds 165625f484a6SHerbert Xu n->truesize += skb->data_len; 16571da177e4SLinus Torvalds n->data_len = skb->data_len; 16581da177e4SLinus Torvalds n->len = skb->len; 16591da177e4SLinus Torvalds 16601da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 16611da177e4SLinus Torvalds int i; 16621da177e4SLinus Torvalds 16631f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 16641f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 16651511022cSDan Carpenter kfree_skb(n); 16661511022cSDan Carpenter n = NULL; 1667a6686f2fSShirley Ma goto out; 1668a6686f2fSShirley Ma } 16691da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16701da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1671ea2ab693SIan Campbell skb_frag_ref(skb, i); 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 16741da177e4SLinus Torvalds } 16751da177e4SLinus Torvalds 167621dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 16771da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 16781da177e4SLinus Torvalds skb_clone_fraglist(n); 16791da177e4SLinus Torvalds } 16801da177e4SLinus Torvalds 168108303c18SIlya Lesokhin skb_copy_header(n, skb); 16821da177e4SLinus Torvalds out: 16831da177e4SLinus Torvalds return n; 16841da177e4SLinus Torvalds } 1685bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 16861da177e4SLinus Torvalds 16871da177e4SLinus Torvalds /** 16881da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 16891da177e4SLinus Torvalds * @skb: buffer to reallocate 16901da177e4SLinus Torvalds * @nhead: room to add at head 16911da177e4SLinus Torvalds * @ntail: room to add at tail 16921da177e4SLinus Torvalds * @gfp_mask: allocation priority 16931da177e4SLinus Torvalds * 1694bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1695bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 16961da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 16971da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 16981da177e4SLinus Torvalds * 16991da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 17001da177e4SLinus Torvalds * reloaded after call to this function. 17011da177e4SLinus Torvalds */ 17021da177e4SLinus Torvalds 170386a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1704dd0fc66fSAl Viro gfp_t gfp_mask) 17051da177e4SLinus Torvalds { 1706158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1707158f323bSEric Dumazet int size = osize + nhead + ntail; 17081da177e4SLinus Torvalds long off; 1709158f323bSEric Dumazet u8 *data; 17101da177e4SLinus Torvalds 17114edd87adSHerbert Xu BUG_ON(nhead < 0); 17124edd87adSHerbert Xu 17139f77fad3STim Hansen BUG_ON(skb_shared(skb)); 17141da177e4SLinus Torvalds 1715*753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 1716*753f1ca4SPavel Begunkov 17171da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 17181da177e4SLinus Torvalds 1719c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1720c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1721c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1722c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 17231da177e4SLinus Torvalds if (!data) 17241da177e4SLinus Torvalds goto nodata; 172587151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 17261da177e4SLinus Torvalds 17271da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 17286602cebbSEric Dumazet * optimized for the cases when header is void. 17296602cebbSEric Dumazet */ 17306602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 17316602cebbSEric Dumazet 17326602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 17336602cebbSEric Dumazet skb_shinfo(skb), 1734fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 17351da177e4SLinus Torvalds 17363e24591aSAlexander Duyck /* 17373e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 17383e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 17393e24591aSAlexander Duyck * be since all we did is relocate the values 17403e24591aSAlexander Duyck */ 17413e24591aSAlexander Duyck if (skb_cloned(skb)) { 174270008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1743a6686f2fSShirley Ma goto nofrags; 17441f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1745c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 17461da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1747ea2ab693SIan Campbell skb_frag_ref(skb, i); 17481da177e4SLinus Torvalds 174921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 17501da177e4SLinus Torvalds skb_clone_fraglist(skb); 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds skb_release_data(skb); 17533e24591aSAlexander Duyck } else { 17543e24591aSAlexander Duyck skb_free_head(skb); 17551fd63041SEric Dumazet } 17561da177e4SLinus Torvalds off = (data + nhead) - skb->head; 17571da177e4SLinus Torvalds 17581da177e4SLinus Torvalds skb->head = data; 1759d3836f21SEric Dumazet skb->head_frag = 0; 17601da177e4SLinus Torvalds skb->data += off; 1761763087daSEric Dumazet 1762763087daSEric Dumazet skb_set_end_offset(skb, size); 17634305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 176456eb8882SPatrick McHardy off = nhead; 176556eb8882SPatrick McHardy #endif 176627a884dcSArnaldo Carvalho de Melo skb->tail += off; 1767b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 17681da177e4SLinus Torvalds skb->cloned = 0; 1769334a8132SPatrick McHardy skb->hdr_len = 0; 17701da177e4SLinus Torvalds skb->nohdr = 0; 17711da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1772158f323bSEric Dumazet 1773de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1774de8f3a83SDaniel Borkmann 1775158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1776158f323bSEric Dumazet * For the moment, we really care of rx path, or 1777158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1778158f323bSEric Dumazet */ 1779158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1780158f323bSEric Dumazet skb->truesize += size - osize; 1781158f323bSEric Dumazet 17821da177e4SLinus Torvalds return 0; 17831da177e4SLinus Torvalds 1784a6686f2fSShirley Ma nofrags: 1785a6686f2fSShirley Ma kfree(data); 17861da177e4SLinus Torvalds nodata: 17871da177e4SLinus Torvalds return -ENOMEM; 17881da177e4SLinus Torvalds } 1789b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 17901da177e4SLinus Torvalds 17911da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 17941da177e4SLinus Torvalds { 17951da177e4SLinus Torvalds struct sk_buff *skb2; 17961da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 17971da177e4SLinus Torvalds 17981da177e4SLinus Torvalds if (delta <= 0) 17991da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 18001da177e4SLinus Torvalds else { 18011da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 18021da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 18031da177e4SLinus Torvalds GFP_ATOMIC)) { 18041da177e4SLinus Torvalds kfree_skb(skb2); 18051da177e4SLinus Torvalds skb2 = NULL; 18061da177e4SLinus Torvalds } 18071da177e4SLinus Torvalds } 18081da177e4SLinus Torvalds return skb2; 18091da177e4SLinus Torvalds } 1810b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 18111da177e4SLinus Torvalds 18122b88cba5SEric Dumazet int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 18132b88cba5SEric Dumazet { 18142b88cba5SEric Dumazet unsigned int saved_end_offset, saved_truesize; 18152b88cba5SEric Dumazet struct skb_shared_info *shinfo; 18162b88cba5SEric Dumazet int res; 18172b88cba5SEric Dumazet 18182b88cba5SEric Dumazet saved_end_offset = skb_end_offset(skb); 18192b88cba5SEric Dumazet saved_truesize = skb->truesize; 18202b88cba5SEric Dumazet 18212b88cba5SEric Dumazet res = pskb_expand_head(skb, 0, 0, pri); 18222b88cba5SEric Dumazet if (res) 18232b88cba5SEric Dumazet return res; 18242b88cba5SEric Dumazet 18252b88cba5SEric Dumazet skb->truesize = saved_truesize; 18262b88cba5SEric Dumazet 18272b88cba5SEric Dumazet if (likely(skb_end_offset(skb) == saved_end_offset)) 18282b88cba5SEric Dumazet return 0; 18292b88cba5SEric Dumazet 18302b88cba5SEric Dumazet shinfo = skb_shinfo(skb); 18312b88cba5SEric Dumazet 18322b88cba5SEric Dumazet /* We are about to change back skb->end, 18332b88cba5SEric Dumazet * we need to move skb_shinfo() to its new location. 18342b88cba5SEric Dumazet */ 18352b88cba5SEric Dumazet memmove(skb->head + saved_end_offset, 18362b88cba5SEric Dumazet shinfo, 18372b88cba5SEric Dumazet offsetof(struct skb_shared_info, frags[shinfo->nr_frags])); 18382b88cba5SEric Dumazet 18392b88cba5SEric Dumazet skb_set_end_offset(skb, saved_end_offset); 18402b88cba5SEric Dumazet 18412b88cba5SEric Dumazet return 0; 18422b88cba5SEric Dumazet } 18432b88cba5SEric Dumazet 18441da177e4SLinus Torvalds /** 1845f1260ff1SVasily Averin * skb_expand_head - reallocate header of &sk_buff 1846f1260ff1SVasily Averin * @skb: buffer to reallocate 1847f1260ff1SVasily Averin * @headroom: needed headroom 1848f1260ff1SVasily Averin * 1849f1260ff1SVasily Averin * Unlike skb_realloc_headroom, this one does not allocate a new skb 1850f1260ff1SVasily Averin * if possible; copies skb->sk to new skb as needed 1851f1260ff1SVasily Averin * and frees original skb in case of failures. 1852f1260ff1SVasily Averin * 1853f1260ff1SVasily Averin * It expect increased headroom and generates warning otherwise. 1854f1260ff1SVasily Averin */ 1855f1260ff1SVasily Averin 1856f1260ff1SVasily Averin struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom) 1857f1260ff1SVasily Averin { 1858f1260ff1SVasily Averin int delta = headroom - skb_headroom(skb); 18597f678defSVasily Averin int osize = skb_end_offset(skb); 18607f678defSVasily Averin struct sock *sk = skb->sk; 1861f1260ff1SVasily Averin 1862f1260ff1SVasily Averin if (WARN_ONCE(delta <= 0, 1863f1260ff1SVasily Averin "%s is expecting an increase in the headroom", __func__)) 1864f1260ff1SVasily Averin return skb; 1865f1260ff1SVasily Averin 18667f678defSVasily Averin delta = SKB_DATA_ALIGN(delta); 18677f678defSVasily Averin /* pskb_expand_head() might crash, if skb is shared. */ 18687f678defSVasily Averin if (skb_shared(skb) || !is_skb_wmem(skb)) { 1869f1260ff1SVasily Averin struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1870f1260ff1SVasily Averin 18717f678defSVasily Averin if (unlikely(!nskb)) 18727f678defSVasily Averin goto fail; 18737f678defSVasily Averin 18747f678defSVasily Averin if (sk) 18757f678defSVasily Averin skb_set_owner_w(nskb, sk); 1876f1260ff1SVasily Averin consume_skb(skb); 1877f1260ff1SVasily Averin skb = nskb; 1878f1260ff1SVasily Averin } 18797f678defSVasily Averin if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC)) 18807f678defSVasily Averin goto fail; 18817f678defSVasily Averin 18827f678defSVasily Averin if (sk && is_skb_wmem(skb)) { 18837f678defSVasily Averin delta = skb_end_offset(skb) - osize; 18847f678defSVasily Averin refcount_add(delta, &sk->sk_wmem_alloc); 18857f678defSVasily Averin skb->truesize += delta; 1886f1260ff1SVasily Averin } 1887f1260ff1SVasily Averin return skb; 18887f678defSVasily Averin 18897f678defSVasily Averin fail: 18907f678defSVasily Averin kfree_skb(skb); 18917f678defSVasily Averin return NULL; 1892f1260ff1SVasily Averin } 1893f1260ff1SVasily Averin EXPORT_SYMBOL(skb_expand_head); 1894f1260ff1SVasily Averin 1895f1260ff1SVasily Averin /** 18961da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 18971da177e4SLinus Torvalds * @skb: buffer to copy 18981da177e4SLinus Torvalds * @newheadroom: new free bytes at head 18991da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 19001da177e4SLinus Torvalds * @gfp_mask: allocation priority 19011da177e4SLinus Torvalds * 19021da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 19031da177e4SLinus Torvalds * allocate additional space. 19041da177e4SLinus Torvalds * 19051da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 19061da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 19071da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 19081da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 19091da177e4SLinus Torvalds * 19101da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 19111da177e4SLinus Torvalds * is called from an interrupt. 19121da177e4SLinus Torvalds */ 19131da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 191486a76cafSVictor Fusco int newheadroom, int newtailroom, 1915dd0fc66fSAl Viro gfp_t gfp_mask) 19161da177e4SLinus Torvalds { 19171da177e4SLinus Torvalds /* 19181da177e4SLinus Torvalds * Allocate the copy buffer 19191da177e4SLinus Torvalds */ 1920c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1921c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1922c93bdd0eSMel Gorman NUMA_NO_NODE); 1923efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 19241da177e4SLinus Torvalds int head_copy_len, head_copy_off; 19251da177e4SLinus Torvalds 19261da177e4SLinus Torvalds if (!n) 19271da177e4SLinus Torvalds return NULL; 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds skb_reserve(n, newheadroom); 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds /* Set the tail pointer and length */ 19321da177e4SLinus Torvalds skb_put(n, skb->len); 19331da177e4SLinus Torvalds 1934efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 19351da177e4SLinus Torvalds head_copy_off = 0; 19361da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 19371da177e4SLinus Torvalds head_copy_len = newheadroom; 19381da177e4SLinus Torvalds else 19391da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 19401da177e4SLinus Torvalds 19411da177e4SLinus Torvalds /* Copy the linear header and data. */ 19429f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 19439f77fad3STim Hansen skb->len + head_copy_len)); 19441da177e4SLinus Torvalds 194508303c18SIlya Lesokhin skb_copy_header(n, skb); 19461da177e4SLinus Torvalds 1947030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1948efd1e8d5SPatrick McHardy 19491da177e4SLinus Torvalds return n; 19501da177e4SLinus Torvalds } 1951b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 19521da177e4SLinus Torvalds 19531da177e4SLinus Torvalds /** 1954cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 19551da177e4SLinus Torvalds * @skb: buffer to pad 19561da177e4SLinus Torvalds * @pad: space to pad 1957cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 19581da177e4SLinus Torvalds * 19591da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 19601da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 19611da177e4SLinus Torvalds * beyond the buffer end onto the wire. 19621da177e4SLinus Torvalds * 1963cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1964cd0a137aSFlorian Fainelli * if @free_on_error is true. 19651da177e4SLinus Torvalds */ 19661da177e4SLinus Torvalds 1967cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 19681da177e4SLinus Torvalds { 19695b057c6bSHerbert Xu int err; 19705b057c6bSHerbert Xu int ntail; 19711da177e4SLinus Torvalds 19721da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 19735b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 19741da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 19755b057c6bSHerbert Xu return 0; 19761da177e4SLinus Torvalds } 19771da177e4SLinus Torvalds 19784305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 19795b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 19805b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 19815b057c6bSHerbert Xu if (unlikely(err)) 19825b057c6bSHerbert Xu goto free_skb; 19835b057c6bSHerbert Xu } 19845b057c6bSHerbert Xu 19855b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 19865b057c6bSHerbert Xu * to be audited. 19875b057c6bSHerbert Xu */ 19885b057c6bSHerbert Xu err = skb_linearize(skb); 19895b057c6bSHerbert Xu if (unlikely(err)) 19905b057c6bSHerbert Xu goto free_skb; 19915b057c6bSHerbert Xu 19925b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 19935b057c6bSHerbert Xu return 0; 19945b057c6bSHerbert Xu 19955b057c6bSHerbert Xu free_skb: 1996cd0a137aSFlorian Fainelli if (free_on_error) 19971da177e4SLinus Torvalds kfree_skb(skb); 19985b057c6bSHerbert Xu return err; 19991da177e4SLinus Torvalds } 2000cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 20011da177e4SLinus Torvalds 20020dde3e16SIlpo Järvinen /** 20030c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 20040c7ddf36SMathias Krause * @skb: start of the buffer to use 20050c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 20060c7ddf36SMathias Krause * @len: amount of data to add 20070c7ddf36SMathias Krause * 20080c7ddf36SMathias Krause * This function extends the used data area of the potentially 20090c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 20100c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 20110c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 20120c7ddf36SMathias Krause * returned. 20130c7ddf36SMathias Krause */ 20140c7ddf36SMathias Krause 20154df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 20160c7ddf36SMathias Krause { 20170c7ddf36SMathias Krause if (tail != skb) { 20180c7ddf36SMathias Krause skb->data_len += len; 20190c7ddf36SMathias Krause skb->len += len; 20200c7ddf36SMathias Krause } 20210c7ddf36SMathias Krause return skb_put(tail, len); 20220c7ddf36SMathias Krause } 20230c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 20240c7ddf36SMathias Krause 20250c7ddf36SMathias Krause /** 20260dde3e16SIlpo Järvinen * skb_put - add data to a buffer 20270dde3e16SIlpo Järvinen * @skb: buffer to use 20280dde3e16SIlpo Järvinen * @len: amount of data to add 20290dde3e16SIlpo Järvinen * 20300dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 20310dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 20320dde3e16SIlpo Järvinen * first byte of the extra data is returned. 20330dde3e16SIlpo Järvinen */ 20344df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 20350dde3e16SIlpo Järvinen { 20364df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 20370dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 20380dde3e16SIlpo Järvinen skb->tail += len; 20390dde3e16SIlpo Järvinen skb->len += len; 20400dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 20410dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 20420dde3e16SIlpo Järvinen return tmp; 20430dde3e16SIlpo Järvinen } 20440dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 20450dde3e16SIlpo Järvinen 20466be8ac2fSIlpo Järvinen /** 2047c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 2048c2aa270aSIlpo Järvinen * @skb: buffer to use 2049c2aa270aSIlpo Järvinen * @len: amount of data to add 2050c2aa270aSIlpo Järvinen * 2051c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 2052c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 2053c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 2054c2aa270aSIlpo Järvinen */ 2055d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 2056c2aa270aSIlpo Järvinen { 2057c2aa270aSIlpo Järvinen skb->data -= len; 2058c2aa270aSIlpo Järvinen skb->len += len; 2059c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 2060c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 2061c2aa270aSIlpo Järvinen return skb->data; 2062c2aa270aSIlpo Järvinen } 2063c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 2064c2aa270aSIlpo Järvinen 2065c2aa270aSIlpo Järvinen /** 20666be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 20676be8ac2fSIlpo Järvinen * @skb: buffer to use 20686be8ac2fSIlpo Järvinen * @len: amount of data to remove 20696be8ac2fSIlpo Järvinen * 20706be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 20716be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 20726be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 20736be8ac2fSIlpo Järvinen * the old data. 20746be8ac2fSIlpo Järvinen */ 2075af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 20766be8ac2fSIlpo Järvinen { 207747d29646SDavid S. Miller return skb_pull_inline(skb, len); 20786be8ac2fSIlpo Järvinen } 20796be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 20806be8ac2fSIlpo Järvinen 2081419ae74eSIlpo Järvinen /** 208213244cccSLuiz Augusto von Dentz * skb_pull_data - remove data from the start of a buffer returning its 208313244cccSLuiz Augusto von Dentz * original position. 208413244cccSLuiz Augusto von Dentz * @skb: buffer to use 208513244cccSLuiz Augusto von Dentz * @len: amount of data to remove 208613244cccSLuiz Augusto von Dentz * 208713244cccSLuiz Augusto von Dentz * This function removes data from the start of a buffer, returning 208813244cccSLuiz Augusto von Dentz * the memory to the headroom. A pointer to the original data in the buffer 208913244cccSLuiz Augusto von Dentz * is returned after checking if there is enough data to pull. Once the 209013244cccSLuiz Augusto von Dentz * data has been pulled future pushes will overwrite the old data. 209113244cccSLuiz Augusto von Dentz */ 209213244cccSLuiz Augusto von Dentz void *skb_pull_data(struct sk_buff *skb, size_t len) 209313244cccSLuiz Augusto von Dentz { 209413244cccSLuiz Augusto von Dentz void *data = skb->data; 209513244cccSLuiz Augusto von Dentz 209613244cccSLuiz Augusto von Dentz if (skb->len < len) 209713244cccSLuiz Augusto von Dentz return NULL; 209813244cccSLuiz Augusto von Dentz 209913244cccSLuiz Augusto von Dentz skb_pull(skb, len); 210013244cccSLuiz Augusto von Dentz 210113244cccSLuiz Augusto von Dentz return data; 210213244cccSLuiz Augusto von Dentz } 210313244cccSLuiz Augusto von Dentz EXPORT_SYMBOL(skb_pull_data); 210413244cccSLuiz Augusto von Dentz 210513244cccSLuiz Augusto von Dentz /** 2106419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 2107419ae74eSIlpo Järvinen * @skb: buffer to alter 2108419ae74eSIlpo Järvinen * @len: new length 2109419ae74eSIlpo Järvinen * 2110419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 2111419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 2112419ae74eSIlpo Järvinen * The skb must be linear. 2113419ae74eSIlpo Järvinen */ 2114419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 2115419ae74eSIlpo Järvinen { 2116419ae74eSIlpo Järvinen if (skb->len > len) 2117419ae74eSIlpo Järvinen __skb_trim(skb, len); 2118419ae74eSIlpo Järvinen } 2119419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 2120419ae74eSIlpo Järvinen 21213cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 21221da177e4SLinus Torvalds */ 21231da177e4SLinus Torvalds 21243cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 21251da177e4SLinus Torvalds { 212627b437c8SHerbert Xu struct sk_buff **fragp; 212727b437c8SHerbert Xu struct sk_buff *frag; 21281da177e4SLinus Torvalds int offset = skb_headlen(skb); 21291da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 21301da177e4SLinus Torvalds int i; 213127b437c8SHerbert Xu int err; 213227b437c8SHerbert Xu 213327b437c8SHerbert Xu if (skb_cloned(skb) && 213427b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 213527b437c8SHerbert Xu return err; 21361da177e4SLinus Torvalds 2137f4d26fb3SHerbert Xu i = 0; 2138f4d26fb3SHerbert Xu if (offset >= len) 2139f4d26fb3SHerbert Xu goto drop_pages; 2140f4d26fb3SHerbert Xu 2141f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 21429e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 214327b437c8SHerbert Xu 214427b437c8SHerbert Xu if (end < len) { 21451da177e4SLinus Torvalds offset = end; 214627b437c8SHerbert Xu continue; 21471da177e4SLinus Torvalds } 21481da177e4SLinus Torvalds 21499e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 215027b437c8SHerbert Xu 2151f4d26fb3SHerbert Xu drop_pages: 215227b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 215327b437c8SHerbert Xu 215427b437c8SHerbert Xu for (; i < nfrags; i++) 2155ea2ab693SIan Campbell skb_frag_unref(skb, i); 215627b437c8SHerbert Xu 215721dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 215827b437c8SHerbert Xu skb_drop_fraglist(skb); 2159f4d26fb3SHerbert Xu goto done; 216027b437c8SHerbert Xu } 216127b437c8SHerbert Xu 216227b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 216327b437c8SHerbert Xu fragp = &frag->next) { 216427b437c8SHerbert Xu int end = offset + frag->len; 216527b437c8SHerbert Xu 216627b437c8SHerbert Xu if (skb_shared(frag)) { 216727b437c8SHerbert Xu struct sk_buff *nfrag; 216827b437c8SHerbert Xu 216927b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 217027b437c8SHerbert Xu if (unlikely(!nfrag)) 217127b437c8SHerbert Xu return -ENOMEM; 217227b437c8SHerbert Xu 217327b437c8SHerbert Xu nfrag->next = frag->next; 217485bb2a60SEric Dumazet consume_skb(frag); 217527b437c8SHerbert Xu frag = nfrag; 217627b437c8SHerbert Xu *fragp = frag; 217727b437c8SHerbert Xu } 217827b437c8SHerbert Xu 217927b437c8SHerbert Xu if (end < len) { 218027b437c8SHerbert Xu offset = end; 218127b437c8SHerbert Xu continue; 218227b437c8SHerbert Xu } 218327b437c8SHerbert Xu 218427b437c8SHerbert Xu if (end > len && 218527b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 218627b437c8SHerbert Xu return err; 218727b437c8SHerbert Xu 218827b437c8SHerbert Xu if (frag->next) 218927b437c8SHerbert Xu skb_drop_list(&frag->next); 219027b437c8SHerbert Xu break; 219127b437c8SHerbert Xu } 219227b437c8SHerbert Xu 2193f4d26fb3SHerbert Xu done: 219427b437c8SHerbert Xu if (len > skb_headlen(skb)) { 21951da177e4SLinus Torvalds skb->data_len -= skb->len - len; 21961da177e4SLinus Torvalds skb->len = len; 21971da177e4SLinus Torvalds } else { 21981da177e4SLinus Torvalds skb->len = len; 21991da177e4SLinus Torvalds skb->data_len = 0; 220027a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 22011da177e4SLinus Torvalds } 22021da177e4SLinus Torvalds 2203c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2204c21b48ccSEric Dumazet skb_condense(skb); 22051da177e4SLinus Torvalds return 0; 22061da177e4SLinus Torvalds } 2207b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 22081da177e4SLinus Torvalds 220988078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 221088078d98SEric Dumazet */ 221188078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 221288078d98SEric Dumazet { 221388078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 221488078d98SEric Dumazet int delta = skb->len - len; 221588078d98SEric Dumazet 2216d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2217d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2218d55bef50SDimitris Michailidis len); 221954970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 222054970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 222154970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 222254970a2fSVasily Averin 222354970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 222454970a2fSVasily Averin return -EINVAL; 222588078d98SEric Dumazet } 222688078d98SEric Dumazet return __pskb_trim(skb, len); 222788078d98SEric Dumazet } 222888078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 222988078d98SEric Dumazet 22301da177e4SLinus Torvalds /** 22311da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 22321da177e4SLinus Torvalds * @skb: buffer to reallocate 22331da177e4SLinus Torvalds * @delta: number of bytes to advance tail 22341da177e4SLinus Torvalds * 22351da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 22361da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 22371da177e4SLinus Torvalds * data from fragmented part. 22381da177e4SLinus Torvalds * 22391da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 22401da177e4SLinus Torvalds * 22411da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 22421da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 22431da177e4SLinus Torvalds * 22441da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 22451da177e4SLinus Torvalds * reloaded after call to this function. 22461da177e4SLinus Torvalds */ 22471da177e4SLinus Torvalds 22481da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 22491da177e4SLinus Torvalds * when it is necessary. 22501da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 22511da177e4SLinus Torvalds * 2. It may change skb pointers. 22521da177e4SLinus Torvalds * 22531da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 22541da177e4SLinus Torvalds */ 2255af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 22561da177e4SLinus Torvalds { 22571da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 22581da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 22591da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 22601da177e4SLinus Torvalds */ 22614305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 22621da177e4SLinus Torvalds 22631da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 22641da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 22651da177e4SLinus Torvalds GFP_ATOMIC)) 22661da177e4SLinus Torvalds return NULL; 22671da177e4SLinus Torvalds } 22681da177e4SLinus Torvalds 22699f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 22709f77fad3STim Hansen skb_tail_pointer(skb), delta)); 22711da177e4SLinus Torvalds 22721da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 22731da177e4SLinus Torvalds * size of pulled pages. Superb. 22741da177e4SLinus Torvalds */ 227521dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 22761da177e4SLinus Torvalds goto pull_pages; 22771da177e4SLinus Torvalds 22781da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 22791da177e4SLinus Torvalds eat = delta; 22801da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22819e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22829e903e08SEric Dumazet 22839e903e08SEric Dumazet if (size >= eat) 22841da177e4SLinus Torvalds goto pull_pages; 22859e903e08SEric Dumazet eat -= size; 22861da177e4SLinus Torvalds } 22871da177e4SLinus Torvalds 22881da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 228909001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 22901da177e4SLinus Torvalds * but taking into account that pulling is expected to 22911da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 22921da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 22931da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 22941da177e4SLinus Torvalds */ 22951da177e4SLinus Torvalds if (eat) { 22961da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 22971da177e4SLinus Torvalds struct sk_buff *clone = NULL; 22981da177e4SLinus Torvalds struct sk_buff *insp = NULL; 22991da177e4SLinus Torvalds 23001da177e4SLinus Torvalds do { 23011da177e4SLinus Torvalds if (list->len <= eat) { 23021da177e4SLinus Torvalds /* Eaten as whole. */ 23031da177e4SLinus Torvalds eat -= list->len; 23041da177e4SLinus Torvalds list = list->next; 23051da177e4SLinus Torvalds insp = list; 23061da177e4SLinus Torvalds } else { 23071da177e4SLinus Torvalds /* Eaten partially. */ 23081da177e4SLinus Torvalds 23091da177e4SLinus Torvalds if (skb_shared(list)) { 23101da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 23111da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 23121da177e4SLinus Torvalds if (!clone) 23131da177e4SLinus Torvalds return NULL; 23141da177e4SLinus Torvalds insp = list->next; 23151da177e4SLinus Torvalds list = clone; 23161da177e4SLinus Torvalds } else { 23171da177e4SLinus Torvalds /* This may be pulled without 23181da177e4SLinus Torvalds * problems. */ 23191da177e4SLinus Torvalds insp = list; 23201da177e4SLinus Torvalds } 23211da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 23221da177e4SLinus Torvalds kfree_skb(clone); 23231da177e4SLinus Torvalds return NULL; 23241da177e4SLinus Torvalds } 23251da177e4SLinus Torvalds break; 23261da177e4SLinus Torvalds } 23271da177e4SLinus Torvalds } while (eat); 23281da177e4SLinus Torvalds 23291da177e4SLinus Torvalds /* Free pulled out fragments. */ 23301da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 23311da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 2332ef527f96SEric Dumazet consume_skb(list); 23331da177e4SLinus Torvalds } 23341da177e4SLinus Torvalds /* And insert new clone at head. */ 23351da177e4SLinus Torvalds if (clone) { 23361da177e4SLinus Torvalds clone->next = list; 23371da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 23381da177e4SLinus Torvalds } 23391da177e4SLinus Torvalds } 23401da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 23411da177e4SLinus Torvalds 23421da177e4SLinus Torvalds pull_pages: 23431da177e4SLinus Torvalds eat = delta; 23441da177e4SLinus Torvalds k = 0; 23451da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 23469e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 23479e903e08SEric Dumazet 23489e903e08SEric Dumazet if (size <= eat) { 2349ea2ab693SIan Campbell skb_frag_unref(skb, i); 23509e903e08SEric Dumazet eat -= size; 23511da177e4SLinus Torvalds } else { 2352b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2353b54c9d5bSJonathan Lemon 2354b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 23551da177e4SLinus Torvalds if (eat) { 2356b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2357b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 23583ccc6c6fSlinzhang if (!i) 23593ccc6c6fSlinzhang goto end; 23601da177e4SLinus Torvalds eat = 0; 23611da177e4SLinus Torvalds } 23621da177e4SLinus Torvalds k++; 23631da177e4SLinus Torvalds } 23641da177e4SLinus Torvalds } 23651da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 23661da177e4SLinus Torvalds 23673ccc6c6fSlinzhang end: 23681da177e4SLinus Torvalds skb->tail += delta; 23691da177e4SLinus Torvalds skb->data_len -= delta; 23701da177e4SLinus Torvalds 23711f8b977aSWillem de Bruijn if (!skb->data_len) 23721f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 23731f8b977aSWillem de Bruijn 237427a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 23751da177e4SLinus Torvalds } 2376b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 23771da177e4SLinus Torvalds 237822019b17SEric Dumazet /** 237922019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 238022019b17SEric Dumazet * @skb: source skb 238122019b17SEric Dumazet * @offset: offset in source 238222019b17SEric Dumazet * @to: destination buffer 238322019b17SEric Dumazet * @len: number of bytes to copy 238422019b17SEric Dumazet * 238522019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 238622019b17SEric Dumazet * destination buffer. 238722019b17SEric Dumazet * 238822019b17SEric Dumazet * CAUTION ! : 238922019b17SEric Dumazet * If its prototype is ever changed, 239022019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 239122019b17SEric Dumazet * since it is called from BPF assembly code. 239222019b17SEric Dumazet */ 23931da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 23941da177e4SLinus Torvalds { 23951a028e50SDavid S. Miller int start = skb_headlen(skb); 2396fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2397fbb398a8SDavid S. Miller int i, copy; 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds if (offset > (int)skb->len - len) 24001da177e4SLinus Torvalds goto fault; 24011da177e4SLinus Torvalds 24021da177e4SLinus Torvalds /* Copy header. */ 24031a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 24041da177e4SLinus Torvalds if (copy > len) 24051da177e4SLinus Torvalds copy = len; 2406d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 24071da177e4SLinus Torvalds if ((len -= copy) == 0) 24081da177e4SLinus Torvalds return 0; 24091da177e4SLinus Torvalds offset += copy; 24101da177e4SLinus Torvalds to += copy; 24111da177e4SLinus Torvalds } 24121da177e4SLinus Torvalds 24131da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 24141a028e50SDavid S. Miller int end; 241551c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 24161da177e4SLinus Torvalds 2417547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24181a028e50SDavid S. Miller 241951c56b00SEric Dumazet end = start + skb_frag_size(f); 24201da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2421c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2422c613c209SWillem de Bruijn struct page *p; 24231da177e4SLinus Torvalds u8 *vaddr; 24241da177e4SLinus Torvalds 24251da177e4SLinus Torvalds if (copy > len) 24261da177e4SLinus Torvalds copy = len; 24271da177e4SLinus Torvalds 2428c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2429b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2430c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2431c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2432c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 243351c56b00SEric Dumazet kunmap_atomic(vaddr); 2434c613c209SWillem de Bruijn } 24351da177e4SLinus Torvalds 24361da177e4SLinus Torvalds if ((len -= copy) == 0) 24371da177e4SLinus Torvalds return 0; 24381da177e4SLinus Torvalds offset += copy; 24391da177e4SLinus Torvalds to += copy; 24401da177e4SLinus Torvalds } 24411a028e50SDavid S. Miller start = end; 24421da177e4SLinus Torvalds } 24431da177e4SLinus Torvalds 2444fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 24451a028e50SDavid S. Miller int end; 24461da177e4SLinus Torvalds 2447547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24481a028e50SDavid S. Miller 2449fbb398a8SDavid S. Miller end = start + frag_iter->len; 24501da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 24511da177e4SLinus Torvalds if (copy > len) 24521da177e4SLinus Torvalds copy = len; 2453fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 24541da177e4SLinus Torvalds goto fault; 24551da177e4SLinus Torvalds if ((len -= copy) == 0) 24561da177e4SLinus Torvalds return 0; 24571da177e4SLinus Torvalds offset += copy; 24581da177e4SLinus Torvalds to += copy; 24591da177e4SLinus Torvalds } 24601a028e50SDavid S. Miller start = end; 24611da177e4SLinus Torvalds } 2462a6686f2fSShirley Ma 24631da177e4SLinus Torvalds if (!len) 24641da177e4SLinus Torvalds return 0; 24651da177e4SLinus Torvalds 24661da177e4SLinus Torvalds fault: 24671da177e4SLinus Torvalds return -EFAULT; 24681da177e4SLinus Torvalds } 2469b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 24701da177e4SLinus Torvalds 24719c55e01cSJens Axboe /* 24729c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 24739c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 24749c55e01cSJens Axboe */ 24759c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 24769c55e01cSJens Axboe { 24778b9d3728SJarek Poplawski put_page(spd->pages[i]); 24788b9d3728SJarek Poplawski } 24799c55e01cSJens Axboe 2480a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 24814fb66994SJarek Poplawski unsigned int *offset, 248218aafc62SEric Dumazet struct sock *sk) 24838b9d3728SJarek Poplawski { 24845640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 24858b9d3728SJarek Poplawski 24865640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 24878b9d3728SJarek Poplawski return NULL; 24884fb66994SJarek Poplawski 24895640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 24904fb66994SJarek Poplawski 24915640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 24925640f768SEric Dumazet page_address(page) + *offset, *len); 24935640f768SEric Dumazet *offset = pfrag->offset; 24945640f768SEric Dumazet pfrag->offset += *len; 24954fb66994SJarek Poplawski 24965640f768SEric Dumazet return pfrag->page; 24979c55e01cSJens Axboe } 24989c55e01cSJens Axboe 249941c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 250041c73a0dSEric Dumazet struct page *page, 250141c73a0dSEric Dumazet unsigned int offset) 250241c73a0dSEric Dumazet { 250341c73a0dSEric Dumazet return spd->nr_pages && 250441c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 250541c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 250641c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 250741c73a0dSEric Dumazet } 250841c73a0dSEric Dumazet 25099c55e01cSJens Axboe /* 25109c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 25119c55e01cSJens Axboe */ 2512a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 251335f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 25144fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 251518aafc62SEric Dumazet bool linear, 25167a67e56fSJarek Poplawski struct sock *sk) 25179c55e01cSJens Axboe { 251841c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2519a108d5f3SDavid S. Miller return true; 25209c55e01cSJens Axboe 25218b9d3728SJarek Poplawski if (linear) { 252218aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 25238b9d3728SJarek Poplawski if (!page) 2524a108d5f3SDavid S. Miller return true; 252541c73a0dSEric Dumazet } 252641c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 252741c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2528a108d5f3SDavid S. Miller return false; 252941c73a0dSEric Dumazet } 25308b9d3728SJarek Poplawski get_page(page); 25319c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 25324fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 25339c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 25349c55e01cSJens Axboe spd->nr_pages++; 25358b9d3728SJarek Poplawski 2536a108d5f3SDavid S. Miller return false; 25379c55e01cSJens Axboe } 25389c55e01cSJens Axboe 2539a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 25402870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 254118aafc62SEric Dumazet unsigned int *len, 2542d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 254335f3d14dSJens Axboe struct sock *sk, 254435f3d14dSJens Axboe struct pipe_inode_info *pipe) 25459c55e01cSJens Axboe { 25462870c43dSOctavian Purdila if (!*len) 2547a108d5f3SDavid S. Miller return true; 25489c55e01cSJens Axboe 25492870c43dSOctavian Purdila /* skip this segment if already processed */ 25502870c43dSOctavian Purdila if (*off >= plen) { 25512870c43dSOctavian Purdila *off -= plen; 2552a108d5f3SDavid S. Miller return false; 25532870c43dSOctavian Purdila } 25542870c43dSOctavian Purdila 25552870c43dSOctavian Purdila /* ignore any bits we already processed */ 25569ca1b22dSEric Dumazet poff += *off; 25579ca1b22dSEric Dumazet plen -= *off; 25582870c43dSOctavian Purdila *off = 0; 25592870c43dSOctavian Purdila 256018aafc62SEric Dumazet do { 256118aafc62SEric Dumazet unsigned int flen = min(*len, plen); 25622870c43dSOctavian Purdila 256318aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 256418aafc62SEric Dumazet linear, sk)) 2565a108d5f3SDavid S. Miller return true; 256618aafc62SEric Dumazet poff += flen; 256718aafc62SEric Dumazet plen -= flen; 25682870c43dSOctavian Purdila *len -= flen; 256918aafc62SEric Dumazet } while (*len && plen); 25702870c43dSOctavian Purdila 2571a108d5f3SDavid S. Miller return false; 2572db43a282SOctavian Purdila } 25739c55e01cSJens Axboe 25749c55e01cSJens Axboe /* 2575a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 25762870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 25779c55e01cSJens Axboe */ 2578a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 257935f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 258035f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 25812870c43dSOctavian Purdila { 25822870c43dSOctavian Purdila int seg; 2583fa9835e5STom Herbert struct sk_buff *iter; 25849c55e01cSJens Axboe 25851d0c0b32SEric Dumazet /* map the linear part : 25862996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 25872996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 25882996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 25899c55e01cSJens Axboe */ 25902870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 25912870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 25922870c43dSOctavian Purdila skb_headlen(skb), 259318aafc62SEric Dumazet offset, len, spd, 25943a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 25951d0c0b32SEric Dumazet sk, pipe)) 2596a108d5f3SDavid S. Miller return true; 25979c55e01cSJens Axboe 25989c55e01cSJens Axboe /* 25999c55e01cSJens Axboe * then map the fragments 26009c55e01cSJens Axboe */ 26019c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 26029c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 26039c55e01cSJens Axboe 2604ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2605b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 260618aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2607a108d5f3SDavid S. Miller return true; 26089c55e01cSJens Axboe } 26099c55e01cSJens Axboe 2610fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2611fa9835e5STom Herbert if (*offset >= iter->len) { 2612fa9835e5STom Herbert *offset -= iter->len; 2613fa9835e5STom Herbert continue; 2614fa9835e5STom Herbert } 2615fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2616fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2617fa9835e5STom Herbert * case. 2618fa9835e5STom Herbert */ 2619fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2620fa9835e5STom Herbert return true; 2621fa9835e5STom Herbert } 2622fa9835e5STom Herbert 2623a108d5f3SDavid S. Miller return false; 26249c55e01cSJens Axboe } 26259c55e01cSJens Axboe 26269c55e01cSJens Axboe /* 26279c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2628fa9835e5STom Herbert * the fragments, and the frag list. 26299c55e01cSJens Axboe */ 2630a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 26319c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 263225869262SAl Viro unsigned int flags) 26339c55e01cSJens Axboe { 263441c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 263541c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 26369c55e01cSJens Axboe struct splice_pipe_desc spd = { 26379c55e01cSJens Axboe .pages = pages, 26389c55e01cSJens Axboe .partial = partial, 2639047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 264028a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 26419c55e01cSJens Axboe .spd_release = sock_spd_release, 26429c55e01cSJens Axboe }; 264335f3d14dSJens Axboe int ret = 0; 264435f3d14dSJens Axboe 2645fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 26469c55e01cSJens Axboe 2647a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 264825869262SAl Viro ret = splice_to_pipe(pipe, &spd); 26499c55e01cSJens Axboe 265035f3d14dSJens Axboe return ret; 26519c55e01cSJens Axboe } 26522b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 26539c55e01cSJens Axboe 26540739cd28SCong Wang static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg, 26550739cd28SCong Wang struct kvec *vec, size_t num, size_t size) 26560739cd28SCong Wang { 26570739cd28SCong Wang struct socket *sock = sk->sk_socket; 26580739cd28SCong Wang 26590739cd28SCong Wang if (!sock) 26600739cd28SCong Wang return -EINVAL; 26610739cd28SCong Wang return kernel_sendmsg(sock, msg, vec, num, size); 26620739cd28SCong Wang } 26630739cd28SCong Wang 26640739cd28SCong Wang static int sendpage_unlocked(struct sock *sk, struct page *page, int offset, 26650739cd28SCong Wang size_t size, int flags) 26660739cd28SCong Wang { 26670739cd28SCong Wang struct socket *sock = sk->sk_socket; 26680739cd28SCong Wang 26690739cd28SCong Wang if (!sock) 26700739cd28SCong Wang return -EINVAL; 26710739cd28SCong Wang return kernel_sendpage(sock, page, offset, size, flags); 26720739cd28SCong Wang } 26730739cd28SCong Wang 26740739cd28SCong Wang typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg, 26750739cd28SCong Wang struct kvec *vec, size_t num, size_t size); 26760739cd28SCong Wang typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset, 26770739cd28SCong Wang size_t size, int flags); 26780739cd28SCong Wang static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, 26790739cd28SCong Wang int len, sendmsg_func sendmsg, sendpage_func sendpage) 268020bf50deSTom Herbert { 268120bf50deSTom Herbert unsigned int orig_len = len; 268220bf50deSTom Herbert struct sk_buff *head = skb; 268320bf50deSTom Herbert unsigned short fragidx; 268420bf50deSTom Herbert int slen, ret; 268520bf50deSTom Herbert 268620bf50deSTom Herbert do_frag_list: 268720bf50deSTom Herbert 268820bf50deSTom Herbert /* Deal with head data */ 268920bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 269020bf50deSTom Herbert struct kvec kv; 269120bf50deSTom Herbert struct msghdr msg; 269220bf50deSTom Herbert 269320bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 269420bf50deSTom Herbert kv.iov_base = skb->data + offset; 2695db5980d8SJohn Fastabend kv.iov_len = slen; 269620bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2697bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 269820bf50deSTom Herbert 26990739cd28SCong Wang ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked, 27000739cd28SCong Wang sendmsg_unlocked, sk, &msg, &kv, 1, slen); 270120bf50deSTom Herbert if (ret <= 0) 270220bf50deSTom Herbert goto error; 270320bf50deSTom Herbert 270420bf50deSTom Herbert offset += ret; 270520bf50deSTom Herbert len -= ret; 270620bf50deSTom Herbert } 270720bf50deSTom Herbert 270820bf50deSTom Herbert /* All the data was skb head? */ 270920bf50deSTom Herbert if (!len) 271020bf50deSTom Herbert goto out; 271120bf50deSTom Herbert 271220bf50deSTom Herbert /* Make offset relative to start of frags */ 271320bf50deSTom Herbert offset -= skb_headlen(skb); 271420bf50deSTom Herbert 271520bf50deSTom Herbert /* Find where we are in frag list */ 271620bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 271720bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 271820bf50deSTom Herbert 2719d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 272020bf50deSTom Herbert break; 272120bf50deSTom Herbert 2722d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 272320bf50deSTom Herbert } 272420bf50deSTom Herbert 272520bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 272620bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 272720bf50deSTom Herbert 2728d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 272920bf50deSTom Herbert 273020bf50deSTom Herbert while (slen) { 27310739cd28SCong Wang ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked, 27320739cd28SCong Wang sendpage_unlocked, sk, 27330739cd28SCong Wang skb_frag_page(frag), 2734b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 273520bf50deSTom Herbert slen, MSG_DONTWAIT); 273620bf50deSTom Herbert if (ret <= 0) 273720bf50deSTom Herbert goto error; 273820bf50deSTom Herbert 273920bf50deSTom Herbert len -= ret; 274020bf50deSTom Herbert offset += ret; 274120bf50deSTom Herbert slen -= ret; 274220bf50deSTom Herbert } 274320bf50deSTom Herbert 274420bf50deSTom Herbert offset = 0; 274520bf50deSTom Herbert } 274620bf50deSTom Herbert 274720bf50deSTom Herbert if (len) { 274820bf50deSTom Herbert /* Process any frag lists */ 274920bf50deSTom Herbert 275020bf50deSTom Herbert if (skb == head) { 275120bf50deSTom Herbert if (skb_has_frag_list(skb)) { 275220bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 275320bf50deSTom Herbert goto do_frag_list; 275420bf50deSTom Herbert } 275520bf50deSTom Herbert } else if (skb->next) { 275620bf50deSTom Herbert skb = skb->next; 275720bf50deSTom Herbert goto do_frag_list; 275820bf50deSTom Herbert } 275920bf50deSTom Herbert } 276020bf50deSTom Herbert 276120bf50deSTom Herbert out: 276220bf50deSTom Herbert return orig_len - len; 276320bf50deSTom Herbert 276420bf50deSTom Herbert error: 276520bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 276620bf50deSTom Herbert } 27670739cd28SCong Wang 27680739cd28SCong Wang /* Send skb data on a socket. Socket must be locked. */ 27690739cd28SCong Wang int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 27700739cd28SCong Wang int len) 27710739cd28SCong Wang { 27720739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked, 27730739cd28SCong Wang kernel_sendpage_locked); 27740739cd28SCong Wang } 277520bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 277620bf50deSTom Herbert 27770739cd28SCong Wang /* Send skb data on a socket. Socket must be unlocked. */ 27780739cd28SCong Wang int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 27790739cd28SCong Wang { 27800739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked, 27810739cd28SCong Wang sendpage_unlocked); 27820739cd28SCong Wang } 27830739cd28SCong Wang 2784357b40a1SHerbert Xu /** 2785357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2786357b40a1SHerbert Xu * @skb: destination buffer 2787357b40a1SHerbert Xu * @offset: offset in destination 2788357b40a1SHerbert Xu * @from: source buffer 2789357b40a1SHerbert Xu * @len: number of bytes to copy 2790357b40a1SHerbert Xu * 2791357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2792357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2793357b40a1SHerbert Xu * traversing fragment lists and such. 2794357b40a1SHerbert Xu */ 2795357b40a1SHerbert Xu 27960c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2797357b40a1SHerbert Xu { 27981a028e50SDavid S. Miller int start = skb_headlen(skb); 2799fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2800fbb398a8SDavid S. Miller int i, copy; 2801357b40a1SHerbert Xu 2802357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2803357b40a1SHerbert Xu goto fault; 2804357b40a1SHerbert Xu 28051a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2806357b40a1SHerbert Xu if (copy > len) 2807357b40a1SHerbert Xu copy = len; 280827d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2809357b40a1SHerbert Xu if ((len -= copy) == 0) 2810357b40a1SHerbert Xu return 0; 2811357b40a1SHerbert Xu offset += copy; 2812357b40a1SHerbert Xu from += copy; 2813357b40a1SHerbert Xu } 2814357b40a1SHerbert Xu 2815357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2816357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 28171a028e50SDavid S. Miller int end; 2818357b40a1SHerbert Xu 2819547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28201a028e50SDavid S. Miller 28219e903e08SEric Dumazet end = start + skb_frag_size(frag); 2822357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2823c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2824c613c209SWillem de Bruijn struct page *p; 2825357b40a1SHerbert Xu u8 *vaddr; 2826357b40a1SHerbert Xu 2827357b40a1SHerbert Xu if (copy > len) 2828357b40a1SHerbert Xu copy = len; 2829357b40a1SHerbert Xu 2830c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2831b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2832c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2833c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2834c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 283551c56b00SEric Dumazet kunmap_atomic(vaddr); 2836c613c209SWillem de Bruijn } 2837357b40a1SHerbert Xu 2838357b40a1SHerbert Xu if ((len -= copy) == 0) 2839357b40a1SHerbert Xu return 0; 2840357b40a1SHerbert Xu offset += copy; 2841357b40a1SHerbert Xu from += copy; 2842357b40a1SHerbert Xu } 28431a028e50SDavid S. Miller start = end; 2844357b40a1SHerbert Xu } 2845357b40a1SHerbert Xu 2846fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 28471a028e50SDavid S. Miller int end; 2848357b40a1SHerbert Xu 2849547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28501a028e50SDavid S. Miller 2851fbb398a8SDavid S. Miller end = start + frag_iter->len; 2852357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2853357b40a1SHerbert Xu if (copy > len) 2854357b40a1SHerbert Xu copy = len; 2855fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 28561a028e50SDavid S. Miller from, copy)) 2857357b40a1SHerbert Xu goto fault; 2858357b40a1SHerbert Xu if ((len -= copy) == 0) 2859357b40a1SHerbert Xu return 0; 2860357b40a1SHerbert Xu offset += copy; 2861357b40a1SHerbert Xu from += copy; 2862357b40a1SHerbert Xu } 28631a028e50SDavid S. Miller start = end; 2864357b40a1SHerbert Xu } 2865357b40a1SHerbert Xu if (!len) 2866357b40a1SHerbert Xu return 0; 2867357b40a1SHerbert Xu 2868357b40a1SHerbert Xu fault: 2869357b40a1SHerbert Xu return -EFAULT; 2870357b40a1SHerbert Xu } 2871357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2872357b40a1SHerbert Xu 28731da177e4SLinus Torvalds /* Checksum skb data. */ 28742817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 28752817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 28761da177e4SLinus Torvalds { 28771a028e50SDavid S. Miller int start = skb_headlen(skb); 28781a028e50SDavid S. Miller int i, copy = start - offset; 2879fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 28801da177e4SLinus Torvalds int pos = 0; 28811da177e4SLinus Torvalds 28821da177e4SLinus Torvalds /* Checksum header. */ 28831da177e4SLinus Torvalds if (copy > 0) { 28841da177e4SLinus Torvalds if (copy > len) 28851da177e4SLinus Torvalds copy = len; 28862544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 28872544af03SMatteo Croce skb->data + offset, copy, csum); 28881da177e4SLinus Torvalds if ((len -= copy) == 0) 28891da177e4SLinus Torvalds return csum; 28901da177e4SLinus Torvalds offset += copy; 28911da177e4SLinus Torvalds pos = copy; 28921da177e4SLinus Torvalds } 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 28951a028e50SDavid S. Miller int end; 289651c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 28971da177e4SLinus Torvalds 2898547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28991a028e50SDavid S. Miller 290051c56b00SEric Dumazet end = start + skb_frag_size(frag); 29011da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2902c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2903c613c209SWillem de Bruijn struct page *p; 290444bb9363SAl Viro __wsum csum2; 29051da177e4SLinus Torvalds u8 *vaddr; 29061da177e4SLinus Torvalds 29071da177e4SLinus Torvalds if (copy > len) 29081da177e4SLinus Torvalds copy = len; 2909c613c209SWillem de Bruijn 2910c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2911b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2912c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2913c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 29142544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 29152544af03SMatteo Croce csum_partial_ext, 29162544af03SMatteo Croce vaddr + p_off, p_len, 0); 291751c56b00SEric Dumazet kunmap_atomic(vaddr); 29182544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 29192544af03SMatteo Croce csum_block_add_ext, csum, 29202544af03SMatteo Croce csum2, pos, p_len); 2921c613c209SWillem de Bruijn pos += p_len; 2922c613c209SWillem de Bruijn } 2923c613c209SWillem de Bruijn 29241da177e4SLinus Torvalds if (!(len -= copy)) 29251da177e4SLinus Torvalds return csum; 29261da177e4SLinus Torvalds offset += copy; 29271da177e4SLinus Torvalds } 29281a028e50SDavid S. Miller start = end; 29291da177e4SLinus Torvalds } 29301da177e4SLinus Torvalds 2931fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 29321a028e50SDavid S. Miller int end; 29331da177e4SLinus Torvalds 2934547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29351a028e50SDavid S. Miller 2936fbb398a8SDavid S. Miller end = start + frag_iter->len; 29371da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 29385f92a738SAl Viro __wsum csum2; 29391da177e4SLinus Torvalds if (copy > len) 29401da177e4SLinus Torvalds copy = len; 29412817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 29422817a336SDaniel Borkmann copy, 0, ops); 29432544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 29442544af03SMatteo Croce csum, csum2, pos, copy); 29451da177e4SLinus Torvalds if ((len -= copy) == 0) 29461da177e4SLinus Torvalds return csum; 29471da177e4SLinus Torvalds offset += copy; 29481da177e4SLinus Torvalds pos += copy; 29491da177e4SLinus Torvalds } 29501a028e50SDavid S. Miller start = end; 29511da177e4SLinus Torvalds } 295209a62660SKris Katterjohn BUG_ON(len); 29531da177e4SLinus Torvalds 29541da177e4SLinus Torvalds return csum; 29551da177e4SLinus Torvalds } 29562817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 29572817a336SDaniel Borkmann 29582817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 29592817a336SDaniel Borkmann int len, __wsum csum) 29602817a336SDaniel Borkmann { 29612817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2962cea80ea8SDaniel Borkmann .update = csum_partial_ext, 29632817a336SDaniel Borkmann .combine = csum_block_add_ext, 29642817a336SDaniel Borkmann }; 29652817a336SDaniel Borkmann 29662817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 29672817a336SDaniel Borkmann } 2968b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 29691da177e4SLinus Torvalds 29701da177e4SLinus Torvalds /* Both of above in one bottle. */ 29711da177e4SLinus Torvalds 297281d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 29738d5930dfSAl Viro u8 *to, int len) 29741da177e4SLinus Torvalds { 29751a028e50SDavid S. Miller int start = skb_headlen(skb); 29761a028e50SDavid S. Miller int i, copy = start - offset; 2977fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 29781da177e4SLinus Torvalds int pos = 0; 29798d5930dfSAl Viro __wsum csum = 0; 29801da177e4SLinus Torvalds 29811da177e4SLinus Torvalds /* Copy header. */ 29821da177e4SLinus Torvalds if (copy > 0) { 29831da177e4SLinus Torvalds if (copy > len) 29841da177e4SLinus Torvalds copy = len; 29851da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2986cc44c17bSAl Viro copy); 29871da177e4SLinus Torvalds if ((len -= copy) == 0) 29881da177e4SLinus Torvalds return csum; 29891da177e4SLinus Torvalds offset += copy; 29901da177e4SLinus Torvalds to += copy; 29911da177e4SLinus Torvalds pos = copy; 29921da177e4SLinus Torvalds } 29931da177e4SLinus Torvalds 29941da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 29951a028e50SDavid S. Miller int end; 29961da177e4SLinus Torvalds 2997547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29981a028e50SDavid S. Miller 29999e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 30001da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 3001c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3002c613c209SWillem de Bruijn u32 p_off, p_len, copied; 3003c613c209SWillem de Bruijn struct page *p; 30045084205fSAl Viro __wsum csum2; 30051da177e4SLinus Torvalds u8 *vaddr; 30061da177e4SLinus Torvalds 30071da177e4SLinus Torvalds if (copy > len) 30081da177e4SLinus Torvalds copy = len; 3009c613c209SWillem de Bruijn 3010c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 3011b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 3012c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 3013c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 3014c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 3015c613c209SWillem de Bruijn to + copied, 3016cc44c17bSAl Viro p_len); 301751c56b00SEric Dumazet kunmap_atomic(vaddr); 30181da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 3019c613c209SWillem de Bruijn pos += p_len; 3020c613c209SWillem de Bruijn } 3021c613c209SWillem de Bruijn 30221da177e4SLinus Torvalds if (!(len -= copy)) 30231da177e4SLinus Torvalds return csum; 30241da177e4SLinus Torvalds offset += copy; 30251da177e4SLinus Torvalds to += copy; 30261da177e4SLinus Torvalds } 30271a028e50SDavid S. Miller start = end; 30281da177e4SLinus Torvalds } 30291da177e4SLinus Torvalds 3030fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 303181d77662SAl Viro __wsum csum2; 30321a028e50SDavid S. Miller int end; 30331da177e4SLinus Torvalds 3034547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30351a028e50SDavid S. Miller 3036fbb398a8SDavid S. Miller end = start + frag_iter->len; 30371da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 30381da177e4SLinus Torvalds if (copy > len) 30391da177e4SLinus Torvalds copy = len; 3040fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 30411a028e50SDavid S. Miller offset - start, 30428d5930dfSAl Viro to, copy); 30431da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 30441da177e4SLinus Torvalds if ((len -= copy) == 0) 30451da177e4SLinus Torvalds return csum; 30461da177e4SLinus Torvalds offset += copy; 30471da177e4SLinus Torvalds to += copy; 30481da177e4SLinus Torvalds pos += copy; 30491da177e4SLinus Torvalds } 30501a028e50SDavid S. Miller start = end; 30511da177e4SLinus Torvalds } 305209a62660SKris Katterjohn BUG_ON(len); 30531da177e4SLinus Torvalds return csum; 30541da177e4SLinus Torvalds } 3055b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 30561da177e4SLinus Torvalds 305749f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 305849f8e832SCong Wang { 305949f8e832SCong Wang __sum16 sum; 306049f8e832SCong Wang 306149f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 306214641931SCong Wang /* See comments in __skb_checksum_complete(). */ 306349f8e832SCong Wang if (likely(!sum)) { 306449f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 306549f8e832SCong Wang !skb->csum_complete_sw) 30667fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 306749f8e832SCong Wang } 306849f8e832SCong Wang if (!skb_shared(skb)) 306949f8e832SCong Wang skb->csum_valid = !sum; 307049f8e832SCong Wang return sum; 307149f8e832SCong Wang } 307249f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 307349f8e832SCong Wang 307414641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 307514641931SCong Wang * which has been changed from the hardware checksum, for example, by 307614641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 307714641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 307814641931SCong Wang * 307914641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 308014641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 308114641931SCong Wang * shared. 308214641931SCong Wang */ 308349f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 308449f8e832SCong Wang { 308549f8e832SCong Wang __wsum csum; 308649f8e832SCong Wang __sum16 sum; 308749f8e832SCong Wang 308849f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 308949f8e832SCong Wang 309049f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 309114641931SCong Wang /* This check is inverted, because we already knew the hardware 309214641931SCong Wang * checksum is invalid before calling this function. So, if the 309314641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 309414641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 309514641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 309614641931SCong Wang * when moving skb->data around. 309714641931SCong Wang */ 309849f8e832SCong Wang if (likely(!sum)) { 309949f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 310049f8e832SCong Wang !skb->csum_complete_sw) 31017fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 310249f8e832SCong Wang } 310349f8e832SCong Wang 310449f8e832SCong Wang if (!skb_shared(skb)) { 310549f8e832SCong Wang /* Save full packet checksum */ 310649f8e832SCong Wang skb->csum = csum; 310749f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 310849f8e832SCong Wang skb->csum_complete_sw = 1; 310949f8e832SCong Wang skb->csum_valid = !sum; 311049f8e832SCong Wang } 311149f8e832SCong Wang 311249f8e832SCong Wang return sum; 311349f8e832SCong Wang } 311449f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 311549f8e832SCong Wang 31169617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 31179617813dSDavide Caratti { 31189617813dSDavide Caratti net_warn_ratelimited( 31199617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 31209617813dSDavide Caratti __func__); 31219617813dSDavide Caratti return 0; 31229617813dSDavide Caratti } 31239617813dSDavide Caratti 31249617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 31259617813dSDavide Caratti int offset, int len) 31269617813dSDavide Caratti { 31279617813dSDavide Caratti net_warn_ratelimited( 31289617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 31299617813dSDavide Caratti __func__); 31309617813dSDavide Caratti return 0; 31319617813dSDavide Caratti } 31329617813dSDavide Caratti 31339617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 31349617813dSDavide Caratti .update = warn_crc32c_csum_update, 31359617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 31369617813dSDavide Caratti }; 31379617813dSDavide Caratti 31389617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 31399617813dSDavide Caratti &default_crc32c_ops; 31409617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 31419617813dSDavide Caratti 3142af2806f8SThomas Graf /** 3143af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 3144af2806f8SThomas Graf * @from: source buffer 3145af2806f8SThomas Graf * 3146af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 3147af2806f8SThomas Graf * into skb_zerocopy(). 3148af2806f8SThomas Graf */ 3149af2806f8SThomas Graf unsigned int 3150af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 3151af2806f8SThomas Graf { 3152af2806f8SThomas Graf unsigned int hlen = 0; 3153af2806f8SThomas Graf 3154af2806f8SThomas Graf if (!from->head_frag || 3155af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 3156a17ad096SPravin B Shelar skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) { 3157af2806f8SThomas Graf hlen = skb_headlen(from); 3158a17ad096SPravin B Shelar if (!hlen) 3159a17ad096SPravin B Shelar hlen = from->len; 3160a17ad096SPravin B Shelar } 3161af2806f8SThomas Graf 3162af2806f8SThomas Graf if (skb_has_frag_list(from)) 3163af2806f8SThomas Graf hlen = from->len; 3164af2806f8SThomas Graf 3165af2806f8SThomas Graf return hlen; 3166af2806f8SThomas Graf } 3167af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 3168af2806f8SThomas Graf 3169af2806f8SThomas Graf /** 3170af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 3171af2806f8SThomas Graf * @to: destination buffer 31727fceb4deSMasanari Iida * @from: source buffer 3173af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 3174af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 3175af2806f8SThomas Graf * 3176af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 3177af2806f8SThomas Graf * to the frags in the source buffer. 3178af2806f8SThomas Graf * 3179af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 3180af2806f8SThomas Graf * headroom in the `to` buffer. 318136d5fe6aSZoltan Kiss * 318236d5fe6aSZoltan Kiss * Return value: 318336d5fe6aSZoltan Kiss * 0: everything is OK 318436d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 318536d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 3186af2806f8SThomas Graf */ 318736d5fe6aSZoltan Kiss int 318836d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 3189af2806f8SThomas Graf { 3190af2806f8SThomas Graf int i, j = 0; 3191af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 319236d5fe6aSZoltan Kiss int ret; 3193af2806f8SThomas Graf struct page *page; 3194af2806f8SThomas Graf unsigned int offset; 3195af2806f8SThomas Graf 3196af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 3197af2806f8SThomas Graf 3198af2806f8SThomas Graf /* dont bother with small payloads */ 319936d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 320036d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 3201af2806f8SThomas Graf 3202af2806f8SThomas Graf if (hlen) { 320336d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 320436d5fe6aSZoltan Kiss if (unlikely(ret)) 320536d5fe6aSZoltan Kiss return ret; 3206af2806f8SThomas Graf len -= hlen; 3207af2806f8SThomas Graf } else { 3208af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 3209af2806f8SThomas Graf if (plen) { 3210af2806f8SThomas Graf page = virt_to_head_page(from->head); 3211af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 3212af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 3213af2806f8SThomas Graf get_page(page); 3214af2806f8SThomas Graf j = 1; 3215af2806f8SThomas Graf len -= plen; 3216af2806f8SThomas Graf } 3217af2806f8SThomas Graf } 3218af2806f8SThomas Graf 3219af2806f8SThomas Graf to->truesize += len + plen; 3220af2806f8SThomas Graf to->len += len + plen; 3221af2806f8SThomas Graf to->data_len += len + plen; 3222af2806f8SThomas Graf 322336d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 322436d5fe6aSZoltan Kiss skb_tx_error(from); 322536d5fe6aSZoltan Kiss return -ENOMEM; 322636d5fe6aSZoltan Kiss } 32271f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 322836d5fe6aSZoltan Kiss 3229af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 3230d8e18a51SMatthew Wilcox (Oracle) int size; 3231d8e18a51SMatthew Wilcox (Oracle) 3232af2806f8SThomas Graf if (!len) 3233af2806f8SThomas Graf break; 3234af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 3235d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 3236d8e18a51SMatthew Wilcox (Oracle) len); 3237d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3238d8e18a51SMatthew Wilcox (Oracle) len -= size; 3239af2806f8SThomas Graf skb_frag_ref(to, j); 3240af2806f8SThomas Graf j++; 3241af2806f8SThomas Graf } 3242af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 324336d5fe6aSZoltan Kiss 324436d5fe6aSZoltan Kiss return 0; 3245af2806f8SThomas Graf } 3246af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3247af2806f8SThomas Graf 32481da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 32491da177e4SLinus Torvalds { 3250d3bc23e7SAl Viro __wsum csum; 32511da177e4SLinus Torvalds long csstart; 32521da177e4SLinus Torvalds 325384fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 325455508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 32551da177e4SLinus Torvalds else 32561da177e4SLinus Torvalds csstart = skb_headlen(skb); 32571da177e4SLinus Torvalds 325809a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 32591da177e4SLinus Torvalds 3260d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 32611da177e4SLinus Torvalds 32621da177e4SLinus Torvalds csum = 0; 32631da177e4SLinus Torvalds if (csstart != skb->len) 32641da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 32658d5930dfSAl Viro skb->len - csstart); 32661da177e4SLinus Torvalds 326784fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3268ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 32691da177e4SLinus Torvalds 3270d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 32711da177e4SLinus Torvalds } 32721da177e4SLinus Torvalds } 3273b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 32741da177e4SLinus Torvalds 32751da177e4SLinus Torvalds /** 32761da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 32771da177e4SLinus Torvalds * @list: list to dequeue from 32781da177e4SLinus Torvalds * 32791da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 32801da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 32811da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32821da177e4SLinus Torvalds */ 32831da177e4SLinus Torvalds 32841da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 32851da177e4SLinus Torvalds { 32861da177e4SLinus Torvalds unsigned long flags; 32871da177e4SLinus Torvalds struct sk_buff *result; 32881da177e4SLinus Torvalds 32891da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32901da177e4SLinus Torvalds result = __skb_dequeue(list); 32911da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32921da177e4SLinus Torvalds return result; 32931da177e4SLinus Torvalds } 3294b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 32951da177e4SLinus Torvalds 32961da177e4SLinus Torvalds /** 32971da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 32981da177e4SLinus Torvalds * @list: list to dequeue from 32991da177e4SLinus Torvalds * 33001da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 33011da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 33021da177e4SLinus Torvalds * returned or %NULL if the list is empty. 33031da177e4SLinus Torvalds */ 33041da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 33051da177e4SLinus Torvalds { 33061da177e4SLinus Torvalds unsigned long flags; 33071da177e4SLinus Torvalds struct sk_buff *result; 33081da177e4SLinus Torvalds 33091da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33101da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 33111da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33121da177e4SLinus Torvalds return result; 33131da177e4SLinus Torvalds } 3314b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 33151da177e4SLinus Torvalds 33161da177e4SLinus Torvalds /** 33171da177e4SLinus Torvalds * skb_queue_purge - empty a list 33181da177e4SLinus Torvalds * @list: list to empty 33191da177e4SLinus Torvalds * 33201da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 33211da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 33221da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 33231da177e4SLinus Torvalds */ 33241da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 33251da177e4SLinus Torvalds { 33261da177e4SLinus Torvalds struct sk_buff *skb; 33271da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 33281da177e4SLinus Torvalds kfree_skb(skb); 33291da177e4SLinus Torvalds } 3330b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 33311da177e4SLinus Torvalds 33321da177e4SLinus Torvalds /** 33339f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 33349f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3335385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 33369f5afeaeSYaogong Wang * 33379f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 33389f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 33399f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 33409f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 33419f5afeaeSYaogong Wang */ 3342385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 33439f5afeaeSYaogong Wang { 33447c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3345385114deSPeter Oskolkov unsigned int sum = 0; 33469f5afeaeSYaogong Wang 33477c90584cSEric Dumazet while (p) { 33487c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 33497c90584cSEric Dumazet 33507c90584cSEric Dumazet p = rb_next(p); 33517c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3352385114deSPeter Oskolkov sum += skb->truesize; 33539f5afeaeSYaogong Wang kfree_skb(skb); 33547c90584cSEric Dumazet } 3355385114deSPeter Oskolkov return sum; 33569f5afeaeSYaogong Wang } 33579f5afeaeSYaogong Wang 33589f5afeaeSYaogong Wang /** 33591da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 33601da177e4SLinus Torvalds * @list: list to use 33611da177e4SLinus Torvalds * @newsk: buffer to queue 33621da177e4SLinus Torvalds * 33631da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 33641da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33651da177e4SLinus Torvalds * safely. 33661da177e4SLinus Torvalds * 33671da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33681da177e4SLinus Torvalds */ 33691da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 33701da177e4SLinus Torvalds { 33711da177e4SLinus Torvalds unsigned long flags; 33721da177e4SLinus Torvalds 33731da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33741da177e4SLinus Torvalds __skb_queue_head(list, newsk); 33751da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33761da177e4SLinus Torvalds } 3377b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 33781da177e4SLinus Torvalds 33791da177e4SLinus Torvalds /** 33801da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 33811da177e4SLinus Torvalds * @list: list to use 33821da177e4SLinus Torvalds * @newsk: buffer to queue 33831da177e4SLinus Torvalds * 33841da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 33851da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33861da177e4SLinus Torvalds * safely. 33871da177e4SLinus Torvalds * 33881da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33891da177e4SLinus Torvalds */ 33901da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 33911da177e4SLinus Torvalds { 33921da177e4SLinus Torvalds unsigned long flags; 33931da177e4SLinus Torvalds 33941da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33951da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 33961da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33971da177e4SLinus Torvalds } 3398b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 33998728b834SDavid S. Miller 34001da177e4SLinus Torvalds /** 34011da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 34021da177e4SLinus Torvalds * @skb: buffer to remove 34038728b834SDavid S. Miller * @list: list to use 34041da177e4SLinus Torvalds * 34058728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 34068728b834SDavid S. Miller * function is atomic with respect to other list locked calls 34071da177e4SLinus Torvalds * 34088728b834SDavid S. Miller * You must know what list the SKB is on. 34091da177e4SLinus Torvalds */ 34108728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 34111da177e4SLinus Torvalds { 34121da177e4SLinus Torvalds unsigned long flags; 34131da177e4SLinus Torvalds 34141da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 34158728b834SDavid S. Miller __skb_unlink(skb, list); 34161da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 34171da177e4SLinus Torvalds } 3418b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 34191da177e4SLinus Torvalds 34201da177e4SLinus Torvalds /** 34211da177e4SLinus Torvalds * skb_append - append a buffer 34221da177e4SLinus Torvalds * @old: buffer to insert after 34231da177e4SLinus Torvalds * @newsk: buffer to insert 34248728b834SDavid S. Miller * @list: list to use 34251da177e4SLinus Torvalds * 34261da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 34271da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 34281da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 34291da177e4SLinus Torvalds */ 34308728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 34311da177e4SLinus Torvalds { 34321da177e4SLinus Torvalds unsigned long flags; 34331da177e4SLinus Torvalds 34348728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 34357de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 34368728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 34371da177e4SLinus Torvalds } 3438b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 34391da177e4SLinus Torvalds 34401da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 34411da177e4SLinus Torvalds struct sk_buff* skb1, 34421da177e4SLinus Torvalds const u32 len, const int pos) 34431da177e4SLinus Torvalds { 34441da177e4SLinus Torvalds int i; 34451da177e4SLinus Torvalds 3446d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3447d626f62bSArnaldo Carvalho de Melo pos - len); 34481da177e4SLinus Torvalds /* And move data appendix as is. */ 34491da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 34501da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 34511da177e4SLinus Torvalds 34521da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 34531da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 34541da177e4SLinus Torvalds skb1->data_len = skb->data_len; 34551da177e4SLinus Torvalds skb1->len += skb1->data_len; 34561da177e4SLinus Torvalds skb->data_len = 0; 34571da177e4SLinus Torvalds skb->len = len; 345827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 34591da177e4SLinus Torvalds } 34601da177e4SLinus Torvalds 34611da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 34621da177e4SLinus Torvalds struct sk_buff* skb1, 34631da177e4SLinus Torvalds const u32 len, int pos) 34641da177e4SLinus Torvalds { 34651da177e4SLinus Torvalds int i, k = 0; 34661da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 34671da177e4SLinus Torvalds 34681da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 34691da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 34701da177e4SLinus Torvalds skb->len = len; 34711da177e4SLinus Torvalds skb->data_len = len - pos; 34721da177e4SLinus Torvalds 34731da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 34749e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 34751da177e4SLinus Torvalds 34761da177e4SLinus Torvalds if (pos + size > len) { 34771da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 34781da177e4SLinus Torvalds 34791da177e4SLinus Torvalds if (pos < len) { 34801da177e4SLinus Torvalds /* Split frag. 34811da177e4SLinus Torvalds * We have two variants in this case: 34821da177e4SLinus Torvalds * 1. Move all the frag to the second 34831da177e4SLinus Torvalds * part, if it is possible. F.e. 34841da177e4SLinus Torvalds * this approach is mandatory for TUX, 34851da177e4SLinus Torvalds * where splitting is expensive. 34861da177e4SLinus Torvalds * 2. Split is accurately. We make this. 34871da177e4SLinus Torvalds */ 3488ea2ab693SIan Campbell skb_frag_ref(skb, i); 3489b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 34909e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 34919e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 34921da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34931da177e4SLinus Torvalds } 34941da177e4SLinus Torvalds k++; 34951da177e4SLinus Torvalds } else 34961da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34971da177e4SLinus Torvalds pos += size; 34981da177e4SLinus Torvalds } 34991da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 35001da177e4SLinus Torvalds } 35011da177e4SLinus Torvalds 35021da177e4SLinus Torvalds /** 35031da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 35041da177e4SLinus Torvalds * @skb: the buffer to split 35051da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 35061da177e4SLinus Torvalds * @len: new length for skb 35071da177e4SLinus Torvalds */ 35081da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 35091da177e4SLinus Torvalds { 35101da177e4SLinus Torvalds int pos = skb_headlen(skb); 35119b65b17dSTalal Ahmad const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY; 35121da177e4SLinus Torvalds 3513*753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 3514*753f1ca4SPavel Begunkov 35159b65b17dSTalal Ahmad skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags; 35161f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 35171da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 35181da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 35191da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 35201da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 35211da177e4SLinus Torvalds } 3522b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 35231da177e4SLinus Torvalds 35249f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 35259f782db3SIlpo Järvinen * 35269f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 35279f782db3SIlpo Järvinen */ 3528832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3529832d11c5SIlpo Järvinen { 3530c4777efaSEric Dumazet return skb_unclone_keeptruesize(skb, GFP_ATOMIC); 3531832d11c5SIlpo Järvinen } 3532832d11c5SIlpo Järvinen 3533832d11c5SIlpo Järvinen /** 3534832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3535832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3536832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3537832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3538832d11c5SIlpo Järvinen * 3539832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 354020e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3541832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3542832d11c5SIlpo Järvinen * 3543832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3544832d11c5SIlpo Järvinen * 3545832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3546832d11c5SIlpo Järvinen * to have non-paged data as well. 3547832d11c5SIlpo Järvinen * 3548832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3549832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3550832d11c5SIlpo Järvinen */ 3551832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3552832d11c5SIlpo Järvinen { 3553832d11c5SIlpo Järvinen int from, to, merge, todo; 3554d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3555832d11c5SIlpo Järvinen 3556832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3557f8071cdeSEric Dumazet 3558f8071cdeSEric Dumazet if (skb_headlen(skb)) 3559f8071cdeSEric Dumazet return 0; 35601f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 35611f8b977aSWillem de Bruijn return 0; 3562832d11c5SIlpo Järvinen 3563832d11c5SIlpo Järvinen todo = shiftlen; 3564832d11c5SIlpo Järvinen from = 0; 3565832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3566832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3567832d11c5SIlpo Järvinen 3568832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3569832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3570832d11c5SIlpo Järvinen */ 3571832d11c5SIlpo Järvinen if (!to || 3572ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3573b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3574832d11c5SIlpo Järvinen merge = -1; 3575832d11c5SIlpo Järvinen } else { 3576832d11c5SIlpo Järvinen merge = to - 1; 3577832d11c5SIlpo Järvinen 35789e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3579832d11c5SIlpo Järvinen if (todo < 0) { 3580832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3581832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3582832d11c5SIlpo Järvinen return 0; 3583832d11c5SIlpo Järvinen 35849f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 35859f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3586832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3587832d11c5SIlpo Järvinen 35889e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 35899e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3590b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3591832d11c5SIlpo Järvinen 3592832d11c5SIlpo Järvinen goto onlymerged; 3593832d11c5SIlpo Järvinen } 3594832d11c5SIlpo Järvinen 3595832d11c5SIlpo Järvinen from++; 3596832d11c5SIlpo Järvinen } 3597832d11c5SIlpo Järvinen 3598832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3599832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3600832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3601832d11c5SIlpo Järvinen return 0; 3602832d11c5SIlpo Järvinen 3603832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3604832d11c5SIlpo Järvinen return 0; 3605832d11c5SIlpo Järvinen 3606832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3607832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3608832d11c5SIlpo Järvinen return 0; 3609832d11c5SIlpo Järvinen 3610832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3611832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3612832d11c5SIlpo Järvinen 36139e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3614832d11c5SIlpo Järvinen *fragto = *fragfrom; 36159e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3616832d11c5SIlpo Järvinen from++; 3617832d11c5SIlpo Järvinen to++; 3618832d11c5SIlpo Järvinen 3619832d11c5SIlpo Järvinen } else { 3620ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3621b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3622b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 36239e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3624832d11c5SIlpo Järvinen 3625b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 36269e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3627832d11c5SIlpo Järvinen todo = 0; 3628832d11c5SIlpo Järvinen 3629832d11c5SIlpo Järvinen to++; 3630832d11c5SIlpo Järvinen break; 3631832d11c5SIlpo Järvinen } 3632832d11c5SIlpo Järvinen } 3633832d11c5SIlpo Järvinen 3634832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3635832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3636832d11c5SIlpo Järvinen 3637832d11c5SIlpo Järvinen if (merge >= 0) { 3638832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3639832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3640832d11c5SIlpo Järvinen 36419e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 36426a5bcd84SIlias Apalodimas __skb_frag_unref(fragfrom, skb->pp_recycle); 3643832d11c5SIlpo Järvinen } 3644832d11c5SIlpo Järvinen 3645832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3646832d11c5SIlpo Järvinen to = 0; 3647832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3648832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3649832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3650832d11c5SIlpo Järvinen 3651832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3652832d11c5SIlpo Järvinen 3653832d11c5SIlpo Järvinen onlymerged: 3654832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3655832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3656832d11c5SIlpo Järvinen */ 3657832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3658832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3659832d11c5SIlpo Järvinen 3660832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3661832d11c5SIlpo Järvinen skb->len -= shiftlen; 3662832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3663832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3664832d11c5SIlpo Järvinen tgt->len += shiftlen; 3665832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3666832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3667832d11c5SIlpo Järvinen 3668832d11c5SIlpo Järvinen return shiftlen; 3669832d11c5SIlpo Järvinen } 3670832d11c5SIlpo Järvinen 3671677e90edSThomas Graf /** 3672677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3673677e90edSThomas Graf * @skb: the buffer to read 3674677e90edSThomas Graf * @from: lower offset of data to be read 3675677e90edSThomas Graf * @to: upper offset of data to be read 3676677e90edSThomas Graf * @st: state variable 3677677e90edSThomas Graf * 3678677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3679677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3680677e90edSThomas Graf */ 3681677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3682677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3683677e90edSThomas Graf { 3684677e90edSThomas Graf st->lower_offset = from; 3685677e90edSThomas Graf st->upper_offset = to; 3686677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3687677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3688677e90edSThomas Graf st->frag_data = NULL; 368997550f6fSWillem de Bruijn st->frag_off = 0; 3690677e90edSThomas Graf } 3691b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3692677e90edSThomas Graf 3693677e90edSThomas Graf /** 3694677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3695677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3696677e90edSThomas Graf * @data: destination pointer for data to be returned 3697677e90edSThomas Graf * @st: state variable 3698677e90edSThomas Graf * 3699bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3700677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3701bc32383cSMathias Krause * the head of the data block to @data and returns the length 3702677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3703677e90edSThomas Graf * offset has been reached. 3704677e90edSThomas Graf * 3705677e90edSThomas Graf * The caller is not required to consume all of the data 3706bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3707677e90edSThomas Graf * of bytes already consumed and the next call to 3708677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3709677e90edSThomas Graf * 371025985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3711e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3712677e90edSThomas Graf * reads of potentially non linear data. 3713677e90edSThomas Graf * 3714bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3715677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3716677e90edSThomas Graf * a stack for this purpose. 3717677e90edSThomas Graf */ 3718677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3719677e90edSThomas Graf struct skb_seq_state *st) 3720677e90edSThomas Graf { 3721677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3722677e90edSThomas Graf skb_frag_t *frag; 3723677e90edSThomas Graf 3724aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3725aeb193eaSWedson Almeida Filho if (st->frag_data) { 3726aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3727aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3728aeb193eaSWedson Almeida Filho } 3729677e90edSThomas Graf return 0; 3730aeb193eaSWedson Almeida Filho } 3731677e90edSThomas Graf 3732677e90edSThomas Graf next_skb: 373395e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3734677e90edSThomas Graf 3735995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 373695e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3737677e90edSThomas Graf return block_limit - abs_offset; 3738677e90edSThomas Graf } 3739677e90edSThomas Graf 3740677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3741677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3742677e90edSThomas Graf 3743677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 374497550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3745677e90edSThomas Graf 374697550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 374797550f6fSWillem de Bruijn 374897550f6fSWillem de Bruijn pg_idx = 0; 374997550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 375097550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 375197550f6fSWillem de Bruijn 375297550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 375397550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 375497550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 375597550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 375697550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 375797550f6fSWillem de Bruijn } 375897550f6fSWillem de Bruijn 375997550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3760677e90edSThomas Graf if (abs_offset < block_limit) { 3761677e90edSThomas Graf if (!st->frag_data) 376297550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3763677e90edSThomas Graf 376497550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3765677e90edSThomas Graf (abs_offset - st->stepped_offset); 3766677e90edSThomas Graf 3767677e90edSThomas Graf return block_limit - abs_offset; 3768677e90edSThomas Graf } 3769677e90edSThomas Graf 3770677e90edSThomas Graf if (st->frag_data) { 377151c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3772677e90edSThomas Graf st->frag_data = NULL; 3773677e90edSThomas Graf } 3774677e90edSThomas Graf 377597550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 377697550f6fSWillem de Bruijn st->frag_off += pg_sz; 377797550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 377897550f6fSWillem de Bruijn st->frag_off = 0; 3779677e90edSThomas Graf st->frag_idx++; 378097550f6fSWillem de Bruijn } 3781677e90edSThomas Graf } 3782677e90edSThomas Graf 37835b5a60daSOlaf Kirch if (st->frag_data) { 378451c56b00SEric Dumazet kunmap_atomic(st->frag_data); 37855b5a60daSOlaf Kirch st->frag_data = NULL; 37865b5a60daSOlaf Kirch } 37875b5a60daSOlaf Kirch 378821dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3789677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 379095e3b24cSHerbert Xu st->frag_idx = 0; 3791677e90edSThomas Graf goto next_skb; 379271b3346dSShyam Iyer } else if (st->cur_skb->next) { 379371b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 379471b3346dSShyam Iyer st->frag_idx = 0; 3795677e90edSThomas Graf goto next_skb; 3796677e90edSThomas Graf } 3797677e90edSThomas Graf 3798677e90edSThomas Graf return 0; 3799677e90edSThomas Graf } 3800b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3801677e90edSThomas Graf 3802677e90edSThomas Graf /** 3803677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3804677e90edSThomas Graf * @st: state variable 3805677e90edSThomas Graf * 3806677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3807677e90edSThomas Graf * returned 0. 3808677e90edSThomas Graf */ 3809677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3810677e90edSThomas Graf { 3811677e90edSThomas Graf if (st->frag_data) 381251c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3813677e90edSThomas Graf } 3814b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3815677e90edSThomas Graf 38163fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 38173fc7e8a6SThomas Graf 38183fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 38193fc7e8a6SThomas Graf struct ts_config *conf, 38203fc7e8a6SThomas Graf struct ts_state *state) 38213fc7e8a6SThomas Graf { 38223fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 38233fc7e8a6SThomas Graf } 38243fc7e8a6SThomas Graf 38253fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 38263fc7e8a6SThomas Graf { 38273fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 38283fc7e8a6SThomas Graf } 38293fc7e8a6SThomas Graf 38303fc7e8a6SThomas Graf /** 38313fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 38323fc7e8a6SThomas Graf * @skb: the buffer to look in 38333fc7e8a6SThomas Graf * @from: search offset 38343fc7e8a6SThomas Graf * @to: search limit 38353fc7e8a6SThomas Graf * @config: textsearch configuration 38363fc7e8a6SThomas Graf * 38373fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 38383fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 38393fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 38403fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 38413fc7e8a6SThomas Graf */ 38423fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3843059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 38443fc7e8a6SThomas Graf { 3845059a2440SBojan Prtvar struct ts_state state; 3846f72b948dSPhil Oester unsigned int ret; 3847f72b948dSPhil Oester 3848b228c9b0SWillem de Bruijn BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb)); 3849b228c9b0SWillem de Bruijn 38503fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 38513fc7e8a6SThomas Graf config->finish = skb_ts_finish; 38523fc7e8a6SThomas Graf 3853059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 38543fc7e8a6SThomas Graf 3855059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3856f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 38573fc7e8a6SThomas Graf } 3858b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 38593fc7e8a6SThomas Graf 3860be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3861be12a1feSHannes Frederic Sowa int offset, size_t size) 3862be12a1feSHannes Frederic Sowa { 3863be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3864be12a1feSHannes Frederic Sowa 3865be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3866be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3867be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3868*753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 3869be12a1feSHannes Frederic Sowa get_page(page); 3870be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3871be12a1feSHannes Frederic Sowa } else { 3872be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3873be12a1feSHannes Frederic Sowa } 3874be12a1feSHannes Frederic Sowa 3875be12a1feSHannes Frederic Sowa return 0; 3876be12a1feSHannes Frederic Sowa } 3877be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3878be12a1feSHannes Frederic Sowa 3879cbb042f9SHerbert Xu /** 3880cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3881cbb042f9SHerbert Xu * @skb: buffer to update 3882cbb042f9SHerbert Xu * @len: length of data pulled 3883cbb042f9SHerbert Xu * 3884cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3885fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 388684fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 388784fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 388884fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3889cbb042f9SHerbert Xu */ 3890af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3891cbb042f9SHerbert Xu { 389231b33dfbSPravin B Shelar unsigned char *data = skb->data; 389331b33dfbSPravin B Shelar 3894cbb042f9SHerbert Xu BUG_ON(len > skb->len); 389531b33dfbSPravin B Shelar __skb_pull(skb, len); 389631b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 389731b33dfbSPravin B Shelar return skb->data; 3898cbb042f9SHerbert Xu } 3899f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3900f94691acSArnaldo Carvalho de Melo 390113acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 390213acc94eSYonghong Song { 390313acc94eSYonghong Song skb_frag_t head_frag; 390413acc94eSYonghong Song struct page *page; 390513acc94eSYonghong Song 390613acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3907d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3908b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3909b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3910d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 391113acc94eSYonghong Song return head_frag; 391213acc94eSYonghong Song } 391313acc94eSYonghong Song 39143a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 39153a1296a3SSteffen Klassert netdev_features_t features, 39163a1296a3SSteffen Klassert unsigned int offset) 39173a1296a3SSteffen Klassert { 39183a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 39193a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 39203a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 39213a1296a3SSteffen Klassert unsigned int delta_len = 0; 39223a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 392353475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 3924cf3ab8d4SLina Wang int len_diff, err; 39253a1296a3SSteffen Klassert 39263a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 39273a1296a3SSteffen Klassert 39283a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 39293a1296a3SSteffen Klassert 39303a1296a3SSteffen Klassert do { 39313a1296a3SSteffen Klassert nskb = list_skb; 39323a1296a3SSteffen Klassert list_skb = list_skb->next; 39333a1296a3SSteffen Klassert 393453475c5dSDongseok Yi err = 0; 3935224102deSlena wang delta_truesize += nskb->truesize; 393653475c5dSDongseok Yi if (skb_shared(nskb)) { 393753475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 393853475c5dSDongseok Yi if (tmp) { 393953475c5dSDongseok Yi consume_skb(nskb); 394053475c5dSDongseok Yi nskb = tmp; 394153475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 394253475c5dSDongseok Yi } else { 394353475c5dSDongseok Yi err = -ENOMEM; 394453475c5dSDongseok Yi } 394553475c5dSDongseok Yi } 394653475c5dSDongseok Yi 39473a1296a3SSteffen Klassert if (!tail) 39483a1296a3SSteffen Klassert skb->next = nskb; 39493a1296a3SSteffen Klassert else 39503a1296a3SSteffen Klassert tail->next = nskb; 39513a1296a3SSteffen Klassert 395253475c5dSDongseok Yi if (unlikely(err)) { 395353475c5dSDongseok Yi nskb->next = list_skb; 395453475c5dSDongseok Yi goto err_linearize; 395553475c5dSDongseok Yi } 395653475c5dSDongseok Yi 39573a1296a3SSteffen Klassert tail = nskb; 39583a1296a3SSteffen Klassert 39593a1296a3SSteffen Klassert delta_len += nskb->len; 39603a1296a3SSteffen Klassert 39613a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 39623a1296a3SSteffen Klassert 3963cf673ed0SFlorian Westphal skb_release_head_state(nskb); 3964cf3ab8d4SLina Wang len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb); 39653a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 39663a1296a3SSteffen Klassert 39673a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 3968cf3ab8d4SLina Wang nskb->transport_header += len_diff; 39693a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 39703a1296a3SSteffen Klassert nskb->data - tnl_hlen, 39713a1296a3SSteffen Klassert offset + tnl_hlen); 39723a1296a3SSteffen Klassert 39733a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 39743a1296a3SSteffen Klassert __skb_linearize(nskb)) 39753a1296a3SSteffen Klassert goto err_linearize; 39763a1296a3SSteffen Klassert 39773a1296a3SSteffen Klassert } while (list_skb); 39783a1296a3SSteffen Klassert 39793a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 39803a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 39813a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 39823a1296a3SSteffen Klassert 39833a1296a3SSteffen Klassert skb_gso_reset(skb); 39843a1296a3SSteffen Klassert 39853a1296a3SSteffen Klassert skb->prev = tail; 39863a1296a3SSteffen Klassert 39873a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 39883a1296a3SSteffen Klassert __skb_linearize(skb)) 39893a1296a3SSteffen Klassert goto err_linearize; 39903a1296a3SSteffen Klassert 39913a1296a3SSteffen Klassert skb_get(skb); 39923a1296a3SSteffen Klassert 39933a1296a3SSteffen Klassert return skb; 39943a1296a3SSteffen Klassert 39953a1296a3SSteffen Klassert err_linearize: 39963a1296a3SSteffen Klassert kfree_skb_list(skb->next); 39973a1296a3SSteffen Klassert skb->next = NULL; 39983a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 39993a1296a3SSteffen Klassert } 40003a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 40013a1296a3SSteffen Klassert 4002f4c50d99SHerbert Xu /** 4003f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 4004df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 4005576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 4006f4c50d99SHerbert Xu * 4007f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 40084c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 40094c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 4010f4c50d99SHerbert Xu */ 4011df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 4012df5771ffSMichael S. Tsirkin netdev_features_t features) 4013f4c50d99SHerbert Xu { 4014f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 4015f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 40161a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 4017df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 4018df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 4019df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 40201fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 4021f4c50d99SHerbert Xu unsigned int offset = doffset; 4022df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 4023802ab55aSAlexander Duyck unsigned int partial_segs = 0; 4024f4c50d99SHerbert Xu unsigned int headroom; 4025802ab55aSAlexander Duyck unsigned int len = head_skb->len; 4026ec5f0615SPravin B Shelar __be16 proto; 402736c98382SAlexander Duyck bool csum, sg; 4028df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 4029f4c50d99SHerbert Xu int err = -ENOMEM; 4030f4c50d99SHerbert Xu int i = 0; 4031f4c50d99SHerbert Xu int pos; 4032f4c50d99SHerbert Xu 40333dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 40343dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 40353dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 40363dcbdb13SShmulik Ladkani * non head_frag head. 40373dcbdb13SShmulik Ladkani * 40383dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 40393dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 40403dcbdb13SShmulik Ladkani * head, then the first one has too). 40413dcbdb13SShmulik Ladkani * 40423dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 40433dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 40443dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 40453dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 40463dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 40473dcbdb13SShmulik Ladkani */ 40483dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 40493dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 40503dcbdb13SShmulik Ladkani } 40513dcbdb13SShmulik Ladkani 40525882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 40532f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 4054ec5f0615SPravin B Shelar if (unlikely(!proto)) 4055ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 4056ec5f0615SPravin B Shelar 405736c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 4058f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 40597e2b10c1STom Herbert 406007b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 406107b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 406207b26c94SSteffen Klassert struct sk_buff *iter; 406343170c4eSIlan Tayari unsigned int frag_len; 406407b26c94SSteffen Klassert 406507b26c94SSteffen Klassert if (!list_skb || 406607b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 406707b26c94SSteffen Klassert goto normal; 406807b26c94SSteffen Klassert 406943170c4eSIlan Tayari /* If we get here then all the required 407043170c4eSIlan Tayari * GSO features except frag_list are supported. 407143170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 407243170c4eSIlan Tayari * with no frag_list. 407343170c4eSIlan Tayari * Currently we can do that only when the buffers don't 407443170c4eSIlan Tayari * have a linear part and all the buffers except 407543170c4eSIlan Tayari * the last are of the same length. 407607b26c94SSteffen Klassert */ 407743170c4eSIlan Tayari frag_len = list_skb->len; 407807b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 407943170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 408043170c4eSIlan Tayari goto normal; 4081eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 408207b26c94SSteffen Klassert goto normal; 408307b26c94SSteffen Klassert 408407b26c94SSteffen Klassert len -= iter->len; 408507b26c94SSteffen Klassert } 408643170c4eSIlan Tayari 408743170c4eSIlan Tayari if (len != frag_len) 408843170c4eSIlan Tayari goto normal; 408907b26c94SSteffen Klassert } 409007b26c94SSteffen Klassert 4091802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 4092802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 4093802ab55aSAlexander Duyck * now. 4094802ab55aSAlexander Duyck */ 4095802ab55aSAlexander Duyck partial_segs = len / mss; 4096d7fb5a80SAlexander Duyck if (partial_segs > 1) 4097802ab55aSAlexander Duyck mss *= partial_segs; 4098d7fb5a80SAlexander Duyck else 4099d7fb5a80SAlexander Duyck partial_segs = 0; 4100802ab55aSAlexander Duyck } 4101802ab55aSAlexander Duyck 410207b26c94SSteffen Klassert normal: 4103df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 4104df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 4105f4c50d99SHerbert Xu 4106f4c50d99SHerbert Xu do { 4107f4c50d99SHerbert Xu struct sk_buff *nskb; 41088cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 4109c8884eddSHerbert Xu int hsize; 4110f4c50d99SHerbert Xu int size; 4111f4c50d99SHerbert Xu 41123953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 41133953c46cSMarcelo Ricardo Leitner len = list_skb->len; 41143953c46cSMarcelo Ricardo Leitner } else { 4115df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 4116f4c50d99SHerbert Xu if (len > mss) 4117f4c50d99SHerbert Xu len = mss; 41183953c46cSMarcelo Ricardo Leitner } 4119f4c50d99SHerbert Xu 4120df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 4121f4c50d99SHerbert Xu 4122dbd50f23SXin Long if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) && 41231a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 41241a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 412589319d38SHerbert Xu 41269d8506ccSHerbert Xu i = 0; 41271a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 41281a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 41291fd819ecSMichael S. Tsirkin frag_skb = list_skb; 41301a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 41319d8506ccSHerbert Xu 41329d8506ccSHerbert Xu while (pos < offset + len) { 41339d8506ccSHerbert Xu BUG_ON(i >= nfrags); 41349d8506ccSHerbert Xu 41354e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 41369d8506ccSHerbert Xu if (pos + size > offset + len) 41379d8506ccSHerbert Xu break; 41389d8506ccSHerbert Xu 41399d8506ccSHerbert Xu i++; 41409d8506ccSHerbert Xu pos += size; 41414e1beba1SMichael S. Tsirkin frag++; 41429d8506ccSHerbert Xu } 41439d8506ccSHerbert Xu 41441a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 41451a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 414689319d38SHerbert Xu 4147f4c50d99SHerbert Xu if (unlikely(!nskb)) 4148f4c50d99SHerbert Xu goto err; 4149f4c50d99SHerbert Xu 41509d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 41519d8506ccSHerbert Xu kfree_skb(nskb); 41529d8506ccSHerbert Xu goto err; 41539d8506ccSHerbert Xu } 41549d8506ccSHerbert Xu 4155ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 415689319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 415789319d38SHerbert Xu kfree_skb(nskb); 415889319d38SHerbert Xu goto err; 415989319d38SHerbert Xu } 416089319d38SHerbert Xu 4161ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 416289319d38SHerbert Xu skb_release_head_state(nskb); 416389319d38SHerbert Xu __skb_push(nskb, doffset); 416489319d38SHerbert Xu } else { 416500b229f7SPaolo Abeni if (hsize < 0) 416600b229f7SPaolo Abeni hsize = 0; 4167dbd50f23SXin Long if (hsize > len || !sg) 4168dbd50f23SXin Long hsize = len; 4169dbd50f23SXin Long 4170c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 4171df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 4172c93bdd0eSMel Gorman NUMA_NO_NODE); 417389319d38SHerbert Xu 417489319d38SHerbert Xu if (unlikely(!nskb)) 417589319d38SHerbert Xu goto err; 417689319d38SHerbert Xu 417789319d38SHerbert Xu skb_reserve(nskb, headroom); 417889319d38SHerbert Xu __skb_put(nskb, doffset); 417989319d38SHerbert Xu } 418089319d38SHerbert Xu 4181f4c50d99SHerbert Xu if (segs) 4182f4c50d99SHerbert Xu tail->next = nskb; 4183f4c50d99SHerbert Xu else 4184f4c50d99SHerbert Xu segs = nskb; 4185f4c50d99SHerbert Xu tail = nskb; 4186f4c50d99SHerbert Xu 4187df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 4188f4c50d99SHerbert Xu 4189030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 4190fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 419168c33163SPravin B Shelar 4192df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 419368c33163SPravin B Shelar nskb->data - tnl_hlen, 419468c33163SPravin B Shelar doffset + tnl_hlen); 419589319d38SHerbert Xu 41969d8506ccSHerbert Xu if (nskb->len == len + doffset) 41971cdbcb79SSimon Horman goto perform_csum_check; 419889319d38SHerbert Xu 41997fbeffedSAlexander Duyck if (!sg) { 42001454c9faSYadu Kishore if (!csum) { 42017fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 42026f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 420376443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 420476443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 42051454c9faSYadu Kishore skb_put(nskb, 42061454c9faSYadu Kishore len), 42078d5930dfSAl Viro len); 42087e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 4209de843723STom Herbert skb_headroom(nskb) + doffset; 42101454c9faSYadu Kishore } else { 42111454c9faSYadu Kishore skb_copy_bits(head_skb, offset, 42121454c9faSYadu Kishore skb_put(nskb, len), 42131454c9faSYadu Kishore len); 42141454c9faSYadu Kishore } 4215f4c50d99SHerbert Xu continue; 4216f4c50d99SHerbert Xu } 4217f4c50d99SHerbert Xu 42188cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 4219f4c50d99SHerbert Xu 4220df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 4221d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4222f4c50d99SHerbert Xu 422306b4feb3SJonathan Lemon skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags & 422406b4feb3SJonathan Lemon SKBFL_SHARED_FRAG; 4225cef401deSEric Dumazet 4226bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4227bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4228bf5c25d6SWillem de Bruijn goto err; 4229bf5c25d6SWillem de Bruijn 42309d8506ccSHerbert Xu while (pos < offset + len) { 42319d8506ccSHerbert Xu if (i >= nfrags) { 42329d8506ccSHerbert Xu i = 0; 42331a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 42341a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 42351fd819ecSMichael S. Tsirkin frag_skb = list_skb; 423613acc94eSYonghong Song if (!skb_headlen(list_skb)) { 42379d8506ccSHerbert Xu BUG_ON(!nfrags); 423813acc94eSYonghong Song } else { 423913acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 42409d8506ccSHerbert Xu 424113acc94eSYonghong Song /* to make room for head_frag. */ 424213acc94eSYonghong Song i--; 424313acc94eSYonghong Song frag--; 424413acc94eSYonghong Song } 4245bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4246bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4247bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4248bf5c25d6SWillem de Bruijn goto err; 4249bf5c25d6SWillem de Bruijn 42501a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 42519d8506ccSHerbert Xu } 42529d8506ccSHerbert Xu 42539d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 42549d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 42559d8506ccSHerbert Xu net_warn_ratelimited( 42569d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 42579d8506ccSHerbert Xu pos, mss); 4258ff907a11SEric Dumazet err = -EINVAL; 42599d8506ccSHerbert Xu goto err; 42609d8506ccSHerbert Xu } 42619d8506ccSHerbert Xu 426213acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 42638cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 42648cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4265f4c50d99SHerbert Xu 4266f4c50d99SHerbert Xu if (pos < offset) { 4267b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 42688cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4269f4c50d99SHerbert Xu } 4270f4c50d99SHerbert Xu 427189319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4272f4c50d99SHerbert Xu 4273f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4274f4c50d99SHerbert Xu i++; 42754e1beba1SMichael S. Tsirkin frag++; 4276f4c50d99SHerbert Xu pos += size; 4277f4c50d99SHerbert Xu } else { 42788cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 427989319d38SHerbert Xu goto skip_fraglist; 4280f4c50d99SHerbert Xu } 4281f4c50d99SHerbert Xu 42828cb19905SMichael S. Tsirkin nskb_frag++; 4283f4c50d99SHerbert Xu } 4284f4c50d99SHerbert Xu 428589319d38SHerbert Xu skip_fraglist: 4286f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4287f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4288f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4289ec5f0615SPravin B Shelar 42901cdbcb79SSimon Horman perform_csum_check: 42917fbeffedSAlexander Duyck if (!csum) { 4292ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4293ff907a11SEric Dumazet __skb_linearize(nskb)) 4294ddff00d4SAlexander Duyck goto err; 4295ff907a11SEric Dumazet 42967fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4297ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 429876443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 429976443456SAlexander Duyck skb_checksum(nskb, doffset, 430076443456SAlexander Duyck nskb->len - doffset, 0); 43017e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 43027e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4303ec5f0615SPravin B Shelar } 4304df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4305f4c50d99SHerbert Xu 4306bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4307bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4308bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4309bec3cfdcSEric Dumazet */ 4310bec3cfdcSEric Dumazet segs->prev = tail; 4311432c856fSToshiaki Makita 4312802ab55aSAlexander Duyck if (partial_segs) { 431307b26c94SSteffen Klassert struct sk_buff *iter; 4314802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 431507b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4316802ab55aSAlexander Duyck 4317802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 431807b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4319802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4320802ab55aSAlexander Duyck 4321802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4322802ab55aSAlexander Duyck * our way back down the stack of protocols. 4323802ab55aSAlexander Duyck */ 432407b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 432507b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 432607b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 432707b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 432807b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 432907b26c94SSteffen Klassert } 433007b26c94SSteffen Klassert 433107b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 433207b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 433307b26c94SSteffen Klassert else if (tail != segs) 433407b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4335802ab55aSAlexander Duyck } 4336802ab55aSAlexander Duyck 4337432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4338432c856fSToshiaki Makita * using skb_set_owner_w(). 4339432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4340432c856fSToshiaki Makita */ 4341432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4342432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4343432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4344432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4345432c856fSToshiaki Makita } 4346f4c50d99SHerbert Xu return segs; 4347f4c50d99SHerbert Xu 4348f4c50d99SHerbert Xu err: 4349289dccbeSEric Dumazet kfree_skb_list(segs); 4350f4c50d99SHerbert Xu return ERR_PTR(err); 4351f4c50d99SHerbert Xu } 4352f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4353f4c50d99SHerbert Xu 4354df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4355df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4356df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4357df5042f4SFlorian Westphal 4358df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4359df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4360df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4361df5042f4SFlorian Westphal #endif 43624165079bSFlorian Westphal #ifdef CONFIG_XFRM 43634165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 43644165079bSFlorian Westphal #endif 436595a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 436695a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 436795a7233cSPaul Blakey #endif 43683ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 43693ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 43703ee17bc7SMat Martineau #endif 437178476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 437278476d31SJeremy Kerr [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow), 437378476d31SJeremy Kerr #endif 4374df5042f4SFlorian Westphal }; 4375df5042f4SFlorian Westphal 4376df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4377df5042f4SFlorian Westphal { 4378df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4379df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4380df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4381df5042f4SFlorian Westphal #endif 43824165079bSFlorian Westphal #ifdef CONFIG_XFRM 43834165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 43844165079bSFlorian Westphal #endif 438595a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 438695a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 438795a7233cSPaul Blakey #endif 43883ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 43893ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 43903ee17bc7SMat Martineau #endif 439178476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 439278476d31SJeremy Kerr skb_ext_type_len[SKB_EXT_MCTP] + 439378476d31SJeremy Kerr #endif 4394df5042f4SFlorian Westphal 0; 4395df5042f4SFlorian Westphal } 4396df5042f4SFlorian Westphal 4397df5042f4SFlorian Westphal static void skb_extensions_init(void) 4398df5042f4SFlorian Westphal { 4399df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4400df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4401df5042f4SFlorian Westphal 4402df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4403df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4404df5042f4SFlorian Westphal 0, 4405df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4406df5042f4SFlorian Westphal NULL); 4407df5042f4SFlorian Westphal } 4408df5042f4SFlorian Westphal #else 4409df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4410df5042f4SFlorian Westphal #endif 4411df5042f4SFlorian Westphal 44121da177e4SLinus Torvalds void __init skb_init(void) 44131da177e4SLinus Torvalds { 441479a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 44151da177e4SLinus Torvalds sizeof(struct sk_buff), 44161da177e4SLinus Torvalds 0, 4417e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 441879a8a642SKees Cook offsetof(struct sk_buff, cb), 441979a8a642SKees Cook sizeof_field(struct sk_buff, cb), 442020c2df83SPaul Mundt NULL); 4421d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4422d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4423d179cd12SDavid S. Miller 0, 4424e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 442520c2df83SPaul Mundt NULL); 4426df5042f4SFlorian Westphal skb_extensions_init(); 44271da177e4SLinus Torvalds } 44281da177e4SLinus Torvalds 442951c739d1SDavid S. Miller static int 443048a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 443148a1df65SJason A. Donenfeld unsigned int recursion_level) 4432716ea3a7SDavid Howells { 44331a028e50SDavid S. Miller int start = skb_headlen(skb); 44341a028e50SDavid S. Miller int i, copy = start - offset; 4435fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4436716ea3a7SDavid Howells int elt = 0; 4437716ea3a7SDavid Howells 443848a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 443948a1df65SJason A. Donenfeld return -EMSGSIZE; 444048a1df65SJason A. Donenfeld 4441716ea3a7SDavid Howells if (copy > 0) { 4442716ea3a7SDavid Howells if (copy > len) 4443716ea3a7SDavid Howells copy = len; 4444642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4445716ea3a7SDavid Howells elt++; 4446716ea3a7SDavid Howells if ((len -= copy) == 0) 4447716ea3a7SDavid Howells return elt; 4448716ea3a7SDavid Howells offset += copy; 4449716ea3a7SDavid Howells } 4450716ea3a7SDavid Howells 4451716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 44521a028e50SDavid S. Miller int end; 4453716ea3a7SDavid Howells 4454547b792cSIlpo Järvinen WARN_ON(start > offset + len); 44551a028e50SDavid S. Miller 44569e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4457716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4458716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 445948a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 446048a1df65SJason A. Donenfeld return -EMSGSIZE; 4461716ea3a7SDavid Howells 4462716ea3a7SDavid Howells if (copy > len) 4463716ea3a7SDavid Howells copy = len; 4464ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4465b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4466716ea3a7SDavid Howells elt++; 4467716ea3a7SDavid Howells if (!(len -= copy)) 4468716ea3a7SDavid Howells return elt; 4469716ea3a7SDavid Howells offset += copy; 4470716ea3a7SDavid Howells } 44711a028e50SDavid S. Miller start = end; 4472716ea3a7SDavid Howells } 4473716ea3a7SDavid Howells 4474fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 447548a1df65SJason A. Donenfeld int end, ret; 4476716ea3a7SDavid Howells 4477547b792cSIlpo Järvinen WARN_ON(start > offset + len); 44781a028e50SDavid S. Miller 4479fbb398a8SDavid S. Miller end = start + frag_iter->len; 4480716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 448148a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 448248a1df65SJason A. Donenfeld return -EMSGSIZE; 448348a1df65SJason A. Donenfeld 4484716ea3a7SDavid Howells if (copy > len) 4485716ea3a7SDavid Howells copy = len; 448648a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 448748a1df65SJason A. Donenfeld copy, recursion_level + 1); 448848a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 448948a1df65SJason A. Donenfeld return ret; 449048a1df65SJason A. Donenfeld elt += ret; 4491716ea3a7SDavid Howells if ((len -= copy) == 0) 4492716ea3a7SDavid Howells return elt; 4493716ea3a7SDavid Howells offset += copy; 4494716ea3a7SDavid Howells } 44951a028e50SDavid S. Miller start = end; 4496716ea3a7SDavid Howells } 4497716ea3a7SDavid Howells BUG_ON(len); 4498716ea3a7SDavid Howells return elt; 4499716ea3a7SDavid Howells } 4500716ea3a7SDavid Howells 450148a1df65SJason A. Donenfeld /** 450248a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 450348a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 450448a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 450548a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 450648a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 450748a1df65SJason A. Donenfeld * 450848a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 450948a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 451048a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 451148a1df65SJason A. Donenfeld * could not fit. 451248a1df65SJason A. Donenfeld */ 451348a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 451448a1df65SJason A. Donenfeld { 451548a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 451648a1df65SJason A. Donenfeld 451748a1df65SJason A. Donenfeld if (nsg <= 0) 451848a1df65SJason A. Donenfeld return nsg; 451948a1df65SJason A. Donenfeld 452048a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 452148a1df65SJason A. Donenfeld 452248a1df65SJason A. Donenfeld return nsg; 452348a1df65SJason A. Donenfeld } 452448a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 452548a1df65SJason A. Donenfeld 452625a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 452725a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 452825a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 452925a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 453025a91d8dSFan Du * 453125a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 453225a91d8dSFan Du * 1. sg_init_table 453325a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 453425a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 453525a91d8dSFan Du * 453625a91d8dSFan Du * This is equivalent to: 453725a91d8dSFan Du * 1. sg_init_table 453825a91d8dSFan Du * 2. skb_to_sgvec(payload1) 453925a91d8dSFan Du * 3. sg_unmark_end 454025a91d8dSFan Du * 4. skb_to_sgvec(payload2) 454125a91d8dSFan Du * 454225a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 454325a91d8dSFan Du * is more preferable. 454425a91d8dSFan Du */ 454525a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 454625a91d8dSFan Du int offset, int len) 454725a91d8dSFan Du { 454848a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 454925a91d8dSFan Du } 455025a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 455125a91d8dSFan Du 455251c739d1SDavid S. Miller 455351c739d1SDavid S. Miller 4554716ea3a7SDavid Howells /** 4555716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4556716ea3a7SDavid Howells * @skb: The socket buffer to check. 4557716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4558716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4559716ea3a7SDavid Howells * 4560716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4561716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4562716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4563716ea3a7SDavid Howells * 4564716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4565716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4566716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4567716ea3a7SDavid Howells * 4568716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4569716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4570716ea3a7SDavid Howells */ 4571716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4572716ea3a7SDavid Howells { 4573716ea3a7SDavid Howells int copyflag; 4574716ea3a7SDavid Howells int elt; 4575716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4576716ea3a7SDavid Howells 4577716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4578716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4579716ea3a7SDavid Howells * at the moment even if they are anonymous). 4580716ea3a7SDavid Howells */ 4581716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4582c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4583716ea3a7SDavid Howells return -ENOMEM; 4584716ea3a7SDavid Howells 4585716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 458621dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4587716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4588716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4589716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4590716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4591716ea3a7SDavid Howells 4592716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4593716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4594716ea3a7SDavid Howells return -ENOMEM; 4595716ea3a7SDavid Howells 4596716ea3a7SDavid Howells /* Voila! */ 4597716ea3a7SDavid Howells *trailer = skb; 4598716ea3a7SDavid Howells return 1; 4599716ea3a7SDavid Howells } 4600716ea3a7SDavid Howells 4601716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4602716ea3a7SDavid Howells 4603716ea3a7SDavid Howells elt = 1; 4604716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4605716ea3a7SDavid Howells copyflag = 0; 4606716ea3a7SDavid Howells 4607716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4608716ea3a7SDavid Howells int ntail = 0; 4609716ea3a7SDavid Howells 4610716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4611716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4612716ea3a7SDavid Howells * after it. */ 4613716ea3a7SDavid Howells 4614716ea3a7SDavid Howells if (skb_shared(skb1)) 4615716ea3a7SDavid Howells copyflag = 1; 4616716ea3a7SDavid Howells 4617716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4618716ea3a7SDavid Howells 4619716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4620716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 462121dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4622716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4623716ea3a7SDavid Howells ntail = tailbits + 128; 4624716ea3a7SDavid Howells } 4625716ea3a7SDavid Howells 4626716ea3a7SDavid Howells if (copyflag || 4627716ea3a7SDavid Howells skb_cloned(skb1) || 4628716ea3a7SDavid Howells ntail || 4629716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 463021dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4631716ea3a7SDavid Howells struct sk_buff *skb2; 4632716ea3a7SDavid Howells 4633716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4634716ea3a7SDavid Howells if (ntail == 0) 4635716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4636716ea3a7SDavid Howells else 4637716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4638716ea3a7SDavid Howells skb_headroom(skb1), 4639716ea3a7SDavid Howells ntail, 4640716ea3a7SDavid Howells GFP_ATOMIC); 4641716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4642716ea3a7SDavid Howells return -ENOMEM; 4643716ea3a7SDavid Howells 4644716ea3a7SDavid Howells if (skb1->sk) 4645716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4646716ea3a7SDavid Howells 4647716ea3a7SDavid Howells /* Looking around. Are we still alive? 4648716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4649716ea3a7SDavid Howells 4650716ea3a7SDavid Howells skb2->next = skb1->next; 4651716ea3a7SDavid Howells *skb_p = skb2; 4652716ea3a7SDavid Howells kfree_skb(skb1); 4653716ea3a7SDavid Howells skb1 = skb2; 4654716ea3a7SDavid Howells } 4655716ea3a7SDavid Howells elt++; 4656716ea3a7SDavid Howells *trailer = skb1; 4657716ea3a7SDavid Howells skb_p = &skb1->next; 4658716ea3a7SDavid Howells } 4659716ea3a7SDavid Howells 4660716ea3a7SDavid Howells return elt; 4661716ea3a7SDavid Howells } 4662b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4663716ea3a7SDavid Howells 4664b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4665b1faf566SEric Dumazet { 4666b1faf566SEric Dumazet struct sock *sk = skb->sk; 4667b1faf566SEric Dumazet 4668b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4669b1faf566SEric Dumazet } 4670b1faf566SEric Dumazet 46718605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 46728605330aSSoheil Hassas Yeganeh { 46738605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 46748605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 46758605330aSSoheil Hassas Yeganeh */ 46768605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 46778605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 46788605330aSSoheil Hassas Yeganeh } 46798605330aSSoheil Hassas Yeganeh 4680b1faf566SEric Dumazet /* 4681b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4682b1faf566SEric Dumazet */ 4683b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4684b1faf566SEric Dumazet { 4685b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4686ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4687b1faf566SEric Dumazet return -ENOMEM; 4688b1faf566SEric Dumazet 4689b1faf566SEric Dumazet skb_orphan(skb); 4690b1faf566SEric Dumazet skb->sk = sk; 4691b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4692b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 46938605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4694b1faf566SEric Dumazet 4695abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4696abb57ea4SEric Dumazet skb_dst_force(skb); 4697abb57ea4SEric Dumazet 4698b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4699b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 4700e3ae2365SAlexander Aring sk_error_report(sk); 4701b1faf566SEric Dumazet return 0; 4702b1faf566SEric Dumazet } 4703b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4704b1faf566SEric Dumazet 470583a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 470683a1a1a7SSoheil Hassas Yeganeh { 470783a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 470883a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 470983a1a1a7SSoheil Hassas Yeganeh } 471083a1a1a7SSoheil Hassas Yeganeh 4711364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4712364a9e93SWillem de Bruijn { 4713364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 471483a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 471583a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4716997d5c3fSEric Dumazet unsigned long flags; 4717364a9e93SWillem de Bruijn 4718997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4719364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 472038b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 472183a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 472238b25793SSoheil Hassas Yeganeh if (icmp_next) 4723985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 472438b25793SSoheil Hassas Yeganeh } 4725997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4726364a9e93SWillem de Bruijn 472783a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 472883a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 472983a1a1a7SSoheil Hassas Yeganeh 473083a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4731e3ae2365SAlexander Aring sk_error_report(sk); 4732364a9e93SWillem de Bruijn 4733364a9e93SWillem de Bruijn return skb; 4734364a9e93SWillem de Bruijn } 4735364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4736364a9e93SWillem de Bruijn 4737cab41c47SAlexander Duyck /** 4738cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4739cab41c47SAlexander Duyck * @skb: the skb to clone 4740cab41c47SAlexander Duyck * 4741cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4742cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4743cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4744cab41c47SAlexander Duyck * 4745cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4746cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4747cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4748cab41c47SAlexander Duyck * the sk_error_queue. 4749cab41c47SAlexander Duyck */ 475062bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 475162bccb8cSAlexander Duyck { 475262bccb8cSAlexander Duyck struct sock *sk = skb->sk; 475362bccb8cSAlexander Duyck struct sk_buff *clone; 475462bccb8cSAlexander Duyck 475541c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 475662bccb8cSAlexander Duyck return NULL; 475762bccb8cSAlexander Duyck 475862bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 475962bccb8cSAlexander Duyck if (!clone) { 476062bccb8cSAlexander Duyck sock_put(sk); 476162bccb8cSAlexander Duyck return NULL; 476262bccb8cSAlexander Duyck } 476362bccb8cSAlexander Duyck 476462bccb8cSAlexander Duyck clone->sk = sk; 476562bccb8cSAlexander Duyck clone->destructor = sock_efree; 476662bccb8cSAlexander Duyck 476762bccb8cSAlexander Duyck return clone; 476862bccb8cSAlexander Duyck } 476962bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 477062bccb8cSAlexander Duyck 477137846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 477237846ef0SAlexander Duyck struct sock *sk, 47734ef1b286SSoheil Hassas Yeganeh int tstype, 47744ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4775ac45f602SPatrick Ohly { 4776ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4777ac45f602SPatrick Ohly int err; 4778ac45f602SPatrick Ohly 47794ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 47804ef1b286SSoheil Hassas Yeganeh 4781ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4782ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4783ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4784ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4785e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 47864ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 47871862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 47884ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 478909c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 479042f67eeaSEric Dumazet if (sk_is_tcp(sk)) 4791a1cdec57SEric Dumazet serr->ee.ee_data -= atomic_read(&sk->sk_tskey); 47924ed2d765SWillem de Bruijn } 479329030374SEric Dumazet 4794ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 479529030374SEric Dumazet 4796ac45f602SPatrick Ohly if (err) 4797ac45f602SPatrick Ohly kfree_skb(skb); 4798ac45f602SPatrick Ohly } 479937846ef0SAlexander Duyck 4800b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4801b245be1fSWillem de Bruijn { 4802b245be1fSWillem de Bruijn bool ret; 4803b245be1fSWillem de Bruijn 4804b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4805b245be1fSWillem de Bruijn return true; 4806b245be1fSWillem de Bruijn 4807b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4808b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4809b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4810b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4811b245be1fSWillem de Bruijn return ret; 4812b245be1fSWillem de Bruijn } 4813b245be1fSWillem de Bruijn 481437846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 481537846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 481637846ef0SAlexander Duyck { 481737846ef0SAlexander Duyck struct sock *sk = skb->sk; 481837846ef0SAlexander Duyck 4819b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 482035b99dffSWillem de Bruijn goto err; 4821b245be1fSWillem de Bruijn 48229ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 48239ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 48249ac25fc0SEric Dumazet */ 482541c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 482637846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 48274ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 482837846ef0SAlexander Duyck sock_put(sk); 482935b99dffSWillem de Bruijn return; 483037846ef0SAlexander Duyck } 483135b99dffSWillem de Bruijn 483235b99dffSWillem de Bruijn err: 483335b99dffSWillem de Bruijn kfree_skb(skb); 48349ac25fc0SEric Dumazet } 483537846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 483637846ef0SAlexander Duyck 483737846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 4838e7ed11eeSYousuk Seung const struct sk_buff *ack_skb, 483937846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 484037846ef0SAlexander Duyck struct sock *sk, int tstype) 484137846ef0SAlexander Duyck { 484237846ef0SAlexander Duyck struct sk_buff *skb; 48434ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 484437846ef0SAlexander Duyck 48453a8dd971SWillem de Bruijn if (!sk) 48463a8dd971SWillem de Bruijn return; 48473a8dd971SWillem de Bruijn 4848b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4849b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4850b50a5c70SMiroslav Lichvar return; 4851b50a5c70SMiroslav Lichvar 48523a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 48533a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 485437846ef0SAlexander Duyck return; 485537846ef0SAlexander Duyck 48561c885808SFrancis Yan if (tsonly) { 48571c885808SFrancis Yan #ifdef CONFIG_INET 48581c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 485942f67eeaSEric Dumazet sk_is_tcp(sk)) { 4860e7ed11eeSYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb, 4861e7ed11eeSYousuk Seung ack_skb); 48624ef1b286SSoheil Hassas Yeganeh opt_stats = true; 48634ef1b286SSoheil Hassas Yeganeh } else 48641c885808SFrancis Yan #endif 48651c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 48661c885808SFrancis Yan } else { 486737846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 48681c885808SFrancis Yan } 486937846ef0SAlexander Duyck if (!skb) 487037846ef0SAlexander Duyck return; 487137846ef0SAlexander Duyck 487249ca0d8bSWillem de Bruijn if (tsonly) { 4873fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4874fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 487549ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 487649ca0d8bSWillem de Bruijn } 487749ca0d8bSWillem de Bruijn 487849ca0d8bSWillem de Bruijn if (hwtstamps) 487949ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 488049ca0d8bSWillem de Bruijn else 4881d93376f5SMartin KaFai Lau __net_timestamp(skb); 488249ca0d8bSWillem de Bruijn 48834ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 488437846ef0SAlexander Duyck } 4885e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4886e7fd2885SWillem de Bruijn 4887e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4888e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4889e7fd2885SWillem de Bruijn { 4890e7ed11eeSYousuk Seung return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk, 4891e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4892e7fd2885SWillem de Bruijn } 4893ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4894ac45f602SPatrick Ohly 48956e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 48966e3e939fSJohannes Berg { 48976e3e939fSJohannes Berg struct sock *sk = skb->sk; 48986e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4899dd4f1072SEric Dumazet int err = 1; 49006e3e939fSJohannes Berg 49016e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 49026e3e939fSJohannes Berg skb->wifi_acked = acked; 49036e3e939fSJohannes Berg 49046e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 49056e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 49066e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 49076e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 49086e3e939fSJohannes Berg 4909dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4910dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4911dd4f1072SEric Dumazet */ 491241c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 49136e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4914dd4f1072SEric Dumazet sock_put(sk); 4915dd4f1072SEric Dumazet } 49166e3e939fSJohannes Berg if (err) 49176e3e939fSJohannes Berg kfree_skb(skb); 49186e3e939fSJohannes Berg } 49196e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 49206e3e939fSJohannes Berg 4921f35d9d8aSRusty Russell /** 4922f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4923f35d9d8aSRusty Russell * @skb: the skb to set 4924f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4925f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4926f35d9d8aSRusty Russell * 4927f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4928f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4929f35d9d8aSRusty Russell * 4930f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4931f35d9d8aSRusty Russell * returns false you should drop the packet. 4932f35d9d8aSRusty Russell */ 4933f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4934f35d9d8aSRusty Russell { 493552b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 493652b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 493752b5d6f5SEric Dumazet 493852b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 493952b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 494052b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4941f35d9d8aSRusty Russell return false; 4942f35d9d8aSRusty Russell } 4943f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 494452b5d6f5SEric Dumazet skb->csum_start = csum_start; 4945f35d9d8aSRusty Russell skb->csum_offset = off; 4946e5d5decaSJason Wang skb_set_transport_header(skb, start); 4947f35d9d8aSRusty Russell return true; 4948f35d9d8aSRusty Russell } 4949b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4950f35d9d8aSRusty Russell 4951ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4952ed1f50c3SPaul Durrant unsigned int max) 4953ed1f50c3SPaul Durrant { 4954ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4955ed1f50c3SPaul Durrant return 0; 4956ed1f50c3SPaul Durrant 4957ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4958ed1f50c3SPaul Durrant * won't need to do it again. 4959ed1f50c3SPaul Durrant */ 4960ed1f50c3SPaul Durrant if (max > skb->len) 4961ed1f50c3SPaul Durrant max = skb->len; 4962ed1f50c3SPaul Durrant 4963ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4964ed1f50c3SPaul Durrant return -ENOMEM; 4965ed1f50c3SPaul Durrant 4966ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4967ed1f50c3SPaul Durrant return -EPROTO; 4968ed1f50c3SPaul Durrant 4969ed1f50c3SPaul Durrant return 0; 4970ed1f50c3SPaul Durrant } 4971ed1f50c3SPaul Durrant 4972f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4973f9708b43SJan Beulich 4974f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4975f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4976f9708b43SJan Beulich unsigned int off) 4977f9708b43SJan Beulich { 4978f9708b43SJan Beulich int err; 4979f9708b43SJan Beulich 4980161d1792SKees Cook switch (proto) { 4981f9708b43SJan Beulich case IPPROTO_TCP: 4982f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4983f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4984f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4985f9708b43SJan Beulich offsetof(struct tcphdr, 4986f9708b43SJan Beulich check))) 4987f9708b43SJan Beulich err = -EPROTO; 4988f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4989f9708b43SJan Beulich 4990f9708b43SJan Beulich case IPPROTO_UDP: 4991f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4992f9708b43SJan Beulich off + sizeof(struct udphdr)); 4993f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4994f9708b43SJan Beulich offsetof(struct udphdr, 4995f9708b43SJan Beulich check))) 4996f9708b43SJan Beulich err = -EPROTO; 4997f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4998f9708b43SJan Beulich } 4999f9708b43SJan Beulich 5000f9708b43SJan Beulich return ERR_PTR(-EPROTO); 5001f9708b43SJan Beulich } 5002f9708b43SJan Beulich 5003ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5004ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 5005ed1f50c3SPaul Durrant */ 5006ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 5007ed1f50c3SPaul Durrant 5008f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 5009ed1f50c3SPaul Durrant { 5010ed1f50c3SPaul Durrant unsigned int off; 5011ed1f50c3SPaul Durrant bool fragment; 5012f9708b43SJan Beulich __sum16 *csum; 5013ed1f50c3SPaul Durrant int err; 5014ed1f50c3SPaul Durrant 5015ed1f50c3SPaul Durrant fragment = false; 5016ed1f50c3SPaul Durrant 5017ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5018ed1f50c3SPaul Durrant sizeof(struct iphdr), 5019ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 5020ed1f50c3SPaul Durrant if (err < 0) 5021ed1f50c3SPaul Durrant goto out; 5022ed1f50c3SPaul Durrant 502311f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 5024ed1f50c3SPaul Durrant fragment = true; 5025ed1f50c3SPaul Durrant 5026ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 5027ed1f50c3SPaul Durrant 5028ed1f50c3SPaul Durrant err = -EPROTO; 5029ed1f50c3SPaul Durrant 5030ed1f50c3SPaul Durrant if (fragment) 5031ed1f50c3SPaul Durrant goto out; 5032ed1f50c3SPaul Durrant 5033f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 5034f9708b43SJan Beulich if (IS_ERR(csum)) 5035f9708b43SJan Beulich return PTR_ERR(csum); 5036ed1f50c3SPaul Durrant 5037ed1f50c3SPaul Durrant if (recalculate) 5038f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 5039ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 5040ed1f50c3SPaul Durrant skb->len - off, 5041f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 5042ed1f50c3SPaul Durrant err = 0; 5043ed1f50c3SPaul Durrant 5044ed1f50c3SPaul Durrant out: 5045ed1f50c3SPaul Durrant return err; 5046ed1f50c3SPaul Durrant } 5047ed1f50c3SPaul Durrant 5048ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5049ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 5050ed1f50c3SPaul Durrant */ 5051ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 5052ed1f50c3SPaul Durrant 5053ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 5054ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 5055ed1f50c3SPaul Durrant 5056ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 5057ed1f50c3SPaul Durrant { 5058ed1f50c3SPaul Durrant int err; 5059ed1f50c3SPaul Durrant u8 nexthdr; 5060ed1f50c3SPaul Durrant unsigned int off; 5061ed1f50c3SPaul Durrant unsigned int len; 5062ed1f50c3SPaul Durrant bool fragment; 5063ed1f50c3SPaul Durrant bool done; 5064f9708b43SJan Beulich __sum16 *csum; 5065ed1f50c3SPaul Durrant 5066ed1f50c3SPaul Durrant fragment = false; 5067ed1f50c3SPaul Durrant done = false; 5068ed1f50c3SPaul Durrant 5069ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 5070ed1f50c3SPaul Durrant 5071ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 5072ed1f50c3SPaul Durrant if (err < 0) 5073ed1f50c3SPaul Durrant goto out; 5074ed1f50c3SPaul Durrant 5075ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 5076ed1f50c3SPaul Durrant 5077ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 5078ed1f50c3SPaul Durrant while (off <= len && !done) { 5079ed1f50c3SPaul Durrant switch (nexthdr) { 5080ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 5081ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 5082ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 5083ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 5084ed1f50c3SPaul Durrant 5085ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5086ed1f50c3SPaul Durrant off + 5087ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 5088ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5089ed1f50c3SPaul Durrant if (err < 0) 5090ed1f50c3SPaul Durrant goto out; 5091ed1f50c3SPaul Durrant 5092ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 5093ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5094ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 5095ed1f50c3SPaul Durrant break; 5096ed1f50c3SPaul Durrant } 5097ed1f50c3SPaul Durrant case IPPROTO_AH: { 5098ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 5099ed1f50c3SPaul Durrant 5100ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5101ed1f50c3SPaul Durrant off + 5102ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 5103ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5104ed1f50c3SPaul Durrant if (err < 0) 5105ed1f50c3SPaul Durrant goto out; 5106ed1f50c3SPaul Durrant 5107ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 5108ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5109ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 5110ed1f50c3SPaul Durrant break; 5111ed1f50c3SPaul Durrant } 5112ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 5113ed1f50c3SPaul Durrant struct frag_hdr *hp; 5114ed1f50c3SPaul Durrant 5115ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5116ed1f50c3SPaul Durrant off + 5117ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5118ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5119ed1f50c3SPaul Durrant if (err < 0) 5120ed1f50c3SPaul Durrant goto out; 5121ed1f50c3SPaul Durrant 5122ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5123ed1f50c3SPaul Durrant 5124ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5125ed1f50c3SPaul Durrant fragment = true; 5126ed1f50c3SPaul Durrant 5127ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5128ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5129ed1f50c3SPaul Durrant break; 5130ed1f50c3SPaul Durrant } 5131ed1f50c3SPaul Durrant default: 5132ed1f50c3SPaul Durrant done = true; 5133ed1f50c3SPaul Durrant break; 5134ed1f50c3SPaul Durrant } 5135ed1f50c3SPaul Durrant } 5136ed1f50c3SPaul Durrant 5137ed1f50c3SPaul Durrant err = -EPROTO; 5138ed1f50c3SPaul Durrant 5139ed1f50c3SPaul Durrant if (!done || fragment) 5140ed1f50c3SPaul Durrant goto out; 5141ed1f50c3SPaul Durrant 5142f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5143f9708b43SJan Beulich if (IS_ERR(csum)) 5144f9708b43SJan Beulich return PTR_ERR(csum); 5145ed1f50c3SPaul Durrant 5146ed1f50c3SPaul Durrant if (recalculate) 5147f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5148ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5149f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5150ed1f50c3SPaul Durrant err = 0; 5151ed1f50c3SPaul Durrant 5152ed1f50c3SPaul Durrant out: 5153ed1f50c3SPaul Durrant return err; 5154ed1f50c3SPaul Durrant } 5155ed1f50c3SPaul Durrant 5156ed1f50c3SPaul Durrant /** 5157ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5158ed1f50c3SPaul Durrant * @skb: the skb to set up 5159ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5160ed1f50c3SPaul Durrant */ 5161ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5162ed1f50c3SPaul Durrant { 5163ed1f50c3SPaul Durrant int err; 5164ed1f50c3SPaul Durrant 5165ed1f50c3SPaul Durrant switch (skb->protocol) { 5166ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5167f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5168ed1f50c3SPaul Durrant break; 5169ed1f50c3SPaul Durrant 5170ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5171ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5172ed1f50c3SPaul Durrant break; 5173ed1f50c3SPaul Durrant 5174ed1f50c3SPaul Durrant default: 5175ed1f50c3SPaul Durrant err = -EPROTO; 5176ed1f50c3SPaul Durrant break; 5177ed1f50c3SPaul Durrant } 5178ed1f50c3SPaul Durrant 5179ed1f50c3SPaul Durrant return err; 5180ed1f50c3SPaul Durrant } 5181ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5182ed1f50c3SPaul Durrant 51839afd85c9SLinus Lüssing /** 51849afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 51859afd85c9SLinus Lüssing * @skb: the skb to check 51869afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51879afd85c9SLinus Lüssing * 51889afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 51899afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 51909afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 51919afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 51929afd85c9SLinus Lüssing * 5193a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5194a516993fSLinus Lüssing * differs from the provided skb. 51959afd85c9SLinus Lüssing */ 51969afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 51979afd85c9SLinus Lüssing unsigned int transport_len) 51989afd85c9SLinus Lüssing { 51999afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52009afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 52019afd85c9SLinus Lüssing int ret; 52029afd85c9SLinus Lüssing 5203a516993fSLinus Lüssing if (skb->len < len) 52049afd85c9SLinus Lüssing return NULL; 5205a516993fSLinus Lüssing else if (skb->len == len) 52069afd85c9SLinus Lüssing return skb; 52079afd85c9SLinus Lüssing 52089afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 52099afd85c9SLinus Lüssing if (!skb_chk) 52109afd85c9SLinus Lüssing return NULL; 52119afd85c9SLinus Lüssing 52129afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 52139afd85c9SLinus Lüssing if (ret) { 52149afd85c9SLinus Lüssing kfree_skb(skb_chk); 52159afd85c9SLinus Lüssing return NULL; 52169afd85c9SLinus Lüssing } 52179afd85c9SLinus Lüssing 52189afd85c9SLinus Lüssing return skb_chk; 52199afd85c9SLinus Lüssing } 52209afd85c9SLinus Lüssing 52219afd85c9SLinus Lüssing /** 52229afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 52239afd85c9SLinus Lüssing * @skb: the skb to check 52249afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 52259afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 52269afd85c9SLinus Lüssing * 52279afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 52289afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 52299afd85c9SLinus Lüssing * 52309afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 52319afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 52329afd85c9SLinus Lüssing * 5233a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5234a516993fSLinus Lüssing * differs from the provided skb. 52359afd85c9SLinus Lüssing */ 52369afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 52379afd85c9SLinus Lüssing unsigned int transport_len, 52389afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 52399afd85c9SLinus Lüssing { 52409afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52419afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5242fcba67c9SLinus Lüssing __sum16 ret; 52439afd85c9SLinus Lüssing 52449afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 52459afd85c9SLinus Lüssing if (!skb_chk) 5246a516993fSLinus Lüssing goto err; 52479afd85c9SLinus Lüssing 5248a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5249a516993fSLinus Lüssing goto err; 52509afd85c9SLinus Lüssing 52519b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 52529afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 52539b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 52549afd85c9SLinus Lüssing 5255a516993fSLinus Lüssing if (ret) 5256a516993fSLinus Lüssing goto err; 52579afd85c9SLinus Lüssing 52589afd85c9SLinus Lüssing return skb_chk; 5259a516993fSLinus Lüssing 5260a516993fSLinus Lüssing err: 5261a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5262a516993fSLinus Lüssing kfree_skb(skb_chk); 5263a516993fSLinus Lüssing 5264a516993fSLinus Lüssing return NULL; 5265a516993fSLinus Lüssing 52669afd85c9SLinus Lüssing } 52679afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 52689afd85c9SLinus Lüssing 52694497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 52704497b076SBen Hutchings { 5271e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5272e87cc472SJoe Perches skb->dev->name); 52734497b076SBen Hutchings } 52744497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5275bad43ca8SEric Dumazet 5276bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5277bad43ca8SEric Dumazet { 52783d861f66SEric Dumazet if (head_stolen) { 52793d861f66SEric Dumazet skb_release_head_state(skb); 5280bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 52813d861f66SEric Dumazet } else { 5282bad43ca8SEric Dumazet __kfree_skb(skb); 5283bad43ca8SEric Dumazet } 52843d861f66SEric Dumazet } 5285bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5286bad43ca8SEric Dumazet 5287bad43ca8SEric Dumazet /** 5288bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5289bad43ca8SEric Dumazet * @to: prior buffer 5290bad43ca8SEric Dumazet * @from: buffer to add 5291bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5292c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5293bad43ca8SEric Dumazet */ 5294bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5295bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5296bad43ca8SEric Dumazet { 5297c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5298bad43ca8SEric Dumazet int i, delta, len = from->len; 5299bad43ca8SEric Dumazet 5300bad43ca8SEric Dumazet *fragstolen = false; 5301bad43ca8SEric Dumazet 5302bad43ca8SEric Dumazet if (skb_cloned(to)) 5303bad43ca8SEric Dumazet return false; 5304bad43ca8SEric Dumazet 53051effe8caSJean-Philippe Brucker /* In general, avoid mixing slab allocated and page_pool allocated 53061effe8caSJean-Philippe Brucker * pages within the same SKB. However when @to is not pp_recycle and 53071effe8caSJean-Philippe Brucker * @from is cloned, we can transition frag pages from page_pool to 53081effe8caSJean-Philippe Brucker * reference counted. 53091effe8caSJean-Philippe Brucker * 53101effe8caSJean-Philippe Brucker * On the other hand, don't allow coalescing two pp_recycle SKBs if 53111effe8caSJean-Philippe Brucker * @from is cloned, in case the SKB is using page_pool fragment 53121effe8caSJean-Philippe Brucker * references (PP_FLAG_PAGE_FRAG). Since we only take full page 53131effe8caSJean-Philippe Brucker * references for cloned SKBs at the moment that would result in 53141effe8caSJean-Philippe Brucker * inconsistent reference counts. 53156a5bcd84SIlias Apalodimas */ 53161effe8caSJean-Philippe Brucker if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from))) 53176a5bcd84SIlias Apalodimas return false; 53186a5bcd84SIlias Apalodimas 5319bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5320e93a0435SEric Dumazet if (len) 5321bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5322bad43ca8SEric Dumazet *delta_truesize = 0; 5323bad43ca8SEric Dumazet return true; 5324bad43ca8SEric Dumazet } 5325bad43ca8SEric Dumazet 5326c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5327c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5328c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5329bad43ca8SEric Dumazet return false; 53301f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 53311f8b977aSWillem de Bruijn return false; 5332bad43ca8SEric Dumazet 5333bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5334bad43ca8SEric Dumazet struct page *page; 5335bad43ca8SEric Dumazet unsigned int offset; 5336bad43ca8SEric Dumazet 5337c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5338c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5339bad43ca8SEric Dumazet return false; 5340bad43ca8SEric Dumazet 5341bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5342bad43ca8SEric Dumazet return false; 5343bad43ca8SEric Dumazet 5344bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5345bad43ca8SEric Dumazet 5346bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5347bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5348bad43ca8SEric Dumazet 5349c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5350bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5351bad43ca8SEric Dumazet *fragstolen = true; 5352bad43ca8SEric Dumazet } else { 5353c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5354c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5355bad43ca8SEric Dumazet return false; 5356bad43ca8SEric Dumazet 5357f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5358bad43ca8SEric Dumazet } 5359bad43ca8SEric Dumazet 5360bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5361bad43ca8SEric Dumazet 5362c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5363c818fa9eSEric Dumazet from_shinfo->frags, 5364c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5365c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5366bad43ca8SEric Dumazet 5367bad43ca8SEric Dumazet if (!skb_cloned(from)) 5368c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5369bad43ca8SEric Dumazet 53708ea853fdSLi RongQing /* if the skb is not cloned this does nothing 53718ea853fdSLi RongQing * since we set nr_frags to 0. 53728ea853fdSLi RongQing */ 5373c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5374c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5375bad43ca8SEric Dumazet 5376bad43ca8SEric Dumazet to->truesize += delta; 5377bad43ca8SEric Dumazet to->len += len; 5378bad43ca8SEric Dumazet to->data_len += len; 5379bad43ca8SEric Dumazet 5380bad43ca8SEric Dumazet *delta_truesize = delta; 5381bad43ca8SEric Dumazet return true; 5382bad43ca8SEric Dumazet } 5383bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5384621e84d6SNicolas Dichtel 5385621e84d6SNicolas Dichtel /** 53868b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5387621e84d6SNicolas Dichtel * 5388621e84d6SNicolas Dichtel * @skb: buffer to clean 53898b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5390621e84d6SNicolas Dichtel * 53918b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 53928b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 53938b27f277SNicolas Dichtel * operations. 53948b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 53958b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 53968b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5397621e84d6SNicolas Dichtel */ 53988b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5399621e84d6SNicolas Dichtel { 5400621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5401621e84d6SNicolas Dichtel skb->skb_iif = 0; 540260ff7467SWANG Cong skb->ignore_df = 0; 5403621e84d6SNicolas Dichtel skb_dst_drop(skb); 5404174e2381SFlorian Westphal skb_ext_reset(skb); 5405895b5c9fSFlorian Westphal nf_reset_ct(skb); 5406621e84d6SNicolas Dichtel nf_reset_trace(skb); 5407213dd74aSHerbert Xu 54086f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 54096f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5410875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 54116f9a5069SPetr Machata #endif 54126f9a5069SPetr Machata 5413213dd74aSHerbert Xu if (!xnet) 5414213dd74aSHerbert Xu return; 5415213dd74aSHerbert Xu 54162b5ec1a5SYe Yin ipvs_reset(skb); 5417213dd74aSHerbert Xu skb->mark = 0; 5418de799101SMartin KaFai Lau skb_clear_tstamp(skb); 5419621e84d6SNicolas Dichtel } 5420621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5421de960aa9SFlorian Westphal 5422de960aa9SFlorian Westphal /** 5423de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5424de960aa9SFlorian Westphal * 5425de960aa9SFlorian Westphal * @skb: GSO skb 5426de960aa9SFlorian Westphal * 5427de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5428de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5429de960aa9SFlorian Westphal * 5430de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5431de960aa9SFlorian Westphal */ 5432a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5433de960aa9SFlorian Westphal { 5434de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5435f993bc25SFlorian Westphal unsigned int thlen = 0; 5436f993bc25SFlorian Westphal 5437f993bc25SFlorian Westphal if (skb->encapsulation) { 5438f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5439f993bc25SFlorian Westphal skb_transport_header(skb); 5440de960aa9SFlorian Westphal 5441de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5442f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5443f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5444f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 54451dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 544690017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5447ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5448ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5449f993bc25SFlorian Westphal } 54506d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 54516d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 54526d39d589SFlorian Westphal * accounted for. 54536d39d589SFlorian Westphal */ 5454f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5455de960aa9SFlorian Westphal } 5456a4a77718SDaniel Axtens 5457a4a77718SDaniel Axtens /** 5458a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5459a4a77718SDaniel Axtens * 5460a4a77718SDaniel Axtens * @skb: GSO skb 5461a4a77718SDaniel Axtens * 5462a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5463a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5464a4a77718SDaniel Axtens * 5465a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5466a4a77718SDaniel Axtens */ 5467a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5468a4a77718SDaniel Axtens { 5469a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5470a4a77718SDaniel Axtens skb_network_header(skb); 5471a4a77718SDaniel Axtens 5472a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5473a4a77718SDaniel Axtens } 5474a4a77718SDaniel Axtens 5475a4a77718SDaniel Axtens /** 5476a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5477a4a77718SDaniel Axtens * 5478a4a77718SDaniel Axtens * @skb: GSO skb 5479a4a77718SDaniel Axtens * 5480a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5481a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5482a4a77718SDaniel Axtens * headers (TCP/UDP). 5483a4a77718SDaniel Axtens */ 5484a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5485a4a77718SDaniel Axtens { 5486a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5487a4a77718SDaniel Axtens 5488a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5489a4a77718SDaniel Axtens } 54900d5501c1SVlad Yasevich 5491ae7ef81eSMarcelo Ricardo Leitner /** 54922b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 54932b16f048SDaniel Axtens * 54942b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 54952b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 54962b16f048SDaniel Axtens * 54972b16f048SDaniel Axtens * We might want to check: 54982b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 54992b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 55002b16f048SDaniel Axtens * 55012b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 55022b16f048SDaniel Axtens * 550349682bfaSMathieu Malaterre * @skb: GSO skb 550449682bfaSMathieu Malaterre * 55052b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 55062b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 55072b16f048SDaniel Axtens * 55082b16f048SDaniel Axtens * @max_len: The maximum permissible length. 55092b16f048SDaniel Axtens * 55102b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 55112b16f048SDaniel Axtens */ 55122b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 55132b16f048SDaniel Axtens unsigned int seg_len, 55142b16f048SDaniel Axtens unsigned int max_len) { 55152b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 55162b16f048SDaniel Axtens const struct sk_buff *iter; 55172b16f048SDaniel Axtens 55182b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 55192b16f048SDaniel Axtens return seg_len <= max_len; 55202b16f048SDaniel Axtens 55212b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 55222b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 55232b16f048SDaniel Axtens 55242b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 55252b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 55262b16f048SDaniel Axtens return false; 55272b16f048SDaniel Axtens } 55282b16f048SDaniel Axtens 55292b16f048SDaniel Axtens return true; 55302b16f048SDaniel Axtens } 55312b16f048SDaniel Axtens 55322b16f048SDaniel Axtens /** 5533779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5534ae7ef81eSMarcelo Ricardo Leitner * 5535ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 553676f21b99SDavid S. Miller * @mtu: MTU to validate against 5537ae7ef81eSMarcelo Ricardo Leitner * 5538779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5539779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5540779b7931SDaniel Axtens * payload. 5541ae7ef81eSMarcelo Ricardo Leitner */ 5542779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5543ae7ef81eSMarcelo Ricardo Leitner { 55442b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5545ae7ef81eSMarcelo Ricardo Leitner } 5546779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5547ae7ef81eSMarcelo Ricardo Leitner 55482b16f048SDaniel Axtens /** 55492b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 55502b16f048SDaniel Axtens * 55512b16f048SDaniel Axtens * @skb: GSO skb 55522b16f048SDaniel Axtens * @len: length to validate against 55532b16f048SDaniel Axtens * 55542b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 55552b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 55562b16f048SDaniel Axtens */ 55572b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 55582b16f048SDaniel Axtens { 55592b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 55602b16f048SDaniel Axtens } 55612b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 55622b16f048SDaniel Axtens 55630d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 55640d5501c1SVlad Yasevich { 5565d85e8be2SYuya Kusakabe int mac_len, meta_len; 5566d85e8be2SYuya Kusakabe void *meta; 55674bbb3e0eSToshiaki Makita 55680d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 55690d5501c1SVlad Yasevich kfree_skb(skb); 55700d5501c1SVlad Yasevich return NULL; 55710d5501c1SVlad Yasevich } 55720d5501c1SVlad Yasevich 55734bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5574ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 55754bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 55764bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5577ae474573SToshiaki Makita } 5578d85e8be2SYuya Kusakabe 5579d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5580d85e8be2SYuya Kusakabe if (meta_len) { 5581d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5582d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5583d85e8be2SYuya Kusakabe } 5584d85e8be2SYuya Kusakabe 55850d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 55860d5501c1SVlad Yasevich return skb; 55870d5501c1SVlad Yasevich } 55880d5501c1SVlad Yasevich 55890d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 55900d5501c1SVlad Yasevich { 55910d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 55920d5501c1SVlad Yasevich u16 vlan_tci; 55930d5501c1SVlad Yasevich 5594df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 55950d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 55960d5501c1SVlad Yasevich return skb; 55970d5501c1SVlad Yasevich } 55980d5501c1SVlad Yasevich 55990d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 56000d5501c1SVlad Yasevich if (unlikely(!skb)) 56010d5501c1SVlad Yasevich goto err_free; 560255eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 560355eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 56040d5501c1SVlad Yasevich goto err_free; 56050d5501c1SVlad Yasevich 56060d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 56070d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 56080d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 56090d5501c1SVlad Yasevich 56100d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 56110d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 56120d5501c1SVlad Yasevich 56130d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 56140d5501c1SVlad Yasevich if (unlikely(!skb)) 56150d5501c1SVlad Yasevich goto err_free; 56160d5501c1SVlad Yasevich 56170d5501c1SVlad Yasevich skb_reset_network_header(skb); 56188be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 56190d5501c1SVlad Yasevich skb_reset_transport_header(skb); 56200d5501c1SVlad Yasevich skb_reset_mac_len(skb); 56210d5501c1SVlad Yasevich 56220d5501c1SVlad Yasevich return skb; 56230d5501c1SVlad Yasevich 56240d5501c1SVlad Yasevich err_free: 56250d5501c1SVlad Yasevich kfree_skb(skb); 56260d5501c1SVlad Yasevich return NULL; 56270d5501c1SVlad Yasevich } 56280d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 56292e4e4410SEric Dumazet 563092ece280SLiu Jian int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len) 5631e2195121SJiri Pirko { 5632e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5633e2195121SJiri Pirko return -ENOMEM; 5634e2195121SJiri Pirko 5635e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5636e2195121SJiri Pirko return 0; 5637e2195121SJiri Pirko 5638e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5639e2195121SJiri Pirko } 5640e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5641e2195121SJiri Pirko 5642bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5643bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5644bfca4c52SShmulik Ladkani */ 5645bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 564693515d53SJiri Pirko { 564793515d53SJiri Pirko struct vlan_hdr *vhdr; 5648b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 564993515d53SJiri Pirko int err; 565093515d53SJiri Pirko 5651b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5652b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5653b6a79208SShmulik Ladkani offset)) { 5654b6a79208SShmulik Ladkani return -EINVAL; 5655b6a79208SShmulik Ladkani } 5656b6a79208SShmulik Ladkani 565793515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 565893515d53SJiri Pirko if (unlikely(err)) 5659b6a79208SShmulik Ladkani return err; 566093515d53SJiri Pirko 566193515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 566293515d53SJiri Pirko 566393515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 566493515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 566593515d53SJiri Pirko 566693515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 566793515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 566893515d53SJiri Pirko 566993515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 567093515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 567193515d53SJiri Pirko 567293515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 567393515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 567493515d53SJiri Pirko 567593515d53SJiri Pirko skb_reset_mac_len(skb); 567693515d53SJiri Pirko 567793515d53SJiri Pirko return err; 567893515d53SJiri Pirko } 5679bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 568093515d53SJiri Pirko 5681b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5682b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5683b6a79208SShmulik Ladkani */ 568493515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 568593515d53SJiri Pirko { 568693515d53SJiri Pirko u16 vlan_tci; 568793515d53SJiri Pirko __be16 vlan_proto; 568893515d53SJiri Pirko int err; 568993515d53SJiri Pirko 5690df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5691b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 569293515d53SJiri Pirko } else { 5693ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 569493515d53SJiri Pirko return 0; 569593515d53SJiri Pirko 569693515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 569793515d53SJiri Pirko if (err) 569893515d53SJiri Pirko return err; 569993515d53SJiri Pirko } 570093515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5701ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 570293515d53SJiri Pirko return 0; 570393515d53SJiri Pirko 570493515d53SJiri Pirko vlan_proto = skb->protocol; 570593515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 570693515d53SJiri Pirko if (unlikely(err)) 570793515d53SJiri Pirko return err; 570893515d53SJiri Pirko 570993515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 571093515d53SJiri Pirko return 0; 571193515d53SJiri Pirko } 571293515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 571393515d53SJiri Pirko 5714b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5715b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5716b6a79208SShmulik Ladkani */ 571793515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 571893515d53SJiri Pirko { 5719df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5720b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 572193515d53SJiri Pirko int err; 572293515d53SJiri Pirko 5723b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5724b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5725b6a79208SShmulik Ladkani offset)) { 5726b6a79208SShmulik Ladkani return -EINVAL; 5727b6a79208SShmulik Ladkani } 5728b6a79208SShmulik Ladkani 572993515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5730df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5731b6a79208SShmulik Ladkani if (err) 573293515d53SJiri Pirko return err; 57339241e2dfSDaniel Borkmann 573493515d53SJiri Pirko skb->protocol = skb->vlan_proto; 573593515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 573693515d53SJiri Pirko 57376b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 573893515d53SJiri Pirko } 573993515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 574093515d53SJiri Pirko return 0; 574193515d53SJiri Pirko } 574293515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 574393515d53SJiri Pirko 574419fbcb36SGuillaume Nault /** 574519fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 574619fbcb36SGuillaume Nault * 574719fbcb36SGuillaume Nault * @skb: Socket buffer to modify 574819fbcb36SGuillaume Nault * 574919fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 575019fbcb36SGuillaume Nault * 575119fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 575219fbcb36SGuillaume Nault * present. 575319fbcb36SGuillaume Nault * 575419fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 575519fbcb36SGuillaume Nault */ 575619fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 575719fbcb36SGuillaume Nault { 575819fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 575919fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 576019fbcb36SGuillaume Nault return -EPROTO; 576119fbcb36SGuillaume Nault 576219fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 576319fbcb36SGuillaume Nault skb_reset_mac_header(skb); 576419fbcb36SGuillaume Nault skb_reset_mac_len(skb); 576519fbcb36SGuillaume Nault 576619fbcb36SGuillaume Nault return 0; 576719fbcb36SGuillaume Nault } 576819fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 576919fbcb36SGuillaume Nault 577019fbcb36SGuillaume Nault /** 577119fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 577219fbcb36SGuillaume Nault * 577319fbcb36SGuillaume Nault * @skb: Socket buffer to modify 577419fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 577519fbcb36SGuillaume Nault * @src: Source MAC address of the new header 577619fbcb36SGuillaume Nault * 577719fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 577819fbcb36SGuillaume Nault * 577919fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 578019fbcb36SGuillaume Nault * 578119fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 578219fbcb36SGuillaume Nault */ 578319fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 578419fbcb36SGuillaume Nault const unsigned char *src) 578519fbcb36SGuillaume Nault { 578619fbcb36SGuillaume Nault struct ethhdr *eth; 578719fbcb36SGuillaume Nault int err; 578819fbcb36SGuillaume Nault 578919fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 579019fbcb36SGuillaume Nault return -EPROTO; 579119fbcb36SGuillaume Nault 579219fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 579319fbcb36SGuillaume Nault if (err < 0) 579419fbcb36SGuillaume Nault return err; 579519fbcb36SGuillaume Nault 579619fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 579719fbcb36SGuillaume Nault skb_reset_mac_header(skb); 579819fbcb36SGuillaume Nault skb_reset_mac_len(skb); 579919fbcb36SGuillaume Nault 580019fbcb36SGuillaume Nault eth = eth_hdr(skb); 580119fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 580219fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 580319fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 580419fbcb36SGuillaume Nault 580519fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 580619fbcb36SGuillaume Nault 580719fbcb36SGuillaume Nault return 0; 580819fbcb36SGuillaume Nault } 580919fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 581019fbcb36SGuillaume Nault 58118822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 58128822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 58138822e270SJohn Hurley __be16 ethertype) 58148822e270SJohn Hurley { 58158822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 58168822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 58178822e270SJohn Hurley 58188822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 58198822e270SJohn Hurley } 58208822e270SJohn Hurley 58218822e270SJohn Hurley hdr->h_proto = ethertype; 58228822e270SJohn Hurley } 58238822e270SJohn Hurley 58248822e270SJohn Hurley /** 5825e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5826e7dbfed1SMartin Varghese * the packet 58278822e270SJohn Hurley * 58288822e270SJohn Hurley * @skb: buffer 58298822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 58308822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5831fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5832e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5833e7dbfed1SMartin Varghese * ethernet 58348822e270SJohn Hurley * 58358822e270SJohn Hurley * Expects skb->data at mac header. 58368822e270SJohn Hurley * 58378822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 58388822e270SJohn Hurley */ 5839fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5840d04ac224SMartin Varghese int mac_len, bool ethernet) 58418822e270SJohn Hurley { 58428822e270SJohn Hurley struct mpls_shim_hdr *lse; 58438822e270SJohn Hurley int err; 58448822e270SJohn Hurley 58458822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 58468822e270SJohn Hurley return -EINVAL; 58478822e270SJohn Hurley 58488822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 58498822e270SJohn Hurley if (skb->encapsulation) 58508822e270SJohn Hurley return -EINVAL; 58518822e270SJohn Hurley 58528822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 58538822e270SJohn Hurley if (unlikely(err)) 58548822e270SJohn Hurley return err; 58558822e270SJohn Hurley 58568822e270SJohn Hurley if (!skb->inner_protocol) { 5857e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 58588822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 58598822e270SJohn Hurley } 58608822e270SJohn Hurley 58618822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 58628822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5863fa4e0f88SDavide Caratti mac_len); 58648822e270SJohn Hurley skb_reset_mac_header(skb); 5865fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5866e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 58678822e270SJohn Hurley 58688822e270SJohn Hurley lse = mpls_hdr(skb); 58698822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 58708822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 58718822e270SJohn Hurley 58724296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 58738822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 58748822e270SJohn Hurley skb->protocol = mpls_proto; 58758822e270SJohn Hurley 58768822e270SJohn Hurley return 0; 58778822e270SJohn Hurley } 58788822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 58798822e270SJohn Hurley 58802e4e4410SEric Dumazet /** 5881ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5882ed246ceeSJohn Hurley * 5883ed246ceeSJohn Hurley * @skb: buffer 5884ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5885fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 588676f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5887ed246ceeSJohn Hurley * 5888ed246ceeSJohn Hurley * Expects skb->data at mac header. 5889ed246ceeSJohn Hurley * 5890ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5891ed246ceeSJohn Hurley */ 5892040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5893040b5cfbSMartin Varghese bool ethernet) 5894ed246ceeSJohn Hurley { 5895ed246ceeSJohn Hurley int err; 5896ed246ceeSJohn Hurley 5897ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5898dedc5a08SDavide Caratti return 0; 5899ed246ceeSJohn Hurley 5900fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5901ed246ceeSJohn Hurley if (unlikely(err)) 5902ed246ceeSJohn Hurley return err; 5903ed246ceeSJohn Hurley 5904ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5905ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5906fa4e0f88SDavide Caratti mac_len); 5907ed246ceeSJohn Hurley 5908ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5909ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5910fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5911ed246ceeSJohn Hurley 59124296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5913ed246ceeSJohn Hurley struct ethhdr *hdr; 5914ed246ceeSJohn Hurley 5915ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5916ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5917ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5918ed246ceeSJohn Hurley } 5919ed246ceeSJohn Hurley skb->protocol = next_proto; 5920ed246ceeSJohn Hurley 5921ed246ceeSJohn Hurley return 0; 5922ed246ceeSJohn Hurley } 5923ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5924ed246ceeSJohn Hurley 5925ed246ceeSJohn Hurley /** 5926d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5927d27cf5c5SJohn Hurley * 5928d27cf5c5SJohn Hurley * @skb: buffer 5929d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5930d27cf5c5SJohn Hurley * 5931d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5932d27cf5c5SJohn Hurley * 5933d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5934d27cf5c5SJohn Hurley */ 5935d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5936d27cf5c5SJohn Hurley { 5937d27cf5c5SJohn Hurley int err; 5938d27cf5c5SJohn Hurley 5939d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5940d27cf5c5SJohn Hurley return -EINVAL; 5941d27cf5c5SJohn Hurley 5942d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5943d27cf5c5SJohn Hurley if (unlikely(err)) 5944d27cf5c5SJohn Hurley return err; 5945d27cf5c5SJohn Hurley 5946d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5947d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5948d27cf5c5SJohn Hurley 5949d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5950d27cf5c5SJohn Hurley } 5951d27cf5c5SJohn Hurley 5952d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 5953d27cf5c5SJohn Hurley 5954d27cf5c5SJohn Hurley return 0; 5955d27cf5c5SJohn Hurley } 5956d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 5957d27cf5c5SJohn Hurley 5958d27cf5c5SJohn Hurley /** 59592a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 59602a2ea508SJohn Hurley * 59612a2ea508SJohn Hurley * @skb: buffer 59622a2ea508SJohn Hurley * 59632a2ea508SJohn Hurley * Expects skb->data at mac header. 59642a2ea508SJohn Hurley * 59652a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 59662a2ea508SJohn Hurley */ 59672a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 59682a2ea508SJohn Hurley { 59692a2ea508SJohn Hurley u32 lse; 59702a2ea508SJohn Hurley u8 ttl; 59712a2ea508SJohn Hurley 59722a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 59732a2ea508SJohn Hurley return -EINVAL; 59742a2ea508SJohn Hurley 597513de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 597613de4ed9SDavide Caratti return -ENOMEM; 597713de4ed9SDavide Caratti 59782a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 59792a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 59802a2ea508SJohn Hurley if (!--ttl) 59812a2ea508SJohn Hurley return -EINVAL; 59822a2ea508SJohn Hurley 59832a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 59842a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 59852a2ea508SJohn Hurley 59862a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 59872a2ea508SJohn Hurley } 59882a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 59892a2ea508SJohn Hurley 59902a2ea508SJohn Hurley /** 59912e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 59922e4e4410SEric Dumazet * 5993de3f0d0eSMasanari Iida * @header_len: size of linear part 5994de3f0d0eSMasanari Iida * @data_len: needed length in frags 5995de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5996de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5997de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 59982e4e4410SEric Dumazet * 59992e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 60002e4e4410SEric Dumazet */ 60012e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 60022e4e4410SEric Dumazet unsigned long data_len, 60032e4e4410SEric Dumazet int max_page_order, 60042e4e4410SEric Dumazet int *errcode, 60052e4e4410SEric Dumazet gfp_t gfp_mask) 60062e4e4410SEric Dumazet { 60072e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 60082e4e4410SEric Dumazet unsigned long chunk; 60092e4e4410SEric Dumazet struct sk_buff *skb; 60102e4e4410SEric Dumazet struct page *page; 60112e4e4410SEric Dumazet int i; 60122e4e4410SEric Dumazet 60132e4e4410SEric Dumazet *errcode = -EMSGSIZE; 60142e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 60152e4e4410SEric Dumazet * high order pages... 60162e4e4410SEric Dumazet */ 60172e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 60182e4e4410SEric Dumazet return NULL; 60192e4e4410SEric Dumazet 60202e4e4410SEric Dumazet *errcode = -ENOBUFS; 6021f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 60222e4e4410SEric Dumazet if (!skb) 60232e4e4410SEric Dumazet return NULL; 60242e4e4410SEric Dumazet 60252e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 60262e4e4410SEric Dumazet 60272e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 60282e4e4410SEric Dumazet int order = max_page_order; 60292e4e4410SEric Dumazet 60302e4e4410SEric Dumazet while (order) { 60312e4e4410SEric Dumazet if (npages >= 1 << order) { 6032d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 60332e4e4410SEric Dumazet __GFP_COMP | 6034d14b56f5SMichal Hocko __GFP_NOWARN, 60352e4e4410SEric Dumazet order); 60362e4e4410SEric Dumazet if (page) 60372e4e4410SEric Dumazet goto fill_page; 60382e4e4410SEric Dumazet /* Do not retry other high order allocations */ 60392e4e4410SEric Dumazet order = 1; 60402e4e4410SEric Dumazet max_page_order = 0; 60412e4e4410SEric Dumazet } 60422e4e4410SEric Dumazet order--; 60432e4e4410SEric Dumazet } 60442e4e4410SEric Dumazet page = alloc_page(gfp_mask); 60452e4e4410SEric Dumazet if (!page) 60462e4e4410SEric Dumazet goto failure; 60472e4e4410SEric Dumazet fill_page: 60482e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 60492e4e4410SEric Dumazet PAGE_SIZE << order); 60502e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 60512e4e4410SEric Dumazet data_len -= chunk; 60522e4e4410SEric Dumazet npages -= 1 << order; 60532e4e4410SEric Dumazet } 60542e4e4410SEric Dumazet return skb; 60552e4e4410SEric Dumazet 60562e4e4410SEric Dumazet failure: 60572e4e4410SEric Dumazet kfree_skb(skb); 60582e4e4410SEric Dumazet return NULL; 60592e4e4410SEric Dumazet } 60602e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 60616fa01ccdSSowmini Varadhan 60626fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 60636fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 60646fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 60656fa01ccdSSowmini Varadhan { 60666fa01ccdSSowmini Varadhan int i; 60676fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 60686fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 60696fa01ccdSSowmini Varadhan u8 *data; 60706fa01ccdSSowmini Varadhan 60716fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 60726fa01ccdSSowmini Varadhan 60736fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 60746fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 60756fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 60766fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 60776fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 60786fa01ccdSSowmini Varadhan if (!data) 60796fa01ccdSSowmini Varadhan return -ENOMEM; 60806fa01ccdSSowmini Varadhan 60816fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 60826fa01ccdSSowmini Varadhan 60836fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 60846fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 60856fa01ccdSSowmini Varadhan skb->len -= off; 60866fa01ccdSSowmini Varadhan 60876fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 60886fa01ccdSSowmini Varadhan skb_shinfo(skb), 60896fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 60906fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 60916fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 60926fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 60936fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 60946fa01ccdSSowmini Varadhan kfree(data); 60956fa01ccdSSowmini Varadhan return -ENOMEM; 60966fa01ccdSSowmini Varadhan } 60976fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 60986fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 60996fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 61006fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 61016fa01ccdSSowmini Varadhan skb_release_data(skb); 61026fa01ccdSSowmini Varadhan } else { 61036fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 61046fa01ccdSSowmini Varadhan * relocate values 61056fa01ccdSSowmini Varadhan */ 61066fa01ccdSSowmini Varadhan skb_free_head(skb); 61076fa01ccdSSowmini Varadhan } 61086fa01ccdSSowmini Varadhan 61096fa01ccdSSowmini Varadhan skb->head = data; 61106fa01ccdSSowmini Varadhan skb->data = data; 61116fa01ccdSSowmini Varadhan skb->head_frag = 0; 6112763087daSEric Dumazet skb_set_end_offset(skb, size); 61136fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 61146fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 61156fa01ccdSSowmini Varadhan skb->cloned = 0; 61166fa01ccdSSowmini Varadhan skb->hdr_len = 0; 61176fa01ccdSSowmini Varadhan skb->nohdr = 0; 61186fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 61196fa01ccdSSowmini Varadhan 61206fa01ccdSSowmini Varadhan return 0; 61216fa01ccdSSowmini Varadhan } 61226fa01ccdSSowmini Varadhan 61236fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 61246fa01ccdSSowmini Varadhan 61256fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 61266fa01ccdSSowmini Varadhan * pskb_carve() 61276fa01ccdSSowmini Varadhan */ 61286fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 61296fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 61306fa01ccdSSowmini Varadhan gfp_t gfp_mask) 61316fa01ccdSSowmini Varadhan { 61326fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 61336fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 61346fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 61356fa01ccdSSowmini Varadhan 61366fa01ccdSSowmini Varadhan do { 61376fa01ccdSSowmini Varadhan if (!list) { 61386fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 61396fa01ccdSSowmini Varadhan return -EFAULT; 61406fa01ccdSSowmini Varadhan } 61416fa01ccdSSowmini Varadhan if (list->len <= eat) { 61426fa01ccdSSowmini Varadhan /* Eaten as whole. */ 61436fa01ccdSSowmini Varadhan eat -= list->len; 61446fa01ccdSSowmini Varadhan list = list->next; 61456fa01ccdSSowmini Varadhan insp = list; 61466fa01ccdSSowmini Varadhan } else { 61476fa01ccdSSowmini Varadhan /* Eaten partially. */ 61486fa01ccdSSowmini Varadhan if (skb_shared(list)) { 61496fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 61506fa01ccdSSowmini Varadhan if (!clone) 61516fa01ccdSSowmini Varadhan return -ENOMEM; 61526fa01ccdSSowmini Varadhan insp = list->next; 61536fa01ccdSSowmini Varadhan list = clone; 61546fa01ccdSSowmini Varadhan } else { 61556fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 61566fa01ccdSSowmini Varadhan insp = list; 61576fa01ccdSSowmini Varadhan } 61586fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 61596fa01ccdSSowmini Varadhan kfree_skb(clone); 61606fa01ccdSSowmini Varadhan return -ENOMEM; 61616fa01ccdSSowmini Varadhan } 61626fa01ccdSSowmini Varadhan break; 61636fa01ccdSSowmini Varadhan } 61646fa01ccdSSowmini Varadhan } while (eat); 61656fa01ccdSSowmini Varadhan 61666fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 61676fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 61686fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 6169ef527f96SEric Dumazet consume_skb(list); 61706fa01ccdSSowmini Varadhan } 61716fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 61726fa01ccdSSowmini Varadhan if (clone) { 61736fa01ccdSSowmini Varadhan clone->next = list; 61746fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 61756fa01ccdSSowmini Varadhan } 61766fa01ccdSSowmini Varadhan return 0; 61776fa01ccdSSowmini Varadhan } 61786fa01ccdSSowmini Varadhan 61796fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 61806fa01ccdSSowmini Varadhan * non-linear part of skb 61816fa01ccdSSowmini Varadhan */ 61826fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 61836fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 61846fa01ccdSSowmini Varadhan { 61856fa01ccdSSowmini Varadhan int i, k = 0; 61866fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 61876fa01ccdSSowmini Varadhan u8 *data; 61886fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 61896fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 61906fa01ccdSSowmini Varadhan 61916fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 61926fa01ccdSSowmini Varadhan 61936fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 61946fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 61956fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 61966fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 61976fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 61986fa01ccdSSowmini Varadhan if (!data) 61996fa01ccdSSowmini Varadhan return -ENOMEM; 62006fa01ccdSSowmini Varadhan 62016fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 62026fa01ccdSSowmini Varadhan 62036fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6204e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 62056fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 62066fa01ccdSSowmini Varadhan kfree(data); 62076fa01ccdSSowmini Varadhan return -ENOMEM; 62086fa01ccdSSowmini Varadhan } 62096fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 62106fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 62116fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 62126fa01ccdSSowmini Varadhan 62136fa01ccdSSowmini Varadhan if (pos + fsize > off) { 62146fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 62156fa01ccdSSowmini Varadhan 62166fa01ccdSSowmini Varadhan if (pos < off) { 62176fa01ccdSSowmini Varadhan /* Split frag. 62186fa01ccdSSowmini Varadhan * We have two variants in this case: 62196fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 62206fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 62216fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 62226fa01ccdSSowmini Varadhan * where splitting is expensive. 62236fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 62246fa01ccdSSowmini Varadhan */ 6225b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 62266fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 62276fa01ccdSSowmini Varadhan } 62286fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 62296fa01ccdSSowmini Varadhan k++; 62306fa01ccdSSowmini Varadhan } 62316fa01ccdSSowmini Varadhan pos += fsize; 62326fa01ccdSSowmini Varadhan } 62336fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 62346fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 62356fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 62366fa01ccdSSowmini Varadhan 62376fa01ccdSSowmini Varadhan /* split line is in frag list */ 6238eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6239eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6240eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6241eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6242eabe8618SMiaohe Lin kfree(data); 6243eabe8618SMiaohe Lin return -ENOMEM; 62446fa01ccdSSowmini Varadhan } 62456fa01ccdSSowmini Varadhan skb_release_data(skb); 62466fa01ccdSSowmini Varadhan 62476fa01ccdSSowmini Varadhan skb->head = data; 62486fa01ccdSSowmini Varadhan skb->head_frag = 0; 62496fa01ccdSSowmini Varadhan skb->data = data; 6250763087daSEric Dumazet skb_set_end_offset(skb, size); 62516fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 62526fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 62536fa01ccdSSowmini Varadhan skb->cloned = 0; 62546fa01ccdSSowmini Varadhan skb->hdr_len = 0; 62556fa01ccdSSowmini Varadhan skb->nohdr = 0; 62566fa01ccdSSowmini Varadhan skb->len -= off; 62576fa01ccdSSowmini Varadhan skb->data_len = skb->len; 62586fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 62596fa01ccdSSowmini Varadhan return 0; 62606fa01ccdSSowmini Varadhan } 62616fa01ccdSSowmini Varadhan 62626fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 62636fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 62646fa01ccdSSowmini Varadhan { 62656fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 62666fa01ccdSSowmini Varadhan 62676fa01ccdSSowmini Varadhan if (len < headlen) 62686fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 62696fa01ccdSSowmini Varadhan else 62706fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 62716fa01ccdSSowmini Varadhan } 62726fa01ccdSSowmini Varadhan 62736fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 62746fa01ccdSSowmini Varadhan * a new skb 62756fa01ccdSSowmini Varadhan */ 62766fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 62776fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 62786fa01ccdSSowmini Varadhan { 62796fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 62806fa01ccdSSowmini Varadhan 62816fa01ccdSSowmini Varadhan if (!clone) 62826fa01ccdSSowmini Varadhan return NULL; 62836fa01ccdSSowmini Varadhan 62846fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 62856fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 62866fa01ccdSSowmini Varadhan kfree_skb(clone); 62876fa01ccdSSowmini Varadhan return NULL; 62886fa01ccdSSowmini Varadhan } 62896fa01ccdSSowmini Varadhan return clone; 62906fa01ccdSSowmini Varadhan } 62916fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6292c8c8b127SEric Dumazet 6293c8c8b127SEric Dumazet /** 6294c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6295c8c8b127SEric Dumazet * @skb: buffer 6296c8c8b127SEric Dumazet * 6297c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6298c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6299c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6300c8c8b127SEric Dumazet * truesize. 6301c8c8b127SEric Dumazet * Notes: 6302c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6303c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6304c8c8b127SEric Dumazet */ 6305c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6306c8c8b127SEric Dumazet { 63073174fed9SEric Dumazet if (skb->data_len) { 63083174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6309c8c8b127SEric Dumazet skb_cloned(skb)) 6310c8c8b127SEric Dumazet return; 6311c8c8b127SEric Dumazet 6312c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6313c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 63143174fed9SEric Dumazet } 63153174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 63163174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 63173174fed9SEric Dumazet * their truesize. 63183174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 63193174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 63203174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6321c8c8b127SEric Dumazet */ 6322c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6323c8c8b127SEric Dumazet } 6324df5042f4SFlorian Westphal 6325df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6326df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6327df5042f4SFlorian Westphal { 6328df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6329df5042f4SFlorian Westphal } 6330df5042f4SFlorian Westphal 63318b69a803SPaolo Abeni /** 63328b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 63338b69a803SPaolo Abeni * 63344930f483SFlorian Westphal * @flags: See kmalloc(). 63354930f483SFlorian Westphal * 63368b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 63378b69a803SPaolo Abeni * skb via __skb_ext_set(). 63388b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 63398b69a803SPaolo Abeni */ 63404930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6341df5042f4SFlorian Westphal { 63424930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6343df5042f4SFlorian Westphal 6344df5042f4SFlorian Westphal if (new) { 6345df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6346df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6347df5042f4SFlorian Westphal } 6348df5042f4SFlorian Westphal 6349df5042f4SFlorian Westphal return new; 6350df5042f4SFlorian Westphal } 6351df5042f4SFlorian Westphal 63524165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 63534165079bSFlorian Westphal unsigned int old_active) 6354df5042f4SFlorian Westphal { 6355df5042f4SFlorian Westphal struct skb_ext *new; 6356df5042f4SFlorian Westphal 6357df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6358df5042f4SFlorian Westphal return old; 6359df5042f4SFlorian Westphal 6360df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6361df5042f4SFlorian Westphal if (!new) 6362df5042f4SFlorian Westphal return NULL; 6363df5042f4SFlorian Westphal 6364df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6365df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6366df5042f4SFlorian Westphal 63674165079bSFlorian Westphal #ifdef CONFIG_XFRM 63684165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 63694165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 63704165079bSFlorian Westphal unsigned int i; 63714165079bSFlorian Westphal 63724165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 63734165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 63744165079bSFlorian Westphal } 63754165079bSFlorian Westphal #endif 6376df5042f4SFlorian Westphal __skb_ext_put(old); 6377df5042f4SFlorian Westphal return new; 6378df5042f4SFlorian Westphal } 6379df5042f4SFlorian Westphal 6380df5042f4SFlorian Westphal /** 63818b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 63828b69a803SPaolo Abeni * @skb: buffer 63838b69a803SPaolo Abeni * @id: extension id 63848b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 63858b69a803SPaolo Abeni * 63868b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 63878b69a803SPaolo Abeni * 63888b69a803SPaolo Abeni * Returns the pointer to the extension. 63898b69a803SPaolo Abeni */ 63908b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 63918b69a803SPaolo Abeni struct skb_ext *ext) 63928b69a803SPaolo Abeni { 63938b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 63948b69a803SPaolo Abeni 63958b69a803SPaolo Abeni skb_ext_put(skb); 63968b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 63978b69a803SPaolo Abeni ext->chunks = newlen; 63988b69a803SPaolo Abeni ext->offset[id] = newoff; 63998b69a803SPaolo Abeni skb->extensions = ext; 64008b69a803SPaolo Abeni skb->active_extensions = 1 << id; 64018b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 64028b69a803SPaolo Abeni } 64038b69a803SPaolo Abeni 64048b69a803SPaolo Abeni /** 6405df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6406df5042f4SFlorian Westphal * @skb: buffer 6407df5042f4SFlorian Westphal * @id: extension to allocate space for 6408df5042f4SFlorian Westphal * 6409df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6410df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6411df5042f4SFlorian Westphal * is returned. 6412df5042f4SFlorian Westphal * 6413df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6414df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6415df5042f4SFlorian Westphal * 6416df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6417df5042f4SFlorian Westphal */ 6418df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6419df5042f4SFlorian Westphal { 6420df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6421df5042f4SFlorian Westphal unsigned int newlen, newoff; 6422df5042f4SFlorian Westphal 6423df5042f4SFlorian Westphal if (skb->active_extensions) { 6424df5042f4SFlorian Westphal old = skb->extensions; 6425df5042f4SFlorian Westphal 64264165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6427df5042f4SFlorian Westphal if (!new) 6428df5042f4SFlorian Westphal return NULL; 6429df5042f4SFlorian Westphal 6430682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6431df5042f4SFlorian Westphal goto set_active; 6432df5042f4SFlorian Westphal 6433e94e50bdSPaolo Abeni newoff = new->chunks; 6434df5042f4SFlorian Westphal } else { 6435df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6436df5042f4SFlorian Westphal 64374930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6438df5042f4SFlorian Westphal if (!new) 6439df5042f4SFlorian Westphal return NULL; 6440df5042f4SFlorian Westphal } 6441df5042f4SFlorian Westphal 6442df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6443df5042f4SFlorian Westphal new->chunks = newlen; 6444df5042f4SFlorian Westphal new->offset[id] = newoff; 6445df5042f4SFlorian Westphal set_active: 6446b0999f38SPaolo Abeni skb->slow_gro = 1; 6447682ec859SPaolo Abeni skb->extensions = new; 6448df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6449df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6450df5042f4SFlorian Westphal } 6451df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6452df5042f4SFlorian Westphal 64534165079bSFlorian Westphal #ifdef CONFIG_XFRM 64544165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 64554165079bSFlorian Westphal { 64564165079bSFlorian Westphal unsigned int i; 64574165079bSFlorian Westphal 64584165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 64594165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 64604165079bSFlorian Westphal } 64614165079bSFlorian Westphal #endif 64624165079bSFlorian Westphal 646378476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 646478476d31SJeremy Kerr static void skb_ext_put_mctp(struct mctp_flow *flow) 646578476d31SJeremy Kerr { 646678476d31SJeremy Kerr if (flow->key) 646778476d31SJeremy Kerr mctp_key_unref(flow->key); 646878476d31SJeremy Kerr } 646978476d31SJeremy Kerr #endif 647078476d31SJeremy Kerr 6471df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6472df5042f4SFlorian Westphal { 6473df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6474df5042f4SFlorian Westphal 6475df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6476df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6477df5042f4SFlorian Westphal skb->extensions = NULL; 6478df5042f4SFlorian Westphal __skb_ext_put(ext); 64794165079bSFlorian Westphal #ifdef CONFIG_XFRM 64804165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 64814165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 64824165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 64834165079bSFlorian Westphal 64844165079bSFlorian Westphal skb_ext_put_sp(sp); 64854165079bSFlorian Westphal sp->len = 0; 64864165079bSFlorian Westphal #endif 6487df5042f4SFlorian Westphal } 6488df5042f4SFlorian Westphal } 6489df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6490df5042f4SFlorian Westphal 6491df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6492df5042f4SFlorian Westphal { 6493df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6494df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6495df5042f4SFlorian Westphal */ 6496df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6497df5042f4SFlorian Westphal goto free_now; 6498df5042f4SFlorian Westphal 6499df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6500df5042f4SFlorian Westphal return; 6501df5042f4SFlorian Westphal free_now: 65024165079bSFlorian Westphal #ifdef CONFIG_XFRM 65034165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 65044165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 65054165079bSFlorian Westphal #endif 650678476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 650778476d31SJeremy Kerr if (__skb_ext_exist(ext, SKB_EXT_MCTP)) 650878476d31SJeremy Kerr skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP)); 650978476d31SJeremy Kerr #endif 65104165079bSFlorian Westphal 6511df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6512df5042f4SFlorian Westphal } 6513df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6514df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 651568822bdfSEric Dumazet 651668822bdfSEric Dumazet /** 651768822bdfSEric Dumazet * skb_attempt_defer_free - queue skb for remote freeing 651868822bdfSEric Dumazet * @skb: buffer 651968822bdfSEric Dumazet * 652068822bdfSEric Dumazet * Put @skb in a per-cpu list, using the cpu which 652168822bdfSEric Dumazet * allocated the skb/pages to reduce false sharing 652268822bdfSEric Dumazet * and memory zone spinlock contention. 652368822bdfSEric Dumazet */ 652468822bdfSEric Dumazet void skb_attempt_defer_free(struct sk_buff *skb) 652568822bdfSEric Dumazet { 652668822bdfSEric Dumazet int cpu = skb->alloc_cpu; 652768822bdfSEric Dumazet struct softnet_data *sd; 652868822bdfSEric Dumazet unsigned long flags; 652939564c3fSEric Dumazet unsigned int defer_max; 653068822bdfSEric Dumazet bool kick; 653168822bdfSEric Dumazet 653268822bdfSEric Dumazet if (WARN_ON_ONCE(cpu >= nr_cpu_ids) || 653368822bdfSEric Dumazet !cpu_online(cpu) || 653468822bdfSEric Dumazet cpu == raw_smp_processor_id()) { 653539564c3fSEric Dumazet nodefer: __kfree_skb(skb); 653668822bdfSEric Dumazet return; 653768822bdfSEric Dumazet } 653868822bdfSEric Dumazet 653968822bdfSEric Dumazet sd = &per_cpu(softnet_data, cpu); 654039564c3fSEric Dumazet defer_max = READ_ONCE(sysctl_skb_defer_max); 654139564c3fSEric Dumazet if (READ_ONCE(sd->defer_count) >= defer_max) 654239564c3fSEric Dumazet goto nodefer; 654339564c3fSEric Dumazet 654468822bdfSEric Dumazet spin_lock_irqsave(&sd->defer_lock, flags); 6545c09b0cd2SJakub Kicinski /* Send an IPI every time queue reaches half capacity. */ 6546c09b0cd2SJakub Kicinski kick = sd->defer_count == (defer_max >> 1); 6547c09b0cd2SJakub Kicinski /* Paired with the READ_ONCE() few lines above */ 6548c09b0cd2SJakub Kicinski WRITE_ONCE(sd->defer_count, sd->defer_count + 1); 6549c09b0cd2SJakub Kicinski 655068822bdfSEric Dumazet skb->next = sd->defer_list; 655168822bdfSEric Dumazet /* Paired with READ_ONCE() in skb_defer_free_flush() */ 655268822bdfSEric Dumazet WRITE_ONCE(sd->defer_list, skb); 655368822bdfSEric Dumazet spin_unlock_irqrestore(&sd->defer_lock, flags); 655468822bdfSEric Dumazet 655568822bdfSEric Dumazet /* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU 655668822bdfSEric Dumazet * if we are unlucky enough (this seems very unlikely). 655768822bdfSEric Dumazet */ 655897e719a8SEric Dumazet if (unlikely(kick) && !cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) 655968822bdfSEric Dumazet smp_call_function_single_async(cpu, &sd->defer_csd); 656068822bdfSEric Dumazet } 6561