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[] = { 970e84afe8SEric Dumazet [SKB_CONSUMED] = "CONSUMED", 989cb252c4SMenglong Dong DEFINE_DROP_REASON(FN, FN) 999cb252c4SMenglong Dong }; 100ec43908dSMenglong Dong EXPORT_SYMBOL(drop_reasons); 101ec43908dSMenglong Dong 1021da177e4SLinus Torvalds /** 103f05de73bSJean Sacren * skb_panic - private function for out-of-line support 1041da177e4SLinus Torvalds * @skb: buffer 1051da177e4SLinus Torvalds * @sz: size 106f05de73bSJean Sacren * @addr: address 10799d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 1081da177e4SLinus Torvalds * 109f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 110f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 111f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 112f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1131da177e4SLinus Torvalds */ 114f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 11599d5851eSJames Hogan const char msg[]) 1161da177e4SLinus Torvalds { 11741a46913SJesper Dangaard Brouer pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n", 11899d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1194305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 12026095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1211da177e4SLinus Torvalds BUG(); 1221da177e4SLinus Torvalds } 1231da177e4SLinus Torvalds 124f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1251da177e4SLinus Torvalds { 126f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1271da177e4SLinus Torvalds } 1281da177e4SLinus Torvalds 129f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 130f05de73bSJean Sacren { 131f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 132f05de73bSJean Sacren } 133c93bdd0eSMel Gorman 13450fad4b5SAlexander Lobakin #define NAPI_SKB_CACHE_SIZE 64 135f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_BULK 16 136f450d539SAlexander Lobakin #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2) 13750fad4b5SAlexander Lobakin 138dbae2b06SPaolo Abeni #if PAGE_SIZE == SZ_4K 139dbae2b06SPaolo Abeni 140dbae2b06SPaolo Abeni #define NAPI_HAS_SMALL_PAGE_FRAG 1 141dbae2b06SPaolo Abeni #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc) 142dbae2b06SPaolo Abeni 143dbae2b06SPaolo Abeni /* specialized page frag allocator using a single order 0 page 144dbae2b06SPaolo Abeni * and slicing it into 1K sized fragment. Constrained to systems 145dbae2b06SPaolo Abeni * with a very limited amount of 1K fragments fitting a single 146dbae2b06SPaolo Abeni * page - to avoid excessive truesize underestimation 147dbae2b06SPaolo Abeni */ 148dbae2b06SPaolo Abeni 149dbae2b06SPaolo Abeni struct page_frag_1k { 150dbae2b06SPaolo Abeni void *va; 151dbae2b06SPaolo Abeni u16 offset; 152dbae2b06SPaolo Abeni bool pfmemalloc; 153dbae2b06SPaolo Abeni }; 154dbae2b06SPaolo Abeni 155dbae2b06SPaolo Abeni static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp) 156dbae2b06SPaolo Abeni { 157dbae2b06SPaolo Abeni struct page *page; 158dbae2b06SPaolo Abeni int offset; 159dbae2b06SPaolo Abeni 160dbae2b06SPaolo Abeni offset = nc->offset - SZ_1K; 161dbae2b06SPaolo Abeni if (likely(offset >= 0)) 162dbae2b06SPaolo Abeni goto use_frag; 163dbae2b06SPaolo Abeni 164dbae2b06SPaolo Abeni page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); 165dbae2b06SPaolo Abeni if (!page) 166dbae2b06SPaolo Abeni return NULL; 167dbae2b06SPaolo Abeni 168dbae2b06SPaolo Abeni nc->va = page_address(page); 169dbae2b06SPaolo Abeni nc->pfmemalloc = page_is_pfmemalloc(page); 170dbae2b06SPaolo Abeni offset = PAGE_SIZE - SZ_1K; 171dbae2b06SPaolo Abeni page_ref_add(page, offset / SZ_1K); 172dbae2b06SPaolo Abeni 173dbae2b06SPaolo Abeni use_frag: 174dbae2b06SPaolo Abeni nc->offset = offset; 175dbae2b06SPaolo Abeni return nc->va + offset; 176dbae2b06SPaolo Abeni } 177dbae2b06SPaolo Abeni #else 178dbae2b06SPaolo Abeni 179dbae2b06SPaolo Abeni /* the small page is actually unused in this build; add dummy helpers 180dbae2b06SPaolo Abeni * to please the compiler and avoid later preprocessor's conditionals 181dbae2b06SPaolo Abeni */ 182dbae2b06SPaolo Abeni #define NAPI_HAS_SMALL_PAGE_FRAG 0 183dbae2b06SPaolo Abeni #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false 184dbae2b06SPaolo Abeni 185dbae2b06SPaolo Abeni struct page_frag_1k { 186dbae2b06SPaolo Abeni }; 187dbae2b06SPaolo Abeni 188dbae2b06SPaolo Abeni static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask) 189dbae2b06SPaolo Abeni { 190dbae2b06SPaolo Abeni return NULL; 191dbae2b06SPaolo Abeni } 192dbae2b06SPaolo Abeni 193dbae2b06SPaolo Abeni #endif 194dbae2b06SPaolo Abeni 19550fad4b5SAlexander Lobakin struct napi_alloc_cache { 19650fad4b5SAlexander Lobakin struct page_frag_cache page; 197dbae2b06SPaolo Abeni struct page_frag_1k page_small; 19850fad4b5SAlexander Lobakin unsigned int skb_count; 19950fad4b5SAlexander Lobakin void *skb_cache[NAPI_SKB_CACHE_SIZE]; 20050fad4b5SAlexander Lobakin }; 20150fad4b5SAlexander Lobakin 20250fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 20350fad4b5SAlexander Lobakin static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 20450fad4b5SAlexander Lobakin 205dbae2b06SPaolo Abeni /* Double check that napi_get_frags() allocates skbs with 206dbae2b06SPaolo Abeni * skb->head being backed by slab, not a page fragment. 207dbae2b06SPaolo Abeni * This is to make sure bug fixed in 3226b158e67c 208dbae2b06SPaolo Abeni * ("net: avoid 32 x truesize under-estimation for tiny skbs") 209dbae2b06SPaolo Abeni * does not accidentally come back. 210dbae2b06SPaolo Abeni */ 211dbae2b06SPaolo Abeni void napi_get_frags_check(struct napi_struct *napi) 212dbae2b06SPaolo Abeni { 213dbae2b06SPaolo Abeni struct sk_buff *skb; 214dbae2b06SPaolo Abeni 215dbae2b06SPaolo Abeni local_bh_disable(); 216dbae2b06SPaolo Abeni skb = napi_get_frags(napi); 217dbae2b06SPaolo Abeni WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag); 218dbae2b06SPaolo Abeni napi_free_frags(napi); 219dbae2b06SPaolo Abeni local_bh_enable(); 220dbae2b06SPaolo Abeni } 221dbae2b06SPaolo Abeni 22232e3573fSYajun Deng void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 22350fad4b5SAlexander Lobakin { 22450fad4b5SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 22550fad4b5SAlexander Lobakin 22650fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 22750fad4b5SAlexander Lobakin 22832e3573fSYajun Deng return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 22950fad4b5SAlexander Lobakin } 23050fad4b5SAlexander Lobakin EXPORT_SYMBOL(__napi_alloc_frag_align); 23150fad4b5SAlexander Lobakin 23250fad4b5SAlexander Lobakin void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask) 23350fad4b5SAlexander Lobakin { 23450fad4b5SAlexander Lobakin void *data; 23550fad4b5SAlexander Lobakin 23650fad4b5SAlexander Lobakin fragsz = SKB_DATA_ALIGN(fragsz); 237afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 23832e3573fSYajun Deng struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache); 23932e3573fSYajun Deng 24050fad4b5SAlexander Lobakin data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask); 24150fad4b5SAlexander Lobakin } else { 24232e3573fSYajun Deng struct napi_alloc_cache *nc; 24332e3573fSYajun Deng 24450fad4b5SAlexander Lobakin local_bh_disable(); 24532e3573fSYajun Deng nc = this_cpu_ptr(&napi_alloc_cache); 24632e3573fSYajun Deng data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask); 24750fad4b5SAlexander Lobakin local_bh_enable(); 24850fad4b5SAlexander Lobakin } 24950fad4b5SAlexander Lobakin return data; 25050fad4b5SAlexander Lobakin } 25150fad4b5SAlexander Lobakin EXPORT_SYMBOL(__netdev_alloc_frag_align); 25250fad4b5SAlexander Lobakin 253f450d539SAlexander Lobakin static struct sk_buff *napi_skb_cache_get(void) 254f450d539SAlexander Lobakin { 255f450d539SAlexander Lobakin struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 256f450d539SAlexander Lobakin struct sk_buff *skb; 257f450d539SAlexander Lobakin 25849ae83fcSSieng Piaw Liew if (unlikely(!nc->skb_count)) { 259f450d539SAlexander Lobakin nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache, 260f450d539SAlexander Lobakin GFP_ATOMIC, 261f450d539SAlexander Lobakin NAPI_SKB_CACHE_BULK, 262f450d539SAlexander Lobakin nc->skb_cache); 263f450d539SAlexander Lobakin if (unlikely(!nc->skb_count)) 264f450d539SAlexander Lobakin return NULL; 26549ae83fcSSieng Piaw Liew } 266f450d539SAlexander Lobakin 267f450d539SAlexander Lobakin skb = nc->skb_cache[--nc->skb_count]; 268f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, skb); 269f450d539SAlexander Lobakin 270f450d539SAlexander Lobakin return skb; 271f450d539SAlexander Lobakin } 272f450d539SAlexander Lobakin 273ce098da1SKees Cook static inline void __finalize_skb_around(struct sk_buff *skb, void *data, 274ce098da1SKees Cook unsigned int size) 275ba0509b6SJesper Dangaard Brouer { 276ba0509b6SJesper Dangaard Brouer struct skb_shared_info *shinfo; 277ba0509b6SJesper Dangaard Brouer 278ba0509b6SJesper Dangaard Brouer size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 279ba0509b6SJesper Dangaard Brouer 280ba0509b6SJesper Dangaard Brouer /* Assumes caller memset cleared SKB */ 281ba0509b6SJesper Dangaard Brouer skb->truesize = SKB_TRUESIZE(size); 282ba0509b6SJesper Dangaard Brouer refcount_set(&skb->users, 1); 283ba0509b6SJesper Dangaard Brouer skb->head = data; 284ba0509b6SJesper Dangaard Brouer skb->data = data; 285ba0509b6SJesper Dangaard Brouer skb_reset_tail_pointer(skb); 286763087daSEric Dumazet skb_set_end_offset(skb, size); 287ba0509b6SJesper Dangaard Brouer skb->mac_header = (typeof(skb->mac_header))~0U; 288ba0509b6SJesper Dangaard Brouer skb->transport_header = (typeof(skb->transport_header))~0U; 28968822bdfSEric Dumazet skb->alloc_cpu = raw_smp_processor_id(); 290ba0509b6SJesper Dangaard Brouer /* make sure we initialize shinfo sequentially */ 291ba0509b6SJesper Dangaard Brouer shinfo = skb_shinfo(skb); 292ba0509b6SJesper Dangaard Brouer memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 293ba0509b6SJesper Dangaard Brouer atomic_set(&shinfo->dataref, 1); 294ba0509b6SJesper Dangaard Brouer 2956370cc3bSAleksandr Nogikh skb_set_kcov_handle(skb, kcov_common_handle()); 296ba0509b6SJesper Dangaard Brouer } 297ba0509b6SJesper Dangaard Brouer 298ce098da1SKees Cook static inline void *__slab_build_skb(struct sk_buff *skb, void *data, 299ce098da1SKees Cook unsigned int *size) 300ce098da1SKees Cook { 301ce098da1SKees Cook void *resized; 302ce098da1SKees Cook 303ce098da1SKees Cook /* Must find the allocation size (and grow it to match). */ 304ce098da1SKees Cook *size = ksize(data); 305ce098da1SKees Cook /* krealloc() will immediately return "data" when 306ce098da1SKees Cook * "ksize(data)" is requested: it is the existing upper 307ce098da1SKees Cook * bounds. As a result, GFP_ATOMIC will be ignored. Note 308ce098da1SKees Cook * that this "new" pointer needs to be passed back to the 309ce098da1SKees Cook * caller for use so the __alloc_size hinting will be 310ce098da1SKees Cook * tracked correctly. 311ce098da1SKees Cook */ 312ce098da1SKees Cook resized = krealloc(data, *size, GFP_ATOMIC); 313ce098da1SKees Cook WARN_ON_ONCE(resized != data); 314ce098da1SKees Cook return resized; 315ce098da1SKees Cook } 316ce098da1SKees Cook 317ce098da1SKees Cook /* build_skb() variant which can operate on slab buffers. 318ce098da1SKees Cook * Note that this should be used sparingly as slab buffers 319ce098da1SKees Cook * cannot be combined efficiently by GRO! 320ce098da1SKees Cook */ 321ce098da1SKees Cook struct sk_buff *slab_build_skb(void *data) 322ce098da1SKees Cook { 323ce098da1SKees Cook struct sk_buff *skb; 324ce098da1SKees Cook unsigned int size; 325ce098da1SKees Cook 326ce098da1SKees Cook skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 327ce098da1SKees Cook if (unlikely(!skb)) 328ce098da1SKees Cook return NULL; 329ce098da1SKees Cook 330ce098da1SKees Cook memset(skb, 0, offsetof(struct sk_buff, tail)); 331ce098da1SKees Cook data = __slab_build_skb(skb, data, &size); 332ce098da1SKees Cook __finalize_skb_around(skb, data, size); 333ce098da1SKees Cook 334ce098da1SKees Cook return skb; 335ce098da1SKees Cook } 336ce098da1SKees Cook EXPORT_SYMBOL(slab_build_skb); 337ce098da1SKees Cook 338ce098da1SKees Cook /* Caller must provide SKB that is memset cleared */ 339ce098da1SKees Cook static void __build_skb_around(struct sk_buff *skb, void *data, 340ce098da1SKees Cook unsigned int frag_size) 341ce098da1SKees Cook { 342ce098da1SKees Cook unsigned int size = frag_size; 343ce098da1SKees Cook 344ce098da1SKees Cook /* frag_size == 0 is considered deprecated now. Callers 345ce098da1SKees Cook * using slab buffer should use slab_build_skb() instead. 346ce098da1SKees Cook */ 347ce098da1SKees Cook if (WARN_ONCE(size == 0, "Use slab_build_skb() instead")) 348ce098da1SKees Cook data = __slab_build_skb(skb, data, &size); 349ce098da1SKees Cook 350ce098da1SKees Cook __finalize_skb_around(skb, data, size); 351ce098da1SKees Cook } 352ce098da1SKees Cook 3531da177e4SLinus Torvalds /** 3542ea2f62cSEric Dumazet * __build_skb - build a network buffer 355b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 356ce098da1SKees Cook * @frag_size: size of data (must not be 0) 357b2b5ce9dSEric Dumazet * 358b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 359ce098da1SKees Cook * skb_shared_info. @data must have been allocated from the page 360ce098da1SKees Cook * allocator or vmalloc(). (A @frag_size of 0 to indicate a kmalloc() 361ce098da1SKees Cook * allocation is deprecated, and callers should use slab_build_skb() 362ce098da1SKees Cook * instead.) 363b2b5ce9dSEric Dumazet * The return is the new skb buffer. 364b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 365b2b5ce9dSEric Dumazet * Notes : 366b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 367b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 368b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 369b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 370b2b5ce9dSEric Dumazet * before giving packet to stack. 371b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 372b2b5ce9dSEric Dumazet */ 3732ea2f62cSEric Dumazet struct sk_buff *__build_skb(void *data, unsigned int frag_size) 374b2b5ce9dSEric Dumazet { 375b2b5ce9dSEric Dumazet struct sk_buff *skb; 376b2b5ce9dSEric Dumazet 377b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 378ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 379b2b5ce9dSEric Dumazet return NULL; 380b2b5ce9dSEric Dumazet 381b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 382483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 383b2b5ce9dSEric Dumazet 384483126b3SAlexander Lobakin return skb; 385b2b5ce9dSEric Dumazet } 3862ea2f62cSEric Dumazet 3872ea2f62cSEric Dumazet /* build_skb() is wrapper over __build_skb(), that specifically 3882ea2f62cSEric Dumazet * takes care of skb->head and skb->pfmemalloc 3892ea2f62cSEric Dumazet */ 3902ea2f62cSEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 3912ea2f62cSEric Dumazet { 3922ea2f62cSEric Dumazet struct sk_buff *skb = __build_skb(data, frag_size); 3932ea2f62cSEric Dumazet 3942ea2f62cSEric Dumazet if (skb && frag_size) { 3952ea2f62cSEric Dumazet skb->head_frag = 1; 3962f064f34SMichal Hocko if (page_is_pfmemalloc(virt_to_head_page(data))) 3972ea2f62cSEric Dumazet skb->pfmemalloc = 1; 3982ea2f62cSEric Dumazet } 3992ea2f62cSEric Dumazet return skb; 4002ea2f62cSEric Dumazet } 401b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 402b2b5ce9dSEric Dumazet 403ba0509b6SJesper Dangaard Brouer /** 404ba0509b6SJesper Dangaard Brouer * build_skb_around - build a network buffer around provided skb 405ba0509b6SJesper Dangaard Brouer * @skb: sk_buff provide by caller, must be memset cleared 406ba0509b6SJesper Dangaard Brouer * @data: data buffer provided by caller 407*12c1604aSJakub Kicinski * @frag_size: size of data 408ba0509b6SJesper Dangaard Brouer */ 409ba0509b6SJesper Dangaard Brouer struct sk_buff *build_skb_around(struct sk_buff *skb, 410ba0509b6SJesper Dangaard Brouer void *data, unsigned int frag_size) 411ba0509b6SJesper Dangaard Brouer { 412ba0509b6SJesper Dangaard Brouer if (unlikely(!skb)) 413ba0509b6SJesper Dangaard Brouer return NULL; 414ba0509b6SJesper Dangaard Brouer 415483126b3SAlexander Lobakin __build_skb_around(skb, data, frag_size); 416ba0509b6SJesper Dangaard Brouer 417483126b3SAlexander Lobakin if (frag_size) { 418ba0509b6SJesper Dangaard Brouer skb->head_frag = 1; 419ba0509b6SJesper Dangaard Brouer if (page_is_pfmemalloc(virt_to_head_page(data))) 420ba0509b6SJesper Dangaard Brouer skb->pfmemalloc = 1; 421ba0509b6SJesper Dangaard Brouer } 422ba0509b6SJesper Dangaard Brouer return skb; 423ba0509b6SJesper Dangaard Brouer } 424ba0509b6SJesper Dangaard Brouer EXPORT_SYMBOL(build_skb_around); 425ba0509b6SJesper Dangaard Brouer 426f450d539SAlexander Lobakin /** 427f450d539SAlexander Lobakin * __napi_build_skb - build a network buffer 428f450d539SAlexander Lobakin * @data: data buffer provided by caller 429*12c1604aSJakub Kicinski * @frag_size: size of data 430f450d539SAlexander Lobakin * 431f450d539SAlexander Lobakin * Version of __build_skb() that uses NAPI percpu caches to obtain 432f450d539SAlexander Lobakin * skbuff_head instead of inplace allocation. 433f450d539SAlexander Lobakin * 434f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 435f450d539SAlexander Lobakin */ 436f450d539SAlexander Lobakin static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size) 437f450d539SAlexander Lobakin { 438f450d539SAlexander Lobakin struct sk_buff *skb; 439f450d539SAlexander Lobakin 440f450d539SAlexander Lobakin skb = napi_skb_cache_get(); 441f450d539SAlexander Lobakin if (unlikely(!skb)) 442f450d539SAlexander Lobakin return NULL; 443f450d539SAlexander Lobakin 444f450d539SAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 445f450d539SAlexander Lobakin __build_skb_around(skb, data, frag_size); 446f450d539SAlexander Lobakin 447f450d539SAlexander Lobakin return skb; 448f450d539SAlexander Lobakin } 449f450d539SAlexander Lobakin 450f450d539SAlexander Lobakin /** 451f450d539SAlexander Lobakin * napi_build_skb - build a network buffer 452f450d539SAlexander Lobakin * @data: data buffer provided by caller 453*12c1604aSJakub Kicinski * @frag_size: size of data 454f450d539SAlexander Lobakin * 455f450d539SAlexander Lobakin * Version of __napi_build_skb() that takes care of skb->head_frag 456f450d539SAlexander Lobakin * and skb->pfmemalloc when the data is a page or page fragment. 457f450d539SAlexander Lobakin * 458f450d539SAlexander Lobakin * Returns a new &sk_buff on success, %NULL on allocation failure. 459f450d539SAlexander Lobakin */ 460f450d539SAlexander Lobakin struct sk_buff *napi_build_skb(void *data, unsigned int frag_size) 461f450d539SAlexander Lobakin { 462f450d539SAlexander Lobakin struct sk_buff *skb = __napi_build_skb(data, frag_size); 463f450d539SAlexander Lobakin 464f450d539SAlexander Lobakin if (likely(skb) && frag_size) { 465f450d539SAlexander Lobakin skb->head_frag = 1; 466f450d539SAlexander Lobakin skb_propagate_pfmemalloc(virt_to_head_page(data), skb); 467f450d539SAlexander Lobakin } 468f450d539SAlexander Lobakin 469f450d539SAlexander Lobakin return skb; 470f450d539SAlexander Lobakin } 471f450d539SAlexander Lobakin EXPORT_SYMBOL(napi_build_skb); 472f450d539SAlexander Lobakin 4735381b23dSAlexander Lobakin /* 4745381b23dSAlexander Lobakin * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 4755381b23dSAlexander Lobakin * the caller if emergency pfmemalloc reserves are being used. If it is and 4765381b23dSAlexander Lobakin * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 4775381b23dSAlexander Lobakin * may be used. Otherwise, the packet data may be discarded until enough 4785381b23dSAlexander Lobakin * memory is free 4795381b23dSAlexander Lobakin */ 480ef28095fSAlexander Lobakin static void *kmalloc_reserve(size_t size, gfp_t flags, int node, 481ef28095fSAlexander Lobakin bool *pfmemalloc) 4825381b23dSAlexander Lobakin { 4835381b23dSAlexander Lobakin void *obj; 4845381b23dSAlexander Lobakin bool ret_pfmemalloc = false; 4855381b23dSAlexander Lobakin 4865381b23dSAlexander Lobakin /* 4875381b23dSAlexander Lobakin * Try a regular allocation, when that fails and we're not entitled 4885381b23dSAlexander Lobakin * to the reserves, fail. 4895381b23dSAlexander Lobakin */ 4905381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, 4915381b23dSAlexander Lobakin flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 4925381b23dSAlexander Lobakin node); 4935381b23dSAlexander Lobakin if (obj || !(gfp_pfmemalloc_allowed(flags))) 4945381b23dSAlexander Lobakin goto out; 4955381b23dSAlexander Lobakin 4965381b23dSAlexander Lobakin /* Try again but now we are using pfmemalloc reserves */ 4975381b23dSAlexander Lobakin ret_pfmemalloc = true; 4985381b23dSAlexander Lobakin obj = kmalloc_node_track_caller(size, flags, node); 4995381b23dSAlexander Lobakin 5005381b23dSAlexander Lobakin out: 5015381b23dSAlexander Lobakin if (pfmemalloc) 5025381b23dSAlexander Lobakin *pfmemalloc = ret_pfmemalloc; 5035381b23dSAlexander Lobakin 5045381b23dSAlexander Lobakin return obj; 5055381b23dSAlexander Lobakin } 5065381b23dSAlexander Lobakin 5075381b23dSAlexander Lobakin /* Allocate a new skbuff. We do this ourselves so we can fill in a few 5085381b23dSAlexander Lobakin * 'private' fields and also do memory statistics to find all the 5095381b23dSAlexander Lobakin * [BEEP] leaks. 5105381b23dSAlexander Lobakin * 5115381b23dSAlexander Lobakin */ 5125381b23dSAlexander Lobakin 5135381b23dSAlexander Lobakin /** 5145381b23dSAlexander Lobakin * __alloc_skb - allocate a network buffer 5155381b23dSAlexander Lobakin * @size: size to allocate 5165381b23dSAlexander Lobakin * @gfp_mask: allocation mask 5175381b23dSAlexander Lobakin * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 5185381b23dSAlexander Lobakin * instead of head cache and allocate a cloned (child) skb. 5195381b23dSAlexander Lobakin * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 5205381b23dSAlexander Lobakin * allocations in case the data is required for writeback 5215381b23dSAlexander Lobakin * @node: numa node to allocate memory on 5225381b23dSAlexander Lobakin * 5235381b23dSAlexander Lobakin * Allocate a new &sk_buff. The returned buffer has no headroom and a 5245381b23dSAlexander Lobakin * tail room of at least size bytes. The object has a reference count 5255381b23dSAlexander Lobakin * of one. The return is the buffer. On a failure the return is %NULL. 5265381b23dSAlexander Lobakin * 5275381b23dSAlexander Lobakin * Buffers may only be allocated from interrupts using a @gfp_mask of 5285381b23dSAlexander Lobakin * %GFP_ATOMIC. 5295381b23dSAlexander Lobakin */ 5305381b23dSAlexander Lobakin struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 5315381b23dSAlexander Lobakin int flags, int node) 5325381b23dSAlexander Lobakin { 5335381b23dSAlexander Lobakin struct kmem_cache *cache; 5345381b23dSAlexander Lobakin struct sk_buff *skb; 535a5df6333SLi RongQing unsigned int osize; 5365381b23dSAlexander Lobakin bool pfmemalloc; 537a5df6333SLi RongQing u8 *data; 5385381b23dSAlexander Lobakin 5395381b23dSAlexander Lobakin cache = (flags & SKB_ALLOC_FCLONE) 5405381b23dSAlexander Lobakin ? skbuff_fclone_cache : skbuff_head_cache; 5415381b23dSAlexander Lobakin 5425381b23dSAlexander Lobakin if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 5435381b23dSAlexander Lobakin gfp_mask |= __GFP_MEMALLOC; 5445381b23dSAlexander Lobakin 5455381b23dSAlexander Lobakin /* Get the HEAD */ 546d13612b5SAlexander Lobakin if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI && 547d13612b5SAlexander Lobakin likely(node == NUMA_NO_NODE || node == numa_mem_id())) 548d13612b5SAlexander Lobakin skb = napi_skb_cache_get(); 549d13612b5SAlexander Lobakin else 550d13612b5SAlexander Lobakin skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node); 551df1ae022SAlexander Lobakin if (unlikely(!skb)) 552df1ae022SAlexander Lobakin return NULL; 5535381b23dSAlexander Lobakin prefetchw(skb); 5545381b23dSAlexander Lobakin 5555381b23dSAlexander Lobakin /* We do our best to align skb_shared_info on a separate cache 5565381b23dSAlexander Lobakin * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 5575381b23dSAlexander Lobakin * aligned memory blocks, unless SLUB/SLAB debug is enabled. 5585381b23dSAlexander Lobakin * Both skb->head and skb_shared_info are cache line aligned. 5595381b23dSAlexander Lobakin */ 5605381b23dSAlexander Lobakin size = SKB_DATA_ALIGN(size); 5615381b23dSAlexander Lobakin size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 56212d6c1d3SKees Cook osize = kmalloc_size_roundup(size); 56312d6c1d3SKees Cook data = kmalloc_reserve(osize, gfp_mask, node, &pfmemalloc); 564df1ae022SAlexander Lobakin if (unlikely(!data)) 5655381b23dSAlexander Lobakin goto nodata; 56612d6c1d3SKees Cook /* kmalloc_size_roundup() might give us more room than requested. 5675381b23dSAlexander Lobakin * Put skb_shared_info exactly at the end of allocated zone, 5685381b23dSAlexander Lobakin * to allow max possible filling before reallocation. 5695381b23dSAlexander Lobakin */ 570a5df6333SLi RongQing size = SKB_WITH_OVERHEAD(osize); 5715381b23dSAlexander Lobakin prefetchw(data + size); 5725381b23dSAlexander Lobakin 5735381b23dSAlexander Lobakin /* 5745381b23dSAlexander Lobakin * Only clear those fields we need to clear, not those that we will 5755381b23dSAlexander Lobakin * actually initialise below. Hence, don't put any more fields after 5765381b23dSAlexander Lobakin * the tail pointer in struct sk_buff! 5775381b23dSAlexander Lobakin */ 5785381b23dSAlexander Lobakin memset(skb, 0, offsetof(struct sk_buff, tail)); 579a5df6333SLi RongQing __build_skb_around(skb, data, osize); 5805381b23dSAlexander Lobakin skb->pfmemalloc = pfmemalloc; 5815381b23dSAlexander Lobakin 5825381b23dSAlexander Lobakin if (flags & SKB_ALLOC_FCLONE) { 5835381b23dSAlexander Lobakin struct sk_buff_fclones *fclones; 5845381b23dSAlexander Lobakin 5855381b23dSAlexander Lobakin fclones = container_of(skb, struct sk_buff_fclones, skb1); 5865381b23dSAlexander Lobakin 5875381b23dSAlexander Lobakin skb->fclone = SKB_FCLONE_ORIG; 5885381b23dSAlexander Lobakin refcount_set(&fclones->fclone_ref, 1); 5895381b23dSAlexander Lobakin } 5905381b23dSAlexander Lobakin 5915381b23dSAlexander Lobakin return skb; 592df1ae022SAlexander Lobakin 5935381b23dSAlexander Lobakin nodata: 5945381b23dSAlexander Lobakin kmem_cache_free(cache, skb); 595df1ae022SAlexander Lobakin return NULL; 5965381b23dSAlexander Lobakin } 5975381b23dSAlexander Lobakin EXPORT_SYMBOL(__alloc_skb); 5985381b23dSAlexander Lobakin 5997ba7aeabSSebastian Andrzej Siewior /** 600fd11a83dSAlexander Duyck * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 601fd11a83dSAlexander Duyck * @dev: network device to receive on 602d7499160SMasanari Iida * @len: length to allocate 603fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb 604fd11a83dSAlexander Duyck * 605fd11a83dSAlexander Duyck * Allocate a new &sk_buff and assign it a usage count of one. The 606fd11a83dSAlexander Duyck * buffer has NET_SKB_PAD headroom built in. Users should allocate 607fd11a83dSAlexander Duyck * the headroom they think they need without accounting for the 608fd11a83dSAlexander Duyck * built in space. The built in space is used for optimisations. 609fd11a83dSAlexander Duyck * 610fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 611fd11a83dSAlexander Duyck */ 6129451980aSAlexander Duyck struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 6139451980aSAlexander Duyck gfp_t gfp_mask) 614fd11a83dSAlexander Duyck { 615b63ae8caSAlexander Duyck struct page_frag_cache *nc; 616fd11a83dSAlexander Duyck struct sk_buff *skb; 6179451980aSAlexander Duyck bool pfmemalloc; 6189451980aSAlexander Duyck void *data; 619fd11a83dSAlexander Duyck 6209451980aSAlexander Duyck len += NET_SKB_PAD; 621fd11a83dSAlexander Duyck 62266c55602SAlexander Lobakin /* If requested length is either too small or too big, 62366c55602SAlexander Lobakin * we use kmalloc() for skb->head allocation. 62466c55602SAlexander Lobakin */ 62566c55602SAlexander Lobakin if (len <= SKB_WITH_OVERHEAD(1024) || 62666c55602SAlexander Lobakin len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 627d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 628a080e7bdSAlexander Duyck skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 629a080e7bdSAlexander Duyck if (!skb) 630a080e7bdSAlexander Duyck goto skb_fail; 631a080e7bdSAlexander Duyck goto skb_success; 632a080e7bdSAlexander Duyck } 6339451980aSAlexander Duyck 6349451980aSAlexander Duyck len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 6359451980aSAlexander Duyck len = SKB_DATA_ALIGN(len); 6369451980aSAlexander Duyck 6379451980aSAlexander Duyck if (sk_memalloc_socks()) 6389451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 6399451980aSAlexander Duyck 640afa79d08SChangbin Du if (in_hardirq() || irqs_disabled()) { 6419451980aSAlexander Duyck nc = this_cpu_ptr(&netdev_alloc_cache); 6428c2dd3e4SAlexander Duyck data = page_frag_alloc(nc, len, gfp_mask); 6439451980aSAlexander Duyck pfmemalloc = nc->pfmemalloc; 64492dcabd7SSebastian Andrzej Siewior } else { 64592dcabd7SSebastian Andrzej Siewior local_bh_disable(); 64692dcabd7SSebastian Andrzej Siewior nc = this_cpu_ptr(&napi_alloc_cache.page); 64792dcabd7SSebastian Andrzej Siewior data = page_frag_alloc(nc, len, gfp_mask); 64892dcabd7SSebastian Andrzej Siewior pfmemalloc = nc->pfmemalloc; 64992dcabd7SSebastian Andrzej Siewior local_bh_enable(); 65092dcabd7SSebastian Andrzej Siewior } 6519451980aSAlexander Duyck 6529451980aSAlexander Duyck if (unlikely(!data)) 6539451980aSAlexander Duyck return NULL; 6549451980aSAlexander Duyck 6559451980aSAlexander Duyck skb = __build_skb(data, len); 6569451980aSAlexander Duyck if (unlikely(!skb)) { 657181edb2bSAlexander Duyck skb_free_frag(data); 6589451980aSAlexander Duyck return NULL; 6599451980aSAlexander Duyck } 6609451980aSAlexander Duyck 6619451980aSAlexander Duyck if (pfmemalloc) 6629451980aSAlexander Duyck skb->pfmemalloc = 1; 6639451980aSAlexander Duyck skb->head_frag = 1; 6649451980aSAlexander Duyck 665a080e7bdSAlexander Duyck skb_success: 6668af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 6677b2e497aSChristoph Hellwig skb->dev = dev; 668fd11a83dSAlexander Duyck 669a080e7bdSAlexander Duyck skb_fail: 6708af27456SChristoph Hellwig return skb; 6718af27456SChristoph Hellwig } 672b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 6731da177e4SLinus Torvalds 674fd11a83dSAlexander Duyck /** 675fd11a83dSAlexander Duyck * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 676fd11a83dSAlexander Duyck * @napi: napi instance this buffer was allocated for 677d7499160SMasanari Iida * @len: length to allocate 678fd11a83dSAlexander Duyck * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 679fd11a83dSAlexander Duyck * 680fd11a83dSAlexander Duyck * Allocate a new sk_buff for use in NAPI receive. This buffer will 681fd11a83dSAlexander Duyck * attempt to allocate the head from a special reserved region used 682fd11a83dSAlexander Duyck * only for NAPI Rx allocation. By doing this we can save several 683fd11a83dSAlexander Duyck * CPU cycles by avoiding having to disable and re-enable IRQs. 684fd11a83dSAlexander Duyck * 685fd11a83dSAlexander Duyck * %NULL is returned if there is no free memory. 686fd11a83dSAlexander Duyck */ 6879451980aSAlexander Duyck struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 6889451980aSAlexander Duyck gfp_t gfp_mask) 689fd11a83dSAlexander Duyck { 6903226b158SEric Dumazet struct napi_alloc_cache *nc; 691fd11a83dSAlexander Duyck struct sk_buff *skb; 692dbae2b06SPaolo Abeni bool pfmemalloc; 6939451980aSAlexander Duyck void *data; 694fd11a83dSAlexander Duyck 695ee2640dfSEric Dumazet DEBUG_NET_WARN_ON_ONCE(!in_softirq()); 6969451980aSAlexander Duyck len += NET_SKB_PAD + NET_IP_ALIGN; 697fd11a83dSAlexander Duyck 6983226b158SEric Dumazet /* If requested length is either too small or too big, 6993226b158SEric Dumazet * we use kmalloc() for skb->head allocation. 700dbae2b06SPaolo Abeni * When the small frag allocator is available, prefer it over kmalloc 701dbae2b06SPaolo Abeni * for small fragments 7023226b158SEric Dumazet */ 703dbae2b06SPaolo Abeni if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) || 7043226b158SEric Dumazet len > SKB_WITH_OVERHEAD(PAGE_SIZE) || 705d0164adcSMel Gorman (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 706cfb8ec65SAlexander Lobakin skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI, 707cfb8ec65SAlexander Lobakin NUMA_NO_NODE); 708a080e7bdSAlexander Duyck if (!skb) 709a080e7bdSAlexander Duyck goto skb_fail; 710a080e7bdSAlexander Duyck goto skb_success; 711a080e7bdSAlexander Duyck } 7129451980aSAlexander Duyck 7133226b158SEric Dumazet nc = this_cpu_ptr(&napi_alloc_cache); 7149451980aSAlexander Duyck 7159451980aSAlexander Duyck if (sk_memalloc_socks()) 7169451980aSAlexander Duyck gfp_mask |= __GFP_MEMALLOC; 7179451980aSAlexander Duyck 718dbae2b06SPaolo Abeni if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) { 719dbae2b06SPaolo Abeni /* we are artificially inflating the allocation size, but 720dbae2b06SPaolo Abeni * that is not as bad as it may look like, as: 721dbae2b06SPaolo Abeni * - 'len' less than GRO_MAX_HEAD makes little sense 722dbae2b06SPaolo Abeni * - On most systems, larger 'len' values lead to fragment 723dbae2b06SPaolo Abeni * size above 512 bytes 724dbae2b06SPaolo Abeni * - kmalloc would use the kmalloc-1k slab for such values 725dbae2b06SPaolo Abeni * - Builds with smaller GRO_MAX_HEAD will very likely do 726dbae2b06SPaolo Abeni * little networking, as that implies no WiFi and no 727dbae2b06SPaolo Abeni * tunnels support, and 32 bits arches. 728dbae2b06SPaolo Abeni */ 729dbae2b06SPaolo Abeni len = SZ_1K; 730dbae2b06SPaolo Abeni 731dbae2b06SPaolo Abeni data = page_frag_alloc_1k(&nc->page_small, gfp_mask); 732dbae2b06SPaolo Abeni pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small); 733dbae2b06SPaolo Abeni } else { 734dbae2b06SPaolo Abeni len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 735dbae2b06SPaolo Abeni len = SKB_DATA_ALIGN(len); 736dbae2b06SPaolo Abeni 7378c2dd3e4SAlexander Duyck data = page_frag_alloc(&nc->page, len, gfp_mask); 738dbae2b06SPaolo Abeni pfmemalloc = nc->page.pfmemalloc; 739dbae2b06SPaolo Abeni } 740dbae2b06SPaolo Abeni 7419451980aSAlexander Duyck if (unlikely(!data)) 7429451980aSAlexander Duyck return NULL; 7439451980aSAlexander Duyck 744cfb8ec65SAlexander Lobakin skb = __napi_build_skb(data, len); 7459451980aSAlexander Duyck if (unlikely(!skb)) { 746181edb2bSAlexander Duyck skb_free_frag(data); 7479451980aSAlexander Duyck return NULL; 7489451980aSAlexander Duyck } 7499451980aSAlexander Duyck 750dbae2b06SPaolo Abeni if (pfmemalloc) 7519451980aSAlexander Duyck skb->pfmemalloc = 1; 7529451980aSAlexander Duyck skb->head_frag = 1; 7539451980aSAlexander Duyck 754a080e7bdSAlexander Duyck skb_success: 755fd11a83dSAlexander Duyck skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 756fd11a83dSAlexander Duyck skb->dev = napi->dev; 757fd11a83dSAlexander Duyck 758a080e7bdSAlexander Duyck skb_fail: 759fd11a83dSAlexander Duyck return skb; 760fd11a83dSAlexander Duyck } 761fd11a83dSAlexander Duyck EXPORT_SYMBOL(__napi_alloc_skb); 762fd11a83dSAlexander Duyck 763654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 76450269e19SEric Dumazet int size, unsigned int truesize) 765654bed16SPeter Zijlstra { 766654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 767654bed16SPeter Zijlstra skb->len += size; 768654bed16SPeter Zijlstra skb->data_len += size; 76950269e19SEric Dumazet skb->truesize += truesize; 770654bed16SPeter Zijlstra } 771654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 772654bed16SPeter Zijlstra 773f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 774f8e617e1SJason Wang unsigned int truesize) 775f8e617e1SJason Wang { 776f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 777f8e617e1SJason Wang 778f8e617e1SJason Wang skb_frag_size_add(frag, size); 779f8e617e1SJason Wang skb->len += size; 780f8e617e1SJason Wang skb->data_len += size; 781f8e617e1SJason Wang skb->truesize += truesize; 782f8e617e1SJason Wang } 783f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 784f8e617e1SJason Wang 78527b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 7861da177e4SLinus Torvalds { 787bd8a7036SEric Dumazet kfree_skb_list(*listp); 78827b437c8SHerbert Xu *listp = NULL; 7891da177e4SLinus Torvalds } 7901da177e4SLinus Torvalds 79127b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 79227b437c8SHerbert Xu { 79327b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 79427b437c8SHerbert Xu } 79527b437c8SHerbert Xu 7961da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 7971da177e4SLinus Torvalds { 7981da177e4SLinus Torvalds struct sk_buff *list; 7991da177e4SLinus Torvalds 800fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 8011da177e4SLinus Torvalds skb_get(list); 8021da177e4SLinus Torvalds } 8031da177e4SLinus Torvalds 8044727bab4SYunsheng Lin static bool skb_pp_recycle(struct sk_buff *skb, void *data) 8054727bab4SYunsheng Lin { 8064727bab4SYunsheng Lin if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle) 8074727bab4SYunsheng Lin return false; 8084727bab4SYunsheng Lin return page_pool_return_skb_page(virt_to_page(data)); 8094727bab4SYunsheng Lin } 8104727bab4SYunsheng Lin 811d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 812d3836f21SEric Dumazet { 813181edb2bSAlexander Duyck unsigned char *head = skb->head; 814181edb2bSAlexander Duyck 8156a5bcd84SIlias Apalodimas if (skb->head_frag) { 8166a5bcd84SIlias Apalodimas if (skb_pp_recycle(skb, head)) 8176a5bcd84SIlias Apalodimas return; 818181edb2bSAlexander Duyck skb_free_frag(head); 8196a5bcd84SIlias Apalodimas } else { 820181edb2bSAlexander Duyck kfree(head); 821d3836f21SEric Dumazet } 8226a5bcd84SIlias Apalodimas } 823d3836f21SEric Dumazet 824511a3edaSEric Dumazet static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason) 8251da177e4SLinus Torvalds { 826ff04a771SEric Dumazet struct skb_shared_info *shinfo = skb_shinfo(skb); 8271da177e4SLinus Torvalds int i; 828ff04a771SEric Dumazet 829ff04a771SEric Dumazet if (skb->cloned && 830ff04a771SEric Dumazet atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 831ff04a771SEric Dumazet &shinfo->dataref)) 8322cc3aeb5SIlias Apalodimas goto exit; 833ff04a771SEric Dumazet 834753f1ca4SPavel Begunkov if (skb_zcopy(skb)) { 835753f1ca4SPavel Begunkov bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS; 836753f1ca4SPavel Begunkov 83770c43167SJonathan Lemon skb_zcopy_clear(skb, true); 838753f1ca4SPavel Begunkov if (skip_unref) 839753f1ca4SPavel Begunkov goto free_head; 840753f1ca4SPavel Begunkov } 84170c43167SJonathan Lemon 842ff04a771SEric Dumazet for (i = 0; i < shinfo->nr_frags; i++) 8436a5bcd84SIlias Apalodimas __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle); 8441da177e4SLinus Torvalds 845753f1ca4SPavel Begunkov free_head: 846ff04a771SEric Dumazet if (shinfo->frag_list) 847511a3edaSEric Dumazet kfree_skb_list_reason(shinfo->frag_list, reason); 8481da177e4SLinus Torvalds 849d3836f21SEric Dumazet skb_free_head(skb); 8502cc3aeb5SIlias Apalodimas exit: 8512cc3aeb5SIlias Apalodimas /* When we clone an SKB we copy the reycling bit. The pp_recycle 8522cc3aeb5SIlias Apalodimas * bit is only set on the head though, so in order to avoid races 8532cc3aeb5SIlias Apalodimas * while trying to recycle fragments on __skb_frag_unref() we need 8542cc3aeb5SIlias Apalodimas * to make one SKB responsible for triggering the recycle path. 8552cc3aeb5SIlias Apalodimas * So disable the recycling bit if an SKB is cloned and we have 85658e61e41STom Rix * additional references to the fragmented part of the SKB. 8572cc3aeb5SIlias Apalodimas * Eventually the last SKB will have the recycling bit set and it's 8582cc3aeb5SIlias Apalodimas * dataref set to 0, which will trigger the recycling 8592cc3aeb5SIlias Apalodimas */ 8602cc3aeb5SIlias Apalodimas skb->pp_recycle = 0; 8611da177e4SLinus Torvalds } 8621da177e4SLinus Torvalds 8631da177e4SLinus Torvalds /* 8641da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 8651da177e4SLinus Torvalds */ 8662d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 8671da177e4SLinus Torvalds { 868d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones; 869d179cd12SDavid S. Miller 870d179cd12SDavid S. Miller switch (skb->fclone) { 871d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 8721da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 8736ffe75ebSEric Dumazet return; 874d179cd12SDavid S. Miller 875d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 876d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb1); 8776ffe75ebSEric Dumazet 8786ffe75ebSEric Dumazet /* We usually free the clone (TX completion) before original skb 8796ffe75ebSEric Dumazet * This test would have no chance to be true for the clone, 8806ffe75ebSEric Dumazet * while here, branch prediction will be good. 8816ffe75ebSEric Dumazet */ 8822638595aSReshetova, Elena if (refcount_read(&fclones->fclone_ref) == 1) 8836ffe75ebSEric Dumazet goto fastpath; 884d179cd12SDavid S. Miller break; 885d179cd12SDavid S. Miller 8866ffe75ebSEric Dumazet default: /* SKB_FCLONE_CLONE */ 887d0bf4a9eSEric Dumazet fclones = container_of(skb, struct sk_buff_fclones, skb2); 888d179cd12SDavid S. Miller break; 8893ff50b79SStephen Hemminger } 8902638595aSReshetova, Elena if (!refcount_dec_and_test(&fclones->fclone_ref)) 8916ffe75ebSEric Dumazet return; 8926ffe75ebSEric Dumazet fastpath: 8936ffe75ebSEric Dumazet kmem_cache_free(skbuff_fclone_cache, fclones); 8941da177e4SLinus Torvalds } 8951da177e4SLinus Torvalds 8960a463c78SPaolo Abeni void skb_release_head_state(struct sk_buff *skb) 8971da177e4SLinus Torvalds { 898adf30907SEric Dumazet skb_dst_drop(skb); 8991da177e4SLinus Torvalds if (skb->destructor) { 9007890e2f0SEric Dumazet DEBUG_NET_WARN_ON_ONCE(in_hardirq()); 9011da177e4SLinus Torvalds skb->destructor(skb); 9021da177e4SLinus Torvalds } 903a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 904cb9c6836SFlorian Westphal nf_conntrack_put(skb_nfct(skb)); 9052fc72c7bSKOVACS Krisztian #endif 906df5042f4SFlorian Westphal skb_ext_put(skb); 90704a4bb55SLennert Buytenhek } 90804a4bb55SLennert Buytenhek 90904a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 910511a3edaSEric Dumazet static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason) 91104a4bb55SLennert Buytenhek { 91204a4bb55SLennert Buytenhek skb_release_head_state(skb); 913a28b1b90SFlorian Westphal if (likely(skb->head)) 914511a3edaSEric Dumazet skb_release_data(skb, reason); 9152d4baff8SHerbert Xu } 9161da177e4SLinus Torvalds 9172d4baff8SHerbert Xu /** 9182d4baff8SHerbert Xu * __kfree_skb - private function 9192d4baff8SHerbert Xu * @skb: buffer 9202d4baff8SHerbert Xu * 9212d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 9222d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 9232d4baff8SHerbert Xu * always call kfree_skb 9242d4baff8SHerbert Xu */ 9252d4baff8SHerbert Xu 9262d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 9272d4baff8SHerbert Xu { 928511a3edaSEric Dumazet skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED); 9291da177e4SLinus Torvalds kfree_skbmem(skb); 9301da177e4SLinus Torvalds } 931b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 9321da177e4SLinus Torvalds 9331da177e4SLinus Torvalds /** 934c504e5c2SMenglong Dong * kfree_skb_reason - free an sk_buff with special reason 935231d06aeSJörn Engel * @skb: buffer to free 936c504e5c2SMenglong Dong * @reason: reason why this skb is dropped 937231d06aeSJörn Engel * 938231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 939c504e5c2SMenglong Dong * hit zero. Meanwhile, pass the drop reason to 'kfree_skb' 940c504e5c2SMenglong Dong * tracepoint. 941231d06aeSJörn Engel */ 942c205cc75SMenglong Dong void __fix_address 943c205cc75SMenglong Dong kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 944231d06aeSJörn Engel { 945c205cc75SMenglong Dong if (unlikely(!skb_unref(skb))) 946231d06aeSJörn Engel return; 9473889a803SPaolo Abeni 94820bbcd0aSMenglong Dong DEBUG_NET_WARN_ON_ONCE(reason <= 0 || reason >= SKB_DROP_REASON_MAX); 94920bbcd0aSMenglong Dong 9500e84afe8SEric Dumazet if (reason == SKB_CONSUMED) 9510e84afe8SEric Dumazet trace_consume_skb(skb); 9520e84afe8SEric Dumazet else 953c504e5c2SMenglong Dong trace_kfree_skb(skb, __builtin_return_address(0), reason); 954231d06aeSJörn Engel __kfree_skb(skb); 955231d06aeSJörn Engel } 956c504e5c2SMenglong Dong EXPORT_SYMBOL(kfree_skb_reason); 957231d06aeSJörn Engel 958215b0f19SMenglong Dong void kfree_skb_list_reason(struct sk_buff *segs, 959215b0f19SMenglong Dong enum skb_drop_reason reason) 960bd8a7036SEric Dumazet { 961bd8a7036SEric Dumazet while (segs) { 962bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 963bd8a7036SEric Dumazet 964215b0f19SMenglong Dong kfree_skb_reason(segs, reason); 965bd8a7036SEric Dumazet segs = next; 966bd8a7036SEric Dumazet } 967bd8a7036SEric Dumazet } 968215b0f19SMenglong Dong EXPORT_SYMBOL(kfree_skb_list_reason); 969bd8a7036SEric Dumazet 9706413139dSWillem de Bruijn /* Dump skb information and contents. 9716413139dSWillem de Bruijn * 9726413139dSWillem de Bruijn * Must only be called from net_ratelimit()-ed paths. 9736413139dSWillem de Bruijn * 974302af7c6SVladimir Oltean * Dumps whole packets if full_pkt, only headers otherwise. 9756413139dSWillem de Bruijn */ 9766413139dSWillem de Bruijn void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt) 9776413139dSWillem de Bruijn { 9786413139dSWillem de Bruijn struct skb_shared_info *sh = skb_shinfo(skb); 9796413139dSWillem de Bruijn struct net_device *dev = skb->dev; 9806413139dSWillem de Bruijn struct sock *sk = skb->sk; 9816413139dSWillem de Bruijn struct sk_buff *list_skb; 9826413139dSWillem de Bruijn bool has_mac, has_trans; 9836413139dSWillem de Bruijn int headroom, tailroom; 9846413139dSWillem de Bruijn int i, len, seg_len; 9856413139dSWillem de Bruijn 9866413139dSWillem de Bruijn if (full_pkt) 9876413139dSWillem de Bruijn len = skb->len; 9886413139dSWillem de Bruijn else 9896413139dSWillem de Bruijn len = min_t(int, skb->len, MAX_HEADER + 128); 9906413139dSWillem de Bruijn 9916413139dSWillem de Bruijn headroom = skb_headroom(skb); 9926413139dSWillem de Bruijn tailroom = skb_tailroom(skb); 9936413139dSWillem de Bruijn 9946413139dSWillem de Bruijn has_mac = skb_mac_header_was_set(skb); 9956413139dSWillem de Bruijn has_trans = skb_transport_header_was_set(skb); 9966413139dSWillem de Bruijn 9976413139dSWillem de Bruijn printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n" 9986413139dSWillem de Bruijn "mac=(%d,%d) net=(%d,%d) trans=%d\n" 9996413139dSWillem de Bruijn "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n" 10006413139dSWillem de Bruijn "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n" 10016413139dSWillem de Bruijn "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n", 10026413139dSWillem de Bruijn level, skb->len, headroom, skb_headlen(skb), tailroom, 10036413139dSWillem de Bruijn has_mac ? skb->mac_header : -1, 10046413139dSWillem de Bruijn has_mac ? skb_mac_header_len(skb) : -1, 10056413139dSWillem de Bruijn skb->network_header, 10066413139dSWillem de Bruijn has_trans ? skb_network_header_len(skb) : -1, 10076413139dSWillem de Bruijn has_trans ? skb->transport_header : -1, 10086413139dSWillem de Bruijn sh->tx_flags, sh->nr_frags, 10096413139dSWillem de Bruijn sh->gso_size, sh->gso_type, sh->gso_segs, 10106413139dSWillem de Bruijn skb->csum, skb->ip_summed, skb->csum_complete_sw, 10116413139dSWillem de Bruijn skb->csum_valid, skb->csum_level, 10126413139dSWillem de Bruijn skb->hash, skb->sw_hash, skb->l4_hash, 10136413139dSWillem de Bruijn ntohs(skb->protocol), skb->pkt_type, skb->skb_iif); 10146413139dSWillem de Bruijn 10156413139dSWillem de Bruijn if (dev) 10168a03ef67SGal Pressman printk("%sdev name=%s feat=%pNF\n", 10176413139dSWillem de Bruijn level, dev->name, &dev->features); 10186413139dSWillem de Bruijn if (sk) 1019db8051f3SQian Cai printk("%ssk family=%hu type=%u proto=%u\n", 10206413139dSWillem de Bruijn level, sk->sk_family, sk->sk_type, sk->sk_protocol); 10216413139dSWillem de Bruijn 10226413139dSWillem de Bruijn if (full_pkt && headroom) 10236413139dSWillem de Bruijn print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET, 10246413139dSWillem de Bruijn 16, 1, skb->head, headroom, false); 10256413139dSWillem de Bruijn 10266413139dSWillem de Bruijn seg_len = min_t(int, skb_headlen(skb), len); 10276413139dSWillem de Bruijn if (seg_len) 10286413139dSWillem de Bruijn print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET, 10296413139dSWillem de Bruijn 16, 1, skb->data, seg_len, false); 10306413139dSWillem de Bruijn len -= seg_len; 10316413139dSWillem de Bruijn 10326413139dSWillem de Bruijn if (full_pkt && tailroom) 10336413139dSWillem de Bruijn print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET, 10346413139dSWillem de Bruijn 16, 1, skb_tail_pointer(skb), tailroom, false); 10356413139dSWillem de Bruijn 10366413139dSWillem de Bruijn for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) { 10376413139dSWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 10386413139dSWillem de Bruijn u32 p_off, p_len, copied; 10396413139dSWillem de Bruijn struct page *p; 10406413139dSWillem de Bruijn u8 *vaddr; 10416413139dSWillem de Bruijn 1042b54c9d5bSJonathan Lemon skb_frag_foreach_page(frag, skb_frag_off(frag), 10436413139dSWillem de Bruijn skb_frag_size(frag), p, p_off, p_len, 10446413139dSWillem de Bruijn copied) { 10456413139dSWillem de Bruijn seg_len = min_t(int, p_len, len); 10466413139dSWillem de Bruijn vaddr = kmap_atomic(p); 10476413139dSWillem de Bruijn print_hex_dump(level, "skb frag: ", 10486413139dSWillem de Bruijn DUMP_PREFIX_OFFSET, 10496413139dSWillem de Bruijn 16, 1, vaddr + p_off, seg_len, false); 10506413139dSWillem de Bruijn kunmap_atomic(vaddr); 10516413139dSWillem de Bruijn len -= seg_len; 10526413139dSWillem de Bruijn if (!len) 10536413139dSWillem de Bruijn break; 10546413139dSWillem de Bruijn } 10556413139dSWillem de Bruijn } 10566413139dSWillem de Bruijn 10576413139dSWillem de Bruijn if (full_pkt && skb_has_frag_list(skb)) { 10586413139dSWillem de Bruijn printk("skb fraglist:\n"); 10596413139dSWillem de Bruijn skb_walk_frags(skb, list_skb) 10606413139dSWillem de Bruijn skb_dump(level, list_skb, true); 10616413139dSWillem de Bruijn } 10626413139dSWillem de Bruijn } 10636413139dSWillem de Bruijn EXPORT_SYMBOL(skb_dump); 10646413139dSWillem de Bruijn 1065d1a203eaSStephen Hemminger /** 106625121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 106725121173SMichael S. Tsirkin * @skb: buffer that triggered an error 106825121173SMichael S. Tsirkin * 106925121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 107025121173SMichael S. Tsirkin * skb must be freed afterwards. 107125121173SMichael S. Tsirkin */ 107225121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 107325121173SMichael S. Tsirkin { 1074753f1ca4SPavel Begunkov if (skb) { 1075753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 10761f8b977aSWillem de Bruijn skb_zcopy_clear(skb, true); 107725121173SMichael S. Tsirkin } 1078753f1ca4SPavel Begunkov } 107925121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 108025121173SMichael S. Tsirkin 1081be769db2SHerbert Xu #ifdef CONFIG_TRACEPOINTS 108225121173SMichael S. Tsirkin /** 1083ead2ceb0SNeil Horman * consume_skb - free an skbuff 1084ead2ceb0SNeil Horman * @skb: buffer to free 1085ead2ceb0SNeil Horman * 1086ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 1087ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 1088ead2ceb0SNeil Horman * is being dropped after a failure and notes that 1089ead2ceb0SNeil Horman */ 1090ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 1091ead2ceb0SNeil Horman { 10923889a803SPaolo Abeni if (!skb_unref(skb)) 1093ead2ceb0SNeil Horman return; 10943889a803SPaolo Abeni 109507dc22e7SKoki Sanagi trace_consume_skb(skb); 1096ead2ceb0SNeil Horman __kfree_skb(skb); 1097ead2ceb0SNeil Horman } 1098ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 1099be769db2SHerbert Xu #endif 1100ead2ceb0SNeil Horman 11010a463c78SPaolo Abeni /** 1102c1639be9SMauro Carvalho Chehab * __consume_stateless_skb - free an skbuff, assuming it is stateless 11030a463c78SPaolo Abeni * @skb: buffer to free 11040a463c78SPaolo Abeni * 1105ca2c1418SPaolo Abeni * Alike consume_skb(), but this variant assumes that this is the last 1106ca2c1418SPaolo Abeni * skb reference and all the head states have been already dropped 11070a463c78SPaolo Abeni */ 1108ca2c1418SPaolo Abeni void __consume_stateless_skb(struct sk_buff *skb) 11090a463c78SPaolo Abeni { 11100a463c78SPaolo Abeni trace_consume_skb(skb); 1111511a3edaSEric Dumazet skb_release_data(skb, SKB_CONSUMED); 11120a463c78SPaolo Abeni kfree_skbmem(skb); 11130a463c78SPaolo Abeni } 11140a463c78SPaolo Abeni 1115f450d539SAlexander Lobakin static void napi_skb_cache_put(struct sk_buff *skb) 1116795bb1c0SJesper Dangaard Brouer { 1117795bb1c0SJesper Dangaard Brouer struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 1118f450d539SAlexander Lobakin u32 i; 1119795bb1c0SJesper Dangaard Brouer 1120f450d539SAlexander Lobakin kasan_poison_object_data(skbuff_head_cache, skb); 1121795bb1c0SJesper Dangaard Brouer nc->skb_cache[nc->skb_count++] = skb; 1122795bb1c0SJesper Dangaard Brouer 1123795bb1c0SJesper Dangaard Brouer if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 1124f450d539SAlexander Lobakin for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++) 1125f450d539SAlexander Lobakin kasan_unpoison_object_data(skbuff_head_cache, 1126f450d539SAlexander Lobakin nc->skb_cache[i]); 1127f450d539SAlexander Lobakin 1128f450d539SAlexander Lobakin kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF, 1129f450d539SAlexander Lobakin nc->skb_cache + NAPI_SKB_CACHE_HALF); 1130f450d539SAlexander Lobakin nc->skb_count = NAPI_SKB_CACHE_HALF; 1131795bb1c0SJesper Dangaard Brouer } 1132795bb1c0SJesper Dangaard Brouer } 1133f450d539SAlexander Lobakin 113415fad714SJesper Dangaard Brouer void __kfree_skb_defer(struct sk_buff *skb) 113515fad714SJesper Dangaard Brouer { 1136511a3edaSEric Dumazet skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED); 11379243adfcSAlexander Lobakin napi_skb_cache_put(skb); 11389243adfcSAlexander Lobakin } 11399243adfcSAlexander Lobakin 11409243adfcSAlexander Lobakin void napi_skb_free_stolen_head(struct sk_buff *skb) 11419243adfcSAlexander Lobakin { 11429efb4b5bSPaolo Abeni if (unlikely(skb->slow_gro)) { 11438550ff8dSPaul Blakey nf_reset_ct(skb); 11449243adfcSAlexander Lobakin skb_dst_drop(skb); 11459243adfcSAlexander Lobakin skb_ext_put(skb); 11465e10da53SPaolo Abeni skb_orphan(skb); 11479efb4b5bSPaolo Abeni skb->slow_gro = 0; 11489efb4b5bSPaolo Abeni } 1149f450d539SAlexander Lobakin napi_skb_cache_put(skb); 115015fad714SJesper Dangaard Brouer } 1151795bb1c0SJesper Dangaard Brouer 1152795bb1c0SJesper Dangaard Brouer void napi_consume_skb(struct sk_buff *skb, int budget) 1153795bb1c0SJesper Dangaard Brouer { 1154885eb0a5SJesper Dangaard Brouer /* Zero budget indicate non-NAPI context called us, like netpoll */ 1155795bb1c0SJesper Dangaard Brouer if (unlikely(!budget)) { 1156885eb0a5SJesper Dangaard Brouer dev_consume_skb_any(skb); 1157795bb1c0SJesper Dangaard Brouer return; 1158795bb1c0SJesper Dangaard Brouer } 1159795bb1c0SJesper Dangaard Brouer 1160ee2640dfSEric Dumazet DEBUG_NET_WARN_ON_ONCE(!in_softirq()); 11616454eca8SYunsheng Lin 11627608894eSPaolo Abeni if (!skb_unref(skb)) 1163795bb1c0SJesper Dangaard Brouer return; 11647608894eSPaolo Abeni 1165795bb1c0SJesper Dangaard Brouer /* if reaching here SKB is ready to free */ 1166795bb1c0SJesper Dangaard Brouer trace_consume_skb(skb); 1167795bb1c0SJesper Dangaard Brouer 1168795bb1c0SJesper Dangaard Brouer /* if SKB is a clone, don't handle this case */ 1169abbdb5a7SEric Dumazet if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 1170795bb1c0SJesper Dangaard Brouer __kfree_skb(skb); 1171795bb1c0SJesper Dangaard Brouer return; 1172795bb1c0SJesper Dangaard Brouer } 1173795bb1c0SJesper Dangaard Brouer 1174511a3edaSEric Dumazet skb_release_all(skb, SKB_CONSUMED); 1175f450d539SAlexander Lobakin napi_skb_cache_put(skb); 1176795bb1c0SJesper Dangaard Brouer } 1177795bb1c0SJesper Dangaard Brouer EXPORT_SYMBOL(napi_consume_skb); 1178795bb1c0SJesper Dangaard Brouer 117903f61041SKees Cook /* Make sure a field is contained by headers group */ 1180b1937227SEric Dumazet #define CHECK_SKB_FIELD(field) \ 118103f61041SKees Cook BUILD_BUG_ON(offsetof(struct sk_buff, field) != \ 118203f61041SKees Cook offsetof(struct sk_buff, headers.field)); \ 1183b1937227SEric Dumazet 1184dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 1185dec18810SHerbert Xu { 1186dec18810SHerbert Xu new->tstamp = old->tstamp; 1187b1937227SEric Dumazet /* We do not copy old->sk */ 1188dec18810SHerbert Xu new->dev = old->dev; 1189b1937227SEric Dumazet memcpy(new->cb, old->cb, sizeof(old->cb)); 11907fee226aSEric Dumazet skb_dst_copy(new, old); 1191df5042f4SFlorian Westphal __skb_ext_copy(new, old); 1192b1937227SEric Dumazet __nf_copy(new, old, false); 11936aa895b0SPatrick McHardy 119403f61041SKees Cook /* Note : this field could be in the headers group. 1195b1937227SEric Dumazet * It is not yet because we do not want to have a 16 bit hole 1196b1937227SEric Dumazet */ 1197b1937227SEric Dumazet new->queue_mapping = old->queue_mapping; 119806021292SEliezer Tamir 119903f61041SKees Cook memcpy(&new->headers, &old->headers, sizeof(new->headers)); 1200b1937227SEric Dumazet CHECK_SKB_FIELD(protocol); 1201b1937227SEric Dumazet CHECK_SKB_FIELD(csum); 1202b1937227SEric Dumazet CHECK_SKB_FIELD(hash); 1203b1937227SEric Dumazet CHECK_SKB_FIELD(priority); 1204b1937227SEric Dumazet CHECK_SKB_FIELD(skb_iif); 1205b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_proto); 1206b1937227SEric Dumazet CHECK_SKB_FIELD(vlan_tci); 1207b1937227SEric Dumazet CHECK_SKB_FIELD(transport_header); 1208b1937227SEric Dumazet CHECK_SKB_FIELD(network_header); 1209b1937227SEric Dumazet CHECK_SKB_FIELD(mac_header); 1210b1937227SEric Dumazet CHECK_SKB_FIELD(inner_protocol); 1211b1937227SEric Dumazet CHECK_SKB_FIELD(inner_transport_header); 1212b1937227SEric Dumazet CHECK_SKB_FIELD(inner_network_header); 1213b1937227SEric Dumazet CHECK_SKB_FIELD(inner_mac_header); 1214b1937227SEric Dumazet CHECK_SKB_FIELD(mark); 1215b1937227SEric Dumazet #ifdef CONFIG_NETWORK_SECMARK 1216b1937227SEric Dumazet CHECK_SKB_FIELD(secmark); 1217b1937227SEric Dumazet #endif 1218e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 1219b1937227SEric Dumazet CHECK_SKB_FIELD(napi_id); 122006021292SEliezer Tamir #endif 122168822bdfSEric Dumazet CHECK_SKB_FIELD(alloc_cpu); 12222bd82484SEric Dumazet #ifdef CONFIG_XPS 12232bd82484SEric Dumazet CHECK_SKB_FIELD(sender_cpu); 12242bd82484SEric Dumazet #endif 1225b1937227SEric Dumazet #ifdef CONFIG_NET_SCHED 1226b1937227SEric Dumazet CHECK_SKB_FIELD(tc_index); 1227b1937227SEric Dumazet #endif 1228b1937227SEric Dumazet 1229dec18810SHerbert Xu } 1230dec18810SHerbert Xu 123182c49a35SHerbert Xu /* 123282c49a35SHerbert Xu * You should not add any new code to this function. Add it to 123382c49a35SHerbert Xu * __copy_skb_header above instead. 123482c49a35SHerbert Xu */ 1235e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 12361da177e4SLinus Torvalds { 12371da177e4SLinus Torvalds #define C(x) n->x = skb->x 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds n->next = n->prev = NULL; 12401da177e4SLinus Torvalds n->sk = NULL; 1241dec18810SHerbert Xu __copy_skb_header(n, skb); 1242dec18810SHerbert Xu 12431da177e4SLinus Torvalds C(len); 12441da177e4SLinus Torvalds C(data_len); 12453e6b3b2eSAlexey Dobriyan C(mac_len); 1246334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 124702f1c89dSPaul Moore n->cloned = 1; 12481da177e4SLinus Torvalds n->nohdr = 0; 1249b13dda9fSEric Dumazet n->peeked = 0; 1250e78bfb07SStefano Brivio C(pfmemalloc); 12516a5bcd84SIlias Apalodimas C(pp_recycle); 12521da177e4SLinus Torvalds n->destructor = NULL; 12531da177e4SLinus Torvalds C(tail); 12541da177e4SLinus Torvalds C(end); 125502f1c89dSPaul Moore C(head); 1256d3836f21SEric Dumazet C(head_frag); 125702f1c89dSPaul Moore C(data); 125802f1c89dSPaul Moore C(truesize); 125963354797SReshetova, Elena refcount_set(&n->users, 1); 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 12621da177e4SLinus Torvalds skb->cloned = 1; 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds return n; 1265e0053ec0SHerbert Xu #undef C 1266e0053ec0SHerbert Xu } 1267e0053ec0SHerbert Xu 1268e0053ec0SHerbert Xu /** 1269da29e4b4SJakub Kicinski * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg 1270da29e4b4SJakub Kicinski * @first: first sk_buff of the msg 1271da29e4b4SJakub Kicinski */ 1272da29e4b4SJakub Kicinski struct sk_buff *alloc_skb_for_msg(struct sk_buff *first) 1273da29e4b4SJakub Kicinski { 1274da29e4b4SJakub Kicinski struct sk_buff *n; 1275da29e4b4SJakub Kicinski 1276da29e4b4SJakub Kicinski n = alloc_skb(0, GFP_ATOMIC); 1277da29e4b4SJakub Kicinski if (!n) 1278da29e4b4SJakub Kicinski return NULL; 1279da29e4b4SJakub Kicinski 1280da29e4b4SJakub Kicinski n->len = first->len; 1281da29e4b4SJakub Kicinski n->data_len = first->len; 1282da29e4b4SJakub Kicinski n->truesize = first->truesize; 1283da29e4b4SJakub Kicinski 1284da29e4b4SJakub Kicinski skb_shinfo(n)->frag_list = first; 1285da29e4b4SJakub Kicinski 1286da29e4b4SJakub Kicinski __copy_skb_header(n, first); 1287da29e4b4SJakub Kicinski n->destructor = NULL; 1288da29e4b4SJakub Kicinski 1289da29e4b4SJakub Kicinski return n; 1290da29e4b4SJakub Kicinski } 1291da29e4b4SJakub Kicinski EXPORT_SYMBOL_GPL(alloc_skb_for_msg); 1292da29e4b4SJakub Kicinski 1293da29e4b4SJakub Kicinski /** 1294e0053ec0SHerbert Xu * skb_morph - morph one skb into another 1295e0053ec0SHerbert Xu * @dst: the skb to receive the contents 1296e0053ec0SHerbert Xu * @src: the skb to supply the contents 1297e0053ec0SHerbert Xu * 1298e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 1299e0053ec0SHerbert Xu * supplied by the user. 1300e0053ec0SHerbert Xu * 1301e0053ec0SHerbert Xu * The target skb is returned upon exit. 1302e0053ec0SHerbert Xu */ 1303e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 1304e0053ec0SHerbert Xu { 1305511a3edaSEric Dumazet skb_release_all(dst, SKB_CONSUMED); 1306e0053ec0SHerbert Xu return __skb_clone(dst, src); 1307e0053ec0SHerbert Xu } 1308e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 1309e0053ec0SHerbert Xu 13106f89dbceSSowmini Varadhan int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 1311a91dbff5SWillem de Bruijn { 1312a91dbff5SWillem de Bruijn unsigned long max_pg, num_pg, new_pg, old_pg; 1313a91dbff5SWillem de Bruijn struct user_struct *user; 1314a91dbff5SWillem de Bruijn 1315a91dbff5SWillem de Bruijn if (capable(CAP_IPC_LOCK) || !size) 1316a91dbff5SWillem de Bruijn return 0; 1317a91dbff5SWillem de Bruijn 1318a91dbff5SWillem de Bruijn num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 1319a91dbff5SWillem de Bruijn max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1320a91dbff5SWillem de Bruijn user = mmp->user ? : current_user(); 1321a91dbff5SWillem de Bruijn 1322a91dbff5SWillem de Bruijn old_pg = atomic_long_read(&user->locked_vm); 132357fc05e8SEric Dumazet do { 1324a91dbff5SWillem de Bruijn new_pg = old_pg + num_pg; 1325a91dbff5SWillem de Bruijn if (new_pg > max_pg) 1326a91dbff5SWillem de Bruijn return -ENOBUFS; 132757fc05e8SEric Dumazet } while (!atomic_long_try_cmpxchg(&user->locked_vm, &old_pg, new_pg)); 1328a91dbff5SWillem de Bruijn 1329a91dbff5SWillem de Bruijn if (!mmp->user) { 1330a91dbff5SWillem de Bruijn mmp->user = get_uid(user); 1331a91dbff5SWillem de Bruijn mmp->num_pg = num_pg; 1332a91dbff5SWillem de Bruijn } else { 1333a91dbff5SWillem de Bruijn mmp->num_pg += num_pg; 1334a91dbff5SWillem de Bruijn } 1335a91dbff5SWillem de Bruijn 1336a91dbff5SWillem de Bruijn return 0; 1337a91dbff5SWillem de Bruijn } 13386f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_account_pinned_pages); 1339a91dbff5SWillem de Bruijn 13406f89dbceSSowmini Varadhan void mm_unaccount_pinned_pages(struct mmpin *mmp) 1341a91dbff5SWillem de Bruijn { 1342a91dbff5SWillem de Bruijn if (mmp->user) { 1343a91dbff5SWillem de Bruijn atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 1344a91dbff5SWillem de Bruijn free_uid(mmp->user); 1345a91dbff5SWillem de Bruijn } 1346a91dbff5SWillem de Bruijn } 13476f89dbceSSowmini Varadhan EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages); 1348a91dbff5SWillem de Bruijn 1349c67b627eSDavid Ahern static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size) 135052267790SWillem de Bruijn { 1351e7d2b510SPavel Begunkov struct ubuf_info_msgzc *uarg; 135252267790SWillem de Bruijn struct sk_buff *skb; 135352267790SWillem de Bruijn 135452267790SWillem de Bruijn WARN_ON_ONCE(!in_task()); 135552267790SWillem de Bruijn 135652267790SWillem de Bruijn skb = sock_omalloc(sk, 0, GFP_KERNEL); 135752267790SWillem de Bruijn if (!skb) 135852267790SWillem de Bruijn return NULL; 135952267790SWillem de Bruijn 136052267790SWillem de Bruijn BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 136152267790SWillem de Bruijn uarg = (void *)skb->cb; 1362a91dbff5SWillem de Bruijn uarg->mmp.user = NULL; 1363a91dbff5SWillem de Bruijn 1364a91dbff5SWillem de Bruijn if (mm_account_pinned_pages(&uarg->mmp, size)) { 1365a91dbff5SWillem de Bruijn kfree_skb(skb); 1366a91dbff5SWillem de Bruijn return NULL; 1367a91dbff5SWillem de Bruijn } 136852267790SWillem de Bruijn 1369e7d2b510SPavel Begunkov uarg->ubuf.callback = msg_zerocopy_callback; 13704ab6c99dSWillem de Bruijn uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 13714ab6c99dSWillem de Bruijn uarg->len = 1; 13724ab6c99dSWillem de Bruijn uarg->bytelen = size; 137352267790SWillem de Bruijn uarg->zerocopy = 1; 1374e7d2b510SPavel Begunkov uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN; 1375e7d2b510SPavel Begunkov refcount_set(&uarg->ubuf.refcnt, 1); 137652267790SWillem de Bruijn sock_hold(sk); 137752267790SWillem de Bruijn 1378e7d2b510SPavel Begunkov return &uarg->ubuf; 137952267790SWillem de Bruijn } 138052267790SWillem de Bruijn 1381e7d2b510SPavel Begunkov static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg) 138252267790SWillem de Bruijn { 138352267790SWillem de Bruijn return container_of((void *)uarg, struct sk_buff, cb); 138452267790SWillem de Bruijn } 138552267790SWillem de Bruijn 13868c793822SJonathan Lemon struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 13874ab6c99dSWillem de Bruijn struct ubuf_info *uarg) 13884ab6c99dSWillem de Bruijn { 13894ab6c99dSWillem de Bruijn if (uarg) { 1390e7d2b510SPavel Begunkov struct ubuf_info_msgzc *uarg_zc; 13914ab6c99dSWillem de Bruijn const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 13924ab6c99dSWillem de Bruijn u32 bytelen, next; 13934ab6c99dSWillem de Bruijn 13941b4b2b09SPavel Begunkov /* there might be non MSG_ZEROCOPY users */ 13951b4b2b09SPavel Begunkov if (uarg->callback != msg_zerocopy_callback) 13961b4b2b09SPavel Begunkov return NULL; 13971b4b2b09SPavel Begunkov 13984ab6c99dSWillem de Bruijn /* realloc only when socket is locked (TCP, UDP cork), 13994ab6c99dSWillem de Bruijn * so uarg->len and sk_zckey access is serialized 14004ab6c99dSWillem de Bruijn */ 14014ab6c99dSWillem de Bruijn if (!sock_owned_by_user(sk)) { 14024ab6c99dSWillem de Bruijn WARN_ON_ONCE(1); 14034ab6c99dSWillem de Bruijn return NULL; 14044ab6c99dSWillem de Bruijn } 14054ab6c99dSWillem de Bruijn 1406e7d2b510SPavel Begunkov uarg_zc = uarg_to_msgzc(uarg); 1407e7d2b510SPavel Begunkov bytelen = uarg_zc->bytelen + size; 1408e7d2b510SPavel Begunkov if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) { 14094ab6c99dSWillem de Bruijn /* TCP can create new skb to attach new uarg */ 14104ab6c99dSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 14114ab6c99dSWillem de Bruijn goto new_alloc; 14124ab6c99dSWillem de Bruijn return NULL; 14134ab6c99dSWillem de Bruijn } 14144ab6c99dSWillem de Bruijn 14154ab6c99dSWillem de Bruijn next = (u32)atomic_read(&sk->sk_zckey); 1416e7d2b510SPavel Begunkov if ((u32)(uarg_zc->id + uarg_zc->len) == next) { 1417e7d2b510SPavel Begunkov if (mm_account_pinned_pages(&uarg_zc->mmp, size)) 1418a91dbff5SWillem de Bruijn return NULL; 1419e7d2b510SPavel Begunkov uarg_zc->len++; 1420e7d2b510SPavel Begunkov uarg_zc->bytelen = bytelen; 14214ab6c99dSWillem de Bruijn atomic_set(&sk->sk_zckey, ++next); 1422100f6d8eSWillem de Bruijn 1423100f6d8eSWillem de Bruijn /* no extra ref when appending to datagram (MSG_MORE) */ 1424100f6d8eSWillem de Bruijn if (sk->sk_type == SOCK_STREAM) 14258e044917SJonathan Lemon net_zcopy_get(uarg); 1426100f6d8eSWillem de Bruijn 14274ab6c99dSWillem de Bruijn return uarg; 14284ab6c99dSWillem de Bruijn } 14294ab6c99dSWillem de Bruijn } 14304ab6c99dSWillem de Bruijn 14314ab6c99dSWillem de Bruijn new_alloc: 14328c793822SJonathan Lemon return msg_zerocopy_alloc(sk, size); 14334ab6c99dSWillem de Bruijn } 14348c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_realloc); 14354ab6c99dSWillem de Bruijn 14364ab6c99dSWillem de Bruijn static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 14374ab6c99dSWillem de Bruijn { 14384ab6c99dSWillem de Bruijn struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 14394ab6c99dSWillem de Bruijn u32 old_lo, old_hi; 14404ab6c99dSWillem de Bruijn u64 sum_len; 14414ab6c99dSWillem de Bruijn 14424ab6c99dSWillem de Bruijn old_lo = serr->ee.ee_info; 14434ab6c99dSWillem de Bruijn old_hi = serr->ee.ee_data; 14444ab6c99dSWillem de Bruijn sum_len = old_hi - old_lo + 1ULL + len; 14454ab6c99dSWillem de Bruijn 14464ab6c99dSWillem de Bruijn if (sum_len >= (1ULL << 32)) 14474ab6c99dSWillem de Bruijn return false; 14484ab6c99dSWillem de Bruijn 14494ab6c99dSWillem de Bruijn if (lo != old_hi + 1) 14504ab6c99dSWillem de Bruijn return false; 14514ab6c99dSWillem de Bruijn 14524ab6c99dSWillem de Bruijn serr->ee.ee_data += len; 14534ab6c99dSWillem de Bruijn return true; 14544ab6c99dSWillem de Bruijn } 14554ab6c99dSWillem de Bruijn 1456e7d2b510SPavel Begunkov static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg) 145752267790SWillem de Bruijn { 14584ab6c99dSWillem de Bruijn struct sk_buff *tail, *skb = skb_from_uarg(uarg); 145952267790SWillem de Bruijn struct sock_exterr_skb *serr; 146052267790SWillem de Bruijn struct sock *sk = skb->sk; 14614ab6c99dSWillem de Bruijn struct sk_buff_head *q; 14624ab6c99dSWillem de Bruijn unsigned long flags; 14633bdd5ee0SWillem de Bruijn bool is_zerocopy; 14644ab6c99dSWillem de Bruijn u32 lo, hi; 14654ab6c99dSWillem de Bruijn u16 len; 146652267790SWillem de Bruijn 1467ccaffff1SWillem de Bruijn mm_unaccount_pinned_pages(&uarg->mmp); 1468ccaffff1SWillem de Bruijn 14694ab6c99dSWillem de Bruijn /* if !len, there was only 1 call, and it was aborted 14704ab6c99dSWillem de Bruijn * so do not queue a completion notification 14714ab6c99dSWillem de Bruijn */ 14724ab6c99dSWillem de Bruijn if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 147352267790SWillem de Bruijn goto release; 147452267790SWillem de Bruijn 14754ab6c99dSWillem de Bruijn len = uarg->len; 14764ab6c99dSWillem de Bruijn lo = uarg->id; 14774ab6c99dSWillem de Bruijn hi = uarg->id + len - 1; 14783bdd5ee0SWillem de Bruijn is_zerocopy = uarg->zerocopy; 14794ab6c99dSWillem de Bruijn 148052267790SWillem de Bruijn serr = SKB_EXT_ERR(skb); 148152267790SWillem de Bruijn memset(serr, 0, sizeof(*serr)); 148252267790SWillem de Bruijn serr->ee.ee_errno = 0; 148352267790SWillem de Bruijn serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 14844ab6c99dSWillem de Bruijn serr->ee.ee_data = hi; 14854ab6c99dSWillem de Bruijn serr->ee.ee_info = lo; 14863bdd5ee0SWillem de Bruijn if (!is_zerocopy) 148752267790SWillem de Bruijn serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 148852267790SWillem de Bruijn 14894ab6c99dSWillem de Bruijn q = &sk->sk_error_queue; 14904ab6c99dSWillem de Bruijn spin_lock_irqsave(&q->lock, flags); 14914ab6c99dSWillem de Bruijn tail = skb_peek_tail(q); 14924ab6c99dSWillem de Bruijn if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 14934ab6c99dSWillem de Bruijn !skb_zerocopy_notify_extend(tail, lo, len)) { 14944ab6c99dSWillem de Bruijn __skb_queue_tail(q, skb); 149552267790SWillem de Bruijn skb = NULL; 14964ab6c99dSWillem de Bruijn } 14974ab6c99dSWillem de Bruijn spin_unlock_irqrestore(&q->lock, flags); 149852267790SWillem de Bruijn 1499e3ae2365SAlexander Aring sk_error_report(sk); 150052267790SWillem de Bruijn 150152267790SWillem de Bruijn release: 150252267790SWillem de Bruijn consume_skb(skb); 150352267790SWillem de Bruijn sock_put(sk); 150452267790SWillem de Bruijn } 150575518851SJonathan Lemon 15068c793822SJonathan Lemon void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 150736177832SJonathan Lemon bool success) 150875518851SJonathan Lemon { 1509e7d2b510SPavel Begunkov struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg); 1510e7d2b510SPavel Begunkov 1511e7d2b510SPavel Begunkov uarg_zc->zerocopy = uarg_zc->zerocopy & success; 151275518851SJonathan Lemon 151375518851SJonathan Lemon if (refcount_dec_and_test(&uarg->refcnt)) 1514e7d2b510SPavel Begunkov __msg_zerocopy_callback(uarg_zc); 151575518851SJonathan Lemon } 15168c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_callback); 151752267790SWillem de Bruijn 15188c793822SJonathan Lemon void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) 151952267790SWillem de Bruijn { 1520e7d2b510SPavel Begunkov struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk; 152152267790SWillem de Bruijn 152252267790SWillem de Bruijn atomic_dec(&sk->sk_zckey); 1523e7d2b510SPavel Begunkov uarg_to_msgzc(uarg)->len--; 152452267790SWillem de Bruijn 152552900d22SWillem de Bruijn if (have_uref) 15268c793822SJonathan Lemon msg_zerocopy_callback(NULL, uarg, true); 152752267790SWillem de Bruijn } 15288c793822SJonathan Lemon EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort); 152952267790SWillem de Bruijn 153052267790SWillem de Bruijn int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 153152267790SWillem de Bruijn struct msghdr *msg, int len, 153252267790SWillem de Bruijn struct ubuf_info *uarg) 153352267790SWillem de Bruijn { 15344ab6c99dSWillem de Bruijn struct ubuf_info *orig_uarg = skb_zcopy(skb); 153552267790SWillem de Bruijn int err, orig_len = skb->len; 153652267790SWillem de Bruijn 15374ab6c99dSWillem de Bruijn /* An skb can only point to one uarg. This edge case happens when 15384ab6c99dSWillem de Bruijn * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 15394ab6c99dSWillem de Bruijn */ 15404ab6c99dSWillem de Bruijn if (orig_uarg && uarg != orig_uarg) 15414ab6c99dSWillem de Bruijn return -EEXIST; 15424ab6c99dSWillem de Bruijn 1543ebe73a28SDavid Ahern err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len); 154452267790SWillem de Bruijn if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 154554d43117SWillem de Bruijn struct sock *save_sk = skb->sk; 154654d43117SWillem de Bruijn 154752267790SWillem de Bruijn /* Streams do not free skb on error. Reset to prev state. */ 1548829b7bddSPavel Begunkov iov_iter_revert(&msg->msg_iter, skb->len - orig_len); 154954d43117SWillem de Bruijn skb->sk = sk; 155052267790SWillem de Bruijn ___pskb_trim(skb, orig_len); 155154d43117SWillem de Bruijn skb->sk = save_sk; 155252267790SWillem de Bruijn return err; 155352267790SWillem de Bruijn } 155452267790SWillem de Bruijn 155552900d22SWillem de Bruijn skb_zcopy_set(skb, uarg, NULL); 155652267790SWillem de Bruijn return skb->len - orig_len; 155752267790SWillem de Bruijn } 155852267790SWillem de Bruijn EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 155952267790SWillem de Bruijn 1560753f1ca4SPavel Begunkov void __skb_zcopy_downgrade_managed(struct sk_buff *skb) 1561753f1ca4SPavel Begunkov { 1562753f1ca4SPavel Begunkov int i; 1563753f1ca4SPavel Begunkov 1564753f1ca4SPavel Begunkov skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS; 1565753f1ca4SPavel Begunkov for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1566753f1ca4SPavel Begunkov skb_frag_ref(skb, i); 1567753f1ca4SPavel Begunkov } 1568753f1ca4SPavel Begunkov EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed); 1569753f1ca4SPavel Begunkov 15701f8b977aSWillem de Bruijn static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 157152267790SWillem de Bruijn gfp_t gfp_mask) 157252267790SWillem de Bruijn { 157352267790SWillem de Bruijn if (skb_zcopy(orig)) { 157452267790SWillem de Bruijn if (skb_zcopy(nskb)) { 157552267790SWillem de Bruijn /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 157652267790SWillem de Bruijn if (!gfp_mask) { 157752267790SWillem de Bruijn WARN_ON_ONCE(1); 157852267790SWillem de Bruijn return -ENOMEM; 157952267790SWillem de Bruijn } 158052267790SWillem de Bruijn if (skb_uarg(nskb) == skb_uarg(orig)) 158152267790SWillem de Bruijn return 0; 158252267790SWillem de Bruijn if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 158352267790SWillem de Bruijn return -EIO; 158452267790SWillem de Bruijn } 158552900d22SWillem de Bruijn skb_zcopy_set(nskb, skb_uarg(orig), NULL); 158652267790SWillem de Bruijn } 158752267790SWillem de Bruijn return 0; 158852267790SWillem de Bruijn } 158952267790SWillem de Bruijn 15902c53040fSBen Hutchings /** 15912c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 159248c83012SMichael S. Tsirkin * @skb: the skb to modify 159348c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 159448c83012SMichael S. Tsirkin * 159506b4feb3SJonathan Lemon * This must be called on skb with SKBFL_ZEROCOPY_ENABLE. 159648c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 159748c83012SMichael S. Tsirkin * to userspace pages. 159848c83012SMichael S. Tsirkin * 159948c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 160048c83012SMichael S. Tsirkin * %GFP_ATOMIC. 160148c83012SMichael S. Tsirkin * 160248c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 160348c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 160448c83012SMichael S. Tsirkin */ 160548c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1606a6686f2fSShirley Ma { 1607a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 1608a6686f2fSShirley Ma struct page *page, *head = NULL; 16093ece7826SWillem de Bruijn int i, new_frags; 16103ece7826SWillem de Bruijn u32 d_off; 1611a6686f2fSShirley Ma 16123ece7826SWillem de Bruijn if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 16133ece7826SWillem de Bruijn return -EINVAL; 16143ece7826SWillem de Bruijn 1615f72c4ac6SWillem de Bruijn if (!num_frags) 1616f72c4ac6SWillem de Bruijn goto release; 1617f72c4ac6SWillem de Bruijn 16183ece7826SWillem de Bruijn new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 16193ece7826SWillem de Bruijn for (i = 0; i < new_frags; i++) { 162002756ed4SKrishna Kumar page = alloc_page(gfp_mask); 1621a6686f2fSShirley Ma if (!page) { 1622a6686f2fSShirley Ma while (head) { 162340dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 1624a6686f2fSShirley Ma put_page(head); 1625a6686f2fSShirley Ma head = next; 1626a6686f2fSShirley Ma } 1627a6686f2fSShirley Ma return -ENOMEM; 1628a6686f2fSShirley Ma } 16293ece7826SWillem de Bruijn set_page_private(page, (unsigned long)head); 16303ece7826SWillem de Bruijn head = page; 16313ece7826SWillem de Bruijn } 16323ece7826SWillem de Bruijn 16333ece7826SWillem de Bruijn page = head; 16343ece7826SWillem de Bruijn d_off = 0; 16353ece7826SWillem de Bruijn for (i = 0; i < num_frags; i++) { 16363ece7826SWillem de Bruijn skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 16373ece7826SWillem de Bruijn u32 p_off, p_len, copied; 16383ece7826SWillem de Bruijn struct page *p; 16393ece7826SWillem de Bruijn u8 *vaddr; 1640c613c209SWillem de Bruijn 1641b54c9d5bSJonathan Lemon skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f), 1642c613c209SWillem de Bruijn p, p_off, p_len, copied) { 16433ece7826SWillem de Bruijn u32 copy, done = 0; 1644c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 16453ece7826SWillem de Bruijn 16463ece7826SWillem de Bruijn while (done < p_len) { 16473ece7826SWillem de Bruijn if (d_off == PAGE_SIZE) { 16483ece7826SWillem de Bruijn d_off = 0; 16493ece7826SWillem de Bruijn page = (struct page *)page_private(page); 16503ece7826SWillem de Bruijn } 16513ece7826SWillem de Bruijn copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 16523ece7826SWillem de Bruijn memcpy(page_address(page) + d_off, 16533ece7826SWillem de Bruijn vaddr + p_off + done, copy); 16543ece7826SWillem de Bruijn done += copy; 16553ece7826SWillem de Bruijn d_off += copy; 16563ece7826SWillem de Bruijn } 165751c56b00SEric Dumazet kunmap_atomic(vaddr); 1658c613c209SWillem de Bruijn } 1659a6686f2fSShirley Ma } 1660a6686f2fSShirley Ma 1661a6686f2fSShirley Ma /* skb frags release userspace buffers */ 166202756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 1663a8605c60SIan Campbell skb_frag_unref(skb, i); 1664a6686f2fSShirley Ma 1665a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 16663ece7826SWillem de Bruijn for (i = 0; i < new_frags - 1; i++) { 16673ece7826SWillem de Bruijn __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 166840dadff2SSunghan Suh head = (struct page *)page_private(head); 1669a6686f2fSShirley Ma } 16703ece7826SWillem de Bruijn __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 16713ece7826SWillem de Bruijn skb_shinfo(skb)->nr_frags = new_frags; 167248c83012SMichael S. Tsirkin 1673b90ddd56SWillem de Bruijn release: 16741f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 1675a6686f2fSShirley Ma return 0; 1676a6686f2fSShirley Ma } 1677dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1678a6686f2fSShirley Ma 1679e0053ec0SHerbert Xu /** 1680e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 1681e0053ec0SHerbert Xu * @skb: buffer to clone 1682e0053ec0SHerbert Xu * @gfp_mask: allocation priority 1683e0053ec0SHerbert Xu * 1684e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1685e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 1686e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 1687e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 1688e0053ec0SHerbert Xu * 1689e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 1690e0053ec0SHerbert Xu * %GFP_ATOMIC. 1691e0053ec0SHerbert Xu */ 1692e0053ec0SHerbert Xu 1693e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1694e0053ec0SHerbert Xu { 1695d0bf4a9eSEric Dumazet struct sk_buff_fclones *fclones = container_of(skb, 1696d0bf4a9eSEric Dumazet struct sk_buff_fclones, 1697d0bf4a9eSEric Dumazet skb1); 16986ffe75ebSEric Dumazet struct sk_buff *n; 1699e0053ec0SHerbert Xu 170070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1701a6686f2fSShirley Ma return NULL; 1702a6686f2fSShirley Ma 1703e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 17042638595aSReshetova, Elena refcount_read(&fclones->fclone_ref) == 1) { 17056ffe75ebSEric Dumazet n = &fclones->skb2; 17062638595aSReshetova, Elena refcount_set(&fclones->fclone_ref, 2); 1707c2dd4059SEric Dumazet n->fclone = SKB_FCLONE_CLONE; 1708e0053ec0SHerbert Xu } else { 1709c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1710c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1711c93bdd0eSMel Gorman 1712e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1713e0053ec0SHerbert Xu if (!n) 1714e0053ec0SHerbert Xu return NULL; 1715fe55f6d5SVegard Nossum 1716e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 1717e0053ec0SHerbert Xu } 1718e0053ec0SHerbert Xu 1719e0053ec0SHerbert Xu return __skb_clone(n, skb); 17201da177e4SLinus Torvalds } 1721b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 17221da177e4SLinus Torvalds 1723b0768a86SToshiaki Makita void skb_headers_offset_update(struct sk_buff *skb, int off) 1724f5b17294SPravin B Shelar { 1725030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 1726030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 1727030737bcSEric Dumazet skb->csum_start += off; 1728f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 1729f5b17294SPravin B Shelar skb->transport_header += off; 1730f5b17294SPravin B Shelar skb->network_header += off; 1731f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 1732f5b17294SPravin B Shelar skb->mac_header += off; 1733f5b17294SPravin B Shelar skb->inner_transport_header += off; 1734f5b17294SPravin B Shelar skb->inner_network_header += off; 1735aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 1736f5b17294SPravin B Shelar } 1737b0768a86SToshiaki Makita EXPORT_SYMBOL(skb_headers_offset_update); 1738f5b17294SPravin B Shelar 173908303c18SIlya Lesokhin void skb_copy_header(struct sk_buff *new, const struct sk_buff *old) 17401da177e4SLinus Torvalds { 1741dec18810SHerbert Xu __copy_skb_header(new, old); 1742dec18810SHerbert Xu 17437967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 17447967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 17457967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 17461da177e4SLinus Torvalds } 174708303c18SIlya Lesokhin EXPORT_SYMBOL(skb_copy_header); 17481da177e4SLinus Torvalds 1749c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1750c93bdd0eSMel Gorman { 1751c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1752c93bdd0eSMel Gorman return SKB_ALLOC_RX; 1753c93bdd0eSMel Gorman return 0; 1754c93bdd0eSMel Gorman } 1755c93bdd0eSMel Gorman 17561da177e4SLinus Torvalds /** 17571da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 17581da177e4SLinus Torvalds * @skb: buffer to copy 17591da177e4SLinus Torvalds * @gfp_mask: allocation priority 17601da177e4SLinus Torvalds * 17611da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 17621da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 17631da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 17641da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 17651da177e4SLinus Torvalds * 17661da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 17671da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 17681da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 17691da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 17701da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 17711da177e4SLinus Torvalds */ 17721da177e4SLinus Torvalds 1773dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 17741da177e4SLinus Torvalds { 17756602cebbSEric Dumazet int headerlen = skb_headroom(skb); 1776ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 1777c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1778c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 17796602cebbSEric Dumazet 17801da177e4SLinus Torvalds if (!n) 17811da177e4SLinus Torvalds return NULL; 17821da177e4SLinus Torvalds 17831da177e4SLinus Torvalds /* Set the data pointer */ 17841da177e4SLinus Torvalds skb_reserve(n, headerlen); 17851da177e4SLinus Torvalds /* Set the tail pointer and length */ 17861da177e4SLinus Torvalds skb_put(n, skb->len); 17871da177e4SLinus Torvalds 17889f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); 17891da177e4SLinus Torvalds 179008303c18SIlya Lesokhin skb_copy_header(n, skb); 17911da177e4SLinus Torvalds return n; 17921da177e4SLinus Torvalds } 1793b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 17941da177e4SLinus Torvalds 17951da177e4SLinus Torvalds /** 1796bad93e9dSOctavian Purdila * __pskb_copy_fclone - create copy of an sk_buff with private head. 17971da177e4SLinus Torvalds * @skb: buffer to copy 1798117632e6SEric Dumazet * @headroom: headroom of new skb 17991da177e4SLinus Torvalds * @gfp_mask: allocation priority 1800bad93e9dSOctavian Purdila * @fclone: if true allocate the copy of the skb from the fclone 1801bad93e9dSOctavian Purdila * cache instead of the head cache; it is recommended to set this 1802bad93e9dSOctavian Purdila * to true for the cases where the copy will likely be cloned 18031da177e4SLinus Torvalds * 18041da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 18051da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 18061da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 18071da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 18081da177e4SLinus Torvalds * or the pointer to the buffer on success. 18091da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 18101da177e4SLinus Torvalds */ 18111da177e4SLinus Torvalds 1812bad93e9dSOctavian Purdila struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1813bad93e9dSOctavian Purdila gfp_t gfp_mask, bool fclone) 18141da177e4SLinus Torvalds { 1815117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1816bad93e9dSOctavian Purdila int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1817bad93e9dSOctavian Purdila struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 18186602cebbSEric Dumazet 18191da177e4SLinus Torvalds if (!n) 18201da177e4SLinus Torvalds goto out; 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds /* Set the data pointer */ 1823117632e6SEric Dumazet skb_reserve(n, headroom); 18241da177e4SLinus Torvalds /* Set the tail pointer and length */ 18251da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 18261da177e4SLinus Torvalds /* Copy the bytes */ 1827d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 18281da177e4SLinus Torvalds 182925f484a6SHerbert Xu n->truesize += skb->data_len; 18301da177e4SLinus Torvalds n->data_len = skb->data_len; 18311da177e4SLinus Torvalds n->len = skb->len; 18321da177e4SLinus Torvalds 18331da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 18341da177e4SLinus Torvalds int i; 18351da177e4SLinus Torvalds 18361f8b977aSWillem de Bruijn if (skb_orphan_frags(skb, gfp_mask) || 18371f8b977aSWillem de Bruijn skb_zerocopy_clone(n, skb, gfp_mask)) { 18381511022cSDan Carpenter kfree_skb(n); 18391511022cSDan Carpenter n = NULL; 1840a6686f2fSShirley Ma goto out; 1841a6686f2fSShirley Ma } 18421da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 18431da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1844ea2ab693SIan Campbell skb_frag_ref(skb, i); 18451da177e4SLinus Torvalds } 18461da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 18471da177e4SLinus Torvalds } 18481da177e4SLinus Torvalds 184921dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 18501da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 18511da177e4SLinus Torvalds skb_clone_fraglist(n); 18521da177e4SLinus Torvalds } 18531da177e4SLinus Torvalds 185408303c18SIlya Lesokhin skb_copy_header(n, skb); 18551da177e4SLinus Torvalds out: 18561da177e4SLinus Torvalds return n; 18571da177e4SLinus Torvalds } 1858bad93e9dSOctavian Purdila EXPORT_SYMBOL(__pskb_copy_fclone); 18591da177e4SLinus Torvalds 18601da177e4SLinus Torvalds /** 18611da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 18621da177e4SLinus Torvalds * @skb: buffer to reallocate 18631da177e4SLinus Torvalds * @nhead: room to add at head 18641da177e4SLinus Torvalds * @ntail: room to add at tail 18651da177e4SLinus Torvalds * @gfp_mask: allocation priority 18661da177e4SLinus Torvalds * 1867bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1868bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 18691da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 18701da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 18711da177e4SLinus Torvalds * 18721da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 18731da177e4SLinus Torvalds * reloaded after call to this function. 18741da177e4SLinus Torvalds */ 18751da177e4SLinus Torvalds 187686a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1877dd0fc66fSAl Viro gfp_t gfp_mask) 18781da177e4SLinus Torvalds { 187912d6c1d3SKees Cook unsigned int osize = skb_end_offset(skb); 188012d6c1d3SKees Cook unsigned int size = osize + nhead + ntail; 18811da177e4SLinus Torvalds long off; 1882158f323bSEric Dumazet u8 *data; 188312d6c1d3SKees Cook int i; 18841da177e4SLinus Torvalds 18854edd87adSHerbert Xu BUG_ON(nhead < 0); 18864edd87adSHerbert Xu 18879f77fad3STim Hansen BUG_ON(skb_shared(skb)); 18881da177e4SLinus Torvalds 1889753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 1890753f1ca4SPavel Begunkov 1891c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1892c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 189312d6c1d3SKees Cook 189412d6c1d3SKees Cook size = SKB_DATA_ALIGN(size); 189512d6c1d3SKees Cook size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 189612d6c1d3SKees Cook size = kmalloc_size_roundup(size); 189712d6c1d3SKees Cook data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL); 18981da177e4SLinus Torvalds if (!data) 18991da177e4SLinus Torvalds goto nodata; 190012d6c1d3SKees Cook size = SKB_WITH_OVERHEAD(size); 19011da177e4SLinus Torvalds 19021da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 19036602cebbSEric Dumazet * optimized for the cases when header is void. 19046602cebbSEric Dumazet */ 19056602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 19066602cebbSEric Dumazet 19076602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 19086602cebbSEric Dumazet skb_shinfo(skb), 1909fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 19101da177e4SLinus Torvalds 19113e24591aSAlexander Duyck /* 19123e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 19133e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 19143e24591aSAlexander Duyck * be since all we did is relocate the values 19153e24591aSAlexander Duyck */ 19163e24591aSAlexander Duyck if (skb_cloned(skb)) { 191770008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1918a6686f2fSShirley Ma goto nofrags; 19191f8b977aSWillem de Bruijn if (skb_zcopy(skb)) 1920c1d1b437SEric Dumazet refcount_inc(&skb_uarg(skb)->refcnt); 19211da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1922ea2ab693SIan Campbell skb_frag_ref(skb, i); 19231da177e4SLinus Torvalds 192421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 19251da177e4SLinus Torvalds skb_clone_fraglist(skb); 19261da177e4SLinus Torvalds 1927511a3edaSEric Dumazet skb_release_data(skb, SKB_CONSUMED); 19283e24591aSAlexander Duyck } else { 19293e24591aSAlexander Duyck skb_free_head(skb); 19301fd63041SEric Dumazet } 19311da177e4SLinus Torvalds off = (data + nhead) - skb->head; 19321da177e4SLinus Torvalds 19331da177e4SLinus Torvalds skb->head = data; 1934d3836f21SEric Dumazet skb->head_frag = 0; 19351da177e4SLinus Torvalds skb->data += off; 1936763087daSEric Dumazet 1937763087daSEric Dumazet skb_set_end_offset(skb, size); 19384305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 193956eb8882SPatrick McHardy off = nhead; 194056eb8882SPatrick McHardy #endif 194127a884dcSArnaldo Carvalho de Melo skb->tail += off; 1942b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 19431da177e4SLinus Torvalds skb->cloned = 0; 1944334a8132SPatrick McHardy skb->hdr_len = 0; 19451da177e4SLinus Torvalds skb->nohdr = 0; 19461da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 1947158f323bSEric Dumazet 1948de8f3a83SDaniel Borkmann skb_metadata_clear(skb); 1949de8f3a83SDaniel Borkmann 1950158f323bSEric Dumazet /* It is not generally safe to change skb->truesize. 1951158f323bSEric Dumazet * For the moment, we really care of rx path, or 1952158f323bSEric Dumazet * when skb is orphaned (not attached to a socket). 1953158f323bSEric Dumazet */ 1954158f323bSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 1955158f323bSEric Dumazet skb->truesize += size - osize; 1956158f323bSEric Dumazet 19571da177e4SLinus Torvalds return 0; 19581da177e4SLinus Torvalds 1959a6686f2fSShirley Ma nofrags: 1960a6686f2fSShirley Ma kfree(data); 19611da177e4SLinus Torvalds nodata: 19621da177e4SLinus Torvalds return -ENOMEM; 19631da177e4SLinus Torvalds } 1964b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 19651da177e4SLinus Torvalds 19661da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 19671da177e4SLinus Torvalds 19681da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 19691da177e4SLinus Torvalds { 19701da177e4SLinus Torvalds struct sk_buff *skb2; 19711da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 19721da177e4SLinus Torvalds 19731da177e4SLinus Torvalds if (delta <= 0) 19741da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 19751da177e4SLinus Torvalds else { 19761da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 19771da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 19781da177e4SLinus Torvalds GFP_ATOMIC)) { 19791da177e4SLinus Torvalds kfree_skb(skb2); 19801da177e4SLinus Torvalds skb2 = NULL; 19811da177e4SLinus Torvalds } 19821da177e4SLinus Torvalds } 19831da177e4SLinus Torvalds return skb2; 19841da177e4SLinus Torvalds } 1985b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 19861da177e4SLinus Torvalds 19872b88cba5SEric Dumazet int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 19882b88cba5SEric Dumazet { 19892b88cba5SEric Dumazet unsigned int saved_end_offset, saved_truesize; 19902b88cba5SEric Dumazet struct skb_shared_info *shinfo; 19912b88cba5SEric Dumazet int res; 19922b88cba5SEric Dumazet 19932b88cba5SEric Dumazet saved_end_offset = skb_end_offset(skb); 19942b88cba5SEric Dumazet saved_truesize = skb->truesize; 19952b88cba5SEric Dumazet 19962b88cba5SEric Dumazet res = pskb_expand_head(skb, 0, 0, pri); 19972b88cba5SEric Dumazet if (res) 19982b88cba5SEric Dumazet return res; 19992b88cba5SEric Dumazet 20002b88cba5SEric Dumazet skb->truesize = saved_truesize; 20012b88cba5SEric Dumazet 20022b88cba5SEric Dumazet if (likely(skb_end_offset(skb) == saved_end_offset)) 20032b88cba5SEric Dumazet return 0; 20042b88cba5SEric Dumazet 20052b88cba5SEric Dumazet shinfo = skb_shinfo(skb); 20062b88cba5SEric Dumazet 20072b88cba5SEric Dumazet /* We are about to change back skb->end, 20082b88cba5SEric Dumazet * we need to move skb_shinfo() to its new location. 20092b88cba5SEric Dumazet */ 20102b88cba5SEric Dumazet memmove(skb->head + saved_end_offset, 20112b88cba5SEric Dumazet shinfo, 20122b88cba5SEric Dumazet offsetof(struct skb_shared_info, frags[shinfo->nr_frags])); 20132b88cba5SEric Dumazet 20142b88cba5SEric Dumazet skb_set_end_offset(skb, saved_end_offset); 20152b88cba5SEric Dumazet 20162b88cba5SEric Dumazet return 0; 20172b88cba5SEric Dumazet } 20182b88cba5SEric Dumazet 20191da177e4SLinus Torvalds /** 2020f1260ff1SVasily Averin * skb_expand_head - reallocate header of &sk_buff 2021f1260ff1SVasily Averin * @skb: buffer to reallocate 2022f1260ff1SVasily Averin * @headroom: needed headroom 2023f1260ff1SVasily Averin * 2024f1260ff1SVasily Averin * Unlike skb_realloc_headroom, this one does not allocate a new skb 2025f1260ff1SVasily Averin * if possible; copies skb->sk to new skb as needed 2026f1260ff1SVasily Averin * and frees original skb in case of failures. 2027f1260ff1SVasily Averin * 2028f1260ff1SVasily Averin * It expect increased headroom and generates warning otherwise. 2029f1260ff1SVasily Averin */ 2030f1260ff1SVasily Averin 2031f1260ff1SVasily Averin struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom) 2032f1260ff1SVasily Averin { 2033f1260ff1SVasily Averin int delta = headroom - skb_headroom(skb); 20347f678defSVasily Averin int osize = skb_end_offset(skb); 20357f678defSVasily Averin struct sock *sk = skb->sk; 2036f1260ff1SVasily Averin 2037f1260ff1SVasily Averin if (WARN_ONCE(delta <= 0, 2038f1260ff1SVasily Averin "%s is expecting an increase in the headroom", __func__)) 2039f1260ff1SVasily Averin return skb; 2040f1260ff1SVasily Averin 20417f678defSVasily Averin delta = SKB_DATA_ALIGN(delta); 20427f678defSVasily Averin /* pskb_expand_head() might crash, if skb is shared. */ 20437f678defSVasily Averin if (skb_shared(skb) || !is_skb_wmem(skb)) { 2044f1260ff1SVasily Averin struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 2045f1260ff1SVasily Averin 20467f678defSVasily Averin if (unlikely(!nskb)) 20477f678defSVasily Averin goto fail; 20487f678defSVasily Averin 20497f678defSVasily Averin if (sk) 20507f678defSVasily Averin skb_set_owner_w(nskb, sk); 2051f1260ff1SVasily Averin consume_skb(skb); 2052f1260ff1SVasily Averin skb = nskb; 2053f1260ff1SVasily Averin } 20547f678defSVasily Averin if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC)) 20557f678defSVasily Averin goto fail; 20567f678defSVasily Averin 20577f678defSVasily Averin if (sk && is_skb_wmem(skb)) { 20587f678defSVasily Averin delta = skb_end_offset(skb) - osize; 20597f678defSVasily Averin refcount_add(delta, &sk->sk_wmem_alloc); 20607f678defSVasily Averin skb->truesize += delta; 2061f1260ff1SVasily Averin } 2062f1260ff1SVasily Averin return skb; 20637f678defSVasily Averin 20647f678defSVasily Averin fail: 20657f678defSVasily Averin kfree_skb(skb); 20667f678defSVasily Averin return NULL; 2067f1260ff1SVasily Averin } 2068f1260ff1SVasily Averin EXPORT_SYMBOL(skb_expand_head); 2069f1260ff1SVasily Averin 2070f1260ff1SVasily Averin /** 20711da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 20721da177e4SLinus Torvalds * @skb: buffer to copy 20731da177e4SLinus Torvalds * @newheadroom: new free bytes at head 20741da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 20751da177e4SLinus Torvalds * @gfp_mask: allocation priority 20761da177e4SLinus Torvalds * 20771da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 20781da177e4SLinus Torvalds * allocate additional space. 20791da177e4SLinus Torvalds * 20801da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 20811da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 20821da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 20831da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 20841da177e4SLinus Torvalds * 20851da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 20861da177e4SLinus Torvalds * is called from an interrupt. 20871da177e4SLinus Torvalds */ 20881da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 208986a76cafSVictor Fusco int newheadroom, int newtailroom, 2090dd0fc66fSAl Viro gfp_t gfp_mask) 20911da177e4SLinus Torvalds { 20921da177e4SLinus Torvalds /* 20931da177e4SLinus Torvalds * Allocate the copy buffer 20941da177e4SLinus Torvalds */ 2095c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 2096c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 2097c93bdd0eSMel Gorman NUMA_NO_NODE); 2098efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 20991da177e4SLinus Torvalds int head_copy_len, head_copy_off; 21001da177e4SLinus Torvalds 21011da177e4SLinus Torvalds if (!n) 21021da177e4SLinus Torvalds return NULL; 21031da177e4SLinus Torvalds 21041da177e4SLinus Torvalds skb_reserve(n, newheadroom); 21051da177e4SLinus Torvalds 21061da177e4SLinus Torvalds /* Set the tail pointer and length */ 21071da177e4SLinus Torvalds skb_put(n, skb->len); 21081da177e4SLinus Torvalds 2109efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 21101da177e4SLinus Torvalds head_copy_off = 0; 21111da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 21121da177e4SLinus Torvalds head_copy_len = newheadroom; 21131da177e4SLinus Torvalds else 21141da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 21151da177e4SLinus Torvalds 21161da177e4SLinus Torvalds /* Copy the linear header and data. */ 21179f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 21189f77fad3STim Hansen skb->len + head_copy_len)); 21191da177e4SLinus Torvalds 212008303c18SIlya Lesokhin skb_copy_header(n, skb); 21211da177e4SLinus Torvalds 2122030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 2123efd1e8d5SPatrick McHardy 21241da177e4SLinus Torvalds return n; 21251da177e4SLinus Torvalds } 2126b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 21271da177e4SLinus Torvalds 21281da177e4SLinus Torvalds /** 2129cd0a137aSFlorian Fainelli * __skb_pad - zero pad the tail of an skb 21301da177e4SLinus Torvalds * @skb: buffer to pad 21311da177e4SLinus Torvalds * @pad: space to pad 2132cd0a137aSFlorian Fainelli * @free_on_error: free buffer on error 21331da177e4SLinus Torvalds * 21341da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 21351da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 21361da177e4SLinus Torvalds * beyond the buffer end onto the wire. 21371da177e4SLinus Torvalds * 2138cd0a137aSFlorian Fainelli * May return error in out of memory cases. The skb is freed on error 2139cd0a137aSFlorian Fainelli * if @free_on_error is true. 21401da177e4SLinus Torvalds */ 21411da177e4SLinus Torvalds 2142cd0a137aSFlorian Fainelli int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 21431da177e4SLinus Torvalds { 21445b057c6bSHerbert Xu int err; 21455b057c6bSHerbert Xu int ntail; 21461da177e4SLinus Torvalds 21471da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 21485b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 21491da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 21505b057c6bSHerbert Xu return 0; 21511da177e4SLinus Torvalds } 21521da177e4SLinus Torvalds 21534305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 21545b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 21555b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 21565b057c6bSHerbert Xu if (unlikely(err)) 21575b057c6bSHerbert Xu goto free_skb; 21585b057c6bSHerbert Xu } 21595b057c6bSHerbert Xu 21605b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 21615b057c6bSHerbert Xu * to be audited. 21625b057c6bSHerbert Xu */ 21635b057c6bSHerbert Xu err = skb_linearize(skb); 21645b057c6bSHerbert Xu if (unlikely(err)) 21655b057c6bSHerbert Xu goto free_skb; 21665b057c6bSHerbert Xu 21675b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 21685b057c6bSHerbert Xu return 0; 21695b057c6bSHerbert Xu 21705b057c6bSHerbert Xu free_skb: 2171cd0a137aSFlorian Fainelli if (free_on_error) 21721da177e4SLinus Torvalds kfree_skb(skb); 21735b057c6bSHerbert Xu return err; 21741da177e4SLinus Torvalds } 2175cd0a137aSFlorian Fainelli EXPORT_SYMBOL(__skb_pad); 21761da177e4SLinus Torvalds 21770dde3e16SIlpo Järvinen /** 21780c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 21790c7ddf36SMathias Krause * @skb: start of the buffer to use 21800c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 21810c7ddf36SMathias Krause * @len: amount of data to add 21820c7ddf36SMathias Krause * 21830c7ddf36SMathias Krause * This function extends the used data area of the potentially 21840c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 21850c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 21860c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 21870c7ddf36SMathias Krause * returned. 21880c7ddf36SMathias Krause */ 21890c7ddf36SMathias Krause 21904df864c1SJohannes Berg void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 21910c7ddf36SMathias Krause { 21920c7ddf36SMathias Krause if (tail != skb) { 21930c7ddf36SMathias Krause skb->data_len += len; 21940c7ddf36SMathias Krause skb->len += len; 21950c7ddf36SMathias Krause } 21960c7ddf36SMathias Krause return skb_put(tail, len); 21970c7ddf36SMathias Krause } 21980c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 21990c7ddf36SMathias Krause 22000c7ddf36SMathias Krause /** 22010dde3e16SIlpo Järvinen * skb_put - add data to a buffer 22020dde3e16SIlpo Järvinen * @skb: buffer to use 22030dde3e16SIlpo Järvinen * @len: amount of data to add 22040dde3e16SIlpo Järvinen * 22050dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 22060dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 22070dde3e16SIlpo Järvinen * first byte of the extra data is returned. 22080dde3e16SIlpo Järvinen */ 22094df864c1SJohannes Berg void *skb_put(struct sk_buff *skb, unsigned int len) 22100dde3e16SIlpo Järvinen { 22114df864c1SJohannes Berg void *tmp = skb_tail_pointer(skb); 22120dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 22130dde3e16SIlpo Järvinen skb->tail += len; 22140dde3e16SIlpo Järvinen skb->len += len; 22150dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 22160dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 22170dde3e16SIlpo Järvinen return tmp; 22180dde3e16SIlpo Järvinen } 22190dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 22200dde3e16SIlpo Järvinen 22216be8ac2fSIlpo Järvinen /** 2222c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 2223c2aa270aSIlpo Järvinen * @skb: buffer to use 2224c2aa270aSIlpo Järvinen * @len: amount of data to add 2225c2aa270aSIlpo Järvinen * 2226c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 2227c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 2228c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 2229c2aa270aSIlpo Järvinen */ 2230d58ff351SJohannes Berg void *skb_push(struct sk_buff *skb, unsigned int len) 2231c2aa270aSIlpo Järvinen { 2232c2aa270aSIlpo Järvinen skb->data -= len; 2233c2aa270aSIlpo Järvinen skb->len += len; 2234c2aa270aSIlpo Järvinen if (unlikely(skb->data < skb->head)) 2235c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 2236c2aa270aSIlpo Järvinen return skb->data; 2237c2aa270aSIlpo Järvinen } 2238c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 2239c2aa270aSIlpo Järvinen 2240c2aa270aSIlpo Järvinen /** 22416be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 22426be8ac2fSIlpo Järvinen * @skb: buffer to use 22436be8ac2fSIlpo Järvinen * @len: amount of data to remove 22446be8ac2fSIlpo Järvinen * 22456be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 22466be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 22476be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 22486be8ac2fSIlpo Järvinen * the old data. 22496be8ac2fSIlpo Järvinen */ 2250af72868bSJohannes Berg void *skb_pull(struct sk_buff *skb, unsigned int len) 22516be8ac2fSIlpo Järvinen { 225247d29646SDavid S. Miller return skb_pull_inline(skb, len); 22536be8ac2fSIlpo Järvinen } 22546be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 22556be8ac2fSIlpo Järvinen 2256419ae74eSIlpo Järvinen /** 225713244cccSLuiz Augusto von Dentz * skb_pull_data - remove data from the start of a buffer returning its 225813244cccSLuiz Augusto von Dentz * original position. 225913244cccSLuiz Augusto von Dentz * @skb: buffer to use 226013244cccSLuiz Augusto von Dentz * @len: amount of data to remove 226113244cccSLuiz Augusto von Dentz * 226213244cccSLuiz Augusto von Dentz * This function removes data from the start of a buffer, returning 226313244cccSLuiz Augusto von Dentz * the memory to the headroom. A pointer to the original data in the buffer 226413244cccSLuiz Augusto von Dentz * is returned after checking if there is enough data to pull. Once the 226513244cccSLuiz Augusto von Dentz * data has been pulled future pushes will overwrite the old data. 226613244cccSLuiz Augusto von Dentz */ 226713244cccSLuiz Augusto von Dentz void *skb_pull_data(struct sk_buff *skb, size_t len) 226813244cccSLuiz Augusto von Dentz { 226913244cccSLuiz Augusto von Dentz void *data = skb->data; 227013244cccSLuiz Augusto von Dentz 227113244cccSLuiz Augusto von Dentz if (skb->len < len) 227213244cccSLuiz Augusto von Dentz return NULL; 227313244cccSLuiz Augusto von Dentz 227413244cccSLuiz Augusto von Dentz skb_pull(skb, len); 227513244cccSLuiz Augusto von Dentz 227613244cccSLuiz Augusto von Dentz return data; 227713244cccSLuiz Augusto von Dentz } 227813244cccSLuiz Augusto von Dentz EXPORT_SYMBOL(skb_pull_data); 227913244cccSLuiz Augusto von Dentz 228013244cccSLuiz Augusto von Dentz /** 2281419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 2282419ae74eSIlpo Järvinen * @skb: buffer to alter 2283419ae74eSIlpo Järvinen * @len: new length 2284419ae74eSIlpo Järvinen * 2285419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 2286419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 2287419ae74eSIlpo Järvinen * The skb must be linear. 2288419ae74eSIlpo Järvinen */ 2289419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 2290419ae74eSIlpo Järvinen { 2291419ae74eSIlpo Järvinen if (skb->len > len) 2292419ae74eSIlpo Järvinen __skb_trim(skb, len); 2293419ae74eSIlpo Järvinen } 2294419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 2295419ae74eSIlpo Järvinen 22963cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 22971da177e4SLinus Torvalds */ 22981da177e4SLinus Torvalds 22993cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 23001da177e4SLinus Torvalds { 230127b437c8SHerbert Xu struct sk_buff **fragp; 230227b437c8SHerbert Xu struct sk_buff *frag; 23031da177e4SLinus Torvalds int offset = skb_headlen(skb); 23041da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 23051da177e4SLinus Torvalds int i; 230627b437c8SHerbert Xu int err; 230727b437c8SHerbert Xu 230827b437c8SHerbert Xu if (skb_cloned(skb) && 230927b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 231027b437c8SHerbert Xu return err; 23111da177e4SLinus Torvalds 2312f4d26fb3SHerbert Xu i = 0; 2313f4d26fb3SHerbert Xu if (offset >= len) 2314f4d26fb3SHerbert Xu goto drop_pages; 2315f4d26fb3SHerbert Xu 2316f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 23179e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 231827b437c8SHerbert Xu 231927b437c8SHerbert Xu if (end < len) { 23201da177e4SLinus Torvalds offset = end; 232127b437c8SHerbert Xu continue; 23221da177e4SLinus Torvalds } 23231da177e4SLinus Torvalds 23249e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 232527b437c8SHerbert Xu 2326f4d26fb3SHerbert Xu drop_pages: 232727b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 232827b437c8SHerbert Xu 232927b437c8SHerbert Xu for (; i < nfrags; i++) 2330ea2ab693SIan Campbell skb_frag_unref(skb, i); 233127b437c8SHerbert Xu 233221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 233327b437c8SHerbert Xu skb_drop_fraglist(skb); 2334f4d26fb3SHerbert Xu goto done; 233527b437c8SHerbert Xu } 233627b437c8SHerbert Xu 233727b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 233827b437c8SHerbert Xu fragp = &frag->next) { 233927b437c8SHerbert Xu int end = offset + frag->len; 234027b437c8SHerbert Xu 234127b437c8SHerbert Xu if (skb_shared(frag)) { 234227b437c8SHerbert Xu struct sk_buff *nfrag; 234327b437c8SHerbert Xu 234427b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 234527b437c8SHerbert Xu if (unlikely(!nfrag)) 234627b437c8SHerbert Xu return -ENOMEM; 234727b437c8SHerbert Xu 234827b437c8SHerbert Xu nfrag->next = frag->next; 234985bb2a60SEric Dumazet consume_skb(frag); 235027b437c8SHerbert Xu frag = nfrag; 235127b437c8SHerbert Xu *fragp = frag; 235227b437c8SHerbert Xu } 235327b437c8SHerbert Xu 235427b437c8SHerbert Xu if (end < len) { 235527b437c8SHerbert Xu offset = end; 235627b437c8SHerbert Xu continue; 235727b437c8SHerbert Xu } 235827b437c8SHerbert Xu 235927b437c8SHerbert Xu if (end > len && 236027b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 236127b437c8SHerbert Xu return err; 236227b437c8SHerbert Xu 236327b437c8SHerbert Xu if (frag->next) 236427b437c8SHerbert Xu skb_drop_list(&frag->next); 236527b437c8SHerbert Xu break; 236627b437c8SHerbert Xu } 236727b437c8SHerbert Xu 2368f4d26fb3SHerbert Xu done: 236927b437c8SHerbert Xu if (len > skb_headlen(skb)) { 23701da177e4SLinus Torvalds skb->data_len -= skb->len - len; 23711da177e4SLinus Torvalds skb->len = len; 23721da177e4SLinus Torvalds } else { 23731da177e4SLinus Torvalds skb->len = len; 23741da177e4SLinus Torvalds skb->data_len = 0; 237527a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 23761da177e4SLinus Torvalds } 23771da177e4SLinus Torvalds 2378c21b48ccSEric Dumazet if (!skb->sk || skb->destructor == sock_edemux) 2379c21b48ccSEric Dumazet skb_condense(skb); 23801da177e4SLinus Torvalds return 0; 23811da177e4SLinus Torvalds } 2382b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 23831da177e4SLinus Torvalds 238488078d98SEric Dumazet /* Note : use pskb_trim_rcsum() instead of calling this directly 238588078d98SEric Dumazet */ 238688078d98SEric Dumazet int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len) 238788078d98SEric Dumazet { 238888078d98SEric Dumazet if (skb->ip_summed == CHECKSUM_COMPLETE) { 238988078d98SEric Dumazet int delta = skb->len - len; 239088078d98SEric Dumazet 2391d55bef50SDimitris Michailidis skb->csum = csum_block_sub(skb->csum, 2392d55bef50SDimitris Michailidis skb_checksum(skb, len, delta, 0), 2393d55bef50SDimitris Michailidis len); 239454970a2fSVasily Averin } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 239554970a2fSVasily Averin int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len; 239654970a2fSVasily Averin int offset = skb_checksum_start_offset(skb) + skb->csum_offset; 239754970a2fSVasily Averin 239854970a2fSVasily Averin if (offset + sizeof(__sum16) > hdlen) 239954970a2fSVasily Averin return -EINVAL; 240088078d98SEric Dumazet } 240188078d98SEric Dumazet return __pskb_trim(skb, len); 240288078d98SEric Dumazet } 240388078d98SEric Dumazet EXPORT_SYMBOL(pskb_trim_rcsum_slow); 240488078d98SEric Dumazet 24051da177e4SLinus Torvalds /** 24061da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 24071da177e4SLinus Torvalds * @skb: buffer to reallocate 24081da177e4SLinus Torvalds * @delta: number of bytes to advance tail 24091da177e4SLinus Torvalds * 24101da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 24111da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 24121da177e4SLinus Torvalds * data from fragmented part. 24131da177e4SLinus Torvalds * 24141da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 24151da177e4SLinus Torvalds * 24161da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 24171da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 24181da177e4SLinus Torvalds * 24191da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 24201da177e4SLinus Torvalds * reloaded after call to this function. 24211da177e4SLinus Torvalds */ 24221da177e4SLinus Torvalds 24231da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 24241da177e4SLinus Torvalds * when it is necessary. 24251da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 24261da177e4SLinus Torvalds * 2. It may change skb pointers. 24271da177e4SLinus Torvalds * 24281da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 24291da177e4SLinus Torvalds */ 2430af72868bSJohannes Berg void *__pskb_pull_tail(struct sk_buff *skb, int delta) 24311da177e4SLinus Torvalds { 24321da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 24331da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 24341da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 24351da177e4SLinus Torvalds */ 24364305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 24371da177e4SLinus Torvalds 24381da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 24391da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 24401da177e4SLinus Torvalds GFP_ATOMIC)) 24411da177e4SLinus Torvalds return NULL; 24421da177e4SLinus Torvalds } 24431da177e4SLinus Torvalds 24449f77fad3STim Hansen BUG_ON(skb_copy_bits(skb, skb_headlen(skb), 24459f77fad3STim Hansen skb_tail_pointer(skb), delta)); 24461da177e4SLinus Torvalds 24471da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 24481da177e4SLinus Torvalds * size of pulled pages. Superb. 24491da177e4SLinus Torvalds */ 245021dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 24511da177e4SLinus Torvalds goto pull_pages; 24521da177e4SLinus Torvalds 24531da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 24541da177e4SLinus Torvalds eat = delta; 24551da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 24569e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 24579e903e08SEric Dumazet 24589e903e08SEric Dumazet if (size >= eat) 24591da177e4SLinus Torvalds goto pull_pages; 24609e903e08SEric Dumazet eat -= size; 24611da177e4SLinus Torvalds } 24621da177e4SLinus Torvalds 24631da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 246409001b03SWenhua Shi * Certainly, it is possible to add an offset to skb data, 24651da177e4SLinus Torvalds * but taking into account that pulling is expected to 24661da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 24671da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 24681da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 24691da177e4SLinus Torvalds */ 24701da177e4SLinus Torvalds if (eat) { 24711da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 24721da177e4SLinus Torvalds struct sk_buff *clone = NULL; 24731da177e4SLinus Torvalds struct sk_buff *insp = NULL; 24741da177e4SLinus Torvalds 24751da177e4SLinus Torvalds do { 24761da177e4SLinus Torvalds if (list->len <= eat) { 24771da177e4SLinus Torvalds /* Eaten as whole. */ 24781da177e4SLinus Torvalds eat -= list->len; 24791da177e4SLinus Torvalds list = list->next; 24801da177e4SLinus Torvalds insp = list; 24811da177e4SLinus Torvalds } else { 24821da177e4SLinus Torvalds /* Eaten partially. */ 24832d7afdcbSSubash Abhinov Kasiviswanathan if (skb_is_gso(skb) && !list->head_frag && 24842d7afdcbSSubash Abhinov Kasiviswanathan skb_headlen(list)) 24852d7afdcbSSubash Abhinov Kasiviswanathan skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 24861da177e4SLinus Torvalds 24871da177e4SLinus Torvalds if (skb_shared(list)) { 24881da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 24891da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 24901da177e4SLinus Torvalds if (!clone) 24911da177e4SLinus Torvalds return NULL; 24921da177e4SLinus Torvalds insp = list->next; 24931da177e4SLinus Torvalds list = clone; 24941da177e4SLinus Torvalds } else { 24951da177e4SLinus Torvalds /* This may be pulled without 24961da177e4SLinus Torvalds * problems. */ 24971da177e4SLinus Torvalds insp = list; 24981da177e4SLinus Torvalds } 24991da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 25001da177e4SLinus Torvalds kfree_skb(clone); 25011da177e4SLinus Torvalds return NULL; 25021da177e4SLinus Torvalds } 25031da177e4SLinus Torvalds break; 25041da177e4SLinus Torvalds } 25051da177e4SLinus Torvalds } while (eat); 25061da177e4SLinus Torvalds 25071da177e4SLinus Torvalds /* Free pulled out fragments. */ 25081da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 25091da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 2510ef527f96SEric Dumazet consume_skb(list); 25111da177e4SLinus Torvalds } 25121da177e4SLinus Torvalds /* And insert new clone at head. */ 25131da177e4SLinus Torvalds if (clone) { 25141da177e4SLinus Torvalds clone->next = list; 25151da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 25161da177e4SLinus Torvalds } 25171da177e4SLinus Torvalds } 25181da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 25191da177e4SLinus Torvalds 25201da177e4SLinus Torvalds pull_pages: 25211da177e4SLinus Torvalds eat = delta; 25221da177e4SLinus Torvalds k = 0; 25231da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 25249e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 25259e903e08SEric Dumazet 25269e903e08SEric Dumazet if (size <= eat) { 2527ea2ab693SIan Campbell skb_frag_unref(skb, i); 25289e903e08SEric Dumazet eat -= size; 25291da177e4SLinus Torvalds } else { 2530b54c9d5bSJonathan Lemon skb_frag_t *frag = &skb_shinfo(skb)->frags[k]; 2531b54c9d5bSJonathan Lemon 2532b54c9d5bSJonathan Lemon *frag = skb_shinfo(skb)->frags[i]; 25331da177e4SLinus Torvalds if (eat) { 2534b54c9d5bSJonathan Lemon skb_frag_off_add(frag, eat); 2535b54c9d5bSJonathan Lemon skb_frag_size_sub(frag, eat); 25363ccc6c6fSlinzhang if (!i) 25373ccc6c6fSlinzhang goto end; 25381da177e4SLinus Torvalds eat = 0; 25391da177e4SLinus Torvalds } 25401da177e4SLinus Torvalds k++; 25411da177e4SLinus Torvalds } 25421da177e4SLinus Torvalds } 25431da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 25441da177e4SLinus Torvalds 25453ccc6c6fSlinzhang end: 25461da177e4SLinus Torvalds skb->tail += delta; 25471da177e4SLinus Torvalds skb->data_len -= delta; 25481da177e4SLinus Torvalds 25491f8b977aSWillem de Bruijn if (!skb->data_len) 25501f8b977aSWillem de Bruijn skb_zcopy_clear(skb, false); 25511f8b977aSWillem de Bruijn 255227a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 25531da177e4SLinus Torvalds } 2554b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 25551da177e4SLinus Torvalds 255622019b17SEric Dumazet /** 255722019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 255822019b17SEric Dumazet * @skb: source skb 255922019b17SEric Dumazet * @offset: offset in source 256022019b17SEric Dumazet * @to: destination buffer 256122019b17SEric Dumazet * @len: number of bytes to copy 256222019b17SEric Dumazet * 256322019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 256422019b17SEric Dumazet * destination buffer. 256522019b17SEric Dumazet * 256622019b17SEric Dumazet * CAUTION ! : 256722019b17SEric Dumazet * If its prototype is ever changed, 256822019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 256922019b17SEric Dumazet * since it is called from BPF assembly code. 257022019b17SEric Dumazet */ 25711da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 25721da177e4SLinus Torvalds { 25731a028e50SDavid S. Miller int start = skb_headlen(skb); 2574fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2575fbb398a8SDavid S. Miller int i, copy; 25761da177e4SLinus Torvalds 25771da177e4SLinus Torvalds if (offset > (int)skb->len - len) 25781da177e4SLinus Torvalds goto fault; 25791da177e4SLinus Torvalds 25801da177e4SLinus Torvalds /* Copy header. */ 25811a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 25821da177e4SLinus Torvalds if (copy > len) 25831da177e4SLinus Torvalds copy = len; 2584d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 25851da177e4SLinus Torvalds if ((len -= copy) == 0) 25861da177e4SLinus Torvalds return 0; 25871da177e4SLinus Torvalds offset += copy; 25881da177e4SLinus Torvalds to += copy; 25891da177e4SLinus Torvalds } 25901da177e4SLinus Torvalds 25911da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 25921a028e50SDavid S. Miller int end; 259351c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 25941da177e4SLinus Torvalds 2595547b792cSIlpo Järvinen WARN_ON(start > offset + len); 25961a028e50SDavid S. Miller 259751c56b00SEric Dumazet end = start + skb_frag_size(f); 25981da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 2599c613c209SWillem de Bruijn u32 p_off, p_len, copied; 2600c613c209SWillem de Bruijn struct page *p; 26011da177e4SLinus Torvalds u8 *vaddr; 26021da177e4SLinus Torvalds 26031da177e4SLinus Torvalds if (copy > len) 26041da177e4SLinus Torvalds copy = len; 26051da177e4SLinus Torvalds 2606c613c209SWillem de Bruijn skb_frag_foreach_page(f, 2607b54c9d5bSJonathan Lemon skb_frag_off(f) + offset - start, 2608c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 2609c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 2610c613c209SWillem de Bruijn memcpy(to + copied, vaddr + p_off, p_len); 261151c56b00SEric Dumazet kunmap_atomic(vaddr); 2612c613c209SWillem de Bruijn } 26131da177e4SLinus Torvalds 26141da177e4SLinus Torvalds if ((len -= copy) == 0) 26151da177e4SLinus Torvalds return 0; 26161da177e4SLinus Torvalds offset += copy; 26171da177e4SLinus Torvalds to += copy; 26181da177e4SLinus Torvalds } 26191a028e50SDavid S. Miller start = end; 26201da177e4SLinus Torvalds } 26211da177e4SLinus Torvalds 2622fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 26231a028e50SDavid S. Miller int end; 26241da177e4SLinus Torvalds 2625547b792cSIlpo Järvinen WARN_ON(start > offset + len); 26261a028e50SDavid S. Miller 2627fbb398a8SDavid S. Miller end = start + frag_iter->len; 26281da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 26291da177e4SLinus Torvalds if (copy > len) 26301da177e4SLinus Torvalds copy = len; 2631fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 26321da177e4SLinus Torvalds goto fault; 26331da177e4SLinus Torvalds if ((len -= copy) == 0) 26341da177e4SLinus Torvalds return 0; 26351da177e4SLinus Torvalds offset += copy; 26361da177e4SLinus Torvalds to += copy; 26371da177e4SLinus Torvalds } 26381a028e50SDavid S. Miller start = end; 26391da177e4SLinus Torvalds } 2640a6686f2fSShirley Ma 26411da177e4SLinus Torvalds if (!len) 26421da177e4SLinus Torvalds return 0; 26431da177e4SLinus Torvalds 26441da177e4SLinus Torvalds fault: 26451da177e4SLinus Torvalds return -EFAULT; 26461da177e4SLinus Torvalds } 2647b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 26481da177e4SLinus Torvalds 26499c55e01cSJens Axboe /* 26509c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 26519c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 26529c55e01cSJens Axboe */ 26539c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 26549c55e01cSJens Axboe { 26558b9d3728SJarek Poplawski put_page(spd->pages[i]); 26568b9d3728SJarek Poplawski } 26579c55e01cSJens Axboe 2658a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 26594fb66994SJarek Poplawski unsigned int *offset, 266018aafc62SEric Dumazet struct sock *sk) 26618b9d3728SJarek Poplawski { 26625640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 26638b9d3728SJarek Poplawski 26645640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 26658b9d3728SJarek Poplawski return NULL; 26664fb66994SJarek Poplawski 26675640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 26684fb66994SJarek Poplawski 26695640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 26705640f768SEric Dumazet page_address(page) + *offset, *len); 26715640f768SEric Dumazet *offset = pfrag->offset; 26725640f768SEric Dumazet pfrag->offset += *len; 26734fb66994SJarek Poplawski 26745640f768SEric Dumazet return pfrag->page; 26759c55e01cSJens Axboe } 26769c55e01cSJens Axboe 267741c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 267841c73a0dSEric Dumazet struct page *page, 267941c73a0dSEric Dumazet unsigned int offset) 268041c73a0dSEric Dumazet { 268141c73a0dSEric Dumazet return spd->nr_pages && 268241c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 268341c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 268441c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 268541c73a0dSEric Dumazet } 268641c73a0dSEric Dumazet 26879c55e01cSJens Axboe /* 26889c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 26899c55e01cSJens Axboe */ 2690a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 269135f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 26924fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 269318aafc62SEric Dumazet bool linear, 26947a67e56fSJarek Poplawski struct sock *sk) 26959c55e01cSJens Axboe { 269641c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2697a108d5f3SDavid S. Miller return true; 26989c55e01cSJens Axboe 26998b9d3728SJarek Poplawski if (linear) { 270018aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 27018b9d3728SJarek Poplawski if (!page) 2702a108d5f3SDavid S. Miller return true; 270341c73a0dSEric Dumazet } 270441c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 270541c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 2706a108d5f3SDavid S. Miller return false; 270741c73a0dSEric Dumazet } 27088b9d3728SJarek Poplawski get_page(page); 27099c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 27104fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 27119c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 27129c55e01cSJens Axboe spd->nr_pages++; 27138b9d3728SJarek Poplawski 2714a108d5f3SDavid S. Miller return false; 27159c55e01cSJens Axboe } 27169c55e01cSJens Axboe 2717a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 27182870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 271918aafc62SEric Dumazet unsigned int *len, 2720d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 272135f3d14dSJens Axboe struct sock *sk, 272235f3d14dSJens Axboe struct pipe_inode_info *pipe) 27239c55e01cSJens Axboe { 27242870c43dSOctavian Purdila if (!*len) 2725a108d5f3SDavid S. Miller return true; 27269c55e01cSJens Axboe 27272870c43dSOctavian Purdila /* skip this segment if already processed */ 27282870c43dSOctavian Purdila if (*off >= plen) { 27292870c43dSOctavian Purdila *off -= plen; 2730a108d5f3SDavid S. Miller return false; 27312870c43dSOctavian Purdila } 27322870c43dSOctavian Purdila 27332870c43dSOctavian Purdila /* ignore any bits we already processed */ 27349ca1b22dSEric Dumazet poff += *off; 27359ca1b22dSEric Dumazet plen -= *off; 27362870c43dSOctavian Purdila *off = 0; 27372870c43dSOctavian Purdila 273818aafc62SEric Dumazet do { 273918aafc62SEric Dumazet unsigned int flen = min(*len, plen); 27402870c43dSOctavian Purdila 274118aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 274218aafc62SEric Dumazet linear, sk)) 2743a108d5f3SDavid S. Miller return true; 274418aafc62SEric Dumazet poff += flen; 274518aafc62SEric Dumazet plen -= flen; 27462870c43dSOctavian Purdila *len -= flen; 274718aafc62SEric Dumazet } while (*len && plen); 27482870c43dSOctavian Purdila 2749a108d5f3SDavid S. Miller return false; 2750db43a282SOctavian Purdila } 27519c55e01cSJens Axboe 27529c55e01cSJens Axboe /* 2753a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 27542870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 27559c55e01cSJens Axboe */ 2756a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 275735f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 275835f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 27592870c43dSOctavian Purdila { 27602870c43dSOctavian Purdila int seg; 2761fa9835e5STom Herbert struct sk_buff *iter; 27629c55e01cSJens Axboe 27631d0c0b32SEric Dumazet /* map the linear part : 27642996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 27652996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 27662996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 27679c55e01cSJens Axboe */ 27682870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 27692870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 27702870c43dSOctavian Purdila skb_headlen(skb), 277118aafc62SEric Dumazet offset, len, spd, 27723a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 27731d0c0b32SEric Dumazet sk, pipe)) 2774a108d5f3SDavid S. Miller return true; 27759c55e01cSJens Axboe 27769c55e01cSJens Axboe /* 27779c55e01cSJens Axboe * then map the fragments 27789c55e01cSJens Axboe */ 27799c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 27809c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 27819c55e01cSJens Axboe 2782ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 2783b54c9d5bSJonathan Lemon skb_frag_off(f), skb_frag_size(f), 278418aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 2785a108d5f3SDavid S. Miller return true; 27869c55e01cSJens Axboe } 27879c55e01cSJens Axboe 2788fa9835e5STom Herbert skb_walk_frags(skb, iter) { 2789fa9835e5STom Herbert if (*offset >= iter->len) { 2790fa9835e5STom Herbert *offset -= iter->len; 2791fa9835e5STom Herbert continue; 2792fa9835e5STom Herbert } 2793fa9835e5STom Herbert /* __skb_splice_bits() only fails if the output has no room 2794fa9835e5STom Herbert * left, so no point in going over the frag_list for the error 2795fa9835e5STom Herbert * case. 2796fa9835e5STom Herbert */ 2797fa9835e5STom Herbert if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2798fa9835e5STom Herbert return true; 2799fa9835e5STom Herbert } 2800fa9835e5STom Herbert 2801a108d5f3SDavid S. Miller return false; 28029c55e01cSJens Axboe } 28039c55e01cSJens Axboe 28049c55e01cSJens Axboe /* 28059c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 2806fa9835e5STom Herbert * the fragments, and the frag list. 28079c55e01cSJens Axboe */ 2808a60e3cc7SHannes Frederic Sowa int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 28099c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 281025869262SAl Viro unsigned int flags) 28119c55e01cSJens Axboe { 281241c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 281341c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 28149c55e01cSJens Axboe struct splice_pipe_desc spd = { 28159c55e01cSJens Axboe .pages = pages, 28169c55e01cSJens Axboe .partial = partial, 2817047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 281828a625cbSMiklos Szeredi .ops = &nosteal_pipe_buf_ops, 28199c55e01cSJens Axboe .spd_release = sock_spd_release, 28209c55e01cSJens Axboe }; 282135f3d14dSJens Axboe int ret = 0; 282235f3d14dSJens Axboe 2823fa9835e5STom Herbert __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 28249c55e01cSJens Axboe 2825a60e3cc7SHannes Frederic Sowa if (spd.nr_pages) 282625869262SAl Viro ret = splice_to_pipe(pipe, &spd); 28279c55e01cSJens Axboe 282835f3d14dSJens Axboe return ret; 28299c55e01cSJens Axboe } 28302b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_splice_bits); 28319c55e01cSJens Axboe 28320739cd28SCong Wang static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg, 28330739cd28SCong Wang struct kvec *vec, size_t num, size_t size) 28340739cd28SCong Wang { 28350739cd28SCong Wang struct socket *sock = sk->sk_socket; 28360739cd28SCong Wang 28370739cd28SCong Wang if (!sock) 28380739cd28SCong Wang return -EINVAL; 28390739cd28SCong Wang return kernel_sendmsg(sock, msg, vec, num, size); 28400739cd28SCong Wang } 28410739cd28SCong Wang 28420739cd28SCong Wang static int sendpage_unlocked(struct sock *sk, struct page *page, int offset, 28430739cd28SCong Wang size_t size, int flags) 28440739cd28SCong Wang { 28450739cd28SCong Wang struct socket *sock = sk->sk_socket; 28460739cd28SCong Wang 28470739cd28SCong Wang if (!sock) 28480739cd28SCong Wang return -EINVAL; 28490739cd28SCong Wang return kernel_sendpage(sock, page, offset, size, flags); 28500739cd28SCong Wang } 28510739cd28SCong Wang 28520739cd28SCong Wang typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg, 28530739cd28SCong Wang struct kvec *vec, size_t num, size_t size); 28540739cd28SCong Wang typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset, 28550739cd28SCong Wang size_t size, int flags); 28560739cd28SCong Wang static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, 28570739cd28SCong Wang int len, sendmsg_func sendmsg, sendpage_func sendpage) 285820bf50deSTom Herbert { 285920bf50deSTom Herbert unsigned int orig_len = len; 286020bf50deSTom Herbert struct sk_buff *head = skb; 286120bf50deSTom Herbert unsigned short fragidx; 286220bf50deSTom Herbert int slen, ret; 286320bf50deSTom Herbert 286420bf50deSTom Herbert do_frag_list: 286520bf50deSTom Herbert 286620bf50deSTom Herbert /* Deal with head data */ 286720bf50deSTom Herbert while (offset < skb_headlen(skb) && len) { 286820bf50deSTom Herbert struct kvec kv; 286920bf50deSTom Herbert struct msghdr msg; 287020bf50deSTom Herbert 287120bf50deSTom Herbert slen = min_t(int, len, skb_headlen(skb) - offset); 287220bf50deSTom Herbert kv.iov_base = skb->data + offset; 2873db5980d8SJohn Fastabend kv.iov_len = slen; 287420bf50deSTom Herbert memset(&msg, 0, sizeof(msg)); 2875bd95e678SJohn Fastabend msg.msg_flags = MSG_DONTWAIT; 287620bf50deSTom Herbert 28770739cd28SCong Wang ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked, 28780739cd28SCong Wang sendmsg_unlocked, sk, &msg, &kv, 1, slen); 287920bf50deSTom Herbert if (ret <= 0) 288020bf50deSTom Herbert goto error; 288120bf50deSTom Herbert 288220bf50deSTom Herbert offset += ret; 288320bf50deSTom Herbert len -= ret; 288420bf50deSTom Herbert } 288520bf50deSTom Herbert 288620bf50deSTom Herbert /* All the data was skb head? */ 288720bf50deSTom Herbert if (!len) 288820bf50deSTom Herbert goto out; 288920bf50deSTom Herbert 289020bf50deSTom Herbert /* Make offset relative to start of frags */ 289120bf50deSTom Herbert offset -= skb_headlen(skb); 289220bf50deSTom Herbert 289320bf50deSTom Herbert /* Find where we are in frag list */ 289420bf50deSTom Herbert for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 289520bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 289620bf50deSTom Herbert 2897d8e18a51SMatthew Wilcox (Oracle) if (offset < skb_frag_size(frag)) 289820bf50deSTom Herbert break; 289920bf50deSTom Herbert 2900d8e18a51SMatthew Wilcox (Oracle) offset -= skb_frag_size(frag); 290120bf50deSTom Herbert } 290220bf50deSTom Herbert 290320bf50deSTom Herbert for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 290420bf50deSTom Herbert skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 290520bf50deSTom Herbert 2906d8e18a51SMatthew Wilcox (Oracle) slen = min_t(size_t, len, skb_frag_size(frag) - offset); 290720bf50deSTom Herbert 290820bf50deSTom Herbert while (slen) { 29090739cd28SCong Wang ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked, 29100739cd28SCong Wang sendpage_unlocked, sk, 29110739cd28SCong Wang skb_frag_page(frag), 2912b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset, 291320bf50deSTom Herbert slen, MSG_DONTWAIT); 291420bf50deSTom Herbert if (ret <= 0) 291520bf50deSTom Herbert goto error; 291620bf50deSTom Herbert 291720bf50deSTom Herbert len -= ret; 291820bf50deSTom Herbert offset += ret; 291920bf50deSTom Herbert slen -= ret; 292020bf50deSTom Herbert } 292120bf50deSTom Herbert 292220bf50deSTom Herbert offset = 0; 292320bf50deSTom Herbert } 292420bf50deSTom Herbert 292520bf50deSTom Herbert if (len) { 292620bf50deSTom Herbert /* Process any frag lists */ 292720bf50deSTom Herbert 292820bf50deSTom Herbert if (skb == head) { 292920bf50deSTom Herbert if (skb_has_frag_list(skb)) { 293020bf50deSTom Herbert skb = skb_shinfo(skb)->frag_list; 293120bf50deSTom Herbert goto do_frag_list; 293220bf50deSTom Herbert } 293320bf50deSTom Herbert } else if (skb->next) { 293420bf50deSTom Herbert skb = skb->next; 293520bf50deSTom Herbert goto do_frag_list; 293620bf50deSTom Herbert } 293720bf50deSTom Herbert } 293820bf50deSTom Herbert 293920bf50deSTom Herbert out: 294020bf50deSTom Herbert return orig_len - len; 294120bf50deSTom Herbert 294220bf50deSTom Herbert error: 294320bf50deSTom Herbert return orig_len == len ? ret : orig_len - len; 294420bf50deSTom Herbert } 29450739cd28SCong Wang 29460739cd28SCong Wang /* Send skb data on a socket. Socket must be locked. */ 29470739cd28SCong Wang int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 29480739cd28SCong Wang int len) 29490739cd28SCong Wang { 29500739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked, 29510739cd28SCong Wang kernel_sendpage_locked); 29520739cd28SCong Wang } 295320bf50deSTom Herbert EXPORT_SYMBOL_GPL(skb_send_sock_locked); 295420bf50deSTom Herbert 29550739cd28SCong Wang /* Send skb data on a socket. Socket must be unlocked. */ 29560739cd28SCong Wang int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 29570739cd28SCong Wang { 29580739cd28SCong Wang return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked, 29590739cd28SCong Wang sendpage_unlocked); 29600739cd28SCong Wang } 29610739cd28SCong Wang 2962357b40a1SHerbert Xu /** 2963357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 2964357b40a1SHerbert Xu * @skb: destination buffer 2965357b40a1SHerbert Xu * @offset: offset in destination 2966357b40a1SHerbert Xu * @from: source buffer 2967357b40a1SHerbert Xu * @len: number of bytes to copy 2968357b40a1SHerbert Xu * 2969357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 2970357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 2971357b40a1SHerbert Xu * traversing fragment lists and such. 2972357b40a1SHerbert Xu */ 2973357b40a1SHerbert Xu 29740c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2975357b40a1SHerbert Xu { 29761a028e50SDavid S. Miller int start = skb_headlen(skb); 2977fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 2978fbb398a8SDavid S. Miller int i, copy; 2979357b40a1SHerbert Xu 2980357b40a1SHerbert Xu if (offset > (int)skb->len - len) 2981357b40a1SHerbert Xu goto fault; 2982357b40a1SHerbert Xu 29831a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 2984357b40a1SHerbert Xu if (copy > len) 2985357b40a1SHerbert Xu copy = len; 298627d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 2987357b40a1SHerbert Xu if ((len -= copy) == 0) 2988357b40a1SHerbert Xu return 0; 2989357b40a1SHerbert Xu offset += copy; 2990357b40a1SHerbert Xu from += copy; 2991357b40a1SHerbert Xu } 2992357b40a1SHerbert Xu 2993357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2994357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 29951a028e50SDavid S. Miller int end; 2996357b40a1SHerbert Xu 2997547b792cSIlpo Järvinen WARN_ON(start > offset + len); 29981a028e50SDavid S. Miller 29999e903e08SEric Dumazet end = start + skb_frag_size(frag); 3000357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 3001c613c209SWillem de Bruijn u32 p_off, p_len, copied; 3002c613c209SWillem de Bruijn struct page *p; 3003357b40a1SHerbert Xu u8 *vaddr; 3004357b40a1SHerbert Xu 3005357b40a1SHerbert Xu if (copy > len) 3006357b40a1SHerbert Xu copy = len; 3007357b40a1SHerbert Xu 3008c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 3009b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 3010c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 3011c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 3012c613c209SWillem de Bruijn memcpy(vaddr + p_off, from + copied, p_len); 301351c56b00SEric Dumazet kunmap_atomic(vaddr); 3014c613c209SWillem de Bruijn } 3015357b40a1SHerbert Xu 3016357b40a1SHerbert Xu if ((len -= copy) == 0) 3017357b40a1SHerbert Xu return 0; 3018357b40a1SHerbert Xu offset += copy; 3019357b40a1SHerbert Xu from += copy; 3020357b40a1SHerbert Xu } 30211a028e50SDavid S. Miller start = end; 3022357b40a1SHerbert Xu } 3023357b40a1SHerbert Xu 3024fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 30251a028e50SDavid S. Miller int end; 3026357b40a1SHerbert Xu 3027547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30281a028e50SDavid S. Miller 3029fbb398a8SDavid S. Miller end = start + frag_iter->len; 3030357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 3031357b40a1SHerbert Xu if (copy > len) 3032357b40a1SHerbert Xu copy = len; 3033fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 30341a028e50SDavid S. Miller from, copy)) 3035357b40a1SHerbert Xu goto fault; 3036357b40a1SHerbert Xu if ((len -= copy) == 0) 3037357b40a1SHerbert Xu return 0; 3038357b40a1SHerbert Xu offset += copy; 3039357b40a1SHerbert Xu from += copy; 3040357b40a1SHerbert Xu } 30411a028e50SDavid S. Miller start = end; 3042357b40a1SHerbert Xu } 3043357b40a1SHerbert Xu if (!len) 3044357b40a1SHerbert Xu return 0; 3045357b40a1SHerbert Xu 3046357b40a1SHerbert Xu fault: 3047357b40a1SHerbert Xu return -EFAULT; 3048357b40a1SHerbert Xu } 3049357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 3050357b40a1SHerbert Xu 30511da177e4SLinus Torvalds /* Checksum skb data. */ 30522817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 30532817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 30541da177e4SLinus Torvalds { 30551a028e50SDavid S. Miller int start = skb_headlen(skb); 30561a028e50SDavid S. Miller int i, copy = start - offset; 3057fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 30581da177e4SLinus Torvalds int pos = 0; 30591da177e4SLinus Torvalds 30601da177e4SLinus Torvalds /* Checksum header. */ 30611da177e4SLinus Torvalds if (copy > 0) { 30621da177e4SLinus Torvalds if (copy > len) 30631da177e4SLinus Torvalds copy = len; 30642544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->update, csum_partial_ext, 30652544af03SMatteo Croce skb->data + offset, copy, csum); 30661da177e4SLinus Torvalds if ((len -= copy) == 0) 30671da177e4SLinus Torvalds return csum; 30681da177e4SLinus Torvalds offset += copy; 30691da177e4SLinus Torvalds pos = copy; 30701da177e4SLinus Torvalds } 30711da177e4SLinus Torvalds 30721da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 30731a028e50SDavid S. Miller int end; 307451c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 30751da177e4SLinus Torvalds 3076547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30771a028e50SDavid S. Miller 307851c56b00SEric Dumazet end = start + skb_frag_size(frag); 30791da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 3080c613c209SWillem de Bruijn u32 p_off, p_len, copied; 3081c613c209SWillem de Bruijn struct page *p; 308244bb9363SAl Viro __wsum csum2; 30831da177e4SLinus Torvalds u8 *vaddr; 30841da177e4SLinus Torvalds 30851da177e4SLinus Torvalds if (copy > len) 30861da177e4SLinus Torvalds copy = len; 3087c613c209SWillem de Bruijn 3088c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 3089b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 3090c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 3091c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 30922544af03SMatteo Croce csum2 = INDIRECT_CALL_1(ops->update, 30932544af03SMatteo Croce csum_partial_ext, 30942544af03SMatteo Croce vaddr + p_off, p_len, 0); 309551c56b00SEric Dumazet kunmap_atomic(vaddr); 30962544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, 30972544af03SMatteo Croce csum_block_add_ext, csum, 30982544af03SMatteo Croce csum2, pos, p_len); 3099c613c209SWillem de Bruijn pos += p_len; 3100c613c209SWillem de Bruijn } 3101c613c209SWillem de Bruijn 31021da177e4SLinus Torvalds if (!(len -= copy)) 31031da177e4SLinus Torvalds return csum; 31041da177e4SLinus Torvalds offset += copy; 31051da177e4SLinus Torvalds } 31061a028e50SDavid S. Miller start = end; 31071da177e4SLinus Torvalds } 31081da177e4SLinus Torvalds 3109fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 31101a028e50SDavid S. Miller int end; 31111da177e4SLinus Torvalds 3112547b792cSIlpo Järvinen WARN_ON(start > offset + len); 31131a028e50SDavid S. Miller 3114fbb398a8SDavid S. Miller end = start + frag_iter->len; 31151da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 31165f92a738SAl Viro __wsum csum2; 31171da177e4SLinus Torvalds if (copy > len) 31181da177e4SLinus Torvalds copy = len; 31192817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 31202817a336SDaniel Borkmann copy, 0, ops); 31212544af03SMatteo Croce csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext, 31222544af03SMatteo Croce csum, csum2, pos, copy); 31231da177e4SLinus Torvalds if ((len -= copy) == 0) 31241da177e4SLinus Torvalds return csum; 31251da177e4SLinus Torvalds offset += copy; 31261da177e4SLinus Torvalds pos += copy; 31271da177e4SLinus Torvalds } 31281a028e50SDavid S. Miller start = end; 31291da177e4SLinus Torvalds } 313009a62660SKris Katterjohn BUG_ON(len); 31311da177e4SLinus Torvalds 31321da177e4SLinus Torvalds return csum; 31331da177e4SLinus Torvalds } 31342817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 31352817a336SDaniel Borkmann 31362817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 31372817a336SDaniel Borkmann int len, __wsum csum) 31382817a336SDaniel Borkmann { 31392817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 3140cea80ea8SDaniel Borkmann .update = csum_partial_ext, 31412817a336SDaniel Borkmann .combine = csum_block_add_ext, 31422817a336SDaniel Borkmann }; 31432817a336SDaniel Borkmann 31442817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 31452817a336SDaniel Borkmann } 3146b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 31471da177e4SLinus Torvalds 31481da177e4SLinus Torvalds /* Both of above in one bottle. */ 31491da177e4SLinus Torvalds 315081d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 31518d5930dfSAl Viro u8 *to, int len) 31521da177e4SLinus Torvalds { 31531a028e50SDavid S. Miller int start = skb_headlen(skb); 31541a028e50SDavid S. Miller int i, copy = start - offset; 3155fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 31561da177e4SLinus Torvalds int pos = 0; 31578d5930dfSAl Viro __wsum csum = 0; 31581da177e4SLinus Torvalds 31591da177e4SLinus Torvalds /* Copy header. */ 31601da177e4SLinus Torvalds if (copy > 0) { 31611da177e4SLinus Torvalds if (copy > len) 31621da177e4SLinus Torvalds copy = len; 31631da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 3164cc44c17bSAl Viro copy); 31651da177e4SLinus Torvalds if ((len -= copy) == 0) 31661da177e4SLinus Torvalds return csum; 31671da177e4SLinus Torvalds offset += copy; 31681da177e4SLinus Torvalds to += copy; 31691da177e4SLinus Torvalds pos = copy; 31701da177e4SLinus Torvalds } 31711da177e4SLinus Torvalds 31721da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 31731a028e50SDavid S. Miller int end; 31741da177e4SLinus Torvalds 3175547b792cSIlpo Järvinen WARN_ON(start > offset + len); 31761a028e50SDavid S. Miller 31779e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 31781da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 3179c613c209SWillem de Bruijn skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3180c613c209SWillem de Bruijn u32 p_off, p_len, copied; 3181c613c209SWillem de Bruijn struct page *p; 31825084205fSAl Viro __wsum csum2; 31831da177e4SLinus Torvalds u8 *vaddr; 31841da177e4SLinus Torvalds 31851da177e4SLinus Torvalds if (copy > len) 31861da177e4SLinus Torvalds copy = len; 3187c613c209SWillem de Bruijn 3188c613c209SWillem de Bruijn skb_frag_foreach_page(frag, 3189b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start, 3190c613c209SWillem de Bruijn copy, p, p_off, p_len, copied) { 3191c613c209SWillem de Bruijn vaddr = kmap_atomic(p); 3192c613c209SWillem de Bruijn csum2 = csum_partial_copy_nocheck(vaddr + p_off, 3193c613c209SWillem de Bruijn to + copied, 3194cc44c17bSAl Viro p_len); 319551c56b00SEric Dumazet kunmap_atomic(vaddr); 31961da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 3197c613c209SWillem de Bruijn pos += p_len; 3198c613c209SWillem de Bruijn } 3199c613c209SWillem de Bruijn 32001da177e4SLinus Torvalds if (!(len -= copy)) 32011da177e4SLinus Torvalds return csum; 32021da177e4SLinus Torvalds offset += copy; 32031da177e4SLinus Torvalds to += copy; 32041da177e4SLinus Torvalds } 32051a028e50SDavid S. Miller start = end; 32061da177e4SLinus Torvalds } 32071da177e4SLinus Torvalds 3208fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 320981d77662SAl Viro __wsum csum2; 32101a028e50SDavid S. Miller int end; 32111da177e4SLinus Torvalds 3212547b792cSIlpo Järvinen WARN_ON(start > offset + len); 32131a028e50SDavid S. Miller 3214fbb398a8SDavid S. Miller end = start + frag_iter->len; 32151da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 32161da177e4SLinus Torvalds if (copy > len) 32171da177e4SLinus Torvalds copy = len; 3218fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 32191a028e50SDavid S. Miller offset - start, 32208d5930dfSAl Viro to, copy); 32211da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 32221da177e4SLinus Torvalds if ((len -= copy) == 0) 32231da177e4SLinus Torvalds return csum; 32241da177e4SLinus Torvalds offset += copy; 32251da177e4SLinus Torvalds to += copy; 32261da177e4SLinus Torvalds pos += copy; 32271da177e4SLinus Torvalds } 32281a028e50SDavid S. Miller start = end; 32291da177e4SLinus Torvalds } 323009a62660SKris Katterjohn BUG_ON(len); 32311da177e4SLinus Torvalds return csum; 32321da177e4SLinus Torvalds } 3233b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 32341da177e4SLinus Torvalds 323549f8e832SCong Wang __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 323649f8e832SCong Wang { 323749f8e832SCong Wang __sum16 sum; 323849f8e832SCong Wang 323949f8e832SCong Wang sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 324014641931SCong Wang /* See comments in __skb_checksum_complete(). */ 324149f8e832SCong Wang if (likely(!sum)) { 324249f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 324349f8e832SCong Wang !skb->csum_complete_sw) 32447fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 324549f8e832SCong Wang } 324649f8e832SCong Wang if (!skb_shared(skb)) 324749f8e832SCong Wang skb->csum_valid = !sum; 324849f8e832SCong Wang return sum; 324949f8e832SCong Wang } 325049f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete_head); 325149f8e832SCong Wang 325214641931SCong Wang /* This function assumes skb->csum already holds pseudo header's checksum, 325314641931SCong Wang * which has been changed from the hardware checksum, for example, by 325414641931SCong Wang * __skb_checksum_validate_complete(). And, the original skb->csum must 325514641931SCong Wang * have been validated unsuccessfully for CHECKSUM_COMPLETE case. 325614641931SCong Wang * 325714641931SCong Wang * It returns non-zero if the recomputed checksum is still invalid, otherwise 325814641931SCong Wang * zero. The new checksum is stored back into skb->csum unless the skb is 325914641931SCong Wang * shared. 326014641931SCong Wang */ 326149f8e832SCong Wang __sum16 __skb_checksum_complete(struct sk_buff *skb) 326249f8e832SCong Wang { 326349f8e832SCong Wang __wsum csum; 326449f8e832SCong Wang __sum16 sum; 326549f8e832SCong Wang 326649f8e832SCong Wang csum = skb_checksum(skb, 0, skb->len, 0); 326749f8e832SCong Wang 326849f8e832SCong Wang sum = csum_fold(csum_add(skb->csum, csum)); 326914641931SCong Wang /* This check is inverted, because we already knew the hardware 327014641931SCong Wang * checksum is invalid before calling this function. So, if the 327114641931SCong Wang * re-computed checksum is valid instead, then we have a mismatch 327214641931SCong Wang * between the original skb->csum and skb_checksum(). This means either 327314641931SCong Wang * the original hardware checksum is incorrect or we screw up skb->csum 327414641931SCong Wang * when moving skb->data around. 327514641931SCong Wang */ 327649f8e832SCong Wang if (likely(!sum)) { 327749f8e832SCong Wang if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 327849f8e832SCong Wang !skb->csum_complete_sw) 32797fe50ac8SCong Wang netdev_rx_csum_fault(skb->dev, skb); 328049f8e832SCong Wang } 328149f8e832SCong Wang 328249f8e832SCong Wang if (!skb_shared(skb)) { 328349f8e832SCong Wang /* Save full packet checksum */ 328449f8e832SCong Wang skb->csum = csum; 328549f8e832SCong Wang skb->ip_summed = CHECKSUM_COMPLETE; 328649f8e832SCong Wang skb->csum_complete_sw = 1; 328749f8e832SCong Wang skb->csum_valid = !sum; 328849f8e832SCong Wang } 328949f8e832SCong Wang 329049f8e832SCong Wang return sum; 329149f8e832SCong Wang } 329249f8e832SCong Wang EXPORT_SYMBOL(__skb_checksum_complete); 329349f8e832SCong Wang 32949617813dSDavide Caratti static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 32959617813dSDavide Caratti { 32969617813dSDavide Caratti net_warn_ratelimited( 32979617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 32989617813dSDavide Caratti __func__); 32999617813dSDavide Caratti return 0; 33009617813dSDavide Caratti } 33019617813dSDavide Caratti 33029617813dSDavide Caratti static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 33039617813dSDavide Caratti int offset, int len) 33049617813dSDavide Caratti { 33059617813dSDavide Caratti net_warn_ratelimited( 33069617813dSDavide Caratti "%s: attempt to compute crc32c without libcrc32c.ko\n", 33079617813dSDavide Caratti __func__); 33089617813dSDavide Caratti return 0; 33099617813dSDavide Caratti } 33109617813dSDavide Caratti 33119617813dSDavide Caratti static const struct skb_checksum_ops default_crc32c_ops = { 33129617813dSDavide Caratti .update = warn_crc32c_csum_update, 33139617813dSDavide Caratti .combine = warn_crc32c_csum_combine, 33149617813dSDavide Caratti }; 33159617813dSDavide Caratti 33169617813dSDavide Caratti const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 33179617813dSDavide Caratti &default_crc32c_ops; 33189617813dSDavide Caratti EXPORT_SYMBOL(crc32c_csum_stub); 33199617813dSDavide Caratti 3320af2806f8SThomas Graf /** 3321af2806f8SThomas Graf * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 3322af2806f8SThomas Graf * @from: source buffer 3323af2806f8SThomas Graf * 3324af2806f8SThomas Graf * Calculates the amount of linear headroom needed in the 'to' skb passed 3325af2806f8SThomas Graf * into skb_zerocopy(). 3326af2806f8SThomas Graf */ 3327af2806f8SThomas Graf unsigned int 3328af2806f8SThomas Graf skb_zerocopy_headlen(const struct sk_buff *from) 3329af2806f8SThomas Graf { 3330af2806f8SThomas Graf unsigned int hlen = 0; 3331af2806f8SThomas Graf 3332af2806f8SThomas Graf if (!from->head_frag || 3333af2806f8SThomas Graf skb_headlen(from) < L1_CACHE_BYTES || 3334a17ad096SPravin B Shelar skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) { 3335af2806f8SThomas Graf hlen = skb_headlen(from); 3336a17ad096SPravin B Shelar if (!hlen) 3337a17ad096SPravin B Shelar hlen = from->len; 3338a17ad096SPravin B Shelar } 3339af2806f8SThomas Graf 3340af2806f8SThomas Graf if (skb_has_frag_list(from)) 3341af2806f8SThomas Graf hlen = from->len; 3342af2806f8SThomas Graf 3343af2806f8SThomas Graf return hlen; 3344af2806f8SThomas Graf } 3345af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 3346af2806f8SThomas Graf 3347af2806f8SThomas Graf /** 3348af2806f8SThomas Graf * skb_zerocopy - Zero copy skb to skb 3349af2806f8SThomas Graf * @to: destination buffer 33507fceb4deSMasanari Iida * @from: source buffer 3351af2806f8SThomas Graf * @len: number of bytes to copy from source buffer 3352af2806f8SThomas Graf * @hlen: size of linear headroom in destination buffer 3353af2806f8SThomas Graf * 3354af2806f8SThomas Graf * Copies up to `len` bytes from `from` to `to` by creating references 3355af2806f8SThomas Graf * to the frags in the source buffer. 3356af2806f8SThomas Graf * 3357af2806f8SThomas Graf * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 3358af2806f8SThomas Graf * headroom in the `to` buffer. 335936d5fe6aSZoltan Kiss * 336036d5fe6aSZoltan Kiss * Return value: 336136d5fe6aSZoltan Kiss * 0: everything is OK 336236d5fe6aSZoltan Kiss * -ENOMEM: couldn't orphan frags of @from due to lack of memory 336336d5fe6aSZoltan Kiss * -EFAULT: skb_copy_bits() found some problem with skb geometry 3364af2806f8SThomas Graf */ 336536d5fe6aSZoltan Kiss int 336636d5fe6aSZoltan Kiss skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 3367af2806f8SThomas Graf { 3368af2806f8SThomas Graf int i, j = 0; 3369af2806f8SThomas Graf int plen = 0; /* length of skb->head fragment */ 337036d5fe6aSZoltan Kiss int ret; 3371af2806f8SThomas Graf struct page *page; 3372af2806f8SThomas Graf unsigned int offset; 3373af2806f8SThomas Graf 3374af2806f8SThomas Graf BUG_ON(!from->head_frag && !hlen); 3375af2806f8SThomas Graf 3376af2806f8SThomas Graf /* dont bother with small payloads */ 337736d5fe6aSZoltan Kiss if (len <= skb_tailroom(to)) 337836d5fe6aSZoltan Kiss return skb_copy_bits(from, 0, skb_put(to, len), len); 3379af2806f8SThomas Graf 3380af2806f8SThomas Graf if (hlen) { 338136d5fe6aSZoltan Kiss ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 338236d5fe6aSZoltan Kiss if (unlikely(ret)) 338336d5fe6aSZoltan Kiss return ret; 3384af2806f8SThomas Graf len -= hlen; 3385af2806f8SThomas Graf } else { 3386af2806f8SThomas Graf plen = min_t(int, skb_headlen(from), len); 3387af2806f8SThomas Graf if (plen) { 3388af2806f8SThomas Graf page = virt_to_head_page(from->head); 3389af2806f8SThomas Graf offset = from->data - (unsigned char *)page_address(page); 3390af2806f8SThomas Graf __skb_fill_page_desc(to, 0, page, offset, plen); 3391af2806f8SThomas Graf get_page(page); 3392af2806f8SThomas Graf j = 1; 3393af2806f8SThomas Graf len -= plen; 3394af2806f8SThomas Graf } 3395af2806f8SThomas Graf } 3396af2806f8SThomas Graf 3397ede57d58SRichard Gobert skb_len_add(to, len + plen); 3398af2806f8SThomas Graf 339936d5fe6aSZoltan Kiss if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 340036d5fe6aSZoltan Kiss skb_tx_error(from); 340136d5fe6aSZoltan Kiss return -ENOMEM; 340236d5fe6aSZoltan Kiss } 34031f8b977aSWillem de Bruijn skb_zerocopy_clone(to, from, GFP_ATOMIC); 340436d5fe6aSZoltan Kiss 3405af2806f8SThomas Graf for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 3406d8e18a51SMatthew Wilcox (Oracle) int size; 3407d8e18a51SMatthew Wilcox (Oracle) 3408af2806f8SThomas Graf if (!len) 3409af2806f8SThomas Graf break; 3410af2806f8SThomas Graf skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 3411d8e18a51SMatthew Wilcox (Oracle) size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]), 3412d8e18a51SMatthew Wilcox (Oracle) len); 3413d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&skb_shinfo(to)->frags[j], size); 3414d8e18a51SMatthew Wilcox (Oracle) len -= size; 3415af2806f8SThomas Graf skb_frag_ref(to, j); 3416af2806f8SThomas Graf j++; 3417af2806f8SThomas Graf } 3418af2806f8SThomas Graf skb_shinfo(to)->nr_frags = j; 341936d5fe6aSZoltan Kiss 342036d5fe6aSZoltan Kiss return 0; 3421af2806f8SThomas Graf } 3422af2806f8SThomas Graf EXPORT_SYMBOL_GPL(skb_zerocopy); 3423af2806f8SThomas Graf 34241da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 34251da177e4SLinus Torvalds { 3426d3bc23e7SAl Viro __wsum csum; 34271da177e4SLinus Torvalds long csstart; 34281da177e4SLinus Torvalds 342984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 343055508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 34311da177e4SLinus Torvalds else 34321da177e4SLinus Torvalds csstart = skb_headlen(skb); 34331da177e4SLinus Torvalds 343409a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 34351da177e4SLinus Torvalds 3436d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 34371da177e4SLinus Torvalds 34381da177e4SLinus Torvalds csum = 0; 34391da177e4SLinus Torvalds if (csstart != skb->len) 34401da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 34418d5930dfSAl Viro skb->len - csstart); 34421da177e4SLinus Torvalds 344384fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 3444ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 34451da177e4SLinus Torvalds 3446d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 34471da177e4SLinus Torvalds } 34481da177e4SLinus Torvalds } 3449b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 34501da177e4SLinus Torvalds 34511da177e4SLinus Torvalds /** 34521da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 34531da177e4SLinus Torvalds * @list: list to dequeue from 34541da177e4SLinus Torvalds * 34551da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 34561da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 34571da177e4SLinus Torvalds * returned or %NULL if the list is empty. 34581da177e4SLinus Torvalds */ 34591da177e4SLinus Torvalds 34601da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 34611da177e4SLinus Torvalds { 34621da177e4SLinus Torvalds unsigned long flags; 34631da177e4SLinus Torvalds struct sk_buff *result; 34641da177e4SLinus Torvalds 34651da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 34661da177e4SLinus Torvalds result = __skb_dequeue(list); 34671da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 34681da177e4SLinus Torvalds return result; 34691da177e4SLinus Torvalds } 3470b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 34711da177e4SLinus Torvalds 34721da177e4SLinus Torvalds /** 34731da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 34741da177e4SLinus Torvalds * @list: list to dequeue from 34751da177e4SLinus Torvalds * 34761da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 34771da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 34781da177e4SLinus Torvalds * returned or %NULL if the list is empty. 34791da177e4SLinus Torvalds */ 34801da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 34811da177e4SLinus Torvalds { 34821da177e4SLinus Torvalds unsigned long flags; 34831da177e4SLinus Torvalds struct sk_buff *result; 34841da177e4SLinus Torvalds 34851da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 34861da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 34871da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 34881da177e4SLinus Torvalds return result; 34891da177e4SLinus Torvalds } 3490b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 34911da177e4SLinus Torvalds 34921da177e4SLinus Torvalds /** 34931da177e4SLinus Torvalds * skb_queue_purge - empty a list 34941da177e4SLinus Torvalds * @list: list to empty 34951da177e4SLinus Torvalds * 34961da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 34971da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 34981da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 34991da177e4SLinus Torvalds */ 35001da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 35011da177e4SLinus Torvalds { 35021da177e4SLinus Torvalds struct sk_buff *skb; 35031da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 35041da177e4SLinus Torvalds kfree_skb(skb); 35051da177e4SLinus Torvalds } 3506b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 35071da177e4SLinus Torvalds 35081da177e4SLinus Torvalds /** 35099f5afeaeSYaogong Wang * skb_rbtree_purge - empty a skb rbtree 35109f5afeaeSYaogong Wang * @root: root of the rbtree to empty 3511385114deSPeter Oskolkov * Return value: the sum of truesizes of all purged skbs. 35129f5afeaeSYaogong Wang * 35139f5afeaeSYaogong Wang * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 35149f5afeaeSYaogong Wang * the list and one reference dropped. This function does not take 35159f5afeaeSYaogong Wang * any lock. Synchronization should be handled by the caller (e.g., TCP 35169f5afeaeSYaogong Wang * out-of-order queue is protected by the socket lock). 35179f5afeaeSYaogong Wang */ 3518385114deSPeter Oskolkov unsigned int skb_rbtree_purge(struct rb_root *root) 35199f5afeaeSYaogong Wang { 35207c90584cSEric Dumazet struct rb_node *p = rb_first(root); 3521385114deSPeter Oskolkov unsigned int sum = 0; 35229f5afeaeSYaogong Wang 35237c90584cSEric Dumazet while (p) { 35247c90584cSEric Dumazet struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode); 35257c90584cSEric Dumazet 35267c90584cSEric Dumazet p = rb_next(p); 35277c90584cSEric Dumazet rb_erase(&skb->rbnode, root); 3528385114deSPeter Oskolkov sum += skb->truesize; 35299f5afeaeSYaogong Wang kfree_skb(skb); 35307c90584cSEric Dumazet } 3531385114deSPeter Oskolkov return sum; 35329f5afeaeSYaogong Wang } 35339f5afeaeSYaogong Wang 35349f5afeaeSYaogong Wang /** 35351da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 35361da177e4SLinus Torvalds * @list: list to use 35371da177e4SLinus Torvalds * @newsk: buffer to queue 35381da177e4SLinus Torvalds * 35391da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 35401da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 35411da177e4SLinus Torvalds * safely. 35421da177e4SLinus Torvalds * 35431da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 35441da177e4SLinus Torvalds */ 35451da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 35461da177e4SLinus Torvalds { 35471da177e4SLinus Torvalds unsigned long flags; 35481da177e4SLinus Torvalds 35491da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 35501da177e4SLinus Torvalds __skb_queue_head(list, newsk); 35511da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 35521da177e4SLinus Torvalds } 3553b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 35541da177e4SLinus Torvalds 35551da177e4SLinus Torvalds /** 35561da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 35571da177e4SLinus Torvalds * @list: list to use 35581da177e4SLinus Torvalds * @newsk: buffer to queue 35591da177e4SLinus Torvalds * 35601da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 35611da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 35621da177e4SLinus Torvalds * safely. 35631da177e4SLinus Torvalds * 35641da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 35651da177e4SLinus Torvalds */ 35661da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 35671da177e4SLinus Torvalds { 35681da177e4SLinus Torvalds unsigned long flags; 35691da177e4SLinus Torvalds 35701da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 35711da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 35721da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 35731da177e4SLinus Torvalds } 3574b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 35758728b834SDavid S. Miller 35761da177e4SLinus Torvalds /** 35771da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 35781da177e4SLinus Torvalds * @skb: buffer to remove 35798728b834SDavid S. Miller * @list: list to use 35801da177e4SLinus Torvalds * 35818728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 35828728b834SDavid S. Miller * function is atomic with respect to other list locked calls 35831da177e4SLinus Torvalds * 35848728b834SDavid S. Miller * You must know what list the SKB is on. 35851da177e4SLinus Torvalds */ 35868728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 35871da177e4SLinus Torvalds { 35881da177e4SLinus Torvalds unsigned long flags; 35891da177e4SLinus Torvalds 35901da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 35918728b834SDavid S. Miller __skb_unlink(skb, list); 35921da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 35931da177e4SLinus Torvalds } 3594b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 35951da177e4SLinus Torvalds 35961da177e4SLinus Torvalds /** 35971da177e4SLinus Torvalds * skb_append - append a buffer 35981da177e4SLinus Torvalds * @old: buffer to insert after 35991da177e4SLinus Torvalds * @newsk: buffer to insert 36008728b834SDavid S. Miller * @list: list to use 36011da177e4SLinus Torvalds * 36021da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 36031da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 36041da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 36051da177e4SLinus Torvalds */ 36068728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 36071da177e4SLinus Torvalds { 36081da177e4SLinus Torvalds unsigned long flags; 36091da177e4SLinus Torvalds 36108728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 36117de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 36128728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 36131da177e4SLinus Torvalds } 3614b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 36151da177e4SLinus Torvalds 36161da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 36171da177e4SLinus Torvalds struct sk_buff* skb1, 36181da177e4SLinus Torvalds const u32 len, const int pos) 36191da177e4SLinus Torvalds { 36201da177e4SLinus Torvalds int i; 36211da177e4SLinus Torvalds 3622d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 3623d626f62bSArnaldo Carvalho de Melo pos - len); 36241da177e4SLinus Torvalds /* And move data appendix as is. */ 36251da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 36261da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 36271da177e4SLinus Torvalds 36281da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 36291da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 36301da177e4SLinus Torvalds skb1->data_len = skb->data_len; 36311da177e4SLinus Torvalds skb1->len += skb1->data_len; 36321da177e4SLinus Torvalds skb->data_len = 0; 36331da177e4SLinus Torvalds skb->len = len; 363427a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 36351da177e4SLinus Torvalds } 36361da177e4SLinus Torvalds 36371da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 36381da177e4SLinus Torvalds struct sk_buff* skb1, 36391da177e4SLinus Torvalds const u32 len, int pos) 36401da177e4SLinus Torvalds { 36411da177e4SLinus Torvalds int i, k = 0; 36421da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 36431da177e4SLinus Torvalds 36441da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 36451da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 36461da177e4SLinus Torvalds skb->len = len; 36471da177e4SLinus Torvalds skb->data_len = len - pos; 36481da177e4SLinus Torvalds 36491da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 36509e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 36511da177e4SLinus Torvalds 36521da177e4SLinus Torvalds if (pos + size > len) { 36531da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 36541da177e4SLinus Torvalds 36551da177e4SLinus Torvalds if (pos < len) { 36561da177e4SLinus Torvalds /* Split frag. 36571da177e4SLinus Torvalds * We have two variants in this case: 36581da177e4SLinus Torvalds * 1. Move all the frag to the second 36591da177e4SLinus Torvalds * part, if it is possible. F.e. 36601da177e4SLinus Torvalds * this approach is mandatory for TUX, 36611da177e4SLinus Torvalds * where splitting is expensive. 36621da177e4SLinus Torvalds * 2. Split is accurately. We make this. 36631da177e4SLinus Torvalds */ 3664ea2ab693SIan Campbell skb_frag_ref(skb, i); 3665b54c9d5bSJonathan Lemon skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos); 36669e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 36679e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 36681da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 36691da177e4SLinus Torvalds } 36701da177e4SLinus Torvalds k++; 36711da177e4SLinus Torvalds } else 36721da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 36731da177e4SLinus Torvalds pos += size; 36741da177e4SLinus Torvalds } 36751da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 36761da177e4SLinus Torvalds } 36771da177e4SLinus Torvalds 36781da177e4SLinus Torvalds /** 36791da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 36801da177e4SLinus Torvalds * @skb: the buffer to split 36811da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 36821da177e4SLinus Torvalds * @len: new length for skb 36831da177e4SLinus Torvalds */ 36841da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 36851da177e4SLinus Torvalds { 36861da177e4SLinus Torvalds int pos = skb_headlen(skb); 36879b65b17dSTalal Ahmad const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY; 36881da177e4SLinus Torvalds 3689753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 3690753f1ca4SPavel Begunkov 36919b65b17dSTalal Ahmad skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags; 36921f8b977aSWillem de Bruijn skb_zerocopy_clone(skb1, skb, 0); 36931da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 36941da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 36951da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 36961da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 36971da177e4SLinus Torvalds } 3698b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 36991da177e4SLinus Torvalds 37009f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 37019f782db3SIlpo Järvinen * 37029f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 37039f782db3SIlpo Järvinen */ 3704832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 3705832d11c5SIlpo Järvinen { 3706c4777efaSEric Dumazet return skb_unclone_keeptruesize(skb, GFP_ATOMIC); 3707832d11c5SIlpo Järvinen } 3708832d11c5SIlpo Järvinen 3709832d11c5SIlpo Järvinen /** 3710832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 3711832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 3712832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 3713832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 3714832d11c5SIlpo Järvinen * 3715832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 371620e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 3717832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 3718832d11c5SIlpo Järvinen * 3719832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 3720832d11c5SIlpo Järvinen * 3721832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 3722832d11c5SIlpo Järvinen * to have non-paged data as well. 3723832d11c5SIlpo Järvinen * 3724832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 3725832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 3726832d11c5SIlpo Järvinen */ 3727832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3728832d11c5SIlpo Järvinen { 3729832d11c5SIlpo Järvinen int from, to, merge, todo; 3730d8e18a51SMatthew Wilcox (Oracle) skb_frag_t *fragfrom, *fragto; 3731832d11c5SIlpo Järvinen 3732832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 3733f8071cdeSEric Dumazet 3734f8071cdeSEric Dumazet if (skb_headlen(skb)) 3735f8071cdeSEric Dumazet return 0; 37361f8b977aSWillem de Bruijn if (skb_zcopy(tgt) || skb_zcopy(skb)) 37371f8b977aSWillem de Bruijn return 0; 3738832d11c5SIlpo Järvinen 3739832d11c5SIlpo Järvinen todo = shiftlen; 3740832d11c5SIlpo Järvinen from = 0; 3741832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 3742832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3743832d11c5SIlpo Järvinen 3744832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 3745832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 3746832d11c5SIlpo Järvinen */ 3747832d11c5SIlpo Järvinen if (!to || 3748ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3749b54c9d5bSJonathan Lemon skb_frag_off(fragfrom))) { 3750832d11c5SIlpo Järvinen merge = -1; 3751832d11c5SIlpo Järvinen } else { 3752832d11c5SIlpo Järvinen merge = to - 1; 3753832d11c5SIlpo Järvinen 37549e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3755832d11c5SIlpo Järvinen if (todo < 0) { 3756832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 3757832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 3758832d11c5SIlpo Järvinen return 0; 3759832d11c5SIlpo Järvinen 37609f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 37619f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3762832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3763832d11c5SIlpo Järvinen 37649e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 37659e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 3766b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, shiftlen); 3767832d11c5SIlpo Järvinen 3768832d11c5SIlpo Järvinen goto onlymerged; 3769832d11c5SIlpo Järvinen } 3770832d11c5SIlpo Järvinen 3771832d11c5SIlpo Järvinen from++; 3772832d11c5SIlpo Järvinen } 3773832d11c5SIlpo Järvinen 3774832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 3775832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 3776832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3777832d11c5SIlpo Järvinen return 0; 3778832d11c5SIlpo Järvinen 3779832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3780832d11c5SIlpo Järvinen return 0; 3781832d11c5SIlpo Järvinen 3782832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3783832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 3784832d11c5SIlpo Järvinen return 0; 3785832d11c5SIlpo Järvinen 3786832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 3787832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 3788832d11c5SIlpo Järvinen 37899e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 3790832d11c5SIlpo Järvinen *fragto = *fragfrom; 37919e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 3792832d11c5SIlpo Järvinen from++; 3793832d11c5SIlpo Järvinen to++; 3794832d11c5SIlpo Järvinen 3795832d11c5SIlpo Järvinen } else { 3796ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 3797b54c9d5bSJonathan Lemon skb_frag_page_copy(fragto, fragfrom); 3798b54c9d5bSJonathan Lemon skb_frag_off_copy(fragto, fragfrom); 37999e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 3800832d11c5SIlpo Järvinen 3801b54c9d5bSJonathan Lemon skb_frag_off_add(fragfrom, todo); 38029e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 3803832d11c5SIlpo Järvinen todo = 0; 3804832d11c5SIlpo Järvinen 3805832d11c5SIlpo Järvinen to++; 3806832d11c5SIlpo Järvinen break; 3807832d11c5SIlpo Järvinen } 3808832d11c5SIlpo Järvinen } 3809832d11c5SIlpo Järvinen 3810832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 3811832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 3812832d11c5SIlpo Järvinen 3813832d11c5SIlpo Järvinen if (merge >= 0) { 3814832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 3815832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 3816832d11c5SIlpo Järvinen 38179e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 38186a5bcd84SIlias Apalodimas __skb_frag_unref(fragfrom, skb->pp_recycle); 3819832d11c5SIlpo Järvinen } 3820832d11c5SIlpo Järvinen 3821832d11c5SIlpo Järvinen /* Reposition in the original skb */ 3822832d11c5SIlpo Järvinen to = 0; 3823832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 3824832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3825832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 3826832d11c5SIlpo Järvinen 3827832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3828832d11c5SIlpo Järvinen 3829832d11c5SIlpo Järvinen onlymerged: 3830832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 3831832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 3832832d11c5SIlpo Järvinen */ 3833832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 3834832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 3835832d11c5SIlpo Järvinen 3836ede57d58SRichard Gobert skb_len_add(skb, -shiftlen); 3837ede57d58SRichard Gobert skb_len_add(tgt, shiftlen); 3838832d11c5SIlpo Järvinen 3839832d11c5SIlpo Järvinen return shiftlen; 3840832d11c5SIlpo Järvinen } 3841832d11c5SIlpo Järvinen 3842677e90edSThomas Graf /** 3843677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 3844677e90edSThomas Graf * @skb: the buffer to read 3845677e90edSThomas Graf * @from: lower offset of data to be read 3846677e90edSThomas Graf * @to: upper offset of data to be read 3847677e90edSThomas Graf * @st: state variable 3848677e90edSThomas Graf * 3849677e90edSThomas Graf * Initializes the specified state variable. Must be called before 3850677e90edSThomas Graf * invoking skb_seq_read() for the first time. 3851677e90edSThomas Graf */ 3852677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3853677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 3854677e90edSThomas Graf { 3855677e90edSThomas Graf st->lower_offset = from; 3856677e90edSThomas Graf st->upper_offset = to; 3857677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 3858677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 3859677e90edSThomas Graf st->frag_data = NULL; 386097550f6fSWillem de Bruijn st->frag_off = 0; 3861677e90edSThomas Graf } 3862b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 3863677e90edSThomas Graf 3864677e90edSThomas Graf /** 3865677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 3866677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 3867677e90edSThomas Graf * @data: destination pointer for data to be returned 3868677e90edSThomas Graf * @st: state variable 3869677e90edSThomas Graf * 3870bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 3871677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 3872bc32383cSMathias Krause * the head of the data block to @data and returns the length 3873677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 3874677e90edSThomas Graf * offset has been reached. 3875677e90edSThomas Graf * 3876677e90edSThomas Graf * The caller is not required to consume all of the data 3877bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 3878677e90edSThomas Graf * of bytes already consumed and the next call to 3879677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 3880677e90edSThomas Graf * 388125985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 3882e793c0f7SMasanari Iida * this limitation is the cost for zerocopy sequential 3883677e90edSThomas Graf * reads of potentially non linear data. 3884677e90edSThomas Graf * 3885bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 3886677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 3887677e90edSThomas Graf * a stack for this purpose. 3888677e90edSThomas Graf */ 3889677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3890677e90edSThomas Graf struct skb_seq_state *st) 3891677e90edSThomas Graf { 3892677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3893677e90edSThomas Graf skb_frag_t *frag; 3894677e90edSThomas Graf 3895aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 3896aeb193eaSWedson Almeida Filho if (st->frag_data) { 3897aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 3898aeb193eaSWedson Almeida Filho st->frag_data = NULL; 3899aeb193eaSWedson Almeida Filho } 3900677e90edSThomas Graf return 0; 3901aeb193eaSWedson Almeida Filho } 3902677e90edSThomas Graf 3903677e90edSThomas Graf next_skb: 390495e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3905677e90edSThomas Graf 3906995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 390795e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3908677e90edSThomas Graf return block_limit - abs_offset; 3909677e90edSThomas Graf } 3910677e90edSThomas Graf 3911677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 3912677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 3913677e90edSThomas Graf 3914677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 391597550f6fSWillem de Bruijn unsigned int pg_idx, pg_off, pg_sz; 3916677e90edSThomas Graf 391797550f6fSWillem de Bruijn frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 391897550f6fSWillem de Bruijn 391997550f6fSWillem de Bruijn pg_idx = 0; 392097550f6fSWillem de Bruijn pg_off = skb_frag_off(frag); 392197550f6fSWillem de Bruijn pg_sz = skb_frag_size(frag); 392297550f6fSWillem de Bruijn 392397550f6fSWillem de Bruijn if (skb_frag_must_loop(skb_frag_page(frag))) { 392497550f6fSWillem de Bruijn pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT; 392597550f6fSWillem de Bruijn pg_off = offset_in_page(pg_off + st->frag_off); 392697550f6fSWillem de Bruijn pg_sz = min_t(unsigned int, pg_sz - st->frag_off, 392797550f6fSWillem de Bruijn PAGE_SIZE - pg_off); 392897550f6fSWillem de Bruijn } 392997550f6fSWillem de Bruijn 393097550f6fSWillem de Bruijn block_limit = pg_sz + st->stepped_offset; 3931677e90edSThomas Graf if (abs_offset < block_limit) { 3932677e90edSThomas Graf if (!st->frag_data) 393397550f6fSWillem de Bruijn st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx); 3934677e90edSThomas Graf 393597550f6fSWillem de Bruijn *data = (u8 *)st->frag_data + pg_off + 3936677e90edSThomas Graf (abs_offset - st->stepped_offset); 3937677e90edSThomas Graf 3938677e90edSThomas Graf return block_limit - abs_offset; 3939677e90edSThomas Graf } 3940677e90edSThomas Graf 3941677e90edSThomas Graf if (st->frag_data) { 394251c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3943677e90edSThomas Graf st->frag_data = NULL; 3944677e90edSThomas Graf } 3945677e90edSThomas Graf 394697550f6fSWillem de Bruijn st->stepped_offset += pg_sz; 394797550f6fSWillem de Bruijn st->frag_off += pg_sz; 394897550f6fSWillem de Bruijn if (st->frag_off == skb_frag_size(frag)) { 394997550f6fSWillem de Bruijn st->frag_off = 0; 3950677e90edSThomas Graf st->frag_idx++; 395197550f6fSWillem de Bruijn } 3952677e90edSThomas Graf } 3953677e90edSThomas Graf 39545b5a60daSOlaf Kirch if (st->frag_data) { 395551c56b00SEric Dumazet kunmap_atomic(st->frag_data); 39565b5a60daSOlaf Kirch st->frag_data = NULL; 39575b5a60daSOlaf Kirch } 39585b5a60daSOlaf Kirch 395921dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3960677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 396195e3b24cSHerbert Xu st->frag_idx = 0; 3962677e90edSThomas Graf goto next_skb; 396371b3346dSShyam Iyer } else if (st->cur_skb->next) { 396471b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 396571b3346dSShyam Iyer st->frag_idx = 0; 3966677e90edSThomas Graf goto next_skb; 3967677e90edSThomas Graf } 3968677e90edSThomas Graf 3969677e90edSThomas Graf return 0; 3970677e90edSThomas Graf } 3971b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 3972677e90edSThomas Graf 3973677e90edSThomas Graf /** 3974677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 3975677e90edSThomas Graf * @st: state variable 3976677e90edSThomas Graf * 3977677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 3978677e90edSThomas Graf * returned 0. 3979677e90edSThomas Graf */ 3980677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 3981677e90edSThomas Graf { 3982677e90edSThomas Graf if (st->frag_data) 398351c56b00SEric Dumazet kunmap_atomic(st->frag_data); 3984677e90edSThomas Graf } 3985b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 3986677e90edSThomas Graf 39873fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 39883fc7e8a6SThomas Graf 39893fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 39903fc7e8a6SThomas Graf struct ts_config *conf, 39913fc7e8a6SThomas Graf struct ts_state *state) 39923fc7e8a6SThomas Graf { 39933fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 39943fc7e8a6SThomas Graf } 39953fc7e8a6SThomas Graf 39963fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 39973fc7e8a6SThomas Graf { 39983fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 39993fc7e8a6SThomas Graf } 40003fc7e8a6SThomas Graf 40013fc7e8a6SThomas Graf /** 40023fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 40033fc7e8a6SThomas Graf * @skb: the buffer to look in 40043fc7e8a6SThomas Graf * @from: search offset 40053fc7e8a6SThomas Graf * @to: search limit 40063fc7e8a6SThomas Graf * @config: textsearch configuration 40073fc7e8a6SThomas Graf * 40083fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 40093fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 40103fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 40113fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 40123fc7e8a6SThomas Graf */ 40133fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 4014059a2440SBojan Prtvar unsigned int to, struct ts_config *config) 40153fc7e8a6SThomas Graf { 4016059a2440SBojan Prtvar struct ts_state state; 4017f72b948dSPhil Oester unsigned int ret; 4018f72b948dSPhil Oester 4019b228c9b0SWillem de Bruijn BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb)); 4020b228c9b0SWillem de Bruijn 40213fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 40223fc7e8a6SThomas Graf config->finish = skb_ts_finish; 40233fc7e8a6SThomas Graf 4024059a2440SBojan Prtvar skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 40253fc7e8a6SThomas Graf 4026059a2440SBojan Prtvar ret = textsearch_find(config, &state); 4027f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 40283fc7e8a6SThomas Graf } 4029b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 40303fc7e8a6SThomas Graf 4031be12a1feSHannes Frederic Sowa int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 4032be12a1feSHannes Frederic Sowa int offset, size_t size) 4033be12a1feSHannes Frederic Sowa { 4034be12a1feSHannes Frederic Sowa int i = skb_shinfo(skb)->nr_frags; 4035be12a1feSHannes Frederic Sowa 4036be12a1feSHannes Frederic Sowa if (skb_can_coalesce(skb, i, page, offset)) { 4037be12a1feSHannes Frederic Sowa skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 4038be12a1feSHannes Frederic Sowa } else if (i < MAX_SKB_FRAGS) { 4039753f1ca4SPavel Begunkov skb_zcopy_downgrade_managed(skb); 4040be12a1feSHannes Frederic Sowa get_page(page); 4041228ebc41SEric Dumazet skb_fill_page_desc_noacc(skb, i, page, offset, size); 4042be12a1feSHannes Frederic Sowa } else { 4043be12a1feSHannes Frederic Sowa return -EMSGSIZE; 4044be12a1feSHannes Frederic Sowa } 4045be12a1feSHannes Frederic Sowa 4046be12a1feSHannes Frederic Sowa return 0; 4047be12a1feSHannes Frederic Sowa } 4048be12a1feSHannes Frederic Sowa EXPORT_SYMBOL_GPL(skb_append_pagefrags); 4049be12a1feSHannes Frederic Sowa 4050cbb042f9SHerbert Xu /** 4051cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 4052cbb042f9SHerbert Xu * @skb: buffer to update 4053cbb042f9SHerbert Xu * @len: length of data pulled 4054cbb042f9SHerbert Xu * 4055cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 4056fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 405784fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 405884fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 405984fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 4060cbb042f9SHerbert Xu */ 4061af72868bSJohannes Berg void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 4062cbb042f9SHerbert Xu { 406331b33dfbSPravin B Shelar unsigned char *data = skb->data; 406431b33dfbSPravin B Shelar 4065cbb042f9SHerbert Xu BUG_ON(len > skb->len); 406631b33dfbSPravin B Shelar __skb_pull(skb, len); 406731b33dfbSPravin B Shelar skb_postpull_rcsum(skb, data, len); 406831b33dfbSPravin B Shelar return skb->data; 4069cbb042f9SHerbert Xu } 4070f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 4071f94691acSArnaldo Carvalho de Melo 407213acc94eSYonghong Song static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb) 407313acc94eSYonghong Song { 407413acc94eSYonghong Song skb_frag_t head_frag; 407513acc94eSYonghong Song struct page *page; 407613acc94eSYonghong Song 407713acc94eSYonghong Song page = virt_to_head_page(frag_skb->head); 4078d8e18a51SMatthew Wilcox (Oracle) __skb_frag_set_page(&head_frag, page); 4079b54c9d5bSJonathan Lemon skb_frag_off_set(&head_frag, frag_skb->data - 4080b54c9d5bSJonathan Lemon (unsigned char *)page_address(page)); 4081d8e18a51SMatthew Wilcox (Oracle) skb_frag_size_set(&head_frag, skb_headlen(frag_skb)); 408213acc94eSYonghong Song return head_frag; 408313acc94eSYonghong Song } 408413acc94eSYonghong Song 40853a1296a3SSteffen Klassert struct sk_buff *skb_segment_list(struct sk_buff *skb, 40863a1296a3SSteffen Klassert netdev_features_t features, 40873a1296a3SSteffen Klassert unsigned int offset) 40883a1296a3SSteffen Klassert { 40893a1296a3SSteffen Klassert struct sk_buff *list_skb = skb_shinfo(skb)->frag_list; 40903a1296a3SSteffen Klassert unsigned int tnl_hlen = skb_tnl_header_len(skb); 40913a1296a3SSteffen Klassert unsigned int delta_truesize = 0; 40923a1296a3SSteffen Klassert unsigned int delta_len = 0; 40933a1296a3SSteffen Klassert struct sk_buff *tail = NULL; 409453475c5dSDongseok Yi struct sk_buff *nskb, *tmp; 4095cf3ab8d4SLina Wang int len_diff, err; 40963a1296a3SSteffen Klassert 40973a1296a3SSteffen Klassert skb_push(skb, -skb_network_offset(skb) + offset); 40983a1296a3SSteffen Klassert 40993a1296a3SSteffen Klassert skb_shinfo(skb)->frag_list = NULL; 41003a1296a3SSteffen Klassert 41013a1296a3SSteffen Klassert do { 41023a1296a3SSteffen Klassert nskb = list_skb; 41033a1296a3SSteffen Klassert list_skb = list_skb->next; 41043a1296a3SSteffen Klassert 410553475c5dSDongseok Yi err = 0; 4106224102deSlena wang delta_truesize += nskb->truesize; 410753475c5dSDongseok Yi if (skb_shared(nskb)) { 410853475c5dSDongseok Yi tmp = skb_clone(nskb, GFP_ATOMIC); 410953475c5dSDongseok Yi if (tmp) { 411053475c5dSDongseok Yi consume_skb(nskb); 411153475c5dSDongseok Yi nskb = tmp; 411253475c5dSDongseok Yi err = skb_unclone(nskb, GFP_ATOMIC); 411353475c5dSDongseok Yi } else { 411453475c5dSDongseok Yi err = -ENOMEM; 411553475c5dSDongseok Yi } 411653475c5dSDongseok Yi } 411753475c5dSDongseok Yi 41183a1296a3SSteffen Klassert if (!tail) 41193a1296a3SSteffen Klassert skb->next = nskb; 41203a1296a3SSteffen Klassert else 41213a1296a3SSteffen Klassert tail->next = nskb; 41223a1296a3SSteffen Klassert 412353475c5dSDongseok Yi if (unlikely(err)) { 412453475c5dSDongseok Yi nskb->next = list_skb; 412553475c5dSDongseok Yi goto err_linearize; 412653475c5dSDongseok Yi } 412753475c5dSDongseok Yi 41283a1296a3SSteffen Klassert tail = nskb; 41293a1296a3SSteffen Klassert 41303a1296a3SSteffen Klassert delta_len += nskb->len; 41313a1296a3SSteffen Klassert 41323a1296a3SSteffen Klassert skb_push(nskb, -skb_network_offset(nskb) + offset); 41333a1296a3SSteffen Klassert 4134cf673ed0SFlorian Westphal skb_release_head_state(nskb); 4135cf3ab8d4SLina Wang len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb); 41363a1296a3SSteffen Klassert __copy_skb_header(nskb, skb); 41373a1296a3SSteffen Klassert 41383a1296a3SSteffen Klassert skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb)); 4139cf3ab8d4SLina Wang nskb->transport_header += len_diff; 41403a1296a3SSteffen Klassert skb_copy_from_linear_data_offset(skb, -tnl_hlen, 41413a1296a3SSteffen Klassert nskb->data - tnl_hlen, 41423a1296a3SSteffen Klassert offset + tnl_hlen); 41433a1296a3SSteffen Klassert 41443a1296a3SSteffen Klassert if (skb_needs_linearize(nskb, features) && 41453a1296a3SSteffen Klassert __skb_linearize(nskb)) 41463a1296a3SSteffen Klassert goto err_linearize; 41473a1296a3SSteffen Klassert 41483a1296a3SSteffen Klassert } while (list_skb); 41493a1296a3SSteffen Klassert 41503a1296a3SSteffen Klassert skb->truesize = skb->truesize - delta_truesize; 41513a1296a3SSteffen Klassert skb->data_len = skb->data_len - delta_len; 41523a1296a3SSteffen Klassert skb->len = skb->len - delta_len; 41533a1296a3SSteffen Klassert 41543a1296a3SSteffen Klassert skb_gso_reset(skb); 41553a1296a3SSteffen Klassert 41563a1296a3SSteffen Klassert skb->prev = tail; 41573a1296a3SSteffen Klassert 41583a1296a3SSteffen Klassert if (skb_needs_linearize(skb, features) && 41593a1296a3SSteffen Klassert __skb_linearize(skb)) 41603a1296a3SSteffen Klassert goto err_linearize; 41613a1296a3SSteffen Klassert 41623a1296a3SSteffen Klassert skb_get(skb); 41633a1296a3SSteffen Klassert 41643a1296a3SSteffen Klassert return skb; 41653a1296a3SSteffen Klassert 41663a1296a3SSteffen Klassert err_linearize: 41673a1296a3SSteffen Klassert kfree_skb_list(skb->next); 41683a1296a3SSteffen Klassert skb->next = NULL; 41693a1296a3SSteffen Klassert return ERR_PTR(-ENOMEM); 41703a1296a3SSteffen Klassert } 41713a1296a3SSteffen Klassert EXPORT_SYMBOL_GPL(skb_segment_list); 41723a1296a3SSteffen Klassert 4173f4c50d99SHerbert Xu /** 4174f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 4175df5771ffSMichael S. Tsirkin * @head_skb: buffer to segment 4176576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 4177f4c50d99SHerbert Xu * 4178f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 41794c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 41804c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 4181f4c50d99SHerbert Xu */ 4182df5771ffSMichael S. Tsirkin struct sk_buff *skb_segment(struct sk_buff *head_skb, 4183df5771ffSMichael S. Tsirkin netdev_features_t features) 4184f4c50d99SHerbert Xu { 4185f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 4186f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 41871a4cedafSMichael S. Tsirkin struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 4188df5771ffSMichael S. Tsirkin skb_frag_t *frag = skb_shinfo(head_skb)->frags; 4189df5771ffSMichael S. Tsirkin unsigned int mss = skb_shinfo(head_skb)->gso_size; 4190df5771ffSMichael S. Tsirkin unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 41911fd819ecSMichael S. Tsirkin struct sk_buff *frag_skb = head_skb; 4192f4c50d99SHerbert Xu unsigned int offset = doffset; 4193df5771ffSMichael S. Tsirkin unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 4194802ab55aSAlexander Duyck unsigned int partial_segs = 0; 4195f4c50d99SHerbert Xu unsigned int headroom; 4196802ab55aSAlexander Duyck unsigned int len = head_skb->len; 4197ec5f0615SPravin B Shelar __be16 proto; 419836c98382SAlexander Duyck bool csum, sg; 4199df5771ffSMichael S. Tsirkin int nfrags = skb_shinfo(head_skb)->nr_frags; 4200f4c50d99SHerbert Xu int err = -ENOMEM; 4201f4c50d99SHerbert Xu int i = 0; 4202f4c50d99SHerbert Xu int pos; 4203f4c50d99SHerbert Xu 42049e4b7a99SJiri Benc if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) && 42059e4b7a99SJiri Benc mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) { 42069e4b7a99SJiri Benc struct sk_buff *check_skb; 42079e4b7a99SJiri Benc 42089e4b7a99SJiri Benc for (check_skb = list_skb; check_skb; check_skb = check_skb->next) { 42099e4b7a99SJiri Benc if (skb_headlen(check_skb) && !check_skb->head_frag) { 42109e4b7a99SJiri Benc /* gso_size is untrusted, and we have a frag_list with 42119e4b7a99SJiri Benc * a linear non head_frag item. 42123dcbdb13SShmulik Ladkani * 42139e4b7a99SJiri Benc * If head_skb's headlen does not fit requested gso_size, 42149e4b7a99SJiri Benc * it means that the frag_list members do NOT terminate 42159e4b7a99SJiri Benc * on exact gso_size boundaries. Hence we cannot perform 42169e4b7a99SJiri Benc * skb_frag_t page sharing. Therefore we must fallback to 42179e4b7a99SJiri Benc * copying the frag_list skbs; we do so by disabling SG. 42183dcbdb13SShmulik Ladkani */ 42193dcbdb13SShmulik Ladkani features &= ~NETIF_F_SG; 42209e4b7a99SJiri Benc break; 42219e4b7a99SJiri Benc } 42229e4b7a99SJiri Benc } 42233dcbdb13SShmulik Ladkani } 42243dcbdb13SShmulik Ladkani 42255882a07cSWei-Chun Chao __skb_push(head_skb, doffset); 42262f631133SMiaohe Lin proto = skb_network_protocol(head_skb, NULL); 4227ec5f0615SPravin B Shelar if (unlikely(!proto)) 4228ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 4229ec5f0615SPravin B Shelar 423036c98382SAlexander Duyck sg = !!(features & NETIF_F_SG); 4231f245d079SAlexander Duyck csum = !!can_checksum_protocol(features, proto); 42327e2b10c1STom Herbert 423307b26c94SSteffen Klassert if (sg && csum && (mss != GSO_BY_FRAGS)) { 423407b26c94SSteffen Klassert if (!(features & NETIF_F_GSO_PARTIAL)) { 423507b26c94SSteffen Klassert struct sk_buff *iter; 423643170c4eSIlan Tayari unsigned int frag_len; 423707b26c94SSteffen Klassert 423807b26c94SSteffen Klassert if (!list_skb || 423907b26c94SSteffen Klassert !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 424007b26c94SSteffen Klassert goto normal; 424107b26c94SSteffen Klassert 424243170c4eSIlan Tayari /* If we get here then all the required 424343170c4eSIlan Tayari * GSO features except frag_list are supported. 424443170c4eSIlan Tayari * Try to split the SKB to multiple GSO SKBs 424543170c4eSIlan Tayari * with no frag_list. 424643170c4eSIlan Tayari * Currently we can do that only when the buffers don't 424743170c4eSIlan Tayari * have a linear part and all the buffers except 424843170c4eSIlan Tayari * the last are of the same length. 424907b26c94SSteffen Klassert */ 425043170c4eSIlan Tayari frag_len = list_skb->len; 425107b26c94SSteffen Klassert skb_walk_frags(head_skb, iter) { 425243170c4eSIlan Tayari if (frag_len != iter->len && iter->next) 425343170c4eSIlan Tayari goto normal; 4254eaffadbbSIlan Tayari if (skb_headlen(iter) && !iter->head_frag) 425507b26c94SSteffen Klassert goto normal; 425607b26c94SSteffen Klassert 425707b26c94SSteffen Klassert len -= iter->len; 425807b26c94SSteffen Klassert } 425943170c4eSIlan Tayari 426043170c4eSIlan Tayari if (len != frag_len) 426143170c4eSIlan Tayari goto normal; 426207b26c94SSteffen Klassert } 426307b26c94SSteffen Klassert 4264802ab55aSAlexander Duyck /* GSO partial only requires that we trim off any excess that 4265802ab55aSAlexander Duyck * doesn't fit into an MSS sized block, so take care of that 4266802ab55aSAlexander Duyck * now. 4267802ab55aSAlexander Duyck */ 4268802ab55aSAlexander Duyck partial_segs = len / mss; 4269d7fb5a80SAlexander Duyck if (partial_segs > 1) 4270802ab55aSAlexander Duyck mss *= partial_segs; 4271d7fb5a80SAlexander Duyck else 4272d7fb5a80SAlexander Duyck partial_segs = 0; 4273802ab55aSAlexander Duyck } 4274802ab55aSAlexander Duyck 427507b26c94SSteffen Klassert normal: 4276df5771ffSMichael S. Tsirkin headroom = skb_headroom(head_skb); 4277df5771ffSMichael S. Tsirkin pos = skb_headlen(head_skb); 4278f4c50d99SHerbert Xu 4279f4c50d99SHerbert Xu do { 4280f4c50d99SHerbert Xu struct sk_buff *nskb; 42818cb19905SMichael S. Tsirkin skb_frag_t *nskb_frag; 4282c8884eddSHerbert Xu int hsize; 4283f4c50d99SHerbert Xu int size; 4284f4c50d99SHerbert Xu 42853953c46cSMarcelo Ricardo Leitner if (unlikely(mss == GSO_BY_FRAGS)) { 42863953c46cSMarcelo Ricardo Leitner len = list_skb->len; 42873953c46cSMarcelo Ricardo Leitner } else { 4288df5771ffSMichael S. Tsirkin len = head_skb->len - offset; 4289f4c50d99SHerbert Xu if (len > mss) 4290f4c50d99SHerbert Xu len = mss; 42913953c46cSMarcelo Ricardo Leitner } 4292f4c50d99SHerbert Xu 4293df5771ffSMichael S. Tsirkin hsize = skb_headlen(head_skb) - offset; 4294f4c50d99SHerbert Xu 4295dbd50f23SXin Long if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) && 42961a4cedafSMichael S. Tsirkin (skb_headlen(list_skb) == len || sg)) { 42971a4cedafSMichael S. Tsirkin BUG_ON(skb_headlen(list_skb) > len); 429889319d38SHerbert Xu 42999d8506ccSHerbert Xu i = 0; 43001a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 43011a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 43021fd819ecSMichael S. Tsirkin frag_skb = list_skb; 43031a4cedafSMichael S. Tsirkin pos += skb_headlen(list_skb); 43049d8506ccSHerbert Xu 43059d8506ccSHerbert Xu while (pos < offset + len) { 43069d8506ccSHerbert Xu BUG_ON(i >= nfrags); 43079d8506ccSHerbert Xu 43084e1beba1SMichael S. Tsirkin size = skb_frag_size(frag); 43099d8506ccSHerbert Xu if (pos + size > offset + len) 43109d8506ccSHerbert Xu break; 43119d8506ccSHerbert Xu 43129d8506ccSHerbert Xu i++; 43139d8506ccSHerbert Xu pos += size; 43144e1beba1SMichael S. Tsirkin frag++; 43159d8506ccSHerbert Xu } 43169d8506ccSHerbert Xu 43171a4cedafSMichael S. Tsirkin nskb = skb_clone(list_skb, GFP_ATOMIC); 43181a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 431989319d38SHerbert Xu 4320f4c50d99SHerbert Xu if (unlikely(!nskb)) 4321f4c50d99SHerbert Xu goto err; 4322f4c50d99SHerbert Xu 43239d8506ccSHerbert Xu if (unlikely(pskb_trim(nskb, len))) { 43249d8506ccSHerbert Xu kfree_skb(nskb); 43259d8506ccSHerbert Xu goto err; 43269d8506ccSHerbert Xu } 43279d8506ccSHerbert Xu 4328ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 432989319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 433089319d38SHerbert Xu kfree_skb(nskb); 433189319d38SHerbert Xu goto err; 433289319d38SHerbert Xu } 433389319d38SHerbert Xu 4334ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 433589319d38SHerbert Xu skb_release_head_state(nskb); 433689319d38SHerbert Xu __skb_push(nskb, doffset); 433789319d38SHerbert Xu } else { 433800b229f7SPaolo Abeni if (hsize < 0) 433900b229f7SPaolo Abeni hsize = 0; 4340dbd50f23SXin Long if (hsize > len || !sg) 4341dbd50f23SXin Long hsize = len; 4342dbd50f23SXin Long 4343c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 4344df5771ffSMichael S. Tsirkin GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 4345c93bdd0eSMel Gorman NUMA_NO_NODE); 434689319d38SHerbert Xu 434789319d38SHerbert Xu if (unlikely(!nskb)) 434889319d38SHerbert Xu goto err; 434989319d38SHerbert Xu 435089319d38SHerbert Xu skb_reserve(nskb, headroom); 435189319d38SHerbert Xu __skb_put(nskb, doffset); 435289319d38SHerbert Xu } 435389319d38SHerbert Xu 4354f4c50d99SHerbert Xu if (segs) 4355f4c50d99SHerbert Xu tail->next = nskb; 4356f4c50d99SHerbert Xu else 4357f4c50d99SHerbert Xu segs = nskb; 4358f4c50d99SHerbert Xu tail = nskb; 4359f4c50d99SHerbert Xu 4360df5771ffSMichael S. Tsirkin __copy_skb_header(nskb, head_skb); 4361f4c50d99SHerbert Xu 4362030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 4363fcdfe3a7SVlad Yasevich skb_reset_mac_len(nskb); 436468c33163SPravin B Shelar 4365df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 436668c33163SPravin B Shelar nskb->data - tnl_hlen, 436768c33163SPravin B Shelar doffset + tnl_hlen); 436889319d38SHerbert Xu 43699d8506ccSHerbert Xu if (nskb->len == len + doffset) 43701cdbcb79SSimon Horman goto perform_csum_check; 437189319d38SHerbert Xu 43727fbeffedSAlexander Duyck if (!sg) { 43731454c9faSYadu Kishore if (!csum) { 43747fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 43756f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 437676443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 437776443456SAlexander Duyck skb_copy_and_csum_bits(head_skb, offset, 43781454c9faSYadu Kishore skb_put(nskb, 43791454c9faSYadu Kishore len), 43808d5930dfSAl Viro len); 43817e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 4382de843723STom Herbert skb_headroom(nskb) + doffset; 43831454c9faSYadu Kishore } else { 4384c624c58eSlily if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len)) 4385c624c58eSlily goto err; 43861454c9faSYadu Kishore } 4387f4c50d99SHerbert Xu continue; 4388f4c50d99SHerbert Xu } 4389f4c50d99SHerbert Xu 43908cb19905SMichael S. Tsirkin nskb_frag = skb_shinfo(nskb)->frags; 4391f4c50d99SHerbert Xu 4392df5771ffSMichael S. Tsirkin skb_copy_from_linear_data_offset(head_skb, offset, 4393d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 4394f4c50d99SHerbert Xu 439506b4feb3SJonathan Lemon skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags & 439606b4feb3SJonathan Lemon SKBFL_SHARED_FRAG; 4397cef401deSEric Dumazet 4398bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4399bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC)) 4400bf5c25d6SWillem de Bruijn goto err; 4401bf5c25d6SWillem de Bruijn 44029d8506ccSHerbert Xu while (pos < offset + len) { 44039d8506ccSHerbert Xu if (i >= nfrags) { 44049d8506ccSHerbert Xu i = 0; 44051a4cedafSMichael S. Tsirkin nfrags = skb_shinfo(list_skb)->nr_frags; 44061a4cedafSMichael S. Tsirkin frag = skb_shinfo(list_skb)->frags; 44071fd819ecSMichael S. Tsirkin frag_skb = list_skb; 440813acc94eSYonghong Song if (!skb_headlen(list_skb)) { 44099d8506ccSHerbert Xu BUG_ON(!nfrags); 441013acc94eSYonghong Song } else { 441113acc94eSYonghong Song BUG_ON(!list_skb->head_frag); 44129d8506ccSHerbert Xu 441313acc94eSYonghong Song /* to make room for head_frag. */ 441413acc94eSYonghong Song i--; 441513acc94eSYonghong Song frag--; 441613acc94eSYonghong Song } 4417bf5c25d6SWillem de Bruijn if (skb_orphan_frags(frag_skb, GFP_ATOMIC) || 4418bf5c25d6SWillem de Bruijn skb_zerocopy_clone(nskb, frag_skb, 4419bf5c25d6SWillem de Bruijn GFP_ATOMIC)) 4420bf5c25d6SWillem de Bruijn goto err; 4421bf5c25d6SWillem de Bruijn 44221a4cedafSMichael S. Tsirkin list_skb = list_skb->next; 44239d8506ccSHerbert Xu } 44249d8506ccSHerbert Xu 44259d8506ccSHerbert Xu if (unlikely(skb_shinfo(nskb)->nr_frags >= 44269d8506ccSHerbert Xu MAX_SKB_FRAGS)) { 44279d8506ccSHerbert Xu net_warn_ratelimited( 44289d8506ccSHerbert Xu "skb_segment: too many frags: %u %u\n", 44299d8506ccSHerbert Xu pos, mss); 4430ff907a11SEric Dumazet err = -EINVAL; 44319d8506ccSHerbert Xu goto err; 44329d8506ccSHerbert Xu } 44339d8506ccSHerbert Xu 443413acc94eSYonghong Song *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag; 44358cb19905SMichael S. Tsirkin __skb_frag_ref(nskb_frag); 44368cb19905SMichael S. Tsirkin size = skb_frag_size(nskb_frag); 4437f4c50d99SHerbert Xu 4438f4c50d99SHerbert Xu if (pos < offset) { 4439b54c9d5bSJonathan Lemon skb_frag_off_add(nskb_frag, offset - pos); 44408cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, offset - pos); 4441f4c50d99SHerbert Xu } 4442f4c50d99SHerbert Xu 444389319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 4444f4c50d99SHerbert Xu 4445f4c50d99SHerbert Xu if (pos + size <= offset + len) { 4446f4c50d99SHerbert Xu i++; 44474e1beba1SMichael S. Tsirkin frag++; 4448f4c50d99SHerbert Xu pos += size; 4449f4c50d99SHerbert Xu } else { 44508cb19905SMichael S. Tsirkin skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 445189319d38SHerbert Xu goto skip_fraglist; 4452f4c50d99SHerbert Xu } 4453f4c50d99SHerbert Xu 44548cb19905SMichael S. Tsirkin nskb_frag++; 4455f4c50d99SHerbert Xu } 4456f4c50d99SHerbert Xu 445789319d38SHerbert Xu skip_fraglist: 4458f4c50d99SHerbert Xu nskb->data_len = len - hsize; 4459f4c50d99SHerbert Xu nskb->len += nskb->data_len; 4460f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 4461ec5f0615SPravin B Shelar 44621cdbcb79SSimon Horman perform_csum_check: 44637fbeffedSAlexander Duyck if (!csum) { 4464ff907a11SEric Dumazet if (skb_has_shared_frag(nskb) && 4465ff907a11SEric Dumazet __skb_linearize(nskb)) 4466ddff00d4SAlexander Duyck goto err; 4467ff907a11SEric Dumazet 44687fbeffedSAlexander Duyck if (!nskb->remcsum_offload) 4469ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 447076443456SAlexander Duyck SKB_GSO_CB(nskb)->csum = 447176443456SAlexander Duyck skb_checksum(nskb, doffset, 447276443456SAlexander Duyck nskb->len - doffset, 0); 44737e2b10c1STom Herbert SKB_GSO_CB(nskb)->csum_start = 44747e2b10c1STom Herbert skb_headroom(nskb) + doffset; 4475ec5f0615SPravin B Shelar } 4476df5771ffSMichael S. Tsirkin } while ((offset += len) < head_skb->len); 4477f4c50d99SHerbert Xu 4478bec3cfdcSEric Dumazet /* Some callers want to get the end of the list. 4479bec3cfdcSEric Dumazet * Put it in segs->prev to avoid walking the list. 4480bec3cfdcSEric Dumazet * (see validate_xmit_skb_list() for example) 4481bec3cfdcSEric Dumazet */ 4482bec3cfdcSEric Dumazet segs->prev = tail; 4483432c856fSToshiaki Makita 4484802ab55aSAlexander Duyck if (partial_segs) { 448507b26c94SSteffen Klassert struct sk_buff *iter; 4486802ab55aSAlexander Duyck int type = skb_shinfo(head_skb)->gso_type; 448707b26c94SSteffen Klassert unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 4488802ab55aSAlexander Duyck 4489802ab55aSAlexander Duyck /* Update type to add partial and then remove dodgy if set */ 449007b26c94SSteffen Klassert type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 4491802ab55aSAlexander Duyck type &= ~SKB_GSO_DODGY; 4492802ab55aSAlexander Duyck 4493802ab55aSAlexander Duyck /* Update GSO info and prepare to start updating headers on 4494802ab55aSAlexander Duyck * our way back down the stack of protocols. 4495802ab55aSAlexander Duyck */ 449607b26c94SSteffen Klassert for (iter = segs; iter; iter = iter->next) { 449707b26c94SSteffen Klassert skb_shinfo(iter)->gso_size = gso_size; 449807b26c94SSteffen Klassert skb_shinfo(iter)->gso_segs = partial_segs; 449907b26c94SSteffen Klassert skb_shinfo(iter)->gso_type = type; 450007b26c94SSteffen Klassert SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 450107b26c94SSteffen Klassert } 450207b26c94SSteffen Klassert 450307b26c94SSteffen Klassert if (tail->len - doffset <= gso_size) 450407b26c94SSteffen Klassert skb_shinfo(tail)->gso_size = 0; 450507b26c94SSteffen Klassert else if (tail != segs) 450607b26c94SSteffen Klassert skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 4507802ab55aSAlexander Duyck } 4508802ab55aSAlexander Duyck 4509432c856fSToshiaki Makita /* Following permits correct backpressure, for protocols 4510432c856fSToshiaki Makita * using skb_set_owner_w(). 4511432c856fSToshiaki Makita * Idea is to tranfert ownership from head_skb to last segment. 4512432c856fSToshiaki Makita */ 4513432c856fSToshiaki Makita if (head_skb->destructor == sock_wfree) { 4514432c856fSToshiaki Makita swap(tail->truesize, head_skb->truesize); 4515432c856fSToshiaki Makita swap(tail->destructor, head_skb->destructor); 4516432c856fSToshiaki Makita swap(tail->sk, head_skb->sk); 4517432c856fSToshiaki Makita } 4518f4c50d99SHerbert Xu return segs; 4519f4c50d99SHerbert Xu 4520f4c50d99SHerbert Xu err: 4521289dccbeSEric Dumazet kfree_skb_list(segs); 4522f4c50d99SHerbert Xu return ERR_PTR(err); 4523f4c50d99SHerbert Xu } 4524f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 4525f4c50d99SHerbert Xu 4526df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 4527df5042f4SFlorian Westphal #define SKB_EXT_ALIGN_VALUE 8 4528df5042f4SFlorian Westphal #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE) 4529df5042f4SFlorian Westphal 4530df5042f4SFlorian Westphal static const u8 skb_ext_type_len[] = { 4531df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4532df5042f4SFlorian Westphal [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info), 4533df5042f4SFlorian Westphal #endif 45344165079bSFlorian Westphal #ifdef CONFIG_XFRM 45354165079bSFlorian Westphal [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path), 45364165079bSFlorian Westphal #endif 453795a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 453895a7233cSPaul Blakey [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext), 453995a7233cSPaul Blakey #endif 45403ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 45413ee17bc7SMat Martineau [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext), 45423ee17bc7SMat Martineau #endif 454378476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 454478476d31SJeremy Kerr [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow), 454578476d31SJeremy Kerr #endif 4546df5042f4SFlorian Westphal }; 4547df5042f4SFlorian Westphal 4548df5042f4SFlorian Westphal static __always_inline unsigned int skb_ext_total_length(void) 4549df5042f4SFlorian Westphal { 4550df5042f4SFlorian Westphal return SKB_EXT_CHUNKSIZEOF(struct skb_ext) + 4551df5042f4SFlorian Westphal #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4552df5042f4SFlorian Westphal skb_ext_type_len[SKB_EXT_BRIDGE_NF] + 4553df5042f4SFlorian Westphal #endif 45544165079bSFlorian Westphal #ifdef CONFIG_XFRM 45554165079bSFlorian Westphal skb_ext_type_len[SKB_EXT_SEC_PATH] + 45564165079bSFlorian Westphal #endif 455795a7233cSPaul Blakey #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 455895a7233cSPaul Blakey skb_ext_type_len[TC_SKB_EXT] + 455995a7233cSPaul Blakey #endif 45603ee17bc7SMat Martineau #if IS_ENABLED(CONFIG_MPTCP) 45613ee17bc7SMat Martineau skb_ext_type_len[SKB_EXT_MPTCP] + 45623ee17bc7SMat Martineau #endif 456378476d31SJeremy Kerr #if IS_ENABLED(CONFIG_MCTP_FLOWS) 456478476d31SJeremy Kerr skb_ext_type_len[SKB_EXT_MCTP] + 456578476d31SJeremy Kerr #endif 4566df5042f4SFlorian Westphal 0; 4567df5042f4SFlorian Westphal } 4568df5042f4SFlorian Westphal 4569df5042f4SFlorian Westphal static void skb_extensions_init(void) 4570df5042f4SFlorian Westphal { 4571df5042f4SFlorian Westphal BUILD_BUG_ON(SKB_EXT_NUM >= 8); 4572df5042f4SFlorian Westphal BUILD_BUG_ON(skb_ext_total_length() > 255); 4573df5042f4SFlorian Westphal 4574df5042f4SFlorian Westphal skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache", 4575df5042f4SFlorian Westphal SKB_EXT_ALIGN_VALUE * skb_ext_total_length(), 4576df5042f4SFlorian Westphal 0, 4577df5042f4SFlorian Westphal SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4578df5042f4SFlorian Westphal NULL); 4579df5042f4SFlorian Westphal } 4580df5042f4SFlorian Westphal #else 4581df5042f4SFlorian Westphal static void skb_extensions_init(void) {} 4582df5042f4SFlorian Westphal #endif 4583df5042f4SFlorian Westphal 45841da177e4SLinus Torvalds void __init skb_init(void) 45851da177e4SLinus Torvalds { 458679a8a642SKees Cook skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache", 45871da177e4SLinus Torvalds sizeof(struct sk_buff), 45881da177e4SLinus Torvalds 0, 4589e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 459079a8a642SKees Cook offsetof(struct sk_buff, cb), 459179a8a642SKees Cook sizeof_field(struct sk_buff, cb), 459220c2df83SPaul Mundt NULL); 4593d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 4594d0bf4a9eSEric Dumazet sizeof(struct sk_buff_fclones), 4595d179cd12SDavid S. Miller 0, 4596e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 459720c2df83SPaul Mundt NULL); 4598df5042f4SFlorian Westphal skb_extensions_init(); 45991da177e4SLinus Torvalds } 46001da177e4SLinus Torvalds 460151c739d1SDavid S. Miller static int 460248a1df65SJason A. Donenfeld __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 460348a1df65SJason A. Donenfeld unsigned int recursion_level) 4604716ea3a7SDavid Howells { 46051a028e50SDavid S. Miller int start = skb_headlen(skb); 46061a028e50SDavid S. Miller int i, copy = start - offset; 4607fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 4608716ea3a7SDavid Howells int elt = 0; 4609716ea3a7SDavid Howells 461048a1df65SJason A. Donenfeld if (unlikely(recursion_level >= 24)) 461148a1df65SJason A. Donenfeld return -EMSGSIZE; 461248a1df65SJason A. Donenfeld 4613716ea3a7SDavid Howells if (copy > 0) { 4614716ea3a7SDavid Howells if (copy > len) 4615716ea3a7SDavid Howells copy = len; 4616642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 4617716ea3a7SDavid Howells elt++; 4618716ea3a7SDavid Howells if ((len -= copy) == 0) 4619716ea3a7SDavid Howells return elt; 4620716ea3a7SDavid Howells offset += copy; 4621716ea3a7SDavid Howells } 4622716ea3a7SDavid Howells 4623716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 46241a028e50SDavid S. Miller int end; 4625716ea3a7SDavid Howells 4626547b792cSIlpo Järvinen WARN_ON(start > offset + len); 46271a028e50SDavid S. Miller 46289e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 4629716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 4630716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 463148a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 463248a1df65SJason A. Donenfeld return -EMSGSIZE; 4633716ea3a7SDavid Howells 4634716ea3a7SDavid Howells if (copy > len) 4635716ea3a7SDavid Howells copy = len; 4636ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 4637b54c9d5bSJonathan Lemon skb_frag_off(frag) + offset - start); 4638716ea3a7SDavid Howells elt++; 4639716ea3a7SDavid Howells if (!(len -= copy)) 4640716ea3a7SDavid Howells return elt; 4641716ea3a7SDavid Howells offset += copy; 4642716ea3a7SDavid Howells } 46431a028e50SDavid S. Miller start = end; 4644716ea3a7SDavid Howells } 4645716ea3a7SDavid Howells 4646fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 464748a1df65SJason A. Donenfeld int end, ret; 4648716ea3a7SDavid Howells 4649547b792cSIlpo Järvinen WARN_ON(start > offset + len); 46501a028e50SDavid S. Miller 4651fbb398a8SDavid S. Miller end = start + frag_iter->len; 4652716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 465348a1df65SJason A. Donenfeld if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 465448a1df65SJason A. Donenfeld return -EMSGSIZE; 465548a1df65SJason A. Donenfeld 4656716ea3a7SDavid Howells if (copy > len) 4657716ea3a7SDavid Howells copy = len; 465848a1df65SJason A. Donenfeld ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 465948a1df65SJason A. Donenfeld copy, recursion_level + 1); 466048a1df65SJason A. Donenfeld if (unlikely(ret < 0)) 466148a1df65SJason A. Donenfeld return ret; 466248a1df65SJason A. Donenfeld elt += ret; 4663716ea3a7SDavid Howells if ((len -= copy) == 0) 4664716ea3a7SDavid Howells return elt; 4665716ea3a7SDavid Howells offset += copy; 4666716ea3a7SDavid Howells } 46671a028e50SDavid S. Miller start = end; 4668716ea3a7SDavid Howells } 4669716ea3a7SDavid Howells BUG_ON(len); 4670716ea3a7SDavid Howells return elt; 4671716ea3a7SDavid Howells } 4672716ea3a7SDavid Howells 467348a1df65SJason A. Donenfeld /** 467448a1df65SJason A. Donenfeld * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 467548a1df65SJason A. Donenfeld * @skb: Socket buffer containing the buffers to be mapped 467648a1df65SJason A. Donenfeld * @sg: The scatter-gather list to map into 467748a1df65SJason A. Donenfeld * @offset: The offset into the buffer's contents to start mapping 467848a1df65SJason A. Donenfeld * @len: Length of buffer space to be mapped 467948a1df65SJason A. Donenfeld * 468048a1df65SJason A. Donenfeld * Fill the specified scatter-gather list with mappings/pointers into a 468148a1df65SJason A. Donenfeld * region of the buffer space attached to a socket buffer. Returns either 468248a1df65SJason A. Donenfeld * the number of scatterlist items used, or -EMSGSIZE if the contents 468348a1df65SJason A. Donenfeld * could not fit. 468448a1df65SJason A. Donenfeld */ 468548a1df65SJason A. Donenfeld int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 468648a1df65SJason A. Donenfeld { 468748a1df65SJason A. Donenfeld int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 468848a1df65SJason A. Donenfeld 468948a1df65SJason A. Donenfeld if (nsg <= 0) 469048a1df65SJason A. Donenfeld return nsg; 469148a1df65SJason A. Donenfeld 469248a1df65SJason A. Donenfeld sg_mark_end(&sg[nsg - 1]); 469348a1df65SJason A. Donenfeld 469448a1df65SJason A. Donenfeld return nsg; 469548a1df65SJason A. Donenfeld } 469648a1df65SJason A. Donenfeld EXPORT_SYMBOL_GPL(skb_to_sgvec); 469748a1df65SJason A. Donenfeld 469825a91d8dSFan Du /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 469925a91d8dSFan Du * sglist without mark the sg which contain last skb data as the end. 470025a91d8dSFan Du * So the caller can mannipulate sg list as will when padding new data after 470125a91d8dSFan Du * the first call without calling sg_unmark_end to expend sg list. 470225a91d8dSFan Du * 470325a91d8dSFan Du * Scenario to use skb_to_sgvec_nomark: 470425a91d8dSFan Du * 1. sg_init_table 470525a91d8dSFan Du * 2. skb_to_sgvec_nomark(payload1) 470625a91d8dSFan Du * 3. skb_to_sgvec_nomark(payload2) 470725a91d8dSFan Du * 470825a91d8dSFan Du * This is equivalent to: 470925a91d8dSFan Du * 1. sg_init_table 471025a91d8dSFan Du * 2. skb_to_sgvec(payload1) 471125a91d8dSFan Du * 3. sg_unmark_end 471225a91d8dSFan Du * 4. skb_to_sgvec(payload2) 471325a91d8dSFan Du * 471425a91d8dSFan Du * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 471525a91d8dSFan Du * is more preferable. 471625a91d8dSFan Du */ 471725a91d8dSFan Du int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 471825a91d8dSFan Du int offset, int len) 471925a91d8dSFan Du { 472048a1df65SJason A. Donenfeld return __skb_to_sgvec(skb, sg, offset, len, 0); 472125a91d8dSFan Du } 472225a91d8dSFan Du EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 472325a91d8dSFan Du 472451c739d1SDavid S. Miller 472551c739d1SDavid S. Miller 4726716ea3a7SDavid Howells /** 4727716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 4728716ea3a7SDavid Howells * @skb: The socket buffer to check. 4729716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 4730716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 4731716ea3a7SDavid Howells * 4732716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 4733716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 4734716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 4735716ea3a7SDavid Howells * 4736716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 4737716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 4738716ea3a7SDavid Howells * set to point to the skb in which this space begins. 4739716ea3a7SDavid Howells * 4740716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 4741716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 4742716ea3a7SDavid Howells */ 4743716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4744716ea3a7SDavid Howells { 4745716ea3a7SDavid Howells int copyflag; 4746716ea3a7SDavid Howells int elt; 4747716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 4748716ea3a7SDavid Howells 4749716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 4750716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 4751716ea3a7SDavid Howells * at the moment even if they are anonymous). 4752716ea3a7SDavid Howells */ 4753716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4754c15fc199SMiaohe Lin !__pskb_pull_tail(skb, __skb_pagelen(skb))) 4755716ea3a7SDavid Howells return -ENOMEM; 4756716ea3a7SDavid Howells 4757716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 475821dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 4759716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 4760716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 4761716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 4762716ea3a7SDavid Howells * space, 128 bytes is fair. */ 4763716ea3a7SDavid Howells 4764716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 4765716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4766716ea3a7SDavid Howells return -ENOMEM; 4767716ea3a7SDavid Howells 4768716ea3a7SDavid Howells /* Voila! */ 4769716ea3a7SDavid Howells *trailer = skb; 4770716ea3a7SDavid Howells return 1; 4771716ea3a7SDavid Howells } 4772716ea3a7SDavid Howells 4773716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 4774716ea3a7SDavid Howells 4775716ea3a7SDavid Howells elt = 1; 4776716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 4777716ea3a7SDavid Howells copyflag = 0; 4778716ea3a7SDavid Howells 4779716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 4780716ea3a7SDavid Howells int ntail = 0; 4781716ea3a7SDavid Howells 4782716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 4783716ea3a7SDavid Howells * this can happen on input. Copy it and everything 4784716ea3a7SDavid Howells * after it. */ 4785716ea3a7SDavid Howells 4786716ea3a7SDavid Howells if (skb_shared(skb1)) 4787716ea3a7SDavid Howells copyflag = 1; 4788716ea3a7SDavid Howells 4789716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 4790716ea3a7SDavid Howells 4791716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 4792716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 479321dc3301SDavid S. Miller skb_has_frag_list(skb1) || 4794716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 4795716ea3a7SDavid Howells ntail = tailbits + 128; 4796716ea3a7SDavid Howells } 4797716ea3a7SDavid Howells 4798716ea3a7SDavid Howells if (copyflag || 4799716ea3a7SDavid Howells skb_cloned(skb1) || 4800716ea3a7SDavid Howells ntail || 4801716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 480221dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 4803716ea3a7SDavid Howells struct sk_buff *skb2; 4804716ea3a7SDavid Howells 4805716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 4806716ea3a7SDavid Howells if (ntail == 0) 4807716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 4808716ea3a7SDavid Howells else 4809716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 4810716ea3a7SDavid Howells skb_headroom(skb1), 4811716ea3a7SDavid Howells ntail, 4812716ea3a7SDavid Howells GFP_ATOMIC); 4813716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 4814716ea3a7SDavid Howells return -ENOMEM; 4815716ea3a7SDavid Howells 4816716ea3a7SDavid Howells if (skb1->sk) 4817716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 4818716ea3a7SDavid Howells 4819716ea3a7SDavid Howells /* Looking around. Are we still alive? 4820716ea3a7SDavid Howells * OK, link new skb, drop old one */ 4821716ea3a7SDavid Howells 4822716ea3a7SDavid Howells skb2->next = skb1->next; 4823716ea3a7SDavid Howells *skb_p = skb2; 4824716ea3a7SDavid Howells kfree_skb(skb1); 4825716ea3a7SDavid Howells skb1 = skb2; 4826716ea3a7SDavid Howells } 4827716ea3a7SDavid Howells elt++; 4828716ea3a7SDavid Howells *trailer = skb1; 4829716ea3a7SDavid Howells skb_p = &skb1->next; 4830716ea3a7SDavid Howells } 4831716ea3a7SDavid Howells 4832716ea3a7SDavid Howells return elt; 4833716ea3a7SDavid Howells } 4834b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 4835716ea3a7SDavid Howells 4836b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 4837b1faf566SEric Dumazet { 4838b1faf566SEric Dumazet struct sock *sk = skb->sk; 4839b1faf566SEric Dumazet 4840b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4841b1faf566SEric Dumazet } 4842b1faf566SEric Dumazet 48438605330aSSoheil Hassas Yeganeh static void skb_set_err_queue(struct sk_buff *skb) 48448605330aSSoheil Hassas Yeganeh { 48458605330aSSoheil Hassas Yeganeh /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 48468605330aSSoheil Hassas Yeganeh * So, it is safe to (mis)use it to mark skbs on the error queue. 48478605330aSSoheil Hassas Yeganeh */ 48488605330aSSoheil Hassas Yeganeh skb->pkt_type = PACKET_OUTGOING; 48498605330aSSoheil Hassas Yeganeh BUILD_BUG_ON(PACKET_OUTGOING == 0); 48508605330aSSoheil Hassas Yeganeh } 48518605330aSSoheil Hassas Yeganeh 4852b1faf566SEric Dumazet /* 4853b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4854b1faf566SEric Dumazet */ 4855b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4856b1faf566SEric Dumazet { 4857b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4858ebb3b78dSEric Dumazet (unsigned int)READ_ONCE(sk->sk_rcvbuf)) 4859b1faf566SEric Dumazet return -ENOMEM; 4860b1faf566SEric Dumazet 4861b1faf566SEric Dumazet skb_orphan(skb); 4862b1faf566SEric Dumazet skb->sk = sk; 4863b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 4864b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 48658605330aSSoheil Hassas Yeganeh skb_set_err_queue(skb); 4866b1faf566SEric Dumazet 4867abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 4868abb57ea4SEric Dumazet skb_dst_force(skb); 4869abb57ea4SEric Dumazet 4870b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 4871b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 4872e3ae2365SAlexander Aring sk_error_report(sk); 4873b1faf566SEric Dumazet return 0; 4874b1faf566SEric Dumazet } 4875b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 4876b1faf566SEric Dumazet 487783a1a1a7SSoheil Hassas Yeganeh static bool is_icmp_err_skb(const struct sk_buff *skb) 487883a1a1a7SSoheil Hassas Yeganeh { 487983a1a1a7SSoheil Hassas Yeganeh return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 488083a1a1a7SSoheil Hassas Yeganeh SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 488183a1a1a7SSoheil Hassas Yeganeh } 488283a1a1a7SSoheil Hassas Yeganeh 4883364a9e93SWillem de Bruijn struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4884364a9e93SWillem de Bruijn { 4885364a9e93SWillem de Bruijn struct sk_buff_head *q = &sk->sk_error_queue; 488683a1a1a7SSoheil Hassas Yeganeh struct sk_buff *skb, *skb_next = NULL; 488783a1a1a7SSoheil Hassas Yeganeh bool icmp_next = false; 4888997d5c3fSEric Dumazet unsigned long flags; 4889364a9e93SWillem de Bruijn 4890997d5c3fSEric Dumazet spin_lock_irqsave(&q->lock, flags); 4891364a9e93SWillem de Bruijn skb = __skb_dequeue(q); 489238b25793SSoheil Hassas Yeganeh if (skb && (skb_next = skb_peek(q))) { 489383a1a1a7SSoheil Hassas Yeganeh icmp_next = is_icmp_err_skb(skb_next); 489438b25793SSoheil Hassas Yeganeh if (icmp_next) 4895985f7337SWillem de Bruijn sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno; 489638b25793SSoheil Hassas Yeganeh } 4897997d5c3fSEric Dumazet spin_unlock_irqrestore(&q->lock, flags); 4898364a9e93SWillem de Bruijn 489983a1a1a7SSoheil Hassas Yeganeh if (is_icmp_err_skb(skb) && !icmp_next) 490083a1a1a7SSoheil Hassas Yeganeh sk->sk_err = 0; 490183a1a1a7SSoheil Hassas Yeganeh 490283a1a1a7SSoheil Hassas Yeganeh if (skb_next) 4903e3ae2365SAlexander Aring sk_error_report(sk); 4904364a9e93SWillem de Bruijn 4905364a9e93SWillem de Bruijn return skb; 4906364a9e93SWillem de Bruijn } 4907364a9e93SWillem de Bruijn EXPORT_SYMBOL(sock_dequeue_err_skb); 4908364a9e93SWillem de Bruijn 4909cab41c47SAlexander Duyck /** 4910cab41c47SAlexander Duyck * skb_clone_sk - create clone of skb, and take reference to socket 4911cab41c47SAlexander Duyck * @skb: the skb to clone 4912cab41c47SAlexander Duyck * 4913cab41c47SAlexander Duyck * This function creates a clone of a buffer that holds a reference on 4914cab41c47SAlexander Duyck * sk_refcnt. Buffers created via this function are meant to be 4915cab41c47SAlexander Duyck * returned using sock_queue_err_skb, or free via kfree_skb. 4916cab41c47SAlexander Duyck * 4917cab41c47SAlexander Duyck * When passing buffers allocated with this function to sock_queue_err_skb 4918cab41c47SAlexander Duyck * it is necessary to wrap the call with sock_hold/sock_put in order to 4919cab41c47SAlexander Duyck * prevent the socket from being released prior to being enqueued on 4920cab41c47SAlexander Duyck * the sk_error_queue. 4921cab41c47SAlexander Duyck */ 492262bccb8cSAlexander Duyck struct sk_buff *skb_clone_sk(struct sk_buff *skb) 492362bccb8cSAlexander Duyck { 492462bccb8cSAlexander Duyck struct sock *sk = skb->sk; 492562bccb8cSAlexander Duyck struct sk_buff *clone; 492662bccb8cSAlexander Duyck 492741c6d650SReshetova, Elena if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 492862bccb8cSAlexander Duyck return NULL; 492962bccb8cSAlexander Duyck 493062bccb8cSAlexander Duyck clone = skb_clone(skb, GFP_ATOMIC); 493162bccb8cSAlexander Duyck if (!clone) { 493262bccb8cSAlexander Duyck sock_put(sk); 493362bccb8cSAlexander Duyck return NULL; 493462bccb8cSAlexander Duyck } 493562bccb8cSAlexander Duyck 493662bccb8cSAlexander Duyck clone->sk = sk; 493762bccb8cSAlexander Duyck clone->destructor = sock_efree; 493862bccb8cSAlexander Duyck 493962bccb8cSAlexander Duyck return clone; 494062bccb8cSAlexander Duyck } 494162bccb8cSAlexander Duyck EXPORT_SYMBOL(skb_clone_sk); 494262bccb8cSAlexander Duyck 494337846ef0SAlexander Duyck static void __skb_complete_tx_timestamp(struct sk_buff *skb, 494437846ef0SAlexander Duyck struct sock *sk, 49454ef1b286SSoheil Hassas Yeganeh int tstype, 49464ef1b286SSoheil Hassas Yeganeh bool opt_stats) 4947ac45f602SPatrick Ohly { 4948ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 4949ac45f602SPatrick Ohly int err; 4950ac45f602SPatrick Ohly 49514ef1b286SSoheil Hassas Yeganeh BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 49524ef1b286SSoheil Hassas Yeganeh 4953ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 4954ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 4955ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 4956ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4957e7fd2885SWillem de Bruijn serr->ee.ee_info = tstype; 49584ef1b286SSoheil Hassas Yeganeh serr->opt_stats = opt_stats; 49591862d620SWillem de Bruijn serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 49604ed2d765SWillem de Bruijn if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 496109c2d251SWillem de Bruijn serr->ee.ee_data = skb_shinfo(skb)->tskey; 496242f67eeaSEric Dumazet if (sk_is_tcp(sk)) 4963a1cdec57SEric Dumazet serr->ee.ee_data -= atomic_read(&sk->sk_tskey); 49644ed2d765SWillem de Bruijn } 496529030374SEric Dumazet 4966ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 496729030374SEric Dumazet 4968ac45f602SPatrick Ohly if (err) 4969ac45f602SPatrick Ohly kfree_skb(skb); 4970ac45f602SPatrick Ohly } 497137846ef0SAlexander Duyck 4972b245be1fSWillem de Bruijn static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4973b245be1fSWillem de Bruijn { 4974b245be1fSWillem de Bruijn bool ret; 4975b245be1fSWillem de Bruijn 4976d2154b0aSKuniyuki Iwashima if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly)) 4977b245be1fSWillem de Bruijn return true; 4978b245be1fSWillem de Bruijn 4979b245be1fSWillem de Bruijn read_lock_bh(&sk->sk_callback_lock); 4980b245be1fSWillem de Bruijn ret = sk->sk_socket && sk->sk_socket->file && 4981b245be1fSWillem de Bruijn file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4982b245be1fSWillem de Bruijn read_unlock_bh(&sk->sk_callback_lock); 4983b245be1fSWillem de Bruijn return ret; 4984b245be1fSWillem de Bruijn } 4985b245be1fSWillem de Bruijn 498637846ef0SAlexander Duyck void skb_complete_tx_timestamp(struct sk_buff *skb, 498737846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps) 498837846ef0SAlexander Duyck { 498937846ef0SAlexander Duyck struct sock *sk = skb->sk; 499037846ef0SAlexander Duyck 4991b245be1fSWillem de Bruijn if (!skb_may_tx_timestamp(sk, false)) 499235b99dffSWillem de Bruijn goto err; 4993b245be1fSWillem de Bruijn 49949ac25fc0SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 49959ac25fc0SEric Dumazet * but only if the socket refcount is not zero. 49969ac25fc0SEric Dumazet */ 499741c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 499837846ef0SAlexander Duyck *skb_hwtstamps(skb) = *hwtstamps; 49994ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 500037846ef0SAlexander Duyck sock_put(sk); 500135b99dffSWillem de Bruijn return; 500237846ef0SAlexander Duyck } 500335b99dffSWillem de Bruijn 500435b99dffSWillem de Bruijn err: 500535b99dffSWillem de Bruijn kfree_skb(skb); 50069ac25fc0SEric Dumazet } 500737846ef0SAlexander Duyck EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 500837846ef0SAlexander Duyck 500937846ef0SAlexander Duyck void __skb_tstamp_tx(struct sk_buff *orig_skb, 5010e7ed11eeSYousuk Seung const struct sk_buff *ack_skb, 501137846ef0SAlexander Duyck struct skb_shared_hwtstamps *hwtstamps, 501237846ef0SAlexander Duyck struct sock *sk, int tstype) 501337846ef0SAlexander Duyck { 501437846ef0SAlexander Duyck struct sk_buff *skb; 50154ef1b286SSoheil Hassas Yeganeh bool tsonly, opt_stats = false; 501637846ef0SAlexander Duyck 50173a8dd971SWillem de Bruijn if (!sk) 50183a8dd971SWillem de Bruijn return; 50193a8dd971SWillem de Bruijn 5020b50a5c70SMiroslav Lichvar if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 5021b50a5c70SMiroslav Lichvar skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 5022b50a5c70SMiroslav Lichvar return; 5023b50a5c70SMiroslav Lichvar 50243a8dd971SWillem de Bruijn tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 50253a8dd971SWillem de Bruijn if (!skb_may_tx_timestamp(sk, tsonly)) 502637846ef0SAlexander Duyck return; 502737846ef0SAlexander Duyck 50281c885808SFrancis Yan if (tsonly) { 50291c885808SFrancis Yan #ifdef CONFIG_INET 50301c885808SFrancis Yan if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 503142f67eeaSEric Dumazet sk_is_tcp(sk)) { 5032e7ed11eeSYousuk Seung skb = tcp_get_timestamping_opt_stats(sk, orig_skb, 5033e7ed11eeSYousuk Seung ack_skb); 50344ef1b286SSoheil Hassas Yeganeh opt_stats = true; 50354ef1b286SSoheil Hassas Yeganeh } else 50361c885808SFrancis Yan #endif 50371c885808SFrancis Yan skb = alloc_skb(0, GFP_ATOMIC); 50381c885808SFrancis Yan } else { 503937846ef0SAlexander Duyck skb = skb_clone(orig_skb, GFP_ATOMIC); 50401c885808SFrancis Yan } 504137846ef0SAlexander Duyck if (!skb) 504237846ef0SAlexander Duyck return; 504337846ef0SAlexander Duyck 504449ca0d8bSWillem de Bruijn if (tsonly) { 5045fff88030SWillem de Bruijn skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 5046fff88030SWillem de Bruijn SKBTX_ANY_TSTAMP; 504749ca0d8bSWillem de Bruijn skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 504849ca0d8bSWillem de Bruijn } 504949ca0d8bSWillem de Bruijn 505049ca0d8bSWillem de Bruijn if (hwtstamps) 505149ca0d8bSWillem de Bruijn *skb_hwtstamps(skb) = *hwtstamps; 505249ca0d8bSWillem de Bruijn else 5053d93376f5SMartin KaFai Lau __net_timestamp(skb); 505449ca0d8bSWillem de Bruijn 50554ef1b286SSoheil Hassas Yeganeh __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 505637846ef0SAlexander Duyck } 5057e7fd2885SWillem de Bruijn EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 5058e7fd2885SWillem de Bruijn 5059e7fd2885SWillem de Bruijn void skb_tstamp_tx(struct sk_buff *orig_skb, 5060e7fd2885SWillem de Bruijn struct skb_shared_hwtstamps *hwtstamps) 5061e7fd2885SWillem de Bruijn { 5062e7ed11eeSYousuk Seung return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk, 5063e7fd2885SWillem de Bruijn SCM_TSTAMP_SND); 5064e7fd2885SWillem de Bruijn } 5065ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 5066ac45f602SPatrick Ohly 50676e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 50686e3e939fSJohannes Berg { 50696e3e939fSJohannes Berg struct sock *sk = skb->sk; 50706e3e939fSJohannes Berg struct sock_exterr_skb *serr; 5071dd4f1072SEric Dumazet int err = 1; 50726e3e939fSJohannes Berg 50736e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 50746e3e939fSJohannes Berg skb->wifi_acked = acked; 50756e3e939fSJohannes Berg 50766e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 50776e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 50786e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 50796e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 50806e3e939fSJohannes Berg 5081dd4f1072SEric Dumazet /* Take a reference to prevent skb_orphan() from freeing the socket, 5082dd4f1072SEric Dumazet * but only if the socket refcount is not zero. 5083dd4f1072SEric Dumazet */ 508441c6d650SReshetova, Elena if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 50856e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 5086dd4f1072SEric Dumazet sock_put(sk); 5087dd4f1072SEric Dumazet } 50886e3e939fSJohannes Berg if (err) 50896e3e939fSJohannes Berg kfree_skb(skb); 50906e3e939fSJohannes Berg } 50916e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 50926e3e939fSJohannes Berg 5093f35d9d8aSRusty Russell /** 5094f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 5095f35d9d8aSRusty Russell * @skb: the skb to set 5096f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 5097f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 5098f35d9d8aSRusty Russell * 5099f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 5100f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 5101f35d9d8aSRusty Russell * 5102f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 5103f35d9d8aSRusty Russell * returns false you should drop the packet. 5104f35d9d8aSRusty Russell */ 5105f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 5106f35d9d8aSRusty Russell { 510752b5d6f5SEric Dumazet u32 csum_end = (u32)start + (u32)off + sizeof(__sum16); 510852b5d6f5SEric Dumazet u32 csum_start = skb_headroom(skb) + (u32)start; 510952b5d6f5SEric Dumazet 511052b5d6f5SEric Dumazet if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) { 511152b5d6f5SEric Dumazet net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n", 511252b5d6f5SEric Dumazet start, off, skb_headroom(skb), skb_headlen(skb)); 5113f35d9d8aSRusty Russell return false; 5114f35d9d8aSRusty Russell } 5115f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 511652b5d6f5SEric Dumazet skb->csum_start = csum_start; 5117f35d9d8aSRusty Russell skb->csum_offset = off; 5118e5d5decaSJason Wang skb_set_transport_header(skb, start); 5119f35d9d8aSRusty Russell return true; 5120f35d9d8aSRusty Russell } 5121b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 5122f35d9d8aSRusty Russell 5123ed1f50c3SPaul Durrant static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 5124ed1f50c3SPaul Durrant unsigned int max) 5125ed1f50c3SPaul Durrant { 5126ed1f50c3SPaul Durrant if (skb_headlen(skb) >= len) 5127ed1f50c3SPaul Durrant return 0; 5128ed1f50c3SPaul Durrant 5129ed1f50c3SPaul Durrant /* If we need to pullup then pullup to the max, so we 5130ed1f50c3SPaul Durrant * won't need to do it again. 5131ed1f50c3SPaul Durrant */ 5132ed1f50c3SPaul Durrant if (max > skb->len) 5133ed1f50c3SPaul Durrant max = skb->len; 5134ed1f50c3SPaul Durrant 5135ed1f50c3SPaul Durrant if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 5136ed1f50c3SPaul Durrant return -ENOMEM; 5137ed1f50c3SPaul Durrant 5138ed1f50c3SPaul Durrant if (skb_headlen(skb) < len) 5139ed1f50c3SPaul Durrant return -EPROTO; 5140ed1f50c3SPaul Durrant 5141ed1f50c3SPaul Durrant return 0; 5142ed1f50c3SPaul Durrant } 5143ed1f50c3SPaul Durrant 5144f9708b43SJan Beulich #define MAX_TCP_HDR_LEN (15 * 4) 5145f9708b43SJan Beulich 5146f9708b43SJan Beulich static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 5147f9708b43SJan Beulich typeof(IPPROTO_IP) proto, 5148f9708b43SJan Beulich unsigned int off) 5149f9708b43SJan Beulich { 5150f9708b43SJan Beulich int err; 5151f9708b43SJan Beulich 5152161d1792SKees Cook switch (proto) { 5153f9708b43SJan Beulich case IPPROTO_TCP: 5154f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 5155f9708b43SJan Beulich off + MAX_TCP_HDR_LEN); 5156f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 5157f9708b43SJan Beulich offsetof(struct tcphdr, 5158f9708b43SJan Beulich check))) 5159f9708b43SJan Beulich err = -EPROTO; 5160f9708b43SJan Beulich return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 5161f9708b43SJan Beulich 5162f9708b43SJan Beulich case IPPROTO_UDP: 5163f9708b43SJan Beulich err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 5164f9708b43SJan Beulich off + sizeof(struct udphdr)); 5165f9708b43SJan Beulich if (!err && !skb_partial_csum_set(skb, off, 5166f9708b43SJan Beulich offsetof(struct udphdr, 5167f9708b43SJan Beulich check))) 5168f9708b43SJan Beulich err = -EPROTO; 5169f9708b43SJan Beulich return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 5170f9708b43SJan Beulich } 5171f9708b43SJan Beulich 5172f9708b43SJan Beulich return ERR_PTR(-EPROTO); 5173f9708b43SJan Beulich } 5174f9708b43SJan Beulich 5175ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5176ed1f50c3SPaul Durrant * maximally sized IP and TCP or UDP headers. 5177ed1f50c3SPaul Durrant */ 5178ed1f50c3SPaul Durrant #define MAX_IP_HDR_LEN 128 5179ed1f50c3SPaul Durrant 5180f9708b43SJan Beulich static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 5181ed1f50c3SPaul Durrant { 5182ed1f50c3SPaul Durrant unsigned int off; 5183ed1f50c3SPaul Durrant bool fragment; 5184f9708b43SJan Beulich __sum16 *csum; 5185ed1f50c3SPaul Durrant int err; 5186ed1f50c3SPaul Durrant 5187ed1f50c3SPaul Durrant fragment = false; 5188ed1f50c3SPaul Durrant 5189ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5190ed1f50c3SPaul Durrant sizeof(struct iphdr), 5191ed1f50c3SPaul Durrant MAX_IP_HDR_LEN); 5192ed1f50c3SPaul Durrant if (err < 0) 5193ed1f50c3SPaul Durrant goto out; 5194ed1f50c3SPaul Durrant 519511f920d2SMiaohe Lin if (ip_is_fragment(ip_hdr(skb))) 5196ed1f50c3SPaul Durrant fragment = true; 5197ed1f50c3SPaul Durrant 5198ed1f50c3SPaul Durrant off = ip_hdrlen(skb); 5199ed1f50c3SPaul Durrant 5200ed1f50c3SPaul Durrant err = -EPROTO; 5201ed1f50c3SPaul Durrant 5202ed1f50c3SPaul Durrant if (fragment) 5203ed1f50c3SPaul Durrant goto out; 5204ed1f50c3SPaul Durrant 5205f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 5206f9708b43SJan Beulich if (IS_ERR(csum)) 5207f9708b43SJan Beulich return PTR_ERR(csum); 5208ed1f50c3SPaul Durrant 5209ed1f50c3SPaul Durrant if (recalculate) 5210f9708b43SJan Beulich *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 5211ed1f50c3SPaul Durrant ip_hdr(skb)->daddr, 5212ed1f50c3SPaul Durrant skb->len - off, 5213f9708b43SJan Beulich ip_hdr(skb)->protocol, 0); 5214ed1f50c3SPaul Durrant err = 0; 5215ed1f50c3SPaul Durrant 5216ed1f50c3SPaul Durrant out: 5217ed1f50c3SPaul Durrant return err; 5218ed1f50c3SPaul Durrant } 5219ed1f50c3SPaul Durrant 5220ed1f50c3SPaul Durrant /* This value should be large enough to cover a tagged ethernet header plus 5221ed1f50c3SPaul Durrant * an IPv6 header, all options, and a maximal TCP or UDP header. 5222ed1f50c3SPaul Durrant */ 5223ed1f50c3SPaul Durrant #define MAX_IPV6_HDR_LEN 256 5224ed1f50c3SPaul Durrant 5225ed1f50c3SPaul Durrant #define OPT_HDR(type, skb, off) \ 5226ed1f50c3SPaul Durrant (type *)(skb_network_header(skb) + (off)) 5227ed1f50c3SPaul Durrant 5228ed1f50c3SPaul Durrant static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 5229ed1f50c3SPaul Durrant { 5230ed1f50c3SPaul Durrant int err; 5231ed1f50c3SPaul Durrant u8 nexthdr; 5232ed1f50c3SPaul Durrant unsigned int off; 5233ed1f50c3SPaul Durrant unsigned int len; 5234ed1f50c3SPaul Durrant bool fragment; 5235ed1f50c3SPaul Durrant bool done; 5236f9708b43SJan Beulich __sum16 *csum; 5237ed1f50c3SPaul Durrant 5238ed1f50c3SPaul Durrant fragment = false; 5239ed1f50c3SPaul Durrant done = false; 5240ed1f50c3SPaul Durrant 5241ed1f50c3SPaul Durrant off = sizeof(struct ipv6hdr); 5242ed1f50c3SPaul Durrant 5243ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 5244ed1f50c3SPaul Durrant if (err < 0) 5245ed1f50c3SPaul Durrant goto out; 5246ed1f50c3SPaul Durrant 5247ed1f50c3SPaul Durrant nexthdr = ipv6_hdr(skb)->nexthdr; 5248ed1f50c3SPaul Durrant 5249ed1f50c3SPaul Durrant len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 5250ed1f50c3SPaul Durrant while (off <= len && !done) { 5251ed1f50c3SPaul Durrant switch (nexthdr) { 5252ed1f50c3SPaul Durrant case IPPROTO_DSTOPTS: 5253ed1f50c3SPaul Durrant case IPPROTO_HOPOPTS: 5254ed1f50c3SPaul Durrant case IPPROTO_ROUTING: { 5255ed1f50c3SPaul Durrant struct ipv6_opt_hdr *hp; 5256ed1f50c3SPaul Durrant 5257ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5258ed1f50c3SPaul Durrant off + 5259ed1f50c3SPaul Durrant sizeof(struct ipv6_opt_hdr), 5260ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5261ed1f50c3SPaul Durrant if (err < 0) 5262ed1f50c3SPaul Durrant goto out; 5263ed1f50c3SPaul Durrant 5264ed1f50c3SPaul Durrant hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 5265ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5266ed1f50c3SPaul Durrant off += ipv6_optlen(hp); 5267ed1f50c3SPaul Durrant break; 5268ed1f50c3SPaul Durrant } 5269ed1f50c3SPaul Durrant case IPPROTO_AH: { 5270ed1f50c3SPaul Durrant struct ip_auth_hdr *hp; 5271ed1f50c3SPaul Durrant 5272ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5273ed1f50c3SPaul Durrant off + 5274ed1f50c3SPaul Durrant sizeof(struct ip_auth_hdr), 5275ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5276ed1f50c3SPaul Durrant if (err < 0) 5277ed1f50c3SPaul Durrant goto out; 5278ed1f50c3SPaul Durrant 5279ed1f50c3SPaul Durrant hp = OPT_HDR(struct ip_auth_hdr, skb, off); 5280ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5281ed1f50c3SPaul Durrant off += ipv6_authlen(hp); 5282ed1f50c3SPaul Durrant break; 5283ed1f50c3SPaul Durrant } 5284ed1f50c3SPaul Durrant case IPPROTO_FRAGMENT: { 5285ed1f50c3SPaul Durrant struct frag_hdr *hp; 5286ed1f50c3SPaul Durrant 5287ed1f50c3SPaul Durrant err = skb_maybe_pull_tail(skb, 5288ed1f50c3SPaul Durrant off + 5289ed1f50c3SPaul Durrant sizeof(struct frag_hdr), 5290ed1f50c3SPaul Durrant MAX_IPV6_HDR_LEN); 5291ed1f50c3SPaul Durrant if (err < 0) 5292ed1f50c3SPaul Durrant goto out; 5293ed1f50c3SPaul Durrant 5294ed1f50c3SPaul Durrant hp = OPT_HDR(struct frag_hdr, skb, off); 5295ed1f50c3SPaul Durrant 5296ed1f50c3SPaul Durrant if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 5297ed1f50c3SPaul Durrant fragment = true; 5298ed1f50c3SPaul Durrant 5299ed1f50c3SPaul Durrant nexthdr = hp->nexthdr; 5300ed1f50c3SPaul Durrant off += sizeof(struct frag_hdr); 5301ed1f50c3SPaul Durrant break; 5302ed1f50c3SPaul Durrant } 5303ed1f50c3SPaul Durrant default: 5304ed1f50c3SPaul Durrant done = true; 5305ed1f50c3SPaul Durrant break; 5306ed1f50c3SPaul Durrant } 5307ed1f50c3SPaul Durrant } 5308ed1f50c3SPaul Durrant 5309ed1f50c3SPaul Durrant err = -EPROTO; 5310ed1f50c3SPaul Durrant 5311ed1f50c3SPaul Durrant if (!done || fragment) 5312ed1f50c3SPaul Durrant goto out; 5313ed1f50c3SPaul Durrant 5314f9708b43SJan Beulich csum = skb_checksum_setup_ip(skb, nexthdr, off); 5315f9708b43SJan Beulich if (IS_ERR(csum)) 5316f9708b43SJan Beulich return PTR_ERR(csum); 5317ed1f50c3SPaul Durrant 5318ed1f50c3SPaul Durrant if (recalculate) 5319f9708b43SJan Beulich *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 5320ed1f50c3SPaul Durrant &ipv6_hdr(skb)->daddr, 5321f9708b43SJan Beulich skb->len - off, nexthdr, 0); 5322ed1f50c3SPaul Durrant err = 0; 5323ed1f50c3SPaul Durrant 5324ed1f50c3SPaul Durrant out: 5325ed1f50c3SPaul Durrant return err; 5326ed1f50c3SPaul Durrant } 5327ed1f50c3SPaul Durrant 5328ed1f50c3SPaul Durrant /** 5329ed1f50c3SPaul Durrant * skb_checksum_setup - set up partial checksum offset 5330ed1f50c3SPaul Durrant * @skb: the skb to set up 5331ed1f50c3SPaul Durrant * @recalculate: if true the pseudo-header checksum will be recalculated 5332ed1f50c3SPaul Durrant */ 5333ed1f50c3SPaul Durrant int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 5334ed1f50c3SPaul Durrant { 5335ed1f50c3SPaul Durrant int err; 5336ed1f50c3SPaul Durrant 5337ed1f50c3SPaul Durrant switch (skb->protocol) { 5338ed1f50c3SPaul Durrant case htons(ETH_P_IP): 5339f9708b43SJan Beulich err = skb_checksum_setup_ipv4(skb, recalculate); 5340ed1f50c3SPaul Durrant break; 5341ed1f50c3SPaul Durrant 5342ed1f50c3SPaul Durrant case htons(ETH_P_IPV6): 5343ed1f50c3SPaul Durrant err = skb_checksum_setup_ipv6(skb, recalculate); 5344ed1f50c3SPaul Durrant break; 5345ed1f50c3SPaul Durrant 5346ed1f50c3SPaul Durrant default: 5347ed1f50c3SPaul Durrant err = -EPROTO; 5348ed1f50c3SPaul Durrant break; 5349ed1f50c3SPaul Durrant } 5350ed1f50c3SPaul Durrant 5351ed1f50c3SPaul Durrant return err; 5352ed1f50c3SPaul Durrant } 5353ed1f50c3SPaul Durrant EXPORT_SYMBOL(skb_checksum_setup); 5354ed1f50c3SPaul Durrant 53559afd85c9SLinus Lüssing /** 53569afd85c9SLinus Lüssing * skb_checksum_maybe_trim - maybe trims the given skb 53579afd85c9SLinus Lüssing * @skb: the skb to check 53589afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 53599afd85c9SLinus Lüssing * 53609afd85c9SLinus Lüssing * Checks whether the given skb has data beyond the given transport length. 53619afd85c9SLinus Lüssing * If so, returns a cloned skb trimmed to this transport length. 53629afd85c9SLinus Lüssing * Otherwise returns the provided skb. Returns NULL in error cases 53639afd85c9SLinus Lüssing * (e.g. transport_len exceeds skb length or out-of-memory). 53649afd85c9SLinus Lüssing * 5365a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5366a516993fSLinus Lüssing * differs from the provided skb. 53679afd85c9SLinus Lüssing */ 53689afd85c9SLinus Lüssing static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 53699afd85c9SLinus Lüssing unsigned int transport_len) 53709afd85c9SLinus Lüssing { 53719afd85c9SLinus Lüssing struct sk_buff *skb_chk; 53729afd85c9SLinus Lüssing unsigned int len = skb_transport_offset(skb) + transport_len; 53739afd85c9SLinus Lüssing int ret; 53749afd85c9SLinus Lüssing 5375a516993fSLinus Lüssing if (skb->len < len) 53769afd85c9SLinus Lüssing return NULL; 5377a516993fSLinus Lüssing else if (skb->len == len) 53789afd85c9SLinus Lüssing return skb; 53799afd85c9SLinus Lüssing 53809afd85c9SLinus Lüssing skb_chk = skb_clone(skb, GFP_ATOMIC); 53819afd85c9SLinus Lüssing if (!skb_chk) 53829afd85c9SLinus Lüssing return NULL; 53839afd85c9SLinus Lüssing 53849afd85c9SLinus Lüssing ret = pskb_trim_rcsum(skb_chk, len); 53859afd85c9SLinus Lüssing if (ret) { 53869afd85c9SLinus Lüssing kfree_skb(skb_chk); 53879afd85c9SLinus Lüssing return NULL; 53889afd85c9SLinus Lüssing } 53899afd85c9SLinus Lüssing 53909afd85c9SLinus Lüssing return skb_chk; 53919afd85c9SLinus Lüssing } 53929afd85c9SLinus Lüssing 53939afd85c9SLinus Lüssing /** 53949afd85c9SLinus Lüssing * skb_checksum_trimmed - validate checksum of an skb 53959afd85c9SLinus Lüssing * @skb: the skb to check 53969afd85c9SLinus Lüssing * @transport_len: the data length beyond the network header 53979afd85c9SLinus Lüssing * @skb_chkf: checksum function to use 53989afd85c9SLinus Lüssing * 53999afd85c9SLinus Lüssing * Applies the given checksum function skb_chkf to the provided skb. 54009afd85c9SLinus Lüssing * Returns a checked and maybe trimmed skb. Returns NULL on error. 54019afd85c9SLinus Lüssing * 54029afd85c9SLinus Lüssing * If the skb has data beyond the given transport length, then a 54039afd85c9SLinus Lüssing * trimmed & cloned skb is checked and returned. 54049afd85c9SLinus Lüssing * 5405a516993fSLinus Lüssing * Caller needs to set the skb transport header and free any returned skb if it 5406a516993fSLinus Lüssing * differs from the provided skb. 54079afd85c9SLinus Lüssing */ 54089afd85c9SLinus Lüssing struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 54099afd85c9SLinus Lüssing unsigned int transport_len, 54109afd85c9SLinus Lüssing __sum16(*skb_chkf)(struct sk_buff *skb)) 54119afd85c9SLinus Lüssing { 54129afd85c9SLinus Lüssing struct sk_buff *skb_chk; 54139afd85c9SLinus Lüssing unsigned int offset = skb_transport_offset(skb); 5414fcba67c9SLinus Lüssing __sum16 ret; 54159afd85c9SLinus Lüssing 54169afd85c9SLinus Lüssing skb_chk = skb_checksum_maybe_trim(skb, transport_len); 54179afd85c9SLinus Lüssing if (!skb_chk) 5418a516993fSLinus Lüssing goto err; 54199afd85c9SLinus Lüssing 5420a516993fSLinus Lüssing if (!pskb_may_pull(skb_chk, offset)) 5421a516993fSLinus Lüssing goto err; 54229afd85c9SLinus Lüssing 54239b368814SLinus Lüssing skb_pull_rcsum(skb_chk, offset); 54249afd85c9SLinus Lüssing ret = skb_chkf(skb_chk); 54259b368814SLinus Lüssing skb_push_rcsum(skb_chk, offset); 54269afd85c9SLinus Lüssing 5427a516993fSLinus Lüssing if (ret) 5428a516993fSLinus Lüssing goto err; 54299afd85c9SLinus Lüssing 54309afd85c9SLinus Lüssing return skb_chk; 5431a516993fSLinus Lüssing 5432a516993fSLinus Lüssing err: 5433a516993fSLinus Lüssing if (skb_chk && skb_chk != skb) 5434a516993fSLinus Lüssing kfree_skb(skb_chk); 5435a516993fSLinus Lüssing 5436a516993fSLinus Lüssing return NULL; 5437a516993fSLinus Lüssing 54389afd85c9SLinus Lüssing } 54399afd85c9SLinus Lüssing EXPORT_SYMBOL(skb_checksum_trimmed); 54409afd85c9SLinus Lüssing 54414497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 54424497b076SBen Hutchings { 5443e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 5444e87cc472SJoe Perches skb->dev->name); 54454497b076SBen Hutchings } 54464497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 5447bad43ca8SEric Dumazet 5448bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 5449bad43ca8SEric Dumazet { 54503d861f66SEric Dumazet if (head_stolen) { 54513d861f66SEric Dumazet skb_release_head_state(skb); 5452bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 54533d861f66SEric Dumazet } else { 5454bad43ca8SEric Dumazet __kfree_skb(skb); 5455bad43ca8SEric Dumazet } 54563d861f66SEric Dumazet } 5457bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 5458bad43ca8SEric Dumazet 5459bad43ca8SEric Dumazet /** 5460bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 5461bad43ca8SEric Dumazet * @to: prior buffer 5462bad43ca8SEric Dumazet * @from: buffer to add 5463bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 5464c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 5465bad43ca8SEric Dumazet */ 5466bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 5467bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 5468bad43ca8SEric Dumazet { 5469c818fa9eSEric Dumazet struct skb_shared_info *to_shinfo, *from_shinfo; 5470bad43ca8SEric Dumazet int i, delta, len = from->len; 5471bad43ca8SEric Dumazet 5472bad43ca8SEric Dumazet *fragstolen = false; 5473bad43ca8SEric Dumazet 5474bad43ca8SEric Dumazet if (skb_cloned(to)) 5475bad43ca8SEric Dumazet return false; 5476bad43ca8SEric Dumazet 54771effe8caSJean-Philippe Brucker /* In general, avoid mixing slab allocated and page_pool allocated 54781effe8caSJean-Philippe Brucker * pages within the same SKB. However when @to is not pp_recycle and 54791effe8caSJean-Philippe Brucker * @from is cloned, we can transition frag pages from page_pool to 54801effe8caSJean-Philippe Brucker * reference counted. 54811effe8caSJean-Philippe Brucker * 54821effe8caSJean-Philippe Brucker * On the other hand, don't allow coalescing two pp_recycle SKBs if 54831effe8caSJean-Philippe Brucker * @from is cloned, in case the SKB is using page_pool fragment 54841effe8caSJean-Philippe Brucker * references (PP_FLAG_PAGE_FRAG). Since we only take full page 54851effe8caSJean-Philippe Brucker * references for cloned SKBs at the moment that would result in 54861effe8caSJean-Philippe Brucker * inconsistent reference counts. 54876a5bcd84SIlias Apalodimas */ 54881effe8caSJean-Philippe Brucker if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from))) 54896a5bcd84SIlias Apalodimas return false; 54906a5bcd84SIlias Apalodimas 5491bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 5492e93a0435SEric Dumazet if (len) 5493bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 5494bad43ca8SEric Dumazet *delta_truesize = 0; 5495bad43ca8SEric Dumazet return true; 5496bad43ca8SEric Dumazet } 5497bad43ca8SEric Dumazet 5498c818fa9eSEric Dumazet to_shinfo = skb_shinfo(to); 5499c818fa9eSEric Dumazet from_shinfo = skb_shinfo(from); 5500c818fa9eSEric Dumazet if (to_shinfo->frag_list || from_shinfo->frag_list) 5501bad43ca8SEric Dumazet return false; 55021f8b977aSWillem de Bruijn if (skb_zcopy(to) || skb_zcopy(from)) 55031f8b977aSWillem de Bruijn return false; 5504bad43ca8SEric Dumazet 5505bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 5506bad43ca8SEric Dumazet struct page *page; 5507bad43ca8SEric Dumazet unsigned int offset; 5508bad43ca8SEric Dumazet 5509c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5510c818fa9eSEric Dumazet from_shinfo->nr_frags >= MAX_SKB_FRAGS) 5511bad43ca8SEric Dumazet return false; 5512bad43ca8SEric Dumazet 5513bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 5514bad43ca8SEric Dumazet return false; 5515bad43ca8SEric Dumazet 5516bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 5517bad43ca8SEric Dumazet 5518bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 5519bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 5520bad43ca8SEric Dumazet 5521c818fa9eSEric Dumazet skb_fill_page_desc(to, to_shinfo->nr_frags, 5522bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 5523bad43ca8SEric Dumazet *fragstolen = true; 5524bad43ca8SEric Dumazet } else { 5525c818fa9eSEric Dumazet if (to_shinfo->nr_frags + 5526c818fa9eSEric Dumazet from_shinfo->nr_frags > MAX_SKB_FRAGS) 5527bad43ca8SEric Dumazet return false; 5528bad43ca8SEric Dumazet 5529f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 5530bad43ca8SEric Dumazet } 5531bad43ca8SEric Dumazet 5532bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 5533bad43ca8SEric Dumazet 5534c818fa9eSEric Dumazet memcpy(to_shinfo->frags + to_shinfo->nr_frags, 5535c818fa9eSEric Dumazet from_shinfo->frags, 5536c818fa9eSEric Dumazet from_shinfo->nr_frags * sizeof(skb_frag_t)); 5537c818fa9eSEric Dumazet to_shinfo->nr_frags += from_shinfo->nr_frags; 5538bad43ca8SEric Dumazet 5539bad43ca8SEric Dumazet if (!skb_cloned(from)) 5540c818fa9eSEric Dumazet from_shinfo->nr_frags = 0; 5541bad43ca8SEric Dumazet 55428ea853fdSLi RongQing /* if the skb is not cloned this does nothing 55438ea853fdSLi RongQing * since we set nr_frags to 0. 55448ea853fdSLi RongQing */ 5545c818fa9eSEric Dumazet for (i = 0; i < from_shinfo->nr_frags; i++) 5546c818fa9eSEric Dumazet __skb_frag_ref(&from_shinfo->frags[i]); 5547bad43ca8SEric Dumazet 5548bad43ca8SEric Dumazet to->truesize += delta; 5549bad43ca8SEric Dumazet to->len += len; 5550bad43ca8SEric Dumazet to->data_len += len; 5551bad43ca8SEric Dumazet 5552bad43ca8SEric Dumazet *delta_truesize = delta; 5553bad43ca8SEric Dumazet return true; 5554bad43ca8SEric Dumazet } 5555bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 5556621e84d6SNicolas Dichtel 5557621e84d6SNicolas Dichtel /** 55588b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 5559621e84d6SNicolas Dichtel * 5560621e84d6SNicolas Dichtel * @skb: buffer to clean 55618b27f277SNicolas Dichtel * @xnet: packet is crossing netns 5562621e84d6SNicolas Dichtel * 55638b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 55648b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 55658b27f277SNicolas Dichtel * operations. 55668b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 55678b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 55688b27f277SNicolas Dichtel * skb that could impact namespace isolation. 5569621e84d6SNicolas Dichtel */ 55708b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 5571621e84d6SNicolas Dichtel { 5572621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 5573621e84d6SNicolas Dichtel skb->skb_iif = 0; 557460ff7467SWANG Cong skb->ignore_df = 0; 5575621e84d6SNicolas Dichtel skb_dst_drop(skb); 5576174e2381SFlorian Westphal skb_ext_reset(skb); 5577895b5c9fSFlorian Westphal nf_reset_ct(skb); 5578621e84d6SNicolas Dichtel nf_reset_trace(skb); 5579213dd74aSHerbert Xu 55806f9a5069SPetr Machata #ifdef CONFIG_NET_SWITCHDEV 55816f9a5069SPetr Machata skb->offload_fwd_mark = 0; 5582875e8939SIdo Schimmel skb->offload_l3_fwd_mark = 0; 55836f9a5069SPetr Machata #endif 55846f9a5069SPetr Machata 5585213dd74aSHerbert Xu if (!xnet) 5586213dd74aSHerbert Xu return; 5587213dd74aSHerbert Xu 55882b5ec1a5SYe Yin ipvs_reset(skb); 5589213dd74aSHerbert Xu skb->mark = 0; 5590de799101SMartin KaFai Lau skb_clear_tstamp(skb); 5591621e84d6SNicolas Dichtel } 5592621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 5593de960aa9SFlorian Westphal 5594de960aa9SFlorian Westphal /** 5595de960aa9SFlorian Westphal * skb_gso_transport_seglen - Return length of individual segments of a gso packet 5596de960aa9SFlorian Westphal * 5597de960aa9SFlorian Westphal * @skb: GSO skb 5598de960aa9SFlorian Westphal * 5599de960aa9SFlorian Westphal * skb_gso_transport_seglen is used to determine the real size of the 5600de960aa9SFlorian Westphal * individual segments, including Layer4 headers (TCP/UDP). 5601de960aa9SFlorian Westphal * 5602de960aa9SFlorian Westphal * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 5603de960aa9SFlorian Westphal */ 5604a4a77718SDaniel Axtens static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 5605de960aa9SFlorian Westphal { 5606de960aa9SFlorian Westphal const struct skb_shared_info *shinfo = skb_shinfo(skb); 5607f993bc25SFlorian Westphal unsigned int thlen = 0; 5608f993bc25SFlorian Westphal 5609f993bc25SFlorian Westphal if (skb->encapsulation) { 5610f993bc25SFlorian Westphal thlen = skb_inner_transport_header(skb) - 5611f993bc25SFlorian Westphal skb_transport_header(skb); 5612de960aa9SFlorian Westphal 5613de960aa9SFlorian Westphal if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 5614f993bc25SFlorian Westphal thlen += inner_tcp_hdrlen(skb); 5615f993bc25SFlorian Westphal } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 5616f993bc25SFlorian Westphal thlen = tcp_hdrlen(skb); 56171dd27cdeSDaniel Axtens } else if (unlikely(skb_is_gso_sctp(skb))) { 561890017accSMarcelo Ricardo Leitner thlen = sizeof(struct sctphdr); 5619ee80d1ebSWillem de Bruijn } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 5620ee80d1ebSWillem de Bruijn thlen = sizeof(struct udphdr); 5621f993bc25SFlorian Westphal } 56226d39d589SFlorian Westphal /* UFO sets gso_size to the size of the fragmentation 56236d39d589SFlorian Westphal * payload, i.e. the size of the L4 (UDP) header is already 56246d39d589SFlorian Westphal * accounted for. 56256d39d589SFlorian Westphal */ 5626f993bc25SFlorian Westphal return thlen + shinfo->gso_size; 5627de960aa9SFlorian Westphal } 5628a4a77718SDaniel Axtens 5629a4a77718SDaniel Axtens /** 5630a4a77718SDaniel Axtens * skb_gso_network_seglen - Return length of individual segments of a gso packet 5631a4a77718SDaniel Axtens * 5632a4a77718SDaniel Axtens * @skb: GSO skb 5633a4a77718SDaniel Axtens * 5634a4a77718SDaniel Axtens * skb_gso_network_seglen is used to determine the real size of the 5635a4a77718SDaniel Axtens * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 5636a4a77718SDaniel Axtens * 5637a4a77718SDaniel Axtens * The MAC/L2 header is not accounted for. 5638a4a77718SDaniel Axtens */ 5639a4a77718SDaniel Axtens static unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 5640a4a77718SDaniel Axtens { 5641a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - 5642a4a77718SDaniel Axtens skb_network_header(skb); 5643a4a77718SDaniel Axtens 5644a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5645a4a77718SDaniel Axtens } 5646a4a77718SDaniel Axtens 5647a4a77718SDaniel Axtens /** 5648a4a77718SDaniel Axtens * skb_gso_mac_seglen - Return length of individual segments of a gso packet 5649a4a77718SDaniel Axtens * 5650a4a77718SDaniel Axtens * @skb: GSO skb 5651a4a77718SDaniel Axtens * 5652a4a77718SDaniel Axtens * skb_gso_mac_seglen is used to determine the real size of the 5653a4a77718SDaniel Axtens * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4 5654a4a77718SDaniel Axtens * headers (TCP/UDP). 5655a4a77718SDaniel Axtens */ 5656a4a77718SDaniel Axtens static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb) 5657a4a77718SDaniel Axtens { 5658a4a77718SDaniel Axtens unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 5659a4a77718SDaniel Axtens 5660a4a77718SDaniel Axtens return hdr_len + skb_gso_transport_seglen(skb); 5661a4a77718SDaniel Axtens } 56620d5501c1SVlad Yasevich 5663ae7ef81eSMarcelo Ricardo Leitner /** 56642b16f048SDaniel Axtens * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS 56652b16f048SDaniel Axtens * 56662b16f048SDaniel Axtens * There are a couple of instances where we have a GSO skb, and we 56672b16f048SDaniel Axtens * want to determine what size it would be after it is segmented. 56682b16f048SDaniel Axtens * 56692b16f048SDaniel Axtens * We might want to check: 56702b16f048SDaniel Axtens * - L3+L4+payload size (e.g. IP forwarding) 56712b16f048SDaniel Axtens * - L2+L3+L4+payload size (e.g. sanity check before passing to driver) 56722b16f048SDaniel Axtens * 56732b16f048SDaniel Axtens * This is a helper to do that correctly considering GSO_BY_FRAGS. 56742b16f048SDaniel Axtens * 567549682bfaSMathieu Malaterre * @skb: GSO skb 567649682bfaSMathieu Malaterre * 56772b16f048SDaniel Axtens * @seg_len: The segmented length (from skb_gso_*_seglen). In the 56782b16f048SDaniel Axtens * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS]. 56792b16f048SDaniel Axtens * 56802b16f048SDaniel Axtens * @max_len: The maximum permissible length. 56812b16f048SDaniel Axtens * 56822b16f048SDaniel Axtens * Returns true if the segmented length <= max length. 56832b16f048SDaniel Axtens */ 56842b16f048SDaniel Axtens static inline bool skb_gso_size_check(const struct sk_buff *skb, 56852b16f048SDaniel Axtens unsigned int seg_len, 56862b16f048SDaniel Axtens unsigned int max_len) { 56872b16f048SDaniel Axtens const struct skb_shared_info *shinfo = skb_shinfo(skb); 56882b16f048SDaniel Axtens const struct sk_buff *iter; 56892b16f048SDaniel Axtens 56902b16f048SDaniel Axtens if (shinfo->gso_size != GSO_BY_FRAGS) 56912b16f048SDaniel Axtens return seg_len <= max_len; 56922b16f048SDaniel Axtens 56932b16f048SDaniel Axtens /* Undo this so we can re-use header sizes */ 56942b16f048SDaniel Axtens seg_len -= GSO_BY_FRAGS; 56952b16f048SDaniel Axtens 56962b16f048SDaniel Axtens skb_walk_frags(skb, iter) { 56972b16f048SDaniel Axtens if (seg_len + skb_headlen(iter) > max_len) 56982b16f048SDaniel Axtens return false; 56992b16f048SDaniel Axtens } 57002b16f048SDaniel Axtens 57012b16f048SDaniel Axtens return true; 57022b16f048SDaniel Axtens } 57032b16f048SDaniel Axtens 57042b16f048SDaniel Axtens /** 5705779b7931SDaniel Axtens * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU? 5706ae7ef81eSMarcelo Ricardo Leitner * 5707ae7ef81eSMarcelo Ricardo Leitner * @skb: GSO skb 570876f21b99SDavid S. Miller * @mtu: MTU to validate against 5709ae7ef81eSMarcelo Ricardo Leitner * 5710779b7931SDaniel Axtens * skb_gso_validate_network_len validates if a given skb will fit a 5711779b7931SDaniel Axtens * wanted MTU once split. It considers L3 headers, L4 headers, and the 5712779b7931SDaniel Axtens * payload. 5713ae7ef81eSMarcelo Ricardo Leitner */ 5714779b7931SDaniel Axtens bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu) 5715ae7ef81eSMarcelo Ricardo Leitner { 57162b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu); 5717ae7ef81eSMarcelo Ricardo Leitner } 5718779b7931SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_network_len); 5719ae7ef81eSMarcelo Ricardo Leitner 57202b16f048SDaniel Axtens /** 57212b16f048SDaniel Axtens * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length? 57222b16f048SDaniel Axtens * 57232b16f048SDaniel Axtens * @skb: GSO skb 57242b16f048SDaniel Axtens * @len: length to validate against 57252b16f048SDaniel Axtens * 57262b16f048SDaniel Axtens * skb_gso_validate_mac_len validates if a given skb will fit a wanted 57272b16f048SDaniel Axtens * length once split, including L2, L3 and L4 headers and the payload. 57282b16f048SDaniel Axtens */ 57292b16f048SDaniel Axtens bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len) 57302b16f048SDaniel Axtens { 57312b16f048SDaniel Axtens return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len); 57322b16f048SDaniel Axtens } 57332b16f048SDaniel Axtens EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len); 57342b16f048SDaniel Axtens 57350d5501c1SVlad Yasevich static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 57360d5501c1SVlad Yasevich { 5737d85e8be2SYuya Kusakabe int mac_len, meta_len; 5738d85e8be2SYuya Kusakabe void *meta; 57394bbb3e0eSToshiaki Makita 57400d5501c1SVlad Yasevich if (skb_cow(skb, skb_headroom(skb)) < 0) { 57410d5501c1SVlad Yasevich kfree_skb(skb); 57420d5501c1SVlad Yasevich return NULL; 57430d5501c1SVlad Yasevich } 57440d5501c1SVlad Yasevich 57454bbb3e0eSToshiaki Makita mac_len = skb->data - skb_mac_header(skb); 5746ae474573SToshiaki Makita if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) { 57474bbb3e0eSToshiaki Makita memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb), 57484bbb3e0eSToshiaki Makita mac_len - VLAN_HLEN - ETH_TLEN); 5749ae474573SToshiaki Makita } 5750d85e8be2SYuya Kusakabe 5751d85e8be2SYuya Kusakabe meta_len = skb_metadata_len(skb); 5752d85e8be2SYuya Kusakabe if (meta_len) { 5753d85e8be2SYuya Kusakabe meta = skb_metadata_end(skb) - meta_len; 5754d85e8be2SYuya Kusakabe memmove(meta + VLAN_HLEN, meta, meta_len); 5755d85e8be2SYuya Kusakabe } 5756d85e8be2SYuya Kusakabe 57570d5501c1SVlad Yasevich skb->mac_header += VLAN_HLEN; 57580d5501c1SVlad Yasevich return skb; 57590d5501c1SVlad Yasevich } 57600d5501c1SVlad Yasevich 57610d5501c1SVlad Yasevich struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 57620d5501c1SVlad Yasevich { 57630d5501c1SVlad Yasevich struct vlan_hdr *vhdr; 57640d5501c1SVlad Yasevich u16 vlan_tci; 57650d5501c1SVlad Yasevich 5766df8a39deSJiri Pirko if (unlikely(skb_vlan_tag_present(skb))) { 57670d5501c1SVlad Yasevich /* vlan_tci is already set-up so leave this for another time */ 57680d5501c1SVlad Yasevich return skb; 57690d5501c1SVlad Yasevich } 57700d5501c1SVlad Yasevich 57710d5501c1SVlad Yasevich skb = skb_share_check(skb, GFP_ATOMIC); 57720d5501c1SVlad Yasevich if (unlikely(!skb)) 57730d5501c1SVlad Yasevich goto err_free; 577455eff0ebSMiaohe Lin /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */ 577555eff0ebSMiaohe Lin if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short)))) 57760d5501c1SVlad Yasevich goto err_free; 57770d5501c1SVlad Yasevich 57780d5501c1SVlad Yasevich vhdr = (struct vlan_hdr *)skb->data; 57790d5501c1SVlad Yasevich vlan_tci = ntohs(vhdr->h_vlan_TCI); 57800d5501c1SVlad Yasevich __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 57810d5501c1SVlad Yasevich 57820d5501c1SVlad Yasevich skb_pull_rcsum(skb, VLAN_HLEN); 57830d5501c1SVlad Yasevich vlan_set_encap_proto(skb, vhdr); 57840d5501c1SVlad Yasevich 57850d5501c1SVlad Yasevich skb = skb_reorder_vlan_header(skb); 57860d5501c1SVlad Yasevich if (unlikely(!skb)) 57870d5501c1SVlad Yasevich goto err_free; 57880d5501c1SVlad Yasevich 57890d5501c1SVlad Yasevich skb_reset_network_header(skb); 57908be33ecfSAlexander Lobakin if (!skb_transport_header_was_set(skb)) 57910d5501c1SVlad Yasevich skb_reset_transport_header(skb); 57920d5501c1SVlad Yasevich skb_reset_mac_len(skb); 57930d5501c1SVlad Yasevich 57940d5501c1SVlad Yasevich return skb; 57950d5501c1SVlad Yasevich 57960d5501c1SVlad Yasevich err_free: 57970d5501c1SVlad Yasevich kfree_skb(skb); 57980d5501c1SVlad Yasevich return NULL; 57990d5501c1SVlad Yasevich } 58000d5501c1SVlad Yasevich EXPORT_SYMBOL(skb_vlan_untag); 58012e4e4410SEric Dumazet 580292ece280SLiu Jian int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len) 5803e2195121SJiri Pirko { 5804e2195121SJiri Pirko if (!pskb_may_pull(skb, write_len)) 5805e2195121SJiri Pirko return -ENOMEM; 5806e2195121SJiri Pirko 5807e2195121SJiri Pirko if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 5808e2195121SJiri Pirko return 0; 5809e2195121SJiri Pirko 5810e2195121SJiri Pirko return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5811e2195121SJiri Pirko } 5812e2195121SJiri Pirko EXPORT_SYMBOL(skb_ensure_writable); 5813e2195121SJiri Pirko 5814bfca4c52SShmulik Ladkani /* remove VLAN header from packet and update csum accordingly. 5815bfca4c52SShmulik Ladkani * expects a non skb_vlan_tag_present skb with a vlan tag payload 5816bfca4c52SShmulik Ladkani */ 5817bfca4c52SShmulik Ladkani int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 581893515d53SJiri Pirko { 581993515d53SJiri Pirko struct vlan_hdr *vhdr; 5820b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 582193515d53SJiri Pirko int err; 582293515d53SJiri Pirko 5823b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5824b6a79208SShmulik Ladkani "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5825b6a79208SShmulik Ladkani offset)) { 5826b6a79208SShmulik Ladkani return -EINVAL; 5827b6a79208SShmulik Ladkani } 5828b6a79208SShmulik Ladkani 582993515d53SJiri Pirko err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 583093515d53SJiri Pirko if (unlikely(err)) 5831b6a79208SShmulik Ladkani return err; 583293515d53SJiri Pirko 583393515d53SJiri Pirko skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 583493515d53SJiri Pirko 583593515d53SJiri Pirko vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 583693515d53SJiri Pirko *vlan_tci = ntohs(vhdr->h_vlan_TCI); 583793515d53SJiri Pirko 583893515d53SJiri Pirko memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 583993515d53SJiri Pirko __skb_pull(skb, VLAN_HLEN); 584093515d53SJiri Pirko 584193515d53SJiri Pirko vlan_set_encap_proto(skb, vhdr); 584293515d53SJiri Pirko skb->mac_header += VLAN_HLEN; 584393515d53SJiri Pirko 584493515d53SJiri Pirko if (skb_network_offset(skb) < ETH_HLEN) 584593515d53SJiri Pirko skb_set_network_header(skb, ETH_HLEN); 584693515d53SJiri Pirko 584793515d53SJiri Pirko skb_reset_mac_len(skb); 584893515d53SJiri Pirko 584993515d53SJiri Pirko return err; 585093515d53SJiri Pirko } 5851bfca4c52SShmulik Ladkani EXPORT_SYMBOL(__skb_vlan_pop); 585293515d53SJiri Pirko 5853b6a79208SShmulik Ladkani /* Pop a vlan tag either from hwaccel or from payload. 5854b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5855b6a79208SShmulik Ladkani */ 585693515d53SJiri Pirko int skb_vlan_pop(struct sk_buff *skb) 585793515d53SJiri Pirko { 585893515d53SJiri Pirko u16 vlan_tci; 585993515d53SJiri Pirko __be16 vlan_proto; 586093515d53SJiri Pirko int err; 586193515d53SJiri Pirko 5862df8a39deSJiri Pirko if (likely(skb_vlan_tag_present(skb))) { 5863b1817524SMichał Mirosław __vlan_hwaccel_clear_tag(skb); 586493515d53SJiri Pirko } else { 5865ecf4ee41SShmulik Ladkani if (unlikely(!eth_type_vlan(skb->protocol))) 586693515d53SJiri Pirko return 0; 586793515d53SJiri Pirko 586893515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 586993515d53SJiri Pirko if (err) 587093515d53SJiri Pirko return err; 587193515d53SJiri Pirko } 587293515d53SJiri Pirko /* move next vlan tag to hw accel tag */ 5873ecf4ee41SShmulik Ladkani if (likely(!eth_type_vlan(skb->protocol))) 587493515d53SJiri Pirko return 0; 587593515d53SJiri Pirko 587693515d53SJiri Pirko vlan_proto = skb->protocol; 587793515d53SJiri Pirko err = __skb_vlan_pop(skb, &vlan_tci); 587893515d53SJiri Pirko if (unlikely(err)) 587993515d53SJiri Pirko return err; 588093515d53SJiri Pirko 588193515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 588293515d53SJiri Pirko return 0; 588393515d53SJiri Pirko } 588493515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_pop); 588593515d53SJiri Pirko 5886b6a79208SShmulik Ladkani /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5887b6a79208SShmulik Ladkani * Expects skb->data at mac header. 5888b6a79208SShmulik Ladkani */ 588993515d53SJiri Pirko int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 589093515d53SJiri Pirko { 5891df8a39deSJiri Pirko if (skb_vlan_tag_present(skb)) { 5892b6a79208SShmulik Ladkani int offset = skb->data - skb_mac_header(skb); 589393515d53SJiri Pirko int err; 589493515d53SJiri Pirko 5895b6a79208SShmulik Ladkani if (WARN_ONCE(offset, 5896b6a79208SShmulik Ladkani "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5897b6a79208SShmulik Ladkani offset)) { 5898b6a79208SShmulik Ladkani return -EINVAL; 5899b6a79208SShmulik Ladkani } 5900b6a79208SShmulik Ladkani 590193515d53SJiri Pirko err = __vlan_insert_tag(skb, skb->vlan_proto, 5902df8a39deSJiri Pirko skb_vlan_tag_get(skb)); 5903b6a79208SShmulik Ladkani if (err) 590493515d53SJiri Pirko return err; 59059241e2dfSDaniel Borkmann 590693515d53SJiri Pirko skb->protocol = skb->vlan_proto; 590793515d53SJiri Pirko skb->mac_len += VLAN_HLEN; 590893515d53SJiri Pirko 59096b83d28aSDaniel Borkmann skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 591093515d53SJiri Pirko } 591193515d53SJiri Pirko __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 591293515d53SJiri Pirko return 0; 591393515d53SJiri Pirko } 591493515d53SJiri Pirko EXPORT_SYMBOL(skb_vlan_push); 591593515d53SJiri Pirko 591619fbcb36SGuillaume Nault /** 591719fbcb36SGuillaume Nault * skb_eth_pop() - Drop the Ethernet header at the head of a packet 591819fbcb36SGuillaume Nault * 591919fbcb36SGuillaume Nault * @skb: Socket buffer to modify 592019fbcb36SGuillaume Nault * 592119fbcb36SGuillaume Nault * Drop the Ethernet header of @skb. 592219fbcb36SGuillaume Nault * 592319fbcb36SGuillaume Nault * Expects that skb->data points to the mac header and that no VLAN tags are 592419fbcb36SGuillaume Nault * present. 592519fbcb36SGuillaume Nault * 592619fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 592719fbcb36SGuillaume Nault */ 592819fbcb36SGuillaume Nault int skb_eth_pop(struct sk_buff *skb) 592919fbcb36SGuillaume Nault { 593019fbcb36SGuillaume Nault if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) || 593119fbcb36SGuillaume Nault skb_network_offset(skb) < ETH_HLEN) 593219fbcb36SGuillaume Nault return -EPROTO; 593319fbcb36SGuillaume Nault 593419fbcb36SGuillaume Nault skb_pull_rcsum(skb, ETH_HLEN); 593519fbcb36SGuillaume Nault skb_reset_mac_header(skb); 593619fbcb36SGuillaume Nault skb_reset_mac_len(skb); 593719fbcb36SGuillaume Nault 593819fbcb36SGuillaume Nault return 0; 593919fbcb36SGuillaume Nault } 594019fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_pop); 594119fbcb36SGuillaume Nault 594219fbcb36SGuillaume Nault /** 594319fbcb36SGuillaume Nault * skb_eth_push() - Add a new Ethernet header at the head of a packet 594419fbcb36SGuillaume Nault * 594519fbcb36SGuillaume Nault * @skb: Socket buffer to modify 594619fbcb36SGuillaume Nault * @dst: Destination MAC address of the new header 594719fbcb36SGuillaume Nault * @src: Source MAC address of the new header 594819fbcb36SGuillaume Nault * 594919fbcb36SGuillaume Nault * Prepend @skb with a new Ethernet header. 595019fbcb36SGuillaume Nault * 595119fbcb36SGuillaume Nault * Expects that skb->data points to the mac header, which must be empty. 595219fbcb36SGuillaume Nault * 595319fbcb36SGuillaume Nault * Returns 0 on success, -errno otherwise. 595419fbcb36SGuillaume Nault */ 595519fbcb36SGuillaume Nault int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 595619fbcb36SGuillaume Nault const unsigned char *src) 595719fbcb36SGuillaume Nault { 595819fbcb36SGuillaume Nault struct ethhdr *eth; 595919fbcb36SGuillaume Nault int err; 596019fbcb36SGuillaume Nault 596119fbcb36SGuillaume Nault if (skb_network_offset(skb) || skb_vlan_tag_present(skb)) 596219fbcb36SGuillaume Nault return -EPROTO; 596319fbcb36SGuillaume Nault 596419fbcb36SGuillaume Nault err = skb_cow_head(skb, sizeof(*eth)); 596519fbcb36SGuillaume Nault if (err < 0) 596619fbcb36SGuillaume Nault return err; 596719fbcb36SGuillaume Nault 596819fbcb36SGuillaume Nault skb_push(skb, sizeof(*eth)); 596919fbcb36SGuillaume Nault skb_reset_mac_header(skb); 597019fbcb36SGuillaume Nault skb_reset_mac_len(skb); 597119fbcb36SGuillaume Nault 597219fbcb36SGuillaume Nault eth = eth_hdr(skb); 597319fbcb36SGuillaume Nault ether_addr_copy(eth->h_dest, dst); 597419fbcb36SGuillaume Nault ether_addr_copy(eth->h_source, src); 597519fbcb36SGuillaume Nault eth->h_proto = skb->protocol; 597619fbcb36SGuillaume Nault 597719fbcb36SGuillaume Nault skb_postpush_rcsum(skb, eth, sizeof(*eth)); 597819fbcb36SGuillaume Nault 597919fbcb36SGuillaume Nault return 0; 598019fbcb36SGuillaume Nault } 598119fbcb36SGuillaume Nault EXPORT_SYMBOL(skb_eth_push); 598219fbcb36SGuillaume Nault 59838822e270SJohn Hurley /* Update the ethertype of hdr and the skb csum value if required. */ 59848822e270SJohn Hurley static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr, 59858822e270SJohn Hurley __be16 ethertype) 59868822e270SJohn Hurley { 59878822e270SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 59888822e270SJohn Hurley __be16 diff[] = { ~hdr->h_proto, ethertype }; 59898822e270SJohn Hurley 59908822e270SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 59918822e270SJohn Hurley } 59928822e270SJohn Hurley 59938822e270SJohn Hurley hdr->h_proto = ethertype; 59948822e270SJohn Hurley } 59958822e270SJohn Hurley 59968822e270SJohn Hurley /** 5997e7dbfed1SMartin Varghese * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of 5998e7dbfed1SMartin Varghese * the packet 59998822e270SJohn Hurley * 60008822e270SJohn Hurley * @skb: buffer 60018822e270SJohn Hurley * @mpls_lse: MPLS label stack entry to push 60028822e270SJohn Hurley * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848) 6003fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 6004e7dbfed1SMartin Varghese * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is 6005e7dbfed1SMartin Varghese * ethernet 60068822e270SJohn Hurley * 60078822e270SJohn Hurley * Expects skb->data at mac header. 60088822e270SJohn Hurley * 60098822e270SJohn Hurley * Returns 0 on success, -errno otherwise. 60108822e270SJohn Hurley */ 6011fa4e0f88SDavide Caratti int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 6012d04ac224SMartin Varghese int mac_len, bool ethernet) 60138822e270SJohn Hurley { 60148822e270SJohn Hurley struct mpls_shim_hdr *lse; 60158822e270SJohn Hurley int err; 60168822e270SJohn Hurley 60178822e270SJohn Hurley if (unlikely(!eth_p_mpls(mpls_proto))) 60188822e270SJohn Hurley return -EINVAL; 60198822e270SJohn Hurley 60208822e270SJohn Hurley /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */ 60218822e270SJohn Hurley if (skb->encapsulation) 60228822e270SJohn Hurley return -EINVAL; 60238822e270SJohn Hurley 60248822e270SJohn Hurley err = skb_cow_head(skb, MPLS_HLEN); 60258822e270SJohn Hurley if (unlikely(err)) 60268822e270SJohn Hurley return err; 60278822e270SJohn Hurley 60288822e270SJohn Hurley if (!skb->inner_protocol) { 6029e7dbfed1SMartin Varghese skb_set_inner_network_header(skb, skb_network_offset(skb)); 60308822e270SJohn Hurley skb_set_inner_protocol(skb, skb->protocol); 60318822e270SJohn Hurley } 60328822e270SJohn Hurley 60338822e270SJohn Hurley skb_push(skb, MPLS_HLEN); 60348822e270SJohn Hurley memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 6035fa4e0f88SDavide Caratti mac_len); 60368822e270SJohn Hurley skb_reset_mac_header(skb); 6037fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 6038e7dbfed1SMartin Varghese skb_reset_mac_len(skb); 60398822e270SJohn Hurley 60408822e270SJohn Hurley lse = mpls_hdr(skb); 60418822e270SJohn Hurley lse->label_stack_entry = mpls_lse; 60428822e270SJohn Hurley skb_postpush_rcsum(skb, lse, MPLS_HLEN); 60438822e270SJohn Hurley 60444296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) 60458822e270SJohn Hurley skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto); 60468822e270SJohn Hurley skb->protocol = mpls_proto; 60478822e270SJohn Hurley 60488822e270SJohn Hurley return 0; 60498822e270SJohn Hurley } 60508822e270SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_push); 60518822e270SJohn Hurley 60522e4e4410SEric Dumazet /** 6053ed246ceeSJohn Hurley * skb_mpls_pop() - pop the outermost MPLS header 6054ed246ceeSJohn Hurley * 6055ed246ceeSJohn Hurley * @skb: buffer 6056ed246ceeSJohn Hurley * @next_proto: ethertype of header after popped MPLS header 6057fa4e0f88SDavide Caratti * @mac_len: length of the MAC header 605876f99f98SMartin Varghese * @ethernet: flag to indicate if the packet is ethernet 6059ed246ceeSJohn Hurley * 6060ed246ceeSJohn Hurley * Expects skb->data at mac header. 6061ed246ceeSJohn Hurley * 6062ed246ceeSJohn Hurley * Returns 0 on success, -errno otherwise. 6063ed246ceeSJohn Hurley */ 6064040b5cfbSMartin Varghese int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 6065040b5cfbSMartin Varghese bool ethernet) 6066ed246ceeSJohn Hurley { 6067ed246ceeSJohn Hurley int err; 6068ed246ceeSJohn Hurley 6069ed246ceeSJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 6070dedc5a08SDavide Caratti return 0; 6071ed246ceeSJohn Hurley 6072fa4e0f88SDavide Caratti err = skb_ensure_writable(skb, mac_len + MPLS_HLEN); 6073ed246ceeSJohn Hurley if (unlikely(err)) 6074ed246ceeSJohn Hurley return err; 6075ed246ceeSJohn Hurley 6076ed246ceeSJohn Hurley skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 6077ed246ceeSJohn Hurley memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 6078fa4e0f88SDavide Caratti mac_len); 6079ed246ceeSJohn Hurley 6080ed246ceeSJohn Hurley __skb_pull(skb, MPLS_HLEN); 6081ed246ceeSJohn Hurley skb_reset_mac_header(skb); 6082fa4e0f88SDavide Caratti skb_set_network_header(skb, mac_len); 6083ed246ceeSJohn Hurley 60844296adc3SGuillaume Nault if (ethernet && mac_len >= ETH_HLEN) { 6085ed246ceeSJohn Hurley struct ethhdr *hdr; 6086ed246ceeSJohn Hurley 6087ed246ceeSJohn Hurley /* use mpls_hdr() to get ethertype to account for VLANs. */ 6088ed246ceeSJohn Hurley hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 6089ed246ceeSJohn Hurley skb_mod_eth_type(skb, hdr, next_proto); 6090ed246ceeSJohn Hurley } 6091ed246ceeSJohn Hurley skb->protocol = next_proto; 6092ed246ceeSJohn Hurley 6093ed246ceeSJohn Hurley return 0; 6094ed246ceeSJohn Hurley } 6095ed246ceeSJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_pop); 6096ed246ceeSJohn Hurley 6097ed246ceeSJohn Hurley /** 6098d27cf5c5SJohn Hurley * skb_mpls_update_lse() - modify outermost MPLS header and update csum 6099d27cf5c5SJohn Hurley * 6100d27cf5c5SJohn Hurley * @skb: buffer 6101d27cf5c5SJohn Hurley * @mpls_lse: new MPLS label stack entry to update to 6102d27cf5c5SJohn Hurley * 6103d27cf5c5SJohn Hurley * Expects skb->data at mac header. 6104d27cf5c5SJohn Hurley * 6105d27cf5c5SJohn Hurley * Returns 0 on success, -errno otherwise. 6106d27cf5c5SJohn Hurley */ 6107d27cf5c5SJohn Hurley int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse) 6108d27cf5c5SJohn Hurley { 6109d27cf5c5SJohn Hurley int err; 6110d27cf5c5SJohn Hurley 6111d27cf5c5SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 6112d27cf5c5SJohn Hurley return -EINVAL; 6113d27cf5c5SJohn Hurley 6114d27cf5c5SJohn Hurley err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 6115d27cf5c5SJohn Hurley if (unlikely(err)) 6116d27cf5c5SJohn Hurley return err; 6117d27cf5c5SJohn Hurley 6118d27cf5c5SJohn Hurley if (skb->ip_summed == CHECKSUM_COMPLETE) { 6119d27cf5c5SJohn Hurley __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse }; 6120d27cf5c5SJohn Hurley 6121d27cf5c5SJohn Hurley skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum); 6122d27cf5c5SJohn Hurley } 6123d27cf5c5SJohn Hurley 6124d27cf5c5SJohn Hurley mpls_hdr(skb)->label_stack_entry = mpls_lse; 6125d27cf5c5SJohn Hurley 6126d27cf5c5SJohn Hurley return 0; 6127d27cf5c5SJohn Hurley } 6128d27cf5c5SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_update_lse); 6129d27cf5c5SJohn Hurley 6130d27cf5c5SJohn Hurley /** 61312a2ea508SJohn Hurley * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header 61322a2ea508SJohn Hurley * 61332a2ea508SJohn Hurley * @skb: buffer 61342a2ea508SJohn Hurley * 61352a2ea508SJohn Hurley * Expects skb->data at mac header. 61362a2ea508SJohn Hurley * 61372a2ea508SJohn Hurley * Returns 0 on success, -errno otherwise. 61382a2ea508SJohn Hurley */ 61392a2ea508SJohn Hurley int skb_mpls_dec_ttl(struct sk_buff *skb) 61402a2ea508SJohn Hurley { 61412a2ea508SJohn Hurley u32 lse; 61422a2ea508SJohn Hurley u8 ttl; 61432a2ea508SJohn Hurley 61442a2ea508SJohn Hurley if (unlikely(!eth_p_mpls(skb->protocol))) 61452a2ea508SJohn Hurley return -EINVAL; 61462a2ea508SJohn Hurley 614713de4ed9SDavide Caratti if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 614813de4ed9SDavide Caratti return -ENOMEM; 614913de4ed9SDavide Caratti 61502a2ea508SJohn Hurley lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry); 61512a2ea508SJohn Hurley ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT; 61522a2ea508SJohn Hurley if (!--ttl) 61532a2ea508SJohn Hurley return -EINVAL; 61542a2ea508SJohn Hurley 61552a2ea508SJohn Hurley lse &= ~MPLS_LS_TTL_MASK; 61562a2ea508SJohn Hurley lse |= ttl << MPLS_LS_TTL_SHIFT; 61572a2ea508SJohn Hurley 61582a2ea508SJohn Hurley return skb_mpls_update_lse(skb, cpu_to_be32(lse)); 61592a2ea508SJohn Hurley } 61602a2ea508SJohn Hurley EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl); 61612a2ea508SJohn Hurley 61622a2ea508SJohn Hurley /** 61632e4e4410SEric Dumazet * alloc_skb_with_frags - allocate skb with page frags 61642e4e4410SEric Dumazet * 6165de3f0d0eSMasanari Iida * @header_len: size of linear part 6166de3f0d0eSMasanari Iida * @data_len: needed length in frags 6167de3f0d0eSMasanari Iida * @max_page_order: max page order desired. 6168de3f0d0eSMasanari Iida * @errcode: pointer to error code if any 6169de3f0d0eSMasanari Iida * @gfp_mask: allocation mask 61702e4e4410SEric Dumazet * 61712e4e4410SEric Dumazet * This can be used to allocate a paged skb, given a maximal order for frags. 61722e4e4410SEric Dumazet */ 61732e4e4410SEric Dumazet struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 61742e4e4410SEric Dumazet unsigned long data_len, 61752e4e4410SEric Dumazet int max_page_order, 61762e4e4410SEric Dumazet int *errcode, 61772e4e4410SEric Dumazet gfp_t gfp_mask) 61782e4e4410SEric Dumazet { 61792e4e4410SEric Dumazet int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 61802e4e4410SEric Dumazet unsigned long chunk; 61812e4e4410SEric Dumazet struct sk_buff *skb; 61822e4e4410SEric Dumazet struct page *page; 61832e4e4410SEric Dumazet int i; 61842e4e4410SEric Dumazet 61852e4e4410SEric Dumazet *errcode = -EMSGSIZE; 61862e4e4410SEric Dumazet /* Note this test could be relaxed, if we succeed to allocate 61872e4e4410SEric Dumazet * high order pages... 61882e4e4410SEric Dumazet */ 61892e4e4410SEric Dumazet if (npages > MAX_SKB_FRAGS) 61902e4e4410SEric Dumazet return NULL; 61912e4e4410SEric Dumazet 61922e4e4410SEric Dumazet *errcode = -ENOBUFS; 6193f8c468e8SDavid Rientjes skb = alloc_skb(header_len, gfp_mask); 61942e4e4410SEric Dumazet if (!skb) 61952e4e4410SEric Dumazet return NULL; 61962e4e4410SEric Dumazet 61972e4e4410SEric Dumazet skb->truesize += npages << PAGE_SHIFT; 61982e4e4410SEric Dumazet 61992e4e4410SEric Dumazet for (i = 0; npages > 0; i++) { 62002e4e4410SEric Dumazet int order = max_page_order; 62012e4e4410SEric Dumazet 62022e4e4410SEric Dumazet while (order) { 62032e4e4410SEric Dumazet if (npages >= 1 << order) { 6204d0164adcSMel Gorman page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 62052e4e4410SEric Dumazet __GFP_COMP | 6206d14b56f5SMichal Hocko __GFP_NOWARN, 62072e4e4410SEric Dumazet order); 62082e4e4410SEric Dumazet if (page) 62092e4e4410SEric Dumazet goto fill_page; 62102e4e4410SEric Dumazet /* Do not retry other high order allocations */ 62112e4e4410SEric Dumazet order = 1; 62122e4e4410SEric Dumazet max_page_order = 0; 62132e4e4410SEric Dumazet } 62142e4e4410SEric Dumazet order--; 62152e4e4410SEric Dumazet } 62162e4e4410SEric Dumazet page = alloc_page(gfp_mask); 62172e4e4410SEric Dumazet if (!page) 62182e4e4410SEric Dumazet goto failure; 62192e4e4410SEric Dumazet fill_page: 62202e4e4410SEric Dumazet chunk = min_t(unsigned long, data_len, 62212e4e4410SEric Dumazet PAGE_SIZE << order); 62222e4e4410SEric Dumazet skb_fill_page_desc(skb, i, page, 0, chunk); 62232e4e4410SEric Dumazet data_len -= chunk; 62242e4e4410SEric Dumazet npages -= 1 << order; 62252e4e4410SEric Dumazet } 62262e4e4410SEric Dumazet return skb; 62272e4e4410SEric Dumazet 62282e4e4410SEric Dumazet failure: 62292e4e4410SEric Dumazet kfree_skb(skb); 62302e4e4410SEric Dumazet return NULL; 62312e4e4410SEric Dumazet } 62322e4e4410SEric Dumazet EXPORT_SYMBOL(alloc_skb_with_frags); 62336fa01ccdSSowmini Varadhan 62346fa01ccdSSowmini Varadhan /* carve out the first off bytes from skb when off < headlen */ 62356fa01ccdSSowmini Varadhan static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 62366fa01ccdSSowmini Varadhan const int headlen, gfp_t gfp_mask) 62376fa01ccdSSowmini Varadhan { 62386fa01ccdSSowmini Varadhan int i; 623912d6c1d3SKees Cook unsigned int size = skb_end_offset(skb); 62406fa01ccdSSowmini Varadhan int new_hlen = headlen - off; 62416fa01ccdSSowmini Varadhan u8 *data; 62426fa01ccdSSowmini Varadhan 62436fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 62446fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 624512d6c1d3SKees Cook 624612d6c1d3SKees Cook size = SKB_DATA_ALIGN(size); 624712d6c1d3SKees Cook size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 624812d6c1d3SKees Cook size = kmalloc_size_roundup(size); 624912d6c1d3SKees Cook data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL); 62506fa01ccdSSowmini Varadhan if (!data) 62516fa01ccdSSowmini Varadhan return -ENOMEM; 625212d6c1d3SKees Cook size = SKB_WITH_OVERHEAD(size); 62536fa01ccdSSowmini Varadhan 62546fa01ccdSSowmini Varadhan /* Copy real data, and all frags */ 62556fa01ccdSSowmini Varadhan skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 62566fa01ccdSSowmini Varadhan skb->len -= off; 62576fa01ccdSSowmini Varadhan 62586fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 62596fa01ccdSSowmini Varadhan skb_shinfo(skb), 62606fa01ccdSSowmini Varadhan offsetof(struct skb_shared_info, 62616fa01ccdSSowmini Varadhan frags[skb_shinfo(skb)->nr_frags])); 62626fa01ccdSSowmini Varadhan if (skb_cloned(skb)) { 62636fa01ccdSSowmini Varadhan /* drop the old head gracefully */ 62646fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 62656fa01ccdSSowmini Varadhan kfree(data); 62666fa01ccdSSowmini Varadhan return -ENOMEM; 62676fa01ccdSSowmini Varadhan } 62686fa01ccdSSowmini Varadhan for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 62696fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 62706fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 62716fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 6272511a3edaSEric Dumazet skb_release_data(skb, SKB_CONSUMED); 62736fa01ccdSSowmini Varadhan } else { 62746fa01ccdSSowmini Varadhan /* we can reuse existing recount- all we did was 62756fa01ccdSSowmini Varadhan * relocate values 62766fa01ccdSSowmini Varadhan */ 62776fa01ccdSSowmini Varadhan skb_free_head(skb); 62786fa01ccdSSowmini Varadhan } 62796fa01ccdSSowmini Varadhan 62806fa01ccdSSowmini Varadhan skb->head = data; 62816fa01ccdSSowmini Varadhan skb->data = data; 62826fa01ccdSSowmini Varadhan skb->head_frag = 0; 6283763087daSEric Dumazet skb_set_end_offset(skb, size); 62846fa01ccdSSowmini Varadhan skb_set_tail_pointer(skb, skb_headlen(skb)); 62856fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 62866fa01ccdSSowmini Varadhan skb->cloned = 0; 62876fa01ccdSSowmini Varadhan skb->hdr_len = 0; 62886fa01ccdSSowmini Varadhan skb->nohdr = 0; 62896fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 62906fa01ccdSSowmini Varadhan 62916fa01ccdSSowmini Varadhan return 0; 62926fa01ccdSSowmini Varadhan } 62936fa01ccdSSowmini Varadhan 62946fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 62956fa01ccdSSowmini Varadhan 62966fa01ccdSSowmini Varadhan /* carve out the first eat bytes from skb's frag_list. May recurse into 62976fa01ccdSSowmini Varadhan * pskb_carve() 62986fa01ccdSSowmini Varadhan */ 62996fa01ccdSSowmini Varadhan static int pskb_carve_frag_list(struct sk_buff *skb, 63006fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo, int eat, 63016fa01ccdSSowmini Varadhan gfp_t gfp_mask) 63026fa01ccdSSowmini Varadhan { 63036fa01ccdSSowmini Varadhan struct sk_buff *list = shinfo->frag_list; 63046fa01ccdSSowmini Varadhan struct sk_buff *clone = NULL; 63056fa01ccdSSowmini Varadhan struct sk_buff *insp = NULL; 63066fa01ccdSSowmini Varadhan 63076fa01ccdSSowmini Varadhan do { 63086fa01ccdSSowmini Varadhan if (!list) { 63096fa01ccdSSowmini Varadhan pr_err("Not enough bytes to eat. Want %d\n", eat); 63106fa01ccdSSowmini Varadhan return -EFAULT; 63116fa01ccdSSowmini Varadhan } 63126fa01ccdSSowmini Varadhan if (list->len <= eat) { 63136fa01ccdSSowmini Varadhan /* Eaten as whole. */ 63146fa01ccdSSowmini Varadhan eat -= list->len; 63156fa01ccdSSowmini Varadhan list = list->next; 63166fa01ccdSSowmini Varadhan insp = list; 63176fa01ccdSSowmini Varadhan } else { 63186fa01ccdSSowmini Varadhan /* Eaten partially. */ 63196fa01ccdSSowmini Varadhan if (skb_shared(list)) { 63206fa01ccdSSowmini Varadhan clone = skb_clone(list, gfp_mask); 63216fa01ccdSSowmini Varadhan if (!clone) 63226fa01ccdSSowmini Varadhan return -ENOMEM; 63236fa01ccdSSowmini Varadhan insp = list->next; 63246fa01ccdSSowmini Varadhan list = clone; 63256fa01ccdSSowmini Varadhan } else { 63266fa01ccdSSowmini Varadhan /* This may be pulled without problems. */ 63276fa01ccdSSowmini Varadhan insp = list; 63286fa01ccdSSowmini Varadhan } 63296fa01ccdSSowmini Varadhan if (pskb_carve(list, eat, gfp_mask) < 0) { 63306fa01ccdSSowmini Varadhan kfree_skb(clone); 63316fa01ccdSSowmini Varadhan return -ENOMEM; 63326fa01ccdSSowmini Varadhan } 63336fa01ccdSSowmini Varadhan break; 63346fa01ccdSSowmini Varadhan } 63356fa01ccdSSowmini Varadhan } while (eat); 63366fa01ccdSSowmini Varadhan 63376fa01ccdSSowmini Varadhan /* Free pulled out fragments. */ 63386fa01ccdSSowmini Varadhan while ((list = shinfo->frag_list) != insp) { 63396fa01ccdSSowmini Varadhan shinfo->frag_list = list->next; 6340ef527f96SEric Dumazet consume_skb(list); 63416fa01ccdSSowmini Varadhan } 63426fa01ccdSSowmini Varadhan /* And insert new clone at head. */ 63436fa01ccdSSowmini Varadhan if (clone) { 63446fa01ccdSSowmini Varadhan clone->next = list; 63456fa01ccdSSowmini Varadhan shinfo->frag_list = clone; 63466fa01ccdSSowmini Varadhan } 63476fa01ccdSSowmini Varadhan return 0; 63486fa01ccdSSowmini Varadhan } 63496fa01ccdSSowmini Varadhan 63506fa01ccdSSowmini Varadhan /* carve off first len bytes from skb. Split line (off) is in the 63516fa01ccdSSowmini Varadhan * non-linear part of skb 63526fa01ccdSSowmini Varadhan */ 63536fa01ccdSSowmini Varadhan static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 63546fa01ccdSSowmini Varadhan int pos, gfp_t gfp_mask) 63556fa01ccdSSowmini Varadhan { 63566fa01ccdSSowmini Varadhan int i, k = 0; 635712d6c1d3SKees Cook unsigned int size = skb_end_offset(skb); 63586fa01ccdSSowmini Varadhan u8 *data; 63596fa01ccdSSowmini Varadhan const int nfrags = skb_shinfo(skb)->nr_frags; 63606fa01ccdSSowmini Varadhan struct skb_shared_info *shinfo; 63616fa01ccdSSowmini Varadhan 63626fa01ccdSSowmini Varadhan if (skb_pfmemalloc(skb)) 63636fa01ccdSSowmini Varadhan gfp_mask |= __GFP_MEMALLOC; 636412d6c1d3SKees Cook 636512d6c1d3SKees Cook size = SKB_DATA_ALIGN(size); 636612d6c1d3SKees Cook size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 636712d6c1d3SKees Cook size = kmalloc_size_roundup(size); 636812d6c1d3SKees Cook data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL); 63696fa01ccdSSowmini Varadhan if (!data) 63706fa01ccdSSowmini Varadhan return -ENOMEM; 637112d6c1d3SKees Cook size = SKB_WITH_OVERHEAD(size); 63726fa01ccdSSowmini Varadhan 63736fa01ccdSSowmini Varadhan memcpy((struct skb_shared_info *)(data + size), 6374e3ec1e8cSMiaohe Lin skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0])); 63756fa01ccdSSowmini Varadhan if (skb_orphan_frags(skb, gfp_mask)) { 63766fa01ccdSSowmini Varadhan kfree(data); 63776fa01ccdSSowmini Varadhan return -ENOMEM; 63786fa01ccdSSowmini Varadhan } 63796fa01ccdSSowmini Varadhan shinfo = (struct skb_shared_info *)(data + size); 63806fa01ccdSSowmini Varadhan for (i = 0; i < nfrags; i++) { 63816fa01ccdSSowmini Varadhan int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 63826fa01ccdSSowmini Varadhan 63836fa01ccdSSowmini Varadhan if (pos + fsize > off) { 63846fa01ccdSSowmini Varadhan shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 63856fa01ccdSSowmini Varadhan 63866fa01ccdSSowmini Varadhan if (pos < off) { 63876fa01ccdSSowmini Varadhan /* Split frag. 63886fa01ccdSSowmini Varadhan * We have two variants in this case: 63896fa01ccdSSowmini Varadhan * 1. Move all the frag to the second 63906fa01ccdSSowmini Varadhan * part, if it is possible. F.e. 63916fa01ccdSSowmini Varadhan * this approach is mandatory for TUX, 63926fa01ccdSSowmini Varadhan * where splitting is expensive. 63936fa01ccdSSowmini Varadhan * 2. Split is accurately. We make this. 63946fa01ccdSSowmini Varadhan */ 6395b54c9d5bSJonathan Lemon skb_frag_off_add(&shinfo->frags[0], off - pos); 63966fa01ccdSSowmini Varadhan skb_frag_size_sub(&shinfo->frags[0], off - pos); 63976fa01ccdSSowmini Varadhan } 63986fa01ccdSSowmini Varadhan skb_frag_ref(skb, i); 63996fa01ccdSSowmini Varadhan k++; 64006fa01ccdSSowmini Varadhan } 64016fa01ccdSSowmini Varadhan pos += fsize; 64026fa01ccdSSowmini Varadhan } 64036fa01ccdSSowmini Varadhan shinfo->nr_frags = k; 64046fa01ccdSSowmini Varadhan if (skb_has_frag_list(skb)) 64056fa01ccdSSowmini Varadhan skb_clone_fraglist(skb); 64066fa01ccdSSowmini Varadhan 64076fa01ccdSSowmini Varadhan /* split line is in frag list */ 6408eabe8618SMiaohe Lin if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) { 6409eabe8618SMiaohe Lin /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */ 6410eabe8618SMiaohe Lin if (skb_has_frag_list(skb)) 6411eabe8618SMiaohe Lin kfree_skb_list(skb_shinfo(skb)->frag_list); 6412eabe8618SMiaohe Lin kfree(data); 6413eabe8618SMiaohe Lin return -ENOMEM; 64146fa01ccdSSowmini Varadhan } 6415511a3edaSEric Dumazet skb_release_data(skb, SKB_CONSUMED); 64166fa01ccdSSowmini Varadhan 64176fa01ccdSSowmini Varadhan skb->head = data; 64186fa01ccdSSowmini Varadhan skb->head_frag = 0; 64196fa01ccdSSowmini Varadhan skb->data = data; 6420763087daSEric Dumazet skb_set_end_offset(skb, size); 64216fa01ccdSSowmini Varadhan skb_reset_tail_pointer(skb); 64226fa01ccdSSowmini Varadhan skb_headers_offset_update(skb, 0); 64236fa01ccdSSowmini Varadhan skb->cloned = 0; 64246fa01ccdSSowmini Varadhan skb->hdr_len = 0; 64256fa01ccdSSowmini Varadhan skb->nohdr = 0; 64266fa01ccdSSowmini Varadhan skb->len -= off; 64276fa01ccdSSowmini Varadhan skb->data_len = skb->len; 64286fa01ccdSSowmini Varadhan atomic_set(&skb_shinfo(skb)->dataref, 1); 64296fa01ccdSSowmini Varadhan return 0; 64306fa01ccdSSowmini Varadhan } 64316fa01ccdSSowmini Varadhan 64326fa01ccdSSowmini Varadhan /* remove len bytes from the beginning of the skb */ 64336fa01ccdSSowmini Varadhan static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 64346fa01ccdSSowmini Varadhan { 64356fa01ccdSSowmini Varadhan int headlen = skb_headlen(skb); 64366fa01ccdSSowmini Varadhan 64376fa01ccdSSowmini Varadhan if (len < headlen) 64386fa01ccdSSowmini Varadhan return pskb_carve_inside_header(skb, len, headlen, gfp); 64396fa01ccdSSowmini Varadhan else 64406fa01ccdSSowmini Varadhan return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 64416fa01ccdSSowmini Varadhan } 64426fa01ccdSSowmini Varadhan 64436fa01ccdSSowmini Varadhan /* Extract to_copy bytes starting at off from skb, and return this in 64446fa01ccdSSowmini Varadhan * a new skb 64456fa01ccdSSowmini Varadhan */ 64466fa01ccdSSowmini Varadhan struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 64476fa01ccdSSowmini Varadhan int to_copy, gfp_t gfp) 64486fa01ccdSSowmini Varadhan { 64496fa01ccdSSowmini Varadhan struct sk_buff *clone = skb_clone(skb, gfp); 64506fa01ccdSSowmini Varadhan 64516fa01ccdSSowmini Varadhan if (!clone) 64526fa01ccdSSowmini Varadhan return NULL; 64536fa01ccdSSowmini Varadhan 64546fa01ccdSSowmini Varadhan if (pskb_carve(clone, off, gfp) < 0 || 64556fa01ccdSSowmini Varadhan pskb_trim(clone, to_copy)) { 64566fa01ccdSSowmini Varadhan kfree_skb(clone); 64576fa01ccdSSowmini Varadhan return NULL; 64586fa01ccdSSowmini Varadhan } 64596fa01ccdSSowmini Varadhan return clone; 64606fa01ccdSSowmini Varadhan } 64616fa01ccdSSowmini Varadhan EXPORT_SYMBOL(pskb_extract); 6462c8c8b127SEric Dumazet 6463c8c8b127SEric Dumazet /** 6464c8c8b127SEric Dumazet * skb_condense - try to get rid of fragments/frag_list if possible 6465c8c8b127SEric Dumazet * @skb: buffer 6466c8c8b127SEric Dumazet * 6467c8c8b127SEric Dumazet * Can be used to save memory before skb is added to a busy queue. 6468c8c8b127SEric Dumazet * If packet has bytes in frags and enough tail room in skb->head, 6469c8c8b127SEric Dumazet * pull all of them, so that we can free the frags right now and adjust 6470c8c8b127SEric Dumazet * truesize. 6471c8c8b127SEric Dumazet * Notes: 6472c8c8b127SEric Dumazet * We do not reallocate skb->head thus can not fail. 6473c8c8b127SEric Dumazet * Caller must re-evaluate skb->truesize if needed. 6474c8c8b127SEric Dumazet */ 6475c8c8b127SEric Dumazet void skb_condense(struct sk_buff *skb) 6476c8c8b127SEric Dumazet { 64773174fed9SEric Dumazet if (skb->data_len) { 64783174fed9SEric Dumazet if (skb->data_len > skb->end - skb->tail || 6479c8c8b127SEric Dumazet skb_cloned(skb)) 6480c8c8b127SEric Dumazet return; 6481c8c8b127SEric Dumazet 6482c8c8b127SEric Dumazet /* Nice, we can free page frag(s) right now */ 6483c8c8b127SEric Dumazet __pskb_pull_tail(skb, skb->data_len); 64843174fed9SEric Dumazet } 64853174fed9SEric Dumazet /* At this point, skb->truesize might be over estimated, 64863174fed9SEric Dumazet * because skb had a fragment, and fragments do not tell 64873174fed9SEric Dumazet * their truesize. 64883174fed9SEric Dumazet * When we pulled its content into skb->head, fragment 64893174fed9SEric Dumazet * was freed, but __pskb_pull_tail() could not possibly 64903174fed9SEric Dumazet * adjust skb->truesize, not knowing the frag truesize. 6491c8c8b127SEric Dumazet */ 6492c8c8b127SEric Dumazet skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6493c8c8b127SEric Dumazet } 6494d57a3a15SDavid Howells EXPORT_SYMBOL(skb_condense); 6495df5042f4SFlorian Westphal 6496df5042f4SFlorian Westphal #ifdef CONFIG_SKB_EXTENSIONS 6497df5042f4SFlorian Westphal static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id) 6498df5042f4SFlorian Westphal { 6499df5042f4SFlorian Westphal return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE); 6500df5042f4SFlorian Westphal } 6501df5042f4SFlorian Westphal 65028b69a803SPaolo Abeni /** 65038b69a803SPaolo Abeni * __skb_ext_alloc - allocate a new skb extensions storage 65048b69a803SPaolo Abeni * 65054930f483SFlorian Westphal * @flags: See kmalloc(). 65064930f483SFlorian Westphal * 65078b69a803SPaolo Abeni * Returns the newly allocated pointer. The pointer can later attached to a 65088b69a803SPaolo Abeni * skb via __skb_ext_set(). 65098b69a803SPaolo Abeni * Note: caller must handle the skb_ext as an opaque data. 65108b69a803SPaolo Abeni */ 65114930f483SFlorian Westphal struct skb_ext *__skb_ext_alloc(gfp_t flags) 6512df5042f4SFlorian Westphal { 65134930f483SFlorian Westphal struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags); 6514df5042f4SFlorian Westphal 6515df5042f4SFlorian Westphal if (new) { 6516df5042f4SFlorian Westphal memset(new->offset, 0, sizeof(new->offset)); 6517df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6518df5042f4SFlorian Westphal } 6519df5042f4SFlorian Westphal 6520df5042f4SFlorian Westphal return new; 6521df5042f4SFlorian Westphal } 6522df5042f4SFlorian Westphal 65234165079bSFlorian Westphal static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old, 65244165079bSFlorian Westphal unsigned int old_active) 6525df5042f4SFlorian Westphal { 6526df5042f4SFlorian Westphal struct skb_ext *new; 6527df5042f4SFlorian Westphal 6528df5042f4SFlorian Westphal if (refcount_read(&old->refcnt) == 1) 6529df5042f4SFlorian Westphal return old; 6530df5042f4SFlorian Westphal 6531df5042f4SFlorian Westphal new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC); 6532df5042f4SFlorian Westphal if (!new) 6533df5042f4SFlorian Westphal return NULL; 6534df5042f4SFlorian Westphal 6535df5042f4SFlorian Westphal memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE); 6536df5042f4SFlorian Westphal refcount_set(&new->refcnt, 1); 6537df5042f4SFlorian Westphal 65384165079bSFlorian Westphal #ifdef CONFIG_XFRM 65394165079bSFlorian Westphal if (old_active & (1 << SKB_EXT_SEC_PATH)) { 65404165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH); 65414165079bSFlorian Westphal unsigned int i; 65424165079bSFlorian Westphal 65434165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 65444165079bSFlorian Westphal xfrm_state_hold(sp->xvec[i]); 65454165079bSFlorian Westphal } 65464165079bSFlorian Westphal #endif 6547df5042f4SFlorian Westphal __skb_ext_put(old); 6548df5042f4SFlorian Westphal return new; 6549df5042f4SFlorian Westphal } 6550df5042f4SFlorian Westphal 6551df5042f4SFlorian Westphal /** 65528b69a803SPaolo Abeni * __skb_ext_set - attach the specified extension storage to this skb 65538b69a803SPaolo Abeni * @skb: buffer 65548b69a803SPaolo Abeni * @id: extension id 65558b69a803SPaolo Abeni * @ext: extension storage previously allocated via __skb_ext_alloc() 65568b69a803SPaolo Abeni * 65578b69a803SPaolo Abeni * Existing extensions, if any, are cleared. 65588b69a803SPaolo Abeni * 65598b69a803SPaolo Abeni * Returns the pointer to the extension. 65608b69a803SPaolo Abeni */ 65618b69a803SPaolo Abeni void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 65628b69a803SPaolo Abeni struct skb_ext *ext) 65638b69a803SPaolo Abeni { 65648b69a803SPaolo Abeni unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext); 65658b69a803SPaolo Abeni 65668b69a803SPaolo Abeni skb_ext_put(skb); 65678b69a803SPaolo Abeni newlen = newoff + skb_ext_type_len[id]; 65688b69a803SPaolo Abeni ext->chunks = newlen; 65698b69a803SPaolo Abeni ext->offset[id] = newoff; 65708b69a803SPaolo Abeni skb->extensions = ext; 65718b69a803SPaolo Abeni skb->active_extensions = 1 << id; 65728b69a803SPaolo Abeni return skb_ext_get_ptr(ext, id); 65738b69a803SPaolo Abeni } 65748b69a803SPaolo Abeni 65758b69a803SPaolo Abeni /** 6576df5042f4SFlorian Westphal * skb_ext_add - allocate space for given extension, COW if needed 6577df5042f4SFlorian Westphal * @skb: buffer 6578df5042f4SFlorian Westphal * @id: extension to allocate space for 6579df5042f4SFlorian Westphal * 6580df5042f4SFlorian Westphal * Allocates enough space for the given extension. 6581df5042f4SFlorian Westphal * If the extension is already present, a pointer to that extension 6582df5042f4SFlorian Westphal * is returned. 6583df5042f4SFlorian Westphal * 6584df5042f4SFlorian Westphal * If the skb was cloned, COW applies and the returned memory can be 6585df5042f4SFlorian Westphal * modified without changing the extension space of clones buffers. 6586df5042f4SFlorian Westphal * 6587df5042f4SFlorian Westphal * Returns pointer to the extension or NULL on allocation failure. 6588df5042f4SFlorian Westphal */ 6589df5042f4SFlorian Westphal void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id) 6590df5042f4SFlorian Westphal { 6591df5042f4SFlorian Westphal struct skb_ext *new, *old = NULL; 6592df5042f4SFlorian Westphal unsigned int newlen, newoff; 6593df5042f4SFlorian Westphal 6594df5042f4SFlorian Westphal if (skb->active_extensions) { 6595df5042f4SFlorian Westphal old = skb->extensions; 6596df5042f4SFlorian Westphal 65974165079bSFlorian Westphal new = skb_ext_maybe_cow(old, skb->active_extensions); 6598df5042f4SFlorian Westphal if (!new) 6599df5042f4SFlorian Westphal return NULL; 6600df5042f4SFlorian Westphal 6601682ec859SPaolo Abeni if (__skb_ext_exist(new, id)) 6602df5042f4SFlorian Westphal goto set_active; 6603df5042f4SFlorian Westphal 6604e94e50bdSPaolo Abeni newoff = new->chunks; 6605df5042f4SFlorian Westphal } else { 6606df5042f4SFlorian Westphal newoff = SKB_EXT_CHUNKSIZEOF(*new); 6607df5042f4SFlorian Westphal 66084930f483SFlorian Westphal new = __skb_ext_alloc(GFP_ATOMIC); 6609df5042f4SFlorian Westphal if (!new) 6610df5042f4SFlorian Westphal return NULL; 6611df5042f4SFlorian Westphal } 6612df5042f4SFlorian Westphal 6613df5042f4SFlorian Westphal newlen = newoff + skb_ext_type_len[id]; 6614df5042f4SFlorian Westphal new->chunks = newlen; 6615df5042f4SFlorian Westphal new->offset[id] = newoff; 6616df5042f4SFlorian Westphal set_active: 6617b0999f38SPaolo Abeni skb->slow_gro = 1; 6618682ec859SPaolo Abeni skb->extensions = new; 6619df5042f4SFlorian Westphal skb->active_extensions |= 1 << id; 6620df5042f4SFlorian Westphal return skb_ext_get_ptr(new, id); 6621df5042f4SFlorian Westphal } 6622df5042f4SFlorian Westphal EXPORT_SYMBOL(skb_ext_add); 6623df5042f4SFlorian Westphal 66244165079bSFlorian Westphal #ifdef CONFIG_XFRM 66254165079bSFlorian Westphal static void skb_ext_put_sp(struct sec_path *sp) 66264165079bSFlorian Westphal { 66274165079bSFlorian Westphal unsigned int i; 66284165079bSFlorian Westphal 66294165079bSFlorian Westphal for (i = 0; i < sp->len; i++) 66304165079bSFlorian Westphal xfrm_state_put(sp->xvec[i]); 66314165079bSFlorian Westphal } 66324165079bSFlorian Westphal #endif 66334165079bSFlorian Westphal 663478476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 663578476d31SJeremy Kerr static void skb_ext_put_mctp(struct mctp_flow *flow) 663678476d31SJeremy Kerr { 663778476d31SJeremy Kerr if (flow->key) 663878476d31SJeremy Kerr mctp_key_unref(flow->key); 663978476d31SJeremy Kerr } 664078476d31SJeremy Kerr #endif 664178476d31SJeremy Kerr 6642df5042f4SFlorian Westphal void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 6643df5042f4SFlorian Westphal { 6644df5042f4SFlorian Westphal struct skb_ext *ext = skb->extensions; 6645df5042f4SFlorian Westphal 6646df5042f4SFlorian Westphal skb->active_extensions &= ~(1 << id); 6647df5042f4SFlorian Westphal if (skb->active_extensions == 0) { 6648df5042f4SFlorian Westphal skb->extensions = NULL; 6649df5042f4SFlorian Westphal __skb_ext_put(ext); 66504165079bSFlorian Westphal #ifdef CONFIG_XFRM 66514165079bSFlorian Westphal } else if (id == SKB_EXT_SEC_PATH && 66524165079bSFlorian Westphal refcount_read(&ext->refcnt) == 1) { 66534165079bSFlorian Westphal struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH); 66544165079bSFlorian Westphal 66554165079bSFlorian Westphal skb_ext_put_sp(sp); 66564165079bSFlorian Westphal sp->len = 0; 66574165079bSFlorian Westphal #endif 6658df5042f4SFlorian Westphal } 6659df5042f4SFlorian Westphal } 6660df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_del); 6661df5042f4SFlorian Westphal 6662df5042f4SFlorian Westphal void __skb_ext_put(struct skb_ext *ext) 6663df5042f4SFlorian Westphal { 6664df5042f4SFlorian Westphal /* If this is last clone, nothing can increment 6665df5042f4SFlorian Westphal * it after check passes. Avoids one atomic op. 6666df5042f4SFlorian Westphal */ 6667df5042f4SFlorian Westphal if (refcount_read(&ext->refcnt) == 1) 6668df5042f4SFlorian Westphal goto free_now; 6669df5042f4SFlorian Westphal 6670df5042f4SFlorian Westphal if (!refcount_dec_and_test(&ext->refcnt)) 6671df5042f4SFlorian Westphal return; 6672df5042f4SFlorian Westphal free_now: 66734165079bSFlorian Westphal #ifdef CONFIG_XFRM 66744165079bSFlorian Westphal if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH)) 66754165079bSFlorian Westphal skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH)); 66764165079bSFlorian Westphal #endif 667778476d31SJeremy Kerr #ifdef CONFIG_MCTP_FLOWS 667878476d31SJeremy Kerr if (__skb_ext_exist(ext, SKB_EXT_MCTP)) 667978476d31SJeremy Kerr skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP)); 668078476d31SJeremy Kerr #endif 66814165079bSFlorian Westphal 6682df5042f4SFlorian Westphal kmem_cache_free(skbuff_ext_cache, ext); 6683df5042f4SFlorian Westphal } 6684df5042f4SFlorian Westphal EXPORT_SYMBOL(__skb_ext_put); 6685df5042f4SFlorian Westphal #endif /* CONFIG_SKB_EXTENSIONS */ 668668822bdfSEric Dumazet 668768822bdfSEric Dumazet /** 668868822bdfSEric Dumazet * skb_attempt_defer_free - queue skb for remote freeing 668968822bdfSEric Dumazet * @skb: buffer 669068822bdfSEric Dumazet * 669168822bdfSEric Dumazet * Put @skb in a per-cpu list, using the cpu which 669268822bdfSEric Dumazet * allocated the skb/pages to reduce false sharing 669368822bdfSEric Dumazet * and memory zone spinlock contention. 669468822bdfSEric Dumazet */ 669568822bdfSEric Dumazet void skb_attempt_defer_free(struct sk_buff *skb) 669668822bdfSEric Dumazet { 669768822bdfSEric Dumazet int cpu = skb->alloc_cpu; 669868822bdfSEric Dumazet struct softnet_data *sd; 669968822bdfSEric Dumazet unsigned long flags; 670039564c3fSEric Dumazet unsigned int defer_max; 670168822bdfSEric Dumazet bool kick; 670268822bdfSEric Dumazet 670368822bdfSEric Dumazet if (WARN_ON_ONCE(cpu >= nr_cpu_ids) || 670468822bdfSEric Dumazet !cpu_online(cpu) || 670568822bdfSEric Dumazet cpu == raw_smp_processor_id()) { 670639564c3fSEric Dumazet nodefer: __kfree_skb(skb); 670768822bdfSEric Dumazet return; 670868822bdfSEric Dumazet } 670968822bdfSEric Dumazet 671068822bdfSEric Dumazet sd = &per_cpu(softnet_data, cpu); 671139564c3fSEric Dumazet defer_max = READ_ONCE(sysctl_skb_defer_max); 671239564c3fSEric Dumazet if (READ_ONCE(sd->defer_count) >= defer_max) 671339564c3fSEric Dumazet goto nodefer; 671439564c3fSEric Dumazet 671568822bdfSEric Dumazet spin_lock_irqsave(&sd->defer_lock, flags); 6716c09b0cd2SJakub Kicinski /* Send an IPI every time queue reaches half capacity. */ 6717c09b0cd2SJakub Kicinski kick = sd->defer_count == (defer_max >> 1); 6718c09b0cd2SJakub Kicinski /* Paired with the READ_ONCE() few lines above */ 6719c09b0cd2SJakub Kicinski WRITE_ONCE(sd->defer_count, sd->defer_count + 1); 6720c09b0cd2SJakub Kicinski 672168822bdfSEric Dumazet skb->next = sd->defer_list; 672268822bdfSEric Dumazet /* Paired with READ_ONCE() in skb_defer_free_flush() */ 672368822bdfSEric Dumazet WRITE_ONCE(sd->defer_list, skb); 672468822bdfSEric Dumazet spin_unlock_irqrestore(&sd->defer_lock, flags); 672568822bdfSEric Dumazet 672668822bdfSEric Dumazet /* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU 672768822bdfSEric Dumazet * if we are unlucky enough (this seems very unlikely). 672868822bdfSEric Dumazet */ 672997e719a8SEric Dumazet if (unlikely(kick) && !cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) 673068822bdfSEric Dumazet smp_call_function_single_async(cpu, &sd->defer_csd); 673168822bdfSEric Dumazet } 6732