12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 41da177e4SLinus Torvalds * 5113aa838SAlan Cox * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 61da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Fixes: 91da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 101da177e4SLinus Torvalds * balancer bugs. 111da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 121da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 131da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 141da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 151da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 161da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 171da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 181da177e4SLinus Torvalds * only put in the headers 191da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 201da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 211da177e4SLinus Torvalds * Andi Kleen : slabified it. 221da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 231da177e4SLinus Torvalds * 241da177e4SLinus Torvalds * NOTE: 251da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 261da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 271da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 281da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 331da177e4SLinus Torvalds */ 341da177e4SLinus Torvalds 35e005d193SJoe Perches #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 36e005d193SJoe Perches 371da177e4SLinus Torvalds #include <linux/module.h> 381da177e4SLinus Torvalds #include <linux/types.h> 391da177e4SLinus Torvalds #include <linux/kernel.h> 401da177e4SLinus Torvalds #include <linux/mm.h> 411da177e4SLinus Torvalds #include <linux/interrupt.h> 421da177e4SLinus Torvalds #include <linux/in.h> 431da177e4SLinus Torvalds #include <linux/inet.h> 441da177e4SLinus Torvalds #include <linux/slab.h> 45de960aa9SFlorian Westphal #include <linux/tcp.h> 46de960aa9SFlorian Westphal #include <linux/udp.h> 4790017accSMarcelo Ricardo Leitner #include <linux/sctp.h> 481da177e4SLinus Torvalds #include <linux/netdevice.h> 491da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 501da177e4SLinus Torvalds #include <net/pkt_sched.h> 511da177e4SLinus Torvalds #endif 521da177e4SLinus Torvalds #include <linux/string.h> 531da177e4SLinus Torvalds #include <linux/skbuff.h> 549c55e01cSJens Axboe #include <linux/splice.h> 551da177e4SLinus Torvalds #include <linux/cache.h> 561da177e4SLinus Torvalds #include <linux/rtnetlink.h> 571da177e4SLinus Torvalds #include <linux/init.h> 58716ea3a7SDavid Howells #include <linux/scatterlist.h> 59ac45f602SPatrick Ohly #include <linux/errqueue.h> 60268bb0ceSLinus Torvalds #include <linux/prefetch.h> 610d5501c1SVlad Yasevich #include <linux/if_vlan.h> 622a2ea508SJohn Hurley #include <linux/mpls.h> 63183f47fcSSebastian Andrzej Siewior #include <linux/kcov.h> 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds #include <net/protocol.h> 661da177e4SLinus Torvalds #include <net/dst.h> 671da177e4SLinus Torvalds #include <net/sock.h> 681da177e4SLinus Torvalds #include <net/checksum.h> 69ed1f50c3SPaul Durrant #include <net/ip6_checksum.h> 701da177e4SLinus Torvalds #include <net/xfrm.h> 718822e270SJohn Hurley #include <net/mpls.h> 723ee17bc7SMat Martineau #include <net/mptcp.h> 7378476d31SJeremy Kerr #include <net/mctp.h> 746a5bcd84SIlias Apalodimas #include <net/page_pool.h> 751da177e4SLinus Torvalds 767c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 77ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7851c56b00SEric Dumazet #include <linux/highmem.h> 79b245be1fSWillem de Bruijn #include <linux/capability.h> 80b245be1fSWillem de Bruijn #include <linux/user_namespace.h> 812544af03SMatteo Croce #include <linux/indirect_call_wrapper.h> 82a1f8e7f7SAl Viro 8339564c3fSEric Dumazet #include "dev.h" 847f678defSVasily Averin #include "sock_destructor.h" 857b7ed885SBart Van Assche 8608009a76SAlexey Dobriyan struct kmem_cache *skbuff_head_cache __ro_after_init; 8708009a76SAlexey Dobriyan static struct kmem_cache *skbuff_fclone_cache __ro_after_init; 88df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 89df5042f4SFlorian Westphal static struct kmem_cache *skbuff_ext_cache __ro_after_init; 90df5042f4SFlorian Westphal #endif 915f74f82eSHans Westgaard Ry int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 925f74f82eSHans Westgaard Ry EXPORT_SYMBOL(sysctl_max_skb_frags); 931da177e4SLinus Torvalds 949cb252c4SMenglong Dong #undef FN 959cb252c4SMenglong Dong #define FN(reason) [SKB_DROP_REASON_##reason] = #reason, 969cb252c4SMenglong Dong const char * const drop_reasons[] = { 979cb252c4SMenglong Dong DEFINE_DROP_REASON(FN, FN) 989cb252c4SMenglong Dong }; 99ec43908dSMenglong Dong EXPORT_SYMBOL(drop_reasons); 100ec43908dSMenglong Dong 1011da177e4SLinus Torvalds /** 102f05de73bSJean Sacren * skb_panic - private function for out-of-line support 1031da177e4SLinus Torvalds * @skb: buffer 1041da177e4SLinus Torvalds * @sz: size 105f05de73bSJean Sacren * @addr: address 10699d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 1071da177e4SLinus Torvalds * 108f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 109f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 110f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 111f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1121da177e4SLinus Torvalds */ 113f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 11499d5851eSJames Hogan const char msg[]) 1151da177e4SLinus Torvalds { 11641a46913SJesper Dangaard Brouer pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", 11799d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1184305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 11926095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1201da177e4SLinus Torvalds BUG(); 1211da177e4SLinus Torvalds } 1221da177e4SLinus Torvalds 123f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1241da177e4SLinus Torvalds { 125f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1261da177e4SLinus Torvalds } 1271da177e4SLinus Torvalds 128f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 129f05de73bSJean Sacren { 130f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 131f05de73bSJean Sacren } 132c93bdd0eSMel Gorman 13350fad4b5SAlexander Lobakin #define NAPI_SKB_CACHE_SIZE 64 134f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_BULK 16 135f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) 13650fad4b5SAlexander Lobakin 13750fad4b5SAlexander Lobakin struct napi_alloc_cache { 13850fad4b5SAlexander Lobakin struct page_frag_cache page; 13950fad4b5SAlexander Lobakin unsigned int skb_count; 14050fad4b5SAlexander Lobakin void *skb_cache[NAPI_SKB_CACHE_SIZE]; 14150fad4b5SAlexander Lobakin }; 14250fad4b5SAlexander Lobakin 14350fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 14450fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 14550fad4b5SAlexander Lobakin 14632e3573fSYajun Deng void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 14750fad4b5SAlexander Lobakin { 14850fad4b5SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 14950fad4b5SAlexander Lobakin 15050fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 15150fad4b5SAlexander Lobakin 15232e3573fSYajun Deng return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 15350fad4b5SAlexander Lobakin } 15450fad4b5SAlexander Lobakin EXPORT_SYMBOL(__napi_alloc_frag_align); 15550fad4b5SAlexander Lobakin 15650fad4b5SAlexander Lobakin void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 15750fad4b5SAlexander Lobakin { 15850fad4b5SAlexander Lobakin void *data; 15950fad4b5SAlexander Lobakin 16050fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 161afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 16232e3573fSYajun Deng struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache); 16332e3573fSYajun Deng 16450fad4b5SAlexander Lobakin data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask); 16550fad4b5SAlexander Lobakin } else { 16632e3573fSYajun Deng struct napi_alloc_cache *nc; 16732e3573fSYajun Deng 16850fad4b5SAlexander Lobakin local_bh_disable(); 16932e3573fSYajun Deng nc = this_cpu_ptr(&napi_alloc_cache); 17032e3573fSYajun Deng data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 17150fad4b5SAlexander Lobakin local_bh_enable(); 17250fad4b5SAlexander Lobakin } 17350fad4b5SAlexander Lobakin return data; 17450fad4b5SAlexander Lobakin } 17550fad4b5SAlexander Lobakin EXPORT_SYMBOL(__netdev_alloc_frag_align); 17650fad4b5SAlexander Lobakin 177f450d539SAlexander Lobakin static struct sk_buff *napi_skb_cache_get(void) 178f450d539SAlexander Lobakin { 179f450d539SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 180f450d539SAlexander Lobakin struct sk_buff *skb; 181f450d539SAlexander Lobakin 18249ae83fcSSieng Piaw Liew if (unlikely(!nc->skb_count)) { 183f450d539SAlexander Lobakin nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache, 184f450d539SAlexander Lobakin GFP_ATOMIC, 185f450d539SAlexander Lobakin NAPI_SKB_CACHE_BULK, 186f450d539SAlexander Lobakin nc->skb_cache); 187f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 188f450d539SAlexander Lobakin return NULL; 18949ae83fcSSieng Piaw Liew } 190f450d539SAlexander Lobakin 191f450d539SAlexander Lobakin skb = nc->skb_cache[--nc->skb_count]; 192f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, skb); 193f450d539SAlexander Lobakin 194f450d539SAlexander Lobakin return skb; 195f450d539SAlexander Lobakin } 196f450d539SAlexander Lobakin 197ba0509b6SJesper Dangaard Brouer /* Caller must provide SKB that is memset cleared */ 198483126b3SAlexander Lobakin static void __build_skb_around(struct sk_buff *skb, void *data, 199483126b3SAlexander Lobakin unsigned int frag_size) 200ba0509b6SJesper Dangaard Brouer { 201ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 202ba0509b6SJesper Dangaard Brouer unsigned int size = frag_size ? : ksize(data); 203ba0509b6SJesper Dangaard Brouer 204ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 205ba0509b6SJesper Dangaard Brouer 206ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 207ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 208ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 209ba0509b6SJesper Dangaard Brouer skb->head = data; 210ba0509b6SJesper Dangaard Brouer skb->data = data; 211ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 212763087daSEric Dumazet skb_set_end_offset(skb, size); 213ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 214ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 21568822bdfSEric Dumazet skb->alloc_cpu = raw_smp_processor_id(); 216ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 217ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 218ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 219ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 220ba0509b6SJesper Dangaard Brouer 2216370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 222ba0509b6SJesper Dangaard Brouer } 223ba0509b6SJesper Dangaard Brouer 2241da177e4SLinus Torvalds /** 2252ea2f62cSEric Dumazet * __build_skb - build a network buffer 226b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 2272ea2f62cSEric Dumazet * @frag_size: size of data, or 0 if head was kmalloced 228b2b5ce9dSEric Dumazet * 229b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 230deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 2312ea2f62cSEric Dumazet * @frag_size is 0, otherwise data should come from the page allocator 2322ea2f62cSEric Dumazet * or vmalloc() 233b2b5ce9dSEric Dumazet * The return is the new skb buffer. 234b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 235b2b5ce9dSEric Dumazet * Notes : 236b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 237b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 238b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 239b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 240b2b5ce9dSEric Dumazet * before giving packet to stack. 241b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 242b2b5ce9dSEric Dumazet */ 2432ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 244b2b5ce9dSEric Dumazet { 245b2b5ce9dSEric Dumazet struct sk_buff *skb; 246b2b5ce9dSEric Dumazet 247b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 248ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 249b2b5ce9dSEric Dumazet return NULL; 250b2b5ce9dSEric Dumazet 251b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 252483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 253b2b5ce9dSEric Dumazet 254483126b3SAlexander Lobakin return skb; 255b2b5ce9dSEric Dumazet } 2562ea2f62cSEric Dumazet 2572ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 2582ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 2592ea2f62cSEric Dumazet * This means that if @frag_size is not zero, then @data must be backed 2602ea2f62cSEric Dumazet * by a page fragment, not kmalloc() or vmalloc() 2612ea2f62cSEric Dumazet */ 2622ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 2632ea2f62cSEric Dumazet { 2642ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 2652ea2f62cSEric Dumazet 2662ea2f62cSEric Dumazet if (skb && frag_size) { 2672ea2f62cSEric Dumazet skb->head_frag = 1; 2682f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 2692ea2f62cSEric Dumazet skb->pfmemalloc = 1; 2702ea2f62cSEric Dumazet } 2712ea2f62cSEric Dumazet return skb; 2722ea2f62cSEric Dumazet } 273b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 274b2b5ce9dSEric Dumazet 275ba0509b6SJesper Dangaard Brouer /** 276ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 277ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 278ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 279ba0509b6SJesper Dangaard Brouer * @frag_size: size of data, or 0 if head was kmalloced 280ba0509b6SJesper Dangaard Brouer */ 281ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 282ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 283ba0509b6SJesper Dangaard Brouer { 284ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 285ba0509b6SJesper Dangaard Brouer return NULL; 286ba0509b6SJesper Dangaard Brouer 287483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 288ba0509b6SJesper Dangaard Brouer 289483126b3SAlexander Lobakin if (frag_size) { 290ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 291ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 292ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 293ba0509b6SJesper Dangaard Brouer } 294ba0509b6SJesper Dangaard Brouer return skb; 295ba0509b6SJesper Dangaard Brouer } 296ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 297ba0509b6SJesper Dangaard Brouer 298f450d539SAlexander Lobakin /** 299f450d539SAlexander Lobakin * __napi_build_skb - build a network buffer 300f450d539SAlexander Lobakin * @data: data buffer provided by caller 301f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 302f450d539SAlexander Lobakin * 303f450d539SAlexander Lobakin * Version of __build_skb() that uses NAPI percpu caches to obtain 304f450d539SAlexander Lobakin * skbuff_head instead of inplace allocation. 305f450d539SAlexander Lobakin * 306f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 307f450d539SAlexander Lobakin */ 308f450d539SAlexander Lobakin static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size) 309f450d539SAlexander Lobakin { 310f450d539SAlexander Lobakin struct sk_buff *skb; 311f450d539SAlexander Lobakin 312f450d539SAlexander Lobakin skb = napi_skb_cache_get(); 313f450d539SAlexander Lobakin if (unlikely(!skb)) 314f450d539SAlexander Lobakin return NULL; 315f450d539SAlexander Lobakin 316f450d539SAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 317f450d539SAlexander Lobakin __build_skb_around(skb, data, frag_size); 318f450d539SAlexander Lobakin 319f450d539SAlexander Lobakin return skb; 320f450d539SAlexander Lobakin } 321f450d539SAlexander Lobakin 322f450d539SAlexander Lobakin /** 323f450d539SAlexander Lobakin * napi_build_skb - build a network buffer 324f450d539SAlexander Lobakin * @data: data buffer provided by caller 325f450d539SAlexander Lobakin * @frag_size: size of data, or 0 if head was kmalloced 326f450d539SAlexander Lobakin * 327f450d539SAlexander Lobakin * Version of __napi_build_skb() that takes care of skb->head_frag 328f450d539SAlexander Lobakin * and skb->pfmemalloc when the data is a page or page fragment. 329f450d539SAlexander Lobakin * 330f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 331f450d539SAlexander Lobakin */ 332f450d539SAlexander Lobakin struct sk_buff *napi_build_skb(void *data, unsigned int frag_size) 333f450d539SAlexander Lobakin { 334f450d539SAlexander Lobakin struct sk_buff *skb = __napi_build_skb(data, frag_size); 335f450d539SAlexander Lobakin 336f450d539SAlexander Lobakin if (likely(skb) && frag_size) { 337f450d539SAlexander Lobakin skb->head_frag = 1; 338f450d539SAlexander Lobakin skb_propagate_pfmemalloc(virt_to_head_page(data), skb); 339f450d539SAlexander Lobakin } 340f450d539SAlexander Lobakin 341f450d539SAlexander Lobakin return skb; 342f450d539SAlexander Lobakin } 343f450d539SAlexander Lobakin EXPORT_SYMBOL(napi_build_skb); 344f450d539SAlexander Lobakin 3455381b23dSAlexander Lobakin /* 3465381b23dSAlexander Lobakin * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 3475381b23dSAlexander Lobakin * the caller if emergency pfmemalloc reserves are being used. If it is and 3485381b23dSAlexander Lobakin * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 3495381b23dSAlexander Lobakin * may be used. Otherwise, the packet data may be discarded until enough 3505381b23dSAlexander Lobakin * memory is free 3515381b23dSAlexander Lobakin */ 352ef28095fSAlexander Lobakin static void *kmalloc_reserve(size_t size, gfp_t flags, int node, 353ef28095fSAlexander Lobakin bool *pfmemalloc) 3545381b23dSAlexander Lobakin { 3555381b23dSAlexander Lobakin void *obj; 3565381b23dSAlexander Lobakin bool ret_pfmemalloc = false; 3575381b23dSAlexander Lobakin 3585381b23dSAlexander Lobakin /* 3595381b23dSAlexander Lobakin * Try a regular allocation, when that fails and we're not entitled 3605381b23dSAlexander Lobakin * to the reserves, fail. 3615381b23dSAlexander Lobakin */ 3625381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, 3635381b23dSAlexander Lobakin flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 3645381b23dSAlexander Lobakin node); 3655381b23dSAlexander Lobakin if (obj || !(gfp_pfmemalloc_allowed(flags))) 3665381b23dSAlexander Lobakin goto out; 3675381b23dSAlexander Lobakin 3685381b23dSAlexander Lobakin /* Try again but now we are using pfmemalloc reserves */ 3695381b23dSAlexander Lobakin ret_pfmemalloc = true; 3705381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, flags, node); 3715381b23dSAlexander Lobakin 3725381b23dSAlexander Lobakin out: 3735381b23dSAlexander Lobakin if (pfmemalloc) 3745381b23dSAlexander Lobakin *pfmemalloc = ret_pfmemalloc; 3755381b23dSAlexander Lobakin 3765381b23dSAlexander Lobakin return obj; 3775381b23dSAlexander Lobakin } 3785381b23dSAlexander Lobakin 3795381b23dSAlexander Lobakin /* Allocate a new skbuff. We do this ourselves so we can fill in a few 3805381b23dSAlexander Lobakin * 'private' fields and also do memory statistics to find all the 3815381b23dSAlexander Lobakin * [BEEP] leaks. 3825381b23dSAlexander Lobakin * 3835381b23dSAlexander Lobakin */ 3845381b23dSAlexander Lobakin 3855381b23dSAlexander Lobakin /** 3865381b23dSAlexander Lobakin * __alloc_skb - allocate a network buffer 3875381b23dSAlexander Lobakin * @size: size to allocate 3885381b23dSAlexander Lobakin * @gfp_mask: allocation mask 3895381b23dSAlexander Lobakin * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 3905381b23dSAlexander Lobakin * instead of head cache and allocate a cloned (child) skb. 3915381b23dSAlexander Lobakin * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 3925381b23dSAlexander Lobakin * allocations in case the data is required for writeback 3935381b23dSAlexander Lobakin * @node: numa node to allocate memory on 3945381b23dSAlexander Lobakin * 3955381b23dSAlexander Lobakin * Allocate a new &sk_buff. The returned buffer has no headroom and a 3965381b23dSAlexander Lobakin * tail room of at least size bytes. The object has a reference count 3975381b23dSAlexander Lobakin * of one. The return is the buffer. On a failure the return is %NULL. 3985381b23dSAlexander Lobakin * 3995381b23dSAlexander Lobakin * Buffers may only be allocated from interrupts using a @gfp_mask of 4005381b23dSAlexander Lobakin * %GFP_ATOMIC. 4015381b23dSAlexander Lobakin */ 4025381b23dSAlexander Lobakin struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 4035381b23dSAlexander Lobakin int flags, int node) 4045381b23dSAlexander Lobakin { 4055381b23dSAlexander Lobakin struct kmem_cache *cache; 4065381b23dSAlexander Lobakin struct sk_buff *skb; 407a5df6333SLi RongQing unsigned int osize; 4085381b23dSAlexander Lobakin bool pfmemalloc; 409a5df6333SLi RongQing u8 *data; 4105381b23dSAlexander Lobakin 4115381b23dSAlexander Lobakin cache = (flags & SKB_ALLOC_FCLONE) 4125381b23dSAlexander Lobakin ? skbuff_fclone_cache : skbuff_head_cache; 4135381b23dSAlexander Lobakin 4145381b23dSAlexander Lobakin if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 4155381b23dSAlexander Lobakin gfp_mask |= __GFP_MEMALLOC; 4165381b23dSAlexander Lobakin 4175381b23dSAlexander Lobakin /* Get the HEAD */ 418d13612b5SAlexander Lobakin if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI && 419d13612b5SAlexander Lobakin likely(node == NUMA_NO_NODE || node == numa_mem_id())) 420d13612b5SAlexander Lobakin skb = napi_skb_cache_get(); 421d13612b5SAlexander Lobakin else 422d13612b5SAlexander Lobakin skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node); 423df1ae022SAlexander Lobakin if (unlikely(!skb)) 424df1ae022SAlexander Lobakin return NULL; 4255381b23dSAlexander Lobakin prefetchw(skb); 4265381b23dSAlexander Lobakin 4275381b23dSAlexander Lobakin /* We do our best to align skb_shared_info on a separate cache 4285381b23dSAlexander Lobakin * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 4295381b23dSAlexander Lobakin * aligned memory blocks, unless SLUB/SLAB debug is enabled. 4305381b23dSAlexander Lobakin * Both skb->head and skb_shared_info are cache line aligned. 4315381b23dSAlexander Lobakin */ 4325381b23dSAlexander Lobakin size = SKB_DATA_ALIGN(size); 4335381b23dSAlexander Lobakin size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4345381b23dSAlexander Lobakin data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 435df1ae022SAlexander Lobakin if (unlikely(!data)) 4365381b23dSAlexander Lobakin goto nodata; 4375381b23dSAlexander Lobakin /* kmalloc(size) might give us more room than requested. 4385381b23dSAlexander Lobakin * Put skb_shared_info exactly at the end of allocated zone, 4395381b23dSAlexander Lobakin * to allow max possible filling before reallocation. 4405381b23dSAlexander Lobakin */ 441a5df6333SLi RongQing osize = ksize(data); 442a5df6333SLi RongQing size = SKB_WITH_OVERHEAD(osize); 4435381b23dSAlexander Lobakin prefetchw(data + size); 4445381b23dSAlexander Lobakin 4455381b23dSAlexander Lobakin /* 4465381b23dSAlexander Lobakin * Only clear those fields we need to clear, not those that we will 4475381b23dSAlexander Lobakin * actually initialise below. Hence, don't put any more fields after 4485381b23dSAlexander Lobakin * the tail pointer in struct sk_buff! 4495381b23dSAlexander Lobakin */ 4505381b23dSAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 451a5df6333SLi RongQing __build_skb_around(skb, data, osize); 4525381b23dSAlexander Lobakin skb->pfmemalloc = pfmemalloc; 4535381b23dSAlexander Lobakin 4545381b23dSAlexander Lobakin if (flags & SKB_ALLOC_FCLONE) { 4555381b23dSAlexander Lobakin struct sk_buff_fclones *fclones; 4565381b23dSAlexander Lobakin 4575381b23dSAlexander Lobakin fclones = container_of(skb, struct sk_buff_fclones, skb1); 4585381b23dSAlexander Lobakin 4595381b23dSAlexander Lobakin skb->fclone = SKB_FCLONE_ORIG; 4605381b23dSAlexander Lobakin refcount_set(&fclones->fclone_ref, 1); 4615381b23dSAlexander Lobakin } 4625381b23dSAlexander Lobakin 4635381b23dSAlexander Lobakin return skb; 464df1ae022SAlexander Lobakin 4655381b23dSAlexander Lobakin nodata: 4665381b23dSAlexander Lobakin kmem_cache_free(cache, skb); 467df1ae022SAlexander Lobakin return NULL; 4685381b23dSAlexander Lobakin } 4695381b23dSAlexander Lobakin EXPORT_SYMBOL(__alloc_skb); 4705381b23dSAlexander Lobakin 4717ba7aeabSSebastian Andrzej Siewior /** 472fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 473fd11a83dSAlexander Duyck * @dev: network device to receive on 474d7499160SMasanari Iida * @len: length to allocate 475fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 476fd11a83dSAlexander Duyck * 477fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 478fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 479fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 480fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 481fd11a83dSAlexander Duyck * 482fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 483fd11a83dSAlexander Duyck */ 4849451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 4859451980aSAlexander Duyck gfp_t gfp_mask) 486fd11a83dSAlexander Duyck { 487b63ae8caSAlexander Duyck struct page_frag_cache *nc; 488fd11a83dSAlexander Duyck struct sk_buff *skb; 4899451980aSAlexander Duyck bool pfmemalloc; 4909451980aSAlexander Duyck void *data; 491fd11a83dSAlexander Duyck 4929451980aSAlexander Duyck len += NET_SKB_PAD; 493fd11a83dSAlexander Duyck 49466c55602SAlexander Lobakin /* If requested length is either too small or too big, 49566c55602SAlexander Lobakin * we use kmalloc() for skb->head allocation. 49666c55602SAlexander Lobakin */ 49766c55602SAlexander Lobakin if (len <= SKB_WITH_OVERHEAD(1024) || 49866c55602SAlexander Lobakin len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 499d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 500a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 501a080e7bdSAlexander Duyck if (!skb) 502a080e7bdSAlexander Duyck goto skb_fail; 503a080e7bdSAlexander Duyck goto skb_success; 504a080e7bdSAlexander Duyck } 5059451980aSAlexander Duyck 5069451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5079451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5089451980aSAlexander Duyck 5099451980aSAlexander Duyck if (sk_memalloc_socks()) 5109451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5119451980aSAlexander Duyck 512afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 5139451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 5148c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 5159451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 51692dcabd7SSebastian Andrzej Siewior } else { 51792dcabd7SSebastian Andrzej Siewior local_bh_disable(); 51892dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 51992dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 52092dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 52192dcabd7SSebastian Andrzej Siewior local_bh_enable(); 52292dcabd7SSebastian Andrzej Siewior } 5239451980aSAlexander Duyck 5249451980aSAlexander Duyck if (unlikely(!data)) 5259451980aSAlexander Duyck return NULL; 5269451980aSAlexander Duyck 5279451980aSAlexander Duyck skb = __build_skb(data, len); 5289451980aSAlexander Duyck if (unlikely(!skb)) { 529181edb2bSAlexander Duyck skb_free_frag(data); 5309451980aSAlexander Duyck return NULL; 5319451980aSAlexander Duyck } 5329451980aSAlexander Duyck 5339451980aSAlexander Duyck if (pfmemalloc) 5349451980aSAlexander Duyck skb->pfmemalloc = 1; 5359451980aSAlexander Duyck skb->head_frag = 1; 5369451980aSAlexander Duyck 537a080e7bdSAlexander Duyck skb_success: 5388af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 5397b2e497aSChristoph Hellwig skb->dev = dev; 540fd11a83dSAlexander Duyck 541a080e7bdSAlexander Duyck skb_fail: 5428af27456SChristoph Hellwig return skb; 5438af27456SChristoph Hellwig } 544b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 5451da177e4SLinus Torvalds 546fd11a83dSAlexander Duyck /** 547fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 548fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 549d7499160SMasanari Iida * @len: length to allocate 550fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 551fd11a83dSAlexander Duyck * 552fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 553fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 554fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 555fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 556fd11a83dSAlexander Duyck * 557fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 558fd11a83dSAlexander Duyck */ 5599451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 5609451980aSAlexander Duyck gfp_t gfp_mask) 561fd11a83dSAlexander Duyck { 5623226b158SEric Dumazet struct napi_alloc_cache *nc; 563fd11a83dSAlexander Duyck struct sk_buff *skb; 5649451980aSAlexander Duyck void *data; 565fd11a83dSAlexander Duyck 566ee2640dfSEric Dumazet DEBUG_NET_WARN_ON_ONCE(!in_softirq()); 5679451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 568fd11a83dSAlexander Duyck 5693226b158SEric Dumazet /* If requested length is either too small or too big, 5703226b158SEric Dumazet * we use kmalloc() for skb->head allocation. 5713226b158SEric Dumazet */ 5723226b158SEric Dumazet if (len <= SKB_WITH_OVERHEAD(1024) || 5733226b158SEric Dumazet len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 574d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 575cfb8ec65SAlexander Lobakin skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, 576cfb8ec65SAlexander Lobakin NUMA_NO_NODE); 577a080e7bdSAlexander Duyck if (!skb) 578a080e7bdSAlexander Duyck goto skb_fail; 579a080e7bdSAlexander Duyck goto skb_success; 580a080e7bdSAlexander Duyck } 5819451980aSAlexander Duyck 5823226b158SEric Dumazet nc = this_cpu_ptr(&napi_alloc_cache); 5839451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 5849451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 5859451980aSAlexander Duyck 5869451980aSAlexander Duyck if (sk_memalloc_socks()) 5879451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 5889451980aSAlexander Duyck 5898c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 5909451980aSAlexander Duyck if (unlikely(!data)) 5919451980aSAlexander Duyck return NULL; 5929451980aSAlexander Duyck 593cfb8ec65SAlexander Lobakin skb = __napi_build_skb(data, len); 5949451980aSAlexander Duyck if (unlikely(!skb)) { 595181edb2bSAlexander Duyck skb_free_frag(data); 5969451980aSAlexander Duyck return NULL; 5979451980aSAlexander Duyck } 5989451980aSAlexander Duyck 599795bb1c0SJesper Dangaard Brouer if (nc->page.pfmemalloc) 6009451980aSAlexander Duyck skb->pfmemalloc = 1; 6019451980aSAlexander Duyck skb->head_frag = 1; 6029451980aSAlexander Duyck 603a080e7bdSAlexander Duyck skb_success: 604fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 605fd11a83dSAlexander Duyck skb->dev = napi->dev; 606fd11a83dSAlexander Duyck 607a080e7bdSAlexander Duyck skb_fail: 608fd11a83dSAlexander Duyck return skb; 609fd11a83dSAlexander Duyck } 610fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 611fd11a83dSAlexander Duyck 612654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 61350269e19SEric Dumazet int size, unsigned int truesize) 614654bed16SPeter Zijlstra { 615654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 616654bed16SPeter Zijlstra skb->len += size; 617654bed16SPeter Zijlstra skb->data_len += size; 61850269e19SEric Dumazet skb->truesize += truesize; 619654bed16SPeter Zijlstra } 620654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 621654bed16SPeter Zijlstra 622f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 623f8e617e1SJason Wang unsigned int truesize) 624f8e617e1SJason Wang { 625f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 626f8e617e1SJason Wang 627f8e617e1SJason Wang skb_frag_size_add(frag, size); 628f8e617e1SJason Wang skb->len += size; 629f8e617e1SJason Wang skb->data_len += size; 630f8e617e1SJason Wang skb->truesize += truesize; 631f8e617e1SJason Wang } 632f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 633f8e617e1SJason Wang 63427b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 6351da177e4SLinus Torvalds { 636bd8a7036SEric Dumazet kfree_skb_list(*listp); 63727b437c8SHerbert Xu *listp = NULL; 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds 64027b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 64127b437c8SHerbert Xu { 64227b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 64327b437c8SHerbert Xu } 64427b437c8SHerbert Xu 6451da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 6461da177e4SLinus Torvalds { 6471da177e4SLinus Torvalds struct sk_buff *list; 6481da177e4SLinus Torvalds 649fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 6501da177e4SLinus Torvalds skb_get(list); 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds 653d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 654d3836f21SEric Dumazet { 655181edb2bSAlexander Duyck unsigned char *head = skb->head; 656181edb2bSAlexander Duyck 6576a5bcd84SIlias Apalodimas if (skb->head_frag) { 6586a5bcd84SIlias Apalodimas if (skb_pp_recycle(skb, head)) 6596a5bcd84SIlias Apalodimas return; 660181edb2bSAlexander Duyck skb_free_frag(head); 6616a5bcd84SIlias Apalodimas } else { 662181edb2bSAlexander Duyck kfree(head); 663d3836f21SEric Dumazet } 6646a5bcd84SIlias Apalodimas } 665d3836f21SEric Dumazet 6665bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 6671da177e4SLinus Torvalds { 668ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 6691da177e4SLinus Torvalds int i; 670ff04a771SEric Dumazet 671ff04a771SEric Dumazet if (skb->cloned && 672ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 673ff04a771SEric Dumazet &shinfo->dataref)) 6742cc3aeb5SIlias Apalodimas goto exit; 675ff04a771SEric Dumazet 676753f1ca4SPavel Begunkov if (skb_zcopy(skb)) { 677753f1ca4SPavel Begunkov bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS; 678753f1ca4SPavel Begunkov 67970c43167SJonathan Lemon skb_zcopy_clear(skb, true); 680753f1ca4SPavel Begunkov if (skip_unref) 681753f1ca4SPavel Begunkov goto free_head; 682753f1ca4SPavel Begunkov } 68370c43167SJonathan Lemon 684ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 6856a5bcd84SIlias Apalodimas __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); 6861da177e4SLinus Torvalds 687753f1ca4SPavel Begunkov free_head: 688ff04a771SEric Dumazet if (shinfo->frag_list) 689ff04a771SEric Dumazet kfree_skb_list(shinfo->frag_list); 6901da177e4SLinus Torvalds 691d3836f21SEric Dumazet skb_free_head(skb); 6922cc3aeb5SIlias Apalodimas exit: 6932cc3aeb5SIlias Apalodimas /* When we clone an SKB we copy the reycling bit. The pp_recycle 6942cc3aeb5SIlias Apalodimas * bit is only set on the head though, so in order to avoid races 6952cc3aeb5SIlias Apalodimas * while trying to recycle fragments on __skb_frag_unref() we need 6962cc3aeb5SIlias Apalodimas * to make one SKB responsible for triggering the recycle path. 6972cc3aeb5SIlias Apalodimas * So disable the recycling bit if an SKB is cloned and we have 69858e61e41STom Rix * additional references to the fragmented part of the SKB. 6992cc3aeb5SIlias Apalodimas * Eventually the last SKB will have the recycling bit set and it's 7002cc3aeb5SIlias Apalodimas * dataref set to 0, which will trigger the recycling 7012cc3aeb5SIlias Apalodimas */ 7022cc3aeb5SIlias Apalodimas skb->pp_recycle = 0; 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds 7051da177e4SLinus Torvalds /* 7061da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 7071da177e4SLinus Torvalds */ 7082d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 7091da177e4SLinus Torvalds { 710d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 711d179cd12SDavid S. Miller 712d179cd12SDavid S. Miller switch (skb->fclone) { 713d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 7141da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 7156ffe75ebSEric Dumazet return; 716d179cd12SDavid S. Miller 717d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 718d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 7196ffe75ebSEric Dumazet 7206ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 7216ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 7226ffe75ebSEric Dumazet * while here, branch prediction will be good. 7236ffe75ebSEric Dumazet */ 7242638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 7256ffe75ebSEric Dumazet goto fastpath; 726d179cd12SDavid S. Miller break; 727d179cd12SDavid S. Miller 7286ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 729d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 730d179cd12SDavid S. Miller break; 7313ff50b79SStephen Hemminger } 7322638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 7336ffe75ebSEric Dumazet return; 7346ffe75ebSEric Dumazet fastpath: 7356ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 7361da177e4SLinus Torvalds } 7371da177e4SLinus Torvalds 7380a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 7391da177e4SLinus Torvalds { 740adf30907SEric Dumazet skb_dst_drop(skb); 7411da177e4SLinus Torvalds if (skb->destructor) { 7427890e2f0SEric Dumazet DEBUG_NET_WARN_ON_ONCE(in_hardirq()); 7431da177e4SLinus Torvalds skb->destructor(skb); 7441da177e4SLinus Torvalds } 745a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 746cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 7472fc72c7bSKOVACS Krisztian #endif 748df5042f4SFlorian Westphal skb_ext_put(skb); 74904a4bb55SLennert Buytenhek } 75004a4bb55SLennert Buytenhek 75104a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 75204a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 75304a4bb55SLennert Buytenhek { 75404a4bb55SLennert Buytenhek skb_release_head_state(skb); 755a28b1b90SFlorian Westphal if (likely(skb->head)) 7562d4baff8SHerbert Xu skb_release_data(skb); 7572d4baff8SHerbert Xu } 7581da177e4SLinus Torvalds 7592d4baff8SHerbert Xu /** 7602d4baff8SHerbert Xu * __kfree_skb - private function 7612d4baff8SHerbert Xu * @skb: buffer 7622d4baff8SHerbert Xu * 7632d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 7642d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 7652d4baff8SHerbert Xu * always call kfree_skb 7662d4baff8SHerbert Xu */ 7672d4baff8SHerbert Xu 7682d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 7692d4baff8SHerbert Xu { 7702d4baff8SHerbert Xu skb_release_all(skb); 7711da177e4SLinus Torvalds kfree_skbmem(skb); 7721da177e4SLinus Torvalds } 773b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds /** 776c504e5c2SMenglong Dong * kfree_skb_reason - free an sk_buff with special reason 777231d06aeSJörn Engel * @skb: buffer to free 778c504e5c2SMenglong Dong * @reason: reason why this skb is dropped 779231d06aeSJörn Engel * 780231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 781c504e5c2SMenglong Dong * hit zero. Meanwhile, pass the drop reason to 'kfree_skb' 782c504e5c2SMenglong Dong * tracepoint. 783231d06aeSJörn Engel */ 784c205cc75SMenglong Dong void __fix_address 785c205cc75SMenglong Dong kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 786231d06aeSJörn Engel { 787c205cc75SMenglong Dong if (unlikely(!skb_unref(skb))) 788231d06aeSJörn Engel return; 7893889a803SPaolo Abeni 79020bbcd0aSMenglong Dong DEBUG_NET_WARN_ON_ONCE(reason <= 0 || reason >= SKB_DROP_REASON_MAX); 79120bbcd0aSMenglong Dong 792c504e5c2SMenglong Dong trace_kfree_skb(skb, __builtin_return_address(0), reason); 793231d06aeSJörn Engel __kfree_skb(skb); 794231d06aeSJörn Engel } 795c504e5c2SMenglong Dong EXPORT_SYMBOL(kfree_skb_reason); 796231d06aeSJörn Engel 797215b0f19SMenglong Dong void kfree_skb_list_reason(struct sk_buff *segs, 798215b0f19SMenglong Dong enum skb_drop_reason reason) 799bd8a7036SEric Dumazet { 800bd8a7036SEric Dumazet while (segs) { 801bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 802bd8a7036SEric Dumazet 803215b0f19SMenglong Dong kfree_skb_reason(segs, reason); 804bd8a7036SEric Dumazet segs = next; 805bd8a7036SEric Dumazet } 806bd8a7036SEric Dumazet } 807215b0f19SMenglong Dong EXPORT_SYMBOL(kfree_skb_list_reason); 808bd8a7036SEric Dumazet 8096413139dSWillem de Bruijn /* Dump skb information and contents. 8106413139dSWillem de Bruijn * 8116413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 8126413139dSWillem de Bruijn * 813302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 8146413139dSWillem de Bruijn */ 8156413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 8166413139dSWillem de Bruijn { 8176413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 8186413139dSWillem de Bruijn struct net_device *dev = skb->dev; 8196413139dSWillem de Bruijn struct sock *sk = skb->sk; 8206413139dSWillem de Bruijn struct sk_buff *list_skb; 8216413139dSWillem de Bruijn bool has_mac, has_trans; 8226413139dSWillem de Bruijn int headroom, tailroom; 8236413139dSWillem de Bruijn int i, len, seg_len; 8246413139dSWillem de Bruijn 8256413139dSWillem de Bruijn if (full_pkt) 8266413139dSWillem de Bruijn len = skb->len; 8276413139dSWillem de Bruijn else 8286413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 8296413139dSWillem de Bruijn 8306413139dSWillem de Bruijn headroom = skb_headroom(skb); 8316413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 8326413139dSWillem de Bruijn 8336413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 8346413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 8356413139dSWillem de Bruijn 8366413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 8376413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 8386413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 8396413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 8406413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 8416413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 8426413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 8436413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 8446413139dSWillem de Bruijn skb->network_header, 8456413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 8466413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 8476413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 8486413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 8496413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 8506413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 8516413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 8526413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 8536413139dSWillem de Bruijn 8546413139dSWillem de Bruijn if (dev) 8558a03ef67SGal Pressman printk("%sdev name=%s feat=%pNF\n", 8566413139dSWillem de Bruijn level, dev->name, &dev->features); 8576413139dSWillem de Bruijn if (sk) 858db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 8596413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 8606413139dSWillem de Bruijn 8616413139dSWillem de Bruijn if (full_pkt && headroom) 8626413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 8636413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 8646413139dSWillem de Bruijn 8656413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 8666413139dSWillem de Bruijn if (seg_len) 8676413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 8686413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 8696413139dSWillem de Bruijn len -= seg_len; 8706413139dSWillem de Bruijn 8716413139dSWillem de Bruijn if (full_pkt && tailroom) 8726413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 8736413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 8746413139dSWillem de Bruijn 8756413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 8766413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 8776413139dSWillem de Bruijn u32 p_off, p_len, copied; 8786413139dSWillem de Bruijn struct page *p; 8796413139dSWillem de Bruijn u8 *vaddr; 8806413139dSWillem de Bruijn 881b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 8826413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 8836413139dSWillem de Bruijn copied) { 8846413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 8856413139dSWillem de Bruijn vaddr = kmap_atomic(p); 8866413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 8876413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 8886413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 8896413139dSWillem de Bruijn kunmap_atomic(vaddr); 8906413139dSWillem de Bruijn len -= seg_len; 8916413139dSWillem de Bruijn if (!len) 8926413139dSWillem de Bruijn break; 8936413139dSWillem de Bruijn } 8946413139dSWillem de Bruijn } 8956413139dSWillem de Bruijn 8966413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 8976413139dSWillem de Bruijn printk("skb fraglist:\n"); 8986413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 8996413139dSWillem de Bruijn skb_dump(level, list_skb, true); 9006413139dSWillem de Bruijn } 9016413139dSWillem de Bruijn } 9026413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 9036413139dSWillem de Bruijn 904d1a203eaSStephen Hemminger /** 90525121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 90625121173SMichael S. Tsirkin * @skb: buffer that triggered an error 90725121173SMichael S. Tsirkin * 90825121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 90925121173SMichael S. Tsirkin * skb must be freed afterwards. 91025121173SMichael S. Tsirkin */ 91125121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 91225121173SMichael S. Tsirkin { 913753f1ca4SPavel Begunkov if (skb) { 914753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 9151f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 91625121173SMichael S. Tsirkin } 917753f1ca4SPavel Begunkov } 91825121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 91925121173SMichael S. Tsirkin 920be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 92125121173SMichael S. Tsirkin /** 922ead2ceb0SNeil Horman * consume_skb - free an skbuff 923ead2ceb0SNeil Horman * @skb: buffer to free 924ead2ceb0SNeil Horman * 925ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 926ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 927ead2ceb0SNeil Horman * is being dropped after a failure and notes that 928ead2ceb0SNeil Horman */ 929ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 930ead2ceb0SNeil Horman { 9313889a803SPaolo Abeni if (!skb_unref(skb)) 932ead2ceb0SNeil Horman return; 9333889a803SPaolo Abeni 93407dc22e7SKoki Sanagi trace_consume_skb(skb); 935ead2ceb0SNeil Horman __kfree_skb(skb); 936ead2ceb0SNeil Horman } 937ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 938be769db2SHerbert Xu #endif 939ead2ceb0SNeil Horman 9400a463c78SPaolo Abeni /** 941c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 9420a463c78SPaolo Abeni * @skb: buffer to free 9430a463c78SPaolo Abeni * 944ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 945ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 9460a463c78SPaolo Abeni */ 947ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 9480a463c78SPaolo Abeni { 9490a463c78SPaolo Abeni trace_consume_skb(skb); 9500a463c78SPaolo Abeni skb_release_data(skb); 9510a463c78SPaolo Abeni kfree_skbmem(skb); 9520a463c78SPaolo Abeni } 9530a463c78SPaolo Abeni 954f450d539SAlexander Lobakin static void napi_skb_cache_put(struct sk_buff *skb) 955795bb1c0SJesper Dangaard Brouer { 956795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 957f450d539SAlexander Lobakin u32 i; 958795bb1c0SJesper Dangaard Brouer 959f450d539SAlexander Lobakin kasan_poison_object_data(skbuff_head_cache, skb); 960795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 961795bb1c0SJesper Dangaard Brouer 962795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 963f450d539SAlexander Lobakin for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++) 964f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, 965f450d539SAlexander Lobakin nc->skb_cache[i]); 966f450d539SAlexander Lobakin 967f450d539SAlexander Lobakin kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF, 968f450d539SAlexander Lobakin nc->skb_cache + NAPI_SKB_CACHE_HALF); 969f450d539SAlexander Lobakin nc->skb_count = NAPI_SKB_CACHE_HALF; 970795bb1c0SJesper Dangaard Brouer } 971795bb1c0SJesper Dangaard Brouer } 972f450d539SAlexander Lobakin 97315fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 97415fad714SJesper Dangaard Brouer { 9759243adfcSAlexander Lobakin skb_release_all(skb); 9769243adfcSAlexander Lobakin napi_skb_cache_put(skb); 9779243adfcSAlexander Lobakin } 9789243adfcSAlexander Lobakin 9799243adfcSAlexander Lobakin void napi_skb_free_stolen_head(struct sk_buff *skb) 9809243adfcSAlexander Lobakin { 9819efb4b5bSPaolo Abeni if (unlikely(skb->slow_gro)) { 9828550ff8dSPaul Blakey nf_reset_ct(skb); 9839243adfcSAlexander Lobakin skb_dst_drop(skb); 9849243adfcSAlexander Lobakin skb_ext_put(skb); 9855e10da53SPaolo Abeni skb_orphan(skb); 9869efb4b5bSPaolo Abeni skb->slow_gro = 0; 9879efb4b5bSPaolo Abeni } 988f450d539SAlexander Lobakin napi_skb_cache_put(skb); 98915fad714SJesper Dangaard Brouer } 990795bb1c0SJesper Dangaard Brouer 991795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 992795bb1c0SJesper Dangaard Brouer { 993885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 994795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 995885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 996795bb1c0SJesper Dangaard Brouer return; 997795bb1c0SJesper Dangaard Brouer } 998795bb1c0SJesper Dangaard Brouer 999ee2640dfSEric Dumazet DEBUG_NET_WARN_ON_ONCE(!in_softirq()); 10006454eca8SYunsheng Lin 10017608894eSPaolo Abeni if (!skb_unref(skb)) 1002795bb1c0SJesper Dangaard Brouer return; 10037608894eSPaolo Abeni 1004795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 1005795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 1006795bb1c0SJesper Dangaard Brouer 1007795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 1008abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 1009795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 1010795bb1c0SJesper Dangaard Brouer return; 1011795bb1c0SJesper Dangaard Brouer } 1012795bb1c0SJesper Dangaard Brouer 10139243adfcSAlexander Lobakin skb_release_all(skb); 1014f450d539SAlexander Lobakin napi_skb_cache_put(skb); 1015795bb1c0SJesper Dangaard Brouer } 1016795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 1017795bb1c0SJesper Dangaard Brouer 101803f61041SKees Cook /* Make sure a field is contained by headers group */ 1019b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 102003f61041SKees Cook BUILD_BUG_ON(offsetof(struct sk_buff, field) != \ 102103f61041SKees Cook offsetof(struct sk_buff, headers.field)); \ 1022b1937227SEric Dumazet 1023dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 1024dec18810SHerbert Xu { 1025dec18810SHerbert Xu new->tstamp = old->tstamp; 1026b1937227SEric Dumazet /* We do not copy old->sk */ 1027dec18810SHerbert Xu new->dev = old->dev; 1028b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 10297fee226aSEric Dumazet skb_dst_copy(new, old); 1030df5042f4SFlorian Westphal __skb_ext_copy(new, old); 1031b1937227SEric Dumazet __nf_copy(new, old, false); 10326aa895b0SPatrick McHardy 103303f61041SKees Cook /* Note : this field could be in the headers group. 1034b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 1035b1937227SEric Dumazet */ 1036b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 103706021292SEliezer Tamir 103803f61041SKees Cook memcpy(&new->headers, &old->headers, sizeof(new->headers)); 1039b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 1040b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 1041b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 1042b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 1043b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 1044b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 1045b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 1046b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 1047b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 1048b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 1049b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 1050b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 1051b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 1052b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 1053b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 1054b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 1055b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 1056b1937227SEric Dumazet #endif 1057e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 1058b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 105906021292SEliezer Tamir #endif 106068822bdfSEric Dumazet CHECK_SKB_FIELD(alloc_cpu); 10612bd82484SEric Dumazet #ifdef CONFIG_XPS 10622bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 10632bd82484SEric Dumazet #endif 1064b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 1065b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 1066b1937227SEric Dumazet #endif 1067b1937227SEric Dumazet 1068dec18810SHerbert Xu } 1069dec18810SHerbert Xu 107082c49a35SHerbert Xu /* 107182c49a35SHerbert Xu * You should not add any new code to this function. Add it to 107282c49a35SHerbert Xu * __copy_skb_header above instead. 107382c49a35SHerbert Xu */ 1074e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 10751da177e4SLinus Torvalds { 10761da177e4SLinus Torvalds #define C(x) n->x = skb->x 10771da177e4SLinus Torvalds 10781da177e4SLinus Torvalds n->next = n->prev = NULL; 10791da177e4SLinus Torvalds n->sk = NULL; 1080dec18810SHerbert Xu __copy_skb_header(n, skb); 1081dec18810SHerbert Xu 10821da177e4SLinus Torvalds C(len); 10831da177e4SLinus Torvalds C(data_len); 10843e6b3b2eSAlexey Dobriyan C(mac_len); 1085334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 108602f1c89dSPaul Moore n->cloned = 1; 10871da177e4SLinus Torvalds n->nohdr = 0; 1088b13dda9fSEric Dumazet n->peeked = 0; 1089e78bfb07SStefano Brivio C(pfmemalloc); 10906a5bcd84SIlias Apalodimas C(pp_recycle); 10911da177e4SLinus Torvalds n->destructor = NULL; 10921da177e4SLinus Torvalds C(tail); 10931da177e4SLinus Torvalds C(end); 109402f1c89dSPaul Moore C(head); 1095d3836f21SEric Dumazet C(head_frag); 109602f1c89dSPaul Moore C(data); 109702f1c89dSPaul Moore C(truesize); 109863354797SReshetova, Elena refcount_set(&n->users, 1); 10991da177e4SLinus Torvalds 11001da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 11011da177e4SLinus Torvalds skb->cloned = 1; 11021da177e4SLinus Torvalds 11031da177e4SLinus Torvalds return n; 1104e0053ec0SHerbert Xu #undef C 1105e0053ec0SHerbert Xu } 1106e0053ec0SHerbert Xu 1107e0053ec0SHerbert Xu /** 1108da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1109da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1110da29e4b4SJakub Kicinski */ 1111da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1112da29e4b4SJakub Kicinski { 1113da29e4b4SJakub Kicinski struct sk_buff *n; 1114da29e4b4SJakub Kicinski 1115da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1116da29e4b4SJakub Kicinski if (!n) 1117da29e4b4SJakub Kicinski return NULL; 1118da29e4b4SJakub Kicinski 1119da29e4b4SJakub Kicinski n->len = first->len; 1120da29e4b4SJakub Kicinski n->data_len = first->len; 1121da29e4b4SJakub Kicinski n->truesize = first->truesize; 1122da29e4b4SJakub Kicinski 1123da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1124da29e4b4SJakub Kicinski 1125da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1126da29e4b4SJakub Kicinski n->destructor = NULL; 1127da29e4b4SJakub Kicinski 1128da29e4b4SJakub Kicinski return n; 1129da29e4b4SJakub Kicinski } 1130da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1131da29e4b4SJakub Kicinski 1132da29e4b4SJakub Kicinski /** 1133e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1134e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1135e0053ec0SHerbert Xu * @src: the skb to supply the contents 1136e0053ec0SHerbert Xu * 1137e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1138e0053ec0SHerbert Xu * supplied by the user. 1139e0053ec0SHerbert Xu * 1140e0053ec0SHerbert Xu * The target skb is returned upon exit. 1141e0053ec0SHerbert Xu */ 1142e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1143e0053ec0SHerbert Xu { 11442d4baff8SHerbert Xu skb_release_all(dst); 1145e0053ec0SHerbert Xu return __skb_clone(dst, src); 1146e0053ec0SHerbert Xu } 1147e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1148e0053ec0SHerbert Xu 11496f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1150a91dbff5SWillem de Bruijn { 1151a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1152a91dbff5SWillem de Bruijn struct user_struct *user; 1153a91dbff5SWillem de Bruijn 1154a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1155a91dbff5SWillem de Bruijn return 0; 1156a91dbff5SWillem de Bruijn 1157a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1158a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1159a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1160a91dbff5SWillem de Bruijn 1161a91dbff5SWillem de Bruijn do { 1162a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 1163a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1164a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1165a91dbff5SWillem de Bruijn return -ENOBUFS; 1166a91dbff5SWillem de Bruijn } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 1167a91dbff5SWillem de Bruijn old_pg); 1168a91dbff5SWillem de Bruijn 1169a91dbff5SWillem de Bruijn if (!mmp->user) { 1170a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1171a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1172a91dbff5SWillem de Bruijn } else { 1173a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1174a91dbff5SWillem de Bruijn } 1175a91dbff5SWillem de Bruijn 1176a91dbff5SWillem de Bruijn return 0; 1177a91dbff5SWillem de Bruijn } 11786f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1179a91dbff5SWillem de Bruijn 11806f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1181a91dbff5SWillem de Bruijn { 1182a91dbff5SWillem de Bruijn if (mmp->user) { 1183a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1184a91dbff5SWillem de Bruijn free_uid(mmp->user); 1185a91dbff5SWillem de Bruijn } 1186a91dbff5SWillem de Bruijn } 11876f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1188a91dbff5SWillem de Bruijn 1189c67b627eSDavid Ahern static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size) 119052267790SWillem de Bruijn { 1191*e7d2b510SPavel Begunkov struct ubuf_info_msgzc *uarg; 119252267790SWillem de Bruijn struct sk_buff *skb; 119352267790SWillem de Bruijn 119452267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 119552267790SWillem de Bruijn 119652267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 119752267790SWillem de Bruijn if (!skb) 119852267790SWillem de Bruijn return NULL; 119952267790SWillem de Bruijn 120052267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 120152267790SWillem de Bruijn uarg = (void *)skb->cb; 1202a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1203a91dbff5SWillem de Bruijn 1204a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1205a91dbff5SWillem de Bruijn kfree_skb(skb); 1206a91dbff5SWillem de Bruijn return NULL; 1207a91dbff5SWillem de Bruijn } 120852267790SWillem de Bruijn 1209*e7d2b510SPavel Begunkov uarg->ubuf.callback = msg_zerocopy_callback; 12104ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 12114ab6c99dSWillem de Bruijn uarg->len = 1; 12124ab6c99dSWillem de Bruijn uarg->bytelen = size; 121352267790SWillem de Bruijn uarg->zerocopy = 1; 1214*e7d2b510SPavel Begunkov uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN; 1215*e7d2b510SPavel Begunkov refcount_set(&uarg->ubuf.refcnt, 1); 121652267790SWillem de Bruijn sock_hold(sk); 121752267790SWillem de Bruijn 1218*e7d2b510SPavel Begunkov return &uarg->ubuf; 121952267790SWillem de Bruijn } 122052267790SWillem de Bruijn 1221*e7d2b510SPavel Begunkov static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg) 122252267790SWillem de Bruijn { 122352267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 122452267790SWillem de Bruijn } 122552267790SWillem de Bruijn 12268c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 12274ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 12284ab6c99dSWillem de Bruijn { 12294ab6c99dSWillem de Bruijn if (uarg) { 1230*e7d2b510SPavel Begunkov struct ubuf_info_msgzc *uarg_zc; 12314ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 12324ab6c99dSWillem de Bruijn u32 bytelen, next; 12334ab6c99dSWillem de Bruijn 12341b4b2b09SPavel Begunkov /* there might be non MSG_ZEROCOPY users */ 12351b4b2b09SPavel Begunkov if (uarg->callback != msg_zerocopy_callback) 12361b4b2b09SPavel Begunkov return NULL; 12371b4b2b09SPavel Begunkov 12384ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 12394ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 12404ab6c99dSWillem de Bruijn */ 12414ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 12424ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 12434ab6c99dSWillem de Bruijn return NULL; 12444ab6c99dSWillem de Bruijn } 12454ab6c99dSWillem de Bruijn 1246*e7d2b510SPavel Begunkov uarg_zc = uarg_to_msgzc(uarg); 1247*e7d2b510SPavel Begunkov bytelen = uarg_zc->bytelen + size; 1248*e7d2b510SPavel Begunkov if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) { 12494ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 12504ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12514ab6c99dSWillem de Bruijn goto new_alloc; 12524ab6c99dSWillem de Bruijn return NULL; 12534ab6c99dSWillem de Bruijn } 12544ab6c99dSWillem de Bruijn 12554ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 1256*e7d2b510SPavel Begunkov if ((u32)(uarg_zc->id + uarg_zc->len) == next) { 1257*e7d2b510SPavel Begunkov if (mm_account_pinned_pages(&uarg_zc->mmp, size)) 1258a91dbff5SWillem de Bruijn return NULL; 1259*e7d2b510SPavel Begunkov uarg_zc->len++; 1260*e7d2b510SPavel Begunkov uarg_zc->bytelen = bytelen; 12614ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1262100f6d8eSWillem de Bruijn 1263100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1264100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 12658e044917SJonathan Lemon net_zcopy_get(uarg); 1266100f6d8eSWillem de Bruijn 12674ab6c99dSWillem de Bruijn return uarg; 12684ab6c99dSWillem de Bruijn } 12694ab6c99dSWillem de Bruijn } 12704ab6c99dSWillem de Bruijn 12714ab6c99dSWillem de Bruijn new_alloc: 12728c793822SJonathan Lemon return msg_zerocopy_alloc(sk, size); 12734ab6c99dSWillem de Bruijn } 12748c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_realloc); 12754ab6c99dSWillem de Bruijn 12764ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 12774ab6c99dSWillem de Bruijn { 12784ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 12794ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 12804ab6c99dSWillem de Bruijn u64 sum_len; 12814ab6c99dSWillem de Bruijn 12824ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 12834ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 12844ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 12854ab6c99dSWillem de Bruijn 12864ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 12874ab6c99dSWillem de Bruijn return false; 12884ab6c99dSWillem de Bruijn 12894ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 12904ab6c99dSWillem de Bruijn return false; 12914ab6c99dSWillem de Bruijn 12924ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 12934ab6c99dSWillem de Bruijn return true; 12944ab6c99dSWillem de Bruijn } 12954ab6c99dSWillem de Bruijn 1296*e7d2b510SPavel Begunkov static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg) 129752267790SWillem de Bruijn { 12984ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 129952267790SWillem de Bruijn struct sock_exterr_skb *serr; 130052267790SWillem de Bruijn struct sock *sk = skb->sk; 13014ab6c99dSWillem de Bruijn struct sk_buff_head *q; 13024ab6c99dSWillem de Bruijn unsigned long flags; 13033bdd5ee0SWillem de Bruijn bool is_zerocopy; 13044ab6c99dSWillem de Bruijn u32 lo, hi; 13054ab6c99dSWillem de Bruijn u16 len; 130652267790SWillem de Bruijn 1307ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1308ccaffff1SWillem de Bruijn 13094ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 13104ab6c99dSWillem de Bruijn * so do not queue a completion notification 13114ab6c99dSWillem de Bruijn */ 13124ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 131352267790SWillem de Bruijn goto release; 131452267790SWillem de Bruijn 13154ab6c99dSWillem de Bruijn len = uarg->len; 13164ab6c99dSWillem de Bruijn lo = uarg->id; 13174ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 13183bdd5ee0SWillem de Bruijn is_zerocopy = uarg->zerocopy; 13194ab6c99dSWillem de Bruijn 132052267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 132152267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 132252267790SWillem de Bruijn serr->ee.ee_errno = 0; 132352267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 13244ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 13254ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 13263bdd5ee0SWillem de Bruijn if (!is_zerocopy) 132752267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 132852267790SWillem de Bruijn 13294ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 13304ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 13314ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 13324ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 13334ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 13344ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 133552267790SWillem de Bruijn skb = NULL; 13364ab6c99dSWillem de Bruijn } 13374ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 133852267790SWillem de Bruijn 1339e3ae2365SAlexander Aring sk_error_report(sk); 134052267790SWillem de Bruijn 134152267790SWillem de Bruijn release: 134252267790SWillem de Bruijn consume_skb(skb); 134352267790SWillem de Bruijn sock_put(sk); 134452267790SWillem de Bruijn } 134575518851SJonathan Lemon 13468c793822SJonathan Lemon void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 134736177832SJonathan Lemon bool success) 134875518851SJonathan Lemon { 1349*e7d2b510SPavel Begunkov struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg); 1350*e7d2b510SPavel Begunkov 1351*e7d2b510SPavel Begunkov uarg_zc->zerocopy = uarg_zc->zerocopy & success; 135275518851SJonathan Lemon 135375518851SJonathan Lemon if (refcount_dec_and_test(&uarg->refcnt)) 1354*e7d2b510SPavel Begunkov __msg_zerocopy_callback(uarg_zc); 135575518851SJonathan Lemon } 13568c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_callback); 135752267790SWillem de Bruijn 13588c793822SJonathan Lemon void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 135952267790SWillem de Bruijn { 1360*e7d2b510SPavel Begunkov struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk; 136152267790SWillem de Bruijn 136252267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 1363*e7d2b510SPavel Begunkov uarg_to_msgzc(uarg)->len--; 136452267790SWillem de Bruijn 136552900d22SWillem de Bruijn if (have_uref) 13668c793822SJonathan Lemon msg_zerocopy_callback(NULL, uarg, true); 136752267790SWillem de Bruijn } 13688c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort); 136952267790SWillem de Bruijn 137052267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 137152267790SWillem de Bruijn struct msghdr *msg, int len, 137252267790SWillem de Bruijn struct ubuf_info *uarg) 137352267790SWillem de Bruijn { 13744ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 137552267790SWillem de Bruijn int err, orig_len = skb->len; 137652267790SWillem de Bruijn 13774ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 13784ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 13794ab6c99dSWillem de Bruijn */ 13804ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 13814ab6c99dSWillem de Bruijn return -EEXIST; 13824ab6c99dSWillem de Bruijn 1383ebe73a28SDavid Ahern err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len); 138452267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 138554d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 138654d43117SWillem de Bruijn 138752267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 1388829b7bddSPavel Begunkov iov_iter_revert(&msg->msg_iter, skb->len - orig_len); 138954d43117SWillem de Bruijn skb->sk = sk; 139052267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 139154d43117SWillem de Bruijn skb->sk = save_sk; 139252267790SWillem de Bruijn return err; 139352267790SWillem de Bruijn } 139452267790SWillem de Bruijn 139552900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 139652267790SWillem de Bruijn return skb->len - orig_len; 139752267790SWillem de Bruijn } 139852267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 139952267790SWillem de Bruijn 1400753f1ca4SPavel Begunkov void __skb_zcopy_downgrade_managed(struct sk_buff *skb) 1401753f1ca4SPavel Begunkov { 1402753f1ca4SPavel Begunkov int i; 1403753f1ca4SPavel Begunkov 1404753f1ca4SPavel Begunkov skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS; 1405753f1ca4SPavel Begunkov for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1406753f1ca4SPavel Begunkov skb_frag_ref(skb, i); 1407753f1ca4SPavel Begunkov } 1408753f1ca4SPavel Begunkov EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed); 1409753f1ca4SPavel Begunkov 14101f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 141152267790SWillem de Bruijn gfp_t gfp_mask) 141252267790SWillem de Bruijn { 141352267790SWillem de Bruijn if (skb_zcopy(orig)) { 141452267790SWillem de Bruijn if (skb_zcopy(nskb)) { 141552267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 141652267790SWillem de Bruijn if (!gfp_mask) { 141752267790SWillem de Bruijn WARN_ON_ONCE(1); 141852267790SWillem de Bruijn return -ENOMEM; 141952267790SWillem de Bruijn } 142052267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 142152267790SWillem de Bruijn return 0; 142252267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 142352267790SWillem de Bruijn return -EIO; 142452267790SWillem de Bruijn } 142552900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 142652267790SWillem de Bruijn } 142752267790SWillem de Bruijn return 0; 142852267790SWillem de Bruijn } 142952267790SWillem de Bruijn 14302c53040fSBen Hutchings /** 14312c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 143248c83012SMichael S. Tsirkin * @skb: the skb to modify 143348c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 143448c83012SMichael S. Tsirkin * 143506b4feb3SJonathan Lemon * This must be called on skb with SKBFL_ZEROCOPY_ENABLE. 143648c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 143748c83012SMichael S. Tsirkin * to userspace pages. 143848c83012SMichael S. Tsirkin * 143948c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 144048c83012SMichael S. Tsirkin * %GFP_ATOMIC. 144148c83012SMichael S. Tsirkin * 144248c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 144348c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 144448c83012SMichael S. Tsirkin */ 144548c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1446a6686f2fSShirley Ma { 1447a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1448a6686f2fSShirley Ma struct page *page, *head = NULL; 14493ece7826SWillem de Bruijn int i, new_frags; 14503ece7826SWillem de Bruijn u32 d_off; 1451a6686f2fSShirley Ma 14523ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 14533ece7826SWillem de Bruijn return -EINVAL; 14543ece7826SWillem de Bruijn 1455f72c4ac6SWillem de Bruijn if (!num_frags) 1456f72c4ac6SWillem de Bruijn goto release; 1457f72c4ac6SWillem de Bruijn 14583ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 14593ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 146002756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1461a6686f2fSShirley Ma if (!page) { 1462a6686f2fSShirley Ma while (head) { 146340dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1464a6686f2fSShirley Ma put_page(head); 1465a6686f2fSShirley Ma head = next; 1466a6686f2fSShirley Ma } 1467a6686f2fSShirley Ma return -ENOMEM; 1468a6686f2fSShirley Ma } 14693ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 14703ece7826SWillem de Bruijn head = page; 14713ece7826SWillem de Bruijn } 14723ece7826SWillem de Bruijn 14733ece7826SWillem de Bruijn page = head; 14743ece7826SWillem de Bruijn d_off = 0; 14753ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 14763ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 14773ece7826SWillem de Bruijn u32 p_off, p_len, copied; 14783ece7826SWillem de Bruijn struct page *p; 14793ece7826SWillem de Bruijn u8 *vaddr; 1480c613c209SWillem de Bruijn 1481b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1482c613c209SWillem de Bruijn p, p_off, p_len, copied) { 14833ece7826SWillem de Bruijn u32 copy, done = 0; 1484c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 14853ece7826SWillem de Bruijn 14863ece7826SWillem de Bruijn while (done < p_len) { 14873ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 14883ece7826SWillem de Bruijn d_off = 0; 14893ece7826SWillem de Bruijn page = (struct page *)page_private(page); 14903ece7826SWillem de Bruijn } 14913ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 14923ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 14933ece7826SWillem de Bruijn vaddr + p_off + done, copy); 14943ece7826SWillem de Bruijn done += copy; 14953ece7826SWillem de Bruijn d_off += copy; 14963ece7826SWillem de Bruijn } 149751c56b00SEric Dumazet kunmap_atomic(vaddr); 1498c613c209SWillem de Bruijn } 1499a6686f2fSShirley Ma } 1500a6686f2fSShirley Ma 1501a6686f2fSShirley Ma /* skb frags release userspace buffers */ 150202756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1503a8605c60SIan Campbell skb_frag_unref(skb, i); 1504a6686f2fSShirley Ma 1505a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 15063ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 15073ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 150840dadff2SSunghan Suh head = (struct page *)page_private(head); 1509a6686f2fSShirley Ma } 15103ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 15113ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 151248c83012SMichael S. Tsirkin 1513b90ddd56SWillem de Bruijn release: 15141f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1515a6686f2fSShirley Ma return 0; 1516a6686f2fSShirley Ma } 1517dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1518a6686f2fSShirley Ma 1519e0053ec0SHerbert Xu /** 1520e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1521e0053ec0SHerbert Xu * @skb: buffer to clone 1522e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1523e0053ec0SHerbert Xu * 1524e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1525e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1526e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1527e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1528e0053ec0SHerbert Xu * 1529e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1530e0053ec0SHerbert Xu * %GFP_ATOMIC. 1531e0053ec0SHerbert Xu */ 1532e0053ec0SHerbert Xu 1533e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1534e0053ec0SHerbert Xu { 1535d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1536d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1537d0bf4a9eSEric Dumazet skb1); 15386ffe75ebSEric Dumazet struct sk_buff *n; 1539e0053ec0SHerbert Xu 154070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1541a6686f2fSShirley Ma return NULL; 1542a6686f2fSShirley Ma 1543e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 15442638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 15456ffe75ebSEric Dumazet n = &fclones->skb2; 15462638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1547c2dd4059SEric Dumazet n->fclone = SKB_FCLONE_CLONE; 1548e0053ec0SHerbert Xu } else { 1549c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1550c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1551c93bdd0eSMel Gorman 1552e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1553e0053ec0SHerbert Xu if (!n) 1554e0053ec0SHerbert Xu return NULL; 1555fe55f6d5SVegard Nossum 1556e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1557e0053ec0SHerbert Xu } 1558e0053ec0SHerbert Xu 1559e0053ec0SHerbert Xu return __skb_clone(n, skb); 15601da177e4SLinus Torvalds } 1561b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 15621da177e4SLinus Torvalds 1563b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1564f5b17294SPravin B Shelar { 1565030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1566030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1567030737bcSEric Dumazet skb->csum_start += off; 1568f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1569f5b17294SPravin B Shelar skb->transport_header += off; 1570f5b17294SPravin B Shelar skb->network_header += off; 1571f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1572f5b17294SPravin B Shelar skb->mac_header += off; 1573f5b17294SPravin B Shelar skb->inner_transport_header += off; 1574f5b17294SPravin B Shelar skb->inner_network_header += off; 1575aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1576f5b17294SPravin B Shelar } 1577b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1578f5b17294SPravin B Shelar 157908303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 15801da177e4SLinus Torvalds { 1581dec18810SHerbert Xu __copy_skb_header(new, old); 1582dec18810SHerbert Xu 15837967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 15847967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 15857967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 15861da177e4SLinus Torvalds } 158708303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 15881da177e4SLinus Torvalds 1589c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1590c93bdd0eSMel Gorman { 1591c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1592c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1593c93bdd0eSMel Gorman return 0; 1594c93bdd0eSMel Gorman } 1595c93bdd0eSMel Gorman 15961da177e4SLinus Torvalds /** 15971da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 15981da177e4SLinus Torvalds * @skb: buffer to copy 15991da177e4SLinus Torvalds * @gfp_mask: allocation priority 16001da177e4SLinus Torvalds * 16011da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 16021da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 16031da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 16041da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 16051da177e4SLinus Torvalds * 16061da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 16071da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 16081da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 16091da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 16101da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 16111da177e4SLinus Torvalds */ 16121da177e4SLinus Torvalds 1613dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 16141da177e4SLinus Torvalds { 16156602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1616ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1617c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1618c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 16196602cebbSEric Dumazet 16201da177e4SLinus Torvalds if (!n) 16211da177e4SLinus Torvalds return NULL; 16221da177e4SLinus Torvalds 16231da177e4SLinus Torvalds /* Set the data pointer */ 16241da177e4SLinus Torvalds skb_reserve(n, headerlen); 16251da177e4SLinus Torvalds /* Set the tail pointer and length */ 16261da177e4SLinus Torvalds skb_put(n, skb->len); 16271da177e4SLinus Torvalds 16289f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 16291da177e4SLinus Torvalds 163008303c18SIlya Lesokhin skb_copy_header(n, skb); 16311da177e4SLinus Torvalds return n; 16321da177e4SLinus Torvalds } 1633b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds /** 1636bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 16371da177e4SLinus Torvalds * @skb: buffer to copy 1638117632e6SEric Dumazet * @headroom: headroom of new skb 16391da177e4SLinus Torvalds * @gfp_mask: allocation priority 1640bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1641bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1642bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 16431da177e4SLinus Torvalds * 16441da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 16451da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 16461da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 16471da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 16481da177e4SLinus Torvalds * or the pointer to the buffer on success. 16491da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 16501da177e4SLinus Torvalds */ 16511da177e4SLinus Torvalds 1652bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1653bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 16541da177e4SLinus Torvalds { 1655117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1656bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1657bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 16586602cebbSEric Dumazet 16591da177e4SLinus Torvalds if (!n) 16601da177e4SLinus Torvalds goto out; 16611da177e4SLinus Torvalds 16621da177e4SLinus Torvalds /* Set the data pointer */ 1663117632e6SEric Dumazet skb_reserve(n, headroom); 16641da177e4SLinus Torvalds /* Set the tail pointer and length */ 16651da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 16661da177e4SLinus Torvalds /* Copy the bytes */ 1667d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 16681da177e4SLinus Torvalds 166925f484a6SHerbert Xu n->truesize += skb->data_len; 16701da177e4SLinus Torvalds n->data_len = skb->data_len; 16711da177e4SLinus Torvalds n->len = skb->len; 16721da177e4SLinus Torvalds 16731da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 16741da177e4SLinus Torvalds int i; 16751da177e4SLinus Torvalds 16761f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 16771f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 16781511022cSDan Carpenter kfree_skb(n); 16791511022cSDan Carpenter n = NULL; 1680a6686f2fSShirley Ma goto out; 1681a6686f2fSShirley Ma } 16821da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16831da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1684ea2ab693SIan Campbell skb_frag_ref(skb, i); 16851da177e4SLinus Torvalds } 16861da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 16871da177e4SLinus Torvalds } 16881da177e4SLinus Torvalds 168921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 16901da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 16911da177e4SLinus Torvalds skb_clone_fraglist(n); 16921da177e4SLinus Torvalds } 16931da177e4SLinus Torvalds 169408303c18SIlya Lesokhin skb_copy_header(n, skb); 16951da177e4SLinus Torvalds out: 16961da177e4SLinus Torvalds return n; 16971da177e4SLinus Torvalds } 1698bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 16991da177e4SLinus Torvalds 17001da177e4SLinus Torvalds /** 17011da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 17021da177e4SLinus Torvalds * @skb: buffer to reallocate 17031da177e4SLinus Torvalds * @nhead: room to add at head 17041da177e4SLinus Torvalds * @ntail: room to add at tail 17051da177e4SLinus Torvalds * @gfp_mask: allocation priority 17061da177e4SLinus Torvalds * 1707bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1708bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 17091da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 17101da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 17111da177e4SLinus Torvalds * 17121da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 17131da177e4SLinus Torvalds * reloaded after call to this function. 17141da177e4SLinus Torvalds */ 17151da177e4SLinus Torvalds 171686a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1717dd0fc66fSAl Viro gfp_t gfp_mask) 17181da177e4SLinus Torvalds { 1719158f323bSEric Dumazet int i, osize = skb_end_offset(skb); 1720158f323bSEric Dumazet int size = osize + nhead + ntail; 17211da177e4SLinus Torvalds long off; 1722158f323bSEric Dumazet u8 *data; 17231da177e4SLinus Torvalds 17244edd87adSHerbert Xu BUG_ON(nhead < 0); 17254edd87adSHerbert Xu 17269f77fad3STim Hansen BUG_ON(skb_shared(skb)); 17271da177e4SLinus Torvalds 1728753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 1729753f1ca4SPavel Begunkov 17301da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 17311da177e4SLinus Torvalds 1732c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1733c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1734c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1735c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 17361da177e4SLinus Torvalds if (!data) 17371da177e4SLinus Torvalds goto nodata; 173887151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 17391da177e4SLinus Torvalds 17401da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 17416602cebbSEric Dumazet * optimized for the cases when header is void. 17426602cebbSEric Dumazet */ 17436602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 17446602cebbSEric Dumazet 17456602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 17466602cebbSEric Dumazet skb_shinfo(skb), 1747fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 17481da177e4SLinus Torvalds 17493e24591aSAlexander Duyck /* 17503e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 17513e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 17523e24591aSAlexander Duyck * be since all we did is relocate the values 17533e24591aSAlexander Duyck */ 17543e24591aSAlexander Duyck if (skb_cloned(skb)) { 175570008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1756a6686f2fSShirley Ma goto nofrags; 17571f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1758c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 17591da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1760ea2ab693SIan Campbell skb_frag_ref(skb, i); 17611da177e4SLinus Torvalds 176221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 17631da177e4SLinus Torvalds skb_clone_fraglist(skb); 17641da177e4SLinus Torvalds 17651da177e4SLinus Torvalds skb_release_data(skb); 17663e24591aSAlexander Duyck } else { 17673e24591aSAlexander Duyck skb_free_head(skb); 17681fd63041SEric Dumazet } 17691da177e4SLinus Torvalds off = (data + nhead) - skb->head; 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds skb->head = data; 1772d3836f21SEric Dumazet skb->head_frag = 0; 17731da177e4SLinus Torvalds skb->data += off; 1774763087daSEric Dumazet 1775763087daSEric Dumazet skb_set_end_offset(skb, size); 17764305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 177756eb8882SPatrick McHardy off = nhead; 177856eb8882SPatrick McHardy #endif 177927a884dcSArnaldo Carvalho de Melo skb->tail += off; 1780b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 17811da177e4SLinus Torvalds skb->cloned = 0; 1782334a8132SPatrick McHardy skb->hdr_len = 0; 17831da177e4SLinus Torvalds skb->nohdr = 0; 17841da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1785158f323bSEric Dumazet 1786de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1787de8f3a83SDaniel Borkmann 1788158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1789158f323bSEric Dumazet * For the moment, we really care of rx path, or 1790158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1791158f323bSEric Dumazet */ 1792158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1793158f323bSEric Dumazet skb->truesize += size - osize; 1794158f323bSEric Dumazet 17951da177e4SLinus Torvalds return 0; 17961da177e4SLinus Torvalds 1797a6686f2fSShirley Ma nofrags: 1798a6686f2fSShirley Ma kfree(data); 17991da177e4SLinus Torvalds nodata: 18001da177e4SLinus Torvalds return -ENOMEM; 18011da177e4SLinus Torvalds } 1802b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 18031da177e4SLinus Torvalds 18041da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 18051da177e4SLinus Torvalds 18061da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 18071da177e4SLinus Torvalds { 18081da177e4SLinus Torvalds struct sk_buff *skb2; 18091da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 18101da177e4SLinus Torvalds 18111da177e4SLinus Torvalds if (delta <= 0) 18121da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 18131da177e4SLinus Torvalds else { 18141da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 18151da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 18161da177e4SLinus Torvalds GFP_ATOMIC)) { 18171da177e4SLinus Torvalds kfree_skb(skb2); 18181da177e4SLinus Torvalds skb2 = NULL; 18191da177e4SLinus Torvalds } 18201da177e4SLinus Torvalds } 18211da177e4SLinus Torvalds return skb2; 18221da177e4SLinus Torvalds } 1823b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 18241da177e4SLinus Torvalds 18252b88cba5SEric Dumazet int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 18262b88cba5SEric Dumazet { 18272b88cba5SEric Dumazet unsigned int saved_end_offset, saved_truesize; 18282b88cba5SEric Dumazet struct skb_shared_info *shinfo; 18292b88cba5SEric Dumazet int res; 18302b88cba5SEric Dumazet 18312b88cba5SEric Dumazet saved_end_offset = skb_end_offset(skb); 18322b88cba5SEric Dumazet saved_truesize = skb->truesize; 18332b88cba5SEric Dumazet 18342b88cba5SEric Dumazet res = pskb_expand_head(skb, 0, 0, pri); 18352b88cba5SEric Dumazet if (res) 18362b88cba5SEric Dumazet return res; 18372b88cba5SEric Dumazet 18382b88cba5SEric Dumazet skb->truesize = saved_truesize; 18392b88cba5SEric Dumazet 18402b88cba5SEric Dumazet if (likely(skb_end_offset(skb) == saved_end_offset)) 18412b88cba5SEric Dumazet return 0; 18422b88cba5SEric Dumazet 18432b88cba5SEric Dumazet shinfo = skb_shinfo(skb); 18442b88cba5SEric Dumazet 18452b88cba5SEric Dumazet /* We are about to change back skb->end, 18462b88cba5SEric Dumazet * we need to move skb_shinfo() to its new location. 18472b88cba5SEric Dumazet */ 18482b88cba5SEric Dumazet memmove(skb->head + saved_end_offset, 18492b88cba5SEric Dumazet shinfo, 18502b88cba5SEric Dumazet offsetof(struct skb_shared_info, frags[shinfo->nr_frags])); 18512b88cba5SEric Dumazet 18522b88cba5SEric Dumazet skb_set_end_offset(skb, saved_end_offset); 18532b88cba5SEric Dumazet 18542b88cba5SEric Dumazet return 0; 18552b88cba5SEric Dumazet } 18562b88cba5SEric Dumazet 18571da177e4SLinus Torvalds /** 1858f1260ff1SVasily Averin * skb_expand_head - reallocate header of &sk_buff 1859f1260ff1SVasily Averin * @skb: buffer to reallocate 1860f1260ff1SVasily Averin * @headroom: needed headroom 1861f1260ff1SVasily Averin * 1862f1260ff1SVasily Averin * Unlike skb_realloc_headroom, this one does not allocate a new skb 1863f1260ff1SVasily Averin * if possible; copies skb->sk to new skb as needed 1864f1260ff1SVasily Averin * and frees original skb in case of failures. 1865f1260ff1SVasily Averin * 1866f1260ff1SVasily Averin * It expect increased headroom and generates warning otherwise. 1867f1260ff1SVasily Averin */ 1868f1260ff1SVasily Averin 1869f1260ff1SVasily Averin struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom) 1870f1260ff1SVasily Averin { 1871f1260ff1SVasily Averin int delta = headroom - skb_headroom(skb); 18727f678defSVasily Averin int osize = skb_end_offset(skb); 18737f678defSVasily Averin struct sock *sk = skb->sk; 1874f1260ff1SVasily Averin 1875f1260ff1SVasily Averin if (WARN_ONCE(delta <= 0, 1876f1260ff1SVasily Averin "%s is expecting an increase in the headroom", __func__)) 1877f1260ff1SVasily Averin return skb; 1878f1260ff1SVasily Averin 18797f678defSVasily Averin delta = SKB_DATA_ALIGN(delta); 18807f678defSVasily Averin /* pskb_expand_head() might crash, if skb is shared. */ 18817f678defSVasily Averin if (skb_shared(skb) || !is_skb_wmem(skb)) { 1882f1260ff1SVasily Averin struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 1883f1260ff1SVasily Averin 18847f678defSVasily Averin if (unlikely(!nskb)) 18857f678defSVasily Averin goto fail; 18867f678defSVasily Averin 18877f678defSVasily Averin if (sk) 18887f678defSVasily Averin skb_set_owner_w(nskb, sk); 1889f1260ff1SVasily Averin consume_skb(skb); 1890f1260ff1SVasily Averin skb = nskb; 1891f1260ff1SVasily Averin } 18927f678defSVasily Averin if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC)) 18937f678defSVasily Averin goto fail; 18947f678defSVasily Averin 18957f678defSVasily Averin if (sk && is_skb_wmem(skb)) { 18967f678defSVasily Averin delta = skb_end_offset(skb) - osize; 18977f678defSVasily Averin refcount_add(delta, &sk->sk_wmem_alloc); 18987f678defSVasily Averin skb->truesize += delta; 1899f1260ff1SVasily Averin } 1900f1260ff1SVasily Averin return skb; 19017f678defSVasily Averin 19027f678defSVasily Averin fail: 19037f678defSVasily Averin kfree_skb(skb); 19047f678defSVasily Averin return NULL; 1905f1260ff1SVasily Averin } 1906f1260ff1SVasily Averin EXPORT_SYMBOL(skb_expand_head); 1907f1260ff1SVasily Averin 1908f1260ff1SVasily Averin /** 19091da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 19101da177e4SLinus Torvalds * @skb: buffer to copy 19111da177e4SLinus Torvalds * @newheadroom: new free bytes at head 19121da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 19131da177e4SLinus Torvalds * @gfp_mask: allocation priority 19141da177e4SLinus Torvalds * 19151da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 19161da177e4SLinus Torvalds * allocate additional space. 19171da177e4SLinus Torvalds * 19181da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 19191da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 19201da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 19211da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 19221da177e4SLinus Torvalds * 19231da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 19241da177e4SLinus Torvalds * is called from an interrupt. 19251da177e4SLinus Torvalds */ 19261da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 192786a76cafSVictor Fusco int newheadroom, int newtailroom, 1928dd0fc66fSAl Viro gfp_t gfp_mask) 19291da177e4SLinus Torvalds { 19301da177e4SLinus Torvalds /* 19311da177e4SLinus Torvalds * Allocate the copy buffer 19321da177e4SLinus Torvalds */ 1933c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1934c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1935c93bdd0eSMel Gorman NUMA_NO_NODE); 1936efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 19371da177e4SLinus Torvalds int head_copy_len, head_copy_off; 19381da177e4SLinus Torvalds 19391da177e4SLinus Torvalds if (!n) 19401da177e4SLinus Torvalds return NULL; 19411da177e4SLinus Torvalds 19421da177e4SLinus Torvalds skb_reserve(n, newheadroom); 19431da177e4SLinus Torvalds 19441da177e4SLinus Torvalds /* Set the tail pointer and length */ 19451da177e4SLinus Torvalds skb_put(n, skb->len); 19461da177e4SLinus Torvalds 1947efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 19481da177e4SLinus Torvalds head_copy_off = 0; 19491da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 19501da177e4SLinus Torvalds head_copy_len = newheadroom; 19511da177e4SLinus Torvalds else 19521da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds /* Copy the linear header and data. */ 19559f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 19569f77fad3STim Hansen skb->len + head_copy_len)); 19571da177e4SLinus Torvalds 195808303c18SIlya Lesokhin skb_copy_header(n, skb); 19591da177e4SLinus Torvalds 1960030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1961efd1e8d5SPatrick McHardy 19621da177e4SLinus Torvalds return n; 19631da177e4SLinus Torvalds } 1964b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 19651da177e4SLinus Torvalds 19661da177e4SLinus Torvalds /** 1967cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 19681da177e4SLinus Torvalds * @skb: buffer to pad 19691da177e4SLinus Torvalds * @pad: space to pad 1970cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 19711da177e4SLinus Torvalds * 19721da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 19731da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 19741da177e4SLinus Torvalds * beyond the buffer end onto the wire. 19751da177e4SLinus Torvalds * 1976cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 1977cd0a137aSFlorian Fainelli * if @free_on_error is true. 19781da177e4SLinus Torvalds */ 19791da177e4SLinus Torvalds 1980cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 19811da177e4SLinus Torvalds { 19825b057c6bSHerbert Xu int err; 19835b057c6bSHerbert Xu int ntail; 19841da177e4SLinus Torvalds 19851da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 19865b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 19871da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 19885b057c6bSHerbert Xu return 0; 19891da177e4SLinus Torvalds } 19901da177e4SLinus Torvalds 19914305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 19925b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 19935b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 19945b057c6bSHerbert Xu if (unlikely(err)) 19955b057c6bSHerbert Xu goto free_skb; 19965b057c6bSHerbert Xu } 19975b057c6bSHerbert Xu 19985b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 19995b057c6bSHerbert Xu * to be audited. 20005b057c6bSHerbert Xu */ 20015b057c6bSHerbert Xu err = skb_linearize(skb); 20025b057c6bSHerbert Xu if (unlikely(err)) 20035b057c6bSHerbert Xu goto free_skb; 20045b057c6bSHerbert Xu 20055b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 20065b057c6bSHerbert Xu return 0; 20075b057c6bSHerbert Xu 20085b057c6bSHerbert Xu free_skb: 2009cd0a137aSFlorian Fainelli if (free_on_error) 20101da177e4SLinus Torvalds kfree_skb(skb); 20115b057c6bSHerbert Xu return err; 20121da177e4SLinus Torvalds } 2013cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 20141da177e4SLinus Torvalds 20150dde3e16SIlpo Järvinen /** 20160c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 20170c7ddf36SMathias Krause * @skb: start of the buffer to use 20180c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 20190c7ddf36SMathias Krause * @len: amount of data to add 20200c7ddf36SMathias Krause * 20210c7ddf36SMathias Krause * This function extends the used data area of the potentially 20220c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 20230c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 20240c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 20250c7ddf36SMathias Krause * returned. 20260c7ddf36SMathias Krause */ 20270c7ddf36SMathias Krause 20284df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 20290c7ddf36SMathias Krause { 20300c7ddf36SMathias Krause if (tail != skb) { 20310c7ddf36SMathias Krause skb->data_len += len; 20320c7ddf36SMathias Krause skb->len += len; 20330c7ddf36SMathias Krause } 20340c7ddf36SMathias Krause return skb_put(tail, len); 20350c7ddf36SMathias Krause } 20360c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 20370c7ddf36SMathias Krause 20380c7ddf36SMathias Krause /** 20390dde3e16SIlpo Järvinen * skb_put - add data to a buffer 20400dde3e16SIlpo Järvinen * @skb: buffer to use 20410dde3e16SIlpo Järvinen * @len: amount of data to add 20420dde3e16SIlpo Järvinen * 20430dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 20440dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 20450dde3e16SIlpo Järvinen * first byte of the extra data is returned. 20460dde3e16SIlpo Järvinen */ 20474df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 20480dde3e16SIlpo Järvinen { 20494df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 20500dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 20510dde3e16SIlpo Järvinen skb->tail += len; 20520dde3e16SIlpo Järvinen skb->len += len; 20530dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 20540dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 20550dde3e16SIlpo Järvinen return tmp; 20560dde3e16SIlpo Järvinen } 20570dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 20580dde3e16SIlpo Järvinen 20596be8ac2fSIlpo Järvinen /** 2060c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 2061c2aa270aSIlpo Järvinen * @skb: buffer to use 2062c2aa270aSIlpo Järvinen * @len: amount of data to add 2063c2aa270aSIlpo Järvinen * 2064c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 2065c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 2066c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 2067c2aa270aSIlpo Järvinen */ 2068d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 2069c2aa270aSIlpo Järvinen { 2070c2aa270aSIlpo Järvinen skb->data -= len; 2071c2aa270aSIlpo Järvinen skb->len += len; 2072c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 2073c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 2074c2aa270aSIlpo Järvinen return skb->data; 2075c2aa270aSIlpo Järvinen } 2076c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 2077c2aa270aSIlpo Järvinen 2078c2aa270aSIlpo Järvinen /** 20796be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 20806be8ac2fSIlpo Järvinen * @skb: buffer to use 20816be8ac2fSIlpo Järvinen * @len: amount of data to remove 20826be8ac2fSIlpo Järvinen * 20836be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 20846be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 20856be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 20866be8ac2fSIlpo Järvinen * the old data. 20876be8ac2fSIlpo Järvinen */ 2088af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 20896be8ac2fSIlpo Järvinen { 209047d29646SDavid S. Miller return skb_pull_inline(skb, len); 20916be8ac2fSIlpo Järvinen } 20926be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 20936be8ac2fSIlpo Järvinen 2094419ae74eSIlpo Järvinen /** 209513244cccSLuiz Augusto von Dentz * skb_pull_data - remove data from the start of a buffer returning its 209613244cccSLuiz Augusto von Dentz * original position. 209713244cccSLuiz Augusto von Dentz * @skb: buffer to use 209813244cccSLuiz Augusto von Dentz * @len: amount of data to remove 209913244cccSLuiz Augusto von Dentz * 210013244cccSLuiz Augusto von Dentz * This function removes data from the start of a buffer, returning 210113244cccSLuiz Augusto von Dentz * the memory to the headroom. A pointer to the original data in the buffer 210213244cccSLuiz Augusto von Dentz * is returned after checking if there is enough data to pull. Once the 210313244cccSLuiz Augusto von Dentz * data has been pulled future pushes will overwrite the old data. 210413244cccSLuiz Augusto von Dentz */ 210513244cccSLuiz Augusto von Dentz void *skb_pull_data(struct sk_buff *skb, size_t len) 210613244cccSLuiz Augusto von Dentz { 210713244cccSLuiz Augusto von Dentz void *data = skb->data; 210813244cccSLuiz Augusto von Dentz 210913244cccSLuiz Augusto von Dentz if (skb->len < len) 211013244cccSLuiz Augusto von Dentz return NULL; 211113244cccSLuiz Augusto von Dentz 211213244cccSLuiz Augusto von Dentz skb_pull(skb, len); 211313244cccSLuiz Augusto von Dentz 211413244cccSLuiz Augusto von Dentz return data; 211513244cccSLuiz Augusto von Dentz } 211613244cccSLuiz Augusto von Dentz EXPORT_SYMBOL(skb_pull_data); 211713244cccSLuiz Augusto von Dentz 211813244cccSLuiz Augusto von Dentz /** 2119419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 2120419ae74eSIlpo Järvinen * @skb: buffer to alter 2121419ae74eSIlpo Järvinen * @len: new length 2122419ae74eSIlpo Järvinen * 2123419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 2124419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 2125419ae74eSIlpo Järvinen * The skb must be linear. 2126419ae74eSIlpo Järvinen */ 2127419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 2128419ae74eSIlpo Järvinen { 2129419ae74eSIlpo Järvinen if (skb->len > len) 2130419ae74eSIlpo Järvinen __skb_trim(skb, len); 2131419ae74eSIlpo Järvinen } 2132419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 2133419ae74eSIlpo Järvinen 21343cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 21351da177e4SLinus Torvalds */ 21361da177e4SLinus Torvalds 21373cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 21381da177e4SLinus Torvalds { 213927b437c8SHerbert Xu struct sk_buff **fragp; 214027b437c8SHerbert Xu struct sk_buff *frag; 21411da177e4SLinus Torvalds int offset = skb_headlen(skb); 21421da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 21431da177e4SLinus Torvalds int i; 214427b437c8SHerbert Xu int err; 214527b437c8SHerbert Xu 214627b437c8SHerbert Xu if (skb_cloned(skb) && 214727b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 214827b437c8SHerbert Xu return err; 21491da177e4SLinus Torvalds 2150f4d26fb3SHerbert Xu i = 0; 2151f4d26fb3SHerbert Xu if (offset >= len) 2152f4d26fb3SHerbert Xu goto drop_pages; 2153f4d26fb3SHerbert Xu 2154f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 21559e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 215627b437c8SHerbert Xu 215727b437c8SHerbert Xu if (end < len) { 21581da177e4SLinus Torvalds offset = end; 215927b437c8SHerbert Xu continue; 21601da177e4SLinus Torvalds } 21611da177e4SLinus Torvalds 21629e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 216327b437c8SHerbert Xu 2164f4d26fb3SHerbert Xu drop_pages: 216527b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 216627b437c8SHerbert Xu 216727b437c8SHerbert Xu for (; i < nfrags; i++) 2168ea2ab693SIan Campbell skb_frag_unref(skb, i); 216927b437c8SHerbert Xu 217021dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 217127b437c8SHerbert Xu skb_drop_fraglist(skb); 2172f4d26fb3SHerbert Xu goto done; 217327b437c8SHerbert Xu } 217427b437c8SHerbert Xu 217527b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 217627b437c8SHerbert Xu fragp = &frag->next) { 217727b437c8SHerbert Xu int end = offset + frag->len; 217827b437c8SHerbert Xu 217927b437c8SHerbert Xu if (skb_shared(frag)) { 218027b437c8SHerbert Xu struct sk_buff *nfrag; 218127b437c8SHerbert Xu 218227b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 218327b437c8SHerbert Xu if (unlikely(!nfrag)) 218427b437c8SHerbert Xu return -ENOMEM; 218527b437c8SHerbert Xu 218627b437c8SHerbert Xu nfrag->next = frag->next; 218785bb2a60SEric Dumazet consume_skb(frag); 218827b437c8SHerbert Xu frag = nfrag; 218927b437c8SHerbert Xu *fragp = frag; 219027b437c8SHerbert Xu } 219127b437c8SHerbert Xu 219227b437c8SHerbert Xu if (end < len) { 219327b437c8SHerbert Xu offset = end; 219427b437c8SHerbert Xu continue; 219527b437c8SHerbert Xu } 219627b437c8SHerbert Xu 219727b437c8SHerbert Xu if (end > len && 219827b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 219927b437c8SHerbert Xu return err; 220027b437c8SHerbert Xu 220127b437c8SHerbert Xu if (frag->next) 220227b437c8SHerbert Xu skb_drop_list(&frag->next); 220327b437c8SHerbert Xu break; 220427b437c8SHerbert Xu } 220527b437c8SHerbert Xu 2206f4d26fb3SHerbert Xu done: 220727b437c8SHerbert Xu if (len > skb_headlen(skb)) { 22081da177e4SLinus Torvalds skb->data_len -= skb->len - len; 22091da177e4SLinus Torvalds skb->len = len; 22101da177e4SLinus Torvalds } else { 22111da177e4SLinus Torvalds skb->len = len; 22121da177e4SLinus Torvalds skb->data_len = 0; 221327a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 22141da177e4SLinus Torvalds } 22151da177e4SLinus Torvalds 2216c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2217c21b48ccSEric Dumazet skb_condense(skb); 22181da177e4SLinus Torvalds return 0; 22191da177e4SLinus Torvalds } 2220b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 22211da177e4SLinus Torvalds 222288078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 222388078d98SEric Dumazet */ 222488078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 222588078d98SEric Dumazet { 222688078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 222788078d98SEric Dumazet int delta = skb->len - len; 222888078d98SEric Dumazet 2229d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2230d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2231d55bef50SDimitris Michailidis len); 223254970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 223354970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 223454970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 223554970a2fSVasily Averin 223654970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 223754970a2fSVasily Averin return -EINVAL; 223888078d98SEric Dumazet } 223988078d98SEric Dumazet return __pskb_trim(skb, len); 224088078d98SEric Dumazet } 224188078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 224288078d98SEric Dumazet 22431da177e4SLinus Torvalds /** 22441da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 22451da177e4SLinus Torvalds * @skb: buffer to reallocate 22461da177e4SLinus Torvalds * @delta: number of bytes to advance tail 22471da177e4SLinus Torvalds * 22481da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 22491da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 22501da177e4SLinus Torvalds * data from fragmented part. 22511da177e4SLinus Torvalds * 22521da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 22531da177e4SLinus Torvalds * 22541da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 22551da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 22561da177e4SLinus Torvalds * 22571da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 22581da177e4SLinus Torvalds * reloaded after call to this function. 22591da177e4SLinus Torvalds */ 22601da177e4SLinus Torvalds 22611da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 22621da177e4SLinus Torvalds * when it is necessary. 22631da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 22641da177e4SLinus Torvalds * 2. It may change skb pointers. 22651da177e4SLinus Torvalds * 22661da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 22671da177e4SLinus Torvalds */ 2268af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 22691da177e4SLinus Torvalds { 22701da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 22711da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 22721da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 22731da177e4SLinus Torvalds */ 22744305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 22751da177e4SLinus Torvalds 22761da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 22771da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 22781da177e4SLinus Torvalds GFP_ATOMIC)) 22791da177e4SLinus Torvalds return NULL; 22801da177e4SLinus Torvalds } 22811da177e4SLinus Torvalds 22829f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 22839f77fad3STim Hansen skb_tail_pointer(skb), delta)); 22841da177e4SLinus Torvalds 22851da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 22861da177e4SLinus Torvalds * size of pulled pages. Superb. 22871da177e4SLinus Torvalds */ 228821dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 22891da177e4SLinus Torvalds goto pull_pages; 22901da177e4SLinus Torvalds 22911da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 22921da177e4SLinus Torvalds eat = delta; 22931da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 22949e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22959e903e08SEric Dumazet 22969e903e08SEric Dumazet if (size >= eat) 22971da177e4SLinus Torvalds goto pull_pages; 22989e903e08SEric Dumazet eat -= size; 22991da177e4SLinus Torvalds } 23001da177e4SLinus Torvalds 23011da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 230209001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 23031da177e4SLinus Torvalds * but taking into account that pulling is expected to 23041da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 23051da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 23061da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 23071da177e4SLinus Torvalds */ 23081da177e4SLinus Torvalds if (eat) { 23091da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 23101da177e4SLinus Torvalds struct sk_buff *clone = NULL; 23111da177e4SLinus Torvalds struct sk_buff *insp = NULL; 23121da177e4SLinus Torvalds 23131da177e4SLinus Torvalds do { 23141da177e4SLinus Torvalds if (list->len <= eat) { 23151da177e4SLinus Torvalds /* Eaten as whole. */ 23161da177e4SLinus Torvalds eat -= list->len; 23171da177e4SLinus Torvalds list = list->next; 23181da177e4SLinus Torvalds insp = list; 23191da177e4SLinus Torvalds } else { 23201da177e4SLinus Torvalds /* Eaten partially. */ 23211da177e4SLinus Torvalds 23221da177e4SLinus Torvalds if (skb_shared(list)) { 23231da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 23241da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 23251da177e4SLinus Torvalds if (!clone) 23261da177e4SLinus Torvalds return NULL; 23271da177e4SLinus Torvalds insp = list->next; 23281da177e4SLinus Torvalds list = clone; 23291da177e4SLinus Torvalds } else { 23301da177e4SLinus Torvalds /* This may be pulled without 23311da177e4SLinus Torvalds * problems. */ 23321da177e4SLinus Torvalds insp = list; 23331da177e4SLinus Torvalds } 23341da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 23351da177e4SLinus Torvalds kfree_skb(clone); 23361da177e4SLinus Torvalds return NULL; 23371da177e4SLinus Torvalds } 23381da177e4SLinus Torvalds break; 23391da177e4SLinus Torvalds } 23401da177e4SLinus Torvalds } while (eat); 23411da177e4SLinus Torvalds 23421da177e4SLinus Torvalds /* Free pulled out fragments. */ 23431da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 23441da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 2345ef527f96SEric Dumazet consume_skb(list); 23461da177e4SLinus Torvalds } 23471da177e4SLinus Torvalds /* And insert new clone at head. */ 23481da177e4SLinus Torvalds if (clone) { 23491da177e4SLinus Torvalds clone->next = list; 23501da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 23511da177e4SLinus Torvalds } 23521da177e4SLinus Torvalds } 23531da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 23541da177e4SLinus Torvalds 23551da177e4SLinus Torvalds pull_pages: 23561da177e4SLinus Torvalds eat = delta; 23571da177e4SLinus Torvalds k = 0; 23581da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 23599e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 23609e903e08SEric Dumazet 23619e903e08SEric Dumazet if (size <= eat) { 2362ea2ab693SIan Campbell skb_frag_unref(skb, i); 23639e903e08SEric Dumazet eat -= size; 23641da177e4SLinus Torvalds } else { 2365b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2366b54c9d5bSJonathan Lemon 2367b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 23681da177e4SLinus Torvalds if (eat) { 2369b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2370b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 23713ccc6c6fSlinzhang if (!i) 23723ccc6c6fSlinzhang goto end; 23731da177e4SLinus Torvalds eat = 0; 23741da177e4SLinus Torvalds } 23751da177e4SLinus Torvalds k++; 23761da177e4SLinus Torvalds } 23771da177e4SLinus Torvalds } 23781da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 23791da177e4SLinus Torvalds 23803ccc6c6fSlinzhang end: 23811da177e4SLinus Torvalds skb->tail += delta; 23821da177e4SLinus Torvalds skb->data_len -= delta; 23831da177e4SLinus Torvalds 23841f8b977aSWillem de Bruijn if (!skb->data_len) 23851f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 23861f8b977aSWillem de Bruijn 238727a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 23881da177e4SLinus Torvalds } 2389b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 23901da177e4SLinus Torvalds 239122019b17SEric Dumazet /** 239222019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 239322019b17SEric Dumazet * @skb: source skb 239422019b17SEric Dumazet * @offset: offset in source 239522019b17SEric Dumazet * @to: destination buffer 239622019b17SEric Dumazet * @len: number of bytes to copy 239722019b17SEric Dumazet * 239822019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 239922019b17SEric Dumazet * destination buffer. 240022019b17SEric Dumazet * 240122019b17SEric Dumazet * CAUTION ! : 240222019b17SEric Dumazet * If its prototype is ever changed, 240322019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 240422019b17SEric Dumazet * since it is called from BPF assembly code. 240522019b17SEric Dumazet */ 24061da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 24071da177e4SLinus Torvalds { 24081a028e50SDavid S. Miller int start = skb_headlen(skb); 2409fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2410fbb398a8SDavid S. Miller int i, copy; 24111da177e4SLinus Torvalds 24121da177e4SLinus Torvalds if (offset > (int)skb->len - len) 24131da177e4SLinus Torvalds goto fault; 24141da177e4SLinus Torvalds 24151da177e4SLinus Torvalds /* Copy header. */ 24161a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 24171da177e4SLinus Torvalds if (copy > len) 24181da177e4SLinus Torvalds copy = len; 2419d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 24201da177e4SLinus Torvalds if ((len -= copy) == 0) 24211da177e4SLinus Torvalds return 0; 24221da177e4SLinus Torvalds offset += copy; 24231da177e4SLinus Torvalds to += copy; 24241da177e4SLinus Torvalds } 24251da177e4SLinus Torvalds 24261da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 24271a028e50SDavid S. Miller int end; 242851c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 24291da177e4SLinus Torvalds 2430547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24311a028e50SDavid S. Miller 243251c56b00SEric Dumazet end = start + skb_frag_size(f); 24331da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2434c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2435c613c209SWillem de Bruijn struct page *p; 24361da177e4SLinus Torvalds u8 *vaddr; 24371da177e4SLinus Torvalds 24381da177e4SLinus Torvalds if (copy > len) 24391da177e4SLinus Torvalds copy = len; 24401da177e4SLinus Torvalds 2441c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2442b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2443c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2444c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2445c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 244651c56b00SEric Dumazet kunmap_atomic(vaddr); 2447c613c209SWillem de Bruijn } 24481da177e4SLinus Torvalds 24491da177e4SLinus Torvalds if ((len -= copy) == 0) 24501da177e4SLinus Torvalds return 0; 24511da177e4SLinus Torvalds offset += copy; 24521da177e4SLinus Torvalds to += copy; 24531da177e4SLinus Torvalds } 24541a028e50SDavid S. Miller start = end; 24551da177e4SLinus Torvalds } 24561da177e4SLinus Torvalds 2457fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 24581a028e50SDavid S. Miller int end; 24591da177e4SLinus Torvalds 2460547b792cSIlpo Järvinen WARN_ON(start > offset + len); 24611a028e50SDavid S. Miller 2462fbb398a8SDavid S. Miller end = start + frag_iter->len; 24631da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 24641da177e4SLinus Torvalds if (copy > len) 24651da177e4SLinus Torvalds copy = len; 2466fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 24671da177e4SLinus Torvalds goto fault; 24681da177e4SLinus Torvalds if ((len -= copy) == 0) 24691da177e4SLinus Torvalds return 0; 24701da177e4SLinus Torvalds offset += copy; 24711da177e4SLinus Torvalds to += copy; 24721da177e4SLinus Torvalds } 24731a028e50SDavid S. Miller start = end; 24741da177e4SLinus Torvalds } 2475a6686f2fSShirley Ma 24761da177e4SLinus Torvalds if (!len) 24771da177e4SLinus Torvalds return 0; 24781da177e4SLinus Torvalds 24791da177e4SLinus Torvalds fault: 24801da177e4SLinus Torvalds return -EFAULT; 24811da177e4SLinus Torvalds } 2482b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 24831da177e4SLinus Torvalds 24849c55e01cSJens Axboe /* 24859c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 24869c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 24879c55e01cSJens Axboe */ 24889c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 24899c55e01cSJens Axboe { 24908b9d3728SJarek Poplawski put_page(spd->pages[i]); 24918b9d3728SJarek Poplawski } 24929c55e01cSJens Axboe 2493a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 24944fb66994SJarek Poplawski unsigned int *offset, 249518aafc62SEric Dumazet struct sock *sk) 24968b9d3728SJarek Poplawski { 24975640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 24988b9d3728SJarek Poplawski 24995640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 25008b9d3728SJarek Poplawski return NULL; 25014fb66994SJarek Poplawski 25025640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 25034fb66994SJarek Poplawski 25045640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 25055640f768SEric Dumazet page_address(page) + *offset, *len); 25065640f768SEric Dumazet *offset = pfrag->offset; 25075640f768SEric Dumazet pfrag->offset += *len; 25084fb66994SJarek Poplawski 25095640f768SEric Dumazet return pfrag->page; 25109c55e01cSJens Axboe } 25119c55e01cSJens Axboe 251241c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 251341c73a0dSEric Dumazet struct page *page, 251441c73a0dSEric Dumazet unsigned int offset) 251541c73a0dSEric Dumazet { 251641c73a0dSEric Dumazet return spd->nr_pages && 251741c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 251841c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 251941c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 252041c73a0dSEric Dumazet } 252141c73a0dSEric Dumazet 25229c55e01cSJens Axboe /* 25239c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 25249c55e01cSJens Axboe */ 2525a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 252635f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 25274fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 252818aafc62SEric Dumazet bool linear, 25297a67e56fSJarek Poplawski struct sock *sk) 25309c55e01cSJens Axboe { 253141c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2532a108d5f3SDavid S. Miller return true; 25339c55e01cSJens Axboe 25348b9d3728SJarek Poplawski if (linear) { 253518aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 25368b9d3728SJarek Poplawski if (!page) 2537a108d5f3SDavid S. Miller return true; 253841c73a0dSEric Dumazet } 253941c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 254041c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2541a108d5f3SDavid S. Miller return false; 254241c73a0dSEric Dumazet } 25438b9d3728SJarek Poplawski get_page(page); 25449c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 25454fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 25469c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 25479c55e01cSJens Axboe spd->nr_pages++; 25488b9d3728SJarek Poplawski 2549a108d5f3SDavid S. Miller return false; 25509c55e01cSJens Axboe } 25519c55e01cSJens Axboe 2552a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 25532870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 255418aafc62SEric Dumazet unsigned int *len, 2555d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 255635f3d14dSJens Axboe struct sock *sk, 255735f3d14dSJens Axboe struct pipe_inode_info *pipe) 25589c55e01cSJens Axboe { 25592870c43dSOctavian Purdila if (!*len) 2560a108d5f3SDavid S. Miller return true; 25619c55e01cSJens Axboe 25622870c43dSOctavian Purdila /* skip this segment if already processed */ 25632870c43dSOctavian Purdila if (*off >= plen) { 25642870c43dSOctavian Purdila *off -= plen; 2565a108d5f3SDavid S. Miller return false; 25662870c43dSOctavian Purdila } 25672870c43dSOctavian Purdila 25682870c43dSOctavian Purdila /* ignore any bits we already processed */ 25699ca1b22dSEric Dumazet poff += *off; 25709ca1b22dSEric Dumazet plen -= *off; 25712870c43dSOctavian Purdila *off = 0; 25722870c43dSOctavian Purdila 257318aafc62SEric Dumazet do { 257418aafc62SEric Dumazet unsigned int flen = min(*len, plen); 25752870c43dSOctavian Purdila 257618aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 257718aafc62SEric Dumazet linear, sk)) 2578a108d5f3SDavid S. Miller return true; 257918aafc62SEric Dumazet poff += flen; 258018aafc62SEric Dumazet plen -= flen; 25812870c43dSOctavian Purdila *len -= flen; 258218aafc62SEric Dumazet } while (*len && plen); 25832870c43dSOctavian Purdila 2584a108d5f3SDavid S. Miller return false; 2585db43a282SOctavian Purdila } 25869c55e01cSJens Axboe 25879c55e01cSJens Axboe /* 2588a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 25892870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 25909c55e01cSJens Axboe */ 2591a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 259235f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 259335f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 25942870c43dSOctavian Purdila { 25952870c43dSOctavian Purdila int seg; 2596fa9835e5STom Herbert struct sk_buff *iter; 25979c55e01cSJens Axboe 25981d0c0b32SEric Dumazet /* map the linear part : 25992996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 26002996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 26012996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 26029c55e01cSJens Axboe */ 26032870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 26042870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 26052870c43dSOctavian Purdila skb_headlen(skb), 260618aafc62SEric Dumazet offset, len, spd, 26073a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 26081d0c0b32SEric Dumazet sk, pipe)) 2609a108d5f3SDavid S. Miller return true; 26109c55e01cSJens Axboe 26119c55e01cSJens Axboe /* 26129c55e01cSJens Axboe * then map the fragments 26139c55e01cSJens Axboe */ 26149c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 26159c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 26169c55e01cSJens Axboe 2617ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2618b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 261918aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2620a108d5f3SDavid S. Miller return true; 26219c55e01cSJens Axboe } 26229c55e01cSJens Axboe 2623fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2624fa9835e5STom Herbert if (*offset >= iter->len) { 2625fa9835e5STom Herbert *offset -= iter->len; 2626fa9835e5STom Herbert continue; 2627fa9835e5STom Herbert } 2628fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2629fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2630fa9835e5STom Herbert * case. 2631fa9835e5STom Herbert */ 2632fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2633fa9835e5STom Herbert return true; 2634fa9835e5STom Herbert } 2635fa9835e5STom Herbert 2636a108d5f3SDavid S. Miller return false; 26379c55e01cSJens Axboe } 26389c55e01cSJens Axboe 26399c55e01cSJens Axboe /* 26409c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2641fa9835e5STom Herbert * the fragments, and the frag list. 26429c55e01cSJens Axboe */ 2643a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 26449c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 264525869262SAl Viro unsigned int flags) 26469c55e01cSJens Axboe { 264741c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 264841c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 26499c55e01cSJens Axboe struct splice_pipe_desc spd = { 26509c55e01cSJens Axboe .pages = pages, 26519c55e01cSJens Axboe .partial = partial, 2652047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 265328a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 26549c55e01cSJens Axboe .spd_release = sock_spd_release, 26559c55e01cSJens Axboe }; 265635f3d14dSJens Axboe int ret = 0; 265735f3d14dSJens Axboe 2658fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 26599c55e01cSJens Axboe 2660a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 266125869262SAl Viro ret = splice_to_pipe(pipe, &spd); 26629c55e01cSJens Axboe 266335f3d14dSJens Axboe return ret; 26649c55e01cSJens Axboe } 26652b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 26669c55e01cSJens Axboe 26670739cd28SCong Wang static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg, 26680739cd28SCong Wang struct kvec *vec, size_t num, size_t size) 26690739cd28SCong Wang { 26700739cd28SCong Wang struct socket *sock = sk->sk_socket; 26710739cd28SCong Wang 26720739cd28SCong Wang if (!sock) 26730739cd28SCong Wang return -EINVAL; 26740739cd28SCong Wang return kernel_sendmsg(sock, msg, vec, num, size); 26750739cd28SCong Wang } 26760739cd28SCong Wang 26770739cd28SCong Wang static int sendpage_unlocked(struct sock *sk, struct page *page, int offset, 26780739cd28SCong Wang size_t size, int flags) 26790739cd28SCong Wang { 26800739cd28SCong Wang struct socket *sock = sk->sk_socket; 26810739cd28SCong Wang 26820739cd28SCong Wang if (!sock) 26830739cd28SCong Wang return -EINVAL; 26840739cd28SCong Wang return kernel_sendpage(sock, page, offset, size, flags); 26850739cd28SCong Wang } 26860739cd28SCong Wang 26870739cd28SCong Wang typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg, 26880739cd28SCong Wang struct kvec *vec, size_t num, size_t size); 26890739cd28SCong Wang typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset, 26900739cd28SCong Wang size_t size, int flags); 26910739cd28SCong Wang static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, 26920739cd28SCong Wang int len, sendmsg_func sendmsg, sendpage_func sendpage) 269320bf50deSTom Herbert { 269420bf50deSTom Herbert unsigned int orig_len = len; 269520bf50deSTom Herbert struct sk_buff *head = skb; 269620bf50deSTom Herbert unsigned short fragidx; 269720bf50deSTom Herbert int slen, ret; 269820bf50deSTom Herbert 269920bf50deSTom Herbert do_frag_list: 270020bf50deSTom Herbert 270120bf50deSTom Herbert /* Deal with head data */ 270220bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 270320bf50deSTom Herbert struct kvec kv; 270420bf50deSTom Herbert struct msghdr msg; 270520bf50deSTom Herbert 270620bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 270720bf50deSTom Herbert kv.iov_base = skb->data + offset; 2708db5980d8SJohn Fastabend kv.iov_len = slen; 270920bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2710bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 271120bf50deSTom Herbert 27120739cd28SCong Wang ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked, 27130739cd28SCong Wang sendmsg_unlocked, sk, &msg, &kv, 1, slen); 271420bf50deSTom Herbert if (ret <= 0) 271520bf50deSTom Herbert goto error; 271620bf50deSTom Herbert 271720bf50deSTom Herbert offset += ret; 271820bf50deSTom Herbert len -= ret; 271920bf50deSTom Herbert } 272020bf50deSTom Herbert 272120bf50deSTom Herbert /* All the data was skb head? */ 272220bf50deSTom Herbert if (!len) 272320bf50deSTom Herbert goto out; 272420bf50deSTom Herbert 272520bf50deSTom Herbert /* Make offset relative to start of frags */ 272620bf50deSTom Herbert offset -= skb_headlen(skb); 272720bf50deSTom Herbert 272820bf50deSTom Herbert /* Find where we are in frag list */ 272920bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 273020bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 273120bf50deSTom Herbert 2732d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 273320bf50deSTom Herbert break; 273420bf50deSTom Herbert 2735d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 273620bf50deSTom Herbert } 273720bf50deSTom Herbert 273820bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 273920bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 274020bf50deSTom Herbert 2741d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 274220bf50deSTom Herbert 274320bf50deSTom Herbert while (slen) { 27440739cd28SCong Wang ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked, 27450739cd28SCong Wang sendpage_unlocked, sk, 27460739cd28SCong Wang skb_frag_page(frag), 2747b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 274820bf50deSTom Herbert slen, MSG_DONTWAIT); 274920bf50deSTom Herbert if (ret <= 0) 275020bf50deSTom Herbert goto error; 275120bf50deSTom Herbert 275220bf50deSTom Herbert len -= ret; 275320bf50deSTom Herbert offset += ret; 275420bf50deSTom Herbert slen -= ret; 275520bf50deSTom Herbert } 275620bf50deSTom Herbert 275720bf50deSTom Herbert offset = 0; 275820bf50deSTom Herbert } 275920bf50deSTom Herbert 276020bf50deSTom Herbert if (len) { 276120bf50deSTom Herbert /* Process any frag lists */ 276220bf50deSTom Herbert 276320bf50deSTom Herbert if (skb == head) { 276420bf50deSTom Herbert if (skb_has_frag_list(skb)) { 276520bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 276620bf50deSTom Herbert goto do_frag_list; 276720bf50deSTom Herbert } 276820bf50deSTom Herbert } else if (skb->next) { 276920bf50deSTom Herbert skb = skb->next; 277020bf50deSTom Herbert goto do_frag_list; 277120bf50deSTom Herbert } 277220bf50deSTom Herbert } 277320bf50deSTom Herbert 277420bf50deSTom Herbert out: 277520bf50deSTom Herbert return orig_len - len; 277620bf50deSTom Herbert 277720bf50deSTom Herbert error: 277820bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 277920bf50deSTom Herbert } 27800739cd28SCong Wang 27810739cd28SCong Wang /* Send skb data on a socket. Socket must be locked. */ 27820739cd28SCong Wang int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 27830739cd28SCong Wang int len) 27840739cd28SCong Wang { 27850739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked, 27860739cd28SCong Wang kernel_sendpage_locked); 27870739cd28SCong Wang } 278820bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 278920bf50deSTom Herbert 27900739cd28SCong Wang /* Send skb data on a socket. Socket must be unlocked. */ 27910739cd28SCong Wang int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 27920739cd28SCong Wang { 27930739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked, 27940739cd28SCong Wang sendpage_unlocked); 27950739cd28SCong Wang } 27960739cd28SCong Wang 2797357b40a1SHerbert Xu /** 2798357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2799357b40a1SHerbert Xu * @skb: destination buffer 2800357b40a1SHerbert Xu * @offset: offset in destination 2801357b40a1SHerbert Xu * @from: source buffer 2802357b40a1SHerbert Xu * @len: number of bytes to copy 2803357b40a1SHerbert Xu * 2804357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2805357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2806357b40a1SHerbert Xu * traversing fragment lists and such. 2807357b40a1SHerbert Xu */ 2808357b40a1SHerbert Xu 28090c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2810357b40a1SHerbert Xu { 28111a028e50SDavid S. Miller int start = skb_headlen(skb); 2812fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2813fbb398a8SDavid S. Miller int i, copy; 2814357b40a1SHerbert Xu 2815357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2816357b40a1SHerbert Xu goto fault; 2817357b40a1SHerbert Xu 28181a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2819357b40a1SHerbert Xu if (copy > len) 2820357b40a1SHerbert Xu copy = len; 282127d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2822357b40a1SHerbert Xu if ((len -= copy) == 0) 2823357b40a1SHerbert Xu return 0; 2824357b40a1SHerbert Xu offset += copy; 2825357b40a1SHerbert Xu from += copy; 2826357b40a1SHerbert Xu } 2827357b40a1SHerbert Xu 2828357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2829357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 28301a028e50SDavid S. Miller int end; 2831357b40a1SHerbert Xu 2832547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28331a028e50SDavid S. Miller 28349e903e08SEric Dumazet end = start + skb_frag_size(frag); 2835357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2836c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2837c613c209SWillem de Bruijn struct page *p; 2838357b40a1SHerbert Xu u8 *vaddr; 2839357b40a1SHerbert Xu 2840357b40a1SHerbert Xu if (copy > len) 2841357b40a1SHerbert Xu copy = len; 2842357b40a1SHerbert Xu 2843c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2844b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2845c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2846c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2847c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 284851c56b00SEric Dumazet kunmap_atomic(vaddr); 2849c613c209SWillem de Bruijn } 2850357b40a1SHerbert Xu 2851357b40a1SHerbert Xu if ((len -= copy) == 0) 2852357b40a1SHerbert Xu return 0; 2853357b40a1SHerbert Xu offset += copy; 2854357b40a1SHerbert Xu from += copy; 2855357b40a1SHerbert Xu } 28561a028e50SDavid S. Miller start = end; 2857357b40a1SHerbert Xu } 2858357b40a1SHerbert Xu 2859fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 28601a028e50SDavid S. Miller int end; 2861357b40a1SHerbert Xu 2862547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28631a028e50SDavid S. Miller 2864fbb398a8SDavid S. Miller end = start + frag_iter->len; 2865357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 2866357b40a1SHerbert Xu if (copy > len) 2867357b40a1SHerbert Xu copy = len; 2868fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 28691a028e50SDavid S. Miller from, copy)) 2870357b40a1SHerbert Xu goto fault; 2871357b40a1SHerbert Xu if ((len -= copy) == 0) 2872357b40a1SHerbert Xu return 0; 2873357b40a1SHerbert Xu offset += copy; 2874357b40a1SHerbert Xu from += copy; 2875357b40a1SHerbert Xu } 28761a028e50SDavid S. Miller start = end; 2877357b40a1SHerbert Xu } 2878357b40a1SHerbert Xu if (!len) 2879357b40a1SHerbert Xu return 0; 2880357b40a1SHerbert Xu 2881357b40a1SHerbert Xu fault: 2882357b40a1SHerbert Xu return -EFAULT; 2883357b40a1SHerbert Xu } 2884357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 2885357b40a1SHerbert Xu 28861da177e4SLinus Torvalds /* Checksum skb data. */ 28872817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 28882817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 28891da177e4SLinus Torvalds { 28901a028e50SDavid S. Miller int start = skb_headlen(skb); 28911a028e50SDavid S. Miller int i, copy = start - offset; 2892fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 28931da177e4SLinus Torvalds int pos = 0; 28941da177e4SLinus Torvalds 28951da177e4SLinus Torvalds /* Checksum header. */ 28961da177e4SLinus Torvalds if (copy > 0) { 28971da177e4SLinus Torvalds if (copy > len) 28981da177e4SLinus Torvalds copy = len; 28992544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 29002544af03SMatteo Croce skb->data + offset, copy, csum); 29011da177e4SLinus Torvalds if ((len -= copy) == 0) 29021da177e4SLinus Torvalds return csum; 29031da177e4SLinus Torvalds offset += copy; 29041da177e4SLinus Torvalds pos = copy; 29051da177e4SLinus Torvalds } 29061da177e4SLinus Torvalds 29071da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 29081a028e50SDavid S. Miller int end; 290951c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 29101da177e4SLinus Torvalds 2911547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29121a028e50SDavid S. Miller 291351c56b00SEric Dumazet end = start + skb_frag_size(frag); 29141da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2915c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2916c613c209SWillem de Bruijn struct page *p; 291744bb9363SAl Viro __wsum csum2; 29181da177e4SLinus Torvalds u8 *vaddr; 29191da177e4SLinus Torvalds 29201da177e4SLinus Torvalds if (copy > len) 29211da177e4SLinus Torvalds copy = len; 2922c613c209SWillem de Bruijn 2923c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 2924b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 2925c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2926c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 29272544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 29282544af03SMatteo Croce csum_partial_ext, 29292544af03SMatteo Croce vaddr + p_off, p_len, 0); 293051c56b00SEric Dumazet kunmap_atomic(vaddr); 29312544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 29322544af03SMatteo Croce csum_block_add_ext, csum, 29332544af03SMatteo Croce csum2, pos, p_len); 2934c613c209SWillem de Bruijn pos += p_len; 2935c613c209SWillem de Bruijn } 2936c613c209SWillem de Bruijn 29371da177e4SLinus Torvalds if (!(len -= copy)) 29381da177e4SLinus Torvalds return csum; 29391da177e4SLinus Torvalds offset += copy; 29401da177e4SLinus Torvalds } 29411a028e50SDavid S. Miller start = end; 29421da177e4SLinus Torvalds } 29431da177e4SLinus Torvalds 2944fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 29451a028e50SDavid S. Miller int end; 29461da177e4SLinus Torvalds 2947547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29481a028e50SDavid S. Miller 2949fbb398a8SDavid S. Miller end = start + frag_iter->len; 29501da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 29515f92a738SAl Viro __wsum csum2; 29521da177e4SLinus Torvalds if (copy > len) 29531da177e4SLinus Torvalds copy = len; 29542817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 29552817a336SDaniel Borkmann copy, 0, ops); 29562544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 29572544af03SMatteo Croce csum, csum2, pos, copy); 29581da177e4SLinus Torvalds if ((len -= copy) == 0) 29591da177e4SLinus Torvalds return csum; 29601da177e4SLinus Torvalds offset += copy; 29611da177e4SLinus Torvalds pos += copy; 29621da177e4SLinus Torvalds } 29631a028e50SDavid S. Miller start = end; 29641da177e4SLinus Torvalds } 296509a62660SKris Katterjohn BUG_ON(len); 29661da177e4SLinus Torvalds 29671da177e4SLinus Torvalds return csum; 29681da177e4SLinus Torvalds } 29692817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 29702817a336SDaniel Borkmann 29712817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 29722817a336SDaniel Borkmann int len, __wsum csum) 29732817a336SDaniel Borkmann { 29742817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2975cea80ea8SDaniel Borkmann .update = csum_partial_ext, 29762817a336SDaniel Borkmann .combine = csum_block_add_ext, 29772817a336SDaniel Borkmann }; 29782817a336SDaniel Borkmann 29792817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 29802817a336SDaniel Borkmann } 2981b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 29821da177e4SLinus Torvalds 29831da177e4SLinus Torvalds /* Both of above in one bottle. */ 29841da177e4SLinus Torvalds 298581d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 29868d5930dfSAl Viro u8 *to, int len) 29871da177e4SLinus Torvalds { 29881a028e50SDavid S. Miller int start = skb_headlen(skb); 29891a028e50SDavid S. Miller int i, copy = start - offset; 2990fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 29911da177e4SLinus Torvalds int pos = 0; 29928d5930dfSAl Viro __wsum csum = 0; 29931da177e4SLinus Torvalds 29941da177e4SLinus Torvalds /* Copy header. */ 29951da177e4SLinus Torvalds if (copy > 0) { 29961da177e4SLinus Torvalds if (copy > len) 29971da177e4SLinus Torvalds copy = len; 29981da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 2999cc44c17bSAl Viro copy); 30001da177e4SLinus Torvalds if ((len -= copy) == 0) 30011da177e4SLinus Torvalds return csum; 30021da177e4SLinus Torvalds offset += copy; 30031da177e4SLinus Torvalds to += copy; 30041da177e4SLinus Torvalds pos = copy; 30051da177e4SLinus Torvalds } 30061da177e4SLinus Torvalds 30071da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 30081a028e50SDavid S. Miller int end; 30091da177e4SLinus Torvalds 3010547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30111a028e50SDavid S. Miller 30129e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 30131da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 3014c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3015c613c209SWillem de Bruijn u32 p_off, p_len, copied; 3016c613c209SWillem de Bruijn struct page *p; 30175084205fSAl Viro __wsum csum2; 30181da177e4SLinus Torvalds u8 *vaddr; 30191da177e4SLinus Torvalds 30201da177e4SLinus Torvalds if (copy > len) 30211da177e4SLinus Torvalds copy = len; 3022c613c209SWillem de Bruijn 3023c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 3024b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 3025c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 3026c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 3027c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 3028c613c209SWillem de Bruijn to + copied, 3029cc44c17bSAl Viro p_len); 303051c56b00SEric Dumazet kunmap_atomic(vaddr); 30311da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 3032c613c209SWillem de Bruijn pos += p_len; 3033c613c209SWillem de Bruijn } 3034c613c209SWillem de Bruijn 30351da177e4SLinus Torvalds if (!(len -= copy)) 30361da177e4SLinus Torvalds return csum; 30371da177e4SLinus Torvalds offset += copy; 30381da177e4SLinus Torvalds to += copy; 30391da177e4SLinus Torvalds } 30401a028e50SDavid S. Miller start = end; 30411da177e4SLinus Torvalds } 30421da177e4SLinus Torvalds 3043fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 304481d77662SAl Viro __wsum csum2; 30451a028e50SDavid S. Miller int end; 30461da177e4SLinus Torvalds 3047547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30481a028e50SDavid S. Miller 3049fbb398a8SDavid S. Miller end = start + frag_iter->len; 30501da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 30511da177e4SLinus Torvalds if (copy > len) 30521da177e4SLinus Torvalds copy = len; 3053fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 30541a028e50SDavid S. Miller offset - start, 30558d5930dfSAl Viro to, copy); 30561da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 30571da177e4SLinus Torvalds if ((len -= copy) == 0) 30581da177e4SLinus Torvalds return csum; 30591da177e4SLinus Torvalds offset += copy; 30601da177e4SLinus Torvalds to += copy; 30611da177e4SLinus Torvalds pos += copy; 30621da177e4SLinus Torvalds } 30631a028e50SDavid S. Miller start = end; 30641da177e4SLinus Torvalds } 306509a62660SKris Katterjohn BUG_ON(len); 30661da177e4SLinus Torvalds return csum; 30671da177e4SLinus Torvalds } 3068b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 30691da177e4SLinus Torvalds 307049f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 307149f8e832SCong Wang { 307249f8e832SCong Wang __sum16 sum; 307349f8e832SCong Wang 307449f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 307514641931SCong Wang /* See comments in __skb_checksum_complete(). */ 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 if (!skb_shared(skb)) 308249f8e832SCong Wang skb->csum_valid = !sum; 308349f8e832SCong Wang return sum; 308449f8e832SCong Wang } 308549f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 308649f8e832SCong Wang 308714641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 308814641931SCong Wang * which has been changed from the hardware checksum, for example, by 308914641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 309014641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 309114641931SCong Wang * 309214641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 309314641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 309414641931SCong Wang * shared. 309514641931SCong Wang */ 309649f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 309749f8e832SCong Wang { 309849f8e832SCong Wang __wsum csum; 309949f8e832SCong Wang __sum16 sum; 310049f8e832SCong Wang 310149f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 310249f8e832SCong Wang 310349f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 310414641931SCong Wang /* This check is inverted, because we already knew the hardware 310514641931SCong Wang * checksum is invalid before calling this function. So, if the 310614641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 310714641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 310814641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 310914641931SCong Wang * when moving skb->data around. 311014641931SCong Wang */ 311149f8e832SCong Wang if (likely(!sum)) { 311249f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 311349f8e832SCong Wang !skb->csum_complete_sw) 31147fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 311549f8e832SCong Wang } 311649f8e832SCong Wang 311749f8e832SCong Wang if (!skb_shared(skb)) { 311849f8e832SCong Wang /* Save full packet checksum */ 311949f8e832SCong Wang skb->csum = csum; 312049f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 312149f8e832SCong Wang skb->csum_complete_sw = 1; 312249f8e832SCong Wang skb->csum_valid = !sum; 312349f8e832SCong Wang } 312449f8e832SCong Wang 312549f8e832SCong Wang return sum; 312649f8e832SCong Wang } 312749f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 312849f8e832SCong Wang 31299617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 31309617813dSDavide Caratti { 31319617813dSDavide Caratti net_warn_ratelimited( 31329617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 31339617813dSDavide Caratti __func__); 31349617813dSDavide Caratti return 0; 31359617813dSDavide Caratti } 31369617813dSDavide Caratti 31379617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 31389617813dSDavide Caratti int offset, int len) 31399617813dSDavide Caratti { 31409617813dSDavide Caratti net_warn_ratelimited( 31419617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 31429617813dSDavide Caratti __func__); 31439617813dSDavide Caratti return 0; 31449617813dSDavide Caratti } 31459617813dSDavide Caratti 31469617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 31479617813dSDavide Caratti .update = warn_crc32c_csum_update, 31489617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 31499617813dSDavide Caratti }; 31509617813dSDavide Caratti 31519617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 31529617813dSDavide Caratti &default_crc32c_ops; 31539617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 31549617813dSDavide Caratti 3155af2806f8SThomas Graf /** 3156af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 3157af2806f8SThomas Graf * @from: source buffer 3158af2806f8SThomas Graf * 3159af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 3160af2806f8SThomas Graf * into skb_zerocopy(). 3161af2806f8SThomas Graf */ 3162af2806f8SThomas Graf unsigned int 3163af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 3164af2806f8SThomas Graf { 3165af2806f8SThomas Graf unsigned int hlen = 0; 3166af2806f8SThomas Graf 3167af2806f8SThomas Graf if (!from->head_frag || 3168af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 3169a17ad096SPravin B Shelar skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) { 3170af2806f8SThomas Graf hlen = skb_headlen(from); 3171a17ad096SPravin B Shelar if (!hlen) 3172a17ad096SPravin B Shelar hlen = from->len; 3173a17ad096SPravin B Shelar } 3174af2806f8SThomas Graf 3175af2806f8SThomas Graf if (skb_has_frag_list(from)) 3176af2806f8SThomas Graf hlen = from->len; 3177af2806f8SThomas Graf 3178af2806f8SThomas Graf return hlen; 3179af2806f8SThomas Graf } 3180af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 3181af2806f8SThomas Graf 3182af2806f8SThomas Graf /** 3183af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 3184af2806f8SThomas Graf * @to: destination buffer 31857fceb4deSMasanari Iida * @from: source buffer 3186af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 3187af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 3188af2806f8SThomas Graf * 3189af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 3190af2806f8SThomas Graf * to the frags in the source buffer. 3191af2806f8SThomas Graf * 3192af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 3193af2806f8SThomas Graf * headroom in the `to` buffer. 319436d5fe6aSZoltan Kiss * 319536d5fe6aSZoltan Kiss * Return value: 319636d5fe6aSZoltan Kiss * 0: everything is OK 319736d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 319836d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 3199af2806f8SThomas Graf */ 320036d5fe6aSZoltan Kiss int 320136d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 3202af2806f8SThomas Graf { 3203af2806f8SThomas Graf int i, j = 0; 3204af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 320536d5fe6aSZoltan Kiss int ret; 3206af2806f8SThomas Graf struct page *page; 3207af2806f8SThomas Graf unsigned int offset; 3208af2806f8SThomas Graf 3209af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 3210af2806f8SThomas Graf 3211af2806f8SThomas Graf /* dont bother with small payloads */ 321236d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 321336d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 3214af2806f8SThomas Graf 3215af2806f8SThomas Graf if (hlen) { 321636d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 321736d5fe6aSZoltan Kiss if (unlikely(ret)) 321836d5fe6aSZoltan Kiss return ret; 3219af2806f8SThomas Graf len -= hlen; 3220af2806f8SThomas Graf } else { 3221af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 3222af2806f8SThomas Graf if (plen) { 3223af2806f8SThomas Graf page = virt_to_head_page(from->head); 3224af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 3225af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 3226af2806f8SThomas Graf get_page(page); 3227af2806f8SThomas Graf j = 1; 3228af2806f8SThomas Graf len -= plen; 3229af2806f8SThomas Graf } 3230af2806f8SThomas Graf } 3231af2806f8SThomas Graf 3232ede57d58SRichard Gobert skb_len_add(to, len + plen); 3233af2806f8SThomas Graf 323436d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 323536d5fe6aSZoltan Kiss skb_tx_error(from); 323636d5fe6aSZoltan Kiss return -ENOMEM; 323736d5fe6aSZoltan Kiss } 32381f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 323936d5fe6aSZoltan Kiss 3240af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 3241d8e18a51SMatthew Wilcox (Oracle) int size; 3242d8e18a51SMatthew Wilcox (Oracle) 3243af2806f8SThomas Graf if (!len) 3244af2806f8SThomas Graf break; 3245af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 3246d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 3247d8e18a51SMatthew Wilcox (Oracle) len); 3248d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3249d8e18a51SMatthew Wilcox (Oracle) len -= size; 3250af2806f8SThomas Graf skb_frag_ref(to, j); 3251af2806f8SThomas Graf j++; 3252af2806f8SThomas Graf } 3253af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 325436d5fe6aSZoltan Kiss 325536d5fe6aSZoltan Kiss return 0; 3256af2806f8SThomas Graf } 3257af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3258af2806f8SThomas Graf 32591da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 32601da177e4SLinus Torvalds { 3261d3bc23e7SAl Viro __wsum csum; 32621da177e4SLinus Torvalds long csstart; 32631da177e4SLinus Torvalds 326484fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 326555508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 32661da177e4SLinus Torvalds else 32671da177e4SLinus Torvalds csstart = skb_headlen(skb); 32681da177e4SLinus Torvalds 326909a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 32701da177e4SLinus Torvalds 3271d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 32721da177e4SLinus Torvalds 32731da177e4SLinus Torvalds csum = 0; 32741da177e4SLinus Torvalds if (csstart != skb->len) 32751da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 32768d5930dfSAl Viro skb->len - csstart); 32771da177e4SLinus Torvalds 327884fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3279ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 32801da177e4SLinus Torvalds 3281d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 32821da177e4SLinus Torvalds } 32831da177e4SLinus Torvalds } 3284b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 32851da177e4SLinus Torvalds 32861da177e4SLinus Torvalds /** 32871da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 32881da177e4SLinus Torvalds * @list: list to dequeue from 32891da177e4SLinus Torvalds * 32901da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 32911da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 32921da177e4SLinus Torvalds * returned or %NULL if the list is empty. 32931da177e4SLinus Torvalds */ 32941da177e4SLinus Torvalds 32951da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 32961da177e4SLinus Torvalds { 32971da177e4SLinus Torvalds unsigned long flags; 32981da177e4SLinus Torvalds struct sk_buff *result; 32991da177e4SLinus Torvalds 33001da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33011da177e4SLinus Torvalds result = __skb_dequeue(list); 33021da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33031da177e4SLinus Torvalds return result; 33041da177e4SLinus Torvalds } 3305b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 33061da177e4SLinus Torvalds 33071da177e4SLinus Torvalds /** 33081da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 33091da177e4SLinus Torvalds * @list: list to dequeue from 33101da177e4SLinus Torvalds * 33111da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 33121da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 33131da177e4SLinus Torvalds * returned or %NULL if the list is empty. 33141da177e4SLinus Torvalds */ 33151da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 33161da177e4SLinus Torvalds { 33171da177e4SLinus Torvalds unsigned long flags; 33181da177e4SLinus Torvalds struct sk_buff *result; 33191da177e4SLinus Torvalds 33201da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33211da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 33221da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33231da177e4SLinus Torvalds return result; 33241da177e4SLinus Torvalds } 3325b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 33261da177e4SLinus Torvalds 33271da177e4SLinus Torvalds /** 33281da177e4SLinus Torvalds * skb_queue_purge - empty a list 33291da177e4SLinus Torvalds * @list: list to empty 33301da177e4SLinus Torvalds * 33311da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 33321da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 33331da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 33341da177e4SLinus Torvalds */ 33351da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 33361da177e4SLinus Torvalds { 33371da177e4SLinus Torvalds struct sk_buff *skb; 33381da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 33391da177e4SLinus Torvalds kfree_skb(skb); 33401da177e4SLinus Torvalds } 3341b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 33421da177e4SLinus Torvalds 33431da177e4SLinus Torvalds /** 33449f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 33459f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3346385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 33479f5afeaeSYaogong Wang * 33489f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 33499f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 33509f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 33519f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 33529f5afeaeSYaogong Wang */ 3353385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 33549f5afeaeSYaogong Wang { 33557c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3356385114deSPeter Oskolkov unsigned int sum = 0; 33579f5afeaeSYaogong Wang 33587c90584cSEric Dumazet while (p) { 33597c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 33607c90584cSEric Dumazet 33617c90584cSEric Dumazet p = rb_next(p); 33627c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3363385114deSPeter Oskolkov sum += skb->truesize; 33649f5afeaeSYaogong Wang kfree_skb(skb); 33657c90584cSEric Dumazet } 3366385114deSPeter Oskolkov return sum; 33679f5afeaeSYaogong Wang } 33689f5afeaeSYaogong Wang 33699f5afeaeSYaogong Wang /** 33701da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 33711da177e4SLinus Torvalds * @list: list to use 33721da177e4SLinus Torvalds * @newsk: buffer to queue 33731da177e4SLinus Torvalds * 33741da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 33751da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33761da177e4SLinus Torvalds * safely. 33771da177e4SLinus Torvalds * 33781da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 33791da177e4SLinus Torvalds */ 33801da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 33811da177e4SLinus Torvalds { 33821da177e4SLinus Torvalds unsigned long flags; 33831da177e4SLinus Torvalds 33841da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 33851da177e4SLinus Torvalds __skb_queue_head(list, newsk); 33861da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 33871da177e4SLinus Torvalds } 3388b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 33891da177e4SLinus Torvalds 33901da177e4SLinus Torvalds /** 33911da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 33921da177e4SLinus Torvalds * @list: list to use 33931da177e4SLinus Torvalds * @newsk: buffer to queue 33941da177e4SLinus Torvalds * 33951da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 33961da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 33971da177e4SLinus Torvalds * safely. 33981da177e4SLinus Torvalds * 33991da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 34001da177e4SLinus Torvalds */ 34011da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 34021da177e4SLinus Torvalds { 34031da177e4SLinus Torvalds unsigned long flags; 34041da177e4SLinus Torvalds 34051da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 34061da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 34071da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 34081da177e4SLinus Torvalds } 3409b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 34108728b834SDavid S. Miller 34111da177e4SLinus Torvalds /** 34121da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 34131da177e4SLinus Torvalds * @skb: buffer to remove 34148728b834SDavid S. Miller * @list: list to use 34151da177e4SLinus Torvalds * 34168728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 34178728b834SDavid S. Miller * function is atomic with respect to other list locked calls 34181da177e4SLinus Torvalds * 34198728b834SDavid S. Miller * You must know what list the SKB is on. 34201da177e4SLinus Torvalds */ 34218728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 34221da177e4SLinus Torvalds { 34231da177e4SLinus Torvalds unsigned long flags; 34241da177e4SLinus Torvalds 34251da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 34268728b834SDavid S. Miller __skb_unlink(skb, list); 34271da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 34281da177e4SLinus Torvalds } 3429b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 34301da177e4SLinus Torvalds 34311da177e4SLinus Torvalds /** 34321da177e4SLinus Torvalds * skb_append - append a buffer 34331da177e4SLinus Torvalds * @old: buffer to insert after 34341da177e4SLinus Torvalds * @newsk: buffer to insert 34358728b834SDavid S. Miller * @list: list to use 34361da177e4SLinus Torvalds * 34371da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 34381da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 34391da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 34401da177e4SLinus Torvalds */ 34418728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 34421da177e4SLinus Torvalds { 34431da177e4SLinus Torvalds unsigned long flags; 34441da177e4SLinus Torvalds 34458728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 34467de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 34478728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 34481da177e4SLinus Torvalds } 3449b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 34501da177e4SLinus Torvalds 34511da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 34521da177e4SLinus Torvalds struct sk_buff* skb1, 34531da177e4SLinus Torvalds const u32 len, const int pos) 34541da177e4SLinus Torvalds { 34551da177e4SLinus Torvalds int i; 34561da177e4SLinus Torvalds 3457d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3458d626f62bSArnaldo Carvalho de Melo pos - len); 34591da177e4SLinus Torvalds /* And move data appendix as is. */ 34601da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 34611da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 34621da177e4SLinus Torvalds 34631da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 34641da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 34651da177e4SLinus Torvalds skb1->data_len = skb->data_len; 34661da177e4SLinus Torvalds skb1->len += skb1->data_len; 34671da177e4SLinus Torvalds skb->data_len = 0; 34681da177e4SLinus Torvalds skb->len = len; 346927a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 34701da177e4SLinus Torvalds } 34711da177e4SLinus Torvalds 34721da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 34731da177e4SLinus Torvalds struct sk_buff* skb1, 34741da177e4SLinus Torvalds const u32 len, int pos) 34751da177e4SLinus Torvalds { 34761da177e4SLinus Torvalds int i, k = 0; 34771da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 34781da177e4SLinus Torvalds 34791da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 34801da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 34811da177e4SLinus Torvalds skb->len = len; 34821da177e4SLinus Torvalds skb->data_len = len - pos; 34831da177e4SLinus Torvalds 34841da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 34859e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 34861da177e4SLinus Torvalds 34871da177e4SLinus Torvalds if (pos + size > len) { 34881da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 34891da177e4SLinus Torvalds 34901da177e4SLinus Torvalds if (pos < len) { 34911da177e4SLinus Torvalds /* Split frag. 34921da177e4SLinus Torvalds * We have two variants in this case: 34931da177e4SLinus Torvalds * 1. Move all the frag to the second 34941da177e4SLinus Torvalds * part, if it is possible. F.e. 34951da177e4SLinus Torvalds * this approach is mandatory for TUX, 34961da177e4SLinus Torvalds * where splitting is expensive. 34971da177e4SLinus Torvalds * 2. Split is accurately. We make this. 34981da177e4SLinus Torvalds */ 3499ea2ab693SIan Campbell skb_frag_ref(skb, i); 3500b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 35019e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 35029e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 35031da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 35041da177e4SLinus Torvalds } 35051da177e4SLinus Torvalds k++; 35061da177e4SLinus Torvalds } else 35071da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 35081da177e4SLinus Torvalds pos += size; 35091da177e4SLinus Torvalds } 35101da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 35111da177e4SLinus Torvalds } 35121da177e4SLinus Torvalds 35131da177e4SLinus Torvalds /** 35141da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 35151da177e4SLinus Torvalds * @skb: the buffer to split 35161da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 35171da177e4SLinus Torvalds * @len: new length for skb 35181da177e4SLinus Torvalds */ 35191da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 35201da177e4SLinus Torvalds { 35211da177e4SLinus Torvalds int pos = skb_headlen(skb); 35229b65b17dSTalal Ahmad const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY; 35231da177e4SLinus Torvalds 3524753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 3525753f1ca4SPavel Begunkov 35269b65b17dSTalal Ahmad skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags; 35271f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 35281da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 35291da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 35301da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 35311da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 35321da177e4SLinus Torvalds } 3533b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 35341da177e4SLinus Torvalds 35359f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 35369f782db3SIlpo Järvinen * 35379f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 35389f782db3SIlpo Järvinen */ 3539832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3540832d11c5SIlpo Järvinen { 3541c4777efaSEric Dumazet return skb_unclone_keeptruesize(skb, GFP_ATOMIC); 3542832d11c5SIlpo Järvinen } 3543832d11c5SIlpo Järvinen 3544832d11c5SIlpo Järvinen /** 3545832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3546832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3547832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3548832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3549832d11c5SIlpo Järvinen * 3550832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 355120e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3552832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3553832d11c5SIlpo Järvinen * 3554832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3555832d11c5SIlpo Järvinen * 3556832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3557832d11c5SIlpo Järvinen * to have non-paged data as well. 3558832d11c5SIlpo Järvinen * 3559832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3560832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3561832d11c5SIlpo Järvinen */ 3562832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3563832d11c5SIlpo Järvinen { 3564832d11c5SIlpo Järvinen int from, to, merge, todo; 3565d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3566832d11c5SIlpo Järvinen 3567832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3568f8071cdeSEric Dumazet 3569f8071cdeSEric Dumazet if (skb_headlen(skb)) 3570f8071cdeSEric Dumazet return 0; 35711f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 35721f8b977aSWillem de Bruijn return 0; 3573832d11c5SIlpo Järvinen 3574832d11c5SIlpo Järvinen todo = shiftlen; 3575832d11c5SIlpo Järvinen from = 0; 3576832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3577832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3578832d11c5SIlpo Järvinen 3579832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3580832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3581832d11c5SIlpo Järvinen */ 3582832d11c5SIlpo Järvinen if (!to || 3583ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3584b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3585832d11c5SIlpo Järvinen merge = -1; 3586832d11c5SIlpo Järvinen } else { 3587832d11c5SIlpo Järvinen merge = to - 1; 3588832d11c5SIlpo Järvinen 35899e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3590832d11c5SIlpo Järvinen if (todo < 0) { 3591832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3592832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3593832d11c5SIlpo Järvinen return 0; 3594832d11c5SIlpo Järvinen 35959f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 35969f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3597832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3598832d11c5SIlpo Järvinen 35999e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 36009e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3601b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3602832d11c5SIlpo Järvinen 3603832d11c5SIlpo Järvinen goto onlymerged; 3604832d11c5SIlpo Järvinen } 3605832d11c5SIlpo Järvinen 3606832d11c5SIlpo Järvinen from++; 3607832d11c5SIlpo Järvinen } 3608832d11c5SIlpo Järvinen 3609832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3610832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3611832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3612832d11c5SIlpo Järvinen return 0; 3613832d11c5SIlpo Järvinen 3614832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3615832d11c5SIlpo Järvinen return 0; 3616832d11c5SIlpo Järvinen 3617832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3618832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3619832d11c5SIlpo Järvinen return 0; 3620832d11c5SIlpo Järvinen 3621832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3622832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3623832d11c5SIlpo Järvinen 36249e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3625832d11c5SIlpo Järvinen *fragto = *fragfrom; 36269e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3627832d11c5SIlpo Järvinen from++; 3628832d11c5SIlpo Järvinen to++; 3629832d11c5SIlpo Järvinen 3630832d11c5SIlpo Järvinen } else { 3631ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3632b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3633b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 36349e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3635832d11c5SIlpo Järvinen 3636b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 36379e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3638832d11c5SIlpo Järvinen todo = 0; 3639832d11c5SIlpo Järvinen 3640832d11c5SIlpo Järvinen to++; 3641832d11c5SIlpo Järvinen break; 3642832d11c5SIlpo Järvinen } 3643832d11c5SIlpo Järvinen } 3644832d11c5SIlpo Järvinen 3645832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3646832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3647832d11c5SIlpo Järvinen 3648832d11c5SIlpo Järvinen if (merge >= 0) { 3649832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3650832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3651832d11c5SIlpo Järvinen 36529e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 36536a5bcd84SIlias Apalodimas __skb_frag_unref(fragfrom, skb->pp_recycle); 3654832d11c5SIlpo Järvinen } 3655832d11c5SIlpo Järvinen 3656832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3657832d11c5SIlpo Järvinen to = 0; 3658832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3659832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3660832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3661832d11c5SIlpo Järvinen 3662832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3663832d11c5SIlpo Järvinen 3664832d11c5SIlpo Järvinen onlymerged: 3665832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3666832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3667832d11c5SIlpo Järvinen */ 3668832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3669832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3670832d11c5SIlpo Järvinen 3671ede57d58SRichard Gobert skb_len_add(skb, -shiftlen); 3672ede57d58SRichard Gobert skb_len_add(tgt, shiftlen); 3673832d11c5SIlpo Järvinen 3674832d11c5SIlpo Järvinen return shiftlen; 3675832d11c5SIlpo Järvinen } 3676832d11c5SIlpo Järvinen 3677677e90edSThomas Graf /** 3678677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3679677e90edSThomas Graf * @skb: the buffer to read 3680677e90edSThomas Graf * @from: lower offset of data to be read 3681677e90edSThomas Graf * @to: upper offset of data to be read 3682677e90edSThomas Graf * @st: state variable 3683677e90edSThomas Graf * 3684677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3685677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3686677e90edSThomas Graf */ 3687677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3688677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3689677e90edSThomas Graf { 3690677e90edSThomas Graf st->lower_offset = from; 3691677e90edSThomas Graf st->upper_offset = to; 3692677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3693677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3694677e90edSThomas Graf st->frag_data = NULL; 369597550f6fSWillem de Bruijn st->frag_off = 0; 3696677e90edSThomas Graf } 3697b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3698677e90edSThomas Graf 3699677e90edSThomas Graf /** 3700677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3701677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3702677e90edSThomas Graf * @data: destination pointer for data to be returned 3703677e90edSThomas Graf * @st: state variable 3704677e90edSThomas Graf * 3705bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3706677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3707bc32383cSMathias Krause * the head of the data block to @data and returns the length 3708677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3709677e90edSThomas Graf * offset has been reached. 3710677e90edSThomas Graf * 3711677e90edSThomas Graf * The caller is not required to consume all of the data 3712bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3713677e90edSThomas Graf * of bytes already consumed and the next call to 3714677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3715677e90edSThomas Graf * 371625985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3717e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3718677e90edSThomas Graf * reads of potentially non linear data. 3719677e90edSThomas Graf * 3720bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3721677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3722677e90edSThomas Graf * a stack for this purpose. 3723677e90edSThomas Graf */ 3724677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3725677e90edSThomas Graf struct skb_seq_state *st) 3726677e90edSThomas Graf { 3727677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3728677e90edSThomas Graf skb_frag_t *frag; 3729677e90edSThomas Graf 3730aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3731aeb193eaSWedson Almeida Filho if (st->frag_data) { 3732aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3733aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3734aeb193eaSWedson Almeida Filho } 3735677e90edSThomas Graf return 0; 3736aeb193eaSWedson Almeida Filho } 3737677e90edSThomas Graf 3738677e90edSThomas Graf next_skb: 373995e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3740677e90edSThomas Graf 3741995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 374295e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3743677e90edSThomas Graf return block_limit - abs_offset; 3744677e90edSThomas Graf } 3745677e90edSThomas Graf 3746677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3747677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3748677e90edSThomas Graf 3749677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 375097550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3751677e90edSThomas Graf 375297550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 375397550f6fSWillem de Bruijn 375497550f6fSWillem de Bruijn pg_idx = 0; 375597550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 375697550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 375797550f6fSWillem de Bruijn 375897550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 375997550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 376097550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 376197550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 376297550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 376397550f6fSWillem de Bruijn } 376497550f6fSWillem de Bruijn 376597550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3766677e90edSThomas Graf if (abs_offset < block_limit) { 3767677e90edSThomas Graf if (!st->frag_data) 376897550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3769677e90edSThomas Graf 377097550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3771677e90edSThomas Graf (abs_offset - st->stepped_offset); 3772677e90edSThomas Graf 3773677e90edSThomas Graf return block_limit - abs_offset; 3774677e90edSThomas Graf } 3775677e90edSThomas Graf 3776677e90edSThomas Graf if (st->frag_data) { 377751c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3778677e90edSThomas Graf st->frag_data = NULL; 3779677e90edSThomas Graf } 3780677e90edSThomas Graf 378197550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 378297550f6fSWillem de Bruijn st->frag_off += pg_sz; 378397550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 378497550f6fSWillem de Bruijn st->frag_off = 0; 3785677e90edSThomas Graf st->frag_idx++; 378697550f6fSWillem de Bruijn } 3787677e90edSThomas Graf } 3788677e90edSThomas Graf 37895b5a60daSOlaf Kirch if (st->frag_data) { 379051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 37915b5a60daSOlaf Kirch st->frag_data = NULL; 37925b5a60daSOlaf Kirch } 37935b5a60daSOlaf Kirch 379421dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3795677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 379695e3b24cSHerbert Xu st->frag_idx = 0; 3797677e90edSThomas Graf goto next_skb; 379871b3346dSShyam Iyer } else if (st->cur_skb->next) { 379971b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 380071b3346dSShyam Iyer st->frag_idx = 0; 3801677e90edSThomas Graf goto next_skb; 3802677e90edSThomas Graf } 3803677e90edSThomas Graf 3804677e90edSThomas Graf return 0; 3805677e90edSThomas Graf } 3806b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3807677e90edSThomas Graf 3808677e90edSThomas Graf /** 3809677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3810677e90edSThomas Graf * @st: state variable 3811677e90edSThomas Graf * 3812677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3813677e90edSThomas Graf * returned 0. 3814677e90edSThomas Graf */ 3815677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3816677e90edSThomas Graf { 3817677e90edSThomas Graf if (st->frag_data) 381851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3819677e90edSThomas Graf } 3820b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3821677e90edSThomas Graf 38223fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 38233fc7e8a6SThomas Graf 38243fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 38253fc7e8a6SThomas Graf struct ts_config *conf, 38263fc7e8a6SThomas Graf struct ts_state *state) 38273fc7e8a6SThomas Graf { 38283fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 38293fc7e8a6SThomas Graf } 38303fc7e8a6SThomas Graf 38313fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 38323fc7e8a6SThomas Graf { 38333fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 38343fc7e8a6SThomas Graf } 38353fc7e8a6SThomas Graf 38363fc7e8a6SThomas Graf /** 38373fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 38383fc7e8a6SThomas Graf * @skb: the buffer to look in 38393fc7e8a6SThomas Graf * @from: search offset 38403fc7e8a6SThomas Graf * @to: search limit 38413fc7e8a6SThomas Graf * @config: textsearch configuration 38423fc7e8a6SThomas Graf * 38433fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 38443fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 38453fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 38463fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 38473fc7e8a6SThomas Graf */ 38483fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3849059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 38503fc7e8a6SThomas Graf { 3851059a2440SBojan Prtvar struct ts_state state; 3852f72b948dSPhil Oester unsigned int ret; 3853f72b948dSPhil Oester 3854b228c9b0SWillem de Bruijn BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb)); 3855b228c9b0SWillem de Bruijn 38563fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 38573fc7e8a6SThomas Graf config->finish = skb_ts_finish; 38583fc7e8a6SThomas Graf 3859059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 38603fc7e8a6SThomas Graf 3861059a2440SBojan Prtvar ret = textsearch_find(config, &state); 3862f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 38633fc7e8a6SThomas Graf } 3864b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 38653fc7e8a6SThomas Graf 3866be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3867be12a1feSHannes Frederic Sowa int offset, size_t size) 3868be12a1feSHannes Frederic Sowa { 3869be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 3870be12a1feSHannes Frederic Sowa 3871be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 3872be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3873be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 3874753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 3875be12a1feSHannes Frederic Sowa get_page(page); 3876be12a1feSHannes Frederic Sowa skb_fill_page_desc(skb, i, page, offset, size); 3877be12a1feSHannes Frederic Sowa } else { 3878be12a1feSHannes Frederic Sowa return -EMSGSIZE; 3879be12a1feSHannes Frederic Sowa } 3880be12a1feSHannes Frederic Sowa 3881be12a1feSHannes Frederic Sowa return 0; 3882be12a1feSHannes Frederic Sowa } 3883be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3884be12a1feSHannes Frederic Sowa 3885cbb042f9SHerbert Xu /** 3886cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 3887cbb042f9SHerbert Xu * @skb: buffer to update 3888cbb042f9SHerbert Xu * @len: length of data pulled 3889cbb042f9SHerbert Xu * 3890cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 3891fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 389284fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 389384fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 389484fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 3895cbb042f9SHerbert Xu */ 3896af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3897cbb042f9SHerbert Xu { 389831b33dfbSPravin B Shelar unsigned char *data = skb->data; 389931b33dfbSPravin B Shelar 3900cbb042f9SHerbert Xu BUG_ON(len > skb->len); 390131b33dfbSPravin B Shelar __skb_pull(skb, len); 390231b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 390331b33dfbSPravin B Shelar return skb->data; 3904cbb042f9SHerbert Xu } 3905f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3906f94691acSArnaldo Carvalho de Melo 390713acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 390813acc94eSYonghong Song { 390913acc94eSYonghong Song skb_frag_t head_frag; 391013acc94eSYonghong Song struct page *page; 391113acc94eSYonghong Song 391213acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 3913d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 3914b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 3915b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 3916d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 391713acc94eSYonghong Song return head_frag; 391813acc94eSYonghong Song } 391913acc94eSYonghong Song 39203a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 39213a1296a3SSteffen Klassert netdev_features_t features, 39223a1296a3SSteffen Klassert unsigned int offset) 39233a1296a3SSteffen Klassert { 39243a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 39253a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 39263a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 39273a1296a3SSteffen Klassert unsigned int delta_len = 0; 39283a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 392953475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 3930cf3ab8d4SLina Wang int len_diff, err; 39313a1296a3SSteffen Klassert 39323a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 39333a1296a3SSteffen Klassert 39343a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 39353a1296a3SSteffen Klassert 39363a1296a3SSteffen Klassert do { 39373a1296a3SSteffen Klassert nskb = list_skb; 39383a1296a3SSteffen Klassert list_skb = list_skb->next; 39393a1296a3SSteffen Klassert 394053475c5dSDongseok Yi err = 0; 3941224102deSlena wang delta_truesize += nskb->truesize; 394253475c5dSDongseok Yi if (skb_shared(nskb)) { 394353475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 394453475c5dSDongseok Yi if (tmp) { 394553475c5dSDongseok Yi consume_skb(nskb); 394653475c5dSDongseok Yi nskb = tmp; 394753475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 394853475c5dSDongseok Yi } else { 394953475c5dSDongseok Yi err = -ENOMEM; 395053475c5dSDongseok Yi } 395153475c5dSDongseok Yi } 395253475c5dSDongseok Yi 39533a1296a3SSteffen Klassert if (!tail) 39543a1296a3SSteffen Klassert skb->next = nskb; 39553a1296a3SSteffen Klassert else 39563a1296a3SSteffen Klassert tail->next = nskb; 39573a1296a3SSteffen Klassert 395853475c5dSDongseok Yi if (unlikely(err)) { 395953475c5dSDongseok Yi nskb->next = list_skb; 396053475c5dSDongseok Yi goto err_linearize; 396153475c5dSDongseok Yi } 396253475c5dSDongseok Yi 39633a1296a3SSteffen Klassert tail = nskb; 39643a1296a3SSteffen Klassert 39653a1296a3SSteffen Klassert delta_len += nskb->len; 39663a1296a3SSteffen Klassert 39673a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 39683a1296a3SSteffen Klassert 3969cf673ed0SFlorian Westphal skb_release_head_state(nskb); 3970cf3ab8d4SLina Wang len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb); 39713a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 39723a1296a3SSteffen Klassert 39733a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 3974cf3ab8d4SLina Wang nskb->transport_header += len_diff; 39753a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 39763a1296a3SSteffen Klassert nskb->data - tnl_hlen, 39773a1296a3SSteffen Klassert offset + tnl_hlen); 39783a1296a3SSteffen Klassert 39793a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 39803a1296a3SSteffen Klassert __skb_linearize(nskb)) 39813a1296a3SSteffen Klassert goto err_linearize; 39823a1296a3SSteffen Klassert 39833a1296a3SSteffen Klassert } while (list_skb); 39843a1296a3SSteffen Klassert 39853a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 39863a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 39873a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 39883a1296a3SSteffen Klassert 39893a1296a3SSteffen Klassert skb_gso_reset(skb); 39903a1296a3SSteffen Klassert 39913a1296a3SSteffen Klassert skb->prev = tail; 39923a1296a3SSteffen Klassert 39933a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 39943a1296a3SSteffen Klassert __skb_linearize(skb)) 39953a1296a3SSteffen Klassert goto err_linearize; 39963a1296a3SSteffen Klassert 39973a1296a3SSteffen Klassert skb_get(skb); 39983a1296a3SSteffen Klassert 39993a1296a3SSteffen Klassert return skb; 40003a1296a3SSteffen Klassert 40013a1296a3SSteffen Klassert err_linearize: 40023a1296a3SSteffen Klassert kfree_skb_list(skb->next); 40033a1296a3SSteffen Klassert skb->next = NULL; 40043a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 40053a1296a3SSteffen Klassert } 40063a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 40073a1296a3SSteffen Klassert 4008f4c50d99SHerbert Xu /** 4009f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 4010df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 4011576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 4012f4c50d99SHerbert Xu * 4013f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 40144c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 40154c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 4016f4c50d99SHerbert Xu */ 4017df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 4018df5771ffSMichael S. Tsirkin netdev_features_t features) 4019f4c50d99SHerbert Xu { 4020f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 4021f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 40221a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 4023df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 4024df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 4025df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 40261fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 4027f4c50d99SHerbert Xu unsigned int offset = doffset; 4028df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 4029802ab55aSAlexander Duyck unsigned int partial_segs = 0; 4030f4c50d99SHerbert Xu unsigned int headroom; 4031802ab55aSAlexander Duyck unsigned int len = head_skb->len; 4032ec5f0615SPravin B Shelar __be16 proto; 403336c98382SAlexander Duyck bool csum, sg; 4034df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 4035f4c50d99SHerbert Xu int err = -ENOMEM; 4036f4c50d99SHerbert Xu int i = 0; 4037f4c50d99SHerbert Xu int pos; 4038f4c50d99SHerbert Xu 40393dcbdb13SShmulik Ladkani if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) && 40403dcbdb13SShmulik Ladkani (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) { 40413dcbdb13SShmulik Ladkani /* gso_size is untrusted, and we have a frag_list with a linear 40423dcbdb13SShmulik Ladkani * non head_frag head. 40433dcbdb13SShmulik Ladkani * 40443dcbdb13SShmulik Ladkani * (we assume checking the first list_skb member suffices; 40453dcbdb13SShmulik Ladkani * i.e if either of the list_skb members have non head_frag 40463dcbdb13SShmulik Ladkani * head, then the first one has too). 40473dcbdb13SShmulik Ladkani * 40483dcbdb13SShmulik Ladkani * If head_skb's headlen does not fit requested gso_size, it 40493dcbdb13SShmulik Ladkani * means that the frag_list members do NOT terminate on exact 40503dcbdb13SShmulik Ladkani * gso_size boundaries. Hence we cannot perform skb_frag_t page 40513dcbdb13SShmulik Ladkani * sharing. Therefore we must fallback to copying the frag_list 40523dcbdb13SShmulik Ladkani * skbs; we do so by disabling SG. 40533dcbdb13SShmulik Ladkani */ 40543dcbdb13SShmulik Ladkani if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) 40553dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 40563dcbdb13SShmulik Ladkani } 40573dcbdb13SShmulik Ladkani 40585882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 40592f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 4060ec5f0615SPravin B Shelar if (unlikely(!proto)) 4061ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 4062ec5f0615SPravin B Shelar 406336c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 4064f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 40657e2b10c1STom Herbert 406607b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 406707b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 406807b26c94SSteffen Klassert struct sk_buff *iter; 406943170c4eSIlan Tayari unsigned int frag_len; 407007b26c94SSteffen Klassert 407107b26c94SSteffen Klassert if (!list_skb || 407207b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 407307b26c94SSteffen Klassert goto normal; 407407b26c94SSteffen Klassert 407543170c4eSIlan Tayari /* If we get here then all the required 407643170c4eSIlan Tayari * GSO features except frag_list are supported. 407743170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 407843170c4eSIlan Tayari * with no frag_list. 407943170c4eSIlan Tayari * Currently we can do that only when the buffers don't 408043170c4eSIlan Tayari * have a linear part and all the buffers except 408143170c4eSIlan Tayari * the last are of the same length. 408207b26c94SSteffen Klassert */ 408343170c4eSIlan Tayari frag_len = list_skb->len; 408407b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 408543170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 408643170c4eSIlan Tayari goto normal; 4087eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 408807b26c94SSteffen Klassert goto normal; 408907b26c94SSteffen Klassert 409007b26c94SSteffen Klassert len -= iter->len; 409107b26c94SSteffen Klassert } 409243170c4eSIlan Tayari 409343170c4eSIlan Tayari if (len != frag_len) 409443170c4eSIlan Tayari goto normal; 409507b26c94SSteffen Klassert } 409607b26c94SSteffen Klassert 4097802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 4098802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 4099802ab55aSAlexander Duyck * now. 4100802ab55aSAlexander Duyck */ 4101802ab55aSAlexander Duyck partial_segs = len / mss; 4102d7fb5a80SAlexander Duyck if (partial_segs > 1) 4103802ab55aSAlexander Duyck mss *= partial_segs; 4104d7fb5a80SAlexander Duyck else 4105d7fb5a80SAlexander Duyck partial_segs = 0; 4106802ab55aSAlexander Duyck } 4107802ab55aSAlexander Duyck 410807b26c94SSteffen Klassert normal: 4109df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 4110df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 4111f4c50d99SHerbert Xu 4112f4c50d99SHerbert Xu do { 4113f4c50d99SHerbert Xu struct sk_buff *nskb; 41148cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 4115c8884eddSHerbert Xu int hsize; 4116f4c50d99SHerbert Xu int size; 4117f4c50d99SHerbert Xu 41183953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 41193953c46cSMarcelo Ricardo Leitner len = list_skb->len; 41203953c46cSMarcelo Ricardo Leitner } else { 4121df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 4122f4c50d99SHerbert Xu if (len > mss) 4123f4c50d99SHerbert Xu len = mss; 41243953c46cSMarcelo Ricardo Leitner } 4125f4c50d99SHerbert Xu 4126df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 4127f4c50d99SHerbert Xu 4128dbd50f23SXin Long if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) && 41291a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 41301a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 413189319d38SHerbert Xu 41329d8506ccSHerbert Xu i = 0; 41331a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 41341a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 41351fd819ecSMichael S. Tsirkin frag_skb = list_skb; 41361a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 41379d8506ccSHerbert Xu 41389d8506ccSHerbert Xu while (pos < offset + len) { 41399d8506ccSHerbert Xu BUG_ON(i >= nfrags); 41409d8506ccSHerbert Xu 41414e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 41429d8506ccSHerbert Xu if (pos + size > offset + len) 41439d8506ccSHerbert Xu break; 41449d8506ccSHerbert Xu 41459d8506ccSHerbert Xu i++; 41469d8506ccSHerbert Xu pos += size; 41474e1beba1SMichael S. Tsirkin frag++; 41489d8506ccSHerbert Xu } 41499d8506ccSHerbert Xu 41501a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 41511a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 415289319d38SHerbert Xu 4153f4c50d99SHerbert Xu if (unlikely(!nskb)) 4154f4c50d99SHerbert Xu goto err; 4155f4c50d99SHerbert Xu 41569d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 41579d8506ccSHerbert Xu kfree_skb(nskb); 41589d8506ccSHerbert Xu goto err; 41599d8506ccSHerbert Xu } 41609d8506ccSHerbert Xu 4161ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 416289319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 416389319d38SHerbert Xu kfree_skb(nskb); 416489319d38SHerbert Xu goto err; 416589319d38SHerbert Xu } 416689319d38SHerbert Xu 4167ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 416889319d38SHerbert Xu skb_release_head_state(nskb); 416989319d38SHerbert Xu __skb_push(nskb, doffset); 417089319d38SHerbert Xu } else { 417100b229f7SPaolo Abeni if (hsize < 0) 417200b229f7SPaolo Abeni hsize = 0; 4173dbd50f23SXin Long if (hsize > len || !sg) 4174dbd50f23SXin Long hsize = len; 4175dbd50f23SXin Long 4176c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 4177df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 4178c93bdd0eSMel Gorman NUMA_NO_NODE); 417989319d38SHerbert Xu 418089319d38SHerbert Xu if (unlikely(!nskb)) 418189319d38SHerbert Xu goto err; 418289319d38SHerbert Xu 418389319d38SHerbert Xu skb_reserve(nskb, headroom); 418489319d38SHerbert Xu __skb_put(nskb, doffset); 418589319d38SHerbert Xu } 418689319d38SHerbert Xu 4187f4c50d99SHerbert Xu if (segs) 4188f4c50d99SHerbert Xu tail->next = nskb; 4189f4c50d99SHerbert Xu else 4190f4c50d99SHerbert Xu segs = nskb; 4191f4c50d99SHerbert Xu tail = nskb; 4192f4c50d99SHerbert Xu 4193df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 4194f4c50d99SHerbert Xu 4195030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 4196fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 419768c33163SPravin B Shelar 4198df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 419968c33163SPravin B Shelar nskb->data - tnl_hlen, 420068c33163SPravin B Shelar doffset + tnl_hlen); 420189319d38SHerbert Xu 42029d8506ccSHerbert Xu if (nskb->len == len + doffset) 42031cdbcb79SSimon Horman goto perform_csum_check; 420489319d38SHerbert Xu 42057fbeffedSAlexander Duyck if (!sg) { 42061454c9faSYadu Kishore if (!csum) { 42077fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 42086f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 420976443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 421076443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 42111454c9faSYadu Kishore skb_put(nskb, 42121454c9faSYadu Kishore len), 42138d5930dfSAl Viro len); 42147e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 4215de843723STom Herbert skb_headroom(nskb) + doffset; 42161454c9faSYadu Kishore } else { 4217c624c58eSlily if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len)) 4218c624c58eSlily goto err; 42191454c9faSYadu Kishore } 4220f4c50d99SHerbert Xu continue; 4221f4c50d99SHerbert Xu } 4222f4c50d99SHerbert Xu 42238cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 4224f4c50d99SHerbert Xu 4225df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 4226d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4227f4c50d99SHerbert Xu 422806b4feb3SJonathan Lemon skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags & 422906b4feb3SJonathan Lemon SKBFL_SHARED_FRAG; 4230cef401deSEric Dumazet 4231bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4232bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4233bf5c25d6SWillem de Bruijn goto err; 4234bf5c25d6SWillem de Bruijn 42359d8506ccSHerbert Xu while (pos < offset + len) { 42369d8506ccSHerbert Xu if (i >= nfrags) { 42379d8506ccSHerbert Xu i = 0; 42381a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 42391a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 42401fd819ecSMichael S. Tsirkin frag_skb = list_skb; 424113acc94eSYonghong Song if (!skb_headlen(list_skb)) { 42429d8506ccSHerbert Xu BUG_ON(!nfrags); 424313acc94eSYonghong Song } else { 424413acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 42459d8506ccSHerbert Xu 424613acc94eSYonghong Song /* to make room for head_frag. */ 424713acc94eSYonghong Song i--; 424813acc94eSYonghong Song frag--; 424913acc94eSYonghong Song } 4250bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4251bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4252bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4253bf5c25d6SWillem de Bruijn goto err; 4254bf5c25d6SWillem de Bruijn 42551a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 42569d8506ccSHerbert Xu } 42579d8506ccSHerbert Xu 42589d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 42599d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 42609d8506ccSHerbert Xu net_warn_ratelimited( 42619d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 42629d8506ccSHerbert Xu pos, mss); 4263ff907a11SEric Dumazet err = -EINVAL; 42649d8506ccSHerbert Xu goto err; 42659d8506ccSHerbert Xu } 42669d8506ccSHerbert Xu 426713acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 42688cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 42698cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4270f4c50d99SHerbert Xu 4271f4c50d99SHerbert Xu if (pos < offset) { 4272b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 42738cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4274f4c50d99SHerbert Xu } 4275f4c50d99SHerbert Xu 427689319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4277f4c50d99SHerbert Xu 4278f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4279f4c50d99SHerbert Xu i++; 42804e1beba1SMichael S. Tsirkin frag++; 4281f4c50d99SHerbert Xu pos += size; 4282f4c50d99SHerbert Xu } else { 42838cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 428489319d38SHerbert Xu goto skip_fraglist; 4285f4c50d99SHerbert Xu } 4286f4c50d99SHerbert Xu 42878cb19905SMichael S. Tsirkin nskb_frag++; 4288f4c50d99SHerbert Xu } 4289f4c50d99SHerbert Xu 429089319d38SHerbert Xu skip_fraglist: 4291f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4292f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4293f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4294ec5f0615SPravin B Shelar 42951cdbcb79SSimon Horman perform_csum_check: 42967fbeffedSAlexander Duyck if (!csum) { 4297ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4298ff907a11SEric Dumazet __skb_linearize(nskb)) 4299ddff00d4SAlexander Duyck goto err; 4300ff907a11SEric Dumazet 43017fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4302ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 430376443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 430476443456SAlexander Duyck skb_checksum(nskb, doffset, 430576443456SAlexander Duyck nskb->len - doffset, 0); 43067e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 43077e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4308ec5f0615SPravin B Shelar } 4309df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4310f4c50d99SHerbert Xu 4311bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4312bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4313bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4314bec3cfdcSEric Dumazet */ 4315bec3cfdcSEric Dumazet segs->prev = tail; 4316432c856fSToshiaki Makita 4317802ab55aSAlexander Duyck if (partial_segs) { 431807b26c94SSteffen Klassert struct sk_buff *iter; 4319802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 432007b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4321802ab55aSAlexander Duyck 4322802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 432307b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4324802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4325802ab55aSAlexander Duyck 4326802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4327802ab55aSAlexander Duyck * our way back down the stack of protocols. 4328802ab55aSAlexander Duyck */ 432907b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 433007b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 433107b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 433207b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 433307b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 433407b26c94SSteffen Klassert } 433507b26c94SSteffen Klassert 433607b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 433707b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 433807b26c94SSteffen Klassert else if (tail != segs) 433907b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4340802ab55aSAlexander Duyck } 4341802ab55aSAlexander Duyck 4342432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4343432c856fSToshiaki Makita * using skb_set_owner_w(). 4344432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4345432c856fSToshiaki Makita */ 4346432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4347432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4348432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4349432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4350432c856fSToshiaki Makita } 4351f4c50d99SHerbert Xu return segs; 4352f4c50d99SHerbert Xu 4353f4c50d99SHerbert Xu err: 4354289dccbeSEric Dumazet kfree_skb_list(segs); 4355f4c50d99SHerbert Xu return ERR_PTR(err); 4356f4c50d99SHerbert Xu } 4357f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4358f4c50d99SHerbert Xu 4359df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4360df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4361df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4362df5042f4SFlorian Westphal 4363df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4364df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4365df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4366df5042f4SFlorian Westphal #endif 43674165079bSFlorian Westphal #ifdef CONFIG_XFRM 43684165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 43694165079bSFlorian Westphal #endif 437095a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 437195a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 437295a7233cSPaul Blakey #endif 43733ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 43743ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 43753ee17bc7SMat Martineau #endif 437678476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 437778476d31SJeremy Kerr [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow), 437878476d31SJeremy Kerr #endif 4379df5042f4SFlorian Westphal }; 4380df5042f4SFlorian Westphal 4381df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4382df5042f4SFlorian Westphal { 4383df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4384df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4385df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4386df5042f4SFlorian Westphal #endif 43874165079bSFlorian Westphal #ifdef CONFIG_XFRM 43884165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 43894165079bSFlorian Westphal #endif 439095a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 439195a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 439295a7233cSPaul Blakey #endif 43933ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 43943ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 43953ee17bc7SMat Martineau #endif 439678476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 439778476d31SJeremy Kerr skb_ext_type_len[SKB_EXT_MCTP] + 439878476d31SJeremy Kerr #endif 4399df5042f4SFlorian Westphal 0; 4400df5042f4SFlorian Westphal } 4401df5042f4SFlorian Westphal 4402df5042f4SFlorian Westphal static void skb_extensions_init(void) 4403df5042f4SFlorian Westphal { 4404df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4405df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4406df5042f4SFlorian Westphal 4407df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4408df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4409df5042f4SFlorian Westphal 0, 4410df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4411df5042f4SFlorian Westphal NULL); 4412df5042f4SFlorian Westphal } 4413df5042f4SFlorian Westphal #else 4414df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4415df5042f4SFlorian Westphal #endif 4416df5042f4SFlorian Westphal 44171da177e4SLinus Torvalds void __init skb_init(void) 44181da177e4SLinus Torvalds { 441979a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 44201da177e4SLinus Torvalds sizeof(struct sk_buff), 44211da177e4SLinus Torvalds 0, 4422e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 442379a8a642SKees Cook offsetof(struct sk_buff, cb), 442479a8a642SKees Cook sizeof_field(struct sk_buff, cb), 442520c2df83SPaul Mundt NULL); 4426d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4427d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4428d179cd12SDavid S. Miller 0, 4429e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 443020c2df83SPaul Mundt NULL); 4431df5042f4SFlorian Westphal skb_extensions_init(); 44321da177e4SLinus Torvalds } 44331da177e4SLinus Torvalds 443451c739d1SDavid S. Miller static int 443548a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 443648a1df65SJason A. Donenfeld unsigned int recursion_level) 4437716ea3a7SDavid Howells { 44381a028e50SDavid S. Miller int start = skb_headlen(skb); 44391a028e50SDavid S. Miller int i, copy = start - offset; 4440fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4441716ea3a7SDavid Howells int elt = 0; 4442716ea3a7SDavid Howells 444348a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 444448a1df65SJason A. Donenfeld return -EMSGSIZE; 444548a1df65SJason A. Donenfeld 4446716ea3a7SDavid Howells if (copy > 0) { 4447716ea3a7SDavid Howells if (copy > len) 4448716ea3a7SDavid Howells copy = len; 4449642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4450716ea3a7SDavid Howells elt++; 4451716ea3a7SDavid Howells if ((len -= copy) == 0) 4452716ea3a7SDavid Howells return elt; 4453716ea3a7SDavid Howells offset += copy; 4454716ea3a7SDavid Howells } 4455716ea3a7SDavid Howells 4456716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 44571a028e50SDavid S. Miller int end; 4458716ea3a7SDavid Howells 4459547b792cSIlpo Järvinen WARN_ON(start > offset + len); 44601a028e50SDavid S. Miller 44619e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4462716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4463716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 446448a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 446548a1df65SJason A. Donenfeld return -EMSGSIZE; 4466716ea3a7SDavid Howells 4467716ea3a7SDavid Howells if (copy > len) 4468716ea3a7SDavid Howells copy = len; 4469ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4470b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4471716ea3a7SDavid Howells elt++; 4472716ea3a7SDavid Howells if (!(len -= copy)) 4473716ea3a7SDavid Howells return elt; 4474716ea3a7SDavid Howells offset += copy; 4475716ea3a7SDavid Howells } 44761a028e50SDavid S. Miller start = end; 4477716ea3a7SDavid Howells } 4478716ea3a7SDavid Howells 4479fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 448048a1df65SJason A. Donenfeld int end, ret; 4481716ea3a7SDavid Howells 4482547b792cSIlpo Järvinen WARN_ON(start > offset + len); 44831a028e50SDavid S. Miller 4484fbb398a8SDavid S. Miller end = start + frag_iter->len; 4485716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 448648a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 448748a1df65SJason A. Donenfeld return -EMSGSIZE; 448848a1df65SJason A. Donenfeld 4489716ea3a7SDavid Howells if (copy > len) 4490716ea3a7SDavid Howells copy = len; 449148a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 449248a1df65SJason A. Donenfeld copy, recursion_level + 1); 449348a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 449448a1df65SJason A. Donenfeld return ret; 449548a1df65SJason A. Donenfeld elt += ret; 4496716ea3a7SDavid Howells if ((len -= copy) == 0) 4497716ea3a7SDavid Howells return elt; 4498716ea3a7SDavid Howells offset += copy; 4499716ea3a7SDavid Howells } 45001a028e50SDavid S. Miller start = end; 4501716ea3a7SDavid Howells } 4502716ea3a7SDavid Howells BUG_ON(len); 4503716ea3a7SDavid Howells return elt; 4504716ea3a7SDavid Howells } 4505716ea3a7SDavid Howells 450648a1df65SJason A. Donenfeld /** 450748a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 450848a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 450948a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 451048a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 451148a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 451248a1df65SJason A. Donenfeld * 451348a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 451448a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 451548a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 451648a1df65SJason A. Donenfeld * could not fit. 451748a1df65SJason A. Donenfeld */ 451848a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 451948a1df65SJason A. Donenfeld { 452048a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 452148a1df65SJason A. Donenfeld 452248a1df65SJason A. Donenfeld if (nsg <= 0) 452348a1df65SJason A. Donenfeld return nsg; 452448a1df65SJason A. Donenfeld 452548a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 452648a1df65SJason A. Donenfeld 452748a1df65SJason A. Donenfeld return nsg; 452848a1df65SJason A. Donenfeld } 452948a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 453048a1df65SJason A. Donenfeld 453125a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 453225a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 453325a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 453425a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 453525a91d8dSFan Du * 453625a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 453725a91d8dSFan Du * 1. sg_init_table 453825a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 453925a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 454025a91d8dSFan Du * 454125a91d8dSFan Du * This is equivalent to: 454225a91d8dSFan Du * 1. sg_init_table 454325a91d8dSFan Du * 2. skb_to_sgvec(payload1) 454425a91d8dSFan Du * 3. sg_unmark_end 454525a91d8dSFan Du * 4. skb_to_sgvec(payload2) 454625a91d8dSFan Du * 454725a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 454825a91d8dSFan Du * is more preferable. 454925a91d8dSFan Du */ 455025a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 455125a91d8dSFan Du int offset, int len) 455225a91d8dSFan Du { 455348a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 455425a91d8dSFan Du } 455525a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 455625a91d8dSFan Du 455751c739d1SDavid S. Miller 455851c739d1SDavid S. Miller 4559716ea3a7SDavid Howells /** 4560716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4561716ea3a7SDavid Howells * @skb: The socket buffer to check. 4562716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4563716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4564716ea3a7SDavid Howells * 4565716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4566716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4567716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4568716ea3a7SDavid Howells * 4569716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4570716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4571716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4572716ea3a7SDavid Howells * 4573716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4574716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4575716ea3a7SDavid Howells */ 4576716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4577716ea3a7SDavid Howells { 4578716ea3a7SDavid Howells int copyflag; 4579716ea3a7SDavid Howells int elt; 4580716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4581716ea3a7SDavid Howells 4582716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4583716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4584716ea3a7SDavid Howells * at the moment even if they are anonymous). 4585716ea3a7SDavid Howells */ 4586716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4587c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4588716ea3a7SDavid Howells return -ENOMEM; 4589716ea3a7SDavid Howells 4590716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 459121dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4592716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4593716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4594716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4595716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4596716ea3a7SDavid Howells 4597716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4598716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4599716ea3a7SDavid Howells return -ENOMEM; 4600716ea3a7SDavid Howells 4601716ea3a7SDavid Howells /* Voila! */ 4602716ea3a7SDavid Howells *trailer = skb; 4603716ea3a7SDavid Howells return 1; 4604716ea3a7SDavid Howells } 4605716ea3a7SDavid Howells 4606716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4607716ea3a7SDavid Howells 4608716ea3a7SDavid Howells elt = 1; 4609716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4610716ea3a7SDavid Howells copyflag = 0; 4611716ea3a7SDavid Howells 4612716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4613716ea3a7SDavid Howells int ntail = 0; 4614716ea3a7SDavid Howells 4615716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4616716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4617716ea3a7SDavid Howells * after it. */ 4618716ea3a7SDavid Howells 4619716ea3a7SDavid Howells if (skb_shared(skb1)) 4620716ea3a7SDavid Howells copyflag = 1; 4621716ea3a7SDavid Howells 4622716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4623716ea3a7SDavid Howells 4624716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4625716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 462621dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4627716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4628716ea3a7SDavid Howells ntail = tailbits + 128; 4629716ea3a7SDavid Howells } 4630716ea3a7SDavid Howells 4631716ea3a7SDavid Howells if (copyflag || 4632716ea3a7SDavid Howells skb_cloned(skb1) || 4633716ea3a7SDavid Howells ntail || 4634716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 463521dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4636716ea3a7SDavid Howells struct sk_buff *skb2; 4637716ea3a7SDavid Howells 4638716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4639716ea3a7SDavid Howells if (ntail == 0) 4640716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4641716ea3a7SDavid Howells else 4642716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4643716ea3a7SDavid Howells skb_headroom(skb1), 4644716ea3a7SDavid Howells ntail, 4645716ea3a7SDavid Howells GFP_ATOMIC); 4646716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4647716ea3a7SDavid Howells return -ENOMEM; 4648716ea3a7SDavid Howells 4649716ea3a7SDavid Howells if (skb1->sk) 4650716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4651716ea3a7SDavid Howells 4652716ea3a7SDavid Howells /* Looking around. Are we still alive? 4653716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4654716ea3a7SDavid Howells 4655716ea3a7SDavid Howells skb2->next = skb1->next; 4656716ea3a7SDavid Howells *skb_p = skb2; 4657716ea3a7SDavid Howells kfree_skb(skb1); 4658716ea3a7SDavid Howells skb1 = skb2; 4659716ea3a7SDavid Howells } 4660716ea3a7SDavid Howells elt++; 4661716ea3a7SDavid Howells *trailer = skb1; 4662716ea3a7SDavid Howells skb_p = &skb1->next; 4663716ea3a7SDavid Howells } 4664716ea3a7SDavid Howells 4665716ea3a7SDavid Howells return elt; 4666716ea3a7SDavid Howells } 4667b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4668716ea3a7SDavid Howells 4669b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4670b1faf566SEric Dumazet { 4671b1faf566SEric Dumazet struct sock *sk = skb->sk; 4672b1faf566SEric Dumazet 4673b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4674b1faf566SEric Dumazet } 4675b1faf566SEric Dumazet 46768605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 46778605330aSSoheil Hassas Yeganeh { 46788605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 46798605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 46808605330aSSoheil Hassas Yeganeh */ 46818605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 46828605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 46838605330aSSoheil Hassas Yeganeh } 46848605330aSSoheil Hassas Yeganeh 4685b1faf566SEric Dumazet /* 4686b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4687b1faf566SEric Dumazet */ 4688b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4689b1faf566SEric Dumazet { 4690b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4691ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4692b1faf566SEric Dumazet return -ENOMEM; 4693b1faf566SEric Dumazet 4694b1faf566SEric Dumazet skb_orphan(skb); 4695b1faf566SEric Dumazet skb->sk = sk; 4696b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4697b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 46988605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4699b1faf566SEric Dumazet 4700abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4701abb57ea4SEric Dumazet skb_dst_force(skb); 4702abb57ea4SEric Dumazet 4703b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4704b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 4705e3ae2365SAlexander Aring sk_error_report(sk); 4706b1faf566SEric Dumazet return 0; 4707b1faf566SEric Dumazet } 4708b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4709b1faf566SEric Dumazet 471083a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 471183a1a1a7SSoheil Hassas Yeganeh { 471283a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 471383a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 471483a1a1a7SSoheil Hassas Yeganeh } 471583a1a1a7SSoheil Hassas Yeganeh 4716364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4717364a9e93SWillem de Bruijn { 4718364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 471983a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 472083a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4721997d5c3fSEric Dumazet unsigned long flags; 4722364a9e93SWillem de Bruijn 4723997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4724364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 472538b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 472683a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 472738b25793SSoheil Hassas Yeganeh if (icmp_next) 4728985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 472938b25793SSoheil Hassas Yeganeh } 4730997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4731364a9e93SWillem de Bruijn 473283a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 473383a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 473483a1a1a7SSoheil Hassas Yeganeh 473583a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4736e3ae2365SAlexander Aring sk_error_report(sk); 4737364a9e93SWillem de Bruijn 4738364a9e93SWillem de Bruijn return skb; 4739364a9e93SWillem de Bruijn } 4740364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4741364a9e93SWillem de Bruijn 4742cab41c47SAlexander Duyck /** 4743cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4744cab41c47SAlexander Duyck * @skb: the skb to clone 4745cab41c47SAlexander Duyck * 4746cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4747cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4748cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4749cab41c47SAlexander Duyck * 4750cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4751cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4752cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4753cab41c47SAlexander Duyck * the sk_error_queue. 4754cab41c47SAlexander Duyck */ 475562bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 475662bccb8cSAlexander Duyck { 475762bccb8cSAlexander Duyck struct sock *sk = skb->sk; 475862bccb8cSAlexander Duyck struct sk_buff *clone; 475962bccb8cSAlexander Duyck 476041c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 476162bccb8cSAlexander Duyck return NULL; 476262bccb8cSAlexander Duyck 476362bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 476462bccb8cSAlexander Duyck if (!clone) { 476562bccb8cSAlexander Duyck sock_put(sk); 476662bccb8cSAlexander Duyck return NULL; 476762bccb8cSAlexander Duyck } 476862bccb8cSAlexander Duyck 476962bccb8cSAlexander Duyck clone->sk = sk; 477062bccb8cSAlexander Duyck clone->destructor = sock_efree; 477162bccb8cSAlexander Duyck 477262bccb8cSAlexander Duyck return clone; 477362bccb8cSAlexander Duyck } 477462bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 477562bccb8cSAlexander Duyck 477637846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 477737846ef0SAlexander Duyck struct sock *sk, 47784ef1b286SSoheil Hassas Yeganeh int tstype, 47794ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4780ac45f602SPatrick Ohly { 4781ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4782ac45f602SPatrick Ohly int err; 4783ac45f602SPatrick Ohly 47844ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 47854ef1b286SSoheil Hassas Yeganeh 4786ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4787ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4788ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4789ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4790e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 47914ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 47921862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 47934ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 479409c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 479542f67eeaSEric Dumazet if (sk_is_tcp(sk)) 4796a1cdec57SEric Dumazet serr->ee.ee_data -= atomic_read(&sk->sk_tskey); 47974ed2d765SWillem de Bruijn } 479829030374SEric Dumazet 4799ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 480029030374SEric Dumazet 4801ac45f602SPatrick Ohly if (err) 4802ac45f602SPatrick Ohly kfree_skb(skb); 4803ac45f602SPatrick Ohly } 480437846ef0SAlexander Duyck 4805b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4806b245be1fSWillem de Bruijn { 4807b245be1fSWillem de Bruijn bool ret; 4808b245be1fSWillem de Bruijn 4809d2154b0aSKuniyuki Iwashima if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly)) 4810b245be1fSWillem de Bruijn return true; 4811b245be1fSWillem de Bruijn 4812b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4813b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4814b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4815b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4816b245be1fSWillem de Bruijn return ret; 4817b245be1fSWillem de Bruijn } 4818b245be1fSWillem de Bruijn 481937846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 482037846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 482137846ef0SAlexander Duyck { 482237846ef0SAlexander Duyck struct sock *sk = skb->sk; 482337846ef0SAlexander Duyck 4824b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 482535b99dffSWillem de Bruijn goto err; 4826b245be1fSWillem de Bruijn 48279ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 48289ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 48299ac25fc0SEric Dumazet */ 483041c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 483137846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 48324ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 483337846ef0SAlexander Duyck sock_put(sk); 483435b99dffSWillem de Bruijn return; 483537846ef0SAlexander Duyck } 483635b99dffSWillem de Bruijn 483735b99dffSWillem de Bruijn err: 483835b99dffSWillem de Bruijn kfree_skb(skb); 48399ac25fc0SEric Dumazet } 484037846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 484137846ef0SAlexander Duyck 484237846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 4843e7ed11eeSYousuk Seung const struct sk_buff *ack_skb, 484437846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 484537846ef0SAlexander Duyck struct sock *sk, int tstype) 484637846ef0SAlexander Duyck { 484737846ef0SAlexander Duyck struct sk_buff *skb; 48484ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 484937846ef0SAlexander Duyck 48503a8dd971SWillem de Bruijn if (!sk) 48513a8dd971SWillem de Bruijn return; 48523a8dd971SWillem de Bruijn 4853b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4854b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4855b50a5c70SMiroslav Lichvar return; 4856b50a5c70SMiroslav Lichvar 48573a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 48583a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 485937846ef0SAlexander Duyck return; 486037846ef0SAlexander Duyck 48611c885808SFrancis Yan if (tsonly) { 48621c885808SFrancis Yan #ifdef CONFIG_INET 48631c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 486442f67eeaSEric Dumazet sk_is_tcp(sk)) { 4865e7ed11eeSYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb, 4866e7ed11eeSYousuk Seung ack_skb); 48674ef1b286SSoheil Hassas Yeganeh opt_stats = true; 48684ef1b286SSoheil Hassas Yeganeh } else 48691c885808SFrancis Yan #endif 48701c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 48711c885808SFrancis Yan } else { 487237846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 48731c885808SFrancis Yan } 487437846ef0SAlexander Duyck if (!skb) 487537846ef0SAlexander Duyck return; 487637846ef0SAlexander Duyck 487749ca0d8bSWillem de Bruijn if (tsonly) { 4878fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4879fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 488049ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 488149ca0d8bSWillem de Bruijn } 488249ca0d8bSWillem de Bruijn 488349ca0d8bSWillem de Bruijn if (hwtstamps) 488449ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 488549ca0d8bSWillem de Bruijn else 4886d93376f5SMartin KaFai Lau __net_timestamp(skb); 488749ca0d8bSWillem de Bruijn 48884ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 488937846ef0SAlexander Duyck } 4890e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4891e7fd2885SWillem de Bruijn 4892e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 4893e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 4894e7fd2885SWillem de Bruijn { 4895e7ed11eeSYousuk Seung return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk, 4896e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 4897e7fd2885SWillem de Bruijn } 4898ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4899ac45f602SPatrick Ohly 49006e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 49016e3e939fSJohannes Berg { 49026e3e939fSJohannes Berg struct sock *sk = skb->sk; 49036e3e939fSJohannes Berg struct sock_exterr_skb *serr; 4904dd4f1072SEric Dumazet int err = 1; 49056e3e939fSJohannes Berg 49066e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 49076e3e939fSJohannes Berg skb->wifi_acked = acked; 49086e3e939fSJohannes Berg 49096e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 49106e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 49116e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 49126e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 49136e3e939fSJohannes Berg 4914dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 4915dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 4916dd4f1072SEric Dumazet */ 491741c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 49186e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 4919dd4f1072SEric Dumazet sock_put(sk); 4920dd4f1072SEric Dumazet } 49216e3e939fSJohannes Berg if (err) 49226e3e939fSJohannes Berg kfree_skb(skb); 49236e3e939fSJohannes Berg } 49246e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 49256e3e939fSJohannes Berg 4926f35d9d8aSRusty Russell /** 4927f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 4928f35d9d8aSRusty Russell * @skb: the skb to set 4929f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 4930f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 4931f35d9d8aSRusty Russell * 4932f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 4933f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4934f35d9d8aSRusty Russell * 4935f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 4936f35d9d8aSRusty Russell * returns false you should drop the packet. 4937f35d9d8aSRusty Russell */ 4938f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4939f35d9d8aSRusty Russell { 494052b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 494152b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 494252b5d6f5SEric Dumazet 494352b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 494452b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 494552b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 4946f35d9d8aSRusty Russell return false; 4947f35d9d8aSRusty Russell } 4948f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 494952b5d6f5SEric Dumazet skb->csum_start = csum_start; 4950f35d9d8aSRusty Russell skb->csum_offset = off; 4951e5d5decaSJason Wang skb_set_transport_header(skb, start); 4952f35d9d8aSRusty Russell return true; 4953f35d9d8aSRusty Russell } 4954b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4955f35d9d8aSRusty Russell 4956ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4957ed1f50c3SPaul Durrant unsigned int max) 4958ed1f50c3SPaul Durrant { 4959ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 4960ed1f50c3SPaul Durrant return 0; 4961ed1f50c3SPaul Durrant 4962ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 4963ed1f50c3SPaul Durrant * won't need to do it again. 4964ed1f50c3SPaul Durrant */ 4965ed1f50c3SPaul Durrant if (max > skb->len) 4966ed1f50c3SPaul Durrant max = skb->len; 4967ed1f50c3SPaul Durrant 4968ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4969ed1f50c3SPaul Durrant return -ENOMEM; 4970ed1f50c3SPaul Durrant 4971ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 4972ed1f50c3SPaul Durrant return -EPROTO; 4973ed1f50c3SPaul Durrant 4974ed1f50c3SPaul Durrant return 0; 4975ed1f50c3SPaul Durrant } 4976ed1f50c3SPaul Durrant 4977f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 4978f9708b43SJan Beulich 4979f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4980f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 4981f9708b43SJan Beulich unsigned int off) 4982f9708b43SJan Beulich { 4983f9708b43SJan Beulich int err; 4984f9708b43SJan Beulich 4985161d1792SKees Cook switch (proto) { 4986f9708b43SJan Beulich case IPPROTO_TCP: 4987f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4988f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 4989f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4990f9708b43SJan Beulich offsetof(struct tcphdr, 4991f9708b43SJan Beulich check))) 4992f9708b43SJan Beulich err = -EPROTO; 4993f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4994f9708b43SJan Beulich 4995f9708b43SJan Beulich case IPPROTO_UDP: 4996f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4997f9708b43SJan Beulich off + sizeof(struct udphdr)); 4998f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 4999f9708b43SJan Beulich offsetof(struct udphdr, 5000f9708b43SJan Beulich check))) 5001f9708b43SJan Beulich err = -EPROTO; 5002f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 5003f9708b43SJan Beulich } 5004f9708b43SJan Beulich 5005f9708b43SJan Beulich return ERR_PTR(-EPROTO); 5006f9708b43SJan Beulich } 5007f9708b43SJan Beulich 5008ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5009ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 5010ed1f50c3SPaul Durrant */ 5011ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 5012ed1f50c3SPaul Durrant 5013f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 5014ed1f50c3SPaul Durrant { 5015ed1f50c3SPaul Durrant unsigned int off; 5016ed1f50c3SPaul Durrant bool fragment; 5017f9708b43SJan Beulich __sum16 *csum; 5018ed1f50c3SPaul Durrant int err; 5019ed1f50c3SPaul Durrant 5020ed1f50c3SPaul Durrant fragment = false; 5021ed1f50c3SPaul Durrant 5022ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5023ed1f50c3SPaul Durrant sizeof(struct iphdr), 5024ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 5025ed1f50c3SPaul Durrant if (err < 0) 5026ed1f50c3SPaul Durrant goto out; 5027ed1f50c3SPaul Durrant 502811f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 5029ed1f50c3SPaul Durrant fragment = true; 5030ed1f50c3SPaul Durrant 5031ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 5032ed1f50c3SPaul Durrant 5033ed1f50c3SPaul Durrant err = -EPROTO; 5034ed1f50c3SPaul Durrant 5035ed1f50c3SPaul Durrant if (fragment) 5036ed1f50c3SPaul Durrant goto out; 5037ed1f50c3SPaul Durrant 5038f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 5039f9708b43SJan Beulich if (IS_ERR(csum)) 5040f9708b43SJan Beulich return PTR_ERR(csum); 5041ed1f50c3SPaul Durrant 5042ed1f50c3SPaul Durrant if (recalculate) 5043f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 5044ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 5045ed1f50c3SPaul Durrant skb->len - off, 5046f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 5047ed1f50c3SPaul Durrant err = 0; 5048ed1f50c3SPaul Durrant 5049ed1f50c3SPaul Durrant out: 5050ed1f50c3SPaul Durrant return err; 5051ed1f50c3SPaul Durrant } 5052ed1f50c3SPaul Durrant 5053ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5054ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 5055ed1f50c3SPaul Durrant */ 5056ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 5057ed1f50c3SPaul Durrant 5058ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 5059ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 5060ed1f50c3SPaul Durrant 5061ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 5062ed1f50c3SPaul Durrant { 5063ed1f50c3SPaul Durrant int err; 5064ed1f50c3SPaul Durrant u8 nexthdr; 5065ed1f50c3SPaul Durrant unsigned int off; 5066ed1f50c3SPaul Durrant unsigned int len; 5067ed1f50c3SPaul Durrant bool fragment; 5068ed1f50c3SPaul Durrant bool done; 5069f9708b43SJan Beulich __sum16 *csum; 5070ed1f50c3SPaul Durrant 5071ed1f50c3SPaul Durrant fragment = false; 5072ed1f50c3SPaul Durrant done = false; 5073ed1f50c3SPaul Durrant 5074ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 5075ed1f50c3SPaul Durrant 5076ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 5077ed1f50c3SPaul Durrant if (err < 0) 5078ed1f50c3SPaul Durrant goto out; 5079ed1f50c3SPaul Durrant 5080ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 5081ed1f50c3SPaul Durrant 5082ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 5083ed1f50c3SPaul Durrant while (off <= len && !done) { 5084ed1f50c3SPaul Durrant switch (nexthdr) { 5085ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 5086ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 5087ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 5088ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 5089ed1f50c3SPaul Durrant 5090ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5091ed1f50c3SPaul Durrant off + 5092ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 5093ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5094ed1f50c3SPaul Durrant if (err < 0) 5095ed1f50c3SPaul Durrant goto out; 5096ed1f50c3SPaul Durrant 5097ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 5098ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5099ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 5100ed1f50c3SPaul Durrant break; 5101ed1f50c3SPaul Durrant } 5102ed1f50c3SPaul Durrant case IPPROTO_AH: { 5103ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 5104ed1f50c3SPaul Durrant 5105ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5106ed1f50c3SPaul Durrant off + 5107ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 5108ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5109ed1f50c3SPaul Durrant if (err < 0) 5110ed1f50c3SPaul Durrant goto out; 5111ed1f50c3SPaul Durrant 5112ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 5113ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5114ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 5115ed1f50c3SPaul Durrant break; 5116ed1f50c3SPaul Durrant } 5117ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 5118ed1f50c3SPaul Durrant struct frag_hdr *hp; 5119ed1f50c3SPaul Durrant 5120ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5121ed1f50c3SPaul Durrant off + 5122ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5123ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5124ed1f50c3SPaul Durrant if (err < 0) 5125ed1f50c3SPaul Durrant goto out; 5126ed1f50c3SPaul Durrant 5127ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5128ed1f50c3SPaul Durrant 5129ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5130ed1f50c3SPaul Durrant fragment = true; 5131ed1f50c3SPaul Durrant 5132ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5133ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5134ed1f50c3SPaul Durrant break; 5135ed1f50c3SPaul Durrant } 5136ed1f50c3SPaul Durrant default: 5137ed1f50c3SPaul Durrant done = true; 5138ed1f50c3SPaul Durrant break; 5139ed1f50c3SPaul Durrant } 5140ed1f50c3SPaul Durrant } 5141ed1f50c3SPaul Durrant 5142ed1f50c3SPaul Durrant err = -EPROTO; 5143ed1f50c3SPaul Durrant 5144ed1f50c3SPaul Durrant if (!done || fragment) 5145ed1f50c3SPaul Durrant goto out; 5146ed1f50c3SPaul Durrant 5147f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5148f9708b43SJan Beulich if (IS_ERR(csum)) 5149f9708b43SJan Beulich return PTR_ERR(csum); 5150ed1f50c3SPaul Durrant 5151ed1f50c3SPaul Durrant if (recalculate) 5152f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5153ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5154f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5155ed1f50c3SPaul Durrant err = 0; 5156ed1f50c3SPaul Durrant 5157ed1f50c3SPaul Durrant out: 5158ed1f50c3SPaul Durrant return err; 5159ed1f50c3SPaul Durrant } 5160ed1f50c3SPaul Durrant 5161ed1f50c3SPaul Durrant /** 5162ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5163ed1f50c3SPaul Durrant * @skb: the skb to set up 5164ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5165ed1f50c3SPaul Durrant */ 5166ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5167ed1f50c3SPaul Durrant { 5168ed1f50c3SPaul Durrant int err; 5169ed1f50c3SPaul Durrant 5170ed1f50c3SPaul Durrant switch (skb->protocol) { 5171ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5172f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5173ed1f50c3SPaul Durrant break; 5174ed1f50c3SPaul Durrant 5175ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5176ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5177ed1f50c3SPaul Durrant break; 5178ed1f50c3SPaul Durrant 5179ed1f50c3SPaul Durrant default: 5180ed1f50c3SPaul Durrant err = -EPROTO; 5181ed1f50c3SPaul Durrant break; 5182ed1f50c3SPaul Durrant } 5183ed1f50c3SPaul Durrant 5184ed1f50c3SPaul Durrant return err; 5185ed1f50c3SPaul Durrant } 5186ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5187ed1f50c3SPaul Durrant 51889afd85c9SLinus Lüssing /** 51899afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 51909afd85c9SLinus Lüssing * @skb: the skb to check 51919afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 51929afd85c9SLinus Lüssing * 51939afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 51949afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 51959afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 51969afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 51979afd85c9SLinus Lüssing * 5198a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5199a516993fSLinus Lüssing * differs from the provided skb. 52009afd85c9SLinus Lüssing */ 52019afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 52029afd85c9SLinus Lüssing unsigned int transport_len) 52039afd85c9SLinus Lüssing { 52049afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52059afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 52069afd85c9SLinus Lüssing int ret; 52079afd85c9SLinus Lüssing 5208a516993fSLinus Lüssing if (skb->len < len) 52099afd85c9SLinus Lüssing return NULL; 5210a516993fSLinus Lüssing else if (skb->len == len) 52119afd85c9SLinus Lüssing return skb; 52129afd85c9SLinus Lüssing 52139afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 52149afd85c9SLinus Lüssing if (!skb_chk) 52159afd85c9SLinus Lüssing return NULL; 52169afd85c9SLinus Lüssing 52179afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 52189afd85c9SLinus Lüssing if (ret) { 52199afd85c9SLinus Lüssing kfree_skb(skb_chk); 52209afd85c9SLinus Lüssing return NULL; 52219afd85c9SLinus Lüssing } 52229afd85c9SLinus Lüssing 52239afd85c9SLinus Lüssing return skb_chk; 52249afd85c9SLinus Lüssing } 52259afd85c9SLinus Lüssing 52269afd85c9SLinus Lüssing /** 52279afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 52289afd85c9SLinus Lüssing * @skb: the skb to check 52299afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 52309afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 52319afd85c9SLinus Lüssing * 52329afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 52339afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 52349afd85c9SLinus Lüssing * 52359afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 52369afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 52379afd85c9SLinus Lüssing * 5238a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5239a516993fSLinus Lüssing * differs from the provided skb. 52409afd85c9SLinus Lüssing */ 52419afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 52429afd85c9SLinus Lüssing unsigned int transport_len, 52439afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 52449afd85c9SLinus Lüssing { 52459afd85c9SLinus Lüssing struct sk_buff *skb_chk; 52469afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5247fcba67c9SLinus Lüssing __sum16 ret; 52489afd85c9SLinus Lüssing 52499afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 52509afd85c9SLinus Lüssing if (!skb_chk) 5251a516993fSLinus Lüssing goto err; 52529afd85c9SLinus Lüssing 5253a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5254a516993fSLinus Lüssing goto err; 52559afd85c9SLinus Lüssing 52569b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 52579afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 52589b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 52599afd85c9SLinus Lüssing 5260a516993fSLinus Lüssing if (ret) 5261a516993fSLinus Lüssing goto err; 52629afd85c9SLinus Lüssing 52639afd85c9SLinus Lüssing return skb_chk; 5264a516993fSLinus Lüssing 5265a516993fSLinus Lüssing err: 5266a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5267a516993fSLinus Lüssing kfree_skb(skb_chk); 5268a516993fSLinus Lüssing 5269a516993fSLinus Lüssing return NULL; 5270a516993fSLinus Lüssing 52719afd85c9SLinus Lüssing } 52729afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 52739afd85c9SLinus Lüssing 52744497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 52754497b076SBen Hutchings { 5276e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5277e87cc472SJoe Perches skb->dev->name); 52784497b076SBen Hutchings } 52794497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5280bad43ca8SEric Dumazet 5281bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5282bad43ca8SEric Dumazet { 52833d861f66SEric Dumazet if (head_stolen) { 52843d861f66SEric Dumazet skb_release_head_state(skb); 5285bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 52863d861f66SEric Dumazet } else { 5287bad43ca8SEric Dumazet __kfree_skb(skb); 5288bad43ca8SEric Dumazet } 52893d861f66SEric Dumazet } 5290bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5291bad43ca8SEric Dumazet 5292bad43ca8SEric Dumazet /** 5293bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5294bad43ca8SEric Dumazet * @to: prior buffer 5295bad43ca8SEric Dumazet * @from: buffer to add 5296bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5297c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5298bad43ca8SEric Dumazet */ 5299bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5300bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5301bad43ca8SEric Dumazet { 5302c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5303bad43ca8SEric Dumazet int i, delta, len = from->len; 5304bad43ca8SEric Dumazet 5305bad43ca8SEric Dumazet *fragstolen = false; 5306bad43ca8SEric Dumazet 5307bad43ca8SEric Dumazet if (skb_cloned(to)) 5308bad43ca8SEric Dumazet return false; 5309bad43ca8SEric Dumazet 53101effe8caSJean-Philippe Brucker /* In general, avoid mixing slab allocated and page_pool allocated 53111effe8caSJean-Philippe Brucker * pages within the same SKB. However when @to is not pp_recycle and 53121effe8caSJean-Philippe Brucker * @from is cloned, we can transition frag pages from page_pool to 53131effe8caSJean-Philippe Brucker * reference counted. 53141effe8caSJean-Philippe Brucker * 53151effe8caSJean-Philippe Brucker * On the other hand, don't allow coalescing two pp_recycle SKBs if 53161effe8caSJean-Philippe Brucker * @from is cloned, in case the SKB is using page_pool fragment 53171effe8caSJean-Philippe Brucker * references (PP_FLAG_PAGE_FRAG). Since we only take full page 53181effe8caSJean-Philippe Brucker * references for cloned SKBs at the moment that would result in 53191effe8caSJean-Philippe Brucker * inconsistent reference counts. 53206a5bcd84SIlias Apalodimas */ 53211effe8caSJean-Philippe Brucker if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from))) 53226a5bcd84SIlias Apalodimas return false; 53236a5bcd84SIlias Apalodimas 5324bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5325e93a0435SEric Dumazet if (len) 5326bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5327bad43ca8SEric Dumazet *delta_truesize = 0; 5328bad43ca8SEric Dumazet return true; 5329bad43ca8SEric Dumazet } 5330bad43ca8SEric Dumazet 5331c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5332c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5333c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5334bad43ca8SEric Dumazet return false; 53351f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 53361f8b977aSWillem de Bruijn return false; 5337bad43ca8SEric Dumazet 5338bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5339bad43ca8SEric Dumazet struct page *page; 5340bad43ca8SEric Dumazet unsigned int offset; 5341bad43ca8SEric Dumazet 5342c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5343c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5344bad43ca8SEric Dumazet return false; 5345bad43ca8SEric Dumazet 5346bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5347bad43ca8SEric Dumazet return false; 5348bad43ca8SEric Dumazet 5349bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5350bad43ca8SEric Dumazet 5351bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5352bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5353bad43ca8SEric Dumazet 5354c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5355bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5356bad43ca8SEric Dumazet *fragstolen = true; 5357bad43ca8SEric Dumazet } else { 5358c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5359c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5360bad43ca8SEric Dumazet return false; 5361bad43ca8SEric Dumazet 5362f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5363bad43ca8SEric Dumazet } 5364bad43ca8SEric Dumazet 5365bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5366bad43ca8SEric Dumazet 5367c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5368c818fa9eSEric Dumazet from_shinfo->frags, 5369c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5370c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5371bad43ca8SEric Dumazet 5372bad43ca8SEric Dumazet if (!skb_cloned(from)) 5373c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5374bad43ca8SEric Dumazet 53758ea853fdSLi RongQing /* if the skb is not cloned this does nothing 53768ea853fdSLi RongQing * since we set nr_frags to 0. 53778ea853fdSLi RongQing */ 5378c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5379c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5380bad43ca8SEric Dumazet 5381bad43ca8SEric Dumazet to->truesize += delta; 5382bad43ca8SEric Dumazet to->len += len; 5383bad43ca8SEric Dumazet to->data_len += len; 5384bad43ca8SEric Dumazet 5385bad43ca8SEric Dumazet *delta_truesize = delta; 5386bad43ca8SEric Dumazet return true; 5387bad43ca8SEric Dumazet } 5388bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5389621e84d6SNicolas Dichtel 5390621e84d6SNicolas Dichtel /** 53918b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5392621e84d6SNicolas Dichtel * 5393621e84d6SNicolas Dichtel * @skb: buffer to clean 53948b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5395621e84d6SNicolas Dichtel * 53968b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 53978b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 53988b27f277SNicolas Dichtel * operations. 53998b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 54008b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 54018b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5402621e84d6SNicolas Dichtel */ 54038b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5404621e84d6SNicolas Dichtel { 5405621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5406621e84d6SNicolas Dichtel skb->skb_iif = 0; 540760ff7467SWANG Cong skb->ignore_df = 0; 5408621e84d6SNicolas Dichtel skb_dst_drop(skb); 5409174e2381SFlorian Westphal skb_ext_reset(skb); 5410895b5c9fSFlorian Westphal nf_reset_ct(skb); 5411621e84d6SNicolas Dichtel nf_reset_trace(skb); 5412213dd74aSHerbert Xu 54136f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 54146f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5415875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 54166f9a5069SPetr Machata #endif 54176f9a5069SPetr Machata 5418213dd74aSHerbert Xu if (!xnet) 5419213dd74aSHerbert Xu return; 5420213dd74aSHerbert Xu 54212b5ec1a5SYe Yin ipvs_reset(skb); 5422213dd74aSHerbert Xu skb->mark = 0; 5423de799101SMartin KaFai Lau skb_clear_tstamp(skb); 5424621e84d6SNicolas Dichtel } 5425621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5426de960aa9SFlorian Westphal 5427de960aa9SFlorian Westphal /** 5428de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5429de960aa9SFlorian Westphal * 5430de960aa9SFlorian Westphal * @skb: GSO skb 5431de960aa9SFlorian Westphal * 5432de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5433de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5434de960aa9SFlorian Westphal * 5435de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5436de960aa9SFlorian Westphal */ 5437a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5438de960aa9SFlorian Westphal { 5439de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5440f993bc25SFlorian Westphal unsigned int thlen = 0; 5441f993bc25SFlorian Westphal 5442f993bc25SFlorian Westphal if (skb->encapsulation) { 5443f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5444f993bc25SFlorian Westphal skb_transport_header(skb); 5445de960aa9SFlorian Westphal 5446de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5447f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5448f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5449f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 54501dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 545190017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5452ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5453ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5454f993bc25SFlorian Westphal } 54556d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 54566d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 54576d39d589SFlorian Westphal * accounted for. 54586d39d589SFlorian Westphal */ 5459f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5460de960aa9SFlorian Westphal } 5461a4a77718SDaniel Axtens 5462a4a77718SDaniel Axtens /** 5463a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5464a4a77718SDaniel Axtens * 5465a4a77718SDaniel Axtens * @skb: GSO skb 5466a4a77718SDaniel Axtens * 5467a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5468a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5469a4a77718SDaniel Axtens * 5470a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5471a4a77718SDaniel Axtens */ 5472a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5473a4a77718SDaniel Axtens { 5474a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5475a4a77718SDaniel Axtens skb_network_header(skb); 5476a4a77718SDaniel Axtens 5477a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5478a4a77718SDaniel Axtens } 5479a4a77718SDaniel Axtens 5480a4a77718SDaniel Axtens /** 5481a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5482a4a77718SDaniel Axtens * 5483a4a77718SDaniel Axtens * @skb: GSO skb 5484a4a77718SDaniel Axtens * 5485a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5486a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5487a4a77718SDaniel Axtens * headers (TCP/UDP). 5488a4a77718SDaniel Axtens */ 5489a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5490a4a77718SDaniel Axtens { 5491a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5492a4a77718SDaniel Axtens 5493a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5494a4a77718SDaniel Axtens } 54950d5501c1SVlad Yasevich 5496ae7ef81eSMarcelo Ricardo Leitner /** 54972b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 54982b16f048SDaniel Axtens * 54992b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 55002b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 55012b16f048SDaniel Axtens * 55022b16f048SDaniel Axtens * We might want to check: 55032b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 55042b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 55052b16f048SDaniel Axtens * 55062b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 55072b16f048SDaniel Axtens * 550849682bfaSMathieu Malaterre * @skb: GSO skb 550949682bfaSMathieu Malaterre * 55102b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 55112b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 55122b16f048SDaniel Axtens * 55132b16f048SDaniel Axtens * @max_len: The maximum permissible length. 55142b16f048SDaniel Axtens * 55152b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 55162b16f048SDaniel Axtens */ 55172b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 55182b16f048SDaniel Axtens unsigned int seg_len, 55192b16f048SDaniel Axtens unsigned int max_len) { 55202b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 55212b16f048SDaniel Axtens const struct sk_buff *iter; 55222b16f048SDaniel Axtens 55232b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 55242b16f048SDaniel Axtens return seg_len <= max_len; 55252b16f048SDaniel Axtens 55262b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 55272b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 55282b16f048SDaniel Axtens 55292b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 55302b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 55312b16f048SDaniel Axtens return false; 55322b16f048SDaniel Axtens } 55332b16f048SDaniel Axtens 55342b16f048SDaniel Axtens return true; 55352b16f048SDaniel Axtens } 55362b16f048SDaniel Axtens 55372b16f048SDaniel Axtens /** 5538779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5539ae7ef81eSMarcelo Ricardo Leitner * 5540ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 554176f21b99SDavid S. Miller * @mtu: MTU to validate against 5542ae7ef81eSMarcelo Ricardo Leitner * 5543779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5544779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5545779b7931SDaniel Axtens * payload. 5546ae7ef81eSMarcelo Ricardo Leitner */ 5547779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5548ae7ef81eSMarcelo Ricardo Leitner { 55492b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5550ae7ef81eSMarcelo Ricardo Leitner } 5551779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5552ae7ef81eSMarcelo Ricardo Leitner 55532b16f048SDaniel Axtens /** 55542b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 55552b16f048SDaniel Axtens * 55562b16f048SDaniel Axtens * @skb: GSO skb 55572b16f048SDaniel Axtens * @len: length to validate against 55582b16f048SDaniel Axtens * 55592b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 55602b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 55612b16f048SDaniel Axtens */ 55622b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 55632b16f048SDaniel Axtens { 55642b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 55652b16f048SDaniel Axtens } 55662b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 55672b16f048SDaniel Axtens 55680d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 55690d5501c1SVlad Yasevich { 5570d85e8be2SYuya Kusakabe int mac_len, meta_len; 5571d85e8be2SYuya Kusakabe void *meta; 55724bbb3e0eSToshiaki Makita 55730d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 55740d5501c1SVlad Yasevich kfree_skb(skb); 55750d5501c1SVlad Yasevich return NULL; 55760d5501c1SVlad Yasevich } 55770d5501c1SVlad Yasevich 55784bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5579ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 55804bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 55814bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5582ae474573SToshiaki Makita } 5583d85e8be2SYuya Kusakabe 5584d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5585d85e8be2SYuya Kusakabe if (meta_len) { 5586d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5587d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5588d85e8be2SYuya Kusakabe } 5589d85e8be2SYuya Kusakabe 55900d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 55910d5501c1SVlad Yasevich return skb; 55920d5501c1SVlad Yasevich } 55930d5501c1SVlad Yasevich 55940d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 55950d5501c1SVlad Yasevich { 55960d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 55970d5501c1SVlad Yasevich u16 vlan_tci; 55980d5501c1SVlad Yasevich 5599df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 56000d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 56010d5501c1SVlad Yasevich return skb; 56020d5501c1SVlad Yasevich } 56030d5501c1SVlad Yasevich 56040d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 56050d5501c1SVlad Yasevich if (unlikely(!skb)) 56060d5501c1SVlad Yasevich goto err_free; 560755eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 560855eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 56090d5501c1SVlad Yasevich goto err_free; 56100d5501c1SVlad Yasevich 56110d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 56120d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 56130d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 56140d5501c1SVlad Yasevich 56150d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 56160d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 56170d5501c1SVlad Yasevich 56180d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 56190d5501c1SVlad Yasevich if (unlikely(!skb)) 56200d5501c1SVlad Yasevich goto err_free; 56210d5501c1SVlad Yasevich 56220d5501c1SVlad Yasevich skb_reset_network_header(skb); 56238be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 56240d5501c1SVlad Yasevich skb_reset_transport_header(skb); 56250d5501c1SVlad Yasevich skb_reset_mac_len(skb); 56260d5501c1SVlad Yasevich 56270d5501c1SVlad Yasevich return skb; 56280d5501c1SVlad Yasevich 56290d5501c1SVlad Yasevich err_free: 56300d5501c1SVlad Yasevich kfree_skb(skb); 56310d5501c1SVlad Yasevich return NULL; 56320d5501c1SVlad Yasevich } 56330d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 56342e4e4410SEric Dumazet 563592ece280SLiu Jian int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len) 5636e2195121SJiri Pirko { 5637e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5638e2195121SJiri Pirko return -ENOMEM; 5639e2195121SJiri Pirko 5640e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5641e2195121SJiri Pirko return 0; 5642e2195121SJiri Pirko 5643e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5644e2195121SJiri Pirko } 5645e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5646e2195121SJiri Pirko 5647bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5648bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5649bfca4c52SShmulik Ladkani */ 5650bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 565193515d53SJiri Pirko { 565293515d53SJiri Pirko struct vlan_hdr *vhdr; 5653b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 565493515d53SJiri Pirko int err; 565593515d53SJiri Pirko 5656b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5657b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5658b6a79208SShmulik Ladkani offset)) { 5659b6a79208SShmulik Ladkani return -EINVAL; 5660b6a79208SShmulik Ladkani } 5661b6a79208SShmulik Ladkani 566293515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 566393515d53SJiri Pirko if (unlikely(err)) 5664b6a79208SShmulik Ladkani return err; 566593515d53SJiri Pirko 566693515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 566793515d53SJiri Pirko 566893515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 566993515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 567093515d53SJiri Pirko 567193515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 567293515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 567393515d53SJiri Pirko 567493515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 567593515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 567693515d53SJiri Pirko 567793515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 567893515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 567993515d53SJiri Pirko 568093515d53SJiri Pirko skb_reset_mac_len(skb); 568193515d53SJiri Pirko 568293515d53SJiri Pirko return err; 568393515d53SJiri Pirko } 5684bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 568593515d53SJiri Pirko 5686b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5687b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5688b6a79208SShmulik Ladkani */ 568993515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 569093515d53SJiri Pirko { 569193515d53SJiri Pirko u16 vlan_tci; 569293515d53SJiri Pirko __be16 vlan_proto; 569393515d53SJiri Pirko int err; 569493515d53SJiri Pirko 5695df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5696b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 569793515d53SJiri Pirko } else { 5698ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 569993515d53SJiri Pirko return 0; 570093515d53SJiri Pirko 570193515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 570293515d53SJiri Pirko if (err) 570393515d53SJiri Pirko return err; 570493515d53SJiri Pirko } 570593515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5706ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 570793515d53SJiri Pirko return 0; 570893515d53SJiri Pirko 570993515d53SJiri Pirko vlan_proto = skb->protocol; 571093515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 571193515d53SJiri Pirko if (unlikely(err)) 571293515d53SJiri Pirko return err; 571393515d53SJiri Pirko 571493515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 571593515d53SJiri Pirko return 0; 571693515d53SJiri Pirko } 571793515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 571893515d53SJiri Pirko 5719b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5720b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5721b6a79208SShmulik Ladkani */ 572293515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 572393515d53SJiri Pirko { 5724df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5725b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 572693515d53SJiri Pirko int err; 572793515d53SJiri Pirko 5728b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5729b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5730b6a79208SShmulik Ladkani offset)) { 5731b6a79208SShmulik Ladkani return -EINVAL; 5732b6a79208SShmulik Ladkani } 5733b6a79208SShmulik Ladkani 573493515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5735df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5736b6a79208SShmulik Ladkani if (err) 573793515d53SJiri Pirko return err; 57389241e2dfSDaniel Borkmann 573993515d53SJiri Pirko skb->protocol = skb->vlan_proto; 574093515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 574193515d53SJiri Pirko 57426b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 574393515d53SJiri Pirko } 574493515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 574593515d53SJiri Pirko return 0; 574693515d53SJiri Pirko } 574793515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 574893515d53SJiri Pirko 574919fbcb36SGuillaume Nault /** 575019fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 575119fbcb36SGuillaume Nault * 575219fbcb36SGuillaume Nault * @skb: Socket buffer to modify 575319fbcb36SGuillaume Nault * 575419fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 575519fbcb36SGuillaume Nault * 575619fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 575719fbcb36SGuillaume Nault * present. 575819fbcb36SGuillaume Nault * 575919fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 576019fbcb36SGuillaume Nault */ 576119fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 576219fbcb36SGuillaume Nault { 576319fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 576419fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 576519fbcb36SGuillaume Nault return -EPROTO; 576619fbcb36SGuillaume Nault 576719fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 576819fbcb36SGuillaume Nault skb_reset_mac_header(skb); 576919fbcb36SGuillaume Nault skb_reset_mac_len(skb); 577019fbcb36SGuillaume Nault 577119fbcb36SGuillaume Nault return 0; 577219fbcb36SGuillaume Nault } 577319fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 577419fbcb36SGuillaume Nault 577519fbcb36SGuillaume Nault /** 577619fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 577719fbcb36SGuillaume Nault * 577819fbcb36SGuillaume Nault * @skb: Socket buffer to modify 577919fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 578019fbcb36SGuillaume Nault * @src: Source MAC address of the new header 578119fbcb36SGuillaume Nault * 578219fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 578319fbcb36SGuillaume Nault * 578419fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 578519fbcb36SGuillaume Nault * 578619fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 578719fbcb36SGuillaume Nault */ 578819fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 578919fbcb36SGuillaume Nault const unsigned char *src) 579019fbcb36SGuillaume Nault { 579119fbcb36SGuillaume Nault struct ethhdr *eth; 579219fbcb36SGuillaume Nault int err; 579319fbcb36SGuillaume Nault 579419fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 579519fbcb36SGuillaume Nault return -EPROTO; 579619fbcb36SGuillaume Nault 579719fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 579819fbcb36SGuillaume Nault if (err < 0) 579919fbcb36SGuillaume Nault return err; 580019fbcb36SGuillaume Nault 580119fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 580219fbcb36SGuillaume Nault skb_reset_mac_header(skb); 580319fbcb36SGuillaume Nault skb_reset_mac_len(skb); 580419fbcb36SGuillaume Nault 580519fbcb36SGuillaume Nault eth = eth_hdr(skb); 580619fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 580719fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 580819fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 580919fbcb36SGuillaume Nault 581019fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 581119fbcb36SGuillaume Nault 581219fbcb36SGuillaume Nault return 0; 581319fbcb36SGuillaume Nault } 581419fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 581519fbcb36SGuillaume Nault 58168822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 58178822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 58188822e270SJohn Hurley __be16 ethertype) 58198822e270SJohn Hurley { 58208822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 58218822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 58228822e270SJohn Hurley 58238822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 58248822e270SJohn Hurley } 58258822e270SJohn Hurley 58268822e270SJohn Hurley hdr->h_proto = ethertype; 58278822e270SJohn Hurley } 58288822e270SJohn Hurley 58298822e270SJohn Hurley /** 5830e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5831e7dbfed1SMartin Varghese * the packet 58328822e270SJohn Hurley * 58338822e270SJohn Hurley * @skb: buffer 58348822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 58358822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 5836fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 5837e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 5838e7dbfed1SMartin Varghese * ethernet 58398822e270SJohn Hurley * 58408822e270SJohn Hurley * Expects skb->data at mac header. 58418822e270SJohn Hurley * 58428822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 58438822e270SJohn Hurley */ 5844fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 5845d04ac224SMartin Varghese int mac_len, bool ethernet) 58468822e270SJohn Hurley { 58478822e270SJohn Hurley struct mpls_shim_hdr *lse; 58488822e270SJohn Hurley int err; 58498822e270SJohn Hurley 58508822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 58518822e270SJohn Hurley return -EINVAL; 58528822e270SJohn Hurley 58538822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 58548822e270SJohn Hurley if (skb->encapsulation) 58558822e270SJohn Hurley return -EINVAL; 58568822e270SJohn Hurley 58578822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 58588822e270SJohn Hurley if (unlikely(err)) 58598822e270SJohn Hurley return err; 58608822e270SJohn Hurley 58618822e270SJohn Hurley if (!skb->inner_protocol) { 5862e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 58638822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 58648822e270SJohn Hurley } 58658822e270SJohn Hurley 58668822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 58678822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 5868fa4e0f88SDavide Caratti mac_len); 58698822e270SJohn Hurley skb_reset_mac_header(skb); 5870fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5871e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 58728822e270SJohn Hurley 58738822e270SJohn Hurley lse = mpls_hdr(skb); 58748822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 58758822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 58768822e270SJohn Hurley 58774296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 58788822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 58798822e270SJohn Hurley skb->protocol = mpls_proto; 58808822e270SJohn Hurley 58818822e270SJohn Hurley return 0; 58828822e270SJohn Hurley } 58838822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 58848822e270SJohn Hurley 58852e4e4410SEric Dumazet /** 5886ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 5887ed246ceeSJohn Hurley * 5888ed246ceeSJohn Hurley * @skb: buffer 5889ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 5890fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 589176f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 5892ed246ceeSJohn Hurley * 5893ed246ceeSJohn Hurley * Expects skb->data at mac header. 5894ed246ceeSJohn Hurley * 5895ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 5896ed246ceeSJohn Hurley */ 5897040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 5898040b5cfbSMartin Varghese bool ethernet) 5899ed246ceeSJohn Hurley { 5900ed246ceeSJohn Hurley int err; 5901ed246ceeSJohn Hurley 5902ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5903dedc5a08SDavide Caratti return 0; 5904ed246ceeSJohn Hurley 5905fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 5906ed246ceeSJohn Hurley if (unlikely(err)) 5907ed246ceeSJohn Hurley return err; 5908ed246ceeSJohn Hurley 5909ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 5910ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 5911fa4e0f88SDavide Caratti mac_len); 5912ed246ceeSJohn Hurley 5913ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 5914ed246ceeSJohn Hurley skb_reset_mac_header(skb); 5915fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 5916ed246ceeSJohn Hurley 59174296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 5918ed246ceeSJohn Hurley struct ethhdr *hdr; 5919ed246ceeSJohn Hurley 5920ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 5921ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 5922ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 5923ed246ceeSJohn Hurley } 5924ed246ceeSJohn Hurley skb->protocol = next_proto; 5925ed246ceeSJohn Hurley 5926ed246ceeSJohn Hurley return 0; 5927ed246ceeSJohn Hurley } 5928ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 5929ed246ceeSJohn Hurley 5930ed246ceeSJohn Hurley /** 5931d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 5932d27cf5c5SJohn Hurley * 5933d27cf5c5SJohn Hurley * @skb: buffer 5934d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 5935d27cf5c5SJohn Hurley * 5936d27cf5c5SJohn Hurley * Expects skb->data at mac header. 5937d27cf5c5SJohn Hurley * 5938d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 5939d27cf5c5SJohn Hurley */ 5940d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 5941d27cf5c5SJohn Hurley { 5942d27cf5c5SJohn Hurley int err; 5943d27cf5c5SJohn Hurley 5944d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 5945d27cf5c5SJohn Hurley return -EINVAL; 5946d27cf5c5SJohn Hurley 5947d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 5948d27cf5c5SJohn Hurley if (unlikely(err)) 5949d27cf5c5SJohn Hurley return err; 5950d27cf5c5SJohn Hurley 5951d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 5952d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 5953d27cf5c5SJohn Hurley 5954d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 5955d27cf5c5SJohn Hurley } 5956d27cf5c5SJohn Hurley 5957d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 5958d27cf5c5SJohn Hurley 5959d27cf5c5SJohn Hurley return 0; 5960d27cf5c5SJohn Hurley } 5961d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 5962d27cf5c5SJohn Hurley 5963d27cf5c5SJohn Hurley /** 59642a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 59652a2ea508SJohn Hurley * 59662a2ea508SJohn Hurley * @skb: buffer 59672a2ea508SJohn Hurley * 59682a2ea508SJohn Hurley * Expects skb->data at mac header. 59692a2ea508SJohn Hurley * 59702a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 59712a2ea508SJohn Hurley */ 59722a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 59732a2ea508SJohn Hurley { 59742a2ea508SJohn Hurley u32 lse; 59752a2ea508SJohn Hurley u8 ttl; 59762a2ea508SJohn Hurley 59772a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 59782a2ea508SJohn Hurley return -EINVAL; 59792a2ea508SJohn Hurley 598013de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 598113de4ed9SDavide Caratti return -ENOMEM; 598213de4ed9SDavide Caratti 59832a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 59842a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 59852a2ea508SJohn Hurley if (!--ttl) 59862a2ea508SJohn Hurley return -EINVAL; 59872a2ea508SJohn Hurley 59882a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 59892a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 59902a2ea508SJohn Hurley 59912a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 59922a2ea508SJohn Hurley } 59932a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 59942a2ea508SJohn Hurley 59952a2ea508SJohn Hurley /** 59962e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 59972e4e4410SEric Dumazet * 5998de3f0d0eSMasanari Iida * @header_len: size of linear part 5999de3f0d0eSMasanari Iida * @data_len: needed length in frags 6000de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 6001de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 6002de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 60032e4e4410SEric Dumazet * 60042e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 60052e4e4410SEric Dumazet */ 60062e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 60072e4e4410SEric Dumazet unsigned long data_len, 60082e4e4410SEric Dumazet int max_page_order, 60092e4e4410SEric Dumazet int *errcode, 60102e4e4410SEric Dumazet gfp_t gfp_mask) 60112e4e4410SEric Dumazet { 60122e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 60132e4e4410SEric Dumazet unsigned long chunk; 60142e4e4410SEric Dumazet struct sk_buff *skb; 60152e4e4410SEric Dumazet struct page *page; 60162e4e4410SEric Dumazet int i; 60172e4e4410SEric Dumazet 60182e4e4410SEric Dumazet *errcode = -EMSGSIZE; 60192e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 60202e4e4410SEric Dumazet * high order pages... 60212e4e4410SEric Dumazet */ 60222e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 60232e4e4410SEric Dumazet return NULL; 60242e4e4410SEric Dumazet 60252e4e4410SEric Dumazet *errcode = -ENOBUFS; 6026f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 60272e4e4410SEric Dumazet if (!skb) 60282e4e4410SEric Dumazet return NULL; 60292e4e4410SEric Dumazet 60302e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 60312e4e4410SEric Dumazet 60322e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 60332e4e4410SEric Dumazet int order = max_page_order; 60342e4e4410SEric Dumazet 60352e4e4410SEric Dumazet while (order) { 60362e4e4410SEric Dumazet if (npages >= 1 << order) { 6037d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 60382e4e4410SEric Dumazet __GFP_COMP | 6039d14b56f5SMichal Hocko __GFP_NOWARN, 60402e4e4410SEric Dumazet order); 60412e4e4410SEric Dumazet if (page) 60422e4e4410SEric Dumazet goto fill_page; 60432e4e4410SEric Dumazet /* Do not retry other high order allocations */ 60442e4e4410SEric Dumazet order = 1; 60452e4e4410SEric Dumazet max_page_order = 0; 60462e4e4410SEric Dumazet } 60472e4e4410SEric Dumazet order--; 60482e4e4410SEric Dumazet } 60492e4e4410SEric Dumazet page = alloc_page(gfp_mask); 60502e4e4410SEric Dumazet if (!page) 60512e4e4410SEric Dumazet goto failure; 60522e4e4410SEric Dumazet fill_page: 60532e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 60542e4e4410SEric Dumazet PAGE_SIZE << order); 60552e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 60562e4e4410SEric Dumazet data_len -= chunk; 60572e4e4410SEric Dumazet npages -= 1 << order; 60582e4e4410SEric Dumazet } 60592e4e4410SEric Dumazet return skb; 60602e4e4410SEric Dumazet 60612e4e4410SEric Dumazet failure: 60622e4e4410SEric Dumazet kfree_skb(skb); 60632e4e4410SEric Dumazet return NULL; 60642e4e4410SEric Dumazet } 60652e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 60666fa01ccdSSowmini Varadhan 60676fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 60686fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 60696fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 60706fa01ccdSSowmini Varadhan { 60716fa01ccdSSowmini Varadhan int i; 60726fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 60736fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 60746fa01ccdSSowmini Varadhan u8 *data; 60756fa01ccdSSowmini Varadhan 60766fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 60776fa01ccdSSowmini Varadhan 60786fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 60796fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 60806fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 60816fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 60826fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 60836fa01ccdSSowmini Varadhan if (!data) 60846fa01ccdSSowmini Varadhan return -ENOMEM; 60856fa01ccdSSowmini Varadhan 60866fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 60876fa01ccdSSowmini Varadhan 60886fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 60896fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 60906fa01ccdSSowmini Varadhan skb->len -= off; 60916fa01ccdSSowmini Varadhan 60926fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 60936fa01ccdSSowmini Varadhan skb_shinfo(skb), 60946fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 60956fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 60966fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 60976fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 60986fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 60996fa01ccdSSowmini Varadhan kfree(data); 61006fa01ccdSSowmini Varadhan return -ENOMEM; 61016fa01ccdSSowmini Varadhan } 61026fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 61036fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 61046fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 61056fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 61066fa01ccdSSowmini Varadhan skb_release_data(skb); 61076fa01ccdSSowmini Varadhan } else { 61086fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 61096fa01ccdSSowmini Varadhan * relocate values 61106fa01ccdSSowmini Varadhan */ 61116fa01ccdSSowmini Varadhan skb_free_head(skb); 61126fa01ccdSSowmini Varadhan } 61136fa01ccdSSowmini Varadhan 61146fa01ccdSSowmini Varadhan skb->head = data; 61156fa01ccdSSowmini Varadhan skb->data = data; 61166fa01ccdSSowmini Varadhan skb->head_frag = 0; 6117763087daSEric Dumazet skb_set_end_offset(skb, size); 61186fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 61196fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 61206fa01ccdSSowmini Varadhan skb->cloned = 0; 61216fa01ccdSSowmini Varadhan skb->hdr_len = 0; 61226fa01ccdSSowmini Varadhan skb->nohdr = 0; 61236fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 61246fa01ccdSSowmini Varadhan 61256fa01ccdSSowmini Varadhan return 0; 61266fa01ccdSSowmini Varadhan } 61276fa01ccdSSowmini Varadhan 61286fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 61296fa01ccdSSowmini Varadhan 61306fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 61316fa01ccdSSowmini Varadhan * pskb_carve() 61326fa01ccdSSowmini Varadhan */ 61336fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 61346fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 61356fa01ccdSSowmini Varadhan gfp_t gfp_mask) 61366fa01ccdSSowmini Varadhan { 61376fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 61386fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 61396fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 61406fa01ccdSSowmini Varadhan 61416fa01ccdSSowmini Varadhan do { 61426fa01ccdSSowmini Varadhan if (!list) { 61436fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 61446fa01ccdSSowmini Varadhan return -EFAULT; 61456fa01ccdSSowmini Varadhan } 61466fa01ccdSSowmini Varadhan if (list->len <= eat) { 61476fa01ccdSSowmini Varadhan /* Eaten as whole. */ 61486fa01ccdSSowmini Varadhan eat -= list->len; 61496fa01ccdSSowmini Varadhan list = list->next; 61506fa01ccdSSowmini Varadhan insp = list; 61516fa01ccdSSowmini Varadhan } else { 61526fa01ccdSSowmini Varadhan /* Eaten partially. */ 61536fa01ccdSSowmini Varadhan if (skb_shared(list)) { 61546fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 61556fa01ccdSSowmini Varadhan if (!clone) 61566fa01ccdSSowmini Varadhan return -ENOMEM; 61576fa01ccdSSowmini Varadhan insp = list->next; 61586fa01ccdSSowmini Varadhan list = clone; 61596fa01ccdSSowmini Varadhan } else { 61606fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 61616fa01ccdSSowmini Varadhan insp = list; 61626fa01ccdSSowmini Varadhan } 61636fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 61646fa01ccdSSowmini Varadhan kfree_skb(clone); 61656fa01ccdSSowmini Varadhan return -ENOMEM; 61666fa01ccdSSowmini Varadhan } 61676fa01ccdSSowmini Varadhan break; 61686fa01ccdSSowmini Varadhan } 61696fa01ccdSSowmini Varadhan } while (eat); 61706fa01ccdSSowmini Varadhan 61716fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 61726fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 61736fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 6174ef527f96SEric Dumazet consume_skb(list); 61756fa01ccdSSowmini Varadhan } 61766fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 61776fa01ccdSSowmini Varadhan if (clone) { 61786fa01ccdSSowmini Varadhan clone->next = list; 61796fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 61806fa01ccdSSowmini Varadhan } 61816fa01ccdSSowmini Varadhan return 0; 61826fa01ccdSSowmini Varadhan } 61836fa01ccdSSowmini Varadhan 61846fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 61856fa01ccdSSowmini Varadhan * non-linear part of skb 61866fa01ccdSSowmini Varadhan */ 61876fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 61886fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 61896fa01ccdSSowmini Varadhan { 61906fa01ccdSSowmini Varadhan int i, k = 0; 61916fa01ccdSSowmini Varadhan int size = skb_end_offset(skb); 61926fa01ccdSSowmini Varadhan u8 *data; 61936fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 61946fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 61956fa01ccdSSowmini Varadhan 61966fa01ccdSSowmini Varadhan size = SKB_DATA_ALIGN(size); 61976fa01ccdSSowmini Varadhan 61986fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 61996fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 62006fa01ccdSSowmini Varadhan data = kmalloc_reserve(size + 62016fa01ccdSSowmini Varadhan SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 62026fa01ccdSSowmini Varadhan gfp_mask, NUMA_NO_NODE, NULL); 62036fa01ccdSSowmini Varadhan if (!data) 62046fa01ccdSSowmini Varadhan return -ENOMEM; 62056fa01ccdSSowmini Varadhan 62066fa01ccdSSowmini Varadhan size = SKB_WITH_OVERHEAD(ksize(data)); 62076fa01ccdSSowmini Varadhan 62086fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6209e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 62106fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 62116fa01ccdSSowmini Varadhan kfree(data); 62126fa01ccdSSowmini Varadhan return -ENOMEM; 62136fa01ccdSSowmini Varadhan } 62146fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 62156fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 62166fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 62176fa01ccdSSowmini Varadhan 62186fa01ccdSSowmini Varadhan if (pos + fsize > off) { 62196fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 62206fa01ccdSSowmini Varadhan 62216fa01ccdSSowmini Varadhan if (pos < off) { 62226fa01ccdSSowmini Varadhan /* Split frag. 62236fa01ccdSSowmini Varadhan * We have two variants in this case: 62246fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 62256fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 62266fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 62276fa01ccdSSowmini Varadhan * where splitting is expensive. 62286fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 62296fa01ccdSSowmini Varadhan */ 6230b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 62316fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 62326fa01ccdSSowmini Varadhan } 62336fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 62346fa01ccdSSowmini Varadhan k++; 62356fa01ccdSSowmini Varadhan } 62366fa01ccdSSowmini Varadhan pos += fsize; 62376fa01ccdSSowmini Varadhan } 62386fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 62396fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 62406fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 62416fa01ccdSSowmini Varadhan 62426fa01ccdSSowmini Varadhan /* split line is in frag list */ 6243eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6244eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6245eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6246eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6247eabe8618SMiaohe Lin kfree(data); 6248eabe8618SMiaohe Lin return -ENOMEM; 62496fa01ccdSSowmini Varadhan } 62506fa01ccdSSowmini Varadhan skb_release_data(skb); 62516fa01ccdSSowmini Varadhan 62526fa01ccdSSowmini Varadhan skb->head = data; 62536fa01ccdSSowmini Varadhan skb->head_frag = 0; 62546fa01ccdSSowmini Varadhan skb->data = data; 6255763087daSEric Dumazet skb_set_end_offset(skb, size); 62566fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 62576fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 62586fa01ccdSSowmini Varadhan skb->cloned = 0; 62596fa01ccdSSowmini Varadhan skb->hdr_len = 0; 62606fa01ccdSSowmini Varadhan skb->nohdr = 0; 62616fa01ccdSSowmini Varadhan skb->len -= off; 62626fa01ccdSSowmini Varadhan skb->data_len = skb->len; 62636fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 62646fa01ccdSSowmini Varadhan return 0; 62656fa01ccdSSowmini Varadhan } 62666fa01ccdSSowmini Varadhan 62676fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 62686fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 62696fa01ccdSSowmini Varadhan { 62706fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 62716fa01ccdSSowmini Varadhan 62726fa01ccdSSowmini Varadhan if (len < headlen) 62736fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 62746fa01ccdSSowmini Varadhan else 62756fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 62766fa01ccdSSowmini Varadhan } 62776fa01ccdSSowmini Varadhan 62786fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 62796fa01ccdSSowmini Varadhan * a new skb 62806fa01ccdSSowmini Varadhan */ 62816fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 62826fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 62836fa01ccdSSowmini Varadhan { 62846fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 62856fa01ccdSSowmini Varadhan 62866fa01ccdSSowmini Varadhan if (!clone) 62876fa01ccdSSowmini Varadhan return NULL; 62886fa01ccdSSowmini Varadhan 62896fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 62906fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 62916fa01ccdSSowmini Varadhan kfree_skb(clone); 62926fa01ccdSSowmini Varadhan return NULL; 62936fa01ccdSSowmini Varadhan } 62946fa01ccdSSowmini Varadhan return clone; 62956fa01ccdSSowmini Varadhan } 62966fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6297c8c8b127SEric Dumazet 6298c8c8b127SEric Dumazet /** 6299c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6300c8c8b127SEric Dumazet * @skb: buffer 6301c8c8b127SEric Dumazet * 6302c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6303c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6304c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6305c8c8b127SEric Dumazet * truesize. 6306c8c8b127SEric Dumazet * Notes: 6307c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6308c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6309c8c8b127SEric Dumazet */ 6310c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6311c8c8b127SEric Dumazet { 63123174fed9SEric Dumazet if (skb->data_len) { 63133174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6314c8c8b127SEric Dumazet skb_cloned(skb)) 6315c8c8b127SEric Dumazet return; 6316c8c8b127SEric Dumazet 6317c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6318c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 63193174fed9SEric Dumazet } 63203174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 63213174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 63223174fed9SEric Dumazet * their truesize. 63233174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 63243174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 63253174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6326c8c8b127SEric Dumazet */ 6327c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6328c8c8b127SEric Dumazet } 6329df5042f4SFlorian Westphal 6330df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6331df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6332df5042f4SFlorian Westphal { 6333df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6334df5042f4SFlorian Westphal } 6335df5042f4SFlorian Westphal 63368b69a803SPaolo Abeni /** 63378b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 63388b69a803SPaolo Abeni * 63394930f483SFlorian Westphal * @flags: See kmalloc(). 63404930f483SFlorian Westphal * 63418b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 63428b69a803SPaolo Abeni * skb via __skb_ext_set(). 63438b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 63448b69a803SPaolo Abeni */ 63454930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6346df5042f4SFlorian Westphal { 63474930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6348df5042f4SFlorian Westphal 6349df5042f4SFlorian Westphal if (new) { 6350df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6351df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6352df5042f4SFlorian Westphal } 6353df5042f4SFlorian Westphal 6354df5042f4SFlorian Westphal return new; 6355df5042f4SFlorian Westphal } 6356df5042f4SFlorian Westphal 63574165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 63584165079bSFlorian Westphal unsigned int old_active) 6359df5042f4SFlorian Westphal { 6360df5042f4SFlorian Westphal struct skb_ext *new; 6361df5042f4SFlorian Westphal 6362df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6363df5042f4SFlorian Westphal return old; 6364df5042f4SFlorian Westphal 6365df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6366df5042f4SFlorian Westphal if (!new) 6367df5042f4SFlorian Westphal return NULL; 6368df5042f4SFlorian Westphal 6369df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6370df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6371df5042f4SFlorian Westphal 63724165079bSFlorian Westphal #ifdef CONFIG_XFRM 63734165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 63744165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 63754165079bSFlorian Westphal unsigned int i; 63764165079bSFlorian Westphal 63774165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 63784165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 63794165079bSFlorian Westphal } 63804165079bSFlorian Westphal #endif 6381df5042f4SFlorian Westphal __skb_ext_put(old); 6382df5042f4SFlorian Westphal return new; 6383df5042f4SFlorian Westphal } 6384df5042f4SFlorian Westphal 6385df5042f4SFlorian Westphal /** 63868b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 63878b69a803SPaolo Abeni * @skb: buffer 63888b69a803SPaolo Abeni * @id: extension id 63898b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 63908b69a803SPaolo Abeni * 63918b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 63928b69a803SPaolo Abeni * 63938b69a803SPaolo Abeni * Returns the pointer to the extension. 63948b69a803SPaolo Abeni */ 63958b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 63968b69a803SPaolo Abeni struct skb_ext *ext) 63978b69a803SPaolo Abeni { 63988b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 63998b69a803SPaolo Abeni 64008b69a803SPaolo Abeni skb_ext_put(skb); 64018b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 64028b69a803SPaolo Abeni ext->chunks = newlen; 64038b69a803SPaolo Abeni ext->offset[id] = newoff; 64048b69a803SPaolo Abeni skb->extensions = ext; 64058b69a803SPaolo Abeni skb->active_extensions = 1 << id; 64068b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 64078b69a803SPaolo Abeni } 64088b69a803SPaolo Abeni 64098b69a803SPaolo Abeni /** 6410df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6411df5042f4SFlorian Westphal * @skb: buffer 6412df5042f4SFlorian Westphal * @id: extension to allocate space for 6413df5042f4SFlorian Westphal * 6414df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6415df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6416df5042f4SFlorian Westphal * is returned. 6417df5042f4SFlorian Westphal * 6418df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6419df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6420df5042f4SFlorian Westphal * 6421df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6422df5042f4SFlorian Westphal */ 6423df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6424df5042f4SFlorian Westphal { 6425df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6426df5042f4SFlorian Westphal unsigned int newlen, newoff; 6427df5042f4SFlorian Westphal 6428df5042f4SFlorian Westphal if (skb->active_extensions) { 6429df5042f4SFlorian Westphal old = skb->extensions; 6430df5042f4SFlorian Westphal 64314165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6432df5042f4SFlorian Westphal if (!new) 6433df5042f4SFlorian Westphal return NULL; 6434df5042f4SFlorian Westphal 6435682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6436df5042f4SFlorian Westphal goto set_active; 6437df5042f4SFlorian Westphal 6438e94e50bdSPaolo Abeni newoff = new->chunks; 6439df5042f4SFlorian Westphal } else { 6440df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6441df5042f4SFlorian Westphal 64424930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6443df5042f4SFlorian Westphal if (!new) 6444df5042f4SFlorian Westphal return NULL; 6445df5042f4SFlorian Westphal } 6446df5042f4SFlorian Westphal 6447df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6448df5042f4SFlorian Westphal new->chunks = newlen; 6449df5042f4SFlorian Westphal new->offset[id] = newoff; 6450df5042f4SFlorian Westphal set_active: 6451b0999f38SPaolo Abeni skb->slow_gro = 1; 6452682ec859SPaolo Abeni skb->extensions = new; 6453df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6454df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6455df5042f4SFlorian Westphal } 6456df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6457df5042f4SFlorian Westphal 64584165079bSFlorian Westphal #ifdef CONFIG_XFRM 64594165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 64604165079bSFlorian Westphal { 64614165079bSFlorian Westphal unsigned int i; 64624165079bSFlorian Westphal 64634165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 64644165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 64654165079bSFlorian Westphal } 64664165079bSFlorian Westphal #endif 64674165079bSFlorian Westphal 646878476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 646978476d31SJeremy Kerr static void skb_ext_put_mctp(struct mctp_flow *flow) 647078476d31SJeremy Kerr { 647178476d31SJeremy Kerr if (flow->key) 647278476d31SJeremy Kerr mctp_key_unref(flow->key); 647378476d31SJeremy Kerr } 647478476d31SJeremy Kerr #endif 647578476d31SJeremy Kerr 6476df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6477df5042f4SFlorian Westphal { 6478df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6479df5042f4SFlorian Westphal 6480df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6481df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6482df5042f4SFlorian Westphal skb->extensions = NULL; 6483df5042f4SFlorian Westphal __skb_ext_put(ext); 64844165079bSFlorian Westphal #ifdef CONFIG_XFRM 64854165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 64864165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 64874165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 64884165079bSFlorian Westphal 64894165079bSFlorian Westphal skb_ext_put_sp(sp); 64904165079bSFlorian Westphal sp->len = 0; 64914165079bSFlorian Westphal #endif 6492df5042f4SFlorian Westphal } 6493df5042f4SFlorian Westphal } 6494df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6495df5042f4SFlorian Westphal 6496df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6497df5042f4SFlorian Westphal { 6498df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6499df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6500df5042f4SFlorian Westphal */ 6501df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6502df5042f4SFlorian Westphal goto free_now; 6503df5042f4SFlorian Westphal 6504df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6505df5042f4SFlorian Westphal return; 6506df5042f4SFlorian Westphal free_now: 65074165079bSFlorian Westphal #ifdef CONFIG_XFRM 65084165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 65094165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 65104165079bSFlorian Westphal #endif 651178476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 651278476d31SJeremy Kerr if (__skb_ext_exist(ext, SKB_EXT_MCTP)) 651378476d31SJeremy Kerr skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP)); 651478476d31SJeremy Kerr #endif 65154165079bSFlorian Westphal 6516df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6517df5042f4SFlorian Westphal } 6518df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6519df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 652068822bdfSEric Dumazet 652168822bdfSEric Dumazet /** 652268822bdfSEric Dumazet * skb_attempt_defer_free - queue skb for remote freeing 652368822bdfSEric Dumazet * @skb: buffer 652468822bdfSEric Dumazet * 652568822bdfSEric Dumazet * Put @skb in a per-cpu list, using the cpu which 652668822bdfSEric Dumazet * allocated the skb/pages to reduce false sharing 652768822bdfSEric Dumazet * and memory zone spinlock contention. 652868822bdfSEric Dumazet */ 652968822bdfSEric Dumazet void skb_attempt_defer_free(struct sk_buff *skb) 653068822bdfSEric Dumazet { 653168822bdfSEric Dumazet int cpu = skb->alloc_cpu; 653268822bdfSEric Dumazet struct softnet_data *sd; 653368822bdfSEric Dumazet unsigned long flags; 653439564c3fSEric Dumazet unsigned int defer_max; 653568822bdfSEric Dumazet bool kick; 653668822bdfSEric Dumazet 653768822bdfSEric Dumazet if (WARN_ON_ONCE(cpu >= nr_cpu_ids) || 653868822bdfSEric Dumazet !cpu_online(cpu) || 653968822bdfSEric Dumazet cpu == raw_smp_processor_id()) { 654039564c3fSEric Dumazet nodefer: __kfree_skb(skb); 654168822bdfSEric Dumazet return; 654268822bdfSEric Dumazet } 654368822bdfSEric Dumazet 654468822bdfSEric Dumazet sd = &per_cpu(softnet_data, cpu); 654539564c3fSEric Dumazet defer_max = READ_ONCE(sysctl_skb_defer_max); 654639564c3fSEric Dumazet if (READ_ONCE(sd->defer_count) >= defer_max) 654739564c3fSEric Dumazet goto nodefer; 654839564c3fSEric Dumazet 654968822bdfSEric Dumazet spin_lock_irqsave(&sd->defer_lock, flags); 6550c09b0cd2SJakub Kicinski /* Send an IPI every time queue reaches half capacity. */ 6551c09b0cd2SJakub Kicinski kick = sd->defer_count == (defer_max >> 1); 6552c09b0cd2SJakub Kicinski /* Paired with the READ_ONCE() few lines above */ 6553c09b0cd2SJakub Kicinski WRITE_ONCE(sd->defer_count, sd->defer_count + 1); 6554c09b0cd2SJakub Kicinski 655568822bdfSEric Dumazet skb->next = sd->defer_list; 655668822bdfSEric Dumazet /* Paired with READ_ONCE() in skb_defer_free_flush() */ 655768822bdfSEric Dumazet WRITE_ONCE(sd->defer_list, skb); 655868822bdfSEric Dumazet spin_unlock_irqrestore(&sd->defer_lock, flags); 655968822bdfSEric Dumazet 656068822bdfSEric Dumazet /* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU 656168822bdfSEric Dumazet * if we are unlucky enough (this seems very unlikely). 656268822bdfSEric Dumazet */ 656397e719a8SEric Dumazet if (unlikely(kick) && !cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) 656468822bdfSEric Dumazet smp_call_function_single_async(cpu, &sd->defer_csd); 656568822bdfSEric Dumazet } 6566