11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 31da177e4SLinus Torvalds * 4113aa838SAlan Cox * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 51da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Fixes: 81da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 91da177e4SLinus Torvalds * balancer bugs. 101da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 111da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 121da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 131da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 141da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 151da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 161da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 171da177e4SLinus Torvalds * only put in the headers 181da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 191da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 201da177e4SLinus Torvalds * Andi Kleen : slabified it. 211da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 221da177e4SLinus Torvalds * 231da177e4SLinus Torvalds * NOTE: 241da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 251da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 261da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 271da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 281da177e4SLinus Torvalds * 291da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or 301da177e4SLinus Torvalds * modify it under the terms of the GNU General Public License 311da177e4SLinus Torvalds * as published by the Free Software Foundation; either version 321da177e4SLinus Torvalds * 2 of the License, or (at your option) any later version. 331da177e4SLinus Torvalds */ 341da177e4SLinus Torvalds 351da177e4SLinus Torvalds /* 361da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 371da177e4SLinus Torvalds */ 381da177e4SLinus Torvalds 39e005d193SJoe Perches #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 40e005d193SJoe Perches 411da177e4SLinus Torvalds #include <linux/module.h> 421da177e4SLinus Torvalds #include <linux/types.h> 431da177e4SLinus Torvalds #include <linux/kernel.h> 44fe55f6d5SVegard Nossum #include <linux/kmemcheck.h> 451da177e4SLinus Torvalds #include <linux/mm.h> 461da177e4SLinus Torvalds #include <linux/interrupt.h> 471da177e4SLinus Torvalds #include <linux/in.h> 481da177e4SLinus Torvalds #include <linux/inet.h> 491da177e4SLinus Torvalds #include <linux/slab.h> 501da177e4SLinus Torvalds #include <linux/netdevice.h> 511da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 521da177e4SLinus Torvalds #include <net/pkt_sched.h> 531da177e4SLinus Torvalds #endif 541da177e4SLinus Torvalds #include <linux/string.h> 551da177e4SLinus Torvalds #include <linux/skbuff.h> 569c55e01cSJens Axboe #include <linux/splice.h> 571da177e4SLinus Torvalds #include <linux/cache.h> 581da177e4SLinus Torvalds #include <linux/rtnetlink.h> 591da177e4SLinus Torvalds #include <linux/init.h> 60716ea3a7SDavid Howells #include <linux/scatterlist.h> 61ac45f602SPatrick Ohly #include <linux/errqueue.h> 62268bb0ceSLinus Torvalds #include <linux/prefetch.h> 631da177e4SLinus Torvalds 641da177e4SLinus Torvalds #include <net/protocol.h> 651da177e4SLinus Torvalds #include <net/dst.h> 661da177e4SLinus Torvalds #include <net/sock.h> 671da177e4SLinus Torvalds #include <net/checksum.h> 681da177e4SLinus Torvalds #include <net/xfrm.h> 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds #include <asm/uaccess.h> 71ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7251c56b00SEric Dumazet #include <linux/highmem.h> 73a1f8e7f7SAl Viro 74d7e8883cSEric Dumazet struct kmem_cache *skbuff_head_cache __read_mostly; 75e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 761da177e4SLinus Torvalds 779c55e01cSJens Axboe static void sock_pipe_buf_release(struct pipe_inode_info *pipe, 789c55e01cSJens Axboe struct pipe_buffer *buf) 799c55e01cSJens Axboe { 808b9d3728SJarek Poplawski put_page(buf->page); 819c55e01cSJens Axboe } 829c55e01cSJens Axboe 839c55e01cSJens Axboe static void sock_pipe_buf_get(struct pipe_inode_info *pipe, 849c55e01cSJens Axboe struct pipe_buffer *buf) 859c55e01cSJens Axboe { 868b9d3728SJarek Poplawski get_page(buf->page); 879c55e01cSJens Axboe } 889c55e01cSJens Axboe 899c55e01cSJens Axboe static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, 909c55e01cSJens Axboe struct pipe_buffer *buf) 919c55e01cSJens Axboe { 929c55e01cSJens Axboe return 1; 939c55e01cSJens Axboe } 949c55e01cSJens Axboe 959c55e01cSJens Axboe 969c55e01cSJens Axboe /* Pipe buffer operations for a socket. */ 9728dfef8fSAlexey Dobriyan static const struct pipe_buf_operations sock_pipe_buf_ops = { 989c55e01cSJens Axboe .can_merge = 0, 999c55e01cSJens Axboe .map = generic_pipe_buf_map, 1009c55e01cSJens Axboe .unmap = generic_pipe_buf_unmap, 1019c55e01cSJens Axboe .confirm = generic_pipe_buf_confirm, 1029c55e01cSJens Axboe .release = sock_pipe_buf_release, 1039c55e01cSJens Axboe .steal = sock_pipe_buf_steal, 1049c55e01cSJens Axboe .get = sock_pipe_buf_get, 1059c55e01cSJens Axboe }; 1069c55e01cSJens Axboe 1071da177e4SLinus Torvalds /** 108f05de73bSJean Sacren * skb_panic - private function for out-of-line support 1091da177e4SLinus Torvalds * @skb: buffer 1101da177e4SLinus Torvalds * @sz: size 111f05de73bSJean Sacren * @addr: address 11299d5851eSJames Hogan * @msg: skb_over_panic or skb_under_panic 1131da177e4SLinus Torvalds * 114f05de73bSJean Sacren * Out-of-line support for skb_put() and skb_push(). 115f05de73bSJean Sacren * Called via the wrapper skb_over_panic() or skb_under_panic(). 116f05de73bSJean Sacren * Keep out of line to prevent kernel bloat. 117f05de73bSJean Sacren * __builtin_return_address is not used because it is not always reliable. 1181da177e4SLinus Torvalds */ 119f05de73bSJean Sacren static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 12099d5851eSJames Hogan const char msg[]) 1211da177e4SLinus Torvalds { 122e005d193SJoe Perches pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 12399d5851eSJames Hogan msg, addr, skb->len, sz, skb->head, skb->data, 1244305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 12526095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1261da177e4SLinus Torvalds BUG(); 1271da177e4SLinus Torvalds } 1281da177e4SLinus Torvalds 129f05de73bSJean Sacren static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 1301da177e4SLinus Torvalds { 131f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 1321da177e4SLinus Torvalds } 1331da177e4SLinus Torvalds 134f05de73bSJean Sacren static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 135f05de73bSJean Sacren { 136f05de73bSJean Sacren skb_panic(skb, sz, addr, __func__); 137f05de73bSJean Sacren } 138c93bdd0eSMel Gorman 139c93bdd0eSMel Gorman /* 140c93bdd0eSMel Gorman * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 141c93bdd0eSMel Gorman * the caller if emergency pfmemalloc reserves are being used. If it is and 142c93bdd0eSMel Gorman * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 143c93bdd0eSMel Gorman * may be used. Otherwise, the packet data may be discarded until enough 144c93bdd0eSMel Gorman * memory is free 145c93bdd0eSMel Gorman */ 146c93bdd0eSMel Gorman #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 147c93bdd0eSMel Gorman __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 14861c5e88aSstephen hemminger 14961c5e88aSstephen hemminger static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 15061c5e88aSstephen hemminger unsigned long ip, bool *pfmemalloc) 151c93bdd0eSMel Gorman { 152c93bdd0eSMel Gorman void *obj; 153c93bdd0eSMel Gorman bool ret_pfmemalloc = false; 154c93bdd0eSMel Gorman 155c93bdd0eSMel Gorman /* 156c93bdd0eSMel Gorman * Try a regular allocation, when that fails and we're not entitled 157c93bdd0eSMel Gorman * to the reserves, fail. 158c93bdd0eSMel Gorman */ 159c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, 160c93bdd0eSMel Gorman flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 161c93bdd0eSMel Gorman node); 162c93bdd0eSMel Gorman if (obj || !(gfp_pfmemalloc_allowed(flags))) 163c93bdd0eSMel Gorman goto out; 164c93bdd0eSMel Gorman 165c93bdd0eSMel Gorman /* Try again but now we are using pfmemalloc reserves */ 166c93bdd0eSMel Gorman ret_pfmemalloc = true; 167c93bdd0eSMel Gorman obj = kmalloc_node_track_caller(size, flags, node); 168c93bdd0eSMel Gorman 169c93bdd0eSMel Gorman out: 170c93bdd0eSMel Gorman if (pfmemalloc) 171c93bdd0eSMel Gorman *pfmemalloc = ret_pfmemalloc; 172c93bdd0eSMel Gorman 173c93bdd0eSMel Gorman return obj; 174c93bdd0eSMel Gorman } 175c93bdd0eSMel Gorman 1761da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1771da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1781da177e4SLinus Torvalds * [BEEP] leaks. 1791da177e4SLinus Torvalds * 1801da177e4SLinus Torvalds */ 1811da177e4SLinus Torvalds 1820ebd0ac5SPatrick McHardy struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node) 1830ebd0ac5SPatrick McHardy { 1840ebd0ac5SPatrick McHardy struct sk_buff *skb; 1850ebd0ac5SPatrick McHardy 1860ebd0ac5SPatrick McHardy /* Get the HEAD */ 1870ebd0ac5SPatrick McHardy skb = kmem_cache_alloc_node(skbuff_head_cache, 1880ebd0ac5SPatrick McHardy gfp_mask & ~__GFP_DMA, node); 1890ebd0ac5SPatrick McHardy if (!skb) 1900ebd0ac5SPatrick McHardy goto out; 1910ebd0ac5SPatrick McHardy 1920ebd0ac5SPatrick McHardy /* 1930ebd0ac5SPatrick McHardy * Only clear those fields we need to clear, not those that we will 1940ebd0ac5SPatrick McHardy * actually initialise below. Hence, don't put any more fields after 1950ebd0ac5SPatrick McHardy * the tail pointer in struct sk_buff! 1960ebd0ac5SPatrick McHardy */ 1970ebd0ac5SPatrick McHardy memset(skb, 0, offsetof(struct sk_buff, tail)); 1985e71d9d7SPablo Neira skb->head = NULL; 1990ebd0ac5SPatrick McHardy skb->truesize = sizeof(struct sk_buff); 2000ebd0ac5SPatrick McHardy atomic_set(&skb->users, 1); 2010ebd0ac5SPatrick McHardy 20235d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 2030ebd0ac5SPatrick McHardy out: 2040ebd0ac5SPatrick McHardy return skb; 2050ebd0ac5SPatrick McHardy } 2060ebd0ac5SPatrick McHardy 2071da177e4SLinus Torvalds /** 208d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 2091da177e4SLinus Torvalds * @size: size to allocate 2101da177e4SLinus Torvalds * @gfp_mask: allocation mask 211c93bdd0eSMel Gorman * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 212c93bdd0eSMel Gorman * instead of head cache and allocate a cloned (child) skb. 213c93bdd0eSMel Gorman * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 214c93bdd0eSMel Gorman * allocations in case the data is required for writeback 215b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 2161da177e4SLinus Torvalds * 2171da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 21894b6042cSBen Hutchings * tail room of at least size bytes. The object has a reference count 21994b6042cSBen Hutchings * of one. The return is the buffer. On a failure the return is %NULL. 2201da177e4SLinus Torvalds * 2211da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 2221da177e4SLinus Torvalds * %GFP_ATOMIC. 2231da177e4SLinus Torvalds */ 224dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 225c93bdd0eSMel Gorman int flags, int node) 2261da177e4SLinus Torvalds { 227e18b890bSChristoph Lameter struct kmem_cache *cache; 2284947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 2291da177e4SLinus Torvalds struct sk_buff *skb; 2301da177e4SLinus Torvalds u8 *data; 231c93bdd0eSMel Gorman bool pfmemalloc; 2321da177e4SLinus Torvalds 233c93bdd0eSMel Gorman cache = (flags & SKB_ALLOC_FCLONE) 234c93bdd0eSMel Gorman ? skbuff_fclone_cache : skbuff_head_cache; 235c93bdd0eSMel Gorman 236c93bdd0eSMel Gorman if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 237c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 2388798b3fbSHerbert Xu 2391da177e4SLinus Torvalds /* Get the HEAD */ 240b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 2411da177e4SLinus Torvalds if (!skb) 2421da177e4SLinus Torvalds goto out; 243ec7d2f2cSEric Dumazet prefetchw(skb); 2441da177e4SLinus Torvalds 24587fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 24687fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 24787fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 24887fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 24987fb4b7bSEric Dumazet */ 250bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 25187fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 252c93bdd0eSMel Gorman data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 2531da177e4SLinus Torvalds if (!data) 2541da177e4SLinus Torvalds goto nodata; 25587fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 25687fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 25787fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 25887fb4b7bSEric Dumazet */ 25987fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 260ec7d2f2cSEric Dumazet prefetchw(data + size); 2611da177e4SLinus Torvalds 262ca0605a7SArnaldo Carvalho de Melo /* 263c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 264c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 265c8005785SJohannes Berg * the tail pointer in struct sk_buff! 266ca0605a7SArnaldo Carvalho de Melo */ 267ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 26887fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 26987fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 270c93bdd0eSMel Gorman skb->pfmemalloc = pfmemalloc; 2711da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2721da177e4SLinus Torvalds skb->head = data; 2731da177e4SLinus Torvalds skb->data = data; 27427a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2754305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 27635d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 27735d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 27819633e12SStephen Hemminger 2794947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2804947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 281ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2824947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 283c2aa3665SEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 2844947d3efSBenjamin LaHaise 285c93bdd0eSMel Gorman if (flags & SKB_ALLOC_FCLONE) { 286d179cd12SDavid S. Miller struct sk_buff *child = skb + 1; 287d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (child + 1); 2881da177e4SLinus Torvalds 289fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags1); 290fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags2); 291d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 292d179cd12SDavid S. Miller atomic_set(fclone_ref, 1); 293d179cd12SDavid S. Miller 294d179cd12SDavid S. Miller child->fclone = SKB_FCLONE_UNAVAILABLE; 295c93bdd0eSMel Gorman child->pfmemalloc = pfmemalloc; 296d179cd12SDavid S. Miller } 2971da177e4SLinus Torvalds out: 2981da177e4SLinus Torvalds return skb; 2991da177e4SLinus Torvalds nodata: 3008798b3fbSHerbert Xu kmem_cache_free(cache, skb); 3011da177e4SLinus Torvalds skb = NULL; 3021da177e4SLinus Torvalds goto out; 3031da177e4SLinus Torvalds } 304b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 3051da177e4SLinus Torvalds 3061da177e4SLinus Torvalds /** 307b2b5ce9dSEric Dumazet * build_skb - build a network buffer 308b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 309d3836f21SEric Dumazet * @frag_size: size of fragment, or 0 if head was kmalloced 310b2b5ce9dSEric Dumazet * 311b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 312deceb4c0SFlorian Fainelli * skb_shared_info. @data must have been allocated by kmalloc() only if 313deceb4c0SFlorian Fainelli * @frag_size is 0, otherwise data should come from the page allocator. 314b2b5ce9dSEric Dumazet * The return is the new skb buffer. 315b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 316b2b5ce9dSEric Dumazet * Notes : 317b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 318b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 319b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 320b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 321b2b5ce9dSEric Dumazet * before giving packet to stack. 322b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 323b2b5ce9dSEric Dumazet */ 324d3836f21SEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 325b2b5ce9dSEric Dumazet { 326b2b5ce9dSEric Dumazet struct skb_shared_info *shinfo; 327b2b5ce9dSEric Dumazet struct sk_buff *skb; 328d3836f21SEric Dumazet unsigned int size = frag_size ? : ksize(data); 329b2b5ce9dSEric Dumazet 330b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 331b2b5ce9dSEric Dumazet if (!skb) 332b2b5ce9dSEric Dumazet return NULL; 333b2b5ce9dSEric Dumazet 334d3836f21SEric Dumazet size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 335b2b5ce9dSEric Dumazet 336b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 337b2b5ce9dSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 338d3836f21SEric Dumazet skb->head_frag = frag_size != 0; 339b2b5ce9dSEric Dumazet atomic_set(&skb->users, 1); 340b2b5ce9dSEric Dumazet skb->head = data; 341b2b5ce9dSEric Dumazet skb->data = data; 342b2b5ce9dSEric Dumazet skb_reset_tail_pointer(skb); 343b2b5ce9dSEric Dumazet skb->end = skb->tail + size; 34435d04610SCong Wang skb->mac_header = (typeof(skb->mac_header))~0U; 34535d04610SCong Wang skb->transport_header = (typeof(skb->transport_header))~0U; 346b2b5ce9dSEric Dumazet 347b2b5ce9dSEric Dumazet /* make sure we initialize shinfo sequentially */ 348b2b5ce9dSEric Dumazet shinfo = skb_shinfo(skb); 349b2b5ce9dSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 350b2b5ce9dSEric Dumazet atomic_set(&shinfo->dataref, 1); 351b2b5ce9dSEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 352b2b5ce9dSEric Dumazet 353b2b5ce9dSEric Dumazet return skb; 354b2b5ce9dSEric Dumazet } 355b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 356b2b5ce9dSEric Dumazet 357a1c7fff7SEric Dumazet struct netdev_alloc_cache { 35869b08f62SEric Dumazet struct page_frag frag; 35969b08f62SEric Dumazet /* we maintain a pagecount bias, so that we dont dirty cache line 36069b08f62SEric Dumazet * containing page->_count every time we allocate a fragment. 36169b08f62SEric Dumazet */ 362540eb7bfSAlexander Duyck unsigned int pagecnt_bias; 363a1c7fff7SEric Dumazet }; 364a1c7fff7SEric Dumazet static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache); 365a1c7fff7SEric Dumazet 366c93bdd0eSMel Gorman static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 3676f532612SEric Dumazet { 3686f532612SEric Dumazet struct netdev_alloc_cache *nc; 3696f532612SEric Dumazet void *data = NULL; 37069b08f62SEric Dumazet int order; 3716f532612SEric Dumazet unsigned long flags; 3726f532612SEric Dumazet 3736f532612SEric Dumazet local_irq_save(flags); 3746f532612SEric Dumazet nc = &__get_cpu_var(netdev_alloc_cache); 37569b08f62SEric Dumazet if (unlikely(!nc->frag.page)) { 3766f532612SEric Dumazet refill: 37769b08f62SEric Dumazet for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) { 37869b08f62SEric Dumazet gfp_t gfp = gfp_mask; 37969b08f62SEric Dumazet 38069b08f62SEric Dumazet if (order) 38169b08f62SEric Dumazet gfp |= __GFP_COMP | __GFP_NOWARN; 38269b08f62SEric Dumazet nc->frag.page = alloc_pages(gfp, order); 38369b08f62SEric Dumazet if (likely(nc->frag.page)) 38469b08f62SEric Dumazet break; 38569b08f62SEric Dumazet if (--order < 0) 386540eb7bfSAlexander Duyck goto end; 38769b08f62SEric Dumazet } 38869b08f62SEric Dumazet nc->frag.size = PAGE_SIZE << order; 389540eb7bfSAlexander Duyck recycle: 39069b08f62SEric Dumazet atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS); 39169b08f62SEric Dumazet nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS; 39269b08f62SEric Dumazet nc->frag.offset = 0; 3936f532612SEric Dumazet } 394540eb7bfSAlexander Duyck 39569b08f62SEric Dumazet if (nc->frag.offset + fragsz > nc->frag.size) { 396540eb7bfSAlexander Duyck /* avoid unnecessary locked operations if possible */ 39769b08f62SEric Dumazet if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) || 39869b08f62SEric Dumazet atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count)) 399540eb7bfSAlexander Duyck goto recycle; 4006f532612SEric Dumazet goto refill; 4016f532612SEric Dumazet } 402540eb7bfSAlexander Duyck 40369b08f62SEric Dumazet data = page_address(nc->frag.page) + nc->frag.offset; 40469b08f62SEric Dumazet nc->frag.offset += fragsz; 405540eb7bfSAlexander Duyck nc->pagecnt_bias--; 406540eb7bfSAlexander Duyck end: 4076f532612SEric Dumazet local_irq_restore(flags); 4086f532612SEric Dumazet return data; 4096f532612SEric Dumazet } 410c93bdd0eSMel Gorman 411c93bdd0eSMel Gorman /** 412c93bdd0eSMel Gorman * netdev_alloc_frag - allocate a page fragment 413c93bdd0eSMel Gorman * @fragsz: fragment size 414c93bdd0eSMel Gorman * 415c93bdd0eSMel Gorman * Allocates a frag from a page for receive buffer. 416c93bdd0eSMel Gorman * Uses GFP_ATOMIC allocations. 417c93bdd0eSMel Gorman */ 418c93bdd0eSMel Gorman void *netdev_alloc_frag(unsigned int fragsz) 419c93bdd0eSMel Gorman { 420c93bdd0eSMel Gorman return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 421c93bdd0eSMel Gorman } 4226f532612SEric Dumazet EXPORT_SYMBOL(netdev_alloc_frag); 4236f532612SEric Dumazet 4246f532612SEric Dumazet /** 4258af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 4268af27456SChristoph Hellwig * @dev: network device to receive on 4278af27456SChristoph Hellwig * @length: length to allocate 4288af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 4298af27456SChristoph Hellwig * 4308af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 4318af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 4328af27456SChristoph Hellwig * the headroom they think they need without accounting for the 4338af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 4348af27456SChristoph Hellwig * 4358af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 4368af27456SChristoph Hellwig */ 4378af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 4388af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 4398af27456SChristoph Hellwig { 4406f532612SEric Dumazet struct sk_buff *skb = NULL; 441a1c7fff7SEric Dumazet unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) + 442a1c7fff7SEric Dumazet SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4438af27456SChristoph Hellwig 444310e158cSEric Dumazet if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) { 445c93bdd0eSMel Gorman void *data; 446c93bdd0eSMel Gorman 447c93bdd0eSMel Gorman if (sk_memalloc_socks()) 448c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 449c93bdd0eSMel Gorman 450c93bdd0eSMel Gorman data = __netdev_alloc_frag(fragsz, gfp_mask); 451a1c7fff7SEric Dumazet 4526f532612SEric Dumazet if (likely(data)) { 4536f532612SEric Dumazet skb = build_skb(data, fragsz); 4546f532612SEric Dumazet if (unlikely(!skb)) 4556f532612SEric Dumazet put_page(virt_to_head_page(data)); 456a1c7fff7SEric Dumazet } 457a1c7fff7SEric Dumazet } else { 458c93bdd0eSMel Gorman skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 459c93bdd0eSMel Gorman SKB_ALLOC_RX, NUMA_NO_NODE); 460a1c7fff7SEric Dumazet } 4617b2e497aSChristoph Hellwig if (likely(skb)) { 4628af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 4637b2e497aSChristoph Hellwig skb->dev = dev; 4647b2e497aSChristoph Hellwig } 4658af27456SChristoph Hellwig return skb; 4668af27456SChristoph Hellwig } 467b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 4681da177e4SLinus Torvalds 469654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 47050269e19SEric Dumazet int size, unsigned int truesize) 471654bed16SPeter Zijlstra { 472654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 473654bed16SPeter Zijlstra skb->len += size; 474654bed16SPeter Zijlstra skb->data_len += size; 47550269e19SEric Dumazet skb->truesize += truesize; 476654bed16SPeter Zijlstra } 477654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 478654bed16SPeter Zijlstra 479f8e617e1SJason Wang void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 480f8e617e1SJason Wang unsigned int truesize) 481f8e617e1SJason Wang { 482f8e617e1SJason Wang skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 483f8e617e1SJason Wang 484f8e617e1SJason Wang skb_frag_size_add(frag, size); 485f8e617e1SJason Wang skb->len += size; 486f8e617e1SJason Wang skb->data_len += size; 487f8e617e1SJason Wang skb->truesize += truesize; 488f8e617e1SJason Wang } 489f8e617e1SJason Wang EXPORT_SYMBOL(skb_coalesce_rx_frag); 490f8e617e1SJason Wang 49127b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 4921da177e4SLinus Torvalds { 493bd8a7036SEric Dumazet kfree_skb_list(*listp); 49427b437c8SHerbert Xu *listp = NULL; 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 49727b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 49827b437c8SHerbert Xu { 49927b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 50027b437c8SHerbert Xu } 50127b437c8SHerbert Xu 5021da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 5031da177e4SLinus Torvalds { 5041da177e4SLinus Torvalds struct sk_buff *list; 5051da177e4SLinus Torvalds 506fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 5071da177e4SLinus Torvalds skb_get(list); 5081da177e4SLinus Torvalds } 5091da177e4SLinus Torvalds 510d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 511d3836f21SEric Dumazet { 512d3836f21SEric Dumazet if (skb->head_frag) 513d3836f21SEric Dumazet put_page(virt_to_head_page(skb->head)); 514d3836f21SEric Dumazet else 515d3836f21SEric Dumazet kfree(skb->head); 516d3836f21SEric Dumazet } 517d3836f21SEric Dumazet 5185bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 5191da177e4SLinus Torvalds { 5201da177e4SLinus Torvalds if (!skb->cloned || 5211da177e4SLinus Torvalds !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 5221da177e4SLinus Torvalds &skb_shinfo(skb)->dataref)) { 5231da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 5241da177e4SLinus Torvalds int i; 5251da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 526ea2ab693SIan Campbell skb_frag_unref(skb, i); 5271da177e4SLinus Torvalds } 5281da177e4SLinus Torvalds 529a6686f2fSShirley Ma /* 530a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 531a6686f2fSShirley Ma * the lower device DMA has done; 532a6686f2fSShirley Ma */ 533a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 534a6686f2fSShirley Ma struct ubuf_info *uarg; 535a6686f2fSShirley Ma 536a6686f2fSShirley Ma uarg = skb_shinfo(skb)->destructor_arg; 537a6686f2fSShirley Ma if (uarg->callback) 538e19d6763SMichael S. Tsirkin uarg->callback(uarg, true); 539a6686f2fSShirley Ma } 540a6686f2fSShirley Ma 54121dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 5421da177e4SLinus Torvalds skb_drop_fraglist(skb); 5431da177e4SLinus Torvalds 544d3836f21SEric Dumazet skb_free_head(skb); 5451da177e4SLinus Torvalds } 5461da177e4SLinus Torvalds } 5471da177e4SLinus Torvalds 5481da177e4SLinus Torvalds /* 5491da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 5501da177e4SLinus Torvalds */ 5512d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 5521da177e4SLinus Torvalds { 553d179cd12SDavid S. Miller struct sk_buff *other; 554d179cd12SDavid S. Miller atomic_t *fclone_ref; 555d179cd12SDavid S. Miller 556d179cd12SDavid S. Miller switch (skb->fclone) { 557d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 5581da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 559d179cd12SDavid S. Miller break; 560d179cd12SDavid S. Miller 561d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 562d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 2); 563d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 564d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, skb); 565d179cd12SDavid S. Miller break; 566d179cd12SDavid S. Miller 567d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 568d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 1); 569d179cd12SDavid S. Miller other = skb - 1; 570d179cd12SDavid S. Miller 571d179cd12SDavid S. Miller /* The clone portion is available for 572d179cd12SDavid S. Miller * fast-cloning again. 573d179cd12SDavid S. Miller */ 574d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_UNAVAILABLE; 575d179cd12SDavid S. Miller 576d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 577d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, other); 578d179cd12SDavid S. Miller break; 5793ff50b79SStephen Hemminger } 5801da177e4SLinus Torvalds } 5811da177e4SLinus Torvalds 58204a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 5831da177e4SLinus Torvalds { 584adf30907SEric Dumazet skb_dst_drop(skb); 5851da177e4SLinus Torvalds #ifdef CONFIG_XFRM 5861da177e4SLinus Torvalds secpath_put(skb->sp); 5871da177e4SLinus Torvalds #endif 5881da177e4SLinus Torvalds if (skb->destructor) { 5899c2b3328SStephen Hemminger WARN_ON(in_irq()); 5901da177e4SLinus Torvalds skb->destructor(skb); 5911da177e4SLinus Torvalds } 592a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5935f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 5942fc72c7bSKOVACS Krisztian #endif 5952fc72c7bSKOVACS Krisztian #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 5969fb9cbb1SYasuyuki Kozakai nf_conntrack_put_reasm(skb->nfct_reasm); 5979fb9cbb1SYasuyuki Kozakai #endif 5981da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 5991da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 6001da177e4SLinus Torvalds #endif 6011da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 6021da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 6031da177e4SLinus Torvalds skb->tc_index = 0; 6041da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 6051da177e4SLinus Torvalds skb->tc_verd = 0; 6061da177e4SLinus Torvalds #endif 6071da177e4SLinus Torvalds #endif 60804a4bb55SLennert Buytenhek } 60904a4bb55SLennert Buytenhek 61004a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 61104a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 61204a4bb55SLennert Buytenhek { 61304a4bb55SLennert Buytenhek skb_release_head_state(skb); 6145e71d9d7SPablo Neira if (likely(skb->head)) 6152d4baff8SHerbert Xu skb_release_data(skb); 6162d4baff8SHerbert Xu } 6171da177e4SLinus Torvalds 6182d4baff8SHerbert Xu /** 6192d4baff8SHerbert Xu * __kfree_skb - private function 6202d4baff8SHerbert Xu * @skb: buffer 6212d4baff8SHerbert Xu * 6222d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 6232d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 6242d4baff8SHerbert Xu * always call kfree_skb 6252d4baff8SHerbert Xu */ 6262d4baff8SHerbert Xu 6272d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 6282d4baff8SHerbert Xu { 6292d4baff8SHerbert Xu skb_release_all(skb); 6301da177e4SLinus Torvalds kfree_skbmem(skb); 6311da177e4SLinus Torvalds } 632b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 6331da177e4SLinus Torvalds 6341da177e4SLinus Torvalds /** 635231d06aeSJörn Engel * kfree_skb - free an sk_buff 636231d06aeSJörn Engel * @skb: buffer to free 637231d06aeSJörn Engel * 638231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 639231d06aeSJörn Engel * hit zero. 640231d06aeSJörn Engel */ 641231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 642231d06aeSJörn Engel { 643231d06aeSJörn Engel if (unlikely(!skb)) 644231d06aeSJörn Engel return; 645231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 646231d06aeSJörn Engel smp_rmb(); 647231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 648231d06aeSJörn Engel return; 649ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 650231d06aeSJörn Engel __kfree_skb(skb); 651231d06aeSJörn Engel } 652b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 653231d06aeSJörn Engel 654bd8a7036SEric Dumazet void kfree_skb_list(struct sk_buff *segs) 655bd8a7036SEric Dumazet { 656bd8a7036SEric Dumazet while (segs) { 657bd8a7036SEric Dumazet struct sk_buff *next = segs->next; 658bd8a7036SEric Dumazet 659bd8a7036SEric Dumazet kfree_skb(segs); 660bd8a7036SEric Dumazet segs = next; 661bd8a7036SEric Dumazet } 662bd8a7036SEric Dumazet } 663bd8a7036SEric Dumazet EXPORT_SYMBOL(kfree_skb_list); 664bd8a7036SEric Dumazet 665d1a203eaSStephen Hemminger /** 66625121173SMichael S. Tsirkin * skb_tx_error - report an sk_buff xmit error 66725121173SMichael S. Tsirkin * @skb: buffer that triggered an error 66825121173SMichael S. Tsirkin * 66925121173SMichael S. Tsirkin * Report xmit error if a device callback is tracking this skb. 67025121173SMichael S. Tsirkin * skb must be freed afterwards. 67125121173SMichael S. Tsirkin */ 67225121173SMichael S. Tsirkin void skb_tx_error(struct sk_buff *skb) 67325121173SMichael S. Tsirkin { 67425121173SMichael S. Tsirkin if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 67525121173SMichael S. Tsirkin struct ubuf_info *uarg; 67625121173SMichael S. Tsirkin 67725121173SMichael S. Tsirkin uarg = skb_shinfo(skb)->destructor_arg; 67825121173SMichael S. Tsirkin if (uarg->callback) 67925121173SMichael S. Tsirkin uarg->callback(uarg, false); 68025121173SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 68125121173SMichael S. Tsirkin } 68225121173SMichael S. Tsirkin } 68325121173SMichael S. Tsirkin EXPORT_SYMBOL(skb_tx_error); 68425121173SMichael S. Tsirkin 68525121173SMichael S. Tsirkin /** 686ead2ceb0SNeil Horman * consume_skb - free an skbuff 687ead2ceb0SNeil Horman * @skb: buffer to free 688ead2ceb0SNeil Horman * 689ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 690ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 691ead2ceb0SNeil Horman * is being dropped after a failure and notes that 692ead2ceb0SNeil Horman */ 693ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 694ead2ceb0SNeil Horman { 695ead2ceb0SNeil Horman if (unlikely(!skb)) 696ead2ceb0SNeil Horman return; 697ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 698ead2ceb0SNeil Horman smp_rmb(); 699ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 700ead2ceb0SNeil Horman return; 70107dc22e7SKoki Sanagi trace_consume_skb(skb); 702ead2ceb0SNeil Horman __kfree_skb(skb); 703ead2ceb0SNeil Horman } 704ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 705ead2ceb0SNeil Horman 706dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 707dec18810SHerbert Xu { 708dec18810SHerbert Xu new->tstamp = old->tstamp; 709dec18810SHerbert Xu new->dev = old->dev; 710dec18810SHerbert Xu new->transport_header = old->transport_header; 711dec18810SHerbert Xu new->network_header = old->network_header; 712dec18810SHerbert Xu new->mac_header = old->mac_header; 7134bc41b84SJoe Stringer new->inner_protocol = old->inner_protocol; 7146a674e9cSJoseph Gasparakis new->inner_transport_header = old->inner_transport_header; 71592df9b21SPravin B Shelar new->inner_network_header = old->inner_network_header; 716aefbd2b3SPravin B Shelar new->inner_mac_header = old->inner_mac_header; 7177fee226aSEric Dumazet skb_dst_copy(new, old); 7180a9627f2STom Herbert new->rxhash = old->rxhash; 7196461be3aSChangli Gao new->ooo_okay = old->ooo_okay; 720bdeab991STom Herbert new->l4_rxhash = old->l4_rxhash; 7213bdc0ebaSBen Greear new->no_fcs = old->no_fcs; 7226a674e9cSJoseph Gasparakis new->encapsulation = old->encapsulation; 723def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 724dec18810SHerbert Xu new->sp = secpath_get(old->sp); 725dec18810SHerbert Xu #endif 726dec18810SHerbert Xu memcpy(new->cb, old->cb, sizeof(old->cb)); 7279bcb97caSHerbert Xu new->csum = old->csum; 728dec18810SHerbert Xu new->local_df = old->local_df; 729dec18810SHerbert Xu new->pkt_type = old->pkt_type; 730dec18810SHerbert Xu new->ip_summed = old->ip_summed; 731dec18810SHerbert Xu skb_copy_queue_mapping(new, old); 732dec18810SHerbert Xu new->priority = old->priority; 733a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_IP_VS) 734dec18810SHerbert Xu new->ipvs_property = old->ipvs_property; 735dec18810SHerbert Xu #endif 736c93bdd0eSMel Gorman new->pfmemalloc = old->pfmemalloc; 737dec18810SHerbert Xu new->protocol = old->protocol; 738dec18810SHerbert Xu new->mark = old->mark; 7398964be4aSEric Dumazet new->skb_iif = old->skb_iif; 740dec18810SHerbert Xu __nf_copy(new, old); 741a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) 742dec18810SHerbert Xu new->nf_trace = old->nf_trace; 743dec18810SHerbert Xu #endif 744dec18810SHerbert Xu #ifdef CONFIG_NET_SCHED 745dec18810SHerbert Xu new->tc_index = old->tc_index; 746dec18810SHerbert Xu #ifdef CONFIG_NET_CLS_ACT 747dec18810SHerbert Xu new->tc_verd = old->tc_verd; 748dec18810SHerbert Xu #endif 749dec18810SHerbert Xu #endif 75086a9bad3SPatrick McHardy new->vlan_proto = old->vlan_proto; 7516aa895b0SPatrick McHardy new->vlan_tci = old->vlan_tci; 7526aa895b0SPatrick McHardy 753dec18810SHerbert Xu skb_copy_secmark(new, old); 75406021292SEliezer Tamir 755e0d1095aSCong Wang #ifdef CONFIG_NET_RX_BUSY_POLL 75606021292SEliezer Tamir new->napi_id = old->napi_id; 75706021292SEliezer Tamir #endif 758dec18810SHerbert Xu } 759dec18810SHerbert Xu 76082c49a35SHerbert Xu /* 76182c49a35SHerbert Xu * You should not add any new code to this function. Add it to 76282c49a35SHerbert Xu * __copy_skb_header above instead. 76382c49a35SHerbert Xu */ 764e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 7651da177e4SLinus Torvalds { 7661da177e4SLinus Torvalds #define C(x) n->x = skb->x 7671da177e4SLinus Torvalds 7681da177e4SLinus Torvalds n->next = n->prev = NULL; 7691da177e4SLinus Torvalds n->sk = NULL; 770dec18810SHerbert Xu __copy_skb_header(n, skb); 771dec18810SHerbert Xu 7721da177e4SLinus Torvalds C(len); 7731da177e4SLinus Torvalds C(data_len); 7743e6b3b2eSAlexey Dobriyan C(mac_len); 775334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 77602f1c89dSPaul Moore n->cloned = 1; 7771da177e4SLinus Torvalds n->nohdr = 0; 7781da177e4SLinus Torvalds n->destructor = NULL; 7791da177e4SLinus Torvalds C(tail); 7801da177e4SLinus Torvalds C(end); 78102f1c89dSPaul Moore C(head); 782d3836f21SEric Dumazet C(head_frag); 78302f1c89dSPaul Moore C(data); 78402f1c89dSPaul Moore C(truesize); 78502f1c89dSPaul Moore atomic_set(&n->users, 1); 7861da177e4SLinus Torvalds 7871da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 7881da177e4SLinus Torvalds skb->cloned = 1; 7891da177e4SLinus Torvalds 7901da177e4SLinus Torvalds return n; 791e0053ec0SHerbert Xu #undef C 792e0053ec0SHerbert Xu } 793e0053ec0SHerbert Xu 794e0053ec0SHerbert Xu /** 795e0053ec0SHerbert Xu * skb_morph - morph one skb into another 796e0053ec0SHerbert Xu * @dst: the skb to receive the contents 797e0053ec0SHerbert Xu * @src: the skb to supply the contents 798e0053ec0SHerbert Xu * 799e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 800e0053ec0SHerbert Xu * supplied by the user. 801e0053ec0SHerbert Xu * 802e0053ec0SHerbert Xu * The target skb is returned upon exit. 803e0053ec0SHerbert Xu */ 804e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 805e0053ec0SHerbert Xu { 8062d4baff8SHerbert Xu skb_release_all(dst); 807e0053ec0SHerbert Xu return __skb_clone(dst, src); 808e0053ec0SHerbert Xu } 809e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 810e0053ec0SHerbert Xu 8112c53040fSBen Hutchings /** 8122c53040fSBen Hutchings * skb_copy_ubufs - copy userspace skb frags buffers to kernel 81348c83012SMichael S. Tsirkin * @skb: the skb to modify 81448c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 81548c83012SMichael S. Tsirkin * 81648c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 81748c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 81848c83012SMichael S. Tsirkin * to userspace pages. 81948c83012SMichael S. Tsirkin * 82048c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 82148c83012SMichael S. Tsirkin * %GFP_ATOMIC. 82248c83012SMichael S. Tsirkin * 82348c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 82448c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 82548c83012SMichael S. Tsirkin */ 82648c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 827a6686f2fSShirley Ma { 828a6686f2fSShirley Ma int i; 829a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 830a6686f2fSShirley Ma struct page *page, *head = NULL; 831a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 832a6686f2fSShirley Ma 833a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 834a6686f2fSShirley Ma u8 *vaddr; 835a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 836a6686f2fSShirley Ma 83702756ed4SKrishna Kumar page = alloc_page(gfp_mask); 838a6686f2fSShirley Ma if (!page) { 839a6686f2fSShirley Ma while (head) { 84040dadff2SSunghan Suh struct page *next = (struct page *)page_private(head); 841a6686f2fSShirley Ma put_page(head); 842a6686f2fSShirley Ma head = next; 843a6686f2fSShirley Ma } 844a6686f2fSShirley Ma return -ENOMEM; 845a6686f2fSShirley Ma } 84651c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 847a6686f2fSShirley Ma memcpy(page_address(page), 8489e903e08SEric Dumazet vaddr + f->page_offset, skb_frag_size(f)); 84951c56b00SEric Dumazet kunmap_atomic(vaddr); 85040dadff2SSunghan Suh set_page_private(page, (unsigned long)head); 851a6686f2fSShirley Ma head = page; 852a6686f2fSShirley Ma } 853a6686f2fSShirley Ma 854a6686f2fSShirley Ma /* skb frags release userspace buffers */ 85502756ed4SKrishna Kumar for (i = 0; i < num_frags; i++) 856a8605c60SIan Campbell skb_frag_unref(skb, i); 857a6686f2fSShirley Ma 858e19d6763SMichael S. Tsirkin uarg->callback(uarg, false); 859a6686f2fSShirley Ma 860a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 86102756ed4SKrishna Kumar for (i = num_frags - 1; i >= 0; i--) { 86202756ed4SKrishna Kumar __skb_fill_page_desc(skb, i, head, 0, 86302756ed4SKrishna Kumar skb_shinfo(skb)->frags[i].size); 86440dadff2SSunghan Suh head = (struct page *)page_private(head); 865a6686f2fSShirley Ma } 86648c83012SMichael S. Tsirkin 86748c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 868a6686f2fSShirley Ma return 0; 869a6686f2fSShirley Ma } 870dcc0fb78SMichael S. Tsirkin EXPORT_SYMBOL_GPL(skb_copy_ubufs); 871a6686f2fSShirley Ma 872e0053ec0SHerbert Xu /** 873e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 874e0053ec0SHerbert Xu * @skb: buffer to clone 875e0053ec0SHerbert Xu * @gfp_mask: allocation priority 876e0053ec0SHerbert Xu * 877e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 878e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 879e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 880e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 881e0053ec0SHerbert Xu * 882e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 883e0053ec0SHerbert Xu * %GFP_ATOMIC. 884e0053ec0SHerbert Xu */ 885e0053ec0SHerbert Xu 886e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 887e0053ec0SHerbert Xu { 888e0053ec0SHerbert Xu struct sk_buff *n; 889e0053ec0SHerbert Xu 89070008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 891a6686f2fSShirley Ma return NULL; 892a6686f2fSShirley Ma 893e0053ec0SHerbert Xu n = skb + 1; 894e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 895e0053ec0SHerbert Xu n->fclone == SKB_FCLONE_UNAVAILABLE) { 896e0053ec0SHerbert Xu atomic_t *fclone_ref = (atomic_t *) (n + 1); 897e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_CLONE; 898e0053ec0SHerbert Xu atomic_inc(fclone_ref); 899e0053ec0SHerbert Xu } else { 900c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 901c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 902c93bdd0eSMel Gorman 903e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 904e0053ec0SHerbert Xu if (!n) 905e0053ec0SHerbert Xu return NULL; 906fe55f6d5SVegard Nossum 907fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 908fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags2); 909e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 910e0053ec0SHerbert Xu } 911e0053ec0SHerbert Xu 912e0053ec0SHerbert Xu return __skb_clone(n, skb); 9131da177e4SLinus Torvalds } 914b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 9151da177e4SLinus Torvalds 916f5b17294SPravin B Shelar static void skb_headers_offset_update(struct sk_buff *skb, int off) 917f5b17294SPravin B Shelar { 918030737bcSEric Dumazet /* Only adjust this if it actually is csum_start rather than csum */ 919030737bcSEric Dumazet if (skb->ip_summed == CHECKSUM_PARTIAL) 920030737bcSEric Dumazet skb->csum_start += off; 921f5b17294SPravin B Shelar /* {transport,network,mac}_header and tail are relative to skb->head */ 922f5b17294SPravin B Shelar skb->transport_header += off; 923f5b17294SPravin B Shelar skb->network_header += off; 924f5b17294SPravin B Shelar if (skb_mac_header_was_set(skb)) 925f5b17294SPravin B Shelar skb->mac_header += off; 926f5b17294SPravin B Shelar skb->inner_transport_header += off; 927f5b17294SPravin B Shelar skb->inner_network_header += off; 928aefbd2b3SPravin B Shelar skb->inner_mac_header += off; 929f5b17294SPravin B Shelar } 930f5b17294SPravin B Shelar 9311da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 9321da177e4SLinus Torvalds { 933dec18810SHerbert Xu __copy_skb_header(new, old); 934dec18810SHerbert Xu 9357967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 9367967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 9377967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 9381da177e4SLinus Torvalds } 9391da177e4SLinus Torvalds 940c93bdd0eSMel Gorman static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 941c93bdd0eSMel Gorman { 942c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 943c93bdd0eSMel Gorman return SKB_ALLOC_RX; 944c93bdd0eSMel Gorman return 0; 945c93bdd0eSMel Gorman } 946c93bdd0eSMel Gorman 9471da177e4SLinus Torvalds /** 9481da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 9491da177e4SLinus Torvalds * @skb: buffer to copy 9501da177e4SLinus Torvalds * @gfp_mask: allocation priority 9511da177e4SLinus Torvalds * 9521da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 9531da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 9541da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 9551da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 9561da177e4SLinus Torvalds * 9571da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 9581da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 9591da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 9601da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 9611da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 9621da177e4SLinus Torvalds */ 9631da177e4SLinus Torvalds 964dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 9651da177e4SLinus Torvalds { 9666602cebbSEric Dumazet int headerlen = skb_headroom(skb); 967ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 968c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 969c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 9706602cebbSEric Dumazet 9711da177e4SLinus Torvalds if (!n) 9721da177e4SLinus Torvalds return NULL; 9731da177e4SLinus Torvalds 9741da177e4SLinus Torvalds /* Set the data pointer */ 9751da177e4SLinus Torvalds skb_reserve(n, headerlen); 9761da177e4SLinus Torvalds /* Set the tail pointer and length */ 9771da177e4SLinus Torvalds skb_put(n, skb->len); 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 9801da177e4SLinus Torvalds BUG(); 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds copy_skb_header(n, skb); 9831da177e4SLinus Torvalds return n; 9841da177e4SLinus Torvalds } 985b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds /** 988117632e6SEric Dumazet * __pskb_copy - create copy of an sk_buff with private head. 9891da177e4SLinus Torvalds * @skb: buffer to copy 990117632e6SEric Dumazet * @headroom: headroom of new skb 9911da177e4SLinus Torvalds * @gfp_mask: allocation priority 9921da177e4SLinus Torvalds * 9931da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 9941da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 9951da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 9961da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 9971da177e4SLinus Torvalds * or the pointer to the buffer on success. 9981da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 9991da177e4SLinus Torvalds */ 10001da177e4SLinus Torvalds 1001117632e6SEric Dumazet struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask) 10021da177e4SLinus Torvalds { 1003117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 1004c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(size, gfp_mask, 1005c93bdd0eSMel Gorman skb_alloc_rx_flag(skb), NUMA_NO_NODE); 10066602cebbSEric Dumazet 10071da177e4SLinus Torvalds if (!n) 10081da177e4SLinus Torvalds goto out; 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds /* Set the data pointer */ 1011117632e6SEric Dumazet skb_reserve(n, headroom); 10121da177e4SLinus Torvalds /* Set the tail pointer and length */ 10131da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 10141da177e4SLinus Torvalds /* Copy the bytes */ 1015d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 10161da177e4SLinus Torvalds 101725f484a6SHerbert Xu n->truesize += skb->data_len; 10181da177e4SLinus Torvalds n->data_len = skb->data_len; 10191da177e4SLinus Torvalds n->len = skb->len; 10201da177e4SLinus Torvalds 10211da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 10221da177e4SLinus Torvalds int i; 10231da177e4SLinus Torvalds 102470008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) { 10251511022cSDan Carpenter kfree_skb(n); 10261511022cSDan Carpenter n = NULL; 1027a6686f2fSShirley Ma goto out; 1028a6686f2fSShirley Ma } 10291da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 10301da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1031ea2ab693SIan Campbell skb_frag_ref(skb, i); 10321da177e4SLinus Torvalds } 10331da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 103621dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 10371da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 10381da177e4SLinus Torvalds skb_clone_fraglist(n); 10391da177e4SLinus Torvalds } 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds copy_skb_header(n, skb); 10421da177e4SLinus Torvalds out: 10431da177e4SLinus Torvalds return n; 10441da177e4SLinus Torvalds } 1045117632e6SEric Dumazet EXPORT_SYMBOL(__pskb_copy); 10461da177e4SLinus Torvalds 10471da177e4SLinus Torvalds /** 10481da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 10491da177e4SLinus Torvalds * @skb: buffer to reallocate 10501da177e4SLinus Torvalds * @nhead: room to add at head 10511da177e4SLinus Torvalds * @ntail: room to add at tail 10521da177e4SLinus Torvalds * @gfp_mask: allocation priority 10531da177e4SLinus Torvalds * 1054bc32383cSMathias Krause * Expands (or creates identical copy, if @nhead and @ntail are zero) 1055bc32383cSMathias Krause * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 10561da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 10571da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 10581da177e4SLinus Torvalds * 10591da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 10601da177e4SLinus Torvalds * reloaded after call to this function. 10611da177e4SLinus Torvalds */ 10621da177e4SLinus Torvalds 106386a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1064dd0fc66fSAl Viro gfp_t gfp_mask) 10651da177e4SLinus Torvalds { 10661da177e4SLinus Torvalds int i; 10671da177e4SLinus Torvalds u8 *data; 1068ec47ea82SAlexander Duyck int size = nhead + skb_end_offset(skb) + ntail; 10691da177e4SLinus Torvalds long off; 10701da177e4SLinus Torvalds 10714edd87adSHerbert Xu BUG_ON(nhead < 0); 10724edd87adSHerbert Xu 10731da177e4SLinus Torvalds if (skb_shared(skb)) 10741da177e4SLinus Torvalds BUG(); 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 10771da177e4SLinus Torvalds 1078c93bdd0eSMel Gorman if (skb_pfmemalloc(skb)) 1079c93bdd0eSMel Gorman gfp_mask |= __GFP_MEMALLOC; 1080c93bdd0eSMel Gorman data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1081c93bdd0eSMel Gorman gfp_mask, NUMA_NO_NODE, NULL); 10821da177e4SLinus Torvalds if (!data) 10831da177e4SLinus Torvalds goto nodata; 108487151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 10876602cebbSEric Dumazet * optimized for the cases when header is void. 10886602cebbSEric Dumazet */ 10896602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 10906602cebbSEric Dumazet 10916602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 10926602cebbSEric Dumazet skb_shinfo(skb), 1093fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 10941da177e4SLinus Torvalds 10953e24591aSAlexander Duyck /* 10963e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 10973e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 10983e24591aSAlexander Duyck * be since all we did is relocate the values 10993e24591aSAlexander Duyck */ 11003e24591aSAlexander Duyck if (skb_cloned(skb)) { 1101a6686f2fSShirley Ma /* copy this zero copy skb frags */ 110270008aa5SMichael S. Tsirkin if (skb_orphan_frags(skb, gfp_mask)) 1103a6686f2fSShirley Ma goto nofrags; 11041da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1105ea2ab693SIan Campbell skb_frag_ref(skb, i); 11061da177e4SLinus Torvalds 110721dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 11081da177e4SLinus Torvalds skb_clone_fraglist(skb); 11091da177e4SLinus Torvalds 11101da177e4SLinus Torvalds skb_release_data(skb); 11113e24591aSAlexander Duyck } else { 11123e24591aSAlexander Duyck skb_free_head(skb); 11131fd63041SEric Dumazet } 11141da177e4SLinus Torvalds off = (data + nhead) - skb->head; 11151da177e4SLinus Torvalds 11161da177e4SLinus Torvalds skb->head = data; 1117d3836f21SEric Dumazet skb->head_frag = 0; 11181da177e4SLinus Torvalds skb->data += off; 11194305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 11204305b541SArnaldo Carvalho de Melo skb->end = size; 112156eb8882SPatrick McHardy off = nhead; 11224305b541SArnaldo Carvalho de Melo #else 11234305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 112456eb8882SPatrick McHardy #endif 112527a884dcSArnaldo Carvalho de Melo skb->tail += off; 1126b41abb42SPeter Pan(潘卫平) skb_headers_offset_update(skb, nhead); 11271da177e4SLinus Torvalds skb->cloned = 0; 1128334a8132SPatrick McHardy skb->hdr_len = 0; 11291da177e4SLinus Torvalds skb->nohdr = 0; 11301da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 11311da177e4SLinus Torvalds return 0; 11321da177e4SLinus Torvalds 1133a6686f2fSShirley Ma nofrags: 1134a6686f2fSShirley Ma kfree(data); 11351da177e4SLinus Torvalds nodata: 11361da177e4SLinus Torvalds return -ENOMEM; 11371da177e4SLinus Torvalds } 1138b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 11391da177e4SLinus Torvalds 11401da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 11431da177e4SLinus Torvalds { 11441da177e4SLinus Torvalds struct sk_buff *skb2; 11451da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds if (delta <= 0) 11481da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 11491da177e4SLinus Torvalds else { 11501da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 11511da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 11521da177e4SLinus Torvalds GFP_ATOMIC)) { 11531da177e4SLinus Torvalds kfree_skb(skb2); 11541da177e4SLinus Torvalds skb2 = NULL; 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds } 11571da177e4SLinus Torvalds return skb2; 11581da177e4SLinus Torvalds } 1159b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds /** 11621da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 11631da177e4SLinus Torvalds * @skb: buffer to copy 11641da177e4SLinus Torvalds * @newheadroom: new free bytes at head 11651da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 11661da177e4SLinus Torvalds * @gfp_mask: allocation priority 11671da177e4SLinus Torvalds * 11681da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 11691da177e4SLinus Torvalds * allocate additional space. 11701da177e4SLinus Torvalds * 11711da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 11721da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 11731da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 11741da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 11751da177e4SLinus Torvalds * 11761da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 11771da177e4SLinus Torvalds * is called from an interrupt. 11781da177e4SLinus Torvalds */ 11791da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 118086a76cafSVictor Fusco int newheadroom, int newtailroom, 1181dd0fc66fSAl Viro gfp_t gfp_mask) 11821da177e4SLinus Torvalds { 11831da177e4SLinus Torvalds /* 11841da177e4SLinus Torvalds * Allocate the copy buffer 11851da177e4SLinus Torvalds */ 1186c93bdd0eSMel Gorman struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1187c93bdd0eSMel Gorman gfp_mask, skb_alloc_rx_flag(skb), 1188c93bdd0eSMel Gorman NUMA_NO_NODE); 1189efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 11901da177e4SLinus Torvalds int head_copy_len, head_copy_off; 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds if (!n) 11931da177e4SLinus Torvalds return NULL; 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds skb_reserve(n, newheadroom); 11961da177e4SLinus Torvalds 11971da177e4SLinus Torvalds /* Set the tail pointer and length */ 11981da177e4SLinus Torvalds skb_put(n, skb->len); 11991da177e4SLinus Torvalds 1200efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 12011da177e4SLinus Torvalds head_copy_off = 0; 12021da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 12031da177e4SLinus Torvalds head_copy_len = newheadroom; 12041da177e4SLinus Torvalds else 12051da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds /* Copy the linear header and data. */ 12081da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 12091da177e4SLinus Torvalds skb->len + head_copy_len)) 12101da177e4SLinus Torvalds BUG(); 12111da177e4SLinus Torvalds 12121da177e4SLinus Torvalds copy_skb_header(n, skb); 12131da177e4SLinus Torvalds 1214030737bcSEric Dumazet skb_headers_offset_update(n, newheadroom - oldheadroom); 1215efd1e8d5SPatrick McHardy 12161da177e4SLinus Torvalds return n; 12171da177e4SLinus Torvalds } 1218b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds /** 12211da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 12221da177e4SLinus Torvalds * @skb: buffer to pad 12231da177e4SLinus Torvalds * @pad: space to pad 12241da177e4SLinus Torvalds * 12251da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 12261da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 12271da177e4SLinus Torvalds * beyond the buffer end onto the wire. 12281da177e4SLinus Torvalds * 12295b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 12301da177e4SLinus Torvalds */ 12311da177e4SLinus Torvalds 12325b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 12331da177e4SLinus Torvalds { 12345b057c6bSHerbert Xu int err; 12355b057c6bSHerbert Xu int ntail; 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 12385b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 12391da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 12405b057c6bSHerbert Xu return 0; 12411da177e4SLinus Torvalds } 12421da177e4SLinus Torvalds 12434305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 12445b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 12455b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 12465b057c6bSHerbert Xu if (unlikely(err)) 12475b057c6bSHerbert Xu goto free_skb; 12485b057c6bSHerbert Xu } 12495b057c6bSHerbert Xu 12505b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 12515b057c6bSHerbert Xu * to be audited. 12525b057c6bSHerbert Xu */ 12535b057c6bSHerbert Xu err = skb_linearize(skb); 12545b057c6bSHerbert Xu if (unlikely(err)) 12555b057c6bSHerbert Xu goto free_skb; 12565b057c6bSHerbert Xu 12575b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 12585b057c6bSHerbert Xu return 0; 12595b057c6bSHerbert Xu 12605b057c6bSHerbert Xu free_skb: 12611da177e4SLinus Torvalds kfree_skb(skb); 12625b057c6bSHerbert Xu return err; 12631da177e4SLinus Torvalds } 1264b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 12651da177e4SLinus Torvalds 12660dde3e16SIlpo Järvinen /** 12670c7ddf36SMathias Krause * pskb_put - add data to the tail of a potentially fragmented buffer 12680c7ddf36SMathias Krause * @skb: start of the buffer to use 12690c7ddf36SMathias Krause * @tail: tail fragment of the buffer to use 12700c7ddf36SMathias Krause * @len: amount of data to add 12710c7ddf36SMathias Krause * 12720c7ddf36SMathias Krause * This function extends the used data area of the potentially 12730c7ddf36SMathias Krause * fragmented buffer. @tail must be the last fragment of @skb -- or 12740c7ddf36SMathias Krause * @skb itself. If this would exceed the total buffer size the kernel 12750c7ddf36SMathias Krause * will panic. A pointer to the first byte of the extra data is 12760c7ddf36SMathias Krause * returned. 12770c7ddf36SMathias Krause */ 12780c7ddf36SMathias Krause 12790c7ddf36SMathias Krause unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 12800c7ddf36SMathias Krause { 12810c7ddf36SMathias Krause if (tail != skb) { 12820c7ddf36SMathias Krause skb->data_len += len; 12830c7ddf36SMathias Krause skb->len += len; 12840c7ddf36SMathias Krause } 12850c7ddf36SMathias Krause return skb_put(tail, len); 12860c7ddf36SMathias Krause } 12870c7ddf36SMathias Krause EXPORT_SYMBOL_GPL(pskb_put); 12880c7ddf36SMathias Krause 12890c7ddf36SMathias Krause /** 12900dde3e16SIlpo Järvinen * skb_put - add data to a buffer 12910dde3e16SIlpo Järvinen * @skb: buffer to use 12920dde3e16SIlpo Järvinen * @len: amount of data to add 12930dde3e16SIlpo Järvinen * 12940dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 12950dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 12960dde3e16SIlpo Järvinen * first byte of the extra data is returned. 12970dde3e16SIlpo Järvinen */ 12980dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 12990dde3e16SIlpo Järvinen { 13000dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 13010dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 13020dde3e16SIlpo Järvinen skb->tail += len; 13030dde3e16SIlpo Järvinen skb->len += len; 13040dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 13050dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 13060dde3e16SIlpo Järvinen return tmp; 13070dde3e16SIlpo Järvinen } 13080dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 13090dde3e16SIlpo Järvinen 13106be8ac2fSIlpo Järvinen /** 1311c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1312c2aa270aSIlpo Järvinen * @skb: buffer to use 1313c2aa270aSIlpo Järvinen * @len: amount of data to add 1314c2aa270aSIlpo Järvinen * 1315c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1316c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1317c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1318c2aa270aSIlpo Järvinen */ 1319c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1320c2aa270aSIlpo Järvinen { 1321c2aa270aSIlpo Järvinen skb->data -= len; 1322c2aa270aSIlpo Järvinen skb->len += len; 1323c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1324c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1325c2aa270aSIlpo Järvinen return skb->data; 1326c2aa270aSIlpo Järvinen } 1327c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1328c2aa270aSIlpo Järvinen 1329c2aa270aSIlpo Järvinen /** 13306be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 13316be8ac2fSIlpo Järvinen * @skb: buffer to use 13326be8ac2fSIlpo Järvinen * @len: amount of data to remove 13336be8ac2fSIlpo Järvinen * 13346be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 13356be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 13366be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 13376be8ac2fSIlpo Järvinen * the old data. 13386be8ac2fSIlpo Järvinen */ 13396be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 13406be8ac2fSIlpo Järvinen { 134147d29646SDavid S. Miller return skb_pull_inline(skb, len); 13426be8ac2fSIlpo Järvinen } 13436be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 13446be8ac2fSIlpo Järvinen 1345419ae74eSIlpo Järvinen /** 1346419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1347419ae74eSIlpo Järvinen * @skb: buffer to alter 1348419ae74eSIlpo Järvinen * @len: new length 1349419ae74eSIlpo Järvinen * 1350419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1351419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1352419ae74eSIlpo Järvinen * The skb must be linear. 1353419ae74eSIlpo Järvinen */ 1354419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1355419ae74eSIlpo Järvinen { 1356419ae74eSIlpo Järvinen if (skb->len > len) 1357419ae74eSIlpo Järvinen __skb_trim(skb, len); 1358419ae74eSIlpo Järvinen } 1359419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1360419ae74eSIlpo Järvinen 13613cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 13621da177e4SLinus Torvalds */ 13631da177e4SLinus Torvalds 13643cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 13651da177e4SLinus Torvalds { 136627b437c8SHerbert Xu struct sk_buff **fragp; 136727b437c8SHerbert Xu struct sk_buff *frag; 13681da177e4SLinus Torvalds int offset = skb_headlen(skb); 13691da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 13701da177e4SLinus Torvalds int i; 137127b437c8SHerbert Xu int err; 137227b437c8SHerbert Xu 137327b437c8SHerbert Xu if (skb_cloned(skb) && 137427b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 137527b437c8SHerbert Xu return err; 13761da177e4SLinus Torvalds 1377f4d26fb3SHerbert Xu i = 0; 1378f4d26fb3SHerbert Xu if (offset >= len) 1379f4d26fb3SHerbert Xu goto drop_pages; 1380f4d26fb3SHerbert Xu 1381f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 13829e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 138327b437c8SHerbert Xu 138427b437c8SHerbert Xu if (end < len) { 13851da177e4SLinus Torvalds offset = end; 138627b437c8SHerbert Xu continue; 13871da177e4SLinus Torvalds } 13881da177e4SLinus Torvalds 13899e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 139027b437c8SHerbert Xu 1391f4d26fb3SHerbert Xu drop_pages: 139227b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 139327b437c8SHerbert Xu 139427b437c8SHerbert Xu for (; i < nfrags; i++) 1395ea2ab693SIan Campbell skb_frag_unref(skb, i); 139627b437c8SHerbert Xu 139721dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 139827b437c8SHerbert Xu skb_drop_fraglist(skb); 1399f4d26fb3SHerbert Xu goto done; 140027b437c8SHerbert Xu } 140127b437c8SHerbert Xu 140227b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 140327b437c8SHerbert Xu fragp = &frag->next) { 140427b437c8SHerbert Xu int end = offset + frag->len; 140527b437c8SHerbert Xu 140627b437c8SHerbert Xu if (skb_shared(frag)) { 140727b437c8SHerbert Xu struct sk_buff *nfrag; 140827b437c8SHerbert Xu 140927b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 141027b437c8SHerbert Xu if (unlikely(!nfrag)) 141127b437c8SHerbert Xu return -ENOMEM; 141227b437c8SHerbert Xu 141327b437c8SHerbert Xu nfrag->next = frag->next; 141485bb2a60SEric Dumazet consume_skb(frag); 141527b437c8SHerbert Xu frag = nfrag; 141627b437c8SHerbert Xu *fragp = frag; 141727b437c8SHerbert Xu } 141827b437c8SHerbert Xu 141927b437c8SHerbert Xu if (end < len) { 142027b437c8SHerbert Xu offset = end; 142127b437c8SHerbert Xu continue; 142227b437c8SHerbert Xu } 142327b437c8SHerbert Xu 142427b437c8SHerbert Xu if (end > len && 142527b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 142627b437c8SHerbert Xu return err; 142727b437c8SHerbert Xu 142827b437c8SHerbert Xu if (frag->next) 142927b437c8SHerbert Xu skb_drop_list(&frag->next); 143027b437c8SHerbert Xu break; 143127b437c8SHerbert Xu } 143227b437c8SHerbert Xu 1433f4d26fb3SHerbert Xu done: 143427b437c8SHerbert Xu if (len > skb_headlen(skb)) { 14351da177e4SLinus Torvalds skb->data_len -= skb->len - len; 14361da177e4SLinus Torvalds skb->len = len; 14371da177e4SLinus Torvalds } else { 14381da177e4SLinus Torvalds skb->len = len; 14391da177e4SLinus Torvalds skb->data_len = 0; 144027a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 14411da177e4SLinus Torvalds } 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds return 0; 14441da177e4SLinus Torvalds } 1445b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 14461da177e4SLinus Torvalds 14471da177e4SLinus Torvalds /** 14481da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 14491da177e4SLinus Torvalds * @skb: buffer to reallocate 14501da177e4SLinus Torvalds * @delta: number of bytes to advance tail 14511da177e4SLinus Torvalds * 14521da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 14531da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 14541da177e4SLinus Torvalds * data from fragmented part. 14551da177e4SLinus Torvalds * 14561da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 14571da177e4SLinus Torvalds * 14581da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 14591da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 14601da177e4SLinus Torvalds * 14611da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 14621da177e4SLinus Torvalds * reloaded after call to this function. 14631da177e4SLinus Torvalds */ 14641da177e4SLinus Torvalds 14651da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 14661da177e4SLinus Torvalds * when it is necessary. 14671da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 14681da177e4SLinus Torvalds * 2. It may change skb pointers. 14691da177e4SLinus Torvalds * 14701da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 14711da177e4SLinus Torvalds */ 14721da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 14731da177e4SLinus Torvalds { 14741da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 14751da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 14761da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 14771da177e4SLinus Torvalds */ 14784305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 14791da177e4SLinus Torvalds 14801da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 14811da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 14821da177e4SLinus Torvalds GFP_ATOMIC)) 14831da177e4SLinus Torvalds return NULL; 14841da177e4SLinus Torvalds } 14851da177e4SLinus Torvalds 148627a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 14871da177e4SLinus Torvalds BUG(); 14881da177e4SLinus Torvalds 14891da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 14901da177e4SLinus Torvalds * size of pulled pages. Superb. 14911da177e4SLinus Torvalds */ 149221dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 14931da177e4SLinus Torvalds goto pull_pages; 14941da177e4SLinus Torvalds 14951da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 14961da177e4SLinus Torvalds eat = delta; 14971da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 14989e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 14999e903e08SEric Dumazet 15009e903e08SEric Dumazet if (size >= eat) 15011da177e4SLinus Torvalds goto pull_pages; 15029e903e08SEric Dumazet eat -= size; 15031da177e4SLinus Torvalds } 15041da177e4SLinus Torvalds 15051da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 15061da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 15071da177e4SLinus Torvalds * but taking into account that pulling is expected to 15081da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 15091da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 15101da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 15111da177e4SLinus Torvalds */ 15121da177e4SLinus Torvalds if (eat) { 15131da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 15141da177e4SLinus Torvalds struct sk_buff *clone = NULL; 15151da177e4SLinus Torvalds struct sk_buff *insp = NULL; 15161da177e4SLinus Torvalds 15171da177e4SLinus Torvalds do { 151809a62660SKris Katterjohn BUG_ON(!list); 15191da177e4SLinus Torvalds 15201da177e4SLinus Torvalds if (list->len <= eat) { 15211da177e4SLinus Torvalds /* Eaten as whole. */ 15221da177e4SLinus Torvalds eat -= list->len; 15231da177e4SLinus Torvalds list = list->next; 15241da177e4SLinus Torvalds insp = list; 15251da177e4SLinus Torvalds } else { 15261da177e4SLinus Torvalds /* Eaten partially. */ 15271da177e4SLinus Torvalds 15281da177e4SLinus Torvalds if (skb_shared(list)) { 15291da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 15301da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 15311da177e4SLinus Torvalds if (!clone) 15321da177e4SLinus Torvalds return NULL; 15331da177e4SLinus Torvalds insp = list->next; 15341da177e4SLinus Torvalds list = clone; 15351da177e4SLinus Torvalds } else { 15361da177e4SLinus Torvalds /* This may be pulled without 15371da177e4SLinus Torvalds * problems. */ 15381da177e4SLinus Torvalds insp = list; 15391da177e4SLinus Torvalds } 15401da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 15411da177e4SLinus Torvalds kfree_skb(clone); 15421da177e4SLinus Torvalds return NULL; 15431da177e4SLinus Torvalds } 15441da177e4SLinus Torvalds break; 15451da177e4SLinus Torvalds } 15461da177e4SLinus Torvalds } while (eat); 15471da177e4SLinus Torvalds 15481da177e4SLinus Torvalds /* Free pulled out fragments. */ 15491da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 15501da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 15511da177e4SLinus Torvalds kfree_skb(list); 15521da177e4SLinus Torvalds } 15531da177e4SLinus Torvalds /* And insert new clone at head. */ 15541da177e4SLinus Torvalds if (clone) { 15551da177e4SLinus Torvalds clone->next = list; 15561da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 15571da177e4SLinus Torvalds } 15581da177e4SLinus Torvalds } 15591da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 15601da177e4SLinus Torvalds 15611da177e4SLinus Torvalds pull_pages: 15621da177e4SLinus Torvalds eat = delta; 15631da177e4SLinus Torvalds k = 0; 15641da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 15659e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 15669e903e08SEric Dumazet 15679e903e08SEric Dumazet if (size <= eat) { 1568ea2ab693SIan Campbell skb_frag_unref(skb, i); 15699e903e08SEric Dumazet eat -= size; 15701da177e4SLinus Torvalds } else { 15711da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 15721da177e4SLinus Torvalds if (eat) { 15731da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 15749e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 15751da177e4SLinus Torvalds eat = 0; 15761da177e4SLinus Torvalds } 15771da177e4SLinus Torvalds k++; 15781da177e4SLinus Torvalds } 15791da177e4SLinus Torvalds } 15801da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 15811da177e4SLinus Torvalds 15821da177e4SLinus Torvalds skb->tail += delta; 15831da177e4SLinus Torvalds skb->data_len -= delta; 15841da177e4SLinus Torvalds 158527a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 15861da177e4SLinus Torvalds } 1587b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 15881da177e4SLinus Torvalds 158922019b17SEric Dumazet /** 159022019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 159122019b17SEric Dumazet * @skb: source skb 159222019b17SEric Dumazet * @offset: offset in source 159322019b17SEric Dumazet * @to: destination buffer 159422019b17SEric Dumazet * @len: number of bytes to copy 159522019b17SEric Dumazet * 159622019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 159722019b17SEric Dumazet * destination buffer. 159822019b17SEric Dumazet * 159922019b17SEric Dumazet * CAUTION ! : 160022019b17SEric Dumazet * If its prototype is ever changed, 160122019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 160222019b17SEric Dumazet * since it is called from BPF assembly code. 160322019b17SEric Dumazet */ 16041da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 16051da177e4SLinus Torvalds { 16061a028e50SDavid S. Miller int start = skb_headlen(skb); 1607fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1608fbb398a8SDavid S. Miller int i, copy; 16091da177e4SLinus Torvalds 16101da177e4SLinus Torvalds if (offset > (int)skb->len - len) 16111da177e4SLinus Torvalds goto fault; 16121da177e4SLinus Torvalds 16131da177e4SLinus Torvalds /* Copy header. */ 16141a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 16151da177e4SLinus Torvalds if (copy > len) 16161da177e4SLinus Torvalds copy = len; 1617d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 16181da177e4SLinus Torvalds if ((len -= copy) == 0) 16191da177e4SLinus Torvalds return 0; 16201da177e4SLinus Torvalds offset += copy; 16211da177e4SLinus Torvalds to += copy; 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16251a028e50SDavid S. Miller int end; 162651c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 16271da177e4SLinus Torvalds 1628547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16291a028e50SDavid S. Miller 163051c56b00SEric Dumazet end = start + skb_frag_size(f); 16311da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 16321da177e4SLinus Torvalds u8 *vaddr; 16331da177e4SLinus Torvalds 16341da177e4SLinus Torvalds if (copy > len) 16351da177e4SLinus Torvalds copy = len; 16361da177e4SLinus Torvalds 163751c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 16381da177e4SLinus Torvalds memcpy(to, 163951c56b00SEric Dumazet vaddr + f->page_offset + offset - start, 164051c56b00SEric Dumazet copy); 164151c56b00SEric Dumazet kunmap_atomic(vaddr); 16421da177e4SLinus Torvalds 16431da177e4SLinus Torvalds if ((len -= copy) == 0) 16441da177e4SLinus Torvalds return 0; 16451da177e4SLinus Torvalds offset += copy; 16461da177e4SLinus Torvalds to += copy; 16471da177e4SLinus Torvalds } 16481a028e50SDavid S. Miller start = end; 16491da177e4SLinus Torvalds } 16501da177e4SLinus Torvalds 1651fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 16521a028e50SDavid S. Miller int end; 16531da177e4SLinus Torvalds 1654547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16551a028e50SDavid S. Miller 1656fbb398a8SDavid S. Miller end = start + frag_iter->len; 16571da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 16581da177e4SLinus Torvalds if (copy > len) 16591da177e4SLinus Torvalds copy = len; 1660fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 16611da177e4SLinus Torvalds goto fault; 16621da177e4SLinus Torvalds if ((len -= copy) == 0) 16631da177e4SLinus Torvalds return 0; 16641da177e4SLinus Torvalds offset += copy; 16651da177e4SLinus Torvalds to += copy; 16661da177e4SLinus Torvalds } 16671a028e50SDavid S. Miller start = end; 16681da177e4SLinus Torvalds } 1669a6686f2fSShirley Ma 16701da177e4SLinus Torvalds if (!len) 16711da177e4SLinus Torvalds return 0; 16721da177e4SLinus Torvalds 16731da177e4SLinus Torvalds fault: 16741da177e4SLinus Torvalds return -EFAULT; 16751da177e4SLinus Torvalds } 1676b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 16771da177e4SLinus Torvalds 16789c55e01cSJens Axboe /* 16799c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 16809c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 16819c55e01cSJens Axboe */ 16829c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 16839c55e01cSJens Axboe { 16848b9d3728SJarek Poplawski put_page(spd->pages[i]); 16858b9d3728SJarek Poplawski } 16869c55e01cSJens Axboe 1687a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 16884fb66994SJarek Poplawski unsigned int *offset, 168918aafc62SEric Dumazet struct sock *sk) 16908b9d3728SJarek Poplawski { 16915640f768SEric Dumazet struct page_frag *pfrag = sk_page_frag(sk); 16928b9d3728SJarek Poplawski 16935640f768SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 16948b9d3728SJarek Poplawski return NULL; 16954fb66994SJarek Poplawski 16965640f768SEric Dumazet *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 16974fb66994SJarek Poplawski 16985640f768SEric Dumazet memcpy(page_address(pfrag->page) + pfrag->offset, 16995640f768SEric Dumazet page_address(page) + *offset, *len); 17005640f768SEric Dumazet *offset = pfrag->offset; 17015640f768SEric Dumazet pfrag->offset += *len; 17024fb66994SJarek Poplawski 17035640f768SEric Dumazet return pfrag->page; 17049c55e01cSJens Axboe } 17059c55e01cSJens Axboe 170641c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 170741c73a0dSEric Dumazet struct page *page, 170841c73a0dSEric Dumazet unsigned int offset) 170941c73a0dSEric Dumazet { 171041c73a0dSEric Dumazet return spd->nr_pages && 171141c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 171241c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 171341c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 171441c73a0dSEric Dumazet } 171541c73a0dSEric Dumazet 17169c55e01cSJens Axboe /* 17179c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 17189c55e01cSJens Axboe */ 1719a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 172035f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 17214fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 172218aafc62SEric Dumazet bool linear, 17237a67e56fSJarek Poplawski struct sock *sk) 17249c55e01cSJens Axboe { 172541c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1726a108d5f3SDavid S. Miller return true; 17279c55e01cSJens Axboe 17288b9d3728SJarek Poplawski if (linear) { 172918aafc62SEric Dumazet page = linear_to_page(page, len, &offset, sk); 17308b9d3728SJarek Poplawski if (!page) 1731a108d5f3SDavid S. Miller return true; 173241c73a0dSEric Dumazet } 173341c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 173441c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 1735a108d5f3SDavid S. Miller return false; 173641c73a0dSEric Dumazet } 17378b9d3728SJarek Poplawski get_page(page); 17389c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 17394fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 17409c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 17419c55e01cSJens Axboe spd->nr_pages++; 17428b9d3728SJarek Poplawski 1743a108d5f3SDavid S. Miller return false; 17449c55e01cSJens Axboe } 17459c55e01cSJens Axboe 1746a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 17472870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 174818aafc62SEric Dumazet unsigned int *len, 1749d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 175035f3d14dSJens Axboe struct sock *sk, 175135f3d14dSJens Axboe struct pipe_inode_info *pipe) 17529c55e01cSJens Axboe { 17532870c43dSOctavian Purdila if (!*len) 1754a108d5f3SDavid S. Miller return true; 17559c55e01cSJens Axboe 17562870c43dSOctavian Purdila /* skip this segment if already processed */ 17572870c43dSOctavian Purdila if (*off >= plen) { 17582870c43dSOctavian Purdila *off -= plen; 1759a108d5f3SDavid S. Miller return false; 17602870c43dSOctavian Purdila } 17612870c43dSOctavian Purdila 17622870c43dSOctavian Purdila /* ignore any bits we already processed */ 17639ca1b22dSEric Dumazet poff += *off; 17649ca1b22dSEric Dumazet plen -= *off; 17652870c43dSOctavian Purdila *off = 0; 17662870c43dSOctavian Purdila 176718aafc62SEric Dumazet do { 176818aafc62SEric Dumazet unsigned int flen = min(*len, plen); 17692870c43dSOctavian Purdila 177018aafc62SEric Dumazet if (spd_fill_page(spd, pipe, page, &flen, poff, 177118aafc62SEric Dumazet linear, sk)) 1772a108d5f3SDavid S. Miller return true; 177318aafc62SEric Dumazet poff += flen; 177418aafc62SEric Dumazet plen -= flen; 17752870c43dSOctavian Purdila *len -= flen; 177618aafc62SEric Dumazet } while (*len && plen); 17772870c43dSOctavian Purdila 1778a108d5f3SDavid S. Miller return false; 1779db43a282SOctavian Purdila } 17809c55e01cSJens Axboe 17819c55e01cSJens Axboe /* 1782a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 17832870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 17849c55e01cSJens Axboe */ 1785a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 178635f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 178735f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 17882870c43dSOctavian Purdila { 17892870c43dSOctavian Purdila int seg; 17909c55e01cSJens Axboe 17911d0c0b32SEric Dumazet /* map the linear part : 17922996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 17932996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 17942996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 17959c55e01cSJens Axboe */ 17962870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 17972870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 17982870c43dSOctavian Purdila skb_headlen(skb), 179918aafc62SEric Dumazet offset, len, spd, 18003a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 18011d0c0b32SEric Dumazet sk, pipe)) 1802a108d5f3SDavid S. Miller return true; 18039c55e01cSJens Axboe 18049c55e01cSJens Axboe /* 18059c55e01cSJens Axboe * then map the fragments 18069c55e01cSJens Axboe */ 18079c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 18089c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 18099c55e01cSJens Axboe 1810ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 18119e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 181218aafc62SEric Dumazet offset, len, spd, false, sk, pipe)) 1813a108d5f3SDavid S. Miller return true; 18149c55e01cSJens Axboe } 18159c55e01cSJens Axboe 1816a108d5f3SDavid S. Miller return false; 18179c55e01cSJens Axboe } 18189c55e01cSJens Axboe 18199c55e01cSJens Axboe /* 18209c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 18219c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 18229c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 18239c55e01cSJens Axboe * handle that cleanly. 18249c55e01cSJens Axboe */ 18258b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 18269c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 18279c55e01cSJens Axboe unsigned int flags) 18289c55e01cSJens Axboe { 182941c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 183041c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 18319c55e01cSJens Axboe struct splice_pipe_desc spd = { 18329c55e01cSJens Axboe .pages = pages, 18339c55e01cSJens Axboe .partial = partial, 1834047fe360SEric Dumazet .nr_pages_max = MAX_SKB_FRAGS, 18359c55e01cSJens Axboe .flags = flags, 18369c55e01cSJens Axboe .ops = &sock_pipe_buf_ops, 18379c55e01cSJens Axboe .spd_release = sock_spd_release, 18389c55e01cSJens Axboe }; 1839fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 18407a67e56fSJarek Poplawski struct sock *sk = skb->sk; 184135f3d14dSJens Axboe int ret = 0; 184235f3d14dSJens Axboe 18439c55e01cSJens Axboe /* 18449c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 18459c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 18469c55e01cSJens Axboe */ 184735f3d14dSJens Axboe if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 18489c55e01cSJens Axboe goto done; 18499c55e01cSJens Axboe else if (!tlen) 18509c55e01cSJens Axboe goto done; 18519c55e01cSJens Axboe 18529c55e01cSJens Axboe /* 18539c55e01cSJens Axboe * now see if we have a frag_list to map 18549c55e01cSJens Axboe */ 1855fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 1856fbb398a8SDavid S. Miller if (!tlen) 18579c55e01cSJens Axboe break; 185835f3d14dSJens Axboe if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1859fbb398a8SDavid S. Miller break; 18609c55e01cSJens Axboe } 18619c55e01cSJens Axboe 18629c55e01cSJens Axboe done: 18639c55e01cSJens Axboe if (spd.nr_pages) { 18649c55e01cSJens Axboe /* 18659c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 18669c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 18679c55e01cSJens Axboe * here as compared to sendfile(). We enter here 18689c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 18699c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 18709c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 18719c55e01cSJens Axboe * and networking will grab the socket lock. 18729c55e01cSJens Axboe */ 1873293ad604SOctavian Purdila release_sock(sk); 18749c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1875293ad604SOctavian Purdila lock_sock(sk); 18769c55e01cSJens Axboe } 18779c55e01cSJens Axboe 187835f3d14dSJens Axboe return ret; 18799c55e01cSJens Axboe } 18809c55e01cSJens Axboe 1881357b40a1SHerbert Xu /** 1882357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1883357b40a1SHerbert Xu * @skb: destination buffer 1884357b40a1SHerbert Xu * @offset: offset in destination 1885357b40a1SHerbert Xu * @from: source buffer 1886357b40a1SHerbert Xu * @len: number of bytes to copy 1887357b40a1SHerbert Xu * 1888357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1889357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1890357b40a1SHerbert Xu * traversing fragment lists and such. 1891357b40a1SHerbert Xu */ 1892357b40a1SHerbert Xu 18930c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1894357b40a1SHerbert Xu { 18951a028e50SDavid S. Miller int start = skb_headlen(skb); 1896fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1897fbb398a8SDavid S. Miller int i, copy; 1898357b40a1SHerbert Xu 1899357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1900357b40a1SHerbert Xu goto fault; 1901357b40a1SHerbert Xu 19021a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1903357b40a1SHerbert Xu if (copy > len) 1904357b40a1SHerbert Xu copy = len; 190527d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1906357b40a1SHerbert Xu if ((len -= copy) == 0) 1907357b40a1SHerbert Xu return 0; 1908357b40a1SHerbert Xu offset += copy; 1909357b40a1SHerbert Xu from += copy; 1910357b40a1SHerbert Xu } 1911357b40a1SHerbert Xu 1912357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1913357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 19141a028e50SDavid S. Miller int end; 1915357b40a1SHerbert Xu 1916547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19171a028e50SDavid S. Miller 19189e903e08SEric Dumazet end = start + skb_frag_size(frag); 1919357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1920357b40a1SHerbert Xu u8 *vaddr; 1921357b40a1SHerbert Xu 1922357b40a1SHerbert Xu if (copy > len) 1923357b40a1SHerbert Xu copy = len; 1924357b40a1SHerbert Xu 192551c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 19261a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 19271a028e50SDavid S. Miller from, copy); 192851c56b00SEric Dumazet kunmap_atomic(vaddr); 1929357b40a1SHerbert Xu 1930357b40a1SHerbert Xu if ((len -= copy) == 0) 1931357b40a1SHerbert Xu return 0; 1932357b40a1SHerbert Xu offset += copy; 1933357b40a1SHerbert Xu from += copy; 1934357b40a1SHerbert Xu } 19351a028e50SDavid S. Miller start = end; 1936357b40a1SHerbert Xu } 1937357b40a1SHerbert Xu 1938fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 19391a028e50SDavid S. Miller int end; 1940357b40a1SHerbert Xu 1941547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19421a028e50SDavid S. Miller 1943fbb398a8SDavid S. Miller end = start + frag_iter->len; 1944357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1945357b40a1SHerbert Xu if (copy > len) 1946357b40a1SHerbert Xu copy = len; 1947fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 19481a028e50SDavid S. Miller from, copy)) 1949357b40a1SHerbert Xu goto fault; 1950357b40a1SHerbert Xu if ((len -= copy) == 0) 1951357b40a1SHerbert Xu return 0; 1952357b40a1SHerbert Xu offset += copy; 1953357b40a1SHerbert Xu from += copy; 1954357b40a1SHerbert Xu } 19551a028e50SDavid S. Miller start = end; 1956357b40a1SHerbert Xu } 1957357b40a1SHerbert Xu if (!len) 1958357b40a1SHerbert Xu return 0; 1959357b40a1SHerbert Xu 1960357b40a1SHerbert Xu fault: 1961357b40a1SHerbert Xu return -EFAULT; 1962357b40a1SHerbert Xu } 1963357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1964357b40a1SHerbert Xu 19651da177e4SLinus Torvalds /* Checksum skb data. */ 19662817a336SDaniel Borkmann __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 19672817a336SDaniel Borkmann __wsum csum, const struct skb_checksum_ops *ops) 19681da177e4SLinus Torvalds { 19691a028e50SDavid S. Miller int start = skb_headlen(skb); 19701a028e50SDavid S. Miller int i, copy = start - offset; 1971fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 19721da177e4SLinus Torvalds int pos = 0; 19731da177e4SLinus Torvalds 19741da177e4SLinus Torvalds /* Checksum header. */ 19751da177e4SLinus Torvalds if (copy > 0) { 19761da177e4SLinus Torvalds if (copy > len) 19771da177e4SLinus Torvalds copy = len; 19782817a336SDaniel Borkmann csum = ops->update(skb->data + offset, copy, csum); 19791da177e4SLinus Torvalds if ((len -= copy) == 0) 19801da177e4SLinus Torvalds return csum; 19811da177e4SLinus Torvalds offset += copy; 19821da177e4SLinus Torvalds pos = copy; 19831da177e4SLinus Torvalds } 19841da177e4SLinus Torvalds 19851da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 19861a028e50SDavid S. Miller int end; 198751c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 19881da177e4SLinus Torvalds 1989547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19901a028e50SDavid S. Miller 199151c56b00SEric Dumazet end = start + skb_frag_size(frag); 19921da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 199344bb9363SAl Viro __wsum csum2; 19941da177e4SLinus Torvalds u8 *vaddr; 19951da177e4SLinus Torvalds 19961da177e4SLinus Torvalds if (copy > len) 19971da177e4SLinus Torvalds copy = len; 199851c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 19992817a336SDaniel Borkmann csum2 = ops->update(vaddr + frag->page_offset + 20001a028e50SDavid S. Miller offset - start, copy, 0); 200151c56b00SEric Dumazet kunmap_atomic(vaddr); 20022817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 20031da177e4SLinus Torvalds if (!(len -= copy)) 20041da177e4SLinus Torvalds return csum; 20051da177e4SLinus Torvalds offset += copy; 20061da177e4SLinus Torvalds pos += copy; 20071da177e4SLinus Torvalds } 20081a028e50SDavid S. Miller start = end; 20091da177e4SLinus Torvalds } 20101da177e4SLinus Torvalds 2011fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 20121a028e50SDavid S. Miller int end; 20131da177e4SLinus Torvalds 2014547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20151a028e50SDavid S. Miller 2016fbb398a8SDavid S. Miller end = start + frag_iter->len; 20171da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 20185f92a738SAl Viro __wsum csum2; 20191da177e4SLinus Torvalds if (copy > len) 20201da177e4SLinus Torvalds copy = len; 20212817a336SDaniel Borkmann csum2 = __skb_checksum(frag_iter, offset - start, 20222817a336SDaniel Borkmann copy, 0, ops); 20232817a336SDaniel Borkmann csum = ops->combine(csum, csum2, pos, copy); 20241da177e4SLinus Torvalds if ((len -= copy) == 0) 20251da177e4SLinus Torvalds return csum; 20261da177e4SLinus Torvalds offset += copy; 20271da177e4SLinus Torvalds pos += copy; 20281da177e4SLinus Torvalds } 20291a028e50SDavid S. Miller start = end; 20301da177e4SLinus Torvalds } 203109a62660SKris Katterjohn BUG_ON(len); 20321da177e4SLinus Torvalds 20331da177e4SLinus Torvalds return csum; 20341da177e4SLinus Torvalds } 20352817a336SDaniel Borkmann EXPORT_SYMBOL(__skb_checksum); 20362817a336SDaniel Borkmann 20372817a336SDaniel Borkmann __wsum skb_checksum(const struct sk_buff *skb, int offset, 20382817a336SDaniel Borkmann int len, __wsum csum) 20392817a336SDaniel Borkmann { 20402817a336SDaniel Borkmann const struct skb_checksum_ops ops = { 2041cea80ea8SDaniel Borkmann .update = csum_partial_ext, 20422817a336SDaniel Borkmann .combine = csum_block_add_ext, 20432817a336SDaniel Borkmann }; 20442817a336SDaniel Borkmann 20452817a336SDaniel Borkmann return __skb_checksum(skb, offset, len, csum, &ops); 20462817a336SDaniel Borkmann } 2047b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 20481da177e4SLinus Torvalds 20491da177e4SLinus Torvalds /* Both of above in one bottle. */ 20501da177e4SLinus Torvalds 205181d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 205281d77662SAl Viro u8 *to, int len, __wsum csum) 20531da177e4SLinus Torvalds { 20541a028e50SDavid S. Miller int start = skb_headlen(skb); 20551a028e50SDavid S. Miller int i, copy = start - offset; 2056fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 20571da177e4SLinus Torvalds int pos = 0; 20581da177e4SLinus Torvalds 20591da177e4SLinus Torvalds /* Copy header. */ 20601da177e4SLinus Torvalds if (copy > 0) { 20611da177e4SLinus Torvalds if (copy > len) 20621da177e4SLinus Torvalds copy = len; 20631da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 20641da177e4SLinus Torvalds copy, csum); 20651da177e4SLinus Torvalds if ((len -= copy) == 0) 20661da177e4SLinus Torvalds return csum; 20671da177e4SLinus Torvalds offset += copy; 20681da177e4SLinus Torvalds to += copy; 20691da177e4SLinus Torvalds pos = copy; 20701da177e4SLinus Torvalds } 20711da177e4SLinus Torvalds 20721da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 20731a028e50SDavid S. Miller int end; 20741da177e4SLinus Torvalds 2075547b792cSIlpo Järvinen WARN_ON(start > offset + len); 20761a028e50SDavid S. Miller 20779e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 20781da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 20795084205fSAl Viro __wsum csum2; 20801da177e4SLinus Torvalds u8 *vaddr; 20811da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 20821da177e4SLinus Torvalds 20831da177e4SLinus Torvalds if (copy > len) 20841da177e4SLinus Torvalds copy = len; 208551c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 20861da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 20871a028e50SDavid S. Miller frag->page_offset + 20881a028e50SDavid S. Miller offset - start, to, 20891a028e50SDavid S. Miller copy, 0); 209051c56b00SEric Dumazet kunmap_atomic(vaddr); 20911da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 20921da177e4SLinus Torvalds if (!(len -= copy)) 20931da177e4SLinus Torvalds return csum; 20941da177e4SLinus Torvalds offset += copy; 20951da177e4SLinus Torvalds to += copy; 20961da177e4SLinus Torvalds pos += copy; 20971da177e4SLinus Torvalds } 20981a028e50SDavid S. Miller start = end; 20991da177e4SLinus Torvalds } 21001da177e4SLinus Torvalds 2101fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 210281d77662SAl Viro __wsum csum2; 21031a028e50SDavid S. Miller int end; 21041da177e4SLinus Torvalds 2105547b792cSIlpo Järvinen WARN_ON(start > offset + len); 21061a028e50SDavid S. Miller 2107fbb398a8SDavid S. Miller end = start + frag_iter->len; 21081da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 21091da177e4SLinus Torvalds if (copy > len) 21101da177e4SLinus Torvalds copy = len; 2111fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 21121a028e50SDavid S. Miller offset - start, 21131da177e4SLinus Torvalds to, copy, 0); 21141da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 21151da177e4SLinus Torvalds if ((len -= copy) == 0) 21161da177e4SLinus Torvalds return csum; 21171da177e4SLinus Torvalds offset += copy; 21181da177e4SLinus Torvalds to += copy; 21191da177e4SLinus Torvalds pos += copy; 21201da177e4SLinus Torvalds } 21211a028e50SDavid S. Miller start = end; 21221da177e4SLinus Torvalds } 212309a62660SKris Katterjohn BUG_ON(len); 21241da177e4SLinus Torvalds return csum; 21251da177e4SLinus Torvalds } 2126b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 21271da177e4SLinus Torvalds 21281da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 21291da177e4SLinus Torvalds { 2130d3bc23e7SAl Viro __wsum csum; 21311da177e4SLinus Torvalds long csstart; 21321da177e4SLinus Torvalds 213384fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 213455508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 21351da177e4SLinus Torvalds else 21361da177e4SLinus Torvalds csstart = skb_headlen(skb); 21371da177e4SLinus Torvalds 213809a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 21391da177e4SLinus Torvalds 2140d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 21411da177e4SLinus Torvalds 21421da177e4SLinus Torvalds csum = 0; 21431da177e4SLinus Torvalds if (csstart != skb->len) 21441da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 21451da177e4SLinus Torvalds skb->len - csstart, 0); 21461da177e4SLinus Torvalds 214784fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2148ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 21491da177e4SLinus Torvalds 2150d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 21511da177e4SLinus Torvalds } 21521da177e4SLinus Torvalds } 2153b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 21541da177e4SLinus Torvalds 21551da177e4SLinus Torvalds /** 21561da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 21571da177e4SLinus Torvalds * @list: list to dequeue from 21581da177e4SLinus Torvalds * 21591da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 21601da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 21611da177e4SLinus Torvalds * returned or %NULL if the list is empty. 21621da177e4SLinus Torvalds */ 21631da177e4SLinus Torvalds 21641da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 21651da177e4SLinus Torvalds { 21661da177e4SLinus Torvalds unsigned long flags; 21671da177e4SLinus Torvalds struct sk_buff *result; 21681da177e4SLinus Torvalds 21691da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 21701da177e4SLinus Torvalds result = __skb_dequeue(list); 21711da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 21721da177e4SLinus Torvalds return result; 21731da177e4SLinus Torvalds } 2174b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 21751da177e4SLinus Torvalds 21761da177e4SLinus Torvalds /** 21771da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 21781da177e4SLinus Torvalds * @list: list to dequeue from 21791da177e4SLinus Torvalds * 21801da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 21811da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 21821da177e4SLinus Torvalds * returned or %NULL if the list is empty. 21831da177e4SLinus Torvalds */ 21841da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 21851da177e4SLinus Torvalds { 21861da177e4SLinus Torvalds unsigned long flags; 21871da177e4SLinus Torvalds struct sk_buff *result; 21881da177e4SLinus Torvalds 21891da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 21901da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 21911da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 21921da177e4SLinus Torvalds return result; 21931da177e4SLinus Torvalds } 2194b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 21951da177e4SLinus Torvalds 21961da177e4SLinus Torvalds /** 21971da177e4SLinus Torvalds * skb_queue_purge - empty a list 21981da177e4SLinus Torvalds * @list: list to empty 21991da177e4SLinus Torvalds * 22001da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 22011da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 22021da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 22031da177e4SLinus Torvalds */ 22041da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 22051da177e4SLinus Torvalds { 22061da177e4SLinus Torvalds struct sk_buff *skb; 22071da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 22081da177e4SLinus Torvalds kfree_skb(skb); 22091da177e4SLinus Torvalds } 2210b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 22111da177e4SLinus Torvalds 22121da177e4SLinus Torvalds /** 22131da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 22141da177e4SLinus Torvalds * @list: list to use 22151da177e4SLinus Torvalds * @newsk: buffer to queue 22161da177e4SLinus Torvalds * 22171da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 22181da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 22191da177e4SLinus Torvalds * safely. 22201da177e4SLinus Torvalds * 22211da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 22221da177e4SLinus Torvalds */ 22231da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 22241da177e4SLinus Torvalds { 22251da177e4SLinus Torvalds unsigned long flags; 22261da177e4SLinus Torvalds 22271da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 22281da177e4SLinus Torvalds __skb_queue_head(list, newsk); 22291da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 22301da177e4SLinus Torvalds } 2231b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 22321da177e4SLinus Torvalds 22331da177e4SLinus Torvalds /** 22341da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 22351da177e4SLinus Torvalds * @list: list to use 22361da177e4SLinus Torvalds * @newsk: buffer to queue 22371da177e4SLinus Torvalds * 22381da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 22391da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 22401da177e4SLinus Torvalds * safely. 22411da177e4SLinus Torvalds * 22421da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 22431da177e4SLinus Torvalds */ 22441da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 22451da177e4SLinus Torvalds { 22461da177e4SLinus Torvalds unsigned long flags; 22471da177e4SLinus Torvalds 22481da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 22491da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 22501da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 22511da177e4SLinus Torvalds } 2252b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 22538728b834SDavid S. Miller 22541da177e4SLinus Torvalds /** 22551da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 22561da177e4SLinus Torvalds * @skb: buffer to remove 22578728b834SDavid S. Miller * @list: list to use 22581da177e4SLinus Torvalds * 22598728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 22608728b834SDavid S. Miller * function is atomic with respect to other list locked calls 22611da177e4SLinus Torvalds * 22628728b834SDavid S. Miller * You must know what list the SKB is on. 22631da177e4SLinus Torvalds */ 22648728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 22651da177e4SLinus Torvalds { 22661da177e4SLinus Torvalds unsigned long flags; 22671da177e4SLinus Torvalds 22681da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 22698728b834SDavid S. Miller __skb_unlink(skb, list); 22701da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 22711da177e4SLinus Torvalds } 2272b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 22731da177e4SLinus Torvalds 22741da177e4SLinus Torvalds /** 22751da177e4SLinus Torvalds * skb_append - append a buffer 22761da177e4SLinus Torvalds * @old: buffer to insert after 22771da177e4SLinus Torvalds * @newsk: buffer to insert 22788728b834SDavid S. Miller * @list: list to use 22791da177e4SLinus Torvalds * 22801da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 22811da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 22821da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 22831da177e4SLinus Torvalds */ 22848728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 22851da177e4SLinus Torvalds { 22861da177e4SLinus Torvalds unsigned long flags; 22871da177e4SLinus Torvalds 22888728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 22897de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 22908728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 22911da177e4SLinus Torvalds } 2292b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds /** 22951da177e4SLinus Torvalds * skb_insert - insert a buffer 22961da177e4SLinus Torvalds * @old: buffer to insert before 22971da177e4SLinus Torvalds * @newsk: buffer to insert 22988728b834SDavid S. Miller * @list: list to use 22991da177e4SLinus Torvalds * 23008728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 23018728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 23028728b834SDavid S. Miller * calls. 23038728b834SDavid S. Miller * 23041da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 23051da177e4SLinus Torvalds */ 23068728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 23071da177e4SLinus Torvalds { 23081da177e4SLinus Torvalds unsigned long flags; 23091da177e4SLinus Torvalds 23108728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 23118728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 23128728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 23131da177e4SLinus Torvalds } 2314b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 23151da177e4SLinus Torvalds 23161da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 23171da177e4SLinus Torvalds struct sk_buff* skb1, 23181da177e4SLinus Torvalds const u32 len, const int pos) 23191da177e4SLinus Torvalds { 23201da177e4SLinus Torvalds int i; 23211da177e4SLinus Torvalds 2322d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2323d626f62bSArnaldo Carvalho de Melo pos - len); 23241da177e4SLinus Torvalds /* And move data appendix as is. */ 23251da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 23261da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 23271da177e4SLinus Torvalds 23281da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 23291da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 23301da177e4SLinus Torvalds skb1->data_len = skb->data_len; 23311da177e4SLinus Torvalds skb1->len += skb1->data_len; 23321da177e4SLinus Torvalds skb->data_len = 0; 23331da177e4SLinus Torvalds skb->len = len; 233427a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 23351da177e4SLinus Torvalds } 23361da177e4SLinus Torvalds 23371da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 23381da177e4SLinus Torvalds struct sk_buff* skb1, 23391da177e4SLinus Torvalds const u32 len, int pos) 23401da177e4SLinus Torvalds { 23411da177e4SLinus Torvalds int i, k = 0; 23421da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 23431da177e4SLinus Torvalds 23441da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 23451da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 23461da177e4SLinus Torvalds skb->len = len; 23471da177e4SLinus Torvalds skb->data_len = len - pos; 23481da177e4SLinus Torvalds 23491da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 23509e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 23511da177e4SLinus Torvalds 23521da177e4SLinus Torvalds if (pos + size > len) { 23531da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 23541da177e4SLinus Torvalds 23551da177e4SLinus Torvalds if (pos < len) { 23561da177e4SLinus Torvalds /* Split frag. 23571da177e4SLinus Torvalds * We have two variants in this case: 23581da177e4SLinus Torvalds * 1. Move all the frag to the second 23591da177e4SLinus Torvalds * part, if it is possible. F.e. 23601da177e4SLinus Torvalds * this approach is mandatory for TUX, 23611da177e4SLinus Torvalds * where splitting is expensive. 23621da177e4SLinus Torvalds * 2. Split is accurately. We make this. 23631da177e4SLinus Torvalds */ 2364ea2ab693SIan Campbell skb_frag_ref(skb, i); 23651da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 23669e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 23679e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 23681da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 23691da177e4SLinus Torvalds } 23701da177e4SLinus Torvalds k++; 23711da177e4SLinus Torvalds } else 23721da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 23731da177e4SLinus Torvalds pos += size; 23741da177e4SLinus Torvalds } 23751da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 23761da177e4SLinus Torvalds } 23771da177e4SLinus Torvalds 23781da177e4SLinus Torvalds /** 23791da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 23801da177e4SLinus Torvalds * @skb: the buffer to split 23811da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 23821da177e4SLinus Torvalds * @len: new length for skb 23831da177e4SLinus Torvalds */ 23841da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 23851da177e4SLinus Torvalds { 23861da177e4SLinus Torvalds int pos = skb_headlen(skb); 23871da177e4SLinus Torvalds 238868534c68SAmerigo Wang skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 23891da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 23901da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 23911da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 23921da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 23931da177e4SLinus Torvalds } 2394b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 23951da177e4SLinus Torvalds 23969f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 23979f782db3SIlpo Järvinen * 23989f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 23999f782db3SIlpo Järvinen */ 2400832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2401832d11c5SIlpo Järvinen { 24020ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2403832d11c5SIlpo Järvinen } 2404832d11c5SIlpo Järvinen 2405832d11c5SIlpo Järvinen /** 2406832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2407832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2408832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2409832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2410832d11c5SIlpo Järvinen * 2411832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 241220e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 2413832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2414832d11c5SIlpo Järvinen * 2415832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2416832d11c5SIlpo Järvinen * 2417832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2418832d11c5SIlpo Järvinen * to have non-paged data as well. 2419832d11c5SIlpo Järvinen * 2420832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2421832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2422832d11c5SIlpo Järvinen */ 2423832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2424832d11c5SIlpo Järvinen { 2425832d11c5SIlpo Järvinen int from, to, merge, todo; 2426832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2427832d11c5SIlpo Järvinen 2428832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2429832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2430832d11c5SIlpo Järvinen 2431832d11c5SIlpo Järvinen todo = shiftlen; 2432832d11c5SIlpo Järvinen from = 0; 2433832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2434832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2435832d11c5SIlpo Järvinen 2436832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2437832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2438832d11c5SIlpo Järvinen */ 2439832d11c5SIlpo Järvinen if (!to || 2440ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2441ea2ab693SIan Campbell fragfrom->page_offset)) { 2442832d11c5SIlpo Järvinen merge = -1; 2443832d11c5SIlpo Järvinen } else { 2444832d11c5SIlpo Järvinen merge = to - 1; 2445832d11c5SIlpo Järvinen 24469e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2447832d11c5SIlpo Järvinen if (todo < 0) { 2448832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2449832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2450832d11c5SIlpo Järvinen return 0; 2451832d11c5SIlpo Järvinen 24529f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 24539f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2454832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2455832d11c5SIlpo Järvinen 24569e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 24579e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 2458832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2459832d11c5SIlpo Järvinen 2460832d11c5SIlpo Järvinen goto onlymerged; 2461832d11c5SIlpo Järvinen } 2462832d11c5SIlpo Järvinen 2463832d11c5SIlpo Järvinen from++; 2464832d11c5SIlpo Järvinen } 2465832d11c5SIlpo Järvinen 2466832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2467832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2468832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2469832d11c5SIlpo Järvinen return 0; 2470832d11c5SIlpo Järvinen 2471832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2472832d11c5SIlpo Järvinen return 0; 2473832d11c5SIlpo Järvinen 2474832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2475832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2476832d11c5SIlpo Järvinen return 0; 2477832d11c5SIlpo Järvinen 2478832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2479832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2480832d11c5SIlpo Järvinen 24819e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 2482832d11c5SIlpo Järvinen *fragto = *fragfrom; 24839e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2484832d11c5SIlpo Järvinen from++; 2485832d11c5SIlpo Järvinen to++; 2486832d11c5SIlpo Järvinen 2487832d11c5SIlpo Järvinen } else { 2488ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2489832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2490832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 24919e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 2492832d11c5SIlpo Järvinen 2493832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 24949e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 2495832d11c5SIlpo Järvinen todo = 0; 2496832d11c5SIlpo Järvinen 2497832d11c5SIlpo Järvinen to++; 2498832d11c5SIlpo Järvinen break; 2499832d11c5SIlpo Järvinen } 2500832d11c5SIlpo Järvinen } 2501832d11c5SIlpo Järvinen 2502832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2503832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2504832d11c5SIlpo Järvinen 2505832d11c5SIlpo Järvinen if (merge >= 0) { 2506832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2507832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2508832d11c5SIlpo Järvinen 25099e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2510ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2511832d11c5SIlpo Järvinen } 2512832d11c5SIlpo Järvinen 2513832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2514832d11c5SIlpo Järvinen to = 0; 2515832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2516832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2517832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2518832d11c5SIlpo Järvinen 2519832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2520832d11c5SIlpo Järvinen 2521832d11c5SIlpo Järvinen onlymerged: 2522832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2523832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2524832d11c5SIlpo Järvinen */ 2525832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2526832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2527832d11c5SIlpo Järvinen 2528832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2529832d11c5SIlpo Järvinen skb->len -= shiftlen; 2530832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2531832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2532832d11c5SIlpo Järvinen tgt->len += shiftlen; 2533832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2534832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2535832d11c5SIlpo Järvinen 2536832d11c5SIlpo Järvinen return shiftlen; 2537832d11c5SIlpo Järvinen } 2538832d11c5SIlpo Järvinen 2539677e90edSThomas Graf /** 2540677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2541677e90edSThomas Graf * @skb: the buffer to read 2542677e90edSThomas Graf * @from: lower offset of data to be read 2543677e90edSThomas Graf * @to: upper offset of data to be read 2544677e90edSThomas Graf * @st: state variable 2545677e90edSThomas Graf * 2546677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2547677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2548677e90edSThomas Graf */ 2549677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2550677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2551677e90edSThomas Graf { 2552677e90edSThomas Graf st->lower_offset = from; 2553677e90edSThomas Graf st->upper_offset = to; 2554677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2555677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2556677e90edSThomas Graf st->frag_data = NULL; 2557677e90edSThomas Graf } 2558b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2559677e90edSThomas Graf 2560677e90edSThomas Graf /** 2561677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2562677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2563677e90edSThomas Graf * @data: destination pointer for data to be returned 2564677e90edSThomas Graf * @st: state variable 2565677e90edSThomas Graf * 2566bc32383cSMathias Krause * Reads a block of skb data at @consumed relative to the 2567677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2568bc32383cSMathias Krause * the head of the data block to @data and returns the length 2569677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2570677e90edSThomas Graf * offset has been reached. 2571677e90edSThomas Graf * 2572677e90edSThomas Graf * The caller is not required to consume all of the data 2573bc32383cSMathias Krause * returned, i.e. @consumed is typically set to the number 2574677e90edSThomas Graf * of bytes already consumed and the next call to 2575677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2576677e90edSThomas Graf * 257725985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2578677e90edSThomas Graf * this limitation is the cost for zerocopy seqeuental 2579677e90edSThomas Graf * reads of potentially non linear data. 2580677e90edSThomas Graf * 2581bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2582677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2583677e90edSThomas Graf * a stack for this purpose. 2584677e90edSThomas Graf */ 2585677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2586677e90edSThomas Graf struct skb_seq_state *st) 2587677e90edSThomas Graf { 2588677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2589677e90edSThomas Graf skb_frag_t *frag; 2590677e90edSThomas Graf 2591aeb193eaSWedson Almeida Filho if (unlikely(abs_offset >= st->upper_offset)) { 2592aeb193eaSWedson Almeida Filho if (st->frag_data) { 2593aeb193eaSWedson Almeida Filho kunmap_atomic(st->frag_data); 2594aeb193eaSWedson Almeida Filho st->frag_data = NULL; 2595aeb193eaSWedson Almeida Filho } 2596677e90edSThomas Graf return 0; 2597aeb193eaSWedson Almeida Filho } 2598677e90edSThomas Graf 2599677e90edSThomas Graf next_skb: 260095e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2601677e90edSThomas Graf 2602995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 260395e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2604677e90edSThomas Graf return block_limit - abs_offset; 2605677e90edSThomas Graf } 2606677e90edSThomas Graf 2607677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2608677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2609677e90edSThomas Graf 2610677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2611677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 26129e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 2613677e90edSThomas Graf 2614677e90edSThomas Graf if (abs_offset < block_limit) { 2615677e90edSThomas Graf if (!st->frag_data) 261651c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 2617677e90edSThomas Graf 2618677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2619677e90edSThomas Graf (abs_offset - st->stepped_offset); 2620677e90edSThomas Graf 2621677e90edSThomas Graf return block_limit - abs_offset; 2622677e90edSThomas Graf } 2623677e90edSThomas Graf 2624677e90edSThomas Graf if (st->frag_data) { 262551c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2626677e90edSThomas Graf st->frag_data = NULL; 2627677e90edSThomas Graf } 2628677e90edSThomas Graf 2629677e90edSThomas Graf st->frag_idx++; 26309e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 2631677e90edSThomas Graf } 2632677e90edSThomas Graf 26335b5a60daSOlaf Kirch if (st->frag_data) { 263451c56b00SEric Dumazet kunmap_atomic(st->frag_data); 26355b5a60daSOlaf Kirch st->frag_data = NULL; 26365b5a60daSOlaf Kirch } 26375b5a60daSOlaf Kirch 263821dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2639677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 264095e3b24cSHerbert Xu st->frag_idx = 0; 2641677e90edSThomas Graf goto next_skb; 264271b3346dSShyam Iyer } else if (st->cur_skb->next) { 264371b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 264471b3346dSShyam Iyer st->frag_idx = 0; 2645677e90edSThomas Graf goto next_skb; 2646677e90edSThomas Graf } 2647677e90edSThomas Graf 2648677e90edSThomas Graf return 0; 2649677e90edSThomas Graf } 2650b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2651677e90edSThomas Graf 2652677e90edSThomas Graf /** 2653677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2654677e90edSThomas Graf * @st: state variable 2655677e90edSThomas Graf * 2656677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2657677e90edSThomas Graf * returned 0. 2658677e90edSThomas Graf */ 2659677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2660677e90edSThomas Graf { 2661677e90edSThomas Graf if (st->frag_data) 266251c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2663677e90edSThomas Graf } 2664b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2665677e90edSThomas Graf 26663fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 26673fc7e8a6SThomas Graf 26683fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 26693fc7e8a6SThomas Graf struct ts_config *conf, 26703fc7e8a6SThomas Graf struct ts_state *state) 26713fc7e8a6SThomas Graf { 26723fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 26733fc7e8a6SThomas Graf } 26743fc7e8a6SThomas Graf 26753fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 26763fc7e8a6SThomas Graf { 26773fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 26783fc7e8a6SThomas Graf } 26793fc7e8a6SThomas Graf 26803fc7e8a6SThomas Graf /** 26813fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 26823fc7e8a6SThomas Graf * @skb: the buffer to look in 26833fc7e8a6SThomas Graf * @from: search offset 26843fc7e8a6SThomas Graf * @to: search limit 26853fc7e8a6SThomas Graf * @config: textsearch configuration 26863fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 26873fc7e8a6SThomas Graf * 26883fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 26893fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 26903fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 26913fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 26923fc7e8a6SThomas Graf */ 26933fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 26943fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 26953fc7e8a6SThomas Graf struct ts_state *state) 26963fc7e8a6SThomas Graf { 2697f72b948dSPhil Oester unsigned int ret; 2698f72b948dSPhil Oester 26993fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 27003fc7e8a6SThomas Graf config->finish = skb_ts_finish; 27013fc7e8a6SThomas Graf 27023fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 27033fc7e8a6SThomas Graf 2704f72b948dSPhil Oester ret = textsearch_find(config, state); 2705f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 27063fc7e8a6SThomas Graf } 2707b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 27083fc7e8a6SThomas Graf 2709e89e9cf5SAnanda Raju /** 27102c53040fSBen Hutchings * skb_append_datato_frags - append the user data to a skb 2711e89e9cf5SAnanda Raju * @sk: sock structure 2712e89e9cf5SAnanda Raju * @skb: skb structure to be appened with user data. 2713e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2714e89e9cf5SAnanda Raju * @from: pointer to user message iov 2715e89e9cf5SAnanda Raju * @length: length of the iov message 2716e89e9cf5SAnanda Raju * 2717e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2718e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2719e89e9cf5SAnanda Raju */ 2720e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2721dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2722e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2723e89e9cf5SAnanda Raju void *from, int length) 2724e89e9cf5SAnanda Raju { 2725b2111724SEric Dumazet int frg_cnt = skb_shinfo(skb)->nr_frags; 2726b2111724SEric Dumazet int copy; 2727e89e9cf5SAnanda Raju int offset = 0; 2728e89e9cf5SAnanda Raju int ret; 2729b2111724SEric Dumazet struct page_frag *pfrag = ¤t->task_frag; 2730e89e9cf5SAnanda Raju 2731e89e9cf5SAnanda Raju do { 2732e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2733e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2734b2111724SEric Dumazet return -EMSGSIZE; 2735e89e9cf5SAnanda Raju 2736b2111724SEric Dumazet if (!sk_page_frag_refill(sk, pfrag)) 2737e89e9cf5SAnanda Raju return -ENOMEM; 2738e89e9cf5SAnanda Raju 2739e89e9cf5SAnanda Raju /* copy the user data to page */ 2740b2111724SEric Dumazet copy = min_t(int, length, pfrag->size - pfrag->offset); 2741e89e9cf5SAnanda Raju 2742b2111724SEric Dumazet ret = getfrag(from, page_address(pfrag->page) + pfrag->offset, 2743e89e9cf5SAnanda Raju offset, copy, 0, skb); 2744e89e9cf5SAnanda Raju if (ret < 0) 2745e89e9cf5SAnanda Raju return -EFAULT; 2746e89e9cf5SAnanda Raju 2747e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2748b2111724SEric Dumazet skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset, 2749b2111724SEric Dumazet copy); 2750b2111724SEric Dumazet frg_cnt++; 2751b2111724SEric Dumazet pfrag->offset += copy; 2752b2111724SEric Dumazet get_page(pfrag->page); 2753b2111724SEric Dumazet 2754b2111724SEric Dumazet skb->truesize += copy; 2755b2111724SEric Dumazet atomic_add(copy, &sk->sk_wmem_alloc); 2756e89e9cf5SAnanda Raju skb->len += copy; 2757e89e9cf5SAnanda Raju skb->data_len += copy; 2758e89e9cf5SAnanda Raju offset += copy; 2759e89e9cf5SAnanda Raju length -= copy; 2760e89e9cf5SAnanda Raju 2761e89e9cf5SAnanda Raju } while (length > 0); 2762e89e9cf5SAnanda Raju 2763e89e9cf5SAnanda Raju return 0; 2764e89e9cf5SAnanda Raju } 2765b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2766e89e9cf5SAnanda Raju 2767cbb042f9SHerbert Xu /** 2768cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2769cbb042f9SHerbert Xu * @skb: buffer to update 2770cbb042f9SHerbert Xu * @len: length of data pulled 2771cbb042f9SHerbert Xu * 2772cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2773fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 277484fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 277584fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 277684fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2777cbb042f9SHerbert Xu */ 2778cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2779cbb042f9SHerbert Xu { 2780cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2781cbb042f9SHerbert Xu skb->len -= len; 2782cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2783cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2784cbb042f9SHerbert Xu return skb->data += len; 2785cbb042f9SHerbert Xu } 2786f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2787f94691acSArnaldo Carvalho de Melo 2788f4c50d99SHerbert Xu /** 2789f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2790f4c50d99SHerbert Xu * @skb: buffer to segment 2791576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2792f4c50d99SHerbert Xu * 2793f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 27944c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 27954c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2796f4c50d99SHerbert Xu */ 2797c8f44affSMichał Mirosław struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features) 2798f4c50d99SHerbert Xu { 2799f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2800f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 280189319d38SHerbert Xu struct sk_buff *fskb = skb_shinfo(skb)->frag_list; 2802f4c50d99SHerbert Xu unsigned int mss = skb_shinfo(skb)->gso_size; 280398e399f8SArnaldo Carvalho de Melo unsigned int doffset = skb->data - skb_mac_header(skb); 2804f4c50d99SHerbert Xu unsigned int offset = doffset; 280568c33163SPravin B Shelar unsigned int tnl_hlen = skb_tnl_header_len(skb); 2806f4c50d99SHerbert Xu unsigned int headroom; 2807f4c50d99SHerbert Xu unsigned int len; 2808ec5f0615SPravin B Shelar __be16 proto; 2809ec5f0615SPravin B Shelar bool csum; 281004ed3e74SMichał Mirosław int sg = !!(features & NETIF_F_SG); 2811f4c50d99SHerbert Xu int nfrags = skb_shinfo(skb)->nr_frags; 2812f4c50d99SHerbert Xu int err = -ENOMEM; 2813f4c50d99SHerbert Xu int i = 0; 2814f4c50d99SHerbert Xu int pos; 2815f4c50d99SHerbert Xu 2816ec5f0615SPravin B Shelar proto = skb_network_protocol(skb); 2817ec5f0615SPravin B Shelar if (unlikely(!proto)) 2818ec5f0615SPravin B Shelar return ERR_PTR(-EINVAL); 2819ec5f0615SPravin B Shelar 2820ec5f0615SPravin B Shelar csum = !!can_checksum_protocol(features, proto); 2821f4c50d99SHerbert Xu __skb_push(skb, doffset); 2822f4c50d99SHerbert Xu headroom = skb_headroom(skb); 2823f4c50d99SHerbert Xu pos = skb_headlen(skb); 2824f4c50d99SHerbert Xu 2825f4c50d99SHerbert Xu do { 2826f4c50d99SHerbert Xu struct sk_buff *nskb; 2827f4c50d99SHerbert Xu skb_frag_t *frag; 2828c8884eddSHerbert Xu int hsize; 2829f4c50d99SHerbert Xu int size; 2830f4c50d99SHerbert Xu 2831f4c50d99SHerbert Xu len = skb->len - offset; 2832f4c50d99SHerbert Xu if (len > mss) 2833f4c50d99SHerbert Xu len = mss; 2834f4c50d99SHerbert Xu 2835f4c50d99SHerbert Xu hsize = skb_headlen(skb) - offset; 2836f4c50d99SHerbert Xu if (hsize < 0) 2837f4c50d99SHerbert Xu hsize = 0; 2838c8884eddSHerbert Xu if (hsize > len || !sg) 2839c8884eddSHerbert Xu hsize = len; 2840f4c50d99SHerbert Xu 284189319d38SHerbert Xu if (!hsize && i >= nfrags) { 284289319d38SHerbert Xu BUG_ON(fskb->len != len); 284389319d38SHerbert Xu 284489319d38SHerbert Xu pos += len; 284589319d38SHerbert Xu nskb = skb_clone(fskb, GFP_ATOMIC); 284689319d38SHerbert Xu fskb = fskb->next; 284789319d38SHerbert Xu 2848f4c50d99SHerbert Xu if (unlikely(!nskb)) 2849f4c50d99SHerbert Xu goto err; 2850f4c50d99SHerbert Xu 2851ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 285289319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 285389319d38SHerbert Xu kfree_skb(nskb); 285489319d38SHerbert Xu goto err; 285589319d38SHerbert Xu } 285689319d38SHerbert Xu 2857ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 285889319d38SHerbert Xu skb_release_head_state(nskb); 285989319d38SHerbert Xu __skb_push(nskb, doffset); 286089319d38SHerbert Xu } else { 2861c93bdd0eSMel Gorman nskb = __alloc_skb(hsize + doffset + headroom, 2862c93bdd0eSMel Gorman GFP_ATOMIC, skb_alloc_rx_flag(skb), 2863c93bdd0eSMel Gorman NUMA_NO_NODE); 286489319d38SHerbert Xu 286589319d38SHerbert Xu if (unlikely(!nskb)) 286689319d38SHerbert Xu goto err; 286789319d38SHerbert Xu 286889319d38SHerbert Xu skb_reserve(nskb, headroom); 286989319d38SHerbert Xu __skb_put(nskb, doffset); 287089319d38SHerbert Xu } 287189319d38SHerbert Xu 2872f4c50d99SHerbert Xu if (segs) 2873f4c50d99SHerbert Xu tail->next = nskb; 2874f4c50d99SHerbert Xu else 2875f4c50d99SHerbert Xu segs = nskb; 2876f4c50d99SHerbert Xu tail = nskb; 2877f4c50d99SHerbert Xu 28786f85a124SHerbert Xu __copy_skb_header(nskb, skb); 2879f4c50d99SHerbert Xu nskb->mac_len = skb->mac_len; 2880f4c50d99SHerbert Xu 2881030737bcSEric Dumazet skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 288268c33163SPravin B Shelar 288368c33163SPravin B Shelar skb_copy_from_linear_data_offset(skb, -tnl_hlen, 288468c33163SPravin B Shelar nskb->data - tnl_hlen, 288568c33163SPravin B Shelar doffset + tnl_hlen); 288689319d38SHerbert Xu 28872f181855SHerbert Xu if (fskb != skb_shinfo(skb)->frag_list) 28881cdbcb79SSimon Horman goto perform_csum_check; 288989319d38SHerbert Xu 2890f4c50d99SHerbert Xu if (!sg) { 28916f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 2892f4c50d99SHerbert Xu nskb->csum = skb_copy_and_csum_bits(skb, offset, 2893f4c50d99SHerbert Xu skb_put(nskb, len), 2894f4c50d99SHerbert Xu len, 0); 2895f4c50d99SHerbert Xu continue; 2896f4c50d99SHerbert Xu } 2897f4c50d99SHerbert Xu 2898f4c50d99SHerbert Xu frag = skb_shinfo(nskb)->frags; 2899f4c50d99SHerbert Xu 2900d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, 2901d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 2902f4c50d99SHerbert Xu 2903c9af6db4SPravin B Shelar skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2904cef401deSEric Dumazet 290589319d38SHerbert Xu while (pos < offset + len && i < nfrags) { 2906f4c50d99SHerbert Xu *frag = skb_shinfo(skb)->frags[i]; 2907ea2ab693SIan Campbell __skb_frag_ref(frag); 29089e903e08SEric Dumazet size = skb_frag_size(frag); 2909f4c50d99SHerbert Xu 2910f4c50d99SHerbert Xu if (pos < offset) { 2911f4c50d99SHerbert Xu frag->page_offset += offset - pos; 29129e903e08SEric Dumazet skb_frag_size_sub(frag, offset - pos); 2913f4c50d99SHerbert Xu } 2914f4c50d99SHerbert Xu 291589319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 2916f4c50d99SHerbert Xu 2917f4c50d99SHerbert Xu if (pos + size <= offset + len) { 2918f4c50d99SHerbert Xu i++; 2919f4c50d99SHerbert Xu pos += size; 2920f4c50d99SHerbert Xu } else { 29219e903e08SEric Dumazet skb_frag_size_sub(frag, pos + size - (offset + len)); 292289319d38SHerbert Xu goto skip_fraglist; 2923f4c50d99SHerbert Xu } 2924f4c50d99SHerbert Xu 2925f4c50d99SHerbert Xu frag++; 2926f4c50d99SHerbert Xu } 2927f4c50d99SHerbert Xu 292889319d38SHerbert Xu if (pos < offset + len) { 292989319d38SHerbert Xu struct sk_buff *fskb2 = fskb; 293089319d38SHerbert Xu 293189319d38SHerbert Xu BUG_ON(pos + fskb->len != offset + len); 293289319d38SHerbert Xu 293389319d38SHerbert Xu pos += fskb->len; 293489319d38SHerbert Xu fskb = fskb->next; 293589319d38SHerbert Xu 293689319d38SHerbert Xu if (fskb2->next) { 293789319d38SHerbert Xu fskb2 = skb_clone(fskb2, GFP_ATOMIC); 293889319d38SHerbert Xu if (!fskb2) 293989319d38SHerbert Xu goto err; 294089319d38SHerbert Xu } else 294189319d38SHerbert Xu skb_get(fskb2); 294289319d38SHerbert Xu 2943fbb398a8SDavid S. Miller SKB_FRAG_ASSERT(nskb); 294489319d38SHerbert Xu skb_shinfo(nskb)->frag_list = fskb2; 294589319d38SHerbert Xu } 294689319d38SHerbert Xu 294789319d38SHerbert Xu skip_fraglist: 2948f4c50d99SHerbert Xu nskb->data_len = len - hsize; 2949f4c50d99SHerbert Xu nskb->len += nskb->data_len; 2950f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 2951ec5f0615SPravin B Shelar 29521cdbcb79SSimon Horman perform_csum_check: 2953ec5f0615SPravin B Shelar if (!csum) { 2954ec5f0615SPravin B Shelar nskb->csum = skb_checksum(nskb, doffset, 2955ec5f0615SPravin B Shelar nskb->len - doffset, 0); 2956ec5f0615SPravin B Shelar nskb->ip_summed = CHECKSUM_NONE; 2957ec5f0615SPravin B Shelar } 2958f4c50d99SHerbert Xu } while ((offset += len) < skb->len); 2959f4c50d99SHerbert Xu 2960f4c50d99SHerbert Xu return segs; 2961f4c50d99SHerbert Xu 2962f4c50d99SHerbert Xu err: 2963f4c50d99SHerbert Xu while ((skb = segs)) { 2964f4c50d99SHerbert Xu segs = skb->next; 2965b08d5840SPatrick McHardy kfree_skb(skb); 2966f4c50d99SHerbert Xu } 2967f4c50d99SHerbert Xu return ERR_PTR(err); 2968f4c50d99SHerbert Xu } 2969f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 2970f4c50d99SHerbert Xu 297171d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 297271d93b39SHerbert Xu { 29738a29111cSEric Dumazet struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 297467147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 297567147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 29768a29111cSEric Dumazet struct sk_buff *nskb, *lp, *p = *head; 29778a29111cSEric Dumazet unsigned int len = skb_gro_len(skb); 2978715dc1f3SEric Dumazet unsigned int delta_truesize; 29798a29111cSEric Dumazet unsigned int headroom; 298071d93b39SHerbert Xu 29818a29111cSEric Dumazet if (unlikely(p->len + len >= 65536)) 298271d93b39SHerbert Xu return -E2BIG; 298371d93b39SHerbert Xu 29848a29111cSEric Dumazet lp = NAPI_GRO_CB(p)->last ?: p; 29858a29111cSEric Dumazet pinfo = skb_shinfo(lp); 29868a29111cSEric Dumazet 29878a29111cSEric Dumazet if (headlen <= offset) { 298842da6994SHerbert Xu skb_frag_t *frag; 298966e92fcfSHerbert Xu skb_frag_t *frag2; 29909aaa156cSHerbert Xu int i = skbinfo->nr_frags; 29919aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 299242da6994SHerbert Xu 299366e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 29948a29111cSEric Dumazet goto merge; 299581705ad1SHerbert Xu 29968a29111cSEric Dumazet offset -= headlen; 29979aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 29989aaa156cSHerbert Xu skbinfo->nr_frags = 0; 2999f5572068SHerbert Xu 30009aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 30019aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 300266e92fcfSHerbert Xu do { 300366e92fcfSHerbert Xu *--frag = *--frag2; 300466e92fcfSHerbert Xu } while (--i); 300566e92fcfSHerbert Xu 300666e92fcfSHerbert Xu frag->page_offset += offset; 30079e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 300866e92fcfSHerbert Xu 3009715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 3010ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 3011ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 3012715dc1f3SEric Dumazet 3013f5572068SHerbert Xu skb->truesize -= skb->data_len; 3014f5572068SHerbert Xu skb->len -= skb->data_len; 3015f5572068SHerbert Xu skb->data_len = 0; 3016f5572068SHerbert Xu 3017715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 30185d38a079SHerbert Xu goto done; 3019d7e8883cSEric Dumazet } else if (skb->head_frag) { 3020d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 3021d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 3022d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 3023d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 3024d7e8883cSEric Dumazet unsigned int first_offset; 3025d7e8883cSEric Dumazet 3026d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 30278a29111cSEric Dumazet goto merge; 3028d7e8883cSEric Dumazet 3029d7e8883cSEric Dumazet first_offset = skb->data - 3030d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 3031d7e8883cSEric Dumazet offset; 3032d7e8883cSEric Dumazet 3033d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3034d7e8883cSEric Dumazet 3035d7e8883cSEric Dumazet frag->page.p = page; 3036d7e8883cSEric Dumazet frag->page_offset = first_offset; 3037d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 3038d7e8883cSEric Dumazet 3039d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3040d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 3041d7e8883cSEric Dumazet 3042715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3043d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3044d7e8883cSEric Dumazet goto done; 30458a29111cSEric Dumazet } 30468a29111cSEric Dumazet if (pinfo->frag_list) 30478a29111cSEric Dumazet goto merge; 30488a29111cSEric Dumazet if (skb_gro_len(p) != pinfo->gso_size) 304969c0cab1SHerbert Xu return -E2BIG; 305071d93b39SHerbert Xu 305171d93b39SHerbert Xu headroom = skb_headroom(p); 30523d3be433SEric Dumazet nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); 305371d93b39SHerbert Xu if (unlikely(!nskb)) 305471d93b39SHerbert Xu return -ENOMEM; 305571d93b39SHerbert Xu 305671d93b39SHerbert Xu __copy_skb_header(nskb, p); 305771d93b39SHerbert Xu nskb->mac_len = p->mac_len; 305871d93b39SHerbert Xu 305971d93b39SHerbert Xu skb_reserve(nskb, headroom); 306086911732SHerbert Xu __skb_put(nskb, skb_gro_offset(p)); 306171d93b39SHerbert Xu 306286911732SHerbert Xu skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 306371d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 306471d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 306571d93b39SHerbert Xu 306686911732SHerbert Xu __skb_pull(p, skb_gro_offset(p)); 306786911732SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), 306886911732SHerbert Xu p->data - skb_mac_header(p)); 306971d93b39SHerbert Xu 307071d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 30719aaa156cSHerbert Xu skb_shinfo(nskb)->gso_size = pinfo->gso_size; 3072622e0ca1SHerbert Xu pinfo->gso_size = 0; 307371d93b39SHerbert Xu skb_header_release(p); 3074c3c7c254SEric Dumazet NAPI_GRO_CB(nskb)->last = p; 307571d93b39SHerbert Xu 307671d93b39SHerbert Xu nskb->data_len += p->len; 3077de8261c2SEric Dumazet nskb->truesize += p->truesize; 307871d93b39SHerbert Xu nskb->len += p->len; 307971d93b39SHerbert Xu 308071d93b39SHerbert Xu *head = nskb; 308171d93b39SHerbert Xu nskb->next = p->next; 308271d93b39SHerbert Xu p->next = NULL; 308371d93b39SHerbert Xu 308471d93b39SHerbert Xu p = nskb; 308571d93b39SHerbert Xu 308671d93b39SHerbert Xu merge: 3087715dc1f3SEric Dumazet delta_truesize = skb->truesize; 308867147ba9SHerbert Xu if (offset > headlen) { 3089d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 3090d1dc7abfSMichal Schmidt 3091d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 30929e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 3093d1dc7abfSMichal Schmidt skb->data_len -= eat; 3094d1dc7abfSMichal Schmidt skb->len -= eat; 309567147ba9SHerbert Xu offset = headlen; 309656035022SHerbert Xu } 309756035022SHerbert Xu 309867147ba9SHerbert Xu __skb_pull(skb, offset); 309956035022SHerbert Xu 31008a29111cSEric Dumazet if (!NAPI_GRO_CB(p)->last) 31018a29111cSEric Dumazet skb_shinfo(p)->frag_list = skb; 31028a29111cSEric Dumazet else 3103c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last->next = skb; 3104c3c7c254SEric Dumazet NAPI_GRO_CB(p)->last = skb; 310571d93b39SHerbert Xu skb_header_release(skb); 31068a29111cSEric Dumazet lp = p; 310771d93b39SHerbert Xu 31085d38a079SHerbert Xu done: 31095d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 311037fe4732SHerbert Xu p->data_len += len; 3111715dc1f3SEric Dumazet p->truesize += delta_truesize; 311237fe4732SHerbert Xu p->len += len; 31138a29111cSEric Dumazet if (lp != p) { 31148a29111cSEric Dumazet lp->data_len += len; 31158a29111cSEric Dumazet lp->truesize += delta_truesize; 31168a29111cSEric Dumazet lp->len += len; 31178a29111cSEric Dumazet } 311871d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 311971d93b39SHerbert Xu return 0; 312071d93b39SHerbert Xu } 312171d93b39SHerbert Xu EXPORT_SYMBOL_GPL(skb_gro_receive); 312271d93b39SHerbert Xu 31231da177e4SLinus Torvalds void __init skb_init(void) 31241da177e4SLinus Torvalds { 31251da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 31261da177e4SLinus Torvalds sizeof(struct sk_buff), 31271da177e4SLinus Torvalds 0, 3128e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 312920c2df83SPaul Mundt NULL); 3130d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3131d179cd12SDavid S. Miller (2*sizeof(struct sk_buff)) + 3132d179cd12SDavid S. Miller sizeof(atomic_t), 3133d179cd12SDavid S. Miller 0, 3134e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 313520c2df83SPaul Mundt NULL); 31361da177e4SLinus Torvalds } 31371da177e4SLinus Torvalds 3138716ea3a7SDavid Howells /** 3139716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3140716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 3141716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 3142716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 3143716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 3144716ea3a7SDavid Howells * 3145716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 3146716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 3147716ea3a7SDavid Howells */ 314851c739d1SDavid S. Miller static int 314951c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3150716ea3a7SDavid Howells { 31511a028e50SDavid S. Miller int start = skb_headlen(skb); 31521a028e50SDavid S. Miller int i, copy = start - offset; 3153fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3154716ea3a7SDavid Howells int elt = 0; 3155716ea3a7SDavid Howells 3156716ea3a7SDavid Howells if (copy > 0) { 3157716ea3a7SDavid Howells if (copy > len) 3158716ea3a7SDavid Howells copy = len; 3159642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 3160716ea3a7SDavid Howells elt++; 3161716ea3a7SDavid Howells if ((len -= copy) == 0) 3162716ea3a7SDavid Howells return elt; 3163716ea3a7SDavid Howells offset += copy; 3164716ea3a7SDavid Howells } 3165716ea3a7SDavid Howells 3166716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 31671a028e50SDavid S. Miller int end; 3168716ea3a7SDavid Howells 3169547b792cSIlpo Järvinen WARN_ON(start > offset + len); 31701a028e50SDavid S. Miller 31719e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3172716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3173716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3174716ea3a7SDavid Howells 3175716ea3a7SDavid Howells if (copy > len) 3176716ea3a7SDavid Howells copy = len; 3177ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3178642f1490SJens Axboe frag->page_offset+offset-start); 3179716ea3a7SDavid Howells elt++; 3180716ea3a7SDavid Howells if (!(len -= copy)) 3181716ea3a7SDavid Howells return elt; 3182716ea3a7SDavid Howells offset += copy; 3183716ea3a7SDavid Howells } 31841a028e50SDavid S. Miller start = end; 3185716ea3a7SDavid Howells } 3186716ea3a7SDavid Howells 3187fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 31881a028e50SDavid S. Miller int end; 3189716ea3a7SDavid Howells 3190547b792cSIlpo Järvinen WARN_ON(start > offset + len); 31911a028e50SDavid S. Miller 3192fbb398a8SDavid S. Miller end = start + frag_iter->len; 3193716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3194716ea3a7SDavid Howells if (copy > len) 3195716ea3a7SDavid Howells copy = len; 3196fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 319751c739d1SDavid S. Miller copy); 3198716ea3a7SDavid Howells if ((len -= copy) == 0) 3199716ea3a7SDavid Howells return elt; 3200716ea3a7SDavid Howells offset += copy; 3201716ea3a7SDavid Howells } 32021a028e50SDavid S. Miller start = end; 3203716ea3a7SDavid Howells } 3204716ea3a7SDavid Howells BUG_ON(len); 3205716ea3a7SDavid Howells return elt; 3206716ea3a7SDavid Howells } 3207716ea3a7SDavid Howells 320851c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 320951c739d1SDavid S. Miller { 321051c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 321151c739d1SDavid S. Miller 3212c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 321351c739d1SDavid S. Miller 321451c739d1SDavid S. Miller return nsg; 321551c739d1SDavid S. Miller } 3216b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 321751c739d1SDavid S. Miller 3218716ea3a7SDavid Howells /** 3219716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 3220716ea3a7SDavid Howells * @skb: The socket buffer to check. 3221716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 3222716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 3223716ea3a7SDavid Howells * 3224716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 3225716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 3226716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 3227716ea3a7SDavid Howells * 3228716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 3229716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 3230716ea3a7SDavid Howells * set to point to the skb in which this space begins. 3231716ea3a7SDavid Howells * 3232716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3233716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3234716ea3a7SDavid Howells */ 3235716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3236716ea3a7SDavid Howells { 3237716ea3a7SDavid Howells int copyflag; 3238716ea3a7SDavid Howells int elt; 3239716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3240716ea3a7SDavid Howells 3241716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3242716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3243716ea3a7SDavid Howells * at the moment even if they are anonymous). 3244716ea3a7SDavid Howells */ 3245716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3246716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3247716ea3a7SDavid Howells return -ENOMEM; 3248716ea3a7SDavid Howells 3249716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 325021dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3251716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3252716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3253716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3254716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3255716ea3a7SDavid Howells 3256716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3257716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3258716ea3a7SDavid Howells return -ENOMEM; 3259716ea3a7SDavid Howells 3260716ea3a7SDavid Howells /* Voila! */ 3261716ea3a7SDavid Howells *trailer = skb; 3262716ea3a7SDavid Howells return 1; 3263716ea3a7SDavid Howells } 3264716ea3a7SDavid Howells 3265716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3266716ea3a7SDavid Howells 3267716ea3a7SDavid Howells elt = 1; 3268716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3269716ea3a7SDavid Howells copyflag = 0; 3270716ea3a7SDavid Howells 3271716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3272716ea3a7SDavid Howells int ntail = 0; 3273716ea3a7SDavid Howells 3274716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3275716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3276716ea3a7SDavid Howells * after it. */ 3277716ea3a7SDavid Howells 3278716ea3a7SDavid Howells if (skb_shared(skb1)) 3279716ea3a7SDavid Howells copyflag = 1; 3280716ea3a7SDavid Howells 3281716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3282716ea3a7SDavid Howells 3283716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3284716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 328521dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3286716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3287716ea3a7SDavid Howells ntail = tailbits + 128; 3288716ea3a7SDavid Howells } 3289716ea3a7SDavid Howells 3290716ea3a7SDavid Howells if (copyflag || 3291716ea3a7SDavid Howells skb_cloned(skb1) || 3292716ea3a7SDavid Howells ntail || 3293716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 329421dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3295716ea3a7SDavid Howells struct sk_buff *skb2; 3296716ea3a7SDavid Howells 3297716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3298716ea3a7SDavid Howells if (ntail == 0) 3299716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3300716ea3a7SDavid Howells else 3301716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3302716ea3a7SDavid Howells skb_headroom(skb1), 3303716ea3a7SDavid Howells ntail, 3304716ea3a7SDavid Howells GFP_ATOMIC); 3305716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3306716ea3a7SDavid Howells return -ENOMEM; 3307716ea3a7SDavid Howells 3308716ea3a7SDavid Howells if (skb1->sk) 3309716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3310716ea3a7SDavid Howells 3311716ea3a7SDavid Howells /* Looking around. Are we still alive? 3312716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3313716ea3a7SDavid Howells 3314716ea3a7SDavid Howells skb2->next = skb1->next; 3315716ea3a7SDavid Howells *skb_p = skb2; 3316716ea3a7SDavid Howells kfree_skb(skb1); 3317716ea3a7SDavid Howells skb1 = skb2; 3318716ea3a7SDavid Howells } 3319716ea3a7SDavid Howells elt++; 3320716ea3a7SDavid Howells *trailer = skb1; 3321716ea3a7SDavid Howells skb_p = &skb1->next; 3322716ea3a7SDavid Howells } 3323716ea3a7SDavid Howells 3324716ea3a7SDavid Howells return elt; 3325716ea3a7SDavid Howells } 3326b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3327716ea3a7SDavid Howells 3328b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3329b1faf566SEric Dumazet { 3330b1faf566SEric Dumazet struct sock *sk = skb->sk; 3331b1faf566SEric Dumazet 3332b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3333b1faf566SEric Dumazet } 3334b1faf566SEric Dumazet 3335b1faf566SEric Dumazet /* 3336b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3337b1faf566SEric Dumazet */ 3338b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3339b1faf566SEric Dumazet { 3340110c4330SEric Dumazet int len = skb->len; 3341110c4330SEric Dumazet 3342b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 334395c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 3344b1faf566SEric Dumazet return -ENOMEM; 3345b1faf566SEric Dumazet 3346b1faf566SEric Dumazet skb_orphan(skb); 3347b1faf566SEric Dumazet skb->sk = sk; 3348b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3349b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3350b1faf566SEric Dumazet 3351abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3352abb57ea4SEric Dumazet skb_dst_force(skb); 3353abb57ea4SEric Dumazet 3354b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3355b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3356110c4330SEric Dumazet sk->sk_data_ready(sk, len); 3357b1faf566SEric Dumazet return 0; 3358b1faf566SEric Dumazet } 3359b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3360b1faf566SEric Dumazet 3361ac45f602SPatrick Ohly void skb_tstamp_tx(struct sk_buff *orig_skb, 3362ac45f602SPatrick Ohly struct skb_shared_hwtstamps *hwtstamps) 3363ac45f602SPatrick Ohly { 3364ac45f602SPatrick Ohly struct sock *sk = orig_skb->sk; 3365ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3366ac45f602SPatrick Ohly struct sk_buff *skb; 3367ac45f602SPatrick Ohly int err; 3368ac45f602SPatrick Ohly 3369ac45f602SPatrick Ohly if (!sk) 3370ac45f602SPatrick Ohly return; 3371ac45f602SPatrick Ohly 3372ac45f602SPatrick Ohly if (hwtstamps) { 33732e31396fSWillem de Bruijn *skb_hwtstamps(orig_skb) = 3374ac45f602SPatrick Ohly *hwtstamps; 3375ac45f602SPatrick Ohly } else { 3376ac45f602SPatrick Ohly /* 3377ac45f602SPatrick Ohly * no hardware time stamps available, 33782244d07bSOliver Hartkopp * so keep the shared tx_flags and only 3379ac45f602SPatrick Ohly * store software time stamp 3380ac45f602SPatrick Ohly */ 33812e31396fSWillem de Bruijn orig_skb->tstamp = ktime_get_real(); 3382ac45f602SPatrick Ohly } 3383ac45f602SPatrick Ohly 33842e31396fSWillem de Bruijn skb = skb_clone(orig_skb, GFP_ATOMIC); 33852e31396fSWillem de Bruijn if (!skb) 33862e31396fSWillem de Bruijn return; 33872e31396fSWillem de Bruijn 3388ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3389ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3390ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3391ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 339229030374SEric Dumazet 3393ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 339429030374SEric Dumazet 3395ac45f602SPatrick Ohly if (err) 3396ac45f602SPatrick Ohly kfree_skb(skb); 3397ac45f602SPatrick Ohly } 3398ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3399ac45f602SPatrick Ohly 34006e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 34016e3e939fSJohannes Berg { 34026e3e939fSJohannes Berg struct sock *sk = skb->sk; 34036e3e939fSJohannes Berg struct sock_exterr_skb *serr; 34046e3e939fSJohannes Berg int err; 34056e3e939fSJohannes Berg 34066e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 34076e3e939fSJohannes Berg skb->wifi_acked = acked; 34086e3e939fSJohannes Berg 34096e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 34106e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 34116e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 34126e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 34136e3e939fSJohannes Berg 34146e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 34156e3e939fSJohannes Berg if (err) 34166e3e939fSJohannes Berg kfree_skb(skb); 34176e3e939fSJohannes Berg } 34186e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 34196e3e939fSJohannes Berg 3420ac45f602SPatrick Ohly 3421f35d9d8aSRusty Russell /** 3422f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3423f35d9d8aSRusty Russell * @skb: the skb to set 3424f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3425f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3426f35d9d8aSRusty Russell * 3427f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3428f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3429f35d9d8aSRusty Russell * 3430f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3431f35d9d8aSRusty Russell * returns false you should drop the packet. 3432f35d9d8aSRusty Russell */ 3433f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3434f35d9d8aSRusty Russell { 34355ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 34365ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3437e87cc472SJoe Perches net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 34385ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3439f35d9d8aSRusty Russell return false; 3440f35d9d8aSRusty Russell } 3441f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3442f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3443f35d9d8aSRusty Russell skb->csum_offset = off; 3444e5d5decaSJason Wang skb_set_transport_header(skb, start); 3445f35d9d8aSRusty Russell return true; 3446f35d9d8aSRusty Russell } 3447b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3448f35d9d8aSRusty Russell 34494497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 34504497b076SBen Hutchings { 3451e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 3452e87cc472SJoe Perches skb->dev->name); 34534497b076SBen Hutchings } 34544497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 3455bad43ca8SEric Dumazet 3456bad43ca8SEric Dumazet void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 3457bad43ca8SEric Dumazet { 34583d861f66SEric Dumazet if (head_stolen) { 34593d861f66SEric Dumazet skb_release_head_state(skb); 3460bad43ca8SEric Dumazet kmem_cache_free(skbuff_head_cache, skb); 34613d861f66SEric Dumazet } else { 3462bad43ca8SEric Dumazet __kfree_skb(skb); 3463bad43ca8SEric Dumazet } 34643d861f66SEric Dumazet } 3465bad43ca8SEric Dumazet EXPORT_SYMBOL(kfree_skb_partial); 3466bad43ca8SEric Dumazet 3467bad43ca8SEric Dumazet /** 3468bad43ca8SEric Dumazet * skb_try_coalesce - try to merge skb to prior one 3469bad43ca8SEric Dumazet * @to: prior buffer 3470bad43ca8SEric Dumazet * @from: buffer to add 3471bad43ca8SEric Dumazet * @fragstolen: pointer to boolean 3472c6c4b97cSRandy Dunlap * @delta_truesize: how much more was allocated than was requested 3473bad43ca8SEric Dumazet */ 3474bad43ca8SEric Dumazet bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 3475bad43ca8SEric Dumazet bool *fragstolen, int *delta_truesize) 3476bad43ca8SEric Dumazet { 3477bad43ca8SEric Dumazet int i, delta, len = from->len; 3478bad43ca8SEric Dumazet 3479bad43ca8SEric Dumazet *fragstolen = false; 3480bad43ca8SEric Dumazet 3481bad43ca8SEric Dumazet if (skb_cloned(to)) 3482bad43ca8SEric Dumazet return false; 3483bad43ca8SEric Dumazet 3484bad43ca8SEric Dumazet if (len <= skb_tailroom(to)) { 3485bad43ca8SEric Dumazet BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 3486bad43ca8SEric Dumazet *delta_truesize = 0; 3487bad43ca8SEric Dumazet return true; 3488bad43ca8SEric Dumazet } 3489bad43ca8SEric Dumazet 3490bad43ca8SEric Dumazet if (skb_has_frag_list(to) || skb_has_frag_list(from)) 3491bad43ca8SEric Dumazet return false; 3492bad43ca8SEric Dumazet 3493bad43ca8SEric Dumazet if (skb_headlen(from) != 0) { 3494bad43ca8SEric Dumazet struct page *page; 3495bad43ca8SEric Dumazet unsigned int offset; 3496bad43ca8SEric Dumazet 3497bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 3498bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 3499bad43ca8SEric Dumazet return false; 3500bad43ca8SEric Dumazet 3501bad43ca8SEric Dumazet if (skb_head_is_locked(from)) 3502bad43ca8SEric Dumazet return false; 3503bad43ca8SEric Dumazet 3504bad43ca8SEric Dumazet delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3505bad43ca8SEric Dumazet 3506bad43ca8SEric Dumazet page = virt_to_head_page(from->head); 3507bad43ca8SEric Dumazet offset = from->data - (unsigned char *)page_address(page); 3508bad43ca8SEric Dumazet 3509bad43ca8SEric Dumazet skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 3510bad43ca8SEric Dumazet page, offset, skb_headlen(from)); 3511bad43ca8SEric Dumazet *fragstolen = true; 3512bad43ca8SEric Dumazet } else { 3513bad43ca8SEric Dumazet if (skb_shinfo(to)->nr_frags + 3514bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 3515bad43ca8SEric Dumazet return false; 3516bad43ca8SEric Dumazet 3517f4b549a5SWeiping Pan delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 3518bad43ca8SEric Dumazet } 3519bad43ca8SEric Dumazet 3520bad43ca8SEric Dumazet WARN_ON_ONCE(delta < len); 3521bad43ca8SEric Dumazet 3522bad43ca8SEric Dumazet memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 3523bad43ca8SEric Dumazet skb_shinfo(from)->frags, 3524bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 3525bad43ca8SEric Dumazet skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 3526bad43ca8SEric Dumazet 3527bad43ca8SEric Dumazet if (!skb_cloned(from)) 3528bad43ca8SEric Dumazet skb_shinfo(from)->nr_frags = 0; 3529bad43ca8SEric Dumazet 35308ea853fdSLi RongQing /* if the skb is not cloned this does nothing 35318ea853fdSLi RongQing * since we set nr_frags to 0. 35328ea853fdSLi RongQing */ 3533bad43ca8SEric Dumazet for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 3534bad43ca8SEric Dumazet skb_frag_ref(from, i); 3535bad43ca8SEric Dumazet 3536bad43ca8SEric Dumazet to->truesize += delta; 3537bad43ca8SEric Dumazet to->len += len; 3538bad43ca8SEric Dumazet to->data_len += len; 3539bad43ca8SEric Dumazet 3540bad43ca8SEric Dumazet *delta_truesize = delta; 3541bad43ca8SEric Dumazet return true; 3542bad43ca8SEric Dumazet } 3543bad43ca8SEric Dumazet EXPORT_SYMBOL(skb_try_coalesce); 3544621e84d6SNicolas Dichtel 3545621e84d6SNicolas Dichtel /** 35468b27f277SNicolas Dichtel * skb_scrub_packet - scrub an skb 3547621e84d6SNicolas Dichtel * 3548621e84d6SNicolas Dichtel * @skb: buffer to clean 35498b27f277SNicolas Dichtel * @xnet: packet is crossing netns 3550621e84d6SNicolas Dichtel * 35518b27f277SNicolas Dichtel * skb_scrub_packet can be used after encapsulating or decapsulting a packet 35528b27f277SNicolas Dichtel * into/from a tunnel. Some information have to be cleared during these 35538b27f277SNicolas Dichtel * operations. 35548b27f277SNicolas Dichtel * skb_scrub_packet can also be used to clean a skb before injecting it in 35558b27f277SNicolas Dichtel * another namespace (@xnet == true). We have to clear all information in the 35568b27f277SNicolas Dichtel * skb that could impact namespace isolation. 3557621e84d6SNicolas Dichtel */ 35588b27f277SNicolas Dichtel void skb_scrub_packet(struct sk_buff *skb, bool xnet) 3559621e84d6SNicolas Dichtel { 35608b27f277SNicolas Dichtel if (xnet) 3561621e84d6SNicolas Dichtel skb_orphan(skb); 3562621e84d6SNicolas Dichtel skb->tstamp.tv64 = 0; 3563621e84d6SNicolas Dichtel skb->pkt_type = PACKET_HOST; 3564621e84d6SNicolas Dichtel skb->skb_iif = 0; 3565621e84d6SNicolas Dichtel skb_dst_drop(skb); 3566621e84d6SNicolas Dichtel skb->mark = 0; 3567621e84d6SNicolas Dichtel secpath_reset(skb); 3568621e84d6SNicolas Dichtel nf_reset(skb); 3569621e84d6SNicolas Dichtel nf_reset_trace(skb); 3570621e84d6SNicolas Dichtel } 3571621e84d6SNicolas Dichtel EXPORT_SYMBOL_GPL(skb_scrub_packet); 3572