12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 41da177e4SLinus Torvalds * 5113aa838SAlan Cox * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 61da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Fixes: 91da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 101da177e4SLinus Torvalds * balancer bugs. 111da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 121da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 131da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 141da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 151da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 161da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 171da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 181da177e4SLinus Torvalds * only put in the headers 191da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 201da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 211da177e4SLinus Torvalds * Andi Kleen : slabified it. 221da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 231da177e4SLinus Torvalds * 241da177e4SLinus Torvalds * NOTE: 251da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 261da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 271da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 281da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 331da177e4SLinus Torvalds */ 341da177e4SLinus Torvalds 35e005d193SJoe Perches #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 36e005d193SJoe Perches 371da177e4SLinus Torvalds #include <linux/module.h> 381da177e4SLinus Torvalds #include <linux/types.h> 391da177e4SLinus Torvalds #include <linux/kernel.h> 401da177e4SLinus Torvalds #include <linux/mm.h> 411da177e4SLinus Torvalds #include <linux/interrupt.h> 421da177e4SLinus Torvalds #include <linux/in.h> 431da177e4SLinus Torvalds #include <linux/inet.h> 441da177e4SLinus Torvalds #include <linux/slab.h> 45de960aa9SFlorian Westphal #include <linux/tcp.h> 46de960aa9SFlorian Westphal #include <linux/udp.h> 4790017accSMarcelo Ricardo Leitner #include <linux/sctp.h> 481da177e4SLinus Torvalds #include <linux/netdevice.h> 491da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 501da177e4SLinus Torvalds #include <net/pkt_sched.h> 511da177e4SLinus Torvalds #endif 521da177e4SLinus Torvalds #include <linux/string.h> 531da177e4SLinus Torvalds #include <linux/skbuff.h> 549c55e01cSJens Axboe #include <linux/splice.h> 551da177e4SLinus Torvalds #include <linux/cache.h> 561da177e4SLinus Torvalds #include <linux/rtnetlink.h> 571da177e4SLinus Torvalds #include <linux/init.h> 58716ea3a7SDavid Howells #include <linux/scatterlist.h> 59ac45f602SPatrick Ohly #include <linux/errqueue.h> 60268bb0ceSLinus Torvalds #include <linux/prefetch.h> 610d5501c1SVlad Yasevich #include <linux/if_vlan.h> 622a2ea508SJohn Hurley #include <linux/mpls.h> 63183f47fcSSebastian Andrzej Siewior #include <linux/kcov.h> 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds #include <net/protocol.h> 661da177e4SLinus Torvalds #include <net/dst.h> 671da177e4SLinus Torvalds #include <net/sock.h> 681da177e4SLinus Torvalds #include <net/checksum.h> 69ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 701da177e4SLinus Torvalds #include <net/xfrm.h> 718822e270SJohn Hurley #include <net/mpls.h> 723ee17bc7SMat Martineau #include <net/mptcp.h> 7378476d31SJeremy Kerr #include <net/mctp.h> 746a5bcd84SIlias Apalodimas #include <net/page_pool.h> 751da177e4SLinus Torvalds 767c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 77ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7851c56b00SEric Dumazet #include <linux/highmem.h> 79b245be1fSWillem de Bruijn #include <linux/capability.h> 80b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 812544af03SMatteo Croce #include <linux/indirect_call_wrapper.h> 82a1f8e7f7SAl Viro 837b7ed885SBart Van Assche #include "datagram.h" 847f678defSVasily Averin #include "sock_destructor.h" 857b7ed885SBart Van Assche 8608009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8708009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 88df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 89df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 90df5042f4SFlorian Westphal #endif 915f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 925f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /** 95f05de73bSJean Sacren * skb_panic - private function for out-of-line support 961da177e4SLinus Torvalds * @skb: buffer 971da177e4SLinus Torvalds * @sz: size 98f05de73bSJean Sacren * @addr: address 9999d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 1001da177e4SLinus Torvalds * 101f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 102f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 103f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 104f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1051da177e4SLinus Torvalds */ 106f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 10799d5851eSJames Hogan const char msg[]) 1081da177e4SLinus Torvalds { 10941a46913SJesper Dangaard Brouer pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", 11099d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1114305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 11226095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1131da177e4SLinus Torvalds BUG(); 1141da177e4SLinus Torvalds } 1151da177e4SLinus Torvalds 116f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1171da177e4SLinus Torvalds { 118f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1191da177e4SLinus Torvalds } 1201da177e4SLinus Torvalds 121f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 122f05de73bSJean Sacren { 123f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 124f05de73bSJean Sacren } 125c93bdd0eSMel Gorman 12650fad4b5SAlexander Lobakin #define NAPI_SKB_CACHE_SIZE 64 127f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_BULK 16 128f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) 12950fad4b5SAlexander Lobakin 13050fad4b5SAlexander Lobakin struct napi_alloc_cache { 13150fad4b5SAlexander Lobakin struct page_frag_cache page; 13250fad4b5SAlexander Lobakin unsigned int skb_count; 13350fad4b5SAlexander Lobakin void *skb_cache[NAPI_SKB_CACHE_SIZE]; 13450fad4b5SAlexander Lobakin }; 13550fad4b5SAlexander Lobakin 13650fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 13750fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 13850fad4b5SAlexander Lobakin 13932e3573fSYajun Deng void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 14050fad4b5SAlexander Lobakin { 14150fad4b5SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 14250fad4b5SAlexander Lobakin 14350fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 14450fad4b5SAlexander Lobakin 14532e3573fSYajun Deng return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 14650fad4b5SAlexander Lobakin } 14750fad4b5SAlexander Lobakin EXPORT_SYMBOL(__napi_alloc_frag_align); 14850fad4b5SAlexander Lobakin 14950fad4b5SAlexander Lobakin void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 15050fad4b5SAlexander Lobakin { 15150fad4b5SAlexander Lobakin void *data; 15250fad4b5SAlexander Lobakin 15350fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 154afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 15532e3573fSYajun Deng struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache); 15632e3573fSYajun Deng 15750fad4b5SAlexander Lobakin data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask); 15850fad4b5SAlexander Lobakin } else { 15932e3573fSYajun Deng struct napi_alloc_cache *nc; 16032e3573fSYajun Deng 16150fad4b5SAlexander Lobakin local_bh_disable(); 16232e3573fSYajun Deng nc = this_cpu_ptr(&napi_alloc_cache); 16332e3573fSYajun Deng data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 16450fad4b5SAlexander Lobakin local_bh_enable(); 16550fad4b5SAlexander Lobakin } 16650fad4b5SAlexander Lobakin return data; 16750fad4b5SAlexander Lobakin } 16850fad4b5SAlexander Lobakin EXPORT_SYMBOL(__netdev_alloc_frag_align); 16950fad4b5SAlexander Lobakin 170f450d539SAlexander Lobakin static struct sk_buff *napi_skb_cache_get(void) 171f450d539SAlexander Lobakin { 172f450d539SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 173f450d539SAlexander Lobakin struct sk_buff *skb; 174f450d539SAlexander Lobakin 175f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 176f450d539SAlexander Lobakin nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache, 177f450d539SAlexander Lobakin GFP_ATOMIC, 178f450d539SAlexander Lobakin NAPI_SKB_CACHE_BULK, 179f450d539SAlexander Lobakin nc->skb_cache); 180f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 181f450d539SAlexander Lobakin return NULL; 182f450d539SAlexander Lobakin 183f450d539SAlexander Lobakin skb = nc->skb_cache[--nc->skb_count]; 184f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, skb); 185f450d539SAlexander Lobakin 186f450d539SAlexander Lobakin return skb; 187f450d539SAlexander Lobakin } 188f450d539SAlexander Lobakin 189ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 190483126b3SAlexander Lobakin static void __build_skb_around(struct sk_buff *skb, void *data, 191483126b3SAlexander Lobakin unsigned int frag_size) 192ba0509b6SJesper Dangaard Brouer { 193ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 194ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 195ba0509b6SJesper Dangaard Brouer 196ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 197ba0509b6SJesper Dangaard Brouer 198ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 199ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 200ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 201ba0509b6SJesper Dangaard Brouer skb->head = data; 202ba0509b6SJesper Dangaard Brouer skb->data = data; 203ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 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; 207ba0509b6SJesper Dangaard Brouer 208ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 209ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 210ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 211ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 212ba0509b6SJesper Dangaard Brouer 2136370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 214ba0509b6SJesper Dangaard Brouer } 215ba0509b6SJesper Dangaard Brouer 2161da177e4SLinus Torvalds /** 2172ea2f62cSEric Dumazet * __build_skb - build a network buffer 218b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2192ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 220b2b5ce9dSEric Dumazet * 221b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 222deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2232ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2242ea2f62cSEric Dumazet * or vmalloc() 225b2b5ce9dSEric Dumazet * The return is the new skb buffer. 226b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 227b2b5ce9dSEric Dumazet * Notes : 228b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 229b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 230b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 231b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 232b2b5ce9dSEric Dumazet * before giving packet to stack. 233b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 234b2b5ce9dSEric Dumazet */ 2352ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 236b2b5ce9dSEric Dumazet { 237b2b5ce9dSEric Dumazet struct sk_buff *skb; 238b2b5ce9dSEric Dumazet 239b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 240ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 241b2b5ce9dSEric Dumazet return NULL; 242b2b5ce9dSEric Dumazet 243b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 244483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 245b2b5ce9dSEric Dumazet 246483126b3SAlexander Lobakin return skb; 247b2b5ce9dSEric Dumazet } 2482ea2f62cSEric Dumazet 2492ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 2502ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 2512ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 2522ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 2532ea2f62cSEric Dumazet */ 2542ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 2552ea2f62cSEric Dumazet { 2562ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 2572ea2f62cSEric Dumazet 2582ea2f62cSEric Dumazet if (skb && frag_size) { 2592ea2f62cSEric Dumazet skb->head_frag = 1; 2602f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 2612ea2f62cSEric Dumazet skb->pfmemalloc = 1; 2622ea2f62cSEric Dumazet } 2632ea2f62cSEric Dumazet return skb; 2642ea2f62cSEric Dumazet } 265b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 266b2b5ce9dSEric Dumazet 267ba0509b6SJesper Dangaard Brouer /** 268ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 269ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 270ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 271ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 272ba0509b6SJesper Dangaard Brouer */ 273ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 274ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 275ba0509b6SJesper Dangaard Brouer { 276ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 277ba0509b6SJesper Dangaard Brouer return NULL; 278ba0509b6SJesper Dangaard Brouer 279483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 280ba0509b6SJesper Dangaard Brouer 281483126b3SAlexander Lobakin if (frag_size) { 282ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 283ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 284ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 285ba0509b6SJesper Dangaard Brouer } 286ba0509b6SJesper Dangaard Brouer return skb; 287ba0509b6SJesper Dangaard Brouer } 288ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 289ba0509b6SJesper Dangaard Brouer 290f450d539SAlexander Lobakin /** 291f450d539SAlexander Lobakin * __napi_build_skb - build a network buffer 292f450d539SAlexander Lobakin * @data: data buffer provided by caller 293f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 294f450d539SAlexander Lobakin * 295f450d539SAlexander Lobakin * Version of __build_skb() that uses NAPI percpu caches to obtain 296f450d539SAlexander Lobakin * skbuff_head instead of inplace allocation. 297f450d539SAlexander Lobakin * 298f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 299f450d539SAlexander Lobakin */ 300f450d539SAlexander Lobakin static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size) 301f450d539SAlexander Lobakin { 302f450d539SAlexander Lobakin struct sk_buff *skb; 303f450d539SAlexander Lobakin 304f450d539SAlexander Lobakin skb = napi_skb_cache_get(); 305f450d539SAlexander Lobakin if (unlikely(!skb)) 306f450d539SAlexander Lobakin return NULL; 307f450d539SAlexander Lobakin 308f450d539SAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 309f450d539SAlexander Lobakin __build_skb_around(skb, data, frag_size); 310f450d539SAlexander Lobakin 311f450d539SAlexander Lobakin return skb; 312f450d539SAlexander Lobakin } 313f450d539SAlexander Lobakin 314f450d539SAlexander Lobakin /** 315f450d539SAlexander Lobakin * napi_build_skb - build a network buffer 316f450d539SAlexander Lobakin * @data: data buffer provided by caller 317f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 318f450d539SAlexander Lobakin * 319f450d539SAlexander Lobakin * Version of __napi_build_skb() that takes care of skb->head_frag 320f450d539SAlexander Lobakin * and skb->pfmemalloc when the data is a page or page fragment. 321f450d539SAlexander Lobakin * 322f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 323f450d539SAlexander Lobakin */ 324f450d539SAlexander Lobakin struct sk_buff *napi_build_skb(void *data, unsigned int frag_size) 325f450d539SAlexander Lobakin { 326f450d539SAlexander Lobakin struct sk_buff *skb = __napi_build_skb(data, frag_size); 327f450d539SAlexander Lobakin 328f450d539SAlexander Lobakin if (likely(skb) && frag_size) { 329f450d539SAlexander Lobakin skb->head_frag = 1; 330f450d539SAlexander Lobakin skb_propagate_pfmemalloc(virt_to_head_page(data), skb); 331f450d539SAlexander Lobakin } 332f450d539SAlexander Lobakin 333f450d539SAlexander Lobakin return skb; 334f450d539SAlexander Lobakin } 335f450d539SAlexander Lobakin EXPORT_SYMBOL(napi_build_skb); 336f450d539SAlexander Lobakin 3375381b23dSAlexander Lobakin /* 3385381b23dSAlexander Lobakin * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 3395381b23dSAlexander Lobakin * the caller if emergency pfmemalloc reserves are being used. If it is and 3405381b23dSAlexander Lobakin * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 3415381b23dSAlexander Lobakin * may be used. Otherwise, the packet data may be discarded until enough 3425381b23dSAlexander Lobakin * memory is free 3435381b23dSAlexander Lobakin */ 344ef28095fSAlexander Lobakin static void *kmalloc_reserve(size_t size, gfp_t flags, int node, 345ef28095fSAlexander Lobakin bool *pfmemalloc) 3465381b23dSAlexander Lobakin { 3475381b23dSAlexander Lobakin void *obj; 3485381b23dSAlexander Lobakin bool ret_pfmemalloc = false; 3495381b23dSAlexander Lobakin 3505381b23dSAlexander Lobakin /* 3515381b23dSAlexander Lobakin * Try a regular allocation, when that fails and we're not entitled 3525381b23dSAlexander Lobakin * to the reserves, fail. 3535381b23dSAlexander Lobakin */ 3545381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, 3555381b23dSAlexander Lobakin flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 3565381b23dSAlexander Lobakin node); 3575381b23dSAlexander Lobakin if (obj || !(gfp_pfmemalloc_allowed(flags))) 3585381b23dSAlexander Lobakin goto out; 3595381b23dSAlexander Lobakin 3605381b23dSAlexander Lobakin /* Try again but now we are using pfmemalloc reserves */ 3615381b23dSAlexander Lobakin ret_pfmemalloc = true; 3625381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, flags, node); 3635381b23dSAlexander Lobakin 3645381b23dSAlexander Lobakin out: 3655381b23dSAlexander Lobakin if (pfmemalloc) 3665381b23dSAlexander Lobakin *pfmemalloc = ret_pfmemalloc; 3675381b23dSAlexander Lobakin 3685381b23dSAlexander Lobakin return obj; 3695381b23dSAlexander Lobakin } 3705381b23dSAlexander Lobakin 3715381b23dSAlexander Lobakin /* Allocate a new skbuff. We do this ourselves so we can fill in a few 3725381b23dSAlexander Lobakin * 'private' fields and also do memory statistics to find all the 3735381b23dSAlexander Lobakin * [BEEP] leaks. 3745381b23dSAlexander Lobakin * 3755381b23dSAlexander Lobakin */ 3765381b23dSAlexander Lobakin 3775381b23dSAlexander Lobakin /** 3785381b23dSAlexander Lobakin * __alloc_skb - allocate a network buffer 3795381b23dSAlexander Lobakin * @size: size to allocate 3805381b23dSAlexander Lobakin * @gfp_mask: allocation mask 3815381b23dSAlexander Lobakin * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 3825381b23dSAlexander Lobakin * instead of head cache and allocate a cloned (child) skb. 3835381b23dSAlexander Lobakin * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 3845381b23dSAlexander Lobakin * allocations in case the data is required for writeback 3855381b23dSAlexander Lobakin * @node: numa node to allocate memory on 3865381b23dSAlexander Lobakin * 3875381b23dSAlexander Lobakin * Allocate a new &sk_buff. The returned buffer has no headroom and a 3885381b23dSAlexander Lobakin * tail room of at least size bytes. The object has a reference count 3895381b23dSAlexander Lobakin * of one. The return is the buffer. On a failure the return is %NULL. 3905381b23dSAlexander Lobakin * 3915381b23dSAlexander Lobakin * Buffers may only be allocated from interrupts using a @gfp_mask of 3925381b23dSAlexander Lobakin * %GFP_ATOMIC. 3935381b23dSAlexander Lobakin */ 3945381b23dSAlexander Lobakin struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 3955381b23dSAlexander Lobakin int flags, int node) 3965381b23dSAlexander Lobakin { 3975381b23dSAlexander Lobakin struct kmem_cache *cache; 3985381b23dSAlexander Lobakin struct sk_buff *skb; 399a5df6333SLi RongQing unsigned int osize; 4005381b23dSAlexander Lobakin bool pfmemalloc; 401a5df6333SLi RongQing u8 *data; 4025381b23dSAlexander Lobakin 4035381b23dSAlexander Lobakin cache = (flags & SKB_ALLOC_FCLONE) 4045381b23dSAlexander Lobakin ? skbuff_fclone_cache : skbuff_head_cache; 4055381b23dSAlexander Lobakin 4065381b23dSAlexander Lobakin if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 4075381b23dSAlexander Lobakin gfp_mask |= __GFP_MEMALLOC; 4085381b23dSAlexander Lobakin 4095381b23dSAlexander Lobakin /* Get the HEAD */ 410d13612b5SAlexander Lobakin if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI && 411d13612b5SAlexander Lobakin likely(node == NUMA_NO_NODE || node == numa_mem_id())) 412d13612b5SAlexander Lobakin skb = napi_skb_cache_get(); 413d13612b5SAlexander Lobakin else 414d13612b5SAlexander Lobakin skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node); 415df1ae022SAlexander Lobakin if (unlikely(!skb)) 416df1ae022SAlexander Lobakin return NULL; 4175381b23dSAlexander Lobakin prefetchw(skb); 4185381b23dSAlexander Lobakin 4195381b23dSAlexander Lobakin /* We do our best to align skb_shared_info on a separate cache 4205381b23dSAlexander Lobakin * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 4215381b23dSAlexander Lobakin * aligned memory blocks, unless SLUB/SLAB debug is enabled. 4225381b23dSAlexander Lobakin * Both skb->head and skb_shared_info are cache line aligned. 4235381b23dSAlexander Lobakin */ 4245381b23dSAlexander Lobakin size = SKB_DATA_ALIGN(size); 4255381b23dSAlexander Lobakin size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4265381b23dSAlexander Lobakin data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 427df1ae022SAlexander Lobakin if (unlikely(!data)) 4285381b23dSAlexander Lobakin goto nodata; 4295381b23dSAlexander Lobakin /* kmalloc(size) might give us more room than requested. 4305381b23dSAlexander Lobakin * Put skb_shared_info exactly at the end of allocated zone, 4315381b23dSAlexander Lobakin * to allow max possible filling before reallocation. 4325381b23dSAlexander Lobakin */ 433a5df6333SLi RongQing osize = ksize(data); 434a5df6333SLi RongQing size = SKB_WITH_OVERHEAD(osize); 4355381b23dSAlexander Lobakin prefetchw(data + size); 4365381b23dSAlexander Lobakin 4375381b23dSAlexander Lobakin /* 4385381b23dSAlexander Lobakin * Only clear those fields we need to clear, not those that we will 4395381b23dSAlexander Lobakin * actually initialise below. Hence, don't put any more fields after 4405381b23dSAlexander Lobakin * the tail pointer in struct sk_buff! 4415381b23dSAlexander Lobakin */ 4425381b23dSAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 443a5df6333SLi RongQing __build_skb_around(skb, data, osize); 4445381b23dSAlexander Lobakin skb->pfmemalloc = pfmemalloc; 4455381b23dSAlexander Lobakin 4465381b23dSAlexander Lobakin if (flags & SKB_ALLOC_FCLONE) { 4475381b23dSAlexander Lobakin struct sk_buff_fclones *fclones; 4485381b23dSAlexander Lobakin 4495381b23dSAlexander Lobakin fclones = container_of(skb, struct sk_buff_fclones, skb1); 4505381b23dSAlexander Lobakin 4515381b23dSAlexander Lobakin skb->fclone = SKB_FCLONE_ORIG; 4525381b23dSAlexander Lobakin refcount_set(&fclones->fclone_ref, 1); 4535381b23dSAlexander Lobakin 4545381b23dSAlexander Lobakin fclones->skb2.fclone = SKB_FCLONE_CLONE; 4555381b23dSAlexander Lobakin } 4565381b23dSAlexander Lobakin 4575381b23dSAlexander Lobakin return skb; 458df1ae022SAlexander Lobakin 4595381b23dSAlexander Lobakin nodata: 4605381b23dSAlexander Lobakin kmem_cache_free(cache, skb); 461df1ae022SAlexander Lobakin return NULL; 4625381b23dSAlexander Lobakin } 4635381b23dSAlexander Lobakin EXPORT_SYMBOL(__alloc_skb); 4645381b23dSAlexander Lobakin 4657ba7aeabSSebastian Andrzej Siewior /** 466fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 467fd11a83dSAlexander Duyck * @dev: network device to receive on 468d7499160SMasanari Iida * @len: length to allocate 469fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 470fd11a83dSAlexander Duyck * 471fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 472fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 473fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 474fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 475fd11a83dSAlexander Duyck * 476fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 477fd11a83dSAlexander Duyck */ 4789451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4799451980aSAlexander Duyck gfp_t gfp_mask) 480fd11a83dSAlexander Duyck { 481b63ae8caSAlexander Duyck struct page_frag_cache *nc; 482fd11a83dSAlexander Duyck struct sk_buff *skb; 4839451980aSAlexander Duyck bool pfmemalloc; 4849451980aSAlexander Duyck void *data; 485fd11a83dSAlexander Duyck 4869451980aSAlexander Duyck len += NET_SKB_PAD; 487fd11a83dSAlexander Duyck 48866c55602SAlexander Lobakin /* If requested length is either too small or too big, 48966c55602SAlexander Lobakin * we use kmalloc() for skb->head allocation. 49066c55602SAlexander Lobakin */ 49166c55602SAlexander Lobakin if (len <= SKB_WITH_OVERHEAD(1024) || 49266c55602SAlexander Lobakin len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 493d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 494a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 495a080e7bdSAlexander Duyck if (!skb) 496a080e7bdSAlexander Duyck goto skb_fail; 497a080e7bdSAlexander Duyck goto skb_success; 498a080e7bdSAlexander Duyck } 4999451980aSAlexander Duyck 5009451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5019451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5029451980aSAlexander Duyck 5039451980aSAlexander Duyck if (sk_memalloc_socks()) 5049451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5059451980aSAlexander Duyck 506afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 5079451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 5088c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 5099451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 51092dcabd7SSebastian Andrzej Siewior } else { 51192dcabd7SSebastian Andrzej Siewior local_bh_disable(); 51292dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 51392dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 51492dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 51592dcabd7SSebastian Andrzej Siewior local_bh_enable(); 51692dcabd7SSebastian Andrzej Siewior } 5179451980aSAlexander Duyck 5189451980aSAlexander Duyck if (unlikely(!data)) 5199451980aSAlexander Duyck return NULL; 5209451980aSAlexander Duyck 5219451980aSAlexander Duyck skb = __build_skb(data, len); 5229451980aSAlexander Duyck if (unlikely(!skb)) { 523181edb2bSAlexander Duyck skb_free_frag(data); 5249451980aSAlexander Duyck return NULL; 5259451980aSAlexander Duyck } 5269451980aSAlexander Duyck 5279451980aSAlexander Duyck if (pfmemalloc) 5289451980aSAlexander Duyck skb->pfmemalloc = 1; 5299451980aSAlexander Duyck skb->head_frag = 1; 5309451980aSAlexander Duyck 531a080e7bdSAlexander Duyck skb_success: 5328af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 5337b2e497aSChristoph Hellwig skb->dev = dev; 534fd11a83dSAlexander Duyck 535a080e7bdSAlexander Duyck skb_fail: 5368af27456SChristoph Hellwig return skb; 5378af27456SChristoph Hellwig } 538b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 5391da177e4SLinus Torvalds 540fd11a83dSAlexander Duyck /** 541fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 542fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 543d7499160SMasanari Iida * @len: length to allocate 544fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 545fd11a83dSAlexander Duyck * 546fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 547fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 548fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 549fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 550fd11a83dSAlexander Duyck * 551fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 552fd11a83dSAlexander Duyck */ 5539451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 5549451980aSAlexander Duyck gfp_t gfp_mask) 555fd11a83dSAlexander Duyck { 5563226b158SEric Dumazet struct napi_alloc_cache *nc; 557fd11a83dSAlexander Duyck struct sk_buff *skb; 5589451980aSAlexander Duyck void *data; 559fd11a83dSAlexander Duyck 5609451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 561fd11a83dSAlexander Duyck 5623226b158SEric Dumazet /* If requested length is either too small or too big, 5633226b158SEric Dumazet * we use kmalloc() for skb->head allocation. 5643226b158SEric Dumazet */ 5653226b158SEric Dumazet if (len <= SKB_WITH_OVERHEAD(1024) || 5663226b158SEric Dumazet len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 567d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 568cfb8ec65SAlexander Lobakin skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, 569cfb8ec65SAlexander Lobakin NUMA_NO_NODE); 570a080e7bdSAlexander Duyck if (!skb) 571a080e7bdSAlexander Duyck goto skb_fail; 572a080e7bdSAlexander Duyck goto skb_success; 573a080e7bdSAlexander Duyck } 5749451980aSAlexander Duyck 5753226b158SEric Dumazet nc = this_cpu_ptr(&napi_alloc_cache); 5769451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5779451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5789451980aSAlexander Duyck 5799451980aSAlexander Duyck if (sk_memalloc_socks()) 5809451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5819451980aSAlexander Duyck 5828c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5839451980aSAlexander Duyck if (unlikely(!data)) 5849451980aSAlexander Duyck return NULL; 5859451980aSAlexander Duyck 586cfb8ec65SAlexander Lobakin skb = __napi_build_skb(data, len); 5879451980aSAlexander Duyck if (unlikely(!skb)) { 588181edb2bSAlexander Duyck skb_free_frag(data); 5899451980aSAlexander Duyck return NULL; 5909451980aSAlexander Duyck } 5919451980aSAlexander Duyck 592795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 5939451980aSAlexander Duyck skb->pfmemalloc = 1; 5949451980aSAlexander Duyck skb->head_frag = 1; 5959451980aSAlexander Duyck 596a080e7bdSAlexander Duyck skb_success: 597fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 598fd11a83dSAlexander Duyck skb->dev = napi->dev; 599fd11a83dSAlexander Duyck 600a080e7bdSAlexander Duyck skb_fail: 601fd11a83dSAlexander Duyck return skb; 602fd11a83dSAlexander Duyck } 603fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 604fd11a83dSAlexander Duyck 605654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 60650269e19SEric Dumazet int size, unsigned int truesize) 607654bed16SPeter Zijlstra { 608654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 609654bed16SPeter Zijlstra skb->len += size; 610654bed16SPeter Zijlstra skb->data_len += size; 61150269e19SEric Dumazet skb->truesize += truesize; 612654bed16SPeter Zijlstra } 613654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 614654bed16SPeter Zijlstra 615f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 616f8e617e1SJason Wang unsigned int truesize) 617f8e617e1SJason Wang { 618f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 619f8e617e1SJason Wang 620f8e617e1SJason Wang skb_frag_size_add(frag, size); 621f8e617e1SJason Wang skb->len += size; 622f8e617e1SJason Wang skb->data_len += size; 623f8e617e1SJason Wang skb->truesize += truesize; 624f8e617e1SJason Wang } 625f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 626f8e617e1SJason Wang 62727b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 6281da177e4SLinus Torvalds { 629bd8a7036SEric Dumazet kfree_skb_list(*listp); 63027b437c8SHerbert Xu *listp = NULL; 6311da177e4SLinus Torvalds } 6321da177e4SLinus Torvalds 63327b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 63427b437c8SHerbert Xu { 63527b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 63627b437c8SHerbert Xu } 63727b437c8SHerbert Xu 6381da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 6391da177e4SLinus Torvalds { 6401da177e4SLinus Torvalds struct sk_buff *list; 6411da177e4SLinus Torvalds 642fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 6431da177e4SLinus Torvalds skb_get(list); 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds 646d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 647d3836f21SEric Dumazet { 648181edb2bSAlexander Duyck unsigned char *head = skb->head; 649181edb2bSAlexander Duyck 6506a5bcd84SIlias Apalodimas if (skb->head_frag) { 6516a5bcd84SIlias Apalodimas if (skb_pp_recycle(skb, head)) 6526a5bcd84SIlias Apalodimas return; 653181edb2bSAlexander Duyck skb_free_frag(head); 6546a5bcd84SIlias Apalodimas } else { 655181edb2bSAlexander Duyck kfree(head); 656d3836f21SEric Dumazet } 6576a5bcd84SIlias Apalodimas } 658d3836f21SEric Dumazet 6595bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 6601da177e4SLinus Torvalds { 661ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6621da177e4SLinus Torvalds int i; 663ff04a771SEric Dumazet 664ff04a771SEric Dumazet if (skb->cloned && 665ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 666ff04a771SEric Dumazet &shinfo->dataref)) 6672cc3aeb5SIlias Apalodimas goto exit; 668ff04a771SEric Dumazet 66970c43167SJonathan Lemon skb_zcopy_clear(skb, true); 67070c43167SJonathan Lemon 671ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 6726a5bcd84SIlias Apalodimas __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); 6731da177e4SLinus Torvalds 674ff04a771SEric Dumazet if (shinfo->frag_list) 675ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6761da177e4SLinus Torvalds 677d3836f21SEric Dumazet skb_free_head(skb); 6782cc3aeb5SIlias Apalodimas exit: 6792cc3aeb5SIlias Apalodimas /* When we clone an SKB we copy the reycling bit. The pp_recycle 6802cc3aeb5SIlias Apalodimas * bit is only set on the head though, so in order to avoid races 6812cc3aeb5SIlias Apalodimas * while trying to recycle fragments on __skb_frag_unref() we need 6822cc3aeb5SIlias Apalodimas * to make one SKB responsible for triggering the recycle path. 6832cc3aeb5SIlias Apalodimas * So disable the recycling bit if an SKB is cloned and we have 68458e61e41STom Rix * additional references to the fragmented part of the SKB. 6852cc3aeb5SIlias Apalodimas * Eventually the last SKB will have the recycling bit set and it's 6862cc3aeb5SIlias Apalodimas * dataref set to 0, which will trigger the recycling 6872cc3aeb5SIlias Apalodimas */ 6882cc3aeb5SIlias Apalodimas skb->pp_recycle = 0; 6891da177e4SLinus Torvalds } 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds /* 6921da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 6931da177e4SLinus Torvalds */ 6942d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 6951da177e4SLinus Torvalds { 696d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 697d179cd12SDavid S. Miller 698d179cd12SDavid S. Miller switch (skb->fclone) { 699d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 7001da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 7016ffe75ebSEric Dumazet return; 702d179cd12SDavid S. Miller 703d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 704d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 7056ffe75ebSEric Dumazet 7066ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 7076ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 7086ffe75ebSEric Dumazet * while here, branch prediction will be good. 7096ffe75ebSEric Dumazet */ 7102638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 7116ffe75ebSEric Dumazet goto fastpath; 712d179cd12SDavid S. Miller break; 713d179cd12SDavid S. Miller 7146ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 715d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 716d179cd12SDavid S. Miller break; 7173ff50b79SStephen Hemminger } 7182638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 7196ffe75ebSEric Dumazet return; 7206ffe75ebSEric Dumazet fastpath: 7216ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds 7240a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 7251da177e4SLinus Torvalds { 726adf30907SEric Dumazet skb_dst_drop(skb); 7271da177e4SLinus Torvalds if (skb->destructor) { 728afa79d08SChangbin Du WARN_ON(in_hardirq()); 7291da177e4SLinus Torvalds skb->destructor(skb); 7301da177e4SLinus Torvalds } 731a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 732cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 7332fc72c7bSKOVACS Krisztian #endif 734df5042f4SFlorian Westphal skb_ext_put(skb); 73504a4bb55SLennert Buytenhek } 73604a4bb55SLennert Buytenhek 73704a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 73804a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 73904a4bb55SLennert Buytenhek { 74004a4bb55SLennert Buytenhek skb_release_head_state(skb); 741a28b1b90SFlorian Westphal if (likely(skb->head)) 7422d4baff8SHerbert Xu skb_release_data(skb); 7432d4baff8SHerbert Xu } 7441da177e4SLinus Torvalds 7452d4baff8SHerbert Xu /** 7462d4baff8SHerbert Xu * __kfree_skb - private function 7472d4baff8SHerbert Xu * @skb: buffer 7482d4baff8SHerbert Xu * 7492d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 7502d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 7512d4baff8SHerbert Xu * always call kfree_skb 7522d4baff8SHerbert Xu */ 7532d4baff8SHerbert Xu 7542d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 7552d4baff8SHerbert Xu { 7562d4baff8SHerbert Xu skb_release_all(skb); 7571da177e4SLinus Torvalds kfree_skbmem(skb); 7581da177e4SLinus Torvalds } 759b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 7601da177e4SLinus Torvalds 7611da177e4SLinus Torvalds /** 762c504e5c2SMenglong Dong * kfree_skb_reason - free an sk_buff with special reason 763231d06aeSJörn Engel * @skb: buffer to free 764c504e5c2SMenglong Dong * @reason: reason why this skb is dropped 765231d06aeSJörn Engel * 766231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 767c504e5c2SMenglong Dong * hit zero. Meanwhile, pass the drop reason to 'kfree_skb' 768c504e5c2SMenglong Dong * tracepoint. 769231d06aeSJörn Engel */ 770c504e5c2SMenglong Dong void kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 771231d06aeSJörn Engel { 7723889a803SPaolo Abeni if (!skb_unref(skb)) 773231d06aeSJörn Engel return; 7743889a803SPaolo Abeni 775c504e5c2SMenglong Dong trace_kfree_skb(skb, __builtin_return_address(0), reason); 776231d06aeSJörn Engel __kfree_skb(skb); 777231d06aeSJörn Engel } 778c504e5c2SMenglong Dong EXPORT_SYMBOL(kfree_skb_reason); 779231d06aeSJörn Engel 780bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 781bd8a7036SEric Dumazet { 782bd8a7036SEric Dumazet while (segs) { 783bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 784bd8a7036SEric Dumazet 785bd8a7036SEric Dumazet kfree_skb(segs); 786bd8a7036SEric Dumazet segs = next; 787bd8a7036SEric Dumazet } 788bd8a7036SEric Dumazet } 789bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 790bd8a7036SEric Dumazet 7916413139dSWillem de Bruijn /* Dump skb information and contents. 7926413139dSWillem de Bruijn * 7936413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 7946413139dSWillem de Bruijn * 795302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 7966413139dSWillem de Bruijn */ 7976413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 7986413139dSWillem de Bruijn { 7996413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 8006413139dSWillem de Bruijn struct net_device *dev = skb->dev; 8016413139dSWillem de Bruijn struct sock *sk = skb->sk; 8026413139dSWillem de Bruijn struct sk_buff *list_skb; 8036413139dSWillem de Bruijn bool has_mac, has_trans; 8046413139dSWillem de Bruijn int headroom, tailroom; 8056413139dSWillem de Bruijn int i, len, seg_len; 8066413139dSWillem de Bruijn 8076413139dSWillem de Bruijn if (full_pkt) 8086413139dSWillem de Bruijn len = skb->len; 8096413139dSWillem de Bruijn else 8106413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 8116413139dSWillem de Bruijn 8126413139dSWillem de Bruijn headroom = skb_headroom(skb); 8136413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 8146413139dSWillem de Bruijn 8156413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 8166413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 8176413139dSWillem de Bruijn 8186413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 8196413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 8206413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 8216413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 8226413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 8236413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 8246413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 8256413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 8266413139dSWillem de Bruijn skb->network_header, 8276413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 8286413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 8296413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 8306413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 8316413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 8326413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 8336413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 8346413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 8356413139dSWillem de Bruijn 8366413139dSWillem de Bruijn if (dev) 8378a03ef67SGal Pressman printk("%sdev name=%s feat=%pNF\n", 8386413139dSWillem de Bruijn level, dev->name, &dev->features); 8396413139dSWillem de Bruijn if (sk) 840db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 8416413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 8426413139dSWillem de Bruijn 8436413139dSWillem de Bruijn if (full_pkt && headroom) 8446413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 8456413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 8466413139dSWillem de Bruijn 8476413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 8486413139dSWillem de Bruijn if (seg_len) 8496413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 8506413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 8516413139dSWillem de Bruijn len -= seg_len; 8526413139dSWillem de Bruijn 8536413139dSWillem de Bruijn if (full_pkt && tailroom) 8546413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 8556413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 8566413139dSWillem de Bruijn 8576413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 8586413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 8596413139dSWillem de Bruijn u32 p_off, p_len, copied; 8606413139dSWillem de Bruijn struct page *p; 8616413139dSWillem de Bruijn u8 *vaddr; 8626413139dSWillem de Bruijn 863b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 8646413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 8656413139dSWillem de Bruijn copied) { 8666413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 8676413139dSWillem de Bruijn vaddr = kmap_atomic(p); 8686413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 8696413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 8706413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 8716413139dSWillem de Bruijn kunmap_atomic(vaddr); 8726413139dSWillem de Bruijn len -= seg_len; 8736413139dSWillem de Bruijn if (!len) 8746413139dSWillem de Bruijn break; 8756413139dSWillem de Bruijn } 8766413139dSWillem de Bruijn } 8776413139dSWillem de Bruijn 8786413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8796413139dSWillem de Bruijn printk("skb fraglist:\n"); 8806413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8816413139dSWillem de Bruijn skb_dump(level, list_skb, true); 8826413139dSWillem de Bruijn } 8836413139dSWillem de Bruijn } 8846413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 8856413139dSWillem de Bruijn 886d1a203eaSStephen Hemminger /** 88725121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 88825121173SMichael S. Tsirkin * @skb: buffer that triggered an error 88925121173SMichael S. Tsirkin * 89025121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 89125121173SMichael S. Tsirkin * skb must be freed afterwards. 89225121173SMichael S. Tsirkin */ 89325121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 89425121173SMichael S. Tsirkin { 8951f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 89625121173SMichael S. Tsirkin } 89725121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 89825121173SMichael S. Tsirkin 899be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 90025121173SMichael S. Tsirkin /** 901ead2ceb0SNeil Horman * consume_skb - free an skbuff 902ead2ceb0SNeil Horman * @skb: buffer to free 903ead2ceb0SNeil Horman * 904ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 905ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 906ead2ceb0SNeil Horman * is being dropped after a failure and notes that 907ead2ceb0SNeil Horman */ 908ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 909ead2ceb0SNeil Horman { 9103889a803SPaolo Abeni if (!skb_unref(skb)) 911ead2ceb0SNeil Horman return; 9123889a803SPaolo Abeni 91307dc22e7SKoki Sanagi trace_consume_skb(skb); 914ead2ceb0SNeil Horman __kfree_skb(skb); 915ead2ceb0SNeil Horman } 916ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 917be769db2SHerbert Xu #endif 918ead2ceb0SNeil Horman 9190a463c78SPaolo Abeni /** 920c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 9210a463c78SPaolo Abeni * @skb: buffer to free 9220a463c78SPaolo Abeni * 923ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 924ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 9250a463c78SPaolo Abeni */ 926ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 9270a463c78SPaolo Abeni { 9280a463c78SPaolo Abeni trace_consume_skb(skb); 9290a463c78SPaolo Abeni skb_release_data(skb); 9300a463c78SPaolo Abeni kfree_skbmem(skb); 9310a463c78SPaolo Abeni } 9320a463c78SPaolo Abeni 933f450d539SAlexander Lobakin static void napi_skb_cache_put(struct sk_buff *skb) 934795bb1c0SJesper Dangaard Brouer { 935795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 936f450d539SAlexander Lobakin u32 i; 937795bb1c0SJesper Dangaard Brouer 938f450d539SAlexander Lobakin kasan_poison_object_data(skbuff_head_cache, skb); 939795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 940795bb1c0SJesper Dangaard Brouer 941795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 942f450d539SAlexander Lobakin for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++) 943f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, 944f450d539SAlexander Lobakin nc->skb_cache[i]); 945f450d539SAlexander Lobakin 946f450d539SAlexander Lobakin kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF, 947f450d539SAlexander Lobakin nc->skb_cache + NAPI_SKB_CACHE_HALF); 948f450d539SAlexander Lobakin nc->skb_count = NAPI_SKB_CACHE_HALF; 949795bb1c0SJesper Dangaard Brouer } 950795bb1c0SJesper Dangaard Brouer } 951f450d539SAlexander Lobakin 95215fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 95315fad714SJesper Dangaard Brouer { 9549243adfcSAlexander Lobakin skb_release_all(skb); 9559243adfcSAlexander Lobakin napi_skb_cache_put(skb); 9569243adfcSAlexander Lobakin } 9579243adfcSAlexander Lobakin 9589243adfcSAlexander Lobakin void napi_skb_free_stolen_head(struct sk_buff *skb) 9599243adfcSAlexander Lobakin { 9609efb4b5bSPaolo Abeni if (unlikely(skb->slow_gro)) { 9618550ff8dSPaul Blakey nf_reset_ct(skb); 9629243adfcSAlexander Lobakin skb_dst_drop(skb); 9639243adfcSAlexander Lobakin skb_ext_put(skb); 9645e10da53SPaolo Abeni skb_orphan(skb); 9659efb4b5bSPaolo Abeni skb->slow_gro = 0; 9669efb4b5bSPaolo Abeni } 967f450d539SAlexander Lobakin napi_skb_cache_put(skb); 96815fad714SJesper Dangaard Brouer } 969795bb1c0SJesper Dangaard Brouer 970795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 971795bb1c0SJesper Dangaard Brouer { 972885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 973795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 974885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 975795bb1c0SJesper Dangaard Brouer return; 976795bb1c0SJesper Dangaard Brouer } 977795bb1c0SJesper Dangaard Brouer 9786454eca8SYunsheng Lin lockdep_assert_in_softirq(); 9796454eca8SYunsheng Lin 9807608894eSPaolo Abeni if (!skb_unref(skb)) 981795bb1c0SJesper Dangaard Brouer return; 9827608894eSPaolo Abeni 983795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 984795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 985795bb1c0SJesper Dangaard Brouer 986795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 987abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 988795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 989795bb1c0SJesper Dangaard Brouer return; 990795bb1c0SJesper Dangaard Brouer } 991795bb1c0SJesper Dangaard Brouer 9929243adfcSAlexander Lobakin skb_release_all(skb); 993f450d539SAlexander Lobakin napi_skb_cache_put(skb); 994795bb1c0SJesper Dangaard Brouer } 995795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 996795bb1c0SJesper Dangaard Brouer 99703f61041SKees Cook /* Make sure a field is contained by headers group */ 998b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 99903f61041SKees Cook BUILD_BUG_ON(offsetof(struct sk_buff, field) != \ 100003f61041SKees Cook offsetof(struct sk_buff, headers.field)); \ 1001b1937227SEric Dumazet 1002dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 1003dec18810SHerbert Xu { 1004dec18810SHerbert Xu new->tstamp = old->tstamp; 1005b1937227SEric Dumazet /* We do not copy old->sk */ 1006dec18810SHerbert Xu new->dev = old->dev; 1007b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 10087fee226aSEric Dumazet skb_dst_copy(new, old); 1009df5042f4SFlorian Westphal __skb_ext_copy(new, old); 1010b1937227SEric Dumazet __nf_copy(new, old, false); 10116aa895b0SPatrick McHardy 101203f61041SKees Cook /* Note : this field could be in the headers group. 1013b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 1014b1937227SEric Dumazet */ 1015b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 101606021292SEliezer Tamir 101703f61041SKees Cook memcpy(&new->headers, &old->headers, sizeof(new->headers)); 1018b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 1019b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 1020b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 1021b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 1022b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 1023b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 1024b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 1025b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 1026b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 1027b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 1028b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 1029b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 1030b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 1031b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 1032b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 1033b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 1034b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 1035b1937227SEric Dumazet #endif 1036e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 1037b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 103806021292SEliezer Tamir #endif 10392bd82484SEric Dumazet #ifdef CONFIG_XPS 10402bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 10412bd82484SEric Dumazet #endif 1042b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 1043b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 1044b1937227SEric Dumazet #endif 1045b1937227SEric Dumazet 1046dec18810SHerbert Xu } 1047dec18810SHerbert Xu 104882c49a35SHerbert Xu /* 104982c49a35SHerbert Xu * You should not add any new code to this function. Add it to 105082c49a35SHerbert Xu * __copy_skb_header above instead. 105182c49a35SHerbert Xu */ 1052e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 10531da177e4SLinus Torvalds { 10541da177e4SLinus Torvalds #define C(x) n->x = skb->x 10551da177e4SLinus Torvalds 10561da177e4SLinus Torvalds n->next = n->prev = NULL; 10571da177e4SLinus Torvalds n->sk = NULL; 1058dec18810SHerbert Xu __copy_skb_header(n, skb); 1059dec18810SHerbert Xu 10601da177e4SLinus Torvalds C(len); 10611da177e4SLinus Torvalds C(data_len); 10623e6b3b2eSAlexey Dobriyan C(mac_len); 1063334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 106402f1c89dSPaul Moore n->cloned = 1; 10651da177e4SLinus Torvalds n->nohdr = 0; 1066b13dda9fSEric Dumazet n->peeked = 0; 1067e78bfb07SStefano Brivio C(pfmemalloc); 10686a5bcd84SIlias Apalodimas C(pp_recycle); 10691da177e4SLinus Torvalds n->destructor = NULL; 10701da177e4SLinus Torvalds C(tail); 10711da177e4SLinus Torvalds C(end); 107202f1c89dSPaul Moore C(head); 1073d3836f21SEric Dumazet C(head_frag); 107402f1c89dSPaul Moore C(data); 107502f1c89dSPaul Moore C(truesize); 107663354797SReshetova, Elena refcount_set(&n->users, 1); 10771da177e4SLinus Torvalds 10781da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 10791da177e4SLinus Torvalds skb->cloned = 1; 10801da177e4SLinus Torvalds 10811da177e4SLinus Torvalds return n; 1082e0053ec0SHerbert Xu #undef C 1083e0053ec0SHerbert Xu } 1084e0053ec0SHerbert Xu 1085e0053ec0SHerbert Xu /** 1086da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1087da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1088da29e4b4SJakub Kicinski */ 1089da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1090da29e4b4SJakub Kicinski { 1091da29e4b4SJakub Kicinski struct sk_buff *n; 1092da29e4b4SJakub Kicinski 1093da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1094da29e4b4SJakub Kicinski if (!n) 1095da29e4b4SJakub Kicinski return NULL; 1096da29e4b4SJakub Kicinski 1097da29e4b4SJakub Kicinski n->len = first->len; 1098da29e4b4SJakub Kicinski n->data_len = first->len; 1099da29e4b4SJakub Kicinski n->truesize = first->truesize; 1100da29e4b4SJakub Kicinski 1101da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1102da29e4b4SJakub Kicinski 1103da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1104da29e4b4SJakub Kicinski n->destructor = NULL; 1105da29e4b4SJakub Kicinski 1106da29e4b4SJakub Kicinski return n; 1107da29e4b4SJakub Kicinski } 1108da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1109da29e4b4SJakub Kicinski 1110da29e4b4SJakub Kicinski /** 1111e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1112e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1113e0053ec0SHerbert Xu * @src: the skb to supply the contents 1114e0053ec0SHerbert Xu * 1115e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1116e0053ec0SHerbert Xu * supplied by the user. 1117e0053ec0SHerbert Xu * 1118e0053ec0SHerbert Xu * The target skb is returned upon exit. 1119e0053ec0SHerbert Xu */ 1120e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1121e0053ec0SHerbert Xu { 11222d4baff8SHerbert Xu skb_release_all(dst); 1123e0053ec0SHerbert Xu return __skb_clone(dst, src); 1124e0053ec0SHerbert Xu } 1125e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1126e0053ec0SHerbert Xu 11276f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1128a91dbff5SWillem de Bruijn { 1129a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1130a91dbff5SWillem de Bruijn struct user_struct *user; 1131a91dbff5SWillem de Bruijn 1132a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1133a91dbff5SWillem de Bruijn return 0; 1134a91dbff5SWillem de Bruijn 1135a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1136a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1137a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1138a91dbff5SWillem de Bruijn 1139a91dbff5SWillem de Bruijn do { 1140a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1141a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1142a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1143a91dbff5SWillem de Bruijn return -ENOBUFS; 1144a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1145a91dbff5SWillem de Bruijn old_pg); 1146a91dbff5SWillem de Bruijn 1147a91dbff5SWillem de Bruijn if (!mmp->user) { 1148a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1149a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1150a91dbff5SWillem de Bruijn } else { 1151a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1152a91dbff5SWillem de Bruijn } 1153a91dbff5SWillem de Bruijn 1154a91dbff5SWillem de Bruijn return 0; 1155a91dbff5SWillem de Bruijn } 11566f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1157a91dbff5SWillem de Bruijn 11586f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1159a91dbff5SWillem de Bruijn { 1160a91dbff5SWillem de Bruijn if (mmp->user) { 1161a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1162a91dbff5SWillem de Bruijn free_uid(mmp->user); 1163a91dbff5SWillem de Bruijn } 1164a91dbff5SWillem de Bruijn } 11656f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1166a91dbff5SWillem de Bruijn 11678c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size) 116852267790SWillem de Bruijn { 116952267790SWillem de Bruijn struct ubuf_info *uarg; 117052267790SWillem de Bruijn struct sk_buff *skb; 117152267790SWillem de Bruijn 117252267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 117352267790SWillem de Bruijn 117452267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 117552267790SWillem de Bruijn if (!skb) 117652267790SWillem de Bruijn return NULL; 117752267790SWillem de Bruijn 117852267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 117952267790SWillem de Bruijn uarg = (void *)skb->cb; 1180a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1181a91dbff5SWillem de Bruijn 1182a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1183a91dbff5SWillem de Bruijn kfree_skb(skb); 1184a91dbff5SWillem de Bruijn return NULL; 1185a91dbff5SWillem de Bruijn } 118652267790SWillem de Bruijn 11878c793822SJonathan Lemon uarg->callback = msg_zerocopy_callback; 11884ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 11894ab6c99dSWillem de Bruijn uarg->len = 1; 11904ab6c99dSWillem de Bruijn uarg->bytelen = size; 119152267790SWillem de Bruijn uarg->zerocopy = 1; 119204c2d33eSJonathan Lemon uarg->flags = SKBFL_ZEROCOPY_FRAG; 1193c1d1b437SEric Dumazet refcount_set(&uarg->refcnt, 1); 119452267790SWillem de Bruijn sock_hold(sk); 119552267790SWillem de Bruijn 119652267790SWillem de Bruijn return uarg; 119752267790SWillem de Bruijn } 11988c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_alloc); 119952267790SWillem de Bruijn 120052267790SWillem de Bruijn static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 120152267790SWillem de Bruijn { 120252267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 120352267790SWillem de Bruijn } 120452267790SWillem de Bruijn 12058c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 12064ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 12074ab6c99dSWillem de Bruijn { 12084ab6c99dSWillem de Bruijn if (uarg) { 12094ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 12104ab6c99dSWillem de Bruijn u32 bytelen, next; 12114ab6c99dSWillem de Bruijn 12124ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 12134ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 12144ab6c99dSWillem de Bruijn */ 12154ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 12164ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 12174ab6c99dSWillem de Bruijn return NULL; 12184ab6c99dSWillem de Bruijn } 12194ab6c99dSWillem de Bruijn 12204ab6c99dSWillem de Bruijn bytelen = uarg->bytelen + size; 12214ab6c99dSWillem de Bruijn if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 12224ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 12234ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12244ab6c99dSWillem de Bruijn goto new_alloc; 12254ab6c99dSWillem de Bruijn return NULL; 12264ab6c99dSWillem de Bruijn } 12274ab6c99dSWillem de Bruijn 12284ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 12294ab6c99dSWillem de Bruijn if ((u32)(uarg->id + uarg->len) == next) { 1230a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) 1231a91dbff5SWillem de Bruijn return NULL; 12324ab6c99dSWillem de Bruijn uarg->len++; 12334ab6c99dSWillem de Bruijn uarg->bytelen = bytelen; 12344ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1235100f6d8eSWillem de Bruijn 1236100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1237100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12388e044917SJonathan Lemon net_zcopy_get(uarg); 1239100f6d8eSWillem de Bruijn 12404ab6c99dSWillem de Bruijn return uarg; 12414ab6c99dSWillem de Bruijn } 12424ab6c99dSWillem de Bruijn } 12434ab6c99dSWillem de Bruijn 12444ab6c99dSWillem de Bruijn new_alloc: 12458c793822SJonathan Lemon return msg_zerocopy_alloc(sk, size); 12464ab6c99dSWillem de Bruijn } 12478c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_realloc); 12484ab6c99dSWillem de Bruijn 12494ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 12504ab6c99dSWillem de Bruijn { 12514ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 12524ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 12534ab6c99dSWillem de Bruijn u64 sum_len; 12544ab6c99dSWillem de Bruijn 12554ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 12564ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 12574ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 12584ab6c99dSWillem de Bruijn 12594ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 12604ab6c99dSWillem de Bruijn return false; 12614ab6c99dSWillem de Bruijn 12624ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 12634ab6c99dSWillem de Bruijn return false; 12644ab6c99dSWillem de Bruijn 12654ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 12664ab6c99dSWillem de Bruijn return true; 12674ab6c99dSWillem de Bruijn } 12684ab6c99dSWillem de Bruijn 12698c793822SJonathan Lemon static void __msg_zerocopy_callback(struct ubuf_info *uarg) 127052267790SWillem de Bruijn { 12714ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 127252267790SWillem de Bruijn struct sock_exterr_skb *serr; 127352267790SWillem de Bruijn struct sock *sk = skb->sk; 12744ab6c99dSWillem de Bruijn struct sk_buff_head *q; 12754ab6c99dSWillem de Bruijn unsigned long flags; 12763bdd5ee0SWillem de Bruijn bool is_zerocopy; 12774ab6c99dSWillem de Bruijn u32 lo, hi; 12784ab6c99dSWillem de Bruijn u16 len; 127952267790SWillem de Bruijn 1280ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1281ccaffff1SWillem de Bruijn 12824ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 12834ab6c99dSWillem de Bruijn * so do not queue a completion notification 12844ab6c99dSWillem de Bruijn */ 12854ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 128652267790SWillem de Bruijn goto release; 128752267790SWillem de Bruijn 12884ab6c99dSWillem de Bruijn len = uarg->len; 12894ab6c99dSWillem de Bruijn lo = uarg->id; 12904ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 12913bdd5ee0SWillem de Bruijn is_zerocopy = uarg->zerocopy; 12924ab6c99dSWillem de Bruijn 129352267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 129452267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 129552267790SWillem de Bruijn serr->ee.ee_errno = 0; 129652267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 12974ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 12984ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 12993bdd5ee0SWillem de Bruijn if (!is_zerocopy) 130052267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 130152267790SWillem de Bruijn 13024ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 13034ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 13044ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 13054ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 13064ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 13074ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 130852267790SWillem de Bruijn skb = NULL; 13094ab6c99dSWillem de Bruijn } 13104ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 131152267790SWillem de Bruijn 1312e3ae2365SAlexander Aring sk_error_report(sk); 131352267790SWillem de Bruijn 131452267790SWillem de Bruijn release: 131552267790SWillem de Bruijn consume_skb(skb); 131652267790SWillem de Bruijn sock_put(sk); 131752267790SWillem de Bruijn } 131875518851SJonathan Lemon 13198c793822SJonathan Lemon void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 132036177832SJonathan Lemon bool success) 132175518851SJonathan Lemon { 132275518851SJonathan Lemon uarg->zerocopy = uarg->zerocopy & success; 132375518851SJonathan Lemon 132475518851SJonathan Lemon if (refcount_dec_and_test(&uarg->refcnt)) 13258c793822SJonathan Lemon __msg_zerocopy_callback(uarg); 132675518851SJonathan Lemon } 13278c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_callback); 132852267790SWillem de Bruijn 13298c793822SJonathan Lemon void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 133052267790SWillem de Bruijn { 133152267790SWillem de Bruijn struct sock *sk = skb_from_uarg(uarg)->sk; 133252267790SWillem de Bruijn 133352267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 13344ab6c99dSWillem de Bruijn uarg->len--; 133552267790SWillem de Bruijn 133652900d22SWillem de Bruijn if (have_uref) 13378c793822SJonathan Lemon msg_zerocopy_callback(NULL, uarg, true); 133852267790SWillem de Bruijn } 13398c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort); 134052267790SWillem de Bruijn 1341b5947e5dSWillem de Bruijn int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) 1342b5947e5dSWillem de Bruijn { 1343b5947e5dSWillem de Bruijn return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1344b5947e5dSWillem de Bruijn } 1345b5947e5dSWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); 1346b5947e5dSWillem de Bruijn 134752267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 134852267790SWillem de Bruijn struct msghdr *msg, int len, 134952267790SWillem de Bruijn struct ubuf_info *uarg) 135052267790SWillem de Bruijn { 13514ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 135252267790SWillem de Bruijn struct iov_iter orig_iter = msg->msg_iter; 135352267790SWillem de Bruijn int err, orig_len = skb->len; 135452267790SWillem de Bruijn 13554ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 13564ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 13574ab6c99dSWillem de Bruijn */ 13584ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 13594ab6c99dSWillem de Bruijn return -EEXIST; 13604ab6c99dSWillem de Bruijn 136152267790SWillem de Bruijn err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 136252267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 136354d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 136454d43117SWillem de Bruijn 136552267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 136652267790SWillem de Bruijn msg->msg_iter = orig_iter; 136754d43117SWillem de Bruijn skb->sk = sk; 136852267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 136954d43117SWillem de Bruijn skb->sk = save_sk; 137052267790SWillem de Bruijn return err; 137152267790SWillem de Bruijn } 137252267790SWillem de Bruijn 137352900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 137452267790SWillem de Bruijn return skb->len - orig_len; 137552267790SWillem de Bruijn } 137652267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 137752267790SWillem de Bruijn 13781f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 137952267790SWillem de Bruijn gfp_t gfp_mask) 138052267790SWillem de Bruijn { 138152267790SWillem de Bruijn if (skb_zcopy(orig)) { 138252267790SWillem de Bruijn if (skb_zcopy(nskb)) { 138352267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 138452267790SWillem de Bruijn if (!gfp_mask) { 138552267790SWillem de Bruijn WARN_ON_ONCE(1); 138652267790SWillem de Bruijn return -ENOMEM; 138752267790SWillem de Bruijn } 138852267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 138952267790SWillem de Bruijn return 0; 139052267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 139152267790SWillem de Bruijn return -EIO; 139252267790SWillem de Bruijn } 139352900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 139452267790SWillem de Bruijn } 139552267790SWillem de Bruijn return 0; 139652267790SWillem de Bruijn } 139752267790SWillem de Bruijn 13982c53040fSBen Hutchings /** 13992c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 140048c83012SMichael S. Tsirkin * @skb: the skb to modify 140148c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 140248c83012SMichael S. Tsirkin * 140306b4feb3SJonathan Lemon * This must be called on skb with SKBFL_ZEROCOPY_ENABLE. 140448c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 140548c83012SMichael S. Tsirkin * to userspace pages. 140648c83012SMichael S. Tsirkin * 140748c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 140848c83012SMichael S. Tsirkin * %GFP_ATOMIC. 140948c83012SMichael S. Tsirkin * 141048c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 141148c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 141248c83012SMichael S. Tsirkin */ 141348c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1414a6686f2fSShirley Ma { 1415a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1416a6686f2fSShirley Ma struct page *page, *head = NULL; 14173ece7826SWillem de Bruijn int i, new_frags; 14183ece7826SWillem de Bruijn u32 d_off; 1419a6686f2fSShirley Ma 14203ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 14213ece7826SWillem de Bruijn return -EINVAL; 14223ece7826SWillem de Bruijn 1423f72c4ac6SWillem de Bruijn if (!num_frags) 1424f72c4ac6SWillem de Bruijn goto release; 1425f72c4ac6SWillem de Bruijn 14263ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 14273ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 142802756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1429a6686f2fSShirley Ma if (!page) { 1430a6686f2fSShirley Ma while (head) { 143140dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1432a6686f2fSShirley Ma put_page(head); 1433a6686f2fSShirley Ma head = next; 1434a6686f2fSShirley Ma } 1435a6686f2fSShirley Ma return -ENOMEM; 1436a6686f2fSShirley Ma } 14373ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 14383ece7826SWillem de Bruijn head = page; 14393ece7826SWillem de Bruijn } 14403ece7826SWillem de Bruijn 14413ece7826SWillem de Bruijn page = head; 14423ece7826SWillem de Bruijn d_off = 0; 14433ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 14443ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 14453ece7826SWillem de Bruijn u32 p_off, p_len, copied; 14463ece7826SWillem de Bruijn struct page *p; 14473ece7826SWillem de Bruijn u8 *vaddr; 1448c613c209SWillem de Bruijn 1449b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1450c613c209SWillem de Bruijn p, p_off, p_len, copied) { 14513ece7826SWillem de Bruijn u32 copy, done = 0; 1452c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 14533ece7826SWillem de Bruijn 14543ece7826SWillem de Bruijn while (done < p_len) { 14553ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 14563ece7826SWillem de Bruijn d_off = 0; 14573ece7826SWillem de Bruijn page = (struct page *)page_private(page); 14583ece7826SWillem de Bruijn } 14593ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 14603ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 14613ece7826SWillem de Bruijn vaddr + p_off + done, copy); 14623ece7826SWillem de Bruijn done += copy; 14633ece7826SWillem de Bruijn d_off += copy; 14643ece7826SWillem de Bruijn } 146551c56b00SEric Dumazet kunmap_atomic(vaddr); 1466c613c209SWillem de Bruijn } 1467a6686f2fSShirley Ma } 1468a6686f2fSShirley Ma 1469a6686f2fSShirley Ma /* skb frags release userspace buffers */ 147002756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1471a8605c60SIan Campbell skb_frag_unref(skb, i); 1472a6686f2fSShirley Ma 1473a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 14743ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 14753ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 147640dadff2SSunghan Suh head = (struct page *)page_private(head); 1477a6686f2fSShirley Ma } 14783ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 14793ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 148048c83012SMichael S. Tsirkin 1481b90ddd56SWillem de Bruijn release: 14821f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1483a6686f2fSShirley Ma return 0; 1484a6686f2fSShirley Ma } 1485dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1486a6686f2fSShirley Ma 1487e0053ec0SHerbert Xu /** 1488e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1489e0053ec0SHerbert Xu * @skb: buffer to clone 1490e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1491e0053ec0SHerbert Xu * 1492e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1493e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1494e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1495e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1496e0053ec0SHerbert Xu * 1497e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1498e0053ec0SHerbert Xu * %GFP_ATOMIC. 1499e0053ec0SHerbert Xu */ 1500e0053ec0SHerbert Xu 1501e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1502e0053ec0SHerbert Xu { 1503d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1504d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1505d0bf4a9eSEric Dumazet skb1); 15066ffe75ebSEric Dumazet struct sk_buff *n; 1507e0053ec0SHerbert Xu 150870008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1509a6686f2fSShirley Ma return NULL; 1510a6686f2fSShirley Ma 1511e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 15122638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 15136ffe75ebSEric Dumazet n = &fclones->skb2; 15142638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1515e0053ec0SHerbert Xu } else { 1516c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1517c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1518c93bdd0eSMel Gorman 1519e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1520e0053ec0SHerbert Xu if (!n) 1521e0053ec0SHerbert Xu return NULL; 1522fe55f6d5SVegard Nossum 1523e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1524e0053ec0SHerbert Xu } 1525e0053ec0SHerbert Xu 1526e0053ec0SHerbert Xu return __skb_clone(n, skb); 15271da177e4SLinus Torvalds } 1528b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 15291da177e4SLinus Torvalds 1530b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1531f5b17294SPravin B Shelar { 1532030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1533030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1534030737bcSEric Dumazet skb->csum_start += off; 1535f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1536f5b17294SPravin B Shelar skb->transport_header += off; 1537f5b17294SPravin B Shelar skb->network_header += off; 1538f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1539f5b17294SPravin B Shelar skb->mac_header += off; 1540f5b17294SPravin B Shelar skb->inner_transport_header += off; 1541f5b17294SPravin B Shelar skb->inner_network_header += off; 1542aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1543f5b17294SPravin B Shelar } 1544b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1545f5b17294SPravin B Shelar 154608303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 15471da177e4SLinus Torvalds { 1548dec18810SHerbert Xu __copy_skb_header(new, old); 1549dec18810SHerbert Xu 15507967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 15517967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 15527967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 15531da177e4SLinus Torvalds } 155408303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 15551da177e4SLinus Torvalds 1556c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1557c93bdd0eSMel Gorman { 1558c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1559c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1560c93bdd0eSMel Gorman return 0; 1561c93bdd0eSMel Gorman } 1562c93bdd0eSMel Gorman 15631da177e4SLinus Torvalds /** 15641da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 15651da177e4SLinus Torvalds * @skb: buffer to copy 15661da177e4SLinus Torvalds * @gfp_mask: allocation priority 15671da177e4SLinus Torvalds * 15681da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 15691da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 15701da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 15711da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 15721da177e4SLinus Torvalds * 15731da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 15741da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 15751da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 15761da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 15771da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 15781da177e4SLinus Torvalds */ 15791da177e4SLinus Torvalds 1580dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 15811da177e4SLinus Torvalds { 15826602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1583ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1584c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1585c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 15866602cebbSEric Dumazet 15871da177e4SLinus Torvalds if (!n) 15881da177e4SLinus Torvalds return NULL; 15891da177e4SLinus Torvalds 15901da177e4SLinus Torvalds /* Set the data pointer */ 15911da177e4SLinus Torvalds skb_reserve(n, headerlen); 15921da177e4SLinus Torvalds /* Set the tail pointer and length */ 15931da177e4SLinus Torvalds skb_put(n, skb->len); 15941da177e4SLinus Torvalds 15959f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 15961da177e4SLinus Torvalds 159708303c18SIlya Lesokhin skb_copy_header(n, skb); 15981da177e4SLinus Torvalds return n; 15991da177e4SLinus Torvalds } 1600b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds /** 1603bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 16041da177e4SLinus Torvalds * @skb: buffer to copy 1605117632e6SEric Dumazet * @headroom: headroom of new skb 16061da177e4SLinus Torvalds * @gfp_mask: allocation priority 1607bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1608bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1609bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 16101da177e4SLinus Torvalds * 16111da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 16121da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 16131da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 16141da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 16151da177e4SLinus Torvalds * or the pointer to the buffer on success. 16161da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 16171da177e4SLinus Torvalds */ 16181da177e4SLinus Torvalds 1619bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1620bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 16211da177e4SLinus Torvalds { 1622117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1623bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1624bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 16256602cebbSEric Dumazet 16261da177e4SLinus Torvalds if (!n) 16271da177e4SLinus Torvalds goto out; 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds /* Set the data pointer */ 1630117632e6SEric Dumazet skb_reserve(n, headroom); 16311da177e4SLinus Torvalds /* Set the tail pointer and length */ 16321da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 16331da177e4SLinus Torvalds /* Copy the bytes */ 1634d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 16351da177e4SLinus Torvalds 163625f484a6SHerbert Xu n->truesize += skb->data_len; 16371da177e4SLinus Torvalds n->data_len = skb->data_len; 16381da177e4SLinus Torvalds n->len = skb->len; 16391da177e4SLinus Torvalds 16401da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 16411da177e4SLinus Torvalds int i; 16421da177e4SLinus Torvalds 16431f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 16441f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 16451511022cSDan Carpenter kfree_skb(n); 16461511022cSDan Carpenter n = NULL; 1647a6686f2fSShirley Ma goto out; 1648a6686f2fSShirley Ma } 16491da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16501da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1651ea2ab693SIan Campbell skb_frag_ref(skb, i); 16521da177e4SLinus Torvalds } 16531da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 16541da177e4SLinus Torvalds } 16551da177e4SLinus Torvalds 165621dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 16571da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 16581da177e4SLinus Torvalds skb_clone_fraglist(n); 16591da177e4SLinus Torvalds } 16601da177e4SLinus Torvalds 166108303c18SIlya Lesokhin skb_copy_header(n, skb); 16621da177e4SLinus Torvalds out: 16631da177e4SLinus Torvalds return n; 16641da177e4SLinus Torvalds } 1665bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds /** 16681da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 16691da177e4SLinus Torvalds * @skb: buffer to reallocate 16701da177e4SLinus Torvalds * @nhead: room to add at head 16711da177e4SLinus Torvalds * @ntail: room to add at tail 16721da177e4SLinus Torvalds * @gfp_mask: allocation priority 16731da177e4SLinus Torvalds * 1674bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1675bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 16761da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 16771da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 16781da177e4SLinus Torvalds * 16791da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 16801da177e4SLinus Torvalds * reloaded after call to this function. 16811da177e4SLinus Torvalds */ 16821da177e4SLinus Torvalds 168386a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1684dd0fc66fSAl Viro gfp_t gfp_mask) 16851da177e4SLinus Torvalds { 1686158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1687158f323bSEric Dumazet int size = osize + nhead + ntail; 16881da177e4SLinus Torvalds long off; 1689158f323bSEric Dumazet u8 *data; 16901da177e4SLinus Torvalds 16914edd87adSHerbert Xu BUG_ON(nhead < 0); 16924edd87adSHerbert Xu 16939f77fad3STim Hansen BUG_ON(skb_shared(skb)); 16941da177e4SLinus Torvalds 16951da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 16961da177e4SLinus Torvalds 1697c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1698c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1699c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1700c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 17011da177e4SLinus Torvalds if (!data) 17021da177e4SLinus Torvalds goto nodata; 170387151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 17041da177e4SLinus Torvalds 17051da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 17066602cebbSEric Dumazet * optimized for the cases when header is void. 17076602cebbSEric Dumazet */ 17086602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 17096602cebbSEric Dumazet 17106602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 17116602cebbSEric Dumazet skb_shinfo(skb), 1712fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 17131da177e4SLinus Torvalds 17143e24591aSAlexander Duyck /* 17153e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 17163e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 17173e24591aSAlexander Duyck * be since all we did is relocate the values 17183e24591aSAlexander Duyck */ 17193e24591aSAlexander Duyck if (skb_cloned(skb)) { 172070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1721a6686f2fSShirley Ma goto nofrags; 17221f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1723c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 17241da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1725ea2ab693SIan Campbell skb_frag_ref(skb, i); 17261da177e4SLinus Torvalds 172721dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 17281da177e4SLinus Torvalds skb_clone_fraglist(skb); 17291da177e4SLinus Torvalds 17301da177e4SLinus Torvalds skb_release_data(skb); 17313e24591aSAlexander Duyck } else { 17323e24591aSAlexander Duyck skb_free_head(skb); 17331fd63041SEric Dumazet } 17341da177e4SLinus Torvalds off = (data + nhead) - skb->head; 17351da177e4SLinus Torvalds 17361da177e4SLinus Torvalds skb->head = data; 1737d3836f21SEric Dumazet skb->head_frag = 0; 17381da177e4SLinus Torvalds skb->data += off; 1739763087daSEric Dumazet 1740763087daSEric Dumazet skb_set_end_offset(skb, size); 17414305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 174256eb8882SPatrick McHardy off = nhead; 174356eb8882SPatrick McHardy #endif 174427a884dcSArnaldo Carvalho de Melo skb->tail += off; 1745b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 17461da177e4SLinus Torvalds skb->cloned = 0; 1747334a8132SPatrick McHardy skb->hdr_len = 0; 17481da177e4SLinus Torvalds skb->nohdr = 0; 17491da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1750158f323bSEric Dumazet 1751de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1752de8f3a83SDaniel Borkmann 1753158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1754158f323bSEric Dumazet * For the moment, we really care of rx path, or 1755158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1756158f323bSEric Dumazet */ 1757158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1758158f323bSEric Dumazet skb->truesize += size - osize; 1759158f323bSEric Dumazet 17601da177e4SLinus Torvalds return 0; 17611da177e4SLinus Torvalds 1762a6686f2fSShirley Ma nofrags: 1763a6686f2fSShirley Ma kfree(data); 17641da177e4SLinus Torvalds nodata: 17651da177e4SLinus Torvalds return -ENOMEM; 17661da177e4SLinus Torvalds } 1767b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 17681da177e4SLinus Torvalds 17691da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 17721da177e4SLinus Torvalds { 17731da177e4SLinus Torvalds struct sk_buff *skb2; 17741da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 17751da177e4SLinus Torvalds 17761da177e4SLinus Torvalds if (delta <= 0) 17771da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 17781da177e4SLinus Torvalds else { 17791da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 17801da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 17811da177e4SLinus Torvalds GFP_ATOMIC)) { 17821da177e4SLinus Torvalds kfree_skb(skb2); 17831da177e4SLinus Torvalds skb2 = NULL; 17841da177e4SLinus Torvalds } 17851da177e4SLinus Torvalds } 17861da177e4SLinus Torvalds return skb2; 17871da177e4SLinus Torvalds } 1788b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 17891da177e4SLinus Torvalds 17902b88cba5SEric Dumazet int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 17912b88cba5SEric Dumazet { 17922b88cba5SEric Dumazet unsigned int saved_end_offset, saved_truesize; 17932b88cba5SEric Dumazet struct skb_shared_info *shinfo; 17942b88cba5SEric Dumazet int res; 17952b88cba5SEric Dumazet 17962b88cba5SEric Dumazet saved_end_offset = skb_end_offset(skb); 17972b88cba5SEric Dumazet saved_truesize = skb->truesize; 17982b88cba5SEric Dumazet 17992b88cba5SEric Dumazet res = pskb_expand_head(skb, 0, 0, pri); 18002b88cba5SEric Dumazet if (res) 18012b88cba5SEric Dumazet return res; 18022b88cba5SEric Dumazet 18032b88cba5SEric Dumazet skb->truesize = saved_truesize; 18042b88cba5SEric Dumazet 18052b88cba5SEric Dumazet if (likely(skb_end_offset(skb) == saved_end_offset)) 18062b88cba5SEric Dumazet return 0; 18072b88cba5SEric Dumazet 18082b88cba5SEric Dumazet shinfo = skb_shinfo(skb); 18092b88cba5SEric Dumazet 18102b88cba5SEric Dumazet /* We are about to change back skb->end, 18112b88cba5SEric Dumazet * we need to move skb_shinfo() to its new location. 18122b88cba5SEric Dumazet */ 18132b88cba5SEric Dumazet memmove(skb->head + saved_end_offset, 18142b88cba5SEric Dumazet shinfo, 18152b88cba5SEric Dumazet offsetof(struct skb_shared_info, frags[shinfo->nr_frags])); 18162b88cba5SEric Dumazet 18172b88cba5SEric Dumazet skb_set_end_offset(skb, saved_end_offset); 18182b88cba5SEric Dumazet 18192b88cba5SEric Dumazet return 0; 18202b88cba5SEric Dumazet } 18212b88cba5SEric Dumazet 18221da177e4SLinus Torvalds /** 1823f1260ff1SVasily Averin * skb_expand_head - reallocate header of &sk_buff 1824f1260ff1SVasily Averin * @skb: buffer to reallocate 1825f1260ff1SVasily Averin * @headroom: needed headroom 1826f1260ff1SVasily Averin * 1827f1260ff1SVasily Averin * Unlike skb_realloc_headroom, this one does not allocate a new skb 1828f1260ff1SVasily Averin * if possible; copies skb->sk to new skb as needed 1829f1260ff1SVasily Averin * and frees original skb in case of failures. 1830f1260ff1SVasily Averin * 1831f1260ff1SVasily Averin * It expect increased headroom and generates warning otherwise. 1832f1260ff1SVasily Averin */ 1833f1260ff1SVasily Averin 1834f1260ff1SVasily Averin struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom) 1835f1260ff1SVasily Averin { 1836f1260ff1SVasily Averin int delta = headroom - skb_headroom(skb); 18377f678defSVasily Averin int osize = skb_end_offset(skb); 18387f678defSVasily Averin struct sock *sk = skb->sk; 1839f1260ff1SVasily Averin 1840f1260ff1SVasily Averin if (WARN_ONCE(delta <= 0, 1841f1260ff1SVasily Averin "%s is expecting an increase in the headroom", __func__)) 1842f1260ff1SVasily Averin return skb; 1843f1260ff1SVasily Averin 18447f678defSVasily Averin delta = SKB_DATA_ALIGN(delta); 18457f678defSVasily Averin /* pskb_expand_head() might crash, if skb is shared. */ 18467f678defSVasily Averin if (skb_shared(skb) || !is_skb_wmem(skb)) { 1847f1260ff1SVasily Averin struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1848f1260ff1SVasily Averin 18497f678defSVasily Averin if (unlikely(!nskb)) 18507f678defSVasily Averin goto fail; 18517f678defSVasily Averin 18527f678defSVasily Averin if (sk) 18537f678defSVasily Averin skb_set_owner_w(nskb, sk); 1854f1260ff1SVasily Averin consume_skb(skb); 1855f1260ff1SVasily Averin skb = nskb; 1856f1260ff1SVasily Averin } 18577f678defSVasily Averin if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC)) 18587f678defSVasily Averin goto fail; 18597f678defSVasily Averin 18607f678defSVasily Averin if (sk && is_skb_wmem(skb)) { 18617f678defSVasily Averin delta = skb_end_offset(skb) - osize; 18627f678defSVasily Averin refcount_add(delta, &sk->sk_wmem_alloc); 18637f678defSVasily Averin skb->truesize += delta; 1864f1260ff1SVasily Averin } 1865f1260ff1SVasily Averin return skb; 18667f678defSVasily Averin 18677f678defSVasily Averin fail: 18687f678defSVasily Averin kfree_skb(skb); 18697f678defSVasily Averin return NULL; 1870f1260ff1SVasily Averin } 1871f1260ff1SVasily Averin EXPORT_SYMBOL(skb_expand_head); 1872f1260ff1SVasily Averin 1873f1260ff1SVasily Averin /** 18741da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 18751da177e4SLinus Torvalds * @skb: buffer to copy 18761da177e4SLinus Torvalds * @newheadroom: new free bytes at head 18771da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 18781da177e4SLinus Torvalds * @gfp_mask: allocation priority 18791da177e4SLinus Torvalds * 18801da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 18811da177e4SLinus Torvalds * allocate additional space. 18821da177e4SLinus Torvalds * 18831da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 18841da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 18851da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 18861da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 18871da177e4SLinus Torvalds * 18881da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 18891da177e4SLinus Torvalds * is called from an interrupt. 18901da177e4SLinus Torvalds */ 18911da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 189286a76cafSVictor Fusco int newheadroom, int newtailroom, 1893dd0fc66fSAl Viro gfp_t gfp_mask) 18941da177e4SLinus Torvalds { 18951da177e4SLinus Torvalds /* 18961da177e4SLinus Torvalds * Allocate the copy buffer 18971da177e4SLinus Torvalds */ 1898c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1899c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1900c93bdd0eSMel Gorman NUMA_NO_NODE); 1901efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 19021da177e4SLinus Torvalds int head_copy_len, head_copy_off; 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds if (!n) 19051da177e4SLinus Torvalds return NULL; 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds skb_reserve(n, newheadroom); 19081da177e4SLinus Torvalds 19091da177e4SLinus Torvalds /* Set the tail pointer and length */ 19101da177e4SLinus Torvalds skb_put(n, skb->len); 19111da177e4SLinus Torvalds 1912efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 19131da177e4SLinus Torvalds head_copy_off = 0; 19141da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 19151da177e4SLinus Torvalds head_copy_len = newheadroom; 19161da177e4SLinus Torvalds else 19171da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 19181da177e4SLinus Torvalds 19191da177e4SLinus Torvalds /* Copy the linear header and data. */ 19209f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 19219f77fad3STim Hansen skb->len + head_copy_len)); 19221da177e4SLinus Torvalds 192308303c18SIlya Lesokhin skb_copy_header(n, skb); 19241da177e4SLinus Torvalds 1925030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1926efd1e8d5SPatrick McHardy 19271da177e4SLinus Torvalds return n; 19281da177e4SLinus Torvalds } 1929b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds /** 1932cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 19331da177e4SLinus Torvalds * @skb: buffer to pad 19341da177e4SLinus Torvalds * @pad: space to pad 1935cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 19361da177e4SLinus Torvalds * 19371da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 19381da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 19391da177e4SLinus Torvalds * beyond the buffer end onto the wire. 19401da177e4SLinus Torvalds * 1941cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1942cd0a137aSFlorian Fainelli * if @free_on_error is true. 19431da177e4SLinus Torvalds */ 19441da177e4SLinus Torvalds 1945cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 19461da177e4SLinus Torvalds { 19475b057c6bSHerbert Xu int err; 19485b057c6bSHerbert Xu int ntail; 19491da177e4SLinus Torvalds 19501da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 19515b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 19521da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 19535b057c6bSHerbert Xu return 0; 19541da177e4SLinus Torvalds } 19551da177e4SLinus Torvalds 19564305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 19575b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 19585b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 19595b057c6bSHerbert Xu if (unlikely(err)) 19605b057c6bSHerbert Xu goto free_skb; 19615b057c6bSHerbert Xu } 19625b057c6bSHerbert Xu 19635b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 19645b057c6bSHerbert Xu * to be audited. 19655b057c6bSHerbert Xu */ 19665b057c6bSHerbert Xu err = skb_linearize(skb); 19675b057c6bSHerbert Xu if (unlikely(err)) 19685b057c6bSHerbert Xu goto free_skb; 19695b057c6bSHerbert Xu 19705b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 19715b057c6bSHerbert Xu return 0; 19725b057c6bSHerbert Xu 19735b057c6bSHerbert Xu free_skb: 1974cd0a137aSFlorian Fainelli if (free_on_error) 19751da177e4SLinus Torvalds kfree_skb(skb); 19765b057c6bSHerbert Xu return err; 19771da177e4SLinus Torvalds } 1978cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 19791da177e4SLinus Torvalds 19800dde3e16SIlpo Järvinen /** 19810c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 19820c7ddf36SMathias Krause * @skb: start of the buffer to use 19830c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 19840c7ddf36SMathias Krause * @len: amount of data to add 19850c7ddf36SMathias Krause * 19860c7ddf36SMathias Krause * This function extends the used data area of the potentially 19870c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 19880c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 19890c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 19900c7ddf36SMathias Krause * returned. 19910c7ddf36SMathias Krause */ 19920c7ddf36SMathias Krause 19934df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 19940c7ddf36SMathias Krause { 19950c7ddf36SMathias Krause if (tail != skb) { 19960c7ddf36SMathias Krause skb->data_len += len; 19970c7ddf36SMathias Krause skb->len += len; 19980c7ddf36SMathias Krause } 19990c7ddf36SMathias Krause return skb_put(tail, len); 20000c7ddf36SMathias Krause } 20010c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 20020c7ddf36SMathias Krause 20030c7ddf36SMathias Krause /** 20040dde3e16SIlpo Järvinen * skb_put - add data to a buffer 20050dde3e16SIlpo Järvinen * @skb: buffer to use 20060dde3e16SIlpo Järvinen * @len: amount of data to add 20070dde3e16SIlpo Järvinen * 20080dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 20090dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 20100dde3e16SIlpo Järvinen * first byte of the extra data is returned. 20110dde3e16SIlpo Järvinen */ 20124df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 20130dde3e16SIlpo Järvinen { 20144df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 20150dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 20160dde3e16SIlpo Järvinen skb->tail += len; 20170dde3e16SIlpo Järvinen skb->len += len; 20180dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 20190dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 20200dde3e16SIlpo Järvinen return tmp; 20210dde3e16SIlpo Järvinen } 20220dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 20230dde3e16SIlpo Järvinen 20246be8ac2fSIlpo Järvinen /** 2025c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 2026c2aa270aSIlpo Järvinen * @skb: buffer to use 2027c2aa270aSIlpo Järvinen * @len: amount of data to add 2028c2aa270aSIlpo Järvinen * 2029c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 2030c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 2031c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 2032c2aa270aSIlpo Järvinen */ 2033d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 2034c2aa270aSIlpo Järvinen { 2035c2aa270aSIlpo Järvinen skb->data -= len; 2036c2aa270aSIlpo Järvinen skb->len += len; 2037c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 2038c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 2039c2aa270aSIlpo Järvinen return skb->data; 2040c2aa270aSIlpo Järvinen } 2041c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 2042c2aa270aSIlpo Järvinen 2043c2aa270aSIlpo Järvinen /** 20446be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 20456be8ac2fSIlpo Järvinen * @skb: buffer to use 20466be8ac2fSIlpo Järvinen * @len: amount of data to remove 20476be8ac2fSIlpo Järvinen * 20486be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 20496be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 20506be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 20516be8ac2fSIlpo Järvinen * the old data. 20526be8ac2fSIlpo Järvinen */ 2053af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 20546be8ac2fSIlpo Järvinen { 205547d29646SDavid S. Miller return skb_pull_inline(skb, len); 20566be8ac2fSIlpo Järvinen } 20576be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 20586be8ac2fSIlpo Järvinen 2059419ae74eSIlpo Järvinen /** 206013244cccSLuiz Augusto von Dentz * skb_pull_data - remove data from the start of a buffer returning its 206113244cccSLuiz Augusto von Dentz * original position. 206213244cccSLuiz Augusto von Dentz * @skb: buffer to use 206313244cccSLuiz Augusto von Dentz * @len: amount of data to remove 206413244cccSLuiz Augusto von Dentz * 206513244cccSLuiz Augusto von Dentz * This function removes data from the start of a buffer, returning 206613244cccSLuiz Augusto von Dentz * the memory to the headroom. A pointer to the original data in the buffer 206713244cccSLuiz Augusto von Dentz * is returned after checking if there is enough data to pull. Once the 206813244cccSLuiz Augusto von Dentz * data has been pulled future pushes will overwrite the old data. 206913244cccSLuiz Augusto von Dentz */ 207013244cccSLuiz Augusto von Dentz void *skb_pull_data(struct sk_buff *skb, size_t len) 207113244cccSLuiz Augusto von Dentz { 207213244cccSLuiz Augusto von Dentz void *data = skb->data; 207313244cccSLuiz Augusto von Dentz 207413244cccSLuiz Augusto von Dentz if (skb->len < len) 207513244cccSLuiz Augusto von Dentz return NULL; 207613244cccSLuiz Augusto von Dentz 207713244cccSLuiz Augusto von Dentz skb_pull(skb, len); 207813244cccSLuiz Augusto von Dentz 207913244cccSLuiz Augusto von Dentz return data; 208013244cccSLuiz Augusto von Dentz } 208113244cccSLuiz Augusto von Dentz EXPORT_SYMBOL(skb_pull_data); 208213244cccSLuiz Augusto von Dentz 208313244cccSLuiz Augusto von Dentz /** 2084419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 2085419ae74eSIlpo Järvinen * @skb: buffer to alter 2086419ae74eSIlpo Järvinen * @len: new length 2087419ae74eSIlpo Järvinen * 2088419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 2089419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 2090419ae74eSIlpo Järvinen * The skb must be linear. 2091419ae74eSIlpo Järvinen */ 2092419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 2093419ae74eSIlpo Järvinen { 2094419ae74eSIlpo Järvinen if (skb->len > len) 2095419ae74eSIlpo Järvinen __skb_trim(skb, len); 2096419ae74eSIlpo Järvinen } 2097419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 2098419ae74eSIlpo Järvinen 20993cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 21001da177e4SLinus Torvalds */ 21011da177e4SLinus Torvalds 21023cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 21031da177e4SLinus Torvalds { 210427b437c8SHerbert Xu struct sk_buff **fragp; 210527b437c8SHerbert Xu struct sk_buff *frag; 21061da177e4SLinus Torvalds int offset = skb_headlen(skb); 21071da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 21081da177e4SLinus Torvalds int i; 210927b437c8SHerbert Xu int err; 211027b437c8SHerbert Xu 211127b437c8SHerbert Xu if (skb_cloned(skb) && 211227b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 211327b437c8SHerbert Xu return err; 21141da177e4SLinus Torvalds 2115f4d26fb3SHerbert Xu i = 0; 2116f4d26fb3SHerbert Xu if (offset >= len) 2117f4d26fb3SHerbert Xu goto drop_pages; 2118f4d26fb3SHerbert Xu 2119f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 21209e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 212127b437c8SHerbert Xu 212227b437c8SHerbert Xu if (end < len) { 21231da177e4SLinus Torvalds offset = end; 212427b437c8SHerbert Xu continue; 21251da177e4SLinus Torvalds } 21261da177e4SLinus Torvalds 21279e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 212827b437c8SHerbert Xu 2129f4d26fb3SHerbert Xu drop_pages: 213027b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 213127b437c8SHerbert Xu 213227b437c8SHerbert Xu for (; i < nfrags; i++) 2133ea2ab693SIan Campbell skb_frag_unref(skb, i); 213427b437c8SHerbert Xu 213521dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 213627b437c8SHerbert Xu skb_drop_fraglist(skb); 2137f4d26fb3SHerbert Xu goto done; 213827b437c8SHerbert Xu } 213927b437c8SHerbert Xu 214027b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 214127b437c8SHerbert Xu fragp = &frag->next) { 214227b437c8SHerbert Xu int end = offset + frag->len; 214327b437c8SHerbert Xu 214427b437c8SHerbert Xu if (skb_shared(frag)) { 214527b437c8SHerbert Xu struct sk_buff *nfrag; 214627b437c8SHerbert Xu 214727b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 214827b437c8SHerbert Xu if (unlikely(!nfrag)) 214927b437c8SHerbert Xu return -ENOMEM; 215027b437c8SHerbert Xu 215127b437c8SHerbert Xu nfrag->next = frag->next; 215285bb2a60SEric Dumazet consume_skb(frag); 215327b437c8SHerbert Xu frag = nfrag; 215427b437c8SHerbert Xu *fragp = frag; 215527b437c8SHerbert Xu } 215627b437c8SHerbert Xu 215727b437c8SHerbert Xu if (end < len) { 215827b437c8SHerbert Xu offset = end; 215927b437c8SHerbert Xu continue; 216027b437c8SHerbert Xu } 216127b437c8SHerbert Xu 216227b437c8SHerbert Xu if (end > len && 216327b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 216427b437c8SHerbert Xu return err; 216527b437c8SHerbert Xu 216627b437c8SHerbert Xu if (frag->next) 216727b437c8SHerbert Xu skb_drop_list(&frag->next); 216827b437c8SHerbert Xu break; 216927b437c8SHerbert Xu } 217027b437c8SHerbert Xu 2171f4d26fb3SHerbert Xu done: 217227b437c8SHerbert Xu if (len > skb_headlen(skb)) { 21731da177e4SLinus Torvalds skb->data_len -= skb->len - len; 21741da177e4SLinus Torvalds skb->len = len; 21751da177e4SLinus Torvalds } else { 21761da177e4SLinus Torvalds skb->len = len; 21771da177e4SLinus Torvalds skb->data_len = 0; 217827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 21791da177e4SLinus Torvalds } 21801da177e4SLinus Torvalds 2181c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2182c21b48ccSEric Dumazet skb_condense(skb); 21831da177e4SLinus Torvalds return 0; 21841da177e4SLinus Torvalds } 2185b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 21861da177e4SLinus Torvalds 218788078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 218888078d98SEric Dumazet */ 218988078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 219088078d98SEric Dumazet { 219188078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 219288078d98SEric Dumazet int delta = skb->len - len; 219388078d98SEric Dumazet 2194d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2195d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2196d55bef50SDimitris Michailidis len); 219754970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 219854970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 219954970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 220054970a2fSVasily Averin 220154970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 220254970a2fSVasily Averin return -EINVAL; 220388078d98SEric Dumazet } 220488078d98SEric Dumazet return __pskb_trim(skb, len); 220588078d98SEric Dumazet } 220688078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 220788078d98SEric Dumazet 22081da177e4SLinus Torvalds /** 22091da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 22101da177e4SLinus Torvalds * @skb: buffer to reallocate 22111da177e4SLinus Torvalds * @delta: number of bytes to advance tail 22121da177e4SLinus Torvalds * 22131da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 22141da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 22151da177e4SLinus Torvalds * data from fragmented part. 22161da177e4SLinus Torvalds * 22171da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 22181da177e4SLinus Torvalds * 22191da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 22201da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 22211da177e4SLinus Torvalds * 22221da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 22231da177e4SLinus Torvalds * reloaded after call to this function. 22241da177e4SLinus Torvalds */ 22251da177e4SLinus Torvalds 22261da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 22271da177e4SLinus Torvalds * when it is necessary. 22281da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 22291da177e4SLinus Torvalds * 2. It may change skb pointers. 22301da177e4SLinus Torvalds * 22311da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 22321da177e4SLinus Torvalds */ 2233af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 22341da177e4SLinus Torvalds { 22351da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 22361da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 22371da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 22381da177e4SLinus Torvalds */ 22394305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 22401da177e4SLinus Torvalds 22411da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 22421da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 22431da177e4SLinus Torvalds GFP_ATOMIC)) 22441da177e4SLinus Torvalds return NULL; 22451da177e4SLinus Torvalds } 22461da177e4SLinus Torvalds 22479f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 22489f77fad3STim Hansen skb_tail_pointer(skb), delta)); 22491da177e4SLinus Torvalds 22501da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 22511da177e4SLinus Torvalds * size of pulled pages. Superb. 22521da177e4SLinus Torvalds */ 225321dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 22541da177e4SLinus Torvalds goto pull_pages; 22551da177e4SLinus Torvalds 22561da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 22571da177e4SLinus Torvalds eat = delta; 22581da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22599e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22609e903e08SEric Dumazet 22619e903e08SEric Dumazet if (size >= eat) 22621da177e4SLinus Torvalds goto pull_pages; 22639e903e08SEric Dumazet eat -= size; 22641da177e4SLinus Torvalds } 22651da177e4SLinus Torvalds 22661da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 226709001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 22681da177e4SLinus Torvalds * but taking into account that pulling is expected to 22691da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 22701da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 22711da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 22721da177e4SLinus Torvalds */ 22731da177e4SLinus Torvalds if (eat) { 22741da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 22751da177e4SLinus Torvalds struct sk_buff *clone = NULL; 22761da177e4SLinus Torvalds struct sk_buff *insp = NULL; 22771da177e4SLinus Torvalds 22781da177e4SLinus Torvalds do { 22791da177e4SLinus Torvalds if (list->len <= eat) { 22801da177e4SLinus Torvalds /* Eaten as whole. */ 22811da177e4SLinus Torvalds eat -= list->len; 22821da177e4SLinus Torvalds list = list->next; 22831da177e4SLinus Torvalds insp = list; 22841da177e4SLinus Torvalds } else { 22851da177e4SLinus Torvalds /* Eaten partially. */ 22861da177e4SLinus Torvalds 22871da177e4SLinus Torvalds if (skb_shared(list)) { 22881da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 22891da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 22901da177e4SLinus Torvalds if (!clone) 22911da177e4SLinus Torvalds return NULL; 22921da177e4SLinus Torvalds insp = list->next; 22931da177e4SLinus Torvalds list = clone; 22941da177e4SLinus Torvalds } else { 22951da177e4SLinus Torvalds /* This may be pulled without 22961da177e4SLinus Torvalds * problems. */ 22971da177e4SLinus Torvalds insp = list; 22981da177e4SLinus Torvalds } 22991da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 23001da177e4SLinus Torvalds kfree_skb(clone); 23011da177e4SLinus Torvalds return NULL; 23021da177e4SLinus Torvalds } 23031da177e4SLinus Torvalds break; 23041da177e4SLinus Torvalds } 23051da177e4SLinus Torvalds } while (eat); 23061da177e4SLinus Torvalds 23071da177e4SLinus Torvalds /* Free pulled out fragments. */ 23081da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 23091da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 2310ef527f96SEric Dumazet consume_skb(list); 23111da177e4SLinus Torvalds } 23121da177e4SLinus Torvalds /* And insert new clone at head. */ 23131da177e4SLinus Torvalds if (clone) { 23141da177e4SLinus Torvalds clone->next = list; 23151da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 23161da177e4SLinus Torvalds } 23171da177e4SLinus Torvalds } 23181da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 23191da177e4SLinus Torvalds 23201da177e4SLinus Torvalds pull_pages: 23211da177e4SLinus Torvalds eat = delta; 23221da177e4SLinus Torvalds k = 0; 23231da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 23249e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 23259e903e08SEric Dumazet 23269e903e08SEric Dumazet if (size <= eat) { 2327ea2ab693SIan Campbell skb_frag_unref(skb, i); 23289e903e08SEric Dumazet eat -= size; 23291da177e4SLinus Torvalds } else { 2330b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2331b54c9d5bSJonathan Lemon 2332b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 23331da177e4SLinus Torvalds if (eat) { 2334b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2335b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 23363ccc6c6fSlinzhang if (!i) 23373ccc6c6fSlinzhang goto end; 23381da177e4SLinus Torvalds eat = 0; 23391da177e4SLinus Torvalds } 23401da177e4SLinus Torvalds k++; 23411da177e4SLinus Torvalds } 23421da177e4SLinus Torvalds } 23431da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 23441da177e4SLinus Torvalds 23453ccc6c6fSlinzhang end: 23461da177e4SLinus Torvalds skb->tail += delta; 23471da177e4SLinus Torvalds skb->data_len -= delta; 23481da177e4SLinus Torvalds 23491f8b977aSWillem de Bruijn if (!skb->data_len) 23501f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 23511f8b977aSWillem de Bruijn 235227a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 23531da177e4SLinus Torvalds } 2354b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 23551da177e4SLinus Torvalds 235622019b17SEric Dumazet /** 235722019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 235822019b17SEric Dumazet * @skb: source skb 235922019b17SEric Dumazet * @offset: offset in source 236022019b17SEric Dumazet * @to: destination buffer 236122019b17SEric Dumazet * @len: number of bytes to copy 236222019b17SEric Dumazet * 236322019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 236422019b17SEric Dumazet * destination buffer. 236522019b17SEric Dumazet * 236622019b17SEric Dumazet * CAUTION ! : 236722019b17SEric Dumazet * If its prototype is ever changed, 236822019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 236922019b17SEric Dumazet * since it is called from BPF assembly code. 237022019b17SEric Dumazet */ 23711da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 23721da177e4SLinus Torvalds { 23731a028e50SDavid S. Miller int start = skb_headlen(skb); 2374fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2375fbb398a8SDavid S. Miller int i, copy; 23761da177e4SLinus Torvalds 23771da177e4SLinus Torvalds if (offset > (int)skb->len - len) 23781da177e4SLinus Torvalds goto fault; 23791da177e4SLinus Torvalds 23801da177e4SLinus Torvalds /* Copy header. */ 23811a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 23821da177e4SLinus Torvalds if (copy > len) 23831da177e4SLinus Torvalds copy = len; 2384d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 23851da177e4SLinus Torvalds if ((len -= copy) == 0) 23861da177e4SLinus Torvalds return 0; 23871da177e4SLinus Torvalds offset += copy; 23881da177e4SLinus Torvalds to += copy; 23891da177e4SLinus Torvalds } 23901da177e4SLinus Torvalds 23911da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 23921a028e50SDavid S. Miller int end; 239351c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 23941da177e4SLinus Torvalds 2395547b792cSIlpo Järvinen WARN_ON(start > offset + len); 23961a028e50SDavid S. Miller 239751c56b00SEric Dumazet end = start + skb_frag_size(f); 23981da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2399c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2400c613c209SWillem de Bruijn struct page *p; 24011da177e4SLinus Torvalds u8 *vaddr; 24021da177e4SLinus Torvalds 24031da177e4SLinus Torvalds if (copy > len) 24041da177e4SLinus Torvalds copy = len; 24051da177e4SLinus Torvalds 2406c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2407b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2408c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2409c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2410c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 241151c56b00SEric Dumazet kunmap_atomic(vaddr); 2412c613c209SWillem de Bruijn } 24131da177e4SLinus Torvalds 24141da177e4SLinus Torvalds if ((len -= copy) == 0) 24151da177e4SLinus Torvalds return 0; 24161da177e4SLinus Torvalds offset += copy; 24171da177e4SLinus Torvalds to += copy; 24181da177e4SLinus Torvalds } 24191a028e50SDavid S. Miller start = end; 24201da177e4SLinus Torvalds } 24211da177e4SLinus Torvalds 2422fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 24231a028e50SDavid S. Miller int end; 24241da177e4SLinus Torvalds 2425547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24261a028e50SDavid S. Miller 2427fbb398a8SDavid S. Miller end = start + frag_iter->len; 24281da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 24291da177e4SLinus Torvalds if (copy > len) 24301da177e4SLinus Torvalds copy = len; 2431fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 24321da177e4SLinus Torvalds goto fault; 24331da177e4SLinus Torvalds if ((len -= copy) == 0) 24341da177e4SLinus Torvalds return 0; 24351da177e4SLinus Torvalds offset += copy; 24361da177e4SLinus Torvalds to += copy; 24371da177e4SLinus Torvalds } 24381a028e50SDavid S. Miller start = end; 24391da177e4SLinus Torvalds } 2440a6686f2fSShirley Ma 24411da177e4SLinus Torvalds if (!len) 24421da177e4SLinus Torvalds return 0; 24431da177e4SLinus Torvalds 24441da177e4SLinus Torvalds fault: 24451da177e4SLinus Torvalds return -EFAULT; 24461da177e4SLinus Torvalds } 2447b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 24481da177e4SLinus Torvalds 24499c55e01cSJens Axboe /* 24509c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 24519c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 24529c55e01cSJens Axboe */ 24539c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 24549c55e01cSJens Axboe { 24558b9d3728SJarek Poplawski put_page(spd->pages[i]); 24568b9d3728SJarek Poplawski } 24579c55e01cSJens Axboe 2458a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 24594fb66994SJarek Poplawski unsigned int *offset, 246018aafc62SEric Dumazet struct sock *sk) 24618b9d3728SJarek Poplawski { 24625640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 24638b9d3728SJarek Poplawski 24645640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 24658b9d3728SJarek Poplawski return NULL; 24664fb66994SJarek Poplawski 24675640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 24684fb66994SJarek Poplawski 24695640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 24705640f768SEric Dumazet page_address(page) + *offset, *len); 24715640f768SEric Dumazet *offset = pfrag->offset; 24725640f768SEric Dumazet pfrag->offset += *len; 24734fb66994SJarek Poplawski 24745640f768SEric Dumazet return pfrag->page; 24759c55e01cSJens Axboe } 24769c55e01cSJens Axboe 247741c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 247841c73a0dSEric Dumazet struct page *page, 247941c73a0dSEric Dumazet unsigned int offset) 248041c73a0dSEric Dumazet { 248141c73a0dSEric Dumazet return spd->nr_pages && 248241c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 248341c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 248441c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 248541c73a0dSEric Dumazet } 248641c73a0dSEric Dumazet 24879c55e01cSJens Axboe /* 24889c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 24899c55e01cSJens Axboe */ 2490a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 249135f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 24924fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 249318aafc62SEric Dumazet bool linear, 24947a67e56fSJarek Poplawski struct sock *sk) 24959c55e01cSJens Axboe { 249641c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2497a108d5f3SDavid S. Miller return true; 24989c55e01cSJens Axboe 24998b9d3728SJarek Poplawski if (linear) { 250018aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 25018b9d3728SJarek Poplawski if (!page) 2502a108d5f3SDavid S. Miller return true; 250341c73a0dSEric Dumazet } 250441c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 250541c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2506a108d5f3SDavid S. Miller return false; 250741c73a0dSEric Dumazet } 25088b9d3728SJarek Poplawski get_page(page); 25099c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 25104fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 25119c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 25129c55e01cSJens Axboe spd->nr_pages++; 25138b9d3728SJarek Poplawski 2514a108d5f3SDavid S. Miller return false; 25159c55e01cSJens Axboe } 25169c55e01cSJens Axboe 2517a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 25182870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 251918aafc62SEric Dumazet unsigned int *len, 2520d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 252135f3d14dSJens Axboe struct sock *sk, 252235f3d14dSJens Axboe struct pipe_inode_info *pipe) 25239c55e01cSJens Axboe { 25242870c43dSOctavian Purdila if (!*len) 2525a108d5f3SDavid S. Miller return true; 25269c55e01cSJens Axboe 25272870c43dSOctavian Purdila /* skip this segment if already processed */ 25282870c43dSOctavian Purdila if (*off >= plen) { 25292870c43dSOctavian Purdila *off -= plen; 2530a108d5f3SDavid S. Miller return false; 25312870c43dSOctavian Purdila } 25322870c43dSOctavian Purdila 25332870c43dSOctavian Purdila /* ignore any bits we already processed */ 25349ca1b22dSEric Dumazet poff += *off; 25359ca1b22dSEric Dumazet plen -= *off; 25362870c43dSOctavian Purdila *off = 0; 25372870c43dSOctavian Purdila 253818aafc62SEric Dumazet do { 253918aafc62SEric Dumazet unsigned int flen = min(*len, plen); 25402870c43dSOctavian Purdila 254118aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 254218aafc62SEric Dumazet linear, sk)) 2543a108d5f3SDavid S. Miller return true; 254418aafc62SEric Dumazet poff += flen; 254518aafc62SEric Dumazet plen -= flen; 25462870c43dSOctavian Purdila *len -= flen; 254718aafc62SEric Dumazet } while (*len && plen); 25482870c43dSOctavian Purdila 2549a108d5f3SDavid S. Miller return false; 2550db43a282SOctavian Purdila } 25519c55e01cSJens Axboe 25529c55e01cSJens Axboe /* 2553a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 25542870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 25559c55e01cSJens Axboe */ 2556a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 255735f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 255835f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 25592870c43dSOctavian Purdila { 25602870c43dSOctavian Purdila int seg; 2561fa9835e5STom Herbert struct sk_buff *iter; 25629c55e01cSJens Axboe 25631d0c0b32SEric Dumazet /* map the linear part : 25642996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 25652996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 25662996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 25679c55e01cSJens Axboe */ 25682870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 25692870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 25702870c43dSOctavian Purdila skb_headlen(skb), 257118aafc62SEric Dumazet offset, len, spd, 25723a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 25731d0c0b32SEric Dumazet sk, pipe)) 2574a108d5f3SDavid S. Miller return true; 25759c55e01cSJens Axboe 25769c55e01cSJens Axboe /* 25779c55e01cSJens Axboe * then map the fragments 25789c55e01cSJens Axboe */ 25799c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 25809c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 25819c55e01cSJens Axboe 2582ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2583b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 258418aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2585a108d5f3SDavid S. Miller return true; 25869c55e01cSJens Axboe } 25879c55e01cSJens Axboe 2588fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2589fa9835e5STom Herbert if (*offset >= iter->len) { 2590fa9835e5STom Herbert *offset -= iter->len; 2591fa9835e5STom Herbert continue; 2592fa9835e5STom Herbert } 2593fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2594fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2595fa9835e5STom Herbert * case. 2596fa9835e5STom Herbert */ 2597fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2598fa9835e5STom Herbert return true; 2599fa9835e5STom Herbert } 2600fa9835e5STom Herbert 2601a108d5f3SDavid S. Miller return false; 26029c55e01cSJens Axboe } 26039c55e01cSJens Axboe 26049c55e01cSJens Axboe /* 26059c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2606fa9835e5STom Herbert * the fragments, and the frag list. 26079c55e01cSJens Axboe */ 2608a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 26099c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 261025869262SAl Viro unsigned int flags) 26119c55e01cSJens Axboe { 261241c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 261341c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 26149c55e01cSJens Axboe struct splice_pipe_desc spd = { 26159c55e01cSJens Axboe .pages = pages, 26169c55e01cSJens Axboe .partial = partial, 2617047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 261828a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 26199c55e01cSJens Axboe .spd_release = sock_spd_release, 26209c55e01cSJens Axboe }; 262135f3d14dSJens Axboe int ret = 0; 262235f3d14dSJens Axboe 2623fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 26249c55e01cSJens Axboe 2625a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 262625869262SAl Viro ret = splice_to_pipe(pipe, &spd); 26279c55e01cSJens Axboe 262835f3d14dSJens Axboe return ret; 26299c55e01cSJens Axboe } 26302b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 26319c55e01cSJens Axboe 26320739cd28SCong Wang static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg, 26330739cd28SCong Wang struct kvec *vec, size_t num, size_t size) 26340739cd28SCong Wang { 26350739cd28SCong Wang struct socket *sock = sk->sk_socket; 26360739cd28SCong Wang 26370739cd28SCong Wang if (!sock) 26380739cd28SCong Wang return -EINVAL; 26390739cd28SCong Wang return kernel_sendmsg(sock, msg, vec, num, size); 26400739cd28SCong Wang } 26410739cd28SCong Wang 26420739cd28SCong Wang static int sendpage_unlocked(struct sock *sk, struct page *page, int offset, 26430739cd28SCong Wang size_t size, int flags) 26440739cd28SCong Wang { 26450739cd28SCong Wang struct socket *sock = sk->sk_socket; 26460739cd28SCong Wang 26470739cd28SCong Wang if (!sock) 26480739cd28SCong Wang return -EINVAL; 26490739cd28SCong Wang return kernel_sendpage(sock, page, offset, size, flags); 26500739cd28SCong Wang } 26510739cd28SCong Wang 26520739cd28SCong Wang typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg, 26530739cd28SCong Wang struct kvec *vec, size_t num, size_t size); 26540739cd28SCong Wang typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset, 26550739cd28SCong Wang size_t size, int flags); 26560739cd28SCong Wang static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, 26570739cd28SCong Wang int len, sendmsg_func sendmsg, sendpage_func sendpage) 265820bf50deSTom Herbert { 265920bf50deSTom Herbert unsigned int orig_len = len; 266020bf50deSTom Herbert struct sk_buff *head = skb; 266120bf50deSTom Herbert unsigned short fragidx; 266220bf50deSTom Herbert int slen, ret; 266320bf50deSTom Herbert 266420bf50deSTom Herbert do_frag_list: 266520bf50deSTom Herbert 266620bf50deSTom Herbert /* Deal with head data */ 266720bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 266820bf50deSTom Herbert struct kvec kv; 266920bf50deSTom Herbert struct msghdr msg; 267020bf50deSTom Herbert 267120bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 267220bf50deSTom Herbert kv.iov_base = skb->data + offset; 2673db5980d8SJohn Fastabend kv.iov_len = slen; 267420bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2675bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 267620bf50deSTom Herbert 26770739cd28SCong Wang ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked, 26780739cd28SCong Wang sendmsg_unlocked, sk, &msg, &kv, 1, slen); 267920bf50deSTom Herbert if (ret <= 0) 268020bf50deSTom Herbert goto error; 268120bf50deSTom Herbert 268220bf50deSTom Herbert offset += ret; 268320bf50deSTom Herbert len -= ret; 268420bf50deSTom Herbert } 268520bf50deSTom Herbert 268620bf50deSTom Herbert /* All the data was skb head? */ 268720bf50deSTom Herbert if (!len) 268820bf50deSTom Herbert goto out; 268920bf50deSTom Herbert 269020bf50deSTom Herbert /* Make offset relative to start of frags */ 269120bf50deSTom Herbert offset -= skb_headlen(skb); 269220bf50deSTom Herbert 269320bf50deSTom Herbert /* Find where we are in frag list */ 269420bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 269520bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 269620bf50deSTom Herbert 2697d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 269820bf50deSTom Herbert break; 269920bf50deSTom Herbert 2700d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 270120bf50deSTom Herbert } 270220bf50deSTom Herbert 270320bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 270420bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 270520bf50deSTom Herbert 2706d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 270720bf50deSTom Herbert 270820bf50deSTom Herbert while (slen) { 27090739cd28SCong Wang ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked, 27100739cd28SCong Wang sendpage_unlocked, sk, 27110739cd28SCong Wang skb_frag_page(frag), 2712b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 271320bf50deSTom Herbert slen, MSG_DONTWAIT); 271420bf50deSTom Herbert if (ret <= 0) 271520bf50deSTom Herbert goto error; 271620bf50deSTom Herbert 271720bf50deSTom Herbert len -= ret; 271820bf50deSTom Herbert offset += ret; 271920bf50deSTom Herbert slen -= ret; 272020bf50deSTom Herbert } 272120bf50deSTom Herbert 272220bf50deSTom Herbert offset = 0; 272320bf50deSTom Herbert } 272420bf50deSTom Herbert 272520bf50deSTom Herbert if (len) { 272620bf50deSTom Herbert /* Process any frag lists */ 272720bf50deSTom Herbert 272820bf50deSTom Herbert if (skb == head) { 272920bf50deSTom Herbert if (skb_has_frag_list(skb)) { 273020bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 273120bf50deSTom Herbert goto do_frag_list; 273220bf50deSTom Herbert } 273320bf50deSTom Herbert } else if (skb->next) { 273420bf50deSTom Herbert skb = skb->next; 273520bf50deSTom Herbert goto do_frag_list; 273620bf50deSTom Herbert } 273720bf50deSTom Herbert } 273820bf50deSTom Herbert 273920bf50deSTom Herbert out: 274020bf50deSTom Herbert return orig_len - len; 274120bf50deSTom Herbert 274220bf50deSTom Herbert error: 274320bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 274420bf50deSTom Herbert } 27450739cd28SCong Wang 27460739cd28SCong Wang /* Send skb data on a socket. Socket must be locked. */ 27470739cd28SCong Wang int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 27480739cd28SCong Wang int len) 27490739cd28SCong Wang { 27500739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked, 27510739cd28SCong Wang kernel_sendpage_locked); 27520739cd28SCong Wang } 275320bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 275420bf50deSTom Herbert 27550739cd28SCong Wang /* Send skb data on a socket. Socket must be unlocked. */ 27560739cd28SCong Wang int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 27570739cd28SCong Wang { 27580739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked, 27590739cd28SCong Wang sendpage_unlocked); 27600739cd28SCong Wang } 27610739cd28SCong Wang 2762357b40a1SHerbert Xu /** 2763357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2764357b40a1SHerbert Xu * @skb: destination buffer 2765357b40a1SHerbert Xu * @offset: offset in destination 2766357b40a1SHerbert Xu * @from: source buffer 2767357b40a1SHerbert Xu * @len: number of bytes to copy 2768357b40a1SHerbert Xu * 2769357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2770357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2771357b40a1SHerbert Xu * traversing fragment lists and such. 2772357b40a1SHerbert Xu */ 2773357b40a1SHerbert Xu 27740c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2775357b40a1SHerbert Xu { 27761a028e50SDavid S. Miller int start = skb_headlen(skb); 2777fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2778fbb398a8SDavid S. Miller int i, copy; 2779357b40a1SHerbert Xu 2780357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2781357b40a1SHerbert Xu goto fault; 2782357b40a1SHerbert Xu 27831a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2784357b40a1SHerbert Xu if (copy > len) 2785357b40a1SHerbert Xu copy = len; 278627d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2787357b40a1SHerbert Xu if ((len -= copy) == 0) 2788357b40a1SHerbert Xu return 0; 2789357b40a1SHerbert Xu offset += copy; 2790357b40a1SHerbert Xu from += copy; 2791357b40a1SHerbert Xu } 2792357b40a1SHerbert Xu 2793357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2794357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 27951a028e50SDavid S. Miller int end; 2796357b40a1SHerbert Xu 2797547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27981a028e50SDavid S. Miller 27999e903e08SEric Dumazet end = start + skb_frag_size(frag); 2800357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2801c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2802c613c209SWillem de Bruijn struct page *p; 2803357b40a1SHerbert Xu u8 *vaddr; 2804357b40a1SHerbert Xu 2805357b40a1SHerbert Xu if (copy > len) 2806357b40a1SHerbert Xu copy = len; 2807357b40a1SHerbert Xu 2808c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2809b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2810c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2811c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2812c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 281351c56b00SEric Dumazet kunmap_atomic(vaddr); 2814c613c209SWillem de Bruijn } 2815357b40a1SHerbert Xu 2816357b40a1SHerbert Xu if ((len -= copy) == 0) 2817357b40a1SHerbert Xu return 0; 2818357b40a1SHerbert Xu offset += copy; 2819357b40a1SHerbert Xu from += copy; 2820357b40a1SHerbert Xu } 28211a028e50SDavid S. Miller start = end; 2822357b40a1SHerbert Xu } 2823357b40a1SHerbert Xu 2824fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 28251a028e50SDavid S. Miller int end; 2826357b40a1SHerbert Xu 2827547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28281a028e50SDavid S. Miller 2829fbb398a8SDavid S. Miller end = start + frag_iter->len; 2830357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2831357b40a1SHerbert Xu if (copy > len) 2832357b40a1SHerbert Xu copy = len; 2833fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 28341a028e50SDavid S. Miller from, copy)) 2835357b40a1SHerbert Xu goto fault; 2836357b40a1SHerbert Xu if ((len -= copy) == 0) 2837357b40a1SHerbert Xu return 0; 2838357b40a1SHerbert Xu offset += copy; 2839357b40a1SHerbert Xu from += copy; 2840357b40a1SHerbert Xu } 28411a028e50SDavid S. Miller start = end; 2842357b40a1SHerbert Xu } 2843357b40a1SHerbert Xu if (!len) 2844357b40a1SHerbert Xu return 0; 2845357b40a1SHerbert Xu 2846357b40a1SHerbert Xu fault: 2847357b40a1SHerbert Xu return -EFAULT; 2848357b40a1SHerbert Xu } 2849357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2850357b40a1SHerbert Xu 28511da177e4SLinus Torvalds /* Checksum skb data. */ 28522817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 28532817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 28541da177e4SLinus Torvalds { 28551a028e50SDavid S. Miller int start = skb_headlen(skb); 28561a028e50SDavid S. Miller int i, copy = start - offset; 2857fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 28581da177e4SLinus Torvalds int pos = 0; 28591da177e4SLinus Torvalds 28601da177e4SLinus Torvalds /* Checksum header. */ 28611da177e4SLinus Torvalds if (copy > 0) { 28621da177e4SLinus Torvalds if (copy > len) 28631da177e4SLinus Torvalds copy = len; 28642544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 28652544af03SMatteo Croce skb->data + offset, copy, csum); 28661da177e4SLinus Torvalds if ((len -= copy) == 0) 28671da177e4SLinus Torvalds return csum; 28681da177e4SLinus Torvalds offset += copy; 28691da177e4SLinus Torvalds pos = copy; 28701da177e4SLinus Torvalds } 28711da177e4SLinus Torvalds 28721da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 28731a028e50SDavid S. Miller int end; 287451c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 28751da177e4SLinus Torvalds 2876547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28771a028e50SDavid S. Miller 287851c56b00SEric Dumazet end = start + skb_frag_size(frag); 28791da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2880c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2881c613c209SWillem de Bruijn struct page *p; 288244bb9363SAl Viro __wsum csum2; 28831da177e4SLinus Torvalds u8 *vaddr; 28841da177e4SLinus Torvalds 28851da177e4SLinus Torvalds if (copy > len) 28861da177e4SLinus Torvalds copy = len; 2887c613c209SWillem de Bruijn 2888c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2889b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2890c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2891c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 28922544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 28932544af03SMatteo Croce csum_partial_ext, 28942544af03SMatteo Croce vaddr + p_off, p_len, 0); 289551c56b00SEric Dumazet kunmap_atomic(vaddr); 28962544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 28972544af03SMatteo Croce csum_block_add_ext, csum, 28982544af03SMatteo Croce csum2, pos, p_len); 2899c613c209SWillem de Bruijn pos += p_len; 2900c613c209SWillem de Bruijn } 2901c613c209SWillem de Bruijn 29021da177e4SLinus Torvalds if (!(len -= copy)) 29031da177e4SLinus Torvalds return csum; 29041da177e4SLinus Torvalds offset += copy; 29051da177e4SLinus Torvalds } 29061a028e50SDavid S. Miller start = end; 29071da177e4SLinus Torvalds } 29081da177e4SLinus Torvalds 2909fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 29101a028e50SDavid S. Miller int end; 29111da177e4SLinus Torvalds 2912547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29131a028e50SDavid S. Miller 2914fbb398a8SDavid S. Miller end = start + frag_iter->len; 29151da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 29165f92a738SAl Viro __wsum csum2; 29171da177e4SLinus Torvalds if (copy > len) 29181da177e4SLinus Torvalds copy = len; 29192817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 29202817a336SDaniel Borkmann copy, 0, ops); 29212544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 29222544af03SMatteo Croce csum, csum2, pos, copy); 29231da177e4SLinus Torvalds if ((len -= copy) == 0) 29241da177e4SLinus Torvalds return csum; 29251da177e4SLinus Torvalds offset += copy; 29261da177e4SLinus Torvalds pos += copy; 29271da177e4SLinus Torvalds } 29281a028e50SDavid S. Miller start = end; 29291da177e4SLinus Torvalds } 293009a62660SKris Katterjohn BUG_ON(len); 29311da177e4SLinus Torvalds 29321da177e4SLinus Torvalds return csum; 29331da177e4SLinus Torvalds } 29342817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 29352817a336SDaniel Borkmann 29362817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 29372817a336SDaniel Borkmann int len, __wsum csum) 29382817a336SDaniel Borkmann { 29392817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2940cea80ea8SDaniel Borkmann .update = csum_partial_ext, 29412817a336SDaniel Borkmann .combine = csum_block_add_ext, 29422817a336SDaniel Borkmann }; 29432817a336SDaniel Borkmann 29442817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 29452817a336SDaniel Borkmann } 2946b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 29471da177e4SLinus Torvalds 29481da177e4SLinus Torvalds /* Both of above in one bottle. */ 29491da177e4SLinus Torvalds 295081d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 29518d5930dfSAl Viro u8 *to, int len) 29521da177e4SLinus Torvalds { 29531a028e50SDavid S. Miller int start = skb_headlen(skb); 29541a028e50SDavid S. Miller int i, copy = start - offset; 2955fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 29561da177e4SLinus Torvalds int pos = 0; 29578d5930dfSAl Viro __wsum csum = 0; 29581da177e4SLinus Torvalds 29591da177e4SLinus Torvalds /* Copy header. */ 29601da177e4SLinus Torvalds if (copy > 0) { 29611da177e4SLinus Torvalds if (copy > len) 29621da177e4SLinus Torvalds copy = len; 29631da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2964cc44c17bSAl Viro copy); 29651da177e4SLinus Torvalds if ((len -= copy) == 0) 29661da177e4SLinus Torvalds return csum; 29671da177e4SLinus Torvalds offset += copy; 29681da177e4SLinus Torvalds to += copy; 29691da177e4SLinus Torvalds pos = copy; 29701da177e4SLinus Torvalds } 29711da177e4SLinus Torvalds 29721da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 29731a028e50SDavid S. Miller int end; 29741da177e4SLinus Torvalds 2975547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29761a028e50SDavid S. Miller 29779e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 29781da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2979c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2980c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2981c613c209SWillem de Bruijn struct page *p; 29825084205fSAl Viro __wsum csum2; 29831da177e4SLinus Torvalds u8 *vaddr; 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds if (copy > len) 29861da177e4SLinus Torvalds copy = len; 2987c613c209SWillem de Bruijn 2988c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2989b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2990c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2991c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2992c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2993c613c209SWillem de Bruijn to + copied, 2994cc44c17bSAl Viro p_len); 299551c56b00SEric Dumazet kunmap_atomic(vaddr); 29961da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 2997c613c209SWillem de Bruijn pos += p_len; 2998c613c209SWillem de Bruijn } 2999c613c209SWillem de Bruijn 30001da177e4SLinus Torvalds if (!(len -= copy)) 30011da177e4SLinus Torvalds return csum; 30021da177e4SLinus Torvalds offset += copy; 30031da177e4SLinus Torvalds to += copy; 30041da177e4SLinus Torvalds } 30051a028e50SDavid S. Miller start = end; 30061da177e4SLinus Torvalds } 30071da177e4SLinus Torvalds 3008fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 300981d77662SAl Viro __wsum csum2; 30101a028e50SDavid S. Miller int end; 30111da177e4SLinus Torvalds 3012547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30131a028e50SDavid S. Miller 3014fbb398a8SDavid S. Miller end = start + frag_iter->len; 30151da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 30161da177e4SLinus Torvalds if (copy > len) 30171da177e4SLinus Torvalds copy = len; 3018fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 30191a028e50SDavid S. Miller offset - start, 30208d5930dfSAl Viro to, copy); 30211da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 30221da177e4SLinus Torvalds if ((len -= copy) == 0) 30231da177e4SLinus Torvalds return csum; 30241da177e4SLinus Torvalds offset += copy; 30251da177e4SLinus Torvalds to += copy; 30261da177e4SLinus Torvalds pos += copy; 30271da177e4SLinus Torvalds } 30281a028e50SDavid S. Miller start = end; 30291da177e4SLinus Torvalds } 303009a62660SKris Katterjohn BUG_ON(len); 30311da177e4SLinus Torvalds return csum; 30321da177e4SLinus Torvalds } 3033b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 30341da177e4SLinus Torvalds 303549f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 303649f8e832SCong Wang { 303749f8e832SCong Wang __sum16 sum; 303849f8e832SCong Wang 303949f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 304014641931SCong Wang /* See comments in __skb_checksum_complete(). */ 304149f8e832SCong Wang if (likely(!sum)) { 304249f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 304349f8e832SCong Wang !skb->csum_complete_sw) 30447fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 304549f8e832SCong Wang } 304649f8e832SCong Wang if (!skb_shared(skb)) 304749f8e832SCong Wang skb->csum_valid = !sum; 304849f8e832SCong Wang return sum; 304949f8e832SCong Wang } 305049f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 305149f8e832SCong Wang 305214641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 305314641931SCong Wang * which has been changed from the hardware checksum, for example, by 305414641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 305514641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 305614641931SCong Wang * 305714641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 305814641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 305914641931SCong Wang * shared. 306014641931SCong Wang */ 306149f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 306249f8e832SCong Wang { 306349f8e832SCong Wang __wsum csum; 306449f8e832SCong Wang __sum16 sum; 306549f8e832SCong Wang 306649f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 306749f8e832SCong Wang 306849f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 306914641931SCong Wang /* This check is inverted, because we already knew the hardware 307014641931SCong Wang * checksum is invalid before calling this function. So, if the 307114641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 307214641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 307314641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 307414641931SCong Wang * when moving skb->data around. 307514641931SCong Wang */ 307649f8e832SCong Wang if (likely(!sum)) { 307749f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 307849f8e832SCong Wang !skb->csum_complete_sw) 30797fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 308049f8e832SCong Wang } 308149f8e832SCong Wang 308249f8e832SCong Wang if (!skb_shared(skb)) { 308349f8e832SCong Wang /* Save full packet checksum */ 308449f8e832SCong Wang skb->csum = csum; 308549f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 308649f8e832SCong Wang skb->csum_complete_sw = 1; 308749f8e832SCong Wang skb->csum_valid = !sum; 308849f8e832SCong Wang } 308949f8e832SCong Wang 309049f8e832SCong Wang return sum; 309149f8e832SCong Wang } 309249f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 309349f8e832SCong Wang 30949617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 30959617813dSDavide Caratti { 30969617813dSDavide Caratti net_warn_ratelimited( 30979617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 30989617813dSDavide Caratti __func__); 30999617813dSDavide Caratti return 0; 31009617813dSDavide Caratti } 31019617813dSDavide Caratti 31029617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 31039617813dSDavide Caratti int offset, int len) 31049617813dSDavide Caratti { 31059617813dSDavide Caratti net_warn_ratelimited( 31069617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 31079617813dSDavide Caratti __func__); 31089617813dSDavide Caratti return 0; 31099617813dSDavide Caratti } 31109617813dSDavide Caratti 31119617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 31129617813dSDavide Caratti .update = warn_crc32c_csum_update, 31139617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 31149617813dSDavide Caratti }; 31159617813dSDavide Caratti 31169617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 31179617813dSDavide Caratti &default_crc32c_ops; 31189617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 31199617813dSDavide Caratti 3120af2806f8SThomas Graf /** 3121af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 3122af2806f8SThomas Graf * @from: source buffer 3123af2806f8SThomas Graf * 3124af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 3125af2806f8SThomas Graf * into skb_zerocopy(). 3126af2806f8SThomas Graf */ 3127af2806f8SThomas Graf unsigned int 3128af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 3129af2806f8SThomas Graf { 3130af2806f8SThomas Graf unsigned int hlen = 0; 3131af2806f8SThomas Graf 3132af2806f8SThomas Graf if (!from->head_frag || 3133af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 3134a17ad096SPravin B Shelar skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) { 3135af2806f8SThomas Graf hlen = skb_headlen(from); 3136a17ad096SPravin B Shelar if (!hlen) 3137a17ad096SPravin B Shelar hlen = from->len; 3138a17ad096SPravin B Shelar } 3139af2806f8SThomas Graf 3140af2806f8SThomas Graf if (skb_has_frag_list(from)) 3141af2806f8SThomas Graf hlen = from->len; 3142af2806f8SThomas Graf 3143af2806f8SThomas Graf return hlen; 3144af2806f8SThomas Graf } 3145af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 3146af2806f8SThomas Graf 3147af2806f8SThomas Graf /** 3148af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 3149af2806f8SThomas Graf * @to: destination buffer 31507fceb4deSMasanari Iida * @from: source buffer 3151af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 3152af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 3153af2806f8SThomas Graf * 3154af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 3155af2806f8SThomas Graf * to the frags in the source buffer. 3156af2806f8SThomas Graf * 3157af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 3158af2806f8SThomas Graf * headroom in the `to` buffer. 315936d5fe6aSZoltan Kiss * 316036d5fe6aSZoltan Kiss * Return value: 316136d5fe6aSZoltan Kiss * 0: everything is OK 316236d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 316336d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 3164af2806f8SThomas Graf */ 316536d5fe6aSZoltan Kiss int 316636d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 3167af2806f8SThomas Graf { 3168af2806f8SThomas Graf int i, j = 0; 3169af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 317036d5fe6aSZoltan Kiss int ret; 3171af2806f8SThomas Graf struct page *page; 3172af2806f8SThomas Graf unsigned int offset; 3173af2806f8SThomas Graf 3174af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 3175af2806f8SThomas Graf 3176af2806f8SThomas Graf /* dont bother with small payloads */ 317736d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 317836d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 3179af2806f8SThomas Graf 3180af2806f8SThomas Graf if (hlen) { 318136d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 318236d5fe6aSZoltan Kiss if (unlikely(ret)) 318336d5fe6aSZoltan Kiss return ret; 3184af2806f8SThomas Graf len -= hlen; 3185af2806f8SThomas Graf } else { 3186af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 3187af2806f8SThomas Graf if (plen) { 3188af2806f8SThomas Graf page = virt_to_head_page(from->head); 3189af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 3190af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 3191af2806f8SThomas Graf get_page(page); 3192af2806f8SThomas Graf j = 1; 3193af2806f8SThomas Graf len -= plen; 3194af2806f8SThomas Graf } 3195af2806f8SThomas Graf } 3196af2806f8SThomas Graf 3197af2806f8SThomas Graf to->truesize += len + plen; 3198af2806f8SThomas Graf to->len += len + plen; 3199af2806f8SThomas Graf to->data_len += len + plen; 3200af2806f8SThomas Graf 320136d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 320236d5fe6aSZoltan Kiss skb_tx_error(from); 320336d5fe6aSZoltan Kiss return -ENOMEM; 320436d5fe6aSZoltan Kiss } 32051f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 320636d5fe6aSZoltan Kiss 3207af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 3208d8e18a51SMatthew Wilcox (Oracle) int size; 3209d8e18a51SMatthew Wilcox (Oracle) 3210af2806f8SThomas Graf if (!len) 3211af2806f8SThomas Graf break; 3212af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 3213d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 3214d8e18a51SMatthew Wilcox (Oracle) len); 3215d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3216d8e18a51SMatthew Wilcox (Oracle) len -= size; 3217af2806f8SThomas Graf skb_frag_ref(to, j); 3218af2806f8SThomas Graf j++; 3219af2806f8SThomas Graf } 3220af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 322136d5fe6aSZoltan Kiss 322236d5fe6aSZoltan Kiss return 0; 3223af2806f8SThomas Graf } 3224af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3225af2806f8SThomas Graf 32261da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 32271da177e4SLinus Torvalds { 3228d3bc23e7SAl Viro __wsum csum; 32291da177e4SLinus Torvalds long csstart; 32301da177e4SLinus Torvalds 323184fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 323255508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 32331da177e4SLinus Torvalds else 32341da177e4SLinus Torvalds csstart = skb_headlen(skb); 32351da177e4SLinus Torvalds 323609a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 32371da177e4SLinus Torvalds 3238d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 32391da177e4SLinus Torvalds 32401da177e4SLinus Torvalds csum = 0; 32411da177e4SLinus Torvalds if (csstart != skb->len) 32421da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 32438d5930dfSAl Viro skb->len - csstart); 32441da177e4SLinus Torvalds 324584fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3246ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 32471da177e4SLinus Torvalds 3248d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 32491da177e4SLinus Torvalds } 32501da177e4SLinus Torvalds } 3251b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 32521da177e4SLinus Torvalds 32531da177e4SLinus Torvalds /** 32541da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 32551da177e4SLinus Torvalds * @list: list to dequeue from 32561da177e4SLinus Torvalds * 32571da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 32581da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 32591da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32601da177e4SLinus Torvalds */ 32611da177e4SLinus Torvalds 32621da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 32631da177e4SLinus Torvalds { 32641da177e4SLinus Torvalds unsigned long flags; 32651da177e4SLinus Torvalds struct sk_buff *result; 32661da177e4SLinus Torvalds 32671da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32681da177e4SLinus Torvalds result = __skb_dequeue(list); 32691da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32701da177e4SLinus Torvalds return result; 32711da177e4SLinus Torvalds } 3272b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 32731da177e4SLinus Torvalds 32741da177e4SLinus Torvalds /** 32751da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 32761da177e4SLinus Torvalds * @list: list to dequeue from 32771da177e4SLinus Torvalds * 32781da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 32791da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 32801da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32811da177e4SLinus Torvalds */ 32821da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 32831da177e4SLinus Torvalds { 32841da177e4SLinus Torvalds unsigned long flags; 32851da177e4SLinus Torvalds struct sk_buff *result; 32861da177e4SLinus Torvalds 32871da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 32881da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 32891da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 32901da177e4SLinus Torvalds return result; 32911da177e4SLinus Torvalds } 3292b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 32931da177e4SLinus Torvalds 32941da177e4SLinus Torvalds /** 32951da177e4SLinus Torvalds * skb_queue_purge - empty a list 32961da177e4SLinus Torvalds * @list: list to empty 32971da177e4SLinus Torvalds * 32981da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 32991da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 33001da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 33011da177e4SLinus Torvalds */ 33021da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 33031da177e4SLinus Torvalds { 33041da177e4SLinus Torvalds struct sk_buff *skb; 33051da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 33061da177e4SLinus Torvalds kfree_skb(skb); 33071da177e4SLinus Torvalds } 3308b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 33091da177e4SLinus Torvalds 33101da177e4SLinus Torvalds /** 33119f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 33129f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3313385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 33149f5afeaeSYaogong Wang * 33159f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 33169f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 33179f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 33189f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 33199f5afeaeSYaogong Wang */ 3320385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 33219f5afeaeSYaogong Wang { 33227c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3323385114deSPeter Oskolkov unsigned int sum = 0; 33249f5afeaeSYaogong Wang 33257c90584cSEric Dumazet while (p) { 33267c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 33277c90584cSEric Dumazet 33287c90584cSEric Dumazet p = rb_next(p); 33297c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3330385114deSPeter Oskolkov sum += skb->truesize; 33319f5afeaeSYaogong Wang kfree_skb(skb); 33327c90584cSEric Dumazet } 3333385114deSPeter Oskolkov return sum; 33349f5afeaeSYaogong Wang } 33359f5afeaeSYaogong Wang 33369f5afeaeSYaogong Wang /** 33371da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 33381da177e4SLinus Torvalds * @list: list to use 33391da177e4SLinus Torvalds * @newsk: buffer to queue 33401da177e4SLinus Torvalds * 33411da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 33421da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33431da177e4SLinus Torvalds * safely. 33441da177e4SLinus Torvalds * 33451da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33461da177e4SLinus Torvalds */ 33471da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 33481da177e4SLinus Torvalds { 33491da177e4SLinus Torvalds unsigned long flags; 33501da177e4SLinus Torvalds 33511da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33521da177e4SLinus Torvalds __skb_queue_head(list, newsk); 33531da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33541da177e4SLinus Torvalds } 3355b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 33561da177e4SLinus Torvalds 33571da177e4SLinus Torvalds /** 33581da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 33591da177e4SLinus Torvalds * @list: list to use 33601da177e4SLinus Torvalds * @newsk: buffer to queue 33611da177e4SLinus Torvalds * 33621da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 33631da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33641da177e4SLinus Torvalds * safely. 33651da177e4SLinus Torvalds * 33661da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33671da177e4SLinus Torvalds */ 33681da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 33691da177e4SLinus Torvalds { 33701da177e4SLinus Torvalds unsigned long flags; 33711da177e4SLinus Torvalds 33721da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33731da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 33741da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33751da177e4SLinus Torvalds } 3376b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 33778728b834SDavid S. Miller 33781da177e4SLinus Torvalds /** 33791da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 33801da177e4SLinus Torvalds * @skb: buffer to remove 33818728b834SDavid S. Miller * @list: list to use 33821da177e4SLinus Torvalds * 33838728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 33848728b834SDavid S. Miller * function is atomic with respect to other list locked calls 33851da177e4SLinus Torvalds * 33868728b834SDavid S. Miller * You must know what list the SKB is on. 33871da177e4SLinus Torvalds */ 33888728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 33891da177e4SLinus Torvalds { 33901da177e4SLinus Torvalds unsigned long flags; 33911da177e4SLinus Torvalds 33921da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33938728b834SDavid S. Miller __skb_unlink(skb, list); 33941da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33951da177e4SLinus Torvalds } 3396b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 33971da177e4SLinus Torvalds 33981da177e4SLinus Torvalds /** 33991da177e4SLinus Torvalds * skb_append - append a buffer 34001da177e4SLinus Torvalds * @old: buffer to insert after 34011da177e4SLinus Torvalds * @newsk: buffer to insert 34028728b834SDavid S. Miller * @list: list to use 34031da177e4SLinus Torvalds * 34041da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 34051da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 34061da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 34071da177e4SLinus Torvalds */ 34088728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 34091da177e4SLinus Torvalds { 34101da177e4SLinus Torvalds unsigned long flags; 34111da177e4SLinus Torvalds 34128728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 34137de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 34148728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 34151da177e4SLinus Torvalds } 3416b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 34171da177e4SLinus Torvalds 34181da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 34191da177e4SLinus Torvalds struct sk_buff* skb1, 34201da177e4SLinus Torvalds const u32 len, const int pos) 34211da177e4SLinus Torvalds { 34221da177e4SLinus Torvalds int i; 34231da177e4SLinus Torvalds 3424d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3425d626f62bSArnaldo Carvalho de Melo pos - len); 34261da177e4SLinus Torvalds /* And move data appendix as is. */ 34271da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 34281da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 34291da177e4SLinus Torvalds 34301da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 34311da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 34321da177e4SLinus Torvalds skb1->data_len = skb->data_len; 34331da177e4SLinus Torvalds skb1->len += skb1->data_len; 34341da177e4SLinus Torvalds skb->data_len = 0; 34351da177e4SLinus Torvalds skb->len = len; 343627a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 34371da177e4SLinus Torvalds } 34381da177e4SLinus Torvalds 34391da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 34401da177e4SLinus Torvalds struct sk_buff* skb1, 34411da177e4SLinus Torvalds const u32 len, int pos) 34421da177e4SLinus Torvalds { 34431da177e4SLinus Torvalds int i, k = 0; 34441da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 34451da177e4SLinus Torvalds 34461da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 34471da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 34481da177e4SLinus Torvalds skb->len = len; 34491da177e4SLinus Torvalds skb->data_len = len - pos; 34501da177e4SLinus Torvalds 34511da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 34529e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 34531da177e4SLinus Torvalds 34541da177e4SLinus Torvalds if (pos + size > len) { 34551da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 34561da177e4SLinus Torvalds 34571da177e4SLinus Torvalds if (pos < len) { 34581da177e4SLinus Torvalds /* Split frag. 34591da177e4SLinus Torvalds * We have two variants in this case: 34601da177e4SLinus Torvalds * 1. Move all the frag to the second 34611da177e4SLinus Torvalds * part, if it is possible. F.e. 34621da177e4SLinus Torvalds * this approach is mandatory for TUX, 34631da177e4SLinus Torvalds * where splitting is expensive. 34641da177e4SLinus Torvalds * 2. Split is accurately. We make this. 34651da177e4SLinus Torvalds */ 3466ea2ab693SIan Campbell skb_frag_ref(skb, i); 3467b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 34689e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 34699e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 34701da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34711da177e4SLinus Torvalds } 34721da177e4SLinus Torvalds k++; 34731da177e4SLinus Torvalds } else 34741da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 34751da177e4SLinus Torvalds pos += size; 34761da177e4SLinus Torvalds } 34771da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 34781da177e4SLinus Torvalds } 34791da177e4SLinus Torvalds 34801da177e4SLinus Torvalds /** 34811da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 34821da177e4SLinus Torvalds * @skb: the buffer to split 34831da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 34841da177e4SLinus Torvalds * @len: new length for skb 34851da177e4SLinus Torvalds */ 34861da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 34871da177e4SLinus Torvalds { 34881da177e4SLinus Torvalds int pos = skb_headlen(skb); 34899b65b17dSTalal Ahmad const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY; 34901da177e4SLinus Torvalds 34919b65b17dSTalal Ahmad skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags; 34921f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 34931da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 34941da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 34951da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 34961da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 34971da177e4SLinus Torvalds } 3498b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 34991da177e4SLinus Torvalds 35009f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 35019f782db3SIlpo Järvinen * 35029f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 35039f782db3SIlpo Järvinen */ 3504832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3505832d11c5SIlpo Järvinen { 3506c4777efaSEric Dumazet return skb_unclone_keeptruesize(skb, GFP_ATOMIC); 3507832d11c5SIlpo Järvinen } 3508832d11c5SIlpo Järvinen 3509832d11c5SIlpo Järvinen /** 3510832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3511832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3512832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3513832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3514832d11c5SIlpo Järvinen * 3515832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 351620e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3517832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3518832d11c5SIlpo Järvinen * 3519832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3520832d11c5SIlpo Järvinen * 3521832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3522832d11c5SIlpo Järvinen * to have non-paged data as well. 3523832d11c5SIlpo Järvinen * 3524832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3525832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3526832d11c5SIlpo Järvinen */ 3527832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3528832d11c5SIlpo Järvinen { 3529832d11c5SIlpo Järvinen int from, to, merge, todo; 3530d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3531832d11c5SIlpo Järvinen 3532832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3533f8071cdeSEric Dumazet 3534f8071cdeSEric Dumazet if (skb_headlen(skb)) 3535f8071cdeSEric Dumazet return 0; 35361f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 35371f8b977aSWillem de Bruijn return 0; 3538832d11c5SIlpo Järvinen 3539832d11c5SIlpo Järvinen todo = shiftlen; 3540832d11c5SIlpo Järvinen from = 0; 3541832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3542832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3543832d11c5SIlpo Järvinen 3544832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3545832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3546832d11c5SIlpo Järvinen */ 3547832d11c5SIlpo Järvinen if (!to || 3548ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3549b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3550832d11c5SIlpo Järvinen merge = -1; 3551832d11c5SIlpo Järvinen } else { 3552832d11c5SIlpo Järvinen merge = to - 1; 3553832d11c5SIlpo Järvinen 35549e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3555832d11c5SIlpo Järvinen if (todo < 0) { 3556832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3557832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3558832d11c5SIlpo Järvinen return 0; 3559832d11c5SIlpo Järvinen 35609f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 35619f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3562832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3563832d11c5SIlpo Järvinen 35649e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 35659e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3566b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3567832d11c5SIlpo Järvinen 3568832d11c5SIlpo Järvinen goto onlymerged; 3569832d11c5SIlpo Järvinen } 3570832d11c5SIlpo Järvinen 3571832d11c5SIlpo Järvinen from++; 3572832d11c5SIlpo Järvinen } 3573832d11c5SIlpo Järvinen 3574832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3575832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3576832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3577832d11c5SIlpo Järvinen return 0; 3578832d11c5SIlpo Järvinen 3579832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3580832d11c5SIlpo Järvinen return 0; 3581832d11c5SIlpo Järvinen 3582832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3583832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3584832d11c5SIlpo Järvinen return 0; 3585832d11c5SIlpo Järvinen 3586832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3587832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3588832d11c5SIlpo Järvinen 35899e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3590832d11c5SIlpo Järvinen *fragto = *fragfrom; 35919e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3592832d11c5SIlpo Järvinen from++; 3593832d11c5SIlpo Järvinen to++; 3594832d11c5SIlpo Järvinen 3595832d11c5SIlpo Järvinen } else { 3596ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3597b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3598b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 35999e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3600832d11c5SIlpo Järvinen 3601b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 36029e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3603832d11c5SIlpo Järvinen todo = 0; 3604832d11c5SIlpo Järvinen 3605832d11c5SIlpo Järvinen to++; 3606832d11c5SIlpo Järvinen break; 3607832d11c5SIlpo Järvinen } 3608832d11c5SIlpo Järvinen } 3609832d11c5SIlpo Järvinen 3610832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3611832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3612832d11c5SIlpo Järvinen 3613832d11c5SIlpo Järvinen if (merge >= 0) { 3614832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3615832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3616832d11c5SIlpo Järvinen 36179e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 36186a5bcd84SIlias Apalodimas __skb_frag_unref(fragfrom, skb->pp_recycle); 3619832d11c5SIlpo Järvinen } 3620832d11c5SIlpo Järvinen 3621832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3622832d11c5SIlpo Järvinen to = 0; 3623832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3624832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3625832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3626832d11c5SIlpo Järvinen 3627832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3628832d11c5SIlpo Järvinen 3629832d11c5SIlpo Järvinen onlymerged: 3630832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3631832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3632832d11c5SIlpo Järvinen */ 3633832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3634832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3635832d11c5SIlpo Järvinen 3636832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 3637832d11c5SIlpo Järvinen skb->len -= shiftlen; 3638832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 3639832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 3640832d11c5SIlpo Järvinen tgt->len += shiftlen; 3641832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 3642832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 3643832d11c5SIlpo Järvinen 3644832d11c5SIlpo Järvinen return shiftlen; 3645832d11c5SIlpo Järvinen } 3646832d11c5SIlpo Järvinen 3647677e90edSThomas Graf /** 3648677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3649677e90edSThomas Graf * @skb: the buffer to read 3650677e90edSThomas Graf * @from: lower offset of data to be read 3651677e90edSThomas Graf * @to: upper offset of data to be read 3652677e90edSThomas Graf * @st: state variable 3653677e90edSThomas Graf * 3654677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3655677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3656677e90edSThomas Graf */ 3657677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3658677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3659677e90edSThomas Graf { 3660677e90edSThomas Graf st->lower_offset = from; 3661677e90edSThomas Graf st->upper_offset = to; 3662677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3663677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3664677e90edSThomas Graf st->frag_data = NULL; 366597550f6fSWillem de Bruijn st->frag_off = 0; 3666677e90edSThomas Graf } 3667b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3668677e90edSThomas Graf 3669677e90edSThomas Graf /** 3670677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3671677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3672677e90edSThomas Graf * @data: destination pointer for data to be returned 3673677e90edSThomas Graf * @st: state variable 3674677e90edSThomas Graf * 3675bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3676677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3677bc32383cSMathias Krause * the head of the data block to @data and returns the length 3678677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3679677e90edSThomas Graf * offset has been reached. 3680677e90edSThomas Graf * 3681677e90edSThomas Graf * The caller is not required to consume all of the data 3682bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3683677e90edSThomas Graf * of bytes already consumed and the next call to 3684677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3685677e90edSThomas Graf * 368625985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3687e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3688677e90edSThomas Graf * reads of potentially non linear data. 3689677e90edSThomas Graf * 3690bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3691677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3692677e90edSThomas Graf * a stack for this purpose. 3693677e90edSThomas Graf */ 3694677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3695677e90edSThomas Graf struct skb_seq_state *st) 3696677e90edSThomas Graf { 3697677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3698677e90edSThomas Graf skb_frag_t *frag; 3699677e90edSThomas Graf 3700aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3701aeb193eaSWedson Almeida Filho if (st->frag_data) { 3702aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3703aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3704aeb193eaSWedson Almeida Filho } 3705677e90edSThomas Graf return 0; 3706aeb193eaSWedson Almeida Filho } 3707677e90edSThomas Graf 3708677e90edSThomas Graf next_skb: 370995e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3710677e90edSThomas Graf 3711995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 371295e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3713677e90edSThomas Graf return block_limit - abs_offset; 3714677e90edSThomas Graf } 3715677e90edSThomas Graf 3716677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3717677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3718677e90edSThomas Graf 3719677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 372097550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3721677e90edSThomas Graf 372297550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 372397550f6fSWillem de Bruijn 372497550f6fSWillem de Bruijn pg_idx = 0; 372597550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 372697550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 372797550f6fSWillem de Bruijn 372897550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 372997550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 373097550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 373197550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 373297550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 373397550f6fSWillem de Bruijn } 373497550f6fSWillem de Bruijn 373597550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3736677e90edSThomas Graf if (abs_offset < block_limit) { 3737677e90edSThomas Graf if (!st->frag_data) 373897550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3739677e90edSThomas Graf 374097550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3741677e90edSThomas Graf (abs_offset - st->stepped_offset); 3742677e90edSThomas Graf 3743677e90edSThomas Graf return block_limit - abs_offset; 3744677e90edSThomas Graf } 3745677e90edSThomas Graf 3746677e90edSThomas Graf if (st->frag_data) { 374751c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3748677e90edSThomas Graf st->frag_data = NULL; 3749677e90edSThomas Graf } 3750677e90edSThomas Graf 375197550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 375297550f6fSWillem de Bruijn st->frag_off += pg_sz; 375397550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 375497550f6fSWillem de Bruijn st->frag_off = 0; 3755677e90edSThomas Graf st->frag_idx++; 375697550f6fSWillem de Bruijn } 3757677e90edSThomas Graf } 3758677e90edSThomas Graf 37595b5a60daSOlaf Kirch if (st->frag_data) { 376051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 37615b5a60daSOlaf Kirch st->frag_data = NULL; 37625b5a60daSOlaf Kirch } 37635b5a60daSOlaf Kirch 376421dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3765677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 376695e3b24cSHerbert Xu st->frag_idx = 0; 3767677e90edSThomas Graf goto next_skb; 376871b3346dSShyam Iyer } else if (st->cur_skb->next) { 376971b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 377071b3346dSShyam Iyer st->frag_idx = 0; 3771677e90edSThomas Graf goto next_skb; 3772677e90edSThomas Graf } 3773677e90edSThomas Graf 3774677e90edSThomas Graf return 0; 3775677e90edSThomas Graf } 3776b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3777677e90edSThomas Graf 3778677e90edSThomas Graf /** 3779677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3780677e90edSThomas Graf * @st: state variable 3781677e90edSThomas Graf * 3782677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3783677e90edSThomas Graf * returned 0. 3784677e90edSThomas Graf */ 3785677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3786677e90edSThomas Graf { 3787677e90edSThomas Graf if (st->frag_data) 378851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3789677e90edSThomas Graf } 3790b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3791677e90edSThomas Graf 37923fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 37933fc7e8a6SThomas Graf 37943fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 37953fc7e8a6SThomas Graf struct ts_config *conf, 37963fc7e8a6SThomas Graf struct ts_state *state) 37973fc7e8a6SThomas Graf { 37983fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 37993fc7e8a6SThomas Graf } 38003fc7e8a6SThomas Graf 38013fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 38023fc7e8a6SThomas Graf { 38033fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 38043fc7e8a6SThomas Graf } 38053fc7e8a6SThomas Graf 38063fc7e8a6SThomas Graf /** 38073fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 38083fc7e8a6SThomas Graf * @skb: the buffer to look in 38093fc7e8a6SThomas Graf * @from: search offset 38103fc7e8a6SThomas Graf * @to: search limit 38113fc7e8a6SThomas Graf * @config: textsearch configuration 38123fc7e8a6SThomas Graf * 38133fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 38143fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 38153fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 38163fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 38173fc7e8a6SThomas Graf */ 38183fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3819059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 38203fc7e8a6SThomas Graf { 3821059a2440SBojan Prtvar struct ts_state state; 3822f72b948dSPhil Oester unsigned int ret; 3823f72b948dSPhil Oester 3824b228c9b0SWillem de Bruijn BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb)); 3825b228c9b0SWillem de Bruijn 38263fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 38273fc7e8a6SThomas Graf config->finish = skb_ts_finish; 38283fc7e8a6SThomas Graf 3829059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 38303fc7e8a6SThomas Graf 3831059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3832f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 38333fc7e8a6SThomas Graf } 3834b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 38353fc7e8a6SThomas Graf 3836be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3837be12a1feSHannes Frederic Sowa int offset, size_t size) 3838be12a1feSHannes Frederic Sowa { 3839be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3840be12a1feSHannes Frederic Sowa 3841be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3842be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3843be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3844be12a1feSHannes Frederic Sowa get_page(page); 3845be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3846be12a1feSHannes Frederic Sowa } else { 3847be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3848be12a1feSHannes Frederic Sowa } 3849be12a1feSHannes Frederic Sowa 3850be12a1feSHannes Frederic Sowa return 0; 3851be12a1feSHannes Frederic Sowa } 3852be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3853be12a1feSHannes Frederic Sowa 3854cbb042f9SHerbert Xu /** 3855cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3856cbb042f9SHerbert Xu * @skb: buffer to update 3857cbb042f9SHerbert Xu * @len: length of data pulled 3858cbb042f9SHerbert Xu * 3859cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3860fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 386184fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 386284fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 386384fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3864cbb042f9SHerbert Xu */ 3865af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3866cbb042f9SHerbert Xu { 386731b33dfbSPravin B Shelar unsigned char *data = skb->data; 386831b33dfbSPravin B Shelar 3869cbb042f9SHerbert Xu BUG_ON(len > skb->len); 387031b33dfbSPravin B Shelar __skb_pull(skb, len); 387131b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 387231b33dfbSPravin B Shelar return skb->data; 3873cbb042f9SHerbert Xu } 3874f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3875f94691acSArnaldo Carvalho de Melo 387613acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 387713acc94eSYonghong Song { 387813acc94eSYonghong Song skb_frag_t head_frag; 387913acc94eSYonghong Song struct page *page; 388013acc94eSYonghong Song 388113acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3882d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3883b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3884b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3885d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 388613acc94eSYonghong Song return head_frag; 388713acc94eSYonghong Song } 388813acc94eSYonghong Song 38893a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 38903a1296a3SSteffen Klassert netdev_features_t features, 38913a1296a3SSteffen Klassert unsigned int offset) 38923a1296a3SSteffen Klassert { 38933a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 38943a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 38953a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 38963a1296a3SSteffen Klassert unsigned int delta_len = 0; 38973a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 389853475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 389953475c5dSDongseok Yi int err; 39003a1296a3SSteffen Klassert 39013a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 39023a1296a3SSteffen Klassert 39033a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 39043a1296a3SSteffen Klassert 39053a1296a3SSteffen Klassert do { 39063a1296a3SSteffen Klassert nskb = list_skb; 39073a1296a3SSteffen Klassert list_skb = list_skb->next; 39083a1296a3SSteffen Klassert 390953475c5dSDongseok Yi err = 0; 391053475c5dSDongseok Yi if (skb_shared(nskb)) { 391153475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 391253475c5dSDongseok Yi if (tmp) { 391353475c5dSDongseok Yi consume_skb(nskb); 391453475c5dSDongseok Yi nskb = tmp; 391553475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 391653475c5dSDongseok Yi } else { 391753475c5dSDongseok Yi err = -ENOMEM; 391853475c5dSDongseok Yi } 391953475c5dSDongseok Yi } 392053475c5dSDongseok Yi 39213a1296a3SSteffen Klassert if (!tail) 39223a1296a3SSteffen Klassert skb->next = nskb; 39233a1296a3SSteffen Klassert else 39243a1296a3SSteffen Klassert tail->next = nskb; 39253a1296a3SSteffen Klassert 392653475c5dSDongseok Yi if (unlikely(err)) { 392753475c5dSDongseok Yi nskb->next = list_skb; 392853475c5dSDongseok Yi goto err_linearize; 392953475c5dSDongseok Yi } 393053475c5dSDongseok Yi 39313a1296a3SSteffen Klassert tail = nskb; 39323a1296a3SSteffen Klassert 39333a1296a3SSteffen Klassert delta_len += nskb->len; 39343a1296a3SSteffen Klassert delta_truesize += nskb->truesize; 39353a1296a3SSteffen Klassert 39363a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 39373a1296a3SSteffen Klassert 3938cf673ed0SFlorian Westphal skb_release_head_state(nskb); 39393a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 39403a1296a3SSteffen Klassert 39413a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 39423a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 39433a1296a3SSteffen Klassert nskb->data - tnl_hlen, 39443a1296a3SSteffen Klassert offset + tnl_hlen); 39453a1296a3SSteffen Klassert 39463a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 39473a1296a3SSteffen Klassert __skb_linearize(nskb)) 39483a1296a3SSteffen Klassert goto err_linearize; 39493a1296a3SSteffen Klassert 39503a1296a3SSteffen Klassert } while (list_skb); 39513a1296a3SSteffen Klassert 39523a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 39533a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 39543a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 39553a1296a3SSteffen Klassert 39563a1296a3SSteffen Klassert skb_gso_reset(skb); 39573a1296a3SSteffen Klassert 39583a1296a3SSteffen Klassert skb->prev = tail; 39593a1296a3SSteffen Klassert 39603a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 39613a1296a3SSteffen Klassert __skb_linearize(skb)) 39623a1296a3SSteffen Klassert goto err_linearize; 39633a1296a3SSteffen Klassert 39643a1296a3SSteffen Klassert skb_get(skb); 39653a1296a3SSteffen Klassert 39663a1296a3SSteffen Klassert return skb; 39673a1296a3SSteffen Klassert 39683a1296a3SSteffen Klassert err_linearize: 39693a1296a3SSteffen Klassert kfree_skb_list(skb->next); 39703a1296a3SSteffen Klassert skb->next = NULL; 39713a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 39723a1296a3SSteffen Klassert } 39733a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 39743a1296a3SSteffen Klassert 3975f4c50d99SHerbert Xu /** 3976f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 3977df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 3978576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 3979f4c50d99SHerbert Xu * 3980f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 39814c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 39824c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 3983f4c50d99SHerbert Xu */ 3984df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 3985df5771ffSMichael S. Tsirkin netdev_features_t features) 3986f4c50d99SHerbert Xu { 3987f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 3988f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 39891a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3990df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3991df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 3992df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 39931fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 3994f4c50d99SHerbert Xu unsigned int offset = doffset; 3995df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3996802ab55aSAlexander Duyck unsigned int partial_segs = 0; 3997f4c50d99SHerbert Xu unsigned int headroom; 3998802ab55aSAlexander Duyck unsigned int len = head_skb->len; 3999ec5f0615SPravin B Shelar __be16 proto; 400036c98382SAlexander Duyck bool csum, sg; 4001df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 4002f4c50d99SHerbert Xu int err = -ENOMEM; 4003f4c50d99SHerbert Xu int i = 0; 4004f4c50d99SHerbert Xu int pos; 4005f4c50d99SHerbert Xu 40063dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 40073dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 40083dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 40093dcbdb13SShmulik Ladkani * non head_frag head. 40103dcbdb13SShmulik Ladkani * 40113dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 40123dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 40133dcbdb13SShmulik Ladkani * head, then the first one has too). 40143dcbdb13SShmulik Ladkani * 40153dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 40163dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 40173dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 40183dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 40193dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 40203dcbdb13SShmulik Ladkani */ 40213dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 40223dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 40233dcbdb13SShmulik Ladkani } 40243dcbdb13SShmulik Ladkani 40255882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 40262f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 4027ec5f0615SPravin B Shelar if (unlikely(!proto)) 4028ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 4029ec5f0615SPravin B Shelar 403036c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 4031f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 40327e2b10c1STom Herbert 403307b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 403407b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 403507b26c94SSteffen Klassert struct sk_buff *iter; 403643170c4eSIlan Tayari unsigned int frag_len; 403707b26c94SSteffen Klassert 403807b26c94SSteffen Klassert if (!list_skb || 403907b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 404007b26c94SSteffen Klassert goto normal; 404107b26c94SSteffen Klassert 404243170c4eSIlan Tayari /* If we get here then all the required 404343170c4eSIlan Tayari * GSO features except frag_list are supported. 404443170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 404543170c4eSIlan Tayari * with no frag_list. 404643170c4eSIlan Tayari * Currently we can do that only when the buffers don't 404743170c4eSIlan Tayari * have a linear part and all the buffers except 404843170c4eSIlan Tayari * the last are of the same length. 404907b26c94SSteffen Klassert */ 405043170c4eSIlan Tayari frag_len = list_skb->len; 405107b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 405243170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 405343170c4eSIlan Tayari goto normal; 4054eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 405507b26c94SSteffen Klassert goto normal; 405607b26c94SSteffen Klassert 405707b26c94SSteffen Klassert len -= iter->len; 405807b26c94SSteffen Klassert } 405943170c4eSIlan Tayari 406043170c4eSIlan Tayari if (len != frag_len) 406143170c4eSIlan Tayari goto normal; 406207b26c94SSteffen Klassert } 406307b26c94SSteffen Klassert 4064802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 4065802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 4066802ab55aSAlexander Duyck * now. 4067802ab55aSAlexander Duyck */ 4068802ab55aSAlexander Duyck partial_segs = len / mss; 4069d7fb5a80SAlexander Duyck if (partial_segs > 1) 4070802ab55aSAlexander Duyck mss *= partial_segs; 4071d7fb5a80SAlexander Duyck else 4072d7fb5a80SAlexander Duyck partial_segs = 0; 4073802ab55aSAlexander Duyck } 4074802ab55aSAlexander Duyck 407507b26c94SSteffen Klassert normal: 4076df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 4077df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 4078f4c50d99SHerbert Xu 4079f4c50d99SHerbert Xu do { 4080f4c50d99SHerbert Xu struct sk_buff *nskb; 40818cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 4082c8884eddSHerbert Xu int hsize; 4083f4c50d99SHerbert Xu int size; 4084f4c50d99SHerbert Xu 40853953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 40863953c46cSMarcelo Ricardo Leitner len = list_skb->len; 40873953c46cSMarcelo Ricardo Leitner } else { 4088df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 4089f4c50d99SHerbert Xu if (len > mss) 4090f4c50d99SHerbert Xu len = mss; 40913953c46cSMarcelo Ricardo Leitner } 4092f4c50d99SHerbert Xu 4093df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 4094f4c50d99SHerbert Xu 4095dbd50f23SXin Long if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) && 40961a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 40971a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 409889319d38SHerbert Xu 40999d8506ccSHerbert Xu i = 0; 41001a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 41011a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 41021fd819ecSMichael S. Tsirkin frag_skb = list_skb; 41031a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 41049d8506ccSHerbert Xu 41059d8506ccSHerbert Xu while (pos < offset + len) { 41069d8506ccSHerbert Xu BUG_ON(i >= nfrags); 41079d8506ccSHerbert Xu 41084e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 41099d8506ccSHerbert Xu if (pos + size > offset + len) 41109d8506ccSHerbert Xu break; 41119d8506ccSHerbert Xu 41129d8506ccSHerbert Xu i++; 41139d8506ccSHerbert Xu pos += size; 41144e1beba1SMichael S. Tsirkin frag++; 41159d8506ccSHerbert Xu } 41169d8506ccSHerbert Xu 41171a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 41181a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 411989319d38SHerbert Xu 4120f4c50d99SHerbert Xu if (unlikely(!nskb)) 4121f4c50d99SHerbert Xu goto err; 4122f4c50d99SHerbert Xu 41239d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 41249d8506ccSHerbert Xu kfree_skb(nskb); 41259d8506ccSHerbert Xu goto err; 41269d8506ccSHerbert Xu } 41279d8506ccSHerbert Xu 4128ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 412989319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 413089319d38SHerbert Xu kfree_skb(nskb); 413189319d38SHerbert Xu goto err; 413289319d38SHerbert Xu } 413389319d38SHerbert Xu 4134ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 413589319d38SHerbert Xu skb_release_head_state(nskb); 413689319d38SHerbert Xu __skb_push(nskb, doffset); 413789319d38SHerbert Xu } else { 413800b229f7SPaolo Abeni if (hsize < 0) 413900b229f7SPaolo Abeni hsize = 0; 4140dbd50f23SXin Long if (hsize > len || !sg) 4141dbd50f23SXin Long hsize = len; 4142dbd50f23SXin Long 4143c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 4144df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 4145c93bdd0eSMel Gorman NUMA_NO_NODE); 414689319d38SHerbert Xu 414789319d38SHerbert Xu if (unlikely(!nskb)) 414889319d38SHerbert Xu goto err; 414989319d38SHerbert Xu 415089319d38SHerbert Xu skb_reserve(nskb, headroom); 415189319d38SHerbert Xu __skb_put(nskb, doffset); 415289319d38SHerbert Xu } 415389319d38SHerbert Xu 4154f4c50d99SHerbert Xu if (segs) 4155f4c50d99SHerbert Xu tail->next = nskb; 4156f4c50d99SHerbert Xu else 4157f4c50d99SHerbert Xu segs = nskb; 4158f4c50d99SHerbert Xu tail = nskb; 4159f4c50d99SHerbert Xu 4160df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 4161f4c50d99SHerbert Xu 4162030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 4163fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 416468c33163SPravin B Shelar 4165df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 416668c33163SPravin B Shelar nskb->data - tnl_hlen, 416768c33163SPravin B Shelar doffset + tnl_hlen); 416889319d38SHerbert Xu 41699d8506ccSHerbert Xu if (nskb->len == len + doffset) 41701cdbcb79SSimon Horman goto perform_csum_check; 417189319d38SHerbert Xu 41727fbeffedSAlexander Duyck if (!sg) { 41731454c9faSYadu Kishore if (!csum) { 41747fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 41756f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 417676443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 417776443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 41781454c9faSYadu Kishore skb_put(nskb, 41791454c9faSYadu Kishore len), 41808d5930dfSAl Viro len); 41817e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 4182de843723STom Herbert skb_headroom(nskb) + doffset; 41831454c9faSYadu Kishore } else { 41841454c9faSYadu Kishore skb_copy_bits(head_skb, offset, 41851454c9faSYadu Kishore skb_put(nskb, len), 41861454c9faSYadu Kishore len); 41871454c9faSYadu Kishore } 4188f4c50d99SHerbert Xu continue; 4189f4c50d99SHerbert Xu } 4190f4c50d99SHerbert Xu 41918cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 4192f4c50d99SHerbert Xu 4193df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 4194d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4195f4c50d99SHerbert Xu 419606b4feb3SJonathan Lemon skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags & 419706b4feb3SJonathan Lemon SKBFL_SHARED_FRAG; 4198cef401deSEric Dumazet 4199bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4200bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4201bf5c25d6SWillem de Bruijn goto err; 4202bf5c25d6SWillem de Bruijn 42039d8506ccSHerbert Xu while (pos < offset + len) { 42049d8506ccSHerbert Xu if (i >= nfrags) { 42059d8506ccSHerbert Xu i = 0; 42061a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 42071a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 42081fd819ecSMichael S. Tsirkin frag_skb = list_skb; 420913acc94eSYonghong Song if (!skb_headlen(list_skb)) { 42109d8506ccSHerbert Xu BUG_ON(!nfrags); 421113acc94eSYonghong Song } else { 421213acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 42139d8506ccSHerbert Xu 421413acc94eSYonghong Song /* to make room for head_frag. */ 421513acc94eSYonghong Song i--; 421613acc94eSYonghong Song frag--; 421713acc94eSYonghong Song } 4218bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4219bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4220bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4221bf5c25d6SWillem de Bruijn goto err; 4222bf5c25d6SWillem de Bruijn 42231a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 42249d8506ccSHerbert Xu } 42259d8506ccSHerbert Xu 42269d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 42279d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 42289d8506ccSHerbert Xu net_warn_ratelimited( 42299d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 42309d8506ccSHerbert Xu pos, mss); 4231ff907a11SEric Dumazet err = -EINVAL; 42329d8506ccSHerbert Xu goto err; 42339d8506ccSHerbert Xu } 42349d8506ccSHerbert Xu 423513acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 42368cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 42378cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4238f4c50d99SHerbert Xu 4239f4c50d99SHerbert Xu if (pos < offset) { 4240b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 42418cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4242f4c50d99SHerbert Xu } 4243f4c50d99SHerbert Xu 424489319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4245f4c50d99SHerbert Xu 4246f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4247f4c50d99SHerbert Xu i++; 42484e1beba1SMichael S. Tsirkin frag++; 4249f4c50d99SHerbert Xu pos += size; 4250f4c50d99SHerbert Xu } else { 42518cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 425289319d38SHerbert Xu goto skip_fraglist; 4253f4c50d99SHerbert Xu } 4254f4c50d99SHerbert Xu 42558cb19905SMichael S. Tsirkin nskb_frag++; 4256f4c50d99SHerbert Xu } 4257f4c50d99SHerbert Xu 425889319d38SHerbert Xu skip_fraglist: 4259f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4260f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4261f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4262ec5f0615SPravin B Shelar 42631cdbcb79SSimon Horman perform_csum_check: 42647fbeffedSAlexander Duyck if (!csum) { 4265ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4266ff907a11SEric Dumazet __skb_linearize(nskb)) 4267ddff00d4SAlexander Duyck goto err; 4268ff907a11SEric Dumazet 42697fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4270ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 427176443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 427276443456SAlexander Duyck skb_checksum(nskb, doffset, 427376443456SAlexander Duyck nskb->len - doffset, 0); 42747e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 42757e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4276ec5f0615SPravin B Shelar } 4277df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4278f4c50d99SHerbert Xu 4279bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4280bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4281bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4282bec3cfdcSEric Dumazet */ 4283bec3cfdcSEric Dumazet segs->prev = tail; 4284432c856fSToshiaki Makita 4285802ab55aSAlexander Duyck if (partial_segs) { 428607b26c94SSteffen Klassert struct sk_buff *iter; 4287802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 428807b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4289802ab55aSAlexander Duyck 4290802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 429107b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4292802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4293802ab55aSAlexander Duyck 4294802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4295802ab55aSAlexander Duyck * our way back down the stack of protocols. 4296802ab55aSAlexander Duyck */ 429707b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 429807b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 429907b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 430007b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 430107b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 430207b26c94SSteffen Klassert } 430307b26c94SSteffen Klassert 430407b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 430507b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 430607b26c94SSteffen Klassert else if (tail != segs) 430707b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4308802ab55aSAlexander Duyck } 4309802ab55aSAlexander Duyck 4310432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4311432c856fSToshiaki Makita * using skb_set_owner_w(). 4312432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4313432c856fSToshiaki Makita */ 4314432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4315432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4316432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4317432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4318432c856fSToshiaki Makita } 4319f4c50d99SHerbert Xu return segs; 4320f4c50d99SHerbert Xu 4321f4c50d99SHerbert Xu err: 4322289dccbeSEric Dumazet kfree_skb_list(segs); 4323f4c50d99SHerbert Xu return ERR_PTR(err); 4324f4c50d99SHerbert Xu } 4325f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4326f4c50d99SHerbert Xu 4327df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4328df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4329df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4330df5042f4SFlorian Westphal 4331df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4332df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4333df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4334df5042f4SFlorian Westphal #endif 43354165079bSFlorian Westphal #ifdef CONFIG_XFRM 43364165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 43374165079bSFlorian Westphal #endif 433895a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 433995a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 434095a7233cSPaul Blakey #endif 43413ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 43423ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 43433ee17bc7SMat Martineau #endif 434478476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 434578476d31SJeremy Kerr [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow), 434678476d31SJeremy Kerr #endif 4347df5042f4SFlorian Westphal }; 4348df5042f4SFlorian Westphal 4349df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4350df5042f4SFlorian Westphal { 4351df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4352df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4353df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4354df5042f4SFlorian Westphal #endif 43554165079bSFlorian Westphal #ifdef CONFIG_XFRM 43564165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 43574165079bSFlorian Westphal #endif 435895a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 435995a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 436095a7233cSPaul Blakey #endif 43613ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 43623ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 43633ee17bc7SMat Martineau #endif 436478476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 436578476d31SJeremy Kerr skb_ext_type_len[SKB_EXT_MCTP] + 436678476d31SJeremy Kerr #endif 4367df5042f4SFlorian Westphal 0; 4368df5042f4SFlorian Westphal } 4369df5042f4SFlorian Westphal 4370df5042f4SFlorian Westphal static void skb_extensions_init(void) 4371df5042f4SFlorian Westphal { 4372df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4373df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4374df5042f4SFlorian Westphal 4375df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4376df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4377df5042f4SFlorian Westphal 0, 4378df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4379df5042f4SFlorian Westphal NULL); 4380df5042f4SFlorian Westphal } 4381df5042f4SFlorian Westphal #else 4382df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4383df5042f4SFlorian Westphal #endif 4384df5042f4SFlorian Westphal 43851da177e4SLinus Torvalds void __init skb_init(void) 43861da177e4SLinus Torvalds { 438779a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 43881da177e4SLinus Torvalds sizeof(struct sk_buff), 43891da177e4SLinus Torvalds 0, 4390e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 439179a8a642SKees Cook offsetof(struct sk_buff, cb), 439279a8a642SKees Cook sizeof_field(struct sk_buff, cb), 439320c2df83SPaul Mundt NULL); 4394d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4395d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4396d179cd12SDavid S. Miller 0, 4397e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 439820c2df83SPaul Mundt NULL); 4399df5042f4SFlorian Westphal skb_extensions_init(); 44001da177e4SLinus Torvalds } 44011da177e4SLinus Torvalds 440251c739d1SDavid S. Miller static int 440348a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 440448a1df65SJason A. Donenfeld unsigned int recursion_level) 4405716ea3a7SDavid Howells { 44061a028e50SDavid S. Miller int start = skb_headlen(skb); 44071a028e50SDavid S. Miller int i, copy = start - offset; 4408fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4409716ea3a7SDavid Howells int elt = 0; 4410716ea3a7SDavid Howells 441148a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 441248a1df65SJason A. Donenfeld return -EMSGSIZE; 441348a1df65SJason A. Donenfeld 4414716ea3a7SDavid Howells if (copy > 0) { 4415716ea3a7SDavid Howells if (copy > len) 4416716ea3a7SDavid Howells copy = len; 4417642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4418716ea3a7SDavid Howells elt++; 4419716ea3a7SDavid Howells if ((len -= copy) == 0) 4420716ea3a7SDavid Howells return elt; 4421716ea3a7SDavid Howells offset += copy; 4422716ea3a7SDavid Howells } 4423716ea3a7SDavid Howells 4424716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 44251a028e50SDavid S. Miller int end; 4426716ea3a7SDavid Howells 4427547b792cSIlpo Järvinen WARN_ON(start > offset + len); 44281a028e50SDavid S. Miller 44299e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4430716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4431716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 443248a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 443348a1df65SJason A. Donenfeld return -EMSGSIZE; 4434716ea3a7SDavid Howells 4435716ea3a7SDavid Howells if (copy > len) 4436716ea3a7SDavid Howells copy = len; 4437ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4438b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4439716ea3a7SDavid Howells elt++; 4440716ea3a7SDavid Howells if (!(len -= copy)) 4441716ea3a7SDavid Howells return elt; 4442716ea3a7SDavid Howells offset += copy; 4443716ea3a7SDavid Howells } 44441a028e50SDavid S. Miller start = end; 4445716ea3a7SDavid Howells } 4446716ea3a7SDavid Howells 4447fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 444848a1df65SJason A. Donenfeld int end, ret; 4449716ea3a7SDavid Howells 4450547b792cSIlpo Järvinen WARN_ON(start > offset + len); 44511a028e50SDavid S. Miller 4452fbb398a8SDavid S. Miller end = start + frag_iter->len; 4453716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 445448a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 445548a1df65SJason A. Donenfeld return -EMSGSIZE; 445648a1df65SJason A. Donenfeld 4457716ea3a7SDavid Howells if (copy > len) 4458716ea3a7SDavid Howells copy = len; 445948a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 446048a1df65SJason A. Donenfeld copy, recursion_level + 1); 446148a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 446248a1df65SJason A. Donenfeld return ret; 446348a1df65SJason A. Donenfeld elt += ret; 4464716ea3a7SDavid Howells if ((len -= copy) == 0) 4465716ea3a7SDavid Howells return elt; 4466716ea3a7SDavid Howells offset += copy; 4467716ea3a7SDavid Howells } 44681a028e50SDavid S. Miller start = end; 4469716ea3a7SDavid Howells } 4470716ea3a7SDavid Howells BUG_ON(len); 4471716ea3a7SDavid Howells return elt; 4472716ea3a7SDavid Howells } 4473716ea3a7SDavid Howells 447448a1df65SJason A. Donenfeld /** 447548a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 447648a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 447748a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 447848a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 447948a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 448048a1df65SJason A. Donenfeld * 448148a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 448248a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 448348a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 448448a1df65SJason A. Donenfeld * could not fit. 448548a1df65SJason A. Donenfeld */ 448648a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 448748a1df65SJason A. Donenfeld { 448848a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 448948a1df65SJason A. Donenfeld 449048a1df65SJason A. Donenfeld if (nsg <= 0) 449148a1df65SJason A. Donenfeld return nsg; 449248a1df65SJason A. Donenfeld 449348a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 449448a1df65SJason A. Donenfeld 449548a1df65SJason A. Donenfeld return nsg; 449648a1df65SJason A. Donenfeld } 449748a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 449848a1df65SJason A. Donenfeld 449925a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 450025a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 450125a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 450225a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 450325a91d8dSFan Du * 450425a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 450525a91d8dSFan Du * 1. sg_init_table 450625a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 450725a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 450825a91d8dSFan Du * 450925a91d8dSFan Du * This is equivalent to: 451025a91d8dSFan Du * 1. sg_init_table 451125a91d8dSFan Du * 2. skb_to_sgvec(payload1) 451225a91d8dSFan Du * 3. sg_unmark_end 451325a91d8dSFan Du * 4. skb_to_sgvec(payload2) 451425a91d8dSFan Du * 451525a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 451625a91d8dSFan Du * is more preferable. 451725a91d8dSFan Du */ 451825a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 451925a91d8dSFan Du int offset, int len) 452025a91d8dSFan Du { 452148a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 452225a91d8dSFan Du } 452325a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 452425a91d8dSFan Du 452551c739d1SDavid S. Miller 452651c739d1SDavid S. Miller 4527716ea3a7SDavid Howells /** 4528716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4529716ea3a7SDavid Howells * @skb: The socket buffer to check. 4530716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4531716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4532716ea3a7SDavid Howells * 4533716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4534716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4535716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4536716ea3a7SDavid Howells * 4537716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4538716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4539716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4540716ea3a7SDavid Howells * 4541716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4542716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4543716ea3a7SDavid Howells */ 4544716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4545716ea3a7SDavid Howells { 4546716ea3a7SDavid Howells int copyflag; 4547716ea3a7SDavid Howells int elt; 4548716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4549716ea3a7SDavid Howells 4550716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4551716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4552716ea3a7SDavid Howells * at the moment even if they are anonymous). 4553716ea3a7SDavid Howells */ 4554716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4555c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4556716ea3a7SDavid Howells return -ENOMEM; 4557716ea3a7SDavid Howells 4558716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 455921dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4560716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4561716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4562716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4563716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4564716ea3a7SDavid Howells 4565716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4566716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4567716ea3a7SDavid Howells return -ENOMEM; 4568716ea3a7SDavid Howells 4569716ea3a7SDavid Howells /* Voila! */ 4570716ea3a7SDavid Howells *trailer = skb; 4571716ea3a7SDavid Howells return 1; 4572716ea3a7SDavid Howells } 4573716ea3a7SDavid Howells 4574716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4575716ea3a7SDavid Howells 4576716ea3a7SDavid Howells elt = 1; 4577716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4578716ea3a7SDavid Howells copyflag = 0; 4579716ea3a7SDavid Howells 4580716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4581716ea3a7SDavid Howells int ntail = 0; 4582716ea3a7SDavid Howells 4583716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4584716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4585716ea3a7SDavid Howells * after it. */ 4586716ea3a7SDavid Howells 4587716ea3a7SDavid Howells if (skb_shared(skb1)) 4588716ea3a7SDavid Howells copyflag = 1; 4589716ea3a7SDavid Howells 4590716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4591716ea3a7SDavid Howells 4592716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4593716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 459421dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4595716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4596716ea3a7SDavid Howells ntail = tailbits + 128; 4597716ea3a7SDavid Howells } 4598716ea3a7SDavid Howells 4599716ea3a7SDavid Howells if (copyflag || 4600716ea3a7SDavid Howells skb_cloned(skb1) || 4601716ea3a7SDavid Howells ntail || 4602716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 460321dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4604716ea3a7SDavid Howells struct sk_buff *skb2; 4605716ea3a7SDavid Howells 4606716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4607716ea3a7SDavid Howells if (ntail == 0) 4608716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4609716ea3a7SDavid Howells else 4610716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4611716ea3a7SDavid Howells skb_headroom(skb1), 4612716ea3a7SDavid Howells ntail, 4613716ea3a7SDavid Howells GFP_ATOMIC); 4614716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4615716ea3a7SDavid Howells return -ENOMEM; 4616716ea3a7SDavid Howells 4617716ea3a7SDavid Howells if (skb1->sk) 4618716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4619716ea3a7SDavid Howells 4620716ea3a7SDavid Howells /* Looking around. Are we still alive? 4621716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4622716ea3a7SDavid Howells 4623716ea3a7SDavid Howells skb2->next = skb1->next; 4624716ea3a7SDavid Howells *skb_p = skb2; 4625716ea3a7SDavid Howells kfree_skb(skb1); 4626716ea3a7SDavid Howells skb1 = skb2; 4627716ea3a7SDavid Howells } 4628716ea3a7SDavid Howells elt++; 4629716ea3a7SDavid Howells *trailer = skb1; 4630716ea3a7SDavid Howells skb_p = &skb1->next; 4631716ea3a7SDavid Howells } 4632716ea3a7SDavid Howells 4633716ea3a7SDavid Howells return elt; 4634716ea3a7SDavid Howells } 4635b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4636716ea3a7SDavid Howells 4637b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4638b1faf566SEric Dumazet { 4639b1faf566SEric Dumazet struct sock *sk = skb->sk; 4640b1faf566SEric Dumazet 4641b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4642b1faf566SEric Dumazet } 4643b1faf566SEric Dumazet 46448605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 46458605330aSSoheil Hassas Yeganeh { 46468605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 46478605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 46488605330aSSoheil Hassas Yeganeh */ 46498605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 46508605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 46518605330aSSoheil Hassas Yeganeh } 46528605330aSSoheil Hassas Yeganeh 4653b1faf566SEric Dumazet /* 4654b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4655b1faf566SEric Dumazet */ 4656b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4657b1faf566SEric Dumazet { 4658b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4659ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4660b1faf566SEric Dumazet return -ENOMEM; 4661b1faf566SEric Dumazet 4662b1faf566SEric Dumazet skb_orphan(skb); 4663b1faf566SEric Dumazet skb->sk = sk; 4664b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4665b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 46668605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4667b1faf566SEric Dumazet 4668abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4669abb57ea4SEric Dumazet skb_dst_force(skb); 4670abb57ea4SEric Dumazet 4671b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4672b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 4673e3ae2365SAlexander Aring sk_error_report(sk); 4674b1faf566SEric Dumazet return 0; 4675b1faf566SEric Dumazet } 4676b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4677b1faf566SEric Dumazet 467883a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 467983a1a1a7SSoheil Hassas Yeganeh { 468083a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 468183a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 468283a1a1a7SSoheil Hassas Yeganeh } 468383a1a1a7SSoheil Hassas Yeganeh 4684364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4685364a9e93SWillem de Bruijn { 4686364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 468783a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 468883a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4689997d5c3fSEric Dumazet unsigned long flags; 4690364a9e93SWillem de Bruijn 4691997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4692364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 469338b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 469483a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 469538b25793SSoheil Hassas Yeganeh if (icmp_next) 4696985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 469738b25793SSoheil Hassas Yeganeh } 4698997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4699364a9e93SWillem de Bruijn 470083a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 470183a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 470283a1a1a7SSoheil Hassas Yeganeh 470383a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4704e3ae2365SAlexander Aring sk_error_report(sk); 4705364a9e93SWillem de Bruijn 4706364a9e93SWillem de Bruijn return skb; 4707364a9e93SWillem de Bruijn } 4708364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4709364a9e93SWillem de Bruijn 4710cab41c47SAlexander Duyck /** 4711cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4712cab41c47SAlexander Duyck * @skb: the skb to clone 4713cab41c47SAlexander Duyck * 4714cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4715cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4716cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4717cab41c47SAlexander Duyck * 4718cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4719cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4720cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4721cab41c47SAlexander Duyck * the sk_error_queue. 4722cab41c47SAlexander Duyck */ 472362bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 472462bccb8cSAlexander Duyck { 472562bccb8cSAlexander Duyck struct sock *sk = skb->sk; 472662bccb8cSAlexander Duyck struct sk_buff *clone; 472762bccb8cSAlexander Duyck 472841c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 472962bccb8cSAlexander Duyck return NULL; 473062bccb8cSAlexander Duyck 473162bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 473262bccb8cSAlexander Duyck if (!clone) { 473362bccb8cSAlexander Duyck sock_put(sk); 473462bccb8cSAlexander Duyck return NULL; 473562bccb8cSAlexander Duyck } 473662bccb8cSAlexander Duyck 473762bccb8cSAlexander Duyck clone->sk = sk; 473862bccb8cSAlexander Duyck clone->destructor = sock_efree; 473962bccb8cSAlexander Duyck 474062bccb8cSAlexander Duyck return clone; 474162bccb8cSAlexander Duyck } 474262bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 474362bccb8cSAlexander Duyck 474437846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 474537846ef0SAlexander Duyck struct sock *sk, 47464ef1b286SSoheil Hassas Yeganeh int tstype, 47474ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4748ac45f602SPatrick Ohly { 4749ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4750ac45f602SPatrick Ohly int err; 4751ac45f602SPatrick Ohly 47524ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 47534ef1b286SSoheil Hassas Yeganeh 4754ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4755ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4756ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4757ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4758e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 47594ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 47601862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 47614ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 476209c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 476342f67eeaSEric Dumazet if (sk_is_tcp(sk)) 4764a1cdec57SEric Dumazet serr->ee.ee_data -= atomic_read(&sk->sk_tskey); 47654ed2d765SWillem de Bruijn } 476629030374SEric Dumazet 4767ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 476829030374SEric Dumazet 4769ac45f602SPatrick Ohly if (err) 4770ac45f602SPatrick Ohly kfree_skb(skb); 4771ac45f602SPatrick Ohly } 477237846ef0SAlexander Duyck 4773b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4774b245be1fSWillem de Bruijn { 4775b245be1fSWillem de Bruijn bool ret; 4776b245be1fSWillem de Bruijn 4777b245be1fSWillem de Bruijn if (likely(sysctl_tstamp_allow_data || tsonly)) 4778b245be1fSWillem de Bruijn return true; 4779b245be1fSWillem de Bruijn 4780b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4781b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4782b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4783b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4784b245be1fSWillem de Bruijn return ret; 4785b245be1fSWillem de Bruijn } 4786b245be1fSWillem de Bruijn 478737846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 478837846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 478937846ef0SAlexander Duyck { 479037846ef0SAlexander Duyck struct sock *sk = skb->sk; 479137846ef0SAlexander Duyck 4792b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 479335b99dffSWillem de Bruijn goto err; 4794b245be1fSWillem de Bruijn 47959ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 47969ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 47979ac25fc0SEric Dumazet */ 479841c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 479937846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 48004ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 480137846ef0SAlexander Duyck sock_put(sk); 480235b99dffSWillem de Bruijn return; 480337846ef0SAlexander Duyck } 480435b99dffSWillem de Bruijn 480535b99dffSWillem de Bruijn err: 480635b99dffSWillem de Bruijn kfree_skb(skb); 48079ac25fc0SEric Dumazet } 480837846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 480937846ef0SAlexander Duyck 481037846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 4811e7ed11eeSYousuk Seung const struct sk_buff *ack_skb, 481237846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 481337846ef0SAlexander Duyck struct sock *sk, int tstype) 481437846ef0SAlexander Duyck { 481537846ef0SAlexander Duyck struct sk_buff *skb; 48164ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 481737846ef0SAlexander Duyck 48183a8dd971SWillem de Bruijn if (!sk) 48193a8dd971SWillem de Bruijn return; 48203a8dd971SWillem de Bruijn 4821b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4822b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4823b50a5c70SMiroslav Lichvar return; 4824b50a5c70SMiroslav Lichvar 48253a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 48263a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 482737846ef0SAlexander Duyck return; 482837846ef0SAlexander Duyck 48291c885808SFrancis Yan if (tsonly) { 48301c885808SFrancis Yan #ifdef CONFIG_INET 48311c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 483242f67eeaSEric Dumazet sk_is_tcp(sk)) { 4833e7ed11eeSYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb, 4834e7ed11eeSYousuk Seung ack_skb); 48354ef1b286SSoheil Hassas Yeganeh opt_stats = true; 48364ef1b286SSoheil Hassas Yeganeh } else 48371c885808SFrancis Yan #endif 48381c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 48391c885808SFrancis Yan } else { 484037846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 48411c885808SFrancis Yan } 484237846ef0SAlexander Duyck if (!skb) 484337846ef0SAlexander Duyck return; 484437846ef0SAlexander Duyck 484549ca0d8bSWillem de Bruijn if (tsonly) { 4846fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4847fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 484849ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 484949ca0d8bSWillem de Bruijn } 485049ca0d8bSWillem de Bruijn 485149ca0d8bSWillem de Bruijn if (hwtstamps) 485249ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 485349ca0d8bSWillem de Bruijn else 485449ca0d8bSWillem de Bruijn skb->tstamp = ktime_get_real(); 485549ca0d8bSWillem de Bruijn 48564ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 485737846ef0SAlexander Duyck } 4858e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4859e7fd2885SWillem de Bruijn 4860e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4861e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4862e7fd2885SWillem de Bruijn { 4863e7ed11eeSYousuk Seung return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk, 4864e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4865e7fd2885SWillem de Bruijn } 4866ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4867ac45f602SPatrick Ohly 48686e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 48696e3e939fSJohannes Berg { 48706e3e939fSJohannes Berg struct sock *sk = skb->sk; 48716e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4872dd4f1072SEric Dumazet int err = 1; 48736e3e939fSJohannes Berg 48746e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 48756e3e939fSJohannes Berg skb->wifi_acked = acked; 48766e3e939fSJohannes Berg 48776e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 48786e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 48796e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 48806e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 48816e3e939fSJohannes Berg 4882dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4883dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4884dd4f1072SEric Dumazet */ 488541c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 48866e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4887dd4f1072SEric Dumazet sock_put(sk); 4888dd4f1072SEric Dumazet } 48896e3e939fSJohannes Berg if (err) 48906e3e939fSJohannes Berg kfree_skb(skb); 48916e3e939fSJohannes Berg } 48926e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 48936e3e939fSJohannes Berg 4894f35d9d8aSRusty Russell /** 4895f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4896f35d9d8aSRusty Russell * @skb: the skb to set 4897f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4898f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4899f35d9d8aSRusty Russell * 4900f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4901f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4902f35d9d8aSRusty Russell * 4903f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4904f35d9d8aSRusty Russell * returns false you should drop the packet. 4905f35d9d8aSRusty Russell */ 4906f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4907f35d9d8aSRusty Russell { 490852b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 490952b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 491052b5d6f5SEric Dumazet 491152b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 491252b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 491352b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4914f35d9d8aSRusty Russell return false; 4915f35d9d8aSRusty Russell } 4916f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 491752b5d6f5SEric Dumazet skb->csum_start = csum_start; 4918f35d9d8aSRusty Russell skb->csum_offset = off; 4919e5d5decaSJason Wang skb_set_transport_header(skb, start); 4920f35d9d8aSRusty Russell return true; 4921f35d9d8aSRusty Russell } 4922b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4923f35d9d8aSRusty Russell 4924ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4925ed1f50c3SPaul Durrant unsigned int max) 4926ed1f50c3SPaul Durrant { 4927ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4928ed1f50c3SPaul Durrant return 0; 4929ed1f50c3SPaul Durrant 4930ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4931ed1f50c3SPaul Durrant * won't need to do it again. 4932ed1f50c3SPaul Durrant */ 4933ed1f50c3SPaul Durrant if (max > skb->len) 4934ed1f50c3SPaul Durrant max = skb->len; 4935ed1f50c3SPaul Durrant 4936ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4937ed1f50c3SPaul Durrant return -ENOMEM; 4938ed1f50c3SPaul Durrant 4939ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4940ed1f50c3SPaul Durrant return -EPROTO; 4941ed1f50c3SPaul Durrant 4942ed1f50c3SPaul Durrant return 0; 4943ed1f50c3SPaul Durrant } 4944ed1f50c3SPaul Durrant 4945f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4946f9708b43SJan Beulich 4947f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4948f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4949f9708b43SJan Beulich unsigned int off) 4950f9708b43SJan Beulich { 4951f9708b43SJan Beulich int err; 4952f9708b43SJan Beulich 4953161d1792SKees Cook switch (proto) { 4954f9708b43SJan Beulich case IPPROTO_TCP: 4955f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4956f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4957f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4958f9708b43SJan Beulich offsetof(struct tcphdr, 4959f9708b43SJan Beulich check))) 4960f9708b43SJan Beulich err = -EPROTO; 4961f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4962f9708b43SJan Beulich 4963f9708b43SJan Beulich case IPPROTO_UDP: 4964f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4965f9708b43SJan Beulich off + sizeof(struct udphdr)); 4966f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4967f9708b43SJan Beulich offsetof(struct udphdr, 4968f9708b43SJan Beulich check))) 4969f9708b43SJan Beulich err = -EPROTO; 4970f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4971f9708b43SJan Beulich } 4972f9708b43SJan Beulich 4973f9708b43SJan Beulich return ERR_PTR(-EPROTO); 4974f9708b43SJan Beulich } 4975f9708b43SJan Beulich 4976ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 4977ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 4978ed1f50c3SPaul Durrant */ 4979ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 4980ed1f50c3SPaul Durrant 4981f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4982ed1f50c3SPaul Durrant { 4983ed1f50c3SPaul Durrant unsigned int off; 4984ed1f50c3SPaul Durrant bool fragment; 4985f9708b43SJan Beulich __sum16 *csum; 4986ed1f50c3SPaul Durrant int err; 4987ed1f50c3SPaul Durrant 4988ed1f50c3SPaul Durrant fragment = false; 4989ed1f50c3SPaul Durrant 4990ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 4991ed1f50c3SPaul Durrant sizeof(struct iphdr), 4992ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 4993ed1f50c3SPaul Durrant if (err < 0) 4994ed1f50c3SPaul Durrant goto out; 4995ed1f50c3SPaul Durrant 499611f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 4997ed1f50c3SPaul Durrant fragment = true; 4998ed1f50c3SPaul Durrant 4999ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 5000ed1f50c3SPaul Durrant 5001ed1f50c3SPaul Durrant err = -EPROTO; 5002ed1f50c3SPaul Durrant 5003ed1f50c3SPaul Durrant if (fragment) 5004ed1f50c3SPaul Durrant goto out; 5005ed1f50c3SPaul Durrant 5006f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 5007f9708b43SJan Beulich if (IS_ERR(csum)) 5008f9708b43SJan Beulich return PTR_ERR(csum); 5009ed1f50c3SPaul Durrant 5010ed1f50c3SPaul Durrant if (recalculate) 5011f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 5012ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 5013ed1f50c3SPaul Durrant skb->len - off, 5014f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 5015ed1f50c3SPaul Durrant err = 0; 5016ed1f50c3SPaul Durrant 5017ed1f50c3SPaul Durrant out: 5018ed1f50c3SPaul Durrant return err; 5019ed1f50c3SPaul Durrant } 5020ed1f50c3SPaul Durrant 5021ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5022ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 5023ed1f50c3SPaul Durrant */ 5024ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 5025ed1f50c3SPaul Durrant 5026ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 5027ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 5028ed1f50c3SPaul Durrant 5029ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 5030ed1f50c3SPaul Durrant { 5031ed1f50c3SPaul Durrant int err; 5032ed1f50c3SPaul Durrant u8 nexthdr; 5033ed1f50c3SPaul Durrant unsigned int off; 5034ed1f50c3SPaul Durrant unsigned int len; 5035ed1f50c3SPaul Durrant bool fragment; 5036ed1f50c3SPaul Durrant bool done; 5037f9708b43SJan Beulich __sum16 *csum; 5038ed1f50c3SPaul Durrant 5039ed1f50c3SPaul Durrant fragment = false; 5040ed1f50c3SPaul Durrant done = false; 5041ed1f50c3SPaul Durrant 5042ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 5043ed1f50c3SPaul Durrant 5044ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 5045ed1f50c3SPaul Durrant if (err < 0) 5046ed1f50c3SPaul Durrant goto out; 5047ed1f50c3SPaul Durrant 5048ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 5049ed1f50c3SPaul Durrant 5050ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 5051ed1f50c3SPaul Durrant while (off <= len && !done) { 5052ed1f50c3SPaul Durrant switch (nexthdr) { 5053ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 5054ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 5055ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 5056ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 5057ed1f50c3SPaul Durrant 5058ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5059ed1f50c3SPaul Durrant off + 5060ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 5061ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5062ed1f50c3SPaul Durrant if (err < 0) 5063ed1f50c3SPaul Durrant goto out; 5064ed1f50c3SPaul Durrant 5065ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 5066ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5067ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 5068ed1f50c3SPaul Durrant break; 5069ed1f50c3SPaul Durrant } 5070ed1f50c3SPaul Durrant case IPPROTO_AH: { 5071ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 5072ed1f50c3SPaul Durrant 5073ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5074ed1f50c3SPaul Durrant off + 5075ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 5076ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5077ed1f50c3SPaul Durrant if (err < 0) 5078ed1f50c3SPaul Durrant goto out; 5079ed1f50c3SPaul Durrant 5080ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 5081ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5082ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 5083ed1f50c3SPaul Durrant break; 5084ed1f50c3SPaul Durrant } 5085ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 5086ed1f50c3SPaul Durrant struct frag_hdr *hp; 5087ed1f50c3SPaul Durrant 5088ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5089ed1f50c3SPaul Durrant off + 5090ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5091ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5092ed1f50c3SPaul Durrant if (err < 0) 5093ed1f50c3SPaul Durrant goto out; 5094ed1f50c3SPaul Durrant 5095ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5096ed1f50c3SPaul Durrant 5097ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5098ed1f50c3SPaul Durrant fragment = true; 5099ed1f50c3SPaul Durrant 5100ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5101ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5102ed1f50c3SPaul Durrant break; 5103ed1f50c3SPaul Durrant } 5104ed1f50c3SPaul Durrant default: 5105ed1f50c3SPaul Durrant done = true; 5106ed1f50c3SPaul Durrant break; 5107ed1f50c3SPaul Durrant } 5108ed1f50c3SPaul Durrant } 5109ed1f50c3SPaul Durrant 5110ed1f50c3SPaul Durrant err = -EPROTO; 5111ed1f50c3SPaul Durrant 5112ed1f50c3SPaul Durrant if (!done || fragment) 5113ed1f50c3SPaul Durrant goto out; 5114ed1f50c3SPaul Durrant 5115f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5116f9708b43SJan Beulich if (IS_ERR(csum)) 5117f9708b43SJan Beulich return PTR_ERR(csum); 5118ed1f50c3SPaul Durrant 5119ed1f50c3SPaul Durrant if (recalculate) 5120f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5121ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5122f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5123ed1f50c3SPaul Durrant err = 0; 5124ed1f50c3SPaul Durrant 5125ed1f50c3SPaul Durrant out: 5126ed1f50c3SPaul Durrant return err; 5127ed1f50c3SPaul Durrant } 5128ed1f50c3SPaul Durrant 5129ed1f50c3SPaul Durrant /** 5130ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5131ed1f50c3SPaul Durrant * @skb: the skb to set up 5132ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5133ed1f50c3SPaul Durrant */ 5134ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5135ed1f50c3SPaul Durrant { 5136ed1f50c3SPaul Durrant int err; 5137ed1f50c3SPaul Durrant 5138ed1f50c3SPaul Durrant switch (skb->protocol) { 5139ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5140f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5141ed1f50c3SPaul Durrant break; 5142ed1f50c3SPaul Durrant 5143ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5144ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5145ed1f50c3SPaul Durrant break; 5146ed1f50c3SPaul Durrant 5147ed1f50c3SPaul Durrant default: 5148ed1f50c3SPaul Durrant err = -EPROTO; 5149ed1f50c3SPaul Durrant break; 5150ed1f50c3SPaul Durrant } 5151ed1f50c3SPaul Durrant 5152ed1f50c3SPaul Durrant return err; 5153ed1f50c3SPaul Durrant } 5154ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5155ed1f50c3SPaul Durrant 51569afd85c9SLinus Lüssing /** 51579afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 51589afd85c9SLinus Lüssing * @skb: the skb to check 51599afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51609afd85c9SLinus Lüssing * 51619afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 51629afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 51639afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 51649afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 51659afd85c9SLinus Lüssing * 5166a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5167a516993fSLinus Lüssing * differs from the provided skb. 51689afd85c9SLinus Lüssing */ 51699afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 51709afd85c9SLinus Lüssing unsigned int transport_len) 51719afd85c9SLinus Lüssing { 51729afd85c9SLinus Lüssing struct sk_buff *skb_chk; 51739afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 51749afd85c9SLinus Lüssing int ret; 51759afd85c9SLinus Lüssing 5176a516993fSLinus Lüssing if (skb->len < len) 51779afd85c9SLinus Lüssing return NULL; 5178a516993fSLinus Lüssing else if (skb->len == len) 51799afd85c9SLinus Lüssing return skb; 51809afd85c9SLinus Lüssing 51819afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 51829afd85c9SLinus Lüssing if (!skb_chk) 51839afd85c9SLinus Lüssing return NULL; 51849afd85c9SLinus Lüssing 51859afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 51869afd85c9SLinus Lüssing if (ret) { 51879afd85c9SLinus Lüssing kfree_skb(skb_chk); 51889afd85c9SLinus Lüssing return NULL; 51899afd85c9SLinus Lüssing } 51909afd85c9SLinus Lüssing 51919afd85c9SLinus Lüssing return skb_chk; 51929afd85c9SLinus Lüssing } 51939afd85c9SLinus Lüssing 51949afd85c9SLinus Lüssing /** 51959afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 51969afd85c9SLinus Lüssing * @skb: the skb to check 51979afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51989afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 51999afd85c9SLinus Lüssing * 52009afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 52019afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 52029afd85c9SLinus Lüssing * 52039afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 52049afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 52059afd85c9SLinus Lüssing * 5206a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5207a516993fSLinus Lüssing * differs from the provided skb. 52089afd85c9SLinus Lüssing */ 52099afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 52109afd85c9SLinus Lüssing unsigned int transport_len, 52119afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 52129afd85c9SLinus Lüssing { 52139afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52149afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5215fcba67c9SLinus Lüssing __sum16 ret; 52169afd85c9SLinus Lüssing 52179afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 52189afd85c9SLinus Lüssing if (!skb_chk) 5219a516993fSLinus Lüssing goto err; 52209afd85c9SLinus Lüssing 5221a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5222a516993fSLinus Lüssing goto err; 52239afd85c9SLinus Lüssing 52249b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 52259afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 52269b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 52279afd85c9SLinus Lüssing 5228a516993fSLinus Lüssing if (ret) 5229a516993fSLinus Lüssing goto err; 52309afd85c9SLinus Lüssing 52319afd85c9SLinus Lüssing return skb_chk; 5232a516993fSLinus Lüssing 5233a516993fSLinus Lüssing err: 5234a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5235a516993fSLinus Lüssing kfree_skb(skb_chk); 5236a516993fSLinus Lüssing 5237a516993fSLinus Lüssing return NULL; 5238a516993fSLinus Lüssing 52399afd85c9SLinus Lüssing } 52409afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 52419afd85c9SLinus Lüssing 52424497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 52434497b076SBen Hutchings { 5244e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5245e87cc472SJoe Perches skb->dev->name); 52464497b076SBen Hutchings } 52474497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5248bad43ca8SEric Dumazet 5249bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5250bad43ca8SEric Dumazet { 52513d861f66SEric Dumazet if (head_stolen) { 52523d861f66SEric Dumazet skb_release_head_state(skb); 5253bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 52543d861f66SEric Dumazet } else { 5255bad43ca8SEric Dumazet __kfree_skb(skb); 5256bad43ca8SEric Dumazet } 52573d861f66SEric Dumazet } 5258bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5259bad43ca8SEric Dumazet 5260bad43ca8SEric Dumazet /** 5261bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5262bad43ca8SEric Dumazet * @to: prior buffer 5263bad43ca8SEric Dumazet * @from: buffer to add 5264bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5265c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5266bad43ca8SEric Dumazet */ 5267bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5268bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5269bad43ca8SEric Dumazet { 5270c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5271bad43ca8SEric Dumazet int i, delta, len = from->len; 5272bad43ca8SEric Dumazet 5273bad43ca8SEric Dumazet *fragstolen = false; 5274bad43ca8SEric Dumazet 5275bad43ca8SEric Dumazet if (skb_cloned(to)) 5276bad43ca8SEric Dumazet return false; 5277bad43ca8SEric Dumazet 52786a5bcd84SIlias Apalodimas /* The page pool signature of struct page will eventually figure out 52796a5bcd84SIlias Apalodimas * which pages can be recycled or not but for now let's prohibit slab 52806a5bcd84SIlias Apalodimas * allocated and page_pool allocated SKBs from being coalesced. 52816a5bcd84SIlias Apalodimas */ 52826a5bcd84SIlias Apalodimas if (to->pp_recycle != from->pp_recycle) 52836a5bcd84SIlias Apalodimas return false; 52846a5bcd84SIlias Apalodimas 5285bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5286e93a0435SEric Dumazet if (len) 5287bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5288bad43ca8SEric Dumazet *delta_truesize = 0; 5289bad43ca8SEric Dumazet return true; 5290bad43ca8SEric Dumazet } 5291bad43ca8SEric Dumazet 5292c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5293c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5294c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5295bad43ca8SEric Dumazet return false; 52961f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 52971f8b977aSWillem de Bruijn return false; 5298bad43ca8SEric Dumazet 5299bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5300bad43ca8SEric Dumazet struct page *page; 5301bad43ca8SEric Dumazet unsigned int offset; 5302bad43ca8SEric Dumazet 5303c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5304c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5305bad43ca8SEric Dumazet return false; 5306bad43ca8SEric Dumazet 5307bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5308bad43ca8SEric Dumazet return false; 5309bad43ca8SEric Dumazet 5310bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5311bad43ca8SEric Dumazet 5312bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5313bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5314bad43ca8SEric Dumazet 5315c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5316bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5317bad43ca8SEric Dumazet *fragstolen = true; 5318bad43ca8SEric Dumazet } else { 5319c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5320c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5321bad43ca8SEric Dumazet return false; 5322bad43ca8SEric Dumazet 5323f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5324bad43ca8SEric Dumazet } 5325bad43ca8SEric Dumazet 5326bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5327bad43ca8SEric Dumazet 5328c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5329c818fa9eSEric Dumazet from_shinfo->frags, 5330c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5331c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5332bad43ca8SEric Dumazet 5333bad43ca8SEric Dumazet if (!skb_cloned(from)) 5334c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5335bad43ca8SEric Dumazet 53368ea853fdSLi RongQing /* if the skb is not cloned this does nothing 53378ea853fdSLi RongQing * since we set nr_frags to 0. 53388ea853fdSLi RongQing */ 5339c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5340c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5341bad43ca8SEric Dumazet 5342bad43ca8SEric Dumazet to->truesize += delta; 5343bad43ca8SEric Dumazet to->len += len; 5344bad43ca8SEric Dumazet to->data_len += len; 5345bad43ca8SEric Dumazet 5346bad43ca8SEric Dumazet *delta_truesize = delta; 5347bad43ca8SEric Dumazet return true; 5348bad43ca8SEric Dumazet } 5349bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5350621e84d6SNicolas Dichtel 5351621e84d6SNicolas Dichtel /** 53528b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5353621e84d6SNicolas Dichtel * 5354621e84d6SNicolas Dichtel * @skb: buffer to clean 53558b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5356621e84d6SNicolas Dichtel * 53578b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 53588b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 53598b27f277SNicolas Dichtel * operations. 53608b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 53618b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 53628b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5363621e84d6SNicolas Dichtel */ 53648b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5365621e84d6SNicolas Dichtel { 5366621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5367621e84d6SNicolas Dichtel skb->skb_iif = 0; 536860ff7467SWANG Cong skb->ignore_df = 0; 5369621e84d6SNicolas Dichtel skb_dst_drop(skb); 5370174e2381SFlorian Westphal skb_ext_reset(skb); 5371895b5c9fSFlorian Westphal nf_reset_ct(skb); 5372621e84d6SNicolas Dichtel nf_reset_trace(skb); 5373213dd74aSHerbert Xu 53746f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 53756f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5376875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 53776f9a5069SPetr Machata #endif 53786f9a5069SPetr Machata 5379213dd74aSHerbert Xu if (!xnet) 5380213dd74aSHerbert Xu return; 5381213dd74aSHerbert Xu 53822b5ec1a5SYe Yin ipvs_reset(skb); 5383213dd74aSHerbert Xu skb->mark = 0; 5384*de799101SMartin KaFai Lau skb_clear_tstamp(skb); 5385621e84d6SNicolas Dichtel } 5386621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5387de960aa9SFlorian Westphal 5388de960aa9SFlorian Westphal /** 5389de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5390de960aa9SFlorian Westphal * 5391de960aa9SFlorian Westphal * @skb: GSO skb 5392de960aa9SFlorian Westphal * 5393de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5394de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5395de960aa9SFlorian Westphal * 5396de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5397de960aa9SFlorian Westphal */ 5398a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5399de960aa9SFlorian Westphal { 5400de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5401f993bc25SFlorian Westphal unsigned int thlen = 0; 5402f993bc25SFlorian Westphal 5403f993bc25SFlorian Westphal if (skb->encapsulation) { 5404f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5405f993bc25SFlorian Westphal skb_transport_header(skb); 5406de960aa9SFlorian Westphal 5407de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5408f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5409f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5410f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 54111dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 541290017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5413ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5414ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5415f993bc25SFlorian Westphal } 54166d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 54176d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 54186d39d589SFlorian Westphal * accounted for. 54196d39d589SFlorian Westphal */ 5420f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5421de960aa9SFlorian Westphal } 5422a4a77718SDaniel Axtens 5423a4a77718SDaniel Axtens /** 5424a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5425a4a77718SDaniel Axtens * 5426a4a77718SDaniel Axtens * @skb: GSO skb 5427a4a77718SDaniel Axtens * 5428a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5429a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5430a4a77718SDaniel Axtens * 5431a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5432a4a77718SDaniel Axtens */ 5433a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5434a4a77718SDaniel Axtens { 5435a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5436a4a77718SDaniel Axtens skb_network_header(skb); 5437a4a77718SDaniel Axtens 5438a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5439a4a77718SDaniel Axtens } 5440a4a77718SDaniel Axtens 5441a4a77718SDaniel Axtens /** 5442a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5443a4a77718SDaniel Axtens * 5444a4a77718SDaniel Axtens * @skb: GSO skb 5445a4a77718SDaniel Axtens * 5446a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5447a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5448a4a77718SDaniel Axtens * headers (TCP/UDP). 5449a4a77718SDaniel Axtens */ 5450a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5451a4a77718SDaniel Axtens { 5452a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5453a4a77718SDaniel Axtens 5454a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5455a4a77718SDaniel Axtens } 54560d5501c1SVlad Yasevich 5457ae7ef81eSMarcelo Ricardo Leitner /** 54582b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 54592b16f048SDaniel Axtens * 54602b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 54612b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 54622b16f048SDaniel Axtens * 54632b16f048SDaniel Axtens * We might want to check: 54642b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 54652b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 54662b16f048SDaniel Axtens * 54672b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 54682b16f048SDaniel Axtens * 546949682bfaSMathieu Malaterre * @skb: GSO skb 547049682bfaSMathieu Malaterre * 54712b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 54722b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 54732b16f048SDaniel Axtens * 54742b16f048SDaniel Axtens * @max_len: The maximum permissible length. 54752b16f048SDaniel Axtens * 54762b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 54772b16f048SDaniel Axtens */ 54782b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 54792b16f048SDaniel Axtens unsigned int seg_len, 54802b16f048SDaniel Axtens unsigned int max_len) { 54812b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 54822b16f048SDaniel Axtens const struct sk_buff *iter; 54832b16f048SDaniel Axtens 54842b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 54852b16f048SDaniel Axtens return seg_len <= max_len; 54862b16f048SDaniel Axtens 54872b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 54882b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 54892b16f048SDaniel Axtens 54902b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 54912b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 54922b16f048SDaniel Axtens return false; 54932b16f048SDaniel Axtens } 54942b16f048SDaniel Axtens 54952b16f048SDaniel Axtens return true; 54962b16f048SDaniel Axtens } 54972b16f048SDaniel Axtens 54982b16f048SDaniel Axtens /** 5499779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5500ae7ef81eSMarcelo Ricardo Leitner * 5501ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 550276f21b99SDavid S. Miller * @mtu: MTU to validate against 5503ae7ef81eSMarcelo Ricardo Leitner * 5504779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5505779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5506779b7931SDaniel Axtens * payload. 5507ae7ef81eSMarcelo Ricardo Leitner */ 5508779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5509ae7ef81eSMarcelo Ricardo Leitner { 55102b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5511ae7ef81eSMarcelo Ricardo Leitner } 5512779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5513ae7ef81eSMarcelo Ricardo Leitner 55142b16f048SDaniel Axtens /** 55152b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 55162b16f048SDaniel Axtens * 55172b16f048SDaniel Axtens * @skb: GSO skb 55182b16f048SDaniel Axtens * @len: length to validate against 55192b16f048SDaniel Axtens * 55202b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 55212b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 55222b16f048SDaniel Axtens */ 55232b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 55242b16f048SDaniel Axtens { 55252b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 55262b16f048SDaniel Axtens } 55272b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 55282b16f048SDaniel Axtens 55290d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 55300d5501c1SVlad Yasevich { 5531d85e8be2SYuya Kusakabe int mac_len, meta_len; 5532d85e8be2SYuya Kusakabe void *meta; 55334bbb3e0eSToshiaki Makita 55340d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 55350d5501c1SVlad Yasevich kfree_skb(skb); 55360d5501c1SVlad Yasevich return NULL; 55370d5501c1SVlad Yasevich } 55380d5501c1SVlad Yasevich 55394bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5540ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 55414bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 55424bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5543ae474573SToshiaki Makita } 5544d85e8be2SYuya Kusakabe 5545d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5546d85e8be2SYuya Kusakabe if (meta_len) { 5547d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5548d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5549d85e8be2SYuya Kusakabe } 5550d85e8be2SYuya Kusakabe 55510d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 55520d5501c1SVlad Yasevich return skb; 55530d5501c1SVlad Yasevich } 55540d5501c1SVlad Yasevich 55550d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 55560d5501c1SVlad Yasevich { 55570d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 55580d5501c1SVlad Yasevich u16 vlan_tci; 55590d5501c1SVlad Yasevich 5560df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 55610d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 55620d5501c1SVlad Yasevich return skb; 55630d5501c1SVlad Yasevich } 55640d5501c1SVlad Yasevich 55650d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 55660d5501c1SVlad Yasevich if (unlikely(!skb)) 55670d5501c1SVlad Yasevich goto err_free; 556855eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 556955eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 55700d5501c1SVlad Yasevich goto err_free; 55710d5501c1SVlad Yasevich 55720d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 55730d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 55740d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 55750d5501c1SVlad Yasevich 55760d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 55770d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 55780d5501c1SVlad Yasevich 55790d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 55800d5501c1SVlad Yasevich if (unlikely(!skb)) 55810d5501c1SVlad Yasevich goto err_free; 55820d5501c1SVlad Yasevich 55830d5501c1SVlad Yasevich skb_reset_network_header(skb); 55848be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 55850d5501c1SVlad Yasevich skb_reset_transport_header(skb); 55860d5501c1SVlad Yasevich skb_reset_mac_len(skb); 55870d5501c1SVlad Yasevich 55880d5501c1SVlad Yasevich return skb; 55890d5501c1SVlad Yasevich 55900d5501c1SVlad Yasevich err_free: 55910d5501c1SVlad Yasevich kfree_skb(skb); 55920d5501c1SVlad Yasevich return NULL; 55930d5501c1SVlad Yasevich } 55940d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 55952e4e4410SEric Dumazet 5596e2195121SJiri Pirko int skb_ensure_writable(struct sk_buff *skb, int write_len) 5597e2195121SJiri Pirko { 5598e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5599e2195121SJiri Pirko return -ENOMEM; 5600e2195121SJiri Pirko 5601e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5602e2195121SJiri Pirko return 0; 5603e2195121SJiri Pirko 5604e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5605e2195121SJiri Pirko } 5606e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5607e2195121SJiri Pirko 5608bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5609bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5610bfca4c52SShmulik Ladkani */ 5611bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 561293515d53SJiri Pirko { 561393515d53SJiri Pirko struct vlan_hdr *vhdr; 5614b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 561593515d53SJiri Pirko int err; 561693515d53SJiri Pirko 5617b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5618b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5619b6a79208SShmulik Ladkani offset)) { 5620b6a79208SShmulik Ladkani return -EINVAL; 5621b6a79208SShmulik Ladkani } 5622b6a79208SShmulik Ladkani 562393515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 562493515d53SJiri Pirko if (unlikely(err)) 5625b6a79208SShmulik Ladkani return err; 562693515d53SJiri Pirko 562793515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 562893515d53SJiri Pirko 562993515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 563093515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 563193515d53SJiri Pirko 563293515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 563393515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 563493515d53SJiri Pirko 563593515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 563693515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 563793515d53SJiri Pirko 563893515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 563993515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 564093515d53SJiri Pirko 564193515d53SJiri Pirko skb_reset_mac_len(skb); 564293515d53SJiri Pirko 564393515d53SJiri Pirko return err; 564493515d53SJiri Pirko } 5645bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 564693515d53SJiri Pirko 5647b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5648b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5649b6a79208SShmulik Ladkani */ 565093515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 565193515d53SJiri Pirko { 565293515d53SJiri Pirko u16 vlan_tci; 565393515d53SJiri Pirko __be16 vlan_proto; 565493515d53SJiri Pirko int err; 565593515d53SJiri Pirko 5656df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5657b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 565893515d53SJiri Pirko } else { 5659ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 566093515d53SJiri Pirko return 0; 566193515d53SJiri Pirko 566293515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 566393515d53SJiri Pirko if (err) 566493515d53SJiri Pirko return err; 566593515d53SJiri Pirko } 566693515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5667ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 566893515d53SJiri Pirko return 0; 566993515d53SJiri Pirko 567093515d53SJiri Pirko vlan_proto = skb->protocol; 567193515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 567293515d53SJiri Pirko if (unlikely(err)) 567393515d53SJiri Pirko return err; 567493515d53SJiri Pirko 567593515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 567693515d53SJiri Pirko return 0; 567793515d53SJiri Pirko } 567893515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 567993515d53SJiri Pirko 5680b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5681b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5682b6a79208SShmulik Ladkani */ 568393515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 568493515d53SJiri Pirko { 5685df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5686b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 568793515d53SJiri Pirko int err; 568893515d53SJiri Pirko 5689b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5690b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5691b6a79208SShmulik Ladkani offset)) { 5692b6a79208SShmulik Ladkani return -EINVAL; 5693b6a79208SShmulik Ladkani } 5694b6a79208SShmulik Ladkani 569593515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5696df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5697b6a79208SShmulik Ladkani if (err) 569893515d53SJiri Pirko return err; 56999241e2dfSDaniel Borkmann 570093515d53SJiri Pirko skb->protocol = skb->vlan_proto; 570193515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 570293515d53SJiri Pirko 57036b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 570493515d53SJiri Pirko } 570593515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 570693515d53SJiri Pirko return 0; 570793515d53SJiri Pirko } 570893515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 570993515d53SJiri Pirko 571019fbcb36SGuillaume Nault /** 571119fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 571219fbcb36SGuillaume Nault * 571319fbcb36SGuillaume Nault * @skb: Socket buffer to modify 571419fbcb36SGuillaume Nault * 571519fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 571619fbcb36SGuillaume Nault * 571719fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 571819fbcb36SGuillaume Nault * present. 571919fbcb36SGuillaume Nault * 572019fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 572119fbcb36SGuillaume Nault */ 572219fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 572319fbcb36SGuillaume Nault { 572419fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 572519fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 572619fbcb36SGuillaume Nault return -EPROTO; 572719fbcb36SGuillaume Nault 572819fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 572919fbcb36SGuillaume Nault skb_reset_mac_header(skb); 573019fbcb36SGuillaume Nault skb_reset_mac_len(skb); 573119fbcb36SGuillaume Nault 573219fbcb36SGuillaume Nault return 0; 573319fbcb36SGuillaume Nault } 573419fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 573519fbcb36SGuillaume Nault 573619fbcb36SGuillaume Nault /** 573719fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 573819fbcb36SGuillaume Nault * 573919fbcb36SGuillaume Nault * @skb: Socket buffer to modify 574019fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 574119fbcb36SGuillaume Nault * @src: Source MAC address of the new header 574219fbcb36SGuillaume Nault * 574319fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 574419fbcb36SGuillaume Nault * 574519fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 574619fbcb36SGuillaume Nault * 574719fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 574819fbcb36SGuillaume Nault */ 574919fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 575019fbcb36SGuillaume Nault const unsigned char *src) 575119fbcb36SGuillaume Nault { 575219fbcb36SGuillaume Nault struct ethhdr *eth; 575319fbcb36SGuillaume Nault int err; 575419fbcb36SGuillaume Nault 575519fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 575619fbcb36SGuillaume Nault return -EPROTO; 575719fbcb36SGuillaume Nault 575819fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 575919fbcb36SGuillaume Nault if (err < 0) 576019fbcb36SGuillaume Nault return err; 576119fbcb36SGuillaume Nault 576219fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 576319fbcb36SGuillaume Nault skb_reset_mac_header(skb); 576419fbcb36SGuillaume Nault skb_reset_mac_len(skb); 576519fbcb36SGuillaume Nault 576619fbcb36SGuillaume Nault eth = eth_hdr(skb); 576719fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 576819fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 576919fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 577019fbcb36SGuillaume Nault 577119fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 577219fbcb36SGuillaume Nault 577319fbcb36SGuillaume Nault return 0; 577419fbcb36SGuillaume Nault } 577519fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 577619fbcb36SGuillaume Nault 57778822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 57788822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 57798822e270SJohn Hurley __be16 ethertype) 57808822e270SJohn Hurley { 57818822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 57828822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 57838822e270SJohn Hurley 57848822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 57858822e270SJohn Hurley } 57868822e270SJohn Hurley 57878822e270SJohn Hurley hdr->h_proto = ethertype; 57888822e270SJohn Hurley } 57898822e270SJohn Hurley 57908822e270SJohn Hurley /** 5791e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5792e7dbfed1SMartin Varghese * the packet 57938822e270SJohn Hurley * 57948822e270SJohn Hurley * @skb: buffer 57958822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 57968822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5797fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5798e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5799e7dbfed1SMartin Varghese * ethernet 58008822e270SJohn Hurley * 58018822e270SJohn Hurley * Expects skb->data at mac header. 58028822e270SJohn Hurley * 58038822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 58048822e270SJohn Hurley */ 5805fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5806d04ac224SMartin Varghese int mac_len, bool ethernet) 58078822e270SJohn Hurley { 58088822e270SJohn Hurley struct mpls_shim_hdr *lse; 58098822e270SJohn Hurley int err; 58108822e270SJohn Hurley 58118822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 58128822e270SJohn Hurley return -EINVAL; 58138822e270SJohn Hurley 58148822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 58158822e270SJohn Hurley if (skb->encapsulation) 58168822e270SJohn Hurley return -EINVAL; 58178822e270SJohn Hurley 58188822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 58198822e270SJohn Hurley if (unlikely(err)) 58208822e270SJohn Hurley return err; 58218822e270SJohn Hurley 58228822e270SJohn Hurley if (!skb->inner_protocol) { 5823e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 58248822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 58258822e270SJohn Hurley } 58268822e270SJohn Hurley 58278822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 58288822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5829fa4e0f88SDavide Caratti mac_len); 58308822e270SJohn Hurley skb_reset_mac_header(skb); 5831fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5832e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 58338822e270SJohn Hurley 58348822e270SJohn Hurley lse = mpls_hdr(skb); 58358822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 58368822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 58378822e270SJohn Hurley 58384296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 58398822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 58408822e270SJohn Hurley skb->protocol = mpls_proto; 58418822e270SJohn Hurley 58428822e270SJohn Hurley return 0; 58438822e270SJohn Hurley } 58448822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 58458822e270SJohn Hurley 58462e4e4410SEric Dumazet /** 5847ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5848ed246ceeSJohn Hurley * 5849ed246ceeSJohn Hurley * @skb: buffer 5850ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5851fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 585276f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5853ed246ceeSJohn Hurley * 5854ed246ceeSJohn Hurley * Expects skb->data at mac header. 5855ed246ceeSJohn Hurley * 5856ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5857ed246ceeSJohn Hurley */ 5858040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5859040b5cfbSMartin Varghese bool ethernet) 5860ed246ceeSJohn Hurley { 5861ed246ceeSJohn Hurley int err; 5862ed246ceeSJohn Hurley 5863ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5864dedc5a08SDavide Caratti return 0; 5865ed246ceeSJohn Hurley 5866fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5867ed246ceeSJohn Hurley if (unlikely(err)) 5868ed246ceeSJohn Hurley return err; 5869ed246ceeSJohn Hurley 5870ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5871ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5872fa4e0f88SDavide Caratti mac_len); 5873ed246ceeSJohn Hurley 5874ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5875ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5876fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5877ed246ceeSJohn Hurley 58784296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5879ed246ceeSJohn Hurley struct ethhdr *hdr; 5880ed246ceeSJohn Hurley 5881ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5882ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5883ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5884ed246ceeSJohn Hurley } 5885ed246ceeSJohn Hurley skb->protocol = next_proto; 5886ed246ceeSJohn Hurley 5887ed246ceeSJohn Hurley return 0; 5888ed246ceeSJohn Hurley } 5889ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5890ed246ceeSJohn Hurley 5891ed246ceeSJohn Hurley /** 5892d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5893d27cf5c5SJohn Hurley * 5894d27cf5c5SJohn Hurley * @skb: buffer 5895d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5896d27cf5c5SJohn Hurley * 5897d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5898d27cf5c5SJohn Hurley * 5899d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5900d27cf5c5SJohn Hurley */ 5901d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5902d27cf5c5SJohn Hurley { 5903d27cf5c5SJohn Hurley int err; 5904d27cf5c5SJohn Hurley 5905d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5906d27cf5c5SJohn Hurley return -EINVAL; 5907d27cf5c5SJohn Hurley 5908d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5909d27cf5c5SJohn Hurley if (unlikely(err)) 5910d27cf5c5SJohn Hurley return err; 5911d27cf5c5SJohn Hurley 5912d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5913d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5914d27cf5c5SJohn Hurley 5915d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5916d27cf5c5SJohn Hurley } 5917d27cf5c5SJohn Hurley 5918d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 5919d27cf5c5SJohn Hurley 5920d27cf5c5SJohn Hurley return 0; 5921d27cf5c5SJohn Hurley } 5922d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 5923d27cf5c5SJohn Hurley 5924d27cf5c5SJohn Hurley /** 59252a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 59262a2ea508SJohn Hurley * 59272a2ea508SJohn Hurley * @skb: buffer 59282a2ea508SJohn Hurley * 59292a2ea508SJohn Hurley * Expects skb->data at mac header. 59302a2ea508SJohn Hurley * 59312a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 59322a2ea508SJohn Hurley */ 59332a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 59342a2ea508SJohn Hurley { 59352a2ea508SJohn Hurley u32 lse; 59362a2ea508SJohn Hurley u8 ttl; 59372a2ea508SJohn Hurley 59382a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 59392a2ea508SJohn Hurley return -EINVAL; 59402a2ea508SJohn Hurley 594113de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 594213de4ed9SDavide Caratti return -ENOMEM; 594313de4ed9SDavide Caratti 59442a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 59452a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 59462a2ea508SJohn Hurley if (!--ttl) 59472a2ea508SJohn Hurley return -EINVAL; 59482a2ea508SJohn Hurley 59492a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 59502a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 59512a2ea508SJohn Hurley 59522a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 59532a2ea508SJohn Hurley } 59542a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 59552a2ea508SJohn Hurley 59562a2ea508SJohn Hurley /** 59572e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 59582e4e4410SEric Dumazet * 5959de3f0d0eSMasanari Iida * @header_len: size of linear part 5960de3f0d0eSMasanari Iida * @data_len: needed length in frags 5961de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 5962de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 5963de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 59642e4e4410SEric Dumazet * 59652e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 59662e4e4410SEric Dumazet */ 59672e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 59682e4e4410SEric Dumazet unsigned long data_len, 59692e4e4410SEric Dumazet int max_page_order, 59702e4e4410SEric Dumazet int *errcode, 59712e4e4410SEric Dumazet gfp_t gfp_mask) 59722e4e4410SEric Dumazet { 59732e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 59742e4e4410SEric Dumazet unsigned long chunk; 59752e4e4410SEric Dumazet struct sk_buff *skb; 59762e4e4410SEric Dumazet struct page *page; 59772e4e4410SEric Dumazet int i; 59782e4e4410SEric Dumazet 59792e4e4410SEric Dumazet *errcode = -EMSGSIZE; 59802e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 59812e4e4410SEric Dumazet * high order pages... 59822e4e4410SEric Dumazet */ 59832e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 59842e4e4410SEric Dumazet return NULL; 59852e4e4410SEric Dumazet 59862e4e4410SEric Dumazet *errcode = -ENOBUFS; 5987f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 59882e4e4410SEric Dumazet if (!skb) 59892e4e4410SEric Dumazet return NULL; 59902e4e4410SEric Dumazet 59912e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 59922e4e4410SEric Dumazet 59932e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 59942e4e4410SEric Dumazet int order = max_page_order; 59952e4e4410SEric Dumazet 59962e4e4410SEric Dumazet while (order) { 59972e4e4410SEric Dumazet if (npages >= 1 << order) { 5998d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 59992e4e4410SEric Dumazet __GFP_COMP | 6000d14b56f5SMichal Hocko __GFP_NOWARN, 60012e4e4410SEric Dumazet order); 60022e4e4410SEric Dumazet if (page) 60032e4e4410SEric Dumazet goto fill_page; 60042e4e4410SEric Dumazet /* Do not retry other high order allocations */ 60052e4e4410SEric Dumazet order = 1; 60062e4e4410SEric Dumazet max_page_order = 0; 60072e4e4410SEric Dumazet } 60082e4e4410SEric Dumazet order--; 60092e4e4410SEric Dumazet } 60102e4e4410SEric Dumazet page = alloc_page(gfp_mask); 60112e4e4410SEric Dumazet if (!page) 60122e4e4410SEric Dumazet goto failure; 60132e4e4410SEric Dumazet fill_page: 60142e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 60152e4e4410SEric Dumazet PAGE_SIZE << order); 60162e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 60172e4e4410SEric Dumazet data_len -= chunk; 60182e4e4410SEric Dumazet npages -= 1 << order; 60192e4e4410SEric Dumazet } 60202e4e4410SEric Dumazet return skb; 60212e4e4410SEric Dumazet 60222e4e4410SEric Dumazet failure: 60232e4e4410SEric Dumazet kfree_skb(skb); 60242e4e4410SEric Dumazet return NULL; 60252e4e4410SEric Dumazet } 60262e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 60276fa01ccdSSowmini Varadhan 60286fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 60296fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 60306fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 60316fa01ccdSSowmini Varadhan { 60326fa01ccdSSowmini Varadhan int i; 60336fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 60346fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 60356fa01ccdSSowmini Varadhan u8 *data; 60366fa01ccdSSowmini Varadhan 60376fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 60386fa01ccdSSowmini Varadhan 60396fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 60406fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 60416fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 60426fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 60436fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 60446fa01ccdSSowmini Varadhan if (!data) 60456fa01ccdSSowmini Varadhan return -ENOMEM; 60466fa01ccdSSowmini Varadhan 60476fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 60486fa01ccdSSowmini Varadhan 60496fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 60506fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 60516fa01ccdSSowmini Varadhan skb->len -= off; 60526fa01ccdSSowmini Varadhan 60536fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 60546fa01ccdSSowmini Varadhan skb_shinfo(skb), 60556fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 60566fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 60576fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 60586fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 60596fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 60606fa01ccdSSowmini Varadhan kfree(data); 60616fa01ccdSSowmini Varadhan return -ENOMEM; 60626fa01ccdSSowmini Varadhan } 60636fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 60646fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 60656fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 60666fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 60676fa01ccdSSowmini Varadhan skb_release_data(skb); 60686fa01ccdSSowmini Varadhan } else { 60696fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 60706fa01ccdSSowmini Varadhan * relocate values 60716fa01ccdSSowmini Varadhan */ 60726fa01ccdSSowmini Varadhan skb_free_head(skb); 60736fa01ccdSSowmini Varadhan } 60746fa01ccdSSowmini Varadhan 60756fa01ccdSSowmini Varadhan skb->head = data; 60766fa01ccdSSowmini Varadhan skb->data = data; 60776fa01ccdSSowmini Varadhan skb->head_frag = 0; 6078763087daSEric Dumazet skb_set_end_offset(skb, size); 60796fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 60806fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 60816fa01ccdSSowmini Varadhan skb->cloned = 0; 60826fa01ccdSSowmini Varadhan skb->hdr_len = 0; 60836fa01ccdSSowmini Varadhan skb->nohdr = 0; 60846fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 60856fa01ccdSSowmini Varadhan 60866fa01ccdSSowmini Varadhan return 0; 60876fa01ccdSSowmini Varadhan } 60886fa01ccdSSowmini Varadhan 60896fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 60906fa01ccdSSowmini Varadhan 60916fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 60926fa01ccdSSowmini Varadhan * pskb_carve() 60936fa01ccdSSowmini Varadhan */ 60946fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 60956fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 60966fa01ccdSSowmini Varadhan gfp_t gfp_mask) 60976fa01ccdSSowmini Varadhan { 60986fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 60996fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 61006fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 61016fa01ccdSSowmini Varadhan 61026fa01ccdSSowmini Varadhan do { 61036fa01ccdSSowmini Varadhan if (!list) { 61046fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 61056fa01ccdSSowmini Varadhan return -EFAULT; 61066fa01ccdSSowmini Varadhan } 61076fa01ccdSSowmini Varadhan if (list->len <= eat) { 61086fa01ccdSSowmini Varadhan /* Eaten as whole. */ 61096fa01ccdSSowmini Varadhan eat -= list->len; 61106fa01ccdSSowmini Varadhan list = list->next; 61116fa01ccdSSowmini Varadhan insp = list; 61126fa01ccdSSowmini Varadhan } else { 61136fa01ccdSSowmini Varadhan /* Eaten partially. */ 61146fa01ccdSSowmini Varadhan if (skb_shared(list)) { 61156fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 61166fa01ccdSSowmini Varadhan if (!clone) 61176fa01ccdSSowmini Varadhan return -ENOMEM; 61186fa01ccdSSowmini Varadhan insp = list->next; 61196fa01ccdSSowmini Varadhan list = clone; 61206fa01ccdSSowmini Varadhan } else { 61216fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 61226fa01ccdSSowmini Varadhan insp = list; 61236fa01ccdSSowmini Varadhan } 61246fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 61256fa01ccdSSowmini Varadhan kfree_skb(clone); 61266fa01ccdSSowmini Varadhan return -ENOMEM; 61276fa01ccdSSowmini Varadhan } 61286fa01ccdSSowmini Varadhan break; 61296fa01ccdSSowmini Varadhan } 61306fa01ccdSSowmini Varadhan } while (eat); 61316fa01ccdSSowmini Varadhan 61326fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 61336fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 61346fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 6135ef527f96SEric Dumazet consume_skb(list); 61366fa01ccdSSowmini Varadhan } 61376fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 61386fa01ccdSSowmini Varadhan if (clone) { 61396fa01ccdSSowmini Varadhan clone->next = list; 61406fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 61416fa01ccdSSowmini Varadhan } 61426fa01ccdSSowmini Varadhan return 0; 61436fa01ccdSSowmini Varadhan } 61446fa01ccdSSowmini Varadhan 61456fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 61466fa01ccdSSowmini Varadhan * non-linear part of skb 61476fa01ccdSSowmini Varadhan */ 61486fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 61496fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 61506fa01ccdSSowmini Varadhan { 61516fa01ccdSSowmini Varadhan int i, k = 0; 61526fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 61536fa01ccdSSowmini Varadhan u8 *data; 61546fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 61556fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 61566fa01ccdSSowmini Varadhan 61576fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 61586fa01ccdSSowmini Varadhan 61596fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 61606fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 61616fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 61626fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 61636fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 61646fa01ccdSSowmini Varadhan if (!data) 61656fa01ccdSSowmini Varadhan return -ENOMEM; 61666fa01ccdSSowmini Varadhan 61676fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 61686fa01ccdSSowmini Varadhan 61696fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6170e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 61716fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 61726fa01ccdSSowmini Varadhan kfree(data); 61736fa01ccdSSowmini Varadhan return -ENOMEM; 61746fa01ccdSSowmini Varadhan } 61756fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 61766fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 61776fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 61786fa01ccdSSowmini Varadhan 61796fa01ccdSSowmini Varadhan if (pos + fsize > off) { 61806fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 61816fa01ccdSSowmini Varadhan 61826fa01ccdSSowmini Varadhan if (pos < off) { 61836fa01ccdSSowmini Varadhan /* Split frag. 61846fa01ccdSSowmini Varadhan * We have two variants in this case: 61856fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 61866fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 61876fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 61886fa01ccdSSowmini Varadhan * where splitting is expensive. 61896fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 61906fa01ccdSSowmini Varadhan */ 6191b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 61926fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 61936fa01ccdSSowmini Varadhan } 61946fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 61956fa01ccdSSowmini Varadhan k++; 61966fa01ccdSSowmini Varadhan } 61976fa01ccdSSowmini Varadhan pos += fsize; 61986fa01ccdSSowmini Varadhan } 61996fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 62006fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 62016fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 62026fa01ccdSSowmini Varadhan 62036fa01ccdSSowmini Varadhan /* split line is in frag list */ 6204eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6205eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6206eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6207eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6208eabe8618SMiaohe Lin kfree(data); 6209eabe8618SMiaohe Lin return -ENOMEM; 62106fa01ccdSSowmini Varadhan } 62116fa01ccdSSowmini Varadhan skb_release_data(skb); 62126fa01ccdSSowmini Varadhan 62136fa01ccdSSowmini Varadhan skb->head = data; 62146fa01ccdSSowmini Varadhan skb->head_frag = 0; 62156fa01ccdSSowmini Varadhan skb->data = data; 6216763087daSEric Dumazet skb_set_end_offset(skb, size); 62176fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 62186fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 62196fa01ccdSSowmini Varadhan skb->cloned = 0; 62206fa01ccdSSowmini Varadhan skb->hdr_len = 0; 62216fa01ccdSSowmini Varadhan skb->nohdr = 0; 62226fa01ccdSSowmini Varadhan skb->len -= off; 62236fa01ccdSSowmini Varadhan skb->data_len = skb->len; 62246fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 62256fa01ccdSSowmini Varadhan return 0; 62266fa01ccdSSowmini Varadhan } 62276fa01ccdSSowmini Varadhan 62286fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 62296fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 62306fa01ccdSSowmini Varadhan { 62316fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 62326fa01ccdSSowmini Varadhan 62336fa01ccdSSowmini Varadhan if (len < headlen) 62346fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 62356fa01ccdSSowmini Varadhan else 62366fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 62376fa01ccdSSowmini Varadhan } 62386fa01ccdSSowmini Varadhan 62396fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 62406fa01ccdSSowmini Varadhan * a new skb 62416fa01ccdSSowmini Varadhan */ 62426fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 62436fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 62446fa01ccdSSowmini Varadhan { 62456fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 62466fa01ccdSSowmini Varadhan 62476fa01ccdSSowmini Varadhan if (!clone) 62486fa01ccdSSowmini Varadhan return NULL; 62496fa01ccdSSowmini Varadhan 62506fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 62516fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 62526fa01ccdSSowmini Varadhan kfree_skb(clone); 62536fa01ccdSSowmini Varadhan return NULL; 62546fa01ccdSSowmini Varadhan } 62556fa01ccdSSowmini Varadhan return clone; 62566fa01ccdSSowmini Varadhan } 62576fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6258c8c8b127SEric Dumazet 6259c8c8b127SEric Dumazet /** 6260c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6261c8c8b127SEric Dumazet * @skb: buffer 6262c8c8b127SEric Dumazet * 6263c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6264c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6265c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6266c8c8b127SEric Dumazet * truesize. 6267c8c8b127SEric Dumazet * Notes: 6268c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6269c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6270c8c8b127SEric Dumazet */ 6271c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6272c8c8b127SEric Dumazet { 62733174fed9SEric Dumazet if (skb->data_len) { 62743174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6275c8c8b127SEric Dumazet skb_cloned(skb)) 6276c8c8b127SEric Dumazet return; 6277c8c8b127SEric Dumazet 6278c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6279c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 62803174fed9SEric Dumazet } 62813174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 62823174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 62833174fed9SEric Dumazet * their truesize. 62843174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 62853174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 62863174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6287c8c8b127SEric Dumazet */ 6288c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6289c8c8b127SEric Dumazet } 6290df5042f4SFlorian Westphal 6291df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6292df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6293df5042f4SFlorian Westphal { 6294df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6295df5042f4SFlorian Westphal } 6296df5042f4SFlorian Westphal 62978b69a803SPaolo Abeni /** 62988b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 62998b69a803SPaolo Abeni * 63004930f483SFlorian Westphal * @flags: See kmalloc(). 63014930f483SFlorian Westphal * 63028b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 63038b69a803SPaolo Abeni * skb via __skb_ext_set(). 63048b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 63058b69a803SPaolo Abeni */ 63064930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6307df5042f4SFlorian Westphal { 63084930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6309df5042f4SFlorian Westphal 6310df5042f4SFlorian Westphal if (new) { 6311df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6312df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6313df5042f4SFlorian Westphal } 6314df5042f4SFlorian Westphal 6315df5042f4SFlorian Westphal return new; 6316df5042f4SFlorian Westphal } 6317df5042f4SFlorian Westphal 63184165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 63194165079bSFlorian Westphal unsigned int old_active) 6320df5042f4SFlorian Westphal { 6321df5042f4SFlorian Westphal struct skb_ext *new; 6322df5042f4SFlorian Westphal 6323df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6324df5042f4SFlorian Westphal return old; 6325df5042f4SFlorian Westphal 6326df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6327df5042f4SFlorian Westphal if (!new) 6328df5042f4SFlorian Westphal return NULL; 6329df5042f4SFlorian Westphal 6330df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6331df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6332df5042f4SFlorian Westphal 63334165079bSFlorian Westphal #ifdef CONFIG_XFRM 63344165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 63354165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 63364165079bSFlorian Westphal unsigned int i; 63374165079bSFlorian Westphal 63384165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 63394165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 63404165079bSFlorian Westphal } 63414165079bSFlorian Westphal #endif 6342df5042f4SFlorian Westphal __skb_ext_put(old); 6343df5042f4SFlorian Westphal return new; 6344df5042f4SFlorian Westphal } 6345df5042f4SFlorian Westphal 6346df5042f4SFlorian Westphal /** 63478b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 63488b69a803SPaolo Abeni * @skb: buffer 63498b69a803SPaolo Abeni * @id: extension id 63508b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 63518b69a803SPaolo Abeni * 63528b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 63538b69a803SPaolo Abeni * 63548b69a803SPaolo Abeni * Returns the pointer to the extension. 63558b69a803SPaolo Abeni */ 63568b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 63578b69a803SPaolo Abeni struct skb_ext *ext) 63588b69a803SPaolo Abeni { 63598b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 63608b69a803SPaolo Abeni 63618b69a803SPaolo Abeni skb_ext_put(skb); 63628b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 63638b69a803SPaolo Abeni ext->chunks = newlen; 63648b69a803SPaolo Abeni ext->offset[id] = newoff; 63658b69a803SPaolo Abeni skb->extensions = ext; 63668b69a803SPaolo Abeni skb->active_extensions = 1 << id; 63678b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 63688b69a803SPaolo Abeni } 63698b69a803SPaolo Abeni 63708b69a803SPaolo Abeni /** 6371df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6372df5042f4SFlorian Westphal * @skb: buffer 6373df5042f4SFlorian Westphal * @id: extension to allocate space for 6374df5042f4SFlorian Westphal * 6375df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6376df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6377df5042f4SFlorian Westphal * is returned. 6378df5042f4SFlorian Westphal * 6379df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6380df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6381df5042f4SFlorian Westphal * 6382df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6383df5042f4SFlorian Westphal */ 6384df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6385df5042f4SFlorian Westphal { 6386df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6387df5042f4SFlorian Westphal unsigned int newlen, newoff; 6388df5042f4SFlorian Westphal 6389df5042f4SFlorian Westphal if (skb->active_extensions) { 6390df5042f4SFlorian Westphal old = skb->extensions; 6391df5042f4SFlorian Westphal 63924165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6393df5042f4SFlorian Westphal if (!new) 6394df5042f4SFlorian Westphal return NULL; 6395df5042f4SFlorian Westphal 6396682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6397df5042f4SFlorian Westphal goto set_active; 6398df5042f4SFlorian Westphal 6399e94e50bdSPaolo Abeni newoff = new->chunks; 6400df5042f4SFlorian Westphal } else { 6401df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6402df5042f4SFlorian Westphal 64034930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6404df5042f4SFlorian Westphal if (!new) 6405df5042f4SFlorian Westphal return NULL; 6406df5042f4SFlorian Westphal } 6407df5042f4SFlorian Westphal 6408df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6409df5042f4SFlorian Westphal new->chunks = newlen; 6410df5042f4SFlorian Westphal new->offset[id] = newoff; 6411df5042f4SFlorian Westphal set_active: 6412b0999f38SPaolo Abeni skb->slow_gro = 1; 6413682ec859SPaolo Abeni skb->extensions = new; 6414df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6415df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6416df5042f4SFlorian Westphal } 6417df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6418df5042f4SFlorian Westphal 64194165079bSFlorian Westphal #ifdef CONFIG_XFRM 64204165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 64214165079bSFlorian Westphal { 64224165079bSFlorian Westphal unsigned int i; 64234165079bSFlorian Westphal 64244165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 64254165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 64264165079bSFlorian Westphal } 64274165079bSFlorian Westphal #endif 64284165079bSFlorian Westphal 642978476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 643078476d31SJeremy Kerr static void skb_ext_put_mctp(struct mctp_flow *flow) 643178476d31SJeremy Kerr { 643278476d31SJeremy Kerr if (flow->key) 643378476d31SJeremy Kerr mctp_key_unref(flow->key); 643478476d31SJeremy Kerr } 643578476d31SJeremy Kerr #endif 643678476d31SJeremy Kerr 6437df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6438df5042f4SFlorian Westphal { 6439df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6440df5042f4SFlorian Westphal 6441df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6442df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6443df5042f4SFlorian Westphal skb->extensions = NULL; 6444df5042f4SFlorian Westphal __skb_ext_put(ext); 64454165079bSFlorian Westphal #ifdef CONFIG_XFRM 64464165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 64474165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 64484165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 64494165079bSFlorian Westphal 64504165079bSFlorian Westphal skb_ext_put_sp(sp); 64514165079bSFlorian Westphal sp->len = 0; 64524165079bSFlorian Westphal #endif 6453df5042f4SFlorian Westphal } 6454df5042f4SFlorian Westphal } 6455df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6456df5042f4SFlorian Westphal 6457df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6458df5042f4SFlorian Westphal { 6459df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6460df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6461df5042f4SFlorian Westphal */ 6462df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6463df5042f4SFlorian Westphal goto free_now; 6464df5042f4SFlorian Westphal 6465df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6466df5042f4SFlorian Westphal return; 6467df5042f4SFlorian Westphal free_now: 64684165079bSFlorian Westphal #ifdef CONFIG_XFRM 64694165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 64704165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 64714165079bSFlorian Westphal #endif 647278476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 647378476d31SJeremy Kerr if (__skb_ext_exist(ext, SKB_EXT_MCTP)) 647478476d31SJeremy Kerr skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP)); 647578476d31SJeremy Kerr #endif 64764165079bSFlorian Westphal 6477df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6478df5042f4SFlorian Westphal } 6479df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6480df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 6481