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 391da177e4SLinus Torvalds #include <linux/module.h> 401da177e4SLinus Torvalds #include <linux/types.h> 411da177e4SLinus Torvalds #include <linux/kernel.h> 42fe55f6d5SVegard Nossum #include <linux/kmemcheck.h> 431da177e4SLinus Torvalds #include <linux/mm.h> 441da177e4SLinus Torvalds #include <linux/interrupt.h> 451da177e4SLinus Torvalds #include <linux/in.h> 461da177e4SLinus Torvalds #include <linux/inet.h> 471da177e4SLinus Torvalds #include <linux/slab.h> 481da177e4SLinus Torvalds #include <linux/netdevice.h> 491da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 501da177e4SLinus Torvalds #include <net/pkt_sched.h> 511da177e4SLinus Torvalds #endif 521da177e4SLinus Torvalds #include <linux/string.h> 531da177e4SLinus Torvalds #include <linux/skbuff.h> 549c55e01cSJens Axboe #include <linux/splice.h> 551da177e4SLinus Torvalds #include <linux/cache.h> 561da177e4SLinus Torvalds #include <linux/rtnetlink.h> 571da177e4SLinus Torvalds #include <linux/init.h> 58716ea3a7SDavid Howells #include <linux/scatterlist.h> 59ac45f602SPatrick Ohly #include <linux/errqueue.h> 60268bb0ceSLinus Torvalds #include <linux/prefetch.h> 611da177e4SLinus Torvalds 621da177e4SLinus Torvalds #include <net/protocol.h> 631da177e4SLinus Torvalds #include <net/dst.h> 641da177e4SLinus Torvalds #include <net/sock.h> 651da177e4SLinus Torvalds #include <net/checksum.h> 661da177e4SLinus Torvalds #include <net/xfrm.h> 671da177e4SLinus Torvalds 681da177e4SLinus Torvalds #include <asm/uaccess.h> 69ad8d75ffSSteven Rostedt #include <trace/events/skb.h> 7051c56b00SEric Dumazet #include <linux/highmem.h> 71a1f8e7f7SAl Viro 72d7e8883cSEric Dumazet struct kmem_cache *skbuff_head_cache __read_mostly; 73e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 741da177e4SLinus Torvalds 759c55e01cSJens Axboe static void sock_pipe_buf_release(struct pipe_inode_info *pipe, 769c55e01cSJens Axboe struct pipe_buffer *buf) 779c55e01cSJens Axboe { 788b9d3728SJarek Poplawski put_page(buf->page); 799c55e01cSJens Axboe } 809c55e01cSJens Axboe 819c55e01cSJens Axboe static void sock_pipe_buf_get(struct pipe_inode_info *pipe, 829c55e01cSJens Axboe struct pipe_buffer *buf) 839c55e01cSJens Axboe { 848b9d3728SJarek Poplawski get_page(buf->page); 859c55e01cSJens Axboe } 869c55e01cSJens Axboe 879c55e01cSJens Axboe static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, 889c55e01cSJens Axboe struct pipe_buffer *buf) 899c55e01cSJens Axboe { 909c55e01cSJens Axboe return 1; 919c55e01cSJens Axboe } 929c55e01cSJens Axboe 939c55e01cSJens Axboe 949c55e01cSJens Axboe /* Pipe buffer operations for a socket. */ 9528dfef8fSAlexey Dobriyan static const struct pipe_buf_operations sock_pipe_buf_ops = { 969c55e01cSJens Axboe .can_merge = 0, 979c55e01cSJens Axboe .map = generic_pipe_buf_map, 989c55e01cSJens Axboe .unmap = generic_pipe_buf_unmap, 999c55e01cSJens Axboe .confirm = generic_pipe_buf_confirm, 1009c55e01cSJens Axboe .release = sock_pipe_buf_release, 1019c55e01cSJens Axboe .steal = sock_pipe_buf_steal, 1029c55e01cSJens Axboe .get = sock_pipe_buf_get, 1039c55e01cSJens Axboe }; 1049c55e01cSJens Axboe 1051da177e4SLinus Torvalds /* 1061da177e4SLinus Torvalds * Keep out-of-line to prevent kernel bloat. 1071da177e4SLinus Torvalds * __builtin_return_address is not used because it is not always 1081da177e4SLinus Torvalds * reliable. 1091da177e4SLinus Torvalds */ 1101da177e4SLinus Torvalds 1111da177e4SLinus Torvalds /** 1121da177e4SLinus Torvalds * skb_over_panic - private function 1131da177e4SLinus Torvalds * @skb: buffer 1141da177e4SLinus Torvalds * @sz: size 1151da177e4SLinus Torvalds * @here: address 1161da177e4SLinus Torvalds * 1171da177e4SLinus Torvalds * Out of line support code for skb_put(). Not user callable. 1181da177e4SLinus Torvalds */ 119ccb7c773SRami Rosen static void skb_over_panic(struct sk_buff *skb, int sz, void *here) 1201da177e4SLinus Torvalds { 12126095455SPatrick McHardy printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " 1224305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 12327a884dcSArnaldo Carvalho de Melo here, 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 1291da177e4SLinus Torvalds /** 1301da177e4SLinus Torvalds * skb_under_panic - private function 1311da177e4SLinus Torvalds * @skb: buffer 1321da177e4SLinus Torvalds * @sz: size 1331da177e4SLinus Torvalds * @here: address 1341da177e4SLinus Torvalds * 1351da177e4SLinus Torvalds * Out of line support code for skb_push(). Not user callable. 1361da177e4SLinus Torvalds */ 1371da177e4SLinus Torvalds 138ccb7c773SRami Rosen static void skb_under_panic(struct sk_buff *skb, int sz, void *here) 1391da177e4SLinus Torvalds { 14026095455SPatrick McHardy printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " 1414305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 14227a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1434305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 14426095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1451da177e4SLinus Torvalds BUG(); 1461da177e4SLinus Torvalds } 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1491da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1501da177e4SLinus Torvalds * [BEEP] leaks. 1511da177e4SLinus Torvalds * 1521da177e4SLinus Torvalds */ 1531da177e4SLinus Torvalds 1541da177e4SLinus Torvalds /** 155d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1561da177e4SLinus Torvalds * @size: size to allocate 1571da177e4SLinus Torvalds * @gfp_mask: allocation mask 158c83c2486SRandy Dunlap * @fclone: allocate from fclone cache instead of head cache 159c83c2486SRandy Dunlap * and allocate a cloned (child) skb 160b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1611da177e4SLinus Torvalds * 1621da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 1631da177e4SLinus Torvalds * tail room of size bytes. The object has a reference count of one. 1641da177e4SLinus Torvalds * The return is the buffer. On a failure the return is %NULL. 1651da177e4SLinus Torvalds * 1661da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1671da177e4SLinus Torvalds * %GFP_ATOMIC. 1681da177e4SLinus Torvalds */ 169dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 170b30973f8SChristoph Hellwig int fclone, int node) 1711da177e4SLinus Torvalds { 172e18b890bSChristoph Lameter struct kmem_cache *cache; 1734947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1741da177e4SLinus Torvalds struct sk_buff *skb; 1751da177e4SLinus Torvalds u8 *data; 1761da177e4SLinus Torvalds 1778798b3fbSHerbert Xu cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; 1788798b3fbSHerbert Xu 1791da177e4SLinus Torvalds /* Get the HEAD */ 180b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1811da177e4SLinus Torvalds if (!skb) 1821da177e4SLinus Torvalds goto out; 183ec7d2f2cSEric Dumazet prefetchw(skb); 1841da177e4SLinus Torvalds 18587fb4b7bSEric Dumazet /* We do our best to align skb_shared_info on a separate cache 18687fb4b7bSEric Dumazet * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 18787fb4b7bSEric Dumazet * aligned memory blocks, unless SLUB/SLAB debug is enabled. 18887fb4b7bSEric Dumazet * Both skb->head and skb_shared_info are cache line aligned. 18987fb4b7bSEric Dumazet */ 190bc417e30STony Lindgren size = SKB_DATA_ALIGN(size); 19187fb4b7bSEric Dumazet size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 19287fb4b7bSEric Dumazet data = kmalloc_node_track_caller(size, gfp_mask, node); 1931da177e4SLinus Torvalds if (!data) 1941da177e4SLinus Torvalds goto nodata; 19587fb4b7bSEric Dumazet /* kmalloc(size) might give us more room than requested. 19687fb4b7bSEric Dumazet * Put skb_shared_info exactly at the end of allocated zone, 19787fb4b7bSEric Dumazet * to allow max possible filling before reallocation. 19887fb4b7bSEric Dumazet */ 19987fb4b7bSEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 200ec7d2f2cSEric Dumazet prefetchw(data + size); 2011da177e4SLinus Torvalds 202ca0605a7SArnaldo Carvalho de Melo /* 203c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 204c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 205c8005785SJohannes Berg * the tail pointer in struct sk_buff! 206ca0605a7SArnaldo Carvalho de Melo */ 207ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 20887fb4b7bSEric Dumazet /* Account for allocated memory : skb + skb->head */ 20987fb4b7bSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 2101da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2111da177e4SLinus Torvalds skb->head = data; 2121da177e4SLinus Torvalds skb->data = data; 21327a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2144305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 21519633e12SStephen Hemminger #ifdef NET_SKBUFF_DATA_USES_OFFSET 21619633e12SStephen Hemminger skb->mac_header = ~0U; 21719633e12SStephen Hemminger #endif 21819633e12SStephen Hemminger 2194947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2204947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 221ec7d2f2cSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 2224947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 223c2aa3665SEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 2244947d3efSBenjamin LaHaise 225d179cd12SDavid S. Miller if (fclone) { 226d179cd12SDavid S. Miller struct sk_buff *child = skb + 1; 227d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (child + 1); 2281da177e4SLinus Torvalds 229fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags1); 230fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(child, flags2); 231d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 232d179cd12SDavid S. Miller atomic_set(fclone_ref, 1); 233d179cd12SDavid S. Miller 234d179cd12SDavid S. Miller child->fclone = SKB_FCLONE_UNAVAILABLE; 235d179cd12SDavid S. Miller } 2361da177e4SLinus Torvalds out: 2371da177e4SLinus Torvalds return skb; 2381da177e4SLinus Torvalds nodata: 2398798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2401da177e4SLinus Torvalds skb = NULL; 2411da177e4SLinus Torvalds goto out; 2421da177e4SLinus Torvalds } 243b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2441da177e4SLinus Torvalds 2451da177e4SLinus Torvalds /** 246b2b5ce9dSEric Dumazet * build_skb - build a network buffer 247b2b5ce9dSEric Dumazet * @data: data buffer provided by caller 248d3836f21SEric Dumazet * @frag_size: size of fragment, or 0 if head was kmalloced 249b2b5ce9dSEric Dumazet * 250b2b5ce9dSEric Dumazet * Allocate a new &sk_buff. Caller provides space holding head and 251b2b5ce9dSEric Dumazet * skb_shared_info. @data must have been allocated by kmalloc() 252b2b5ce9dSEric Dumazet * The return is the new skb buffer. 253b2b5ce9dSEric Dumazet * On a failure the return is %NULL, and @data is not freed. 254b2b5ce9dSEric Dumazet * Notes : 255b2b5ce9dSEric Dumazet * Before IO, driver allocates only data buffer where NIC put incoming frame 256b2b5ce9dSEric Dumazet * Driver should add room at head (NET_SKB_PAD) and 257b2b5ce9dSEric Dumazet * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 258b2b5ce9dSEric Dumazet * After IO, driver calls build_skb(), to allocate sk_buff and populate it 259b2b5ce9dSEric Dumazet * before giving packet to stack. 260b2b5ce9dSEric Dumazet * RX rings only contains data buffers, not full skbs. 261b2b5ce9dSEric Dumazet */ 262d3836f21SEric Dumazet struct sk_buff *build_skb(void *data, unsigned int frag_size) 263b2b5ce9dSEric Dumazet { 264b2b5ce9dSEric Dumazet struct skb_shared_info *shinfo; 265b2b5ce9dSEric Dumazet struct sk_buff *skb; 266d3836f21SEric Dumazet unsigned int size = frag_size ? : ksize(data); 267b2b5ce9dSEric Dumazet 268b2b5ce9dSEric Dumazet skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 269b2b5ce9dSEric Dumazet if (!skb) 270b2b5ce9dSEric Dumazet return NULL; 271b2b5ce9dSEric Dumazet 272d3836f21SEric Dumazet size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 273b2b5ce9dSEric Dumazet 274b2b5ce9dSEric Dumazet memset(skb, 0, offsetof(struct sk_buff, tail)); 275b2b5ce9dSEric Dumazet skb->truesize = SKB_TRUESIZE(size); 276d3836f21SEric Dumazet skb->head_frag = frag_size != 0; 277b2b5ce9dSEric Dumazet atomic_set(&skb->users, 1); 278b2b5ce9dSEric Dumazet skb->head = data; 279b2b5ce9dSEric Dumazet skb->data = data; 280b2b5ce9dSEric Dumazet skb_reset_tail_pointer(skb); 281b2b5ce9dSEric Dumazet skb->end = skb->tail + size; 282b2b5ce9dSEric Dumazet #ifdef NET_SKBUFF_DATA_USES_OFFSET 283b2b5ce9dSEric Dumazet skb->mac_header = ~0U; 284b2b5ce9dSEric Dumazet #endif 285b2b5ce9dSEric Dumazet 286b2b5ce9dSEric Dumazet /* make sure we initialize shinfo sequentially */ 287b2b5ce9dSEric Dumazet shinfo = skb_shinfo(skb); 288b2b5ce9dSEric Dumazet memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 289b2b5ce9dSEric Dumazet atomic_set(&shinfo->dataref, 1); 290b2b5ce9dSEric Dumazet kmemcheck_annotate_variable(shinfo->destructor_arg); 291b2b5ce9dSEric Dumazet 292b2b5ce9dSEric Dumazet return skb; 293b2b5ce9dSEric Dumazet } 294b2b5ce9dSEric Dumazet EXPORT_SYMBOL(build_skb); 295b2b5ce9dSEric Dumazet 296b2b5ce9dSEric Dumazet /** 2978af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 2988af27456SChristoph Hellwig * @dev: network device to receive on 2998af27456SChristoph Hellwig * @length: length to allocate 3008af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 3018af27456SChristoph Hellwig * 3028af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 3038af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 3048af27456SChristoph Hellwig * the headroom they think they need without accounting for the 3058af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 3068af27456SChristoph Hellwig * 3078af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 3088af27456SChristoph Hellwig */ 3098af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 3108af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 3118af27456SChristoph Hellwig { 3128af27456SChristoph Hellwig struct sk_buff *skb; 3138af27456SChristoph Hellwig 314564824b0SEric Dumazet skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE); 3157b2e497aSChristoph Hellwig if (likely(skb)) { 3168af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 3177b2e497aSChristoph Hellwig skb->dev = dev; 3187b2e497aSChristoph Hellwig } 3198af27456SChristoph Hellwig return skb; 3208af27456SChristoph Hellwig } 321b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 3221da177e4SLinus Torvalds 323654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 32450269e19SEric Dumazet int size, unsigned int truesize) 325654bed16SPeter Zijlstra { 326654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 327654bed16SPeter Zijlstra skb->len += size; 328654bed16SPeter Zijlstra skb->data_len += size; 32950269e19SEric Dumazet skb->truesize += truesize; 330654bed16SPeter Zijlstra } 331654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 332654bed16SPeter Zijlstra 333f58518e6SIlpo Järvinen /** 334f58518e6SIlpo Järvinen * dev_alloc_skb - allocate an skbuff for receiving 335f58518e6SIlpo Järvinen * @length: length to allocate 336f58518e6SIlpo Järvinen * 337f58518e6SIlpo Järvinen * Allocate a new &sk_buff and assign it a usage count of one. The 338f58518e6SIlpo Järvinen * buffer has unspecified headroom built in. Users should allocate 339f58518e6SIlpo Järvinen * the headroom they think they need without accounting for the 340f58518e6SIlpo Järvinen * built in space. The built in space is used for optimisations. 341f58518e6SIlpo Järvinen * 342f58518e6SIlpo Järvinen * %NULL is returned if there is no free memory. Although this function 343f58518e6SIlpo Järvinen * allocates memory it can be called from an interrupt. 344f58518e6SIlpo Järvinen */ 345f58518e6SIlpo Järvinen struct sk_buff *dev_alloc_skb(unsigned int length) 346f58518e6SIlpo Järvinen { 3471483b874SDenys Vlasenko /* 3481483b874SDenys Vlasenko * There is more code here than it seems: 349a0f55e0eSDavid S. Miller * __dev_alloc_skb is an inline 3501483b874SDenys Vlasenko */ 351f58518e6SIlpo Järvinen return __dev_alloc_skb(length, GFP_ATOMIC); 352f58518e6SIlpo Järvinen } 353f58518e6SIlpo Järvinen EXPORT_SYMBOL(dev_alloc_skb); 354f58518e6SIlpo Järvinen 35527b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 3561da177e4SLinus Torvalds { 35727b437c8SHerbert Xu struct sk_buff *list = *listp; 3581da177e4SLinus Torvalds 35927b437c8SHerbert Xu *listp = NULL; 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds do { 3621da177e4SLinus Torvalds struct sk_buff *this = list; 3631da177e4SLinus Torvalds list = list->next; 3641da177e4SLinus Torvalds kfree_skb(this); 3651da177e4SLinus Torvalds } while (list); 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds 36827b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 36927b437c8SHerbert Xu { 37027b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 37127b437c8SHerbert Xu } 37227b437c8SHerbert Xu 3731da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 3741da177e4SLinus Torvalds { 3751da177e4SLinus Torvalds struct sk_buff *list; 3761da177e4SLinus Torvalds 377fbb398a8SDavid S. Miller skb_walk_frags(skb, list) 3781da177e4SLinus Torvalds skb_get(list); 3791da177e4SLinus Torvalds } 3801da177e4SLinus Torvalds 381d3836f21SEric Dumazet static void skb_free_head(struct sk_buff *skb) 382d3836f21SEric Dumazet { 383d3836f21SEric Dumazet if (skb->head_frag) 384d3836f21SEric Dumazet put_page(virt_to_head_page(skb->head)); 385d3836f21SEric Dumazet else 386d3836f21SEric Dumazet kfree(skb->head); 387d3836f21SEric Dumazet } 388d3836f21SEric Dumazet 3895bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 3901da177e4SLinus Torvalds { 3911da177e4SLinus Torvalds if (!skb->cloned || 3921da177e4SLinus Torvalds !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 3931da177e4SLinus Torvalds &skb_shinfo(skb)->dataref)) { 3941da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 3951da177e4SLinus Torvalds int i; 3961da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 397ea2ab693SIan Campbell skb_frag_unref(skb, i); 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds 400a6686f2fSShirley Ma /* 401a6686f2fSShirley Ma * If skb buf is from userspace, we need to notify the caller 402a6686f2fSShirley Ma * the lower device DMA has done; 403a6686f2fSShirley Ma */ 404a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 405a6686f2fSShirley Ma struct ubuf_info *uarg; 406a6686f2fSShirley Ma 407a6686f2fSShirley Ma uarg = skb_shinfo(skb)->destructor_arg; 408a6686f2fSShirley Ma if (uarg->callback) 409a6686f2fSShirley Ma uarg->callback(uarg); 410a6686f2fSShirley Ma } 411a6686f2fSShirley Ma 41221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 4131da177e4SLinus Torvalds skb_drop_fraglist(skb); 4141da177e4SLinus Torvalds 415d3836f21SEric Dumazet skb_free_head(skb); 4161da177e4SLinus Torvalds } 4171da177e4SLinus Torvalds } 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds /* 4201da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 4211da177e4SLinus Torvalds */ 4222d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 4231da177e4SLinus Torvalds { 424d179cd12SDavid S. Miller struct sk_buff *other; 425d179cd12SDavid S. Miller atomic_t *fclone_ref; 426d179cd12SDavid S. Miller 427d179cd12SDavid S. Miller switch (skb->fclone) { 428d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 4291da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 430d179cd12SDavid S. Miller break; 431d179cd12SDavid S. Miller 432d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 433d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 2); 434d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 435d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, skb); 436d179cd12SDavid S. Miller break; 437d179cd12SDavid S. Miller 438d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 439d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 1); 440d179cd12SDavid S. Miller other = skb - 1; 441d179cd12SDavid S. Miller 442d179cd12SDavid S. Miller /* The clone portion is available for 443d179cd12SDavid S. Miller * fast-cloning again. 444d179cd12SDavid S. Miller */ 445d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_UNAVAILABLE; 446d179cd12SDavid S. Miller 447d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 448d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, other); 449d179cd12SDavid S. Miller break; 4503ff50b79SStephen Hemminger } 4511da177e4SLinus Torvalds } 4521da177e4SLinus Torvalds 45304a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 4541da177e4SLinus Torvalds { 455adf30907SEric Dumazet skb_dst_drop(skb); 4561da177e4SLinus Torvalds #ifdef CONFIG_XFRM 4571da177e4SLinus Torvalds secpath_put(skb->sp); 4581da177e4SLinus Torvalds #endif 4591da177e4SLinus Torvalds if (skb->destructor) { 4609c2b3328SStephen Hemminger WARN_ON(in_irq()); 4611da177e4SLinus Torvalds skb->destructor(skb); 4621da177e4SLinus Torvalds } 463a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4645f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 4652fc72c7bSKOVACS Krisztian #endif 4662fc72c7bSKOVACS Krisztian #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 4679fb9cbb1SYasuyuki Kozakai nf_conntrack_put_reasm(skb->nfct_reasm); 4689fb9cbb1SYasuyuki Kozakai #endif 4691da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 4701da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 4711da177e4SLinus Torvalds #endif 4721da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 4731da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 4741da177e4SLinus Torvalds skb->tc_index = 0; 4751da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 4761da177e4SLinus Torvalds skb->tc_verd = 0; 4771da177e4SLinus Torvalds #endif 4781da177e4SLinus Torvalds #endif 47904a4bb55SLennert Buytenhek } 48004a4bb55SLennert Buytenhek 48104a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 48204a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 48304a4bb55SLennert Buytenhek { 48404a4bb55SLennert Buytenhek skb_release_head_state(skb); 4852d4baff8SHerbert Xu skb_release_data(skb); 4862d4baff8SHerbert Xu } 4871da177e4SLinus Torvalds 4882d4baff8SHerbert Xu /** 4892d4baff8SHerbert Xu * __kfree_skb - private function 4902d4baff8SHerbert Xu * @skb: buffer 4912d4baff8SHerbert Xu * 4922d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 4932d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 4942d4baff8SHerbert Xu * always call kfree_skb 4952d4baff8SHerbert Xu */ 4962d4baff8SHerbert Xu 4972d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 4982d4baff8SHerbert Xu { 4992d4baff8SHerbert Xu skb_release_all(skb); 5001da177e4SLinus Torvalds kfree_skbmem(skb); 5011da177e4SLinus Torvalds } 502b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 5031da177e4SLinus Torvalds 5041da177e4SLinus Torvalds /** 505231d06aeSJörn Engel * kfree_skb - free an sk_buff 506231d06aeSJörn Engel * @skb: buffer to free 507231d06aeSJörn Engel * 508231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 509231d06aeSJörn Engel * hit zero. 510231d06aeSJörn Engel */ 511231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 512231d06aeSJörn Engel { 513231d06aeSJörn Engel if (unlikely(!skb)) 514231d06aeSJörn Engel return; 515231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 516231d06aeSJörn Engel smp_rmb(); 517231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 518231d06aeSJörn Engel return; 519ead2ceb0SNeil Horman trace_kfree_skb(skb, __builtin_return_address(0)); 520231d06aeSJörn Engel __kfree_skb(skb); 521231d06aeSJörn Engel } 522b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 523231d06aeSJörn Engel 524d1a203eaSStephen Hemminger /** 525ead2ceb0SNeil Horman * consume_skb - free an skbuff 526ead2ceb0SNeil Horman * @skb: buffer to free 527ead2ceb0SNeil Horman * 528ead2ceb0SNeil Horman * Drop a ref to the buffer and free it if the usage count has hit zero 529ead2ceb0SNeil Horman * Functions identically to kfree_skb, but kfree_skb assumes that the frame 530ead2ceb0SNeil Horman * is being dropped after a failure and notes that 531ead2ceb0SNeil Horman */ 532ead2ceb0SNeil Horman void consume_skb(struct sk_buff *skb) 533ead2ceb0SNeil Horman { 534ead2ceb0SNeil Horman if (unlikely(!skb)) 535ead2ceb0SNeil Horman return; 536ead2ceb0SNeil Horman if (likely(atomic_read(&skb->users) == 1)) 537ead2ceb0SNeil Horman smp_rmb(); 538ead2ceb0SNeil Horman else if (likely(!atomic_dec_and_test(&skb->users))) 539ead2ceb0SNeil Horman return; 54007dc22e7SKoki Sanagi trace_consume_skb(skb); 541ead2ceb0SNeil Horman __kfree_skb(skb); 542ead2ceb0SNeil Horman } 543ead2ceb0SNeil Horman EXPORT_SYMBOL(consume_skb); 544ead2ceb0SNeil Horman 545ead2ceb0SNeil Horman /** 5463d153a7cSAndy Fleming * skb_recycle - clean up an skb for reuse 5473d153a7cSAndy Fleming * @skb: buffer 5483d153a7cSAndy Fleming * 5493d153a7cSAndy Fleming * Recycles the skb to be reused as a receive buffer. This 5503d153a7cSAndy Fleming * function does any necessary reference count dropping, and 5513d153a7cSAndy Fleming * cleans up the skbuff as if it just came from __alloc_skb(). 5523d153a7cSAndy Fleming */ 5533d153a7cSAndy Fleming void skb_recycle(struct sk_buff *skb) 5543d153a7cSAndy Fleming { 5553d153a7cSAndy Fleming struct skb_shared_info *shinfo; 5563d153a7cSAndy Fleming 5573d153a7cSAndy Fleming skb_release_head_state(skb); 5583d153a7cSAndy Fleming 5593d153a7cSAndy Fleming shinfo = skb_shinfo(skb); 5603d153a7cSAndy Fleming memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 5613d153a7cSAndy Fleming atomic_set(&shinfo->dataref, 1); 5623d153a7cSAndy Fleming 5633d153a7cSAndy Fleming memset(skb, 0, offsetof(struct sk_buff, tail)); 5643d153a7cSAndy Fleming skb->data = skb->head + NET_SKB_PAD; 5653d153a7cSAndy Fleming skb_reset_tail_pointer(skb); 5663d153a7cSAndy Fleming } 5673d153a7cSAndy Fleming EXPORT_SYMBOL(skb_recycle); 5683d153a7cSAndy Fleming 5693d153a7cSAndy Fleming /** 570d1a203eaSStephen Hemminger * skb_recycle_check - check if skb can be reused for receive 571d1a203eaSStephen Hemminger * @skb: buffer 572d1a203eaSStephen Hemminger * @skb_size: minimum receive buffer size 573d1a203eaSStephen Hemminger * 574d1a203eaSStephen Hemminger * Checks that the skb passed in is not shared or cloned, and 575d1a203eaSStephen Hemminger * that it is linear and its head portion at least as large as 576d1a203eaSStephen Hemminger * skb_size so that it can be recycled as a receive buffer. 577d1a203eaSStephen Hemminger * If these conditions are met, this function does any necessary 578d1a203eaSStephen Hemminger * reference count dropping and cleans up the skbuff as if it 579d1a203eaSStephen Hemminger * just came from __alloc_skb(). 580d1a203eaSStephen Hemminger */ 5815b0daa34SChangli Gao bool skb_recycle_check(struct sk_buff *skb, int skb_size) 58204a4bb55SLennert Buytenhek { 5833d153a7cSAndy Fleming if (!skb_is_recycleable(skb, skb_size)) 5845b0daa34SChangli Gao return false; 585e84af6ddSAnton Vorontsov 5863d153a7cSAndy Fleming skb_recycle(skb); 58704a4bb55SLennert Buytenhek 5885b0daa34SChangli Gao return true; 58904a4bb55SLennert Buytenhek } 59004a4bb55SLennert Buytenhek EXPORT_SYMBOL(skb_recycle_check); 59104a4bb55SLennert Buytenhek 592dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 593dec18810SHerbert Xu { 594dec18810SHerbert Xu new->tstamp = old->tstamp; 595dec18810SHerbert Xu new->dev = old->dev; 596dec18810SHerbert Xu new->transport_header = old->transport_header; 597dec18810SHerbert Xu new->network_header = old->network_header; 598dec18810SHerbert Xu new->mac_header = old->mac_header; 5997fee226aSEric Dumazet skb_dst_copy(new, old); 6000a9627f2STom Herbert new->rxhash = old->rxhash; 6016461be3aSChangli Gao new->ooo_okay = old->ooo_okay; 602bdeab991STom Herbert new->l4_rxhash = old->l4_rxhash; 6033bdc0ebaSBen Greear new->no_fcs = old->no_fcs; 604def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 605dec18810SHerbert Xu new->sp = secpath_get(old->sp); 606dec18810SHerbert Xu #endif 607dec18810SHerbert Xu memcpy(new->cb, old->cb, sizeof(old->cb)); 6089bcb97caSHerbert Xu new->csum = old->csum; 609dec18810SHerbert Xu new->local_df = old->local_df; 610dec18810SHerbert Xu new->pkt_type = old->pkt_type; 611dec18810SHerbert Xu new->ip_summed = old->ip_summed; 612dec18810SHerbert Xu skb_copy_queue_mapping(new, old); 613dec18810SHerbert Xu new->priority = old->priority; 614a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_IP_VS) 615dec18810SHerbert Xu new->ipvs_property = old->ipvs_property; 616dec18810SHerbert Xu #endif 617dec18810SHerbert Xu new->protocol = old->protocol; 618dec18810SHerbert Xu new->mark = old->mark; 6198964be4aSEric Dumazet new->skb_iif = old->skb_iif; 620dec18810SHerbert Xu __nf_copy(new, old); 621a3bf7ae9SIgor Maravić #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) 622dec18810SHerbert Xu new->nf_trace = old->nf_trace; 623dec18810SHerbert Xu #endif 624dec18810SHerbert Xu #ifdef CONFIG_NET_SCHED 625dec18810SHerbert Xu new->tc_index = old->tc_index; 626dec18810SHerbert Xu #ifdef CONFIG_NET_CLS_ACT 627dec18810SHerbert Xu new->tc_verd = old->tc_verd; 628dec18810SHerbert Xu #endif 629dec18810SHerbert Xu #endif 6306aa895b0SPatrick McHardy new->vlan_tci = old->vlan_tci; 6316aa895b0SPatrick McHardy 632dec18810SHerbert Xu skb_copy_secmark(new, old); 633dec18810SHerbert Xu } 634dec18810SHerbert Xu 63582c49a35SHerbert Xu /* 63682c49a35SHerbert Xu * You should not add any new code to this function. Add it to 63782c49a35SHerbert Xu * __copy_skb_header above instead. 63882c49a35SHerbert Xu */ 639e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 6401da177e4SLinus Torvalds { 6411da177e4SLinus Torvalds #define C(x) n->x = skb->x 6421da177e4SLinus Torvalds 6431da177e4SLinus Torvalds n->next = n->prev = NULL; 6441da177e4SLinus Torvalds n->sk = NULL; 645dec18810SHerbert Xu __copy_skb_header(n, skb); 646dec18810SHerbert Xu 6471da177e4SLinus Torvalds C(len); 6481da177e4SLinus Torvalds C(data_len); 6493e6b3b2eSAlexey Dobriyan C(mac_len); 650334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 65102f1c89dSPaul Moore n->cloned = 1; 6521da177e4SLinus Torvalds n->nohdr = 0; 6531da177e4SLinus Torvalds n->destructor = NULL; 6541da177e4SLinus Torvalds C(tail); 6551da177e4SLinus Torvalds C(end); 65602f1c89dSPaul Moore C(head); 657d3836f21SEric Dumazet C(head_frag); 65802f1c89dSPaul Moore C(data); 65902f1c89dSPaul Moore C(truesize); 66002f1c89dSPaul Moore atomic_set(&n->users, 1); 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 6631da177e4SLinus Torvalds skb->cloned = 1; 6641da177e4SLinus Torvalds 6651da177e4SLinus Torvalds return n; 666e0053ec0SHerbert Xu #undef C 667e0053ec0SHerbert Xu } 668e0053ec0SHerbert Xu 669e0053ec0SHerbert Xu /** 670e0053ec0SHerbert Xu * skb_morph - morph one skb into another 671e0053ec0SHerbert Xu * @dst: the skb to receive the contents 672e0053ec0SHerbert Xu * @src: the skb to supply the contents 673e0053ec0SHerbert Xu * 674e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 675e0053ec0SHerbert Xu * supplied by the user. 676e0053ec0SHerbert Xu * 677e0053ec0SHerbert Xu * The target skb is returned upon exit. 678e0053ec0SHerbert Xu */ 679e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 680e0053ec0SHerbert Xu { 6812d4baff8SHerbert Xu skb_release_all(dst); 682e0053ec0SHerbert Xu return __skb_clone(dst, src); 683e0053ec0SHerbert Xu } 684e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 685e0053ec0SHerbert Xu 68648c83012SMichael S. Tsirkin /* skb_copy_ubufs - copy userspace skb frags buffers to kernel 68748c83012SMichael S. Tsirkin * @skb: the skb to modify 68848c83012SMichael S. Tsirkin * @gfp_mask: allocation priority 68948c83012SMichael S. Tsirkin * 69048c83012SMichael S. Tsirkin * This must be called on SKBTX_DEV_ZEROCOPY skb. 69148c83012SMichael S. Tsirkin * It will copy all frags into kernel and drop the reference 69248c83012SMichael S. Tsirkin * to userspace pages. 69348c83012SMichael S. Tsirkin * 69448c83012SMichael S. Tsirkin * If this function is called from an interrupt gfp_mask() must be 69548c83012SMichael S. Tsirkin * %GFP_ATOMIC. 69648c83012SMichael S. Tsirkin * 69748c83012SMichael S. Tsirkin * Returns 0 on success or a negative error code on failure 69848c83012SMichael S. Tsirkin * to allocate kernel memory to copy to. 69948c83012SMichael S. Tsirkin */ 70048c83012SMichael S. Tsirkin int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 701a6686f2fSShirley Ma { 702a6686f2fSShirley Ma int i; 703a6686f2fSShirley Ma int num_frags = skb_shinfo(skb)->nr_frags; 704a6686f2fSShirley Ma struct page *page, *head = NULL; 705a6686f2fSShirley Ma struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 706a6686f2fSShirley Ma 707a6686f2fSShirley Ma for (i = 0; i < num_frags; i++) { 708a6686f2fSShirley Ma u8 *vaddr; 709a6686f2fSShirley Ma skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 710a6686f2fSShirley Ma 711a6686f2fSShirley Ma page = alloc_page(GFP_ATOMIC); 712a6686f2fSShirley Ma if (!page) { 713a6686f2fSShirley Ma while (head) { 714a6686f2fSShirley Ma struct page *next = (struct page *)head->private; 715a6686f2fSShirley Ma put_page(head); 716a6686f2fSShirley Ma head = next; 717a6686f2fSShirley Ma } 718a6686f2fSShirley Ma return -ENOMEM; 719a6686f2fSShirley Ma } 72051c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 721a6686f2fSShirley Ma memcpy(page_address(page), 7229e903e08SEric Dumazet vaddr + f->page_offset, skb_frag_size(f)); 72351c56b00SEric Dumazet kunmap_atomic(vaddr); 724a6686f2fSShirley Ma page->private = (unsigned long)head; 725a6686f2fSShirley Ma head = page; 726a6686f2fSShirley Ma } 727a6686f2fSShirley Ma 728a6686f2fSShirley Ma /* skb frags release userspace buffers */ 729a6686f2fSShirley Ma for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 730a8605c60SIan Campbell skb_frag_unref(skb, i); 731a6686f2fSShirley Ma 732a6686f2fSShirley Ma uarg->callback(uarg); 733a6686f2fSShirley Ma 734a6686f2fSShirley Ma /* skb frags point to kernel buffers */ 735a6686f2fSShirley Ma for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) { 736a8605c60SIan Campbell __skb_fill_page_desc(skb, i-1, head, 0, 737a8605c60SIan Campbell skb_shinfo(skb)->frags[i - 1].size); 738a6686f2fSShirley Ma head = (struct page *)head->private; 739a6686f2fSShirley Ma } 74048c83012SMichael S. Tsirkin 74148c83012SMichael S. Tsirkin skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 742a6686f2fSShirley Ma return 0; 743a6686f2fSShirley Ma } 744a6686f2fSShirley Ma 745a6686f2fSShirley Ma 746e0053ec0SHerbert Xu /** 747e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 748e0053ec0SHerbert Xu * @skb: buffer to clone 749e0053ec0SHerbert Xu * @gfp_mask: allocation priority 750e0053ec0SHerbert Xu * 751e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 752e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 753e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 754e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 755e0053ec0SHerbert Xu * 756e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 757e0053ec0SHerbert Xu * %GFP_ATOMIC. 758e0053ec0SHerbert Xu */ 759e0053ec0SHerbert Xu 760e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 761e0053ec0SHerbert Xu { 762e0053ec0SHerbert Xu struct sk_buff *n; 763e0053ec0SHerbert Xu 764a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 765a6686f2fSShirley Ma if (skb_copy_ubufs(skb, gfp_mask)) 766a6686f2fSShirley Ma return NULL; 767a6686f2fSShirley Ma } 768a6686f2fSShirley Ma 769e0053ec0SHerbert Xu n = skb + 1; 770e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 771e0053ec0SHerbert Xu n->fclone == SKB_FCLONE_UNAVAILABLE) { 772e0053ec0SHerbert Xu atomic_t *fclone_ref = (atomic_t *) (n + 1); 773e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_CLONE; 774e0053ec0SHerbert Xu atomic_inc(fclone_ref); 775e0053ec0SHerbert Xu } else { 776e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 777e0053ec0SHerbert Xu if (!n) 778e0053ec0SHerbert Xu return NULL; 779fe55f6d5SVegard Nossum 780fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags1); 781fe55f6d5SVegard Nossum kmemcheck_annotate_bitfield(n, flags2); 782e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 783e0053ec0SHerbert Xu } 784e0053ec0SHerbert Xu 785e0053ec0SHerbert Xu return __skb_clone(n, skb); 7861da177e4SLinus Torvalds } 787b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 7881da177e4SLinus Torvalds 7891da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 7901da177e4SLinus Torvalds { 7912e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 7921da177e4SLinus Torvalds /* 7931da177e4SLinus Torvalds * Shift between the two data areas in bytes 7941da177e4SLinus Torvalds */ 7951da177e4SLinus Torvalds unsigned long offset = new->data - old->data; 7962e07fa9cSArnaldo Carvalho de Melo #endif 797dec18810SHerbert Xu 798dec18810SHerbert Xu __copy_skb_header(new, old); 799dec18810SHerbert Xu 8002e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 8012e07fa9cSArnaldo Carvalho de Melo /* {transport,network,mac}_header are relative to skb->head */ 8022e07fa9cSArnaldo Carvalho de Melo new->transport_header += offset; 8032e07fa9cSArnaldo Carvalho de Melo new->network_header += offset; 804603a8bbeSStephen Hemminger if (skb_mac_header_was_set(new)) 8052e07fa9cSArnaldo Carvalho de Melo new->mac_header += offset; 8062e07fa9cSArnaldo Carvalho de Melo #endif 8077967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 8087967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 8097967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds 8121da177e4SLinus Torvalds /** 8131da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 8141da177e4SLinus Torvalds * @skb: buffer to copy 8151da177e4SLinus Torvalds * @gfp_mask: allocation priority 8161da177e4SLinus Torvalds * 8171da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 8181da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 8191da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 8201da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 8211da177e4SLinus Torvalds * 8221da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 8231da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 8241da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 8251da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 8261da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 8271da177e4SLinus Torvalds */ 8281da177e4SLinus Torvalds 829dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 8301da177e4SLinus Torvalds { 8316602cebbSEric Dumazet int headerlen = skb_headroom(skb); 832ec47ea82SAlexander Duyck unsigned int size = skb_end_offset(skb) + skb->data_len; 8336602cebbSEric Dumazet struct sk_buff *n = alloc_skb(size, gfp_mask); 8346602cebbSEric Dumazet 8351da177e4SLinus Torvalds if (!n) 8361da177e4SLinus Torvalds return NULL; 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds /* Set the data pointer */ 8391da177e4SLinus Torvalds skb_reserve(n, headerlen); 8401da177e4SLinus Torvalds /* Set the tail pointer and length */ 8411da177e4SLinus Torvalds skb_put(n, skb->len); 8421da177e4SLinus Torvalds 8431da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 8441da177e4SLinus Torvalds BUG(); 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds copy_skb_header(n, skb); 8471da177e4SLinus Torvalds return n; 8481da177e4SLinus Torvalds } 849b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 8501da177e4SLinus Torvalds 8511da177e4SLinus Torvalds /** 852117632e6SEric Dumazet * __pskb_copy - create copy of an sk_buff with private head. 8531da177e4SLinus Torvalds * @skb: buffer to copy 854117632e6SEric Dumazet * @headroom: headroom of new skb 8551da177e4SLinus Torvalds * @gfp_mask: allocation priority 8561da177e4SLinus Torvalds * 8571da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 8581da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 8591da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 8601da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 8611da177e4SLinus Torvalds * or the pointer to the buffer on success. 8621da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 8631da177e4SLinus Torvalds */ 8641da177e4SLinus Torvalds 865117632e6SEric Dumazet struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask) 8661da177e4SLinus Torvalds { 867117632e6SEric Dumazet unsigned int size = skb_headlen(skb) + headroom; 8686602cebbSEric Dumazet struct sk_buff *n = alloc_skb(size, gfp_mask); 8696602cebbSEric Dumazet 8701da177e4SLinus Torvalds if (!n) 8711da177e4SLinus Torvalds goto out; 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds /* Set the data pointer */ 874117632e6SEric Dumazet skb_reserve(n, headroom); 8751da177e4SLinus Torvalds /* Set the tail pointer and length */ 8761da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 8771da177e4SLinus Torvalds /* Copy the bytes */ 878d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 8791da177e4SLinus Torvalds 88025f484a6SHerbert Xu n->truesize += skb->data_len; 8811da177e4SLinus Torvalds n->data_len = skb->data_len; 8821da177e4SLinus Torvalds n->len = skb->len; 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 8851da177e4SLinus Torvalds int i; 8861da177e4SLinus Torvalds 887a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 888a6686f2fSShirley Ma if (skb_copy_ubufs(skb, gfp_mask)) { 8891511022cSDan Carpenter kfree_skb(n); 8901511022cSDan Carpenter n = NULL; 891a6686f2fSShirley Ma goto out; 892a6686f2fSShirley Ma } 893a6686f2fSShirley Ma } 8941da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 8951da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 896ea2ab693SIan Campbell skb_frag_ref(skb, i); 8971da177e4SLinus Torvalds } 8981da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 8991da177e4SLinus Torvalds } 9001da177e4SLinus Torvalds 90121dc3301SDavid S. Miller if (skb_has_frag_list(skb)) { 9021da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 9031da177e4SLinus Torvalds skb_clone_fraglist(n); 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds copy_skb_header(n, skb); 9071da177e4SLinus Torvalds out: 9081da177e4SLinus Torvalds return n; 9091da177e4SLinus Torvalds } 910117632e6SEric Dumazet EXPORT_SYMBOL(__pskb_copy); 9111da177e4SLinus Torvalds 9121da177e4SLinus Torvalds /** 9131da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 9141da177e4SLinus Torvalds * @skb: buffer to reallocate 9151da177e4SLinus Torvalds * @nhead: room to add at head 9161da177e4SLinus Torvalds * @ntail: room to add at tail 9171da177e4SLinus Torvalds * @gfp_mask: allocation priority 9181da177e4SLinus Torvalds * 9191da177e4SLinus Torvalds * Expands (or creates identical copy, if &nhead and &ntail are zero) 9201da177e4SLinus Torvalds * header of skb. &sk_buff itself is not changed. &sk_buff MUST have 9211da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 9221da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 9231da177e4SLinus Torvalds * 9241da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 9251da177e4SLinus Torvalds * reloaded after call to this function. 9261da177e4SLinus Torvalds */ 9271da177e4SLinus Torvalds 92886a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 929dd0fc66fSAl Viro gfp_t gfp_mask) 9301da177e4SLinus Torvalds { 9311da177e4SLinus Torvalds int i; 9321da177e4SLinus Torvalds u8 *data; 933ec47ea82SAlexander Duyck int size = nhead + skb_end_offset(skb) + ntail; 9341da177e4SLinus Torvalds long off; 9351da177e4SLinus Torvalds 9364edd87adSHerbert Xu BUG_ON(nhead < 0); 9374edd87adSHerbert Xu 9381da177e4SLinus Torvalds if (skb_shared(skb)) 9391da177e4SLinus Torvalds BUG(); 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 9421da177e4SLinus Torvalds 94387151b86SEric Dumazet data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 94487151b86SEric Dumazet gfp_mask); 9451da177e4SLinus Torvalds if (!data) 9461da177e4SLinus Torvalds goto nodata; 94787151b86SEric Dumazet size = SKB_WITH_OVERHEAD(ksize(data)); 9481da177e4SLinus Torvalds 9491da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 9506602cebbSEric Dumazet * optimized for the cases when header is void. 9516602cebbSEric Dumazet */ 9526602cebbSEric Dumazet memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 9536602cebbSEric Dumazet 9546602cebbSEric Dumazet memcpy((struct skb_shared_info *)(data + size), 9556602cebbSEric Dumazet skb_shinfo(skb), 956fed66381SEric Dumazet offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 9571da177e4SLinus Torvalds 9583e24591aSAlexander Duyck /* 9593e24591aSAlexander Duyck * if shinfo is shared we must drop the old head gracefully, but if it 9603e24591aSAlexander Duyck * is not we can just drop the old head and let the existing refcount 9613e24591aSAlexander Duyck * be since all we did is relocate the values 9623e24591aSAlexander Duyck */ 9633e24591aSAlexander Duyck if (skb_cloned(skb)) { 964a6686f2fSShirley Ma /* copy this zero copy skb frags */ 965a6686f2fSShirley Ma if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 966a6686f2fSShirley Ma if (skb_copy_ubufs(skb, gfp_mask)) 967a6686f2fSShirley Ma goto nofrags; 968a6686f2fSShirley Ma } 9691da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 970ea2ab693SIan Campbell skb_frag_ref(skb, i); 9711da177e4SLinus Torvalds 97221dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 9731da177e4SLinus Torvalds skb_clone_fraglist(skb); 9741da177e4SLinus Torvalds 9751da177e4SLinus Torvalds skb_release_data(skb); 9763e24591aSAlexander Duyck } else { 9773e24591aSAlexander Duyck skb_free_head(skb); 9781fd63041SEric Dumazet } 9791da177e4SLinus Torvalds off = (data + nhead) - skb->head; 9801da177e4SLinus Torvalds 9811da177e4SLinus Torvalds skb->head = data; 982d3836f21SEric Dumazet skb->head_frag = 0; 9831da177e4SLinus Torvalds skb->data += off; 9844305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 9854305b541SArnaldo Carvalho de Melo skb->end = size; 98656eb8882SPatrick McHardy off = nhead; 9874305b541SArnaldo Carvalho de Melo #else 9884305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 98956eb8882SPatrick McHardy #endif 99027a884dcSArnaldo Carvalho de Melo /* {transport,network,mac}_header and tail are relative to skb->head */ 99127a884dcSArnaldo Carvalho de Melo skb->tail += off; 992b0e380b1SArnaldo Carvalho de Melo skb->transport_header += off; 993b0e380b1SArnaldo Carvalho de Melo skb->network_header += off; 994603a8bbeSStephen Hemminger if (skb_mac_header_was_set(skb)) 995b0e380b1SArnaldo Carvalho de Melo skb->mac_header += off; 99600c5a983SAndrea Shepard /* Only adjust this if it actually is csum_start rather than csum */ 99700c5a983SAndrea Shepard if (skb->ip_summed == CHECKSUM_PARTIAL) 998172a863fSHerbert Xu skb->csum_start += nhead; 9991da177e4SLinus Torvalds skb->cloned = 0; 1000334a8132SPatrick McHardy skb->hdr_len = 0; 10011da177e4SLinus Torvalds skb->nohdr = 0; 10021da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 10031da177e4SLinus Torvalds return 0; 10041da177e4SLinus Torvalds 1005a6686f2fSShirley Ma nofrags: 1006a6686f2fSShirley Ma kfree(data); 10071da177e4SLinus Torvalds nodata: 10081da177e4SLinus Torvalds return -ENOMEM; 10091da177e4SLinus Torvalds } 1010b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 10111da177e4SLinus Torvalds 10121da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 10131da177e4SLinus Torvalds 10141da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 10151da177e4SLinus Torvalds { 10161da177e4SLinus Torvalds struct sk_buff *skb2; 10171da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds if (delta <= 0) 10201da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 10211da177e4SLinus Torvalds else { 10221da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 10231da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 10241da177e4SLinus Torvalds GFP_ATOMIC)) { 10251da177e4SLinus Torvalds kfree_skb(skb2); 10261da177e4SLinus Torvalds skb2 = NULL; 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds return skb2; 10301da177e4SLinus Torvalds } 1031b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 10321da177e4SLinus Torvalds 10331da177e4SLinus Torvalds /** 10341da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 10351da177e4SLinus Torvalds * @skb: buffer to copy 10361da177e4SLinus Torvalds * @newheadroom: new free bytes at head 10371da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 10381da177e4SLinus Torvalds * @gfp_mask: allocation priority 10391da177e4SLinus Torvalds * 10401da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 10411da177e4SLinus Torvalds * allocate additional space. 10421da177e4SLinus Torvalds * 10431da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 10441da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 10451da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 10461da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 10471da177e4SLinus Torvalds * 10481da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 10491da177e4SLinus Torvalds * is called from an interrupt. 10501da177e4SLinus Torvalds */ 10511da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 105286a76cafSVictor Fusco int newheadroom, int newtailroom, 1053dd0fc66fSAl Viro gfp_t gfp_mask) 10541da177e4SLinus Torvalds { 10551da177e4SLinus Torvalds /* 10561da177e4SLinus Torvalds * Allocate the copy buffer 10571da177e4SLinus Torvalds */ 10581da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom, 10591da177e4SLinus Torvalds gfp_mask); 1060efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 10611da177e4SLinus Torvalds int head_copy_len, head_copy_off; 106252886051SHerbert Xu int off; 10631da177e4SLinus Torvalds 10641da177e4SLinus Torvalds if (!n) 10651da177e4SLinus Torvalds return NULL; 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds skb_reserve(n, newheadroom); 10681da177e4SLinus Torvalds 10691da177e4SLinus Torvalds /* Set the tail pointer and length */ 10701da177e4SLinus Torvalds skb_put(n, skb->len); 10711da177e4SLinus Torvalds 1072efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 10731da177e4SLinus Torvalds head_copy_off = 0; 10741da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 10751da177e4SLinus Torvalds head_copy_len = newheadroom; 10761da177e4SLinus Torvalds else 10771da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds /* Copy the linear header and data. */ 10801da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 10811da177e4SLinus Torvalds skb->len + head_copy_len)) 10821da177e4SLinus Torvalds BUG(); 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds copy_skb_header(n, skb); 10851da177e4SLinus Torvalds 1086efd1e8d5SPatrick McHardy off = newheadroom - oldheadroom; 1087be2b6e62SDavid S. Miller if (n->ip_summed == CHECKSUM_PARTIAL) 108852886051SHerbert Xu n->csum_start += off; 108952886051SHerbert Xu #ifdef NET_SKBUFF_DATA_USES_OFFSET 1090efd1e8d5SPatrick McHardy n->transport_header += off; 1091efd1e8d5SPatrick McHardy n->network_header += off; 1092603a8bbeSStephen Hemminger if (skb_mac_header_was_set(skb)) 1093efd1e8d5SPatrick McHardy n->mac_header += off; 109452886051SHerbert Xu #endif 1095efd1e8d5SPatrick McHardy 10961da177e4SLinus Torvalds return n; 10971da177e4SLinus Torvalds } 1098b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 10991da177e4SLinus Torvalds 11001da177e4SLinus Torvalds /** 11011da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 11021da177e4SLinus Torvalds * @skb: buffer to pad 11031da177e4SLinus Torvalds * @pad: space to pad 11041da177e4SLinus Torvalds * 11051da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 11061da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 11071da177e4SLinus Torvalds * beyond the buffer end onto the wire. 11081da177e4SLinus Torvalds * 11095b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 11101da177e4SLinus Torvalds */ 11111da177e4SLinus Torvalds 11125b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 11131da177e4SLinus Torvalds { 11145b057c6bSHerbert Xu int err; 11155b057c6bSHerbert Xu int ntail; 11161da177e4SLinus Torvalds 11171da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 11185b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 11191da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 11205b057c6bSHerbert Xu return 0; 11211da177e4SLinus Torvalds } 11221da177e4SLinus Torvalds 11234305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 11245b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 11255b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 11265b057c6bSHerbert Xu if (unlikely(err)) 11275b057c6bSHerbert Xu goto free_skb; 11285b057c6bSHerbert Xu } 11295b057c6bSHerbert Xu 11305b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 11315b057c6bSHerbert Xu * to be audited. 11325b057c6bSHerbert Xu */ 11335b057c6bSHerbert Xu err = skb_linearize(skb); 11345b057c6bSHerbert Xu if (unlikely(err)) 11355b057c6bSHerbert Xu goto free_skb; 11365b057c6bSHerbert Xu 11375b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 11385b057c6bSHerbert Xu return 0; 11395b057c6bSHerbert Xu 11405b057c6bSHerbert Xu free_skb: 11411da177e4SLinus Torvalds kfree_skb(skb); 11425b057c6bSHerbert Xu return err; 11431da177e4SLinus Torvalds } 1144b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 11451da177e4SLinus Torvalds 11460dde3e16SIlpo Järvinen /** 11470dde3e16SIlpo Järvinen * skb_put - add data to a buffer 11480dde3e16SIlpo Järvinen * @skb: buffer to use 11490dde3e16SIlpo Järvinen * @len: amount of data to add 11500dde3e16SIlpo Järvinen * 11510dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 11520dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 11530dde3e16SIlpo Järvinen * first byte of the extra data is returned. 11540dde3e16SIlpo Järvinen */ 11550dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 11560dde3e16SIlpo Järvinen { 11570dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 11580dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 11590dde3e16SIlpo Järvinen skb->tail += len; 11600dde3e16SIlpo Järvinen skb->len += len; 11610dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 11620dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 11630dde3e16SIlpo Järvinen return tmp; 11640dde3e16SIlpo Järvinen } 11650dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 11660dde3e16SIlpo Järvinen 11676be8ac2fSIlpo Järvinen /** 1168c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 1169c2aa270aSIlpo Järvinen * @skb: buffer to use 1170c2aa270aSIlpo Järvinen * @len: amount of data to add 1171c2aa270aSIlpo Järvinen * 1172c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 1173c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 1174c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 1175c2aa270aSIlpo Järvinen */ 1176c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1177c2aa270aSIlpo Järvinen { 1178c2aa270aSIlpo Järvinen skb->data -= len; 1179c2aa270aSIlpo Järvinen skb->len += len; 1180c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1181c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1182c2aa270aSIlpo Järvinen return skb->data; 1183c2aa270aSIlpo Järvinen } 1184c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1185c2aa270aSIlpo Järvinen 1186c2aa270aSIlpo Järvinen /** 11876be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 11886be8ac2fSIlpo Järvinen * @skb: buffer to use 11896be8ac2fSIlpo Järvinen * @len: amount of data to remove 11906be8ac2fSIlpo Järvinen * 11916be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 11926be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 11936be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 11946be8ac2fSIlpo Järvinen * the old data. 11956be8ac2fSIlpo Järvinen */ 11966be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 11976be8ac2fSIlpo Järvinen { 119847d29646SDavid S. Miller return skb_pull_inline(skb, len); 11996be8ac2fSIlpo Järvinen } 12006be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 12016be8ac2fSIlpo Järvinen 1202419ae74eSIlpo Järvinen /** 1203419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1204419ae74eSIlpo Järvinen * @skb: buffer to alter 1205419ae74eSIlpo Järvinen * @len: new length 1206419ae74eSIlpo Järvinen * 1207419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1208419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1209419ae74eSIlpo Järvinen * The skb must be linear. 1210419ae74eSIlpo Järvinen */ 1211419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1212419ae74eSIlpo Järvinen { 1213419ae74eSIlpo Järvinen if (skb->len > len) 1214419ae74eSIlpo Järvinen __skb_trim(skb, len); 1215419ae74eSIlpo Järvinen } 1216419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1217419ae74eSIlpo Järvinen 12183cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 12191da177e4SLinus Torvalds */ 12201da177e4SLinus Torvalds 12213cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 12221da177e4SLinus Torvalds { 122327b437c8SHerbert Xu struct sk_buff **fragp; 122427b437c8SHerbert Xu struct sk_buff *frag; 12251da177e4SLinus Torvalds int offset = skb_headlen(skb); 12261da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 12271da177e4SLinus Torvalds int i; 122827b437c8SHerbert Xu int err; 122927b437c8SHerbert Xu 123027b437c8SHerbert Xu if (skb_cloned(skb) && 123127b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 123227b437c8SHerbert Xu return err; 12331da177e4SLinus Torvalds 1234f4d26fb3SHerbert Xu i = 0; 1235f4d26fb3SHerbert Xu if (offset >= len) 1236f4d26fb3SHerbert Xu goto drop_pages; 1237f4d26fb3SHerbert Xu 1238f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 12399e903e08SEric Dumazet int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 124027b437c8SHerbert Xu 124127b437c8SHerbert Xu if (end < len) { 12421da177e4SLinus Torvalds offset = end; 124327b437c8SHerbert Xu continue; 12441da177e4SLinus Torvalds } 12451da177e4SLinus Torvalds 12469e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 124727b437c8SHerbert Xu 1248f4d26fb3SHerbert Xu drop_pages: 124927b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 125027b437c8SHerbert Xu 125127b437c8SHerbert Xu for (; i < nfrags; i++) 1252ea2ab693SIan Campbell skb_frag_unref(skb, i); 125327b437c8SHerbert Xu 125421dc3301SDavid S. Miller if (skb_has_frag_list(skb)) 125527b437c8SHerbert Xu skb_drop_fraglist(skb); 1256f4d26fb3SHerbert Xu goto done; 125727b437c8SHerbert Xu } 125827b437c8SHerbert Xu 125927b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 126027b437c8SHerbert Xu fragp = &frag->next) { 126127b437c8SHerbert Xu int end = offset + frag->len; 126227b437c8SHerbert Xu 126327b437c8SHerbert Xu if (skb_shared(frag)) { 126427b437c8SHerbert Xu struct sk_buff *nfrag; 126527b437c8SHerbert Xu 126627b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 126727b437c8SHerbert Xu if (unlikely(!nfrag)) 126827b437c8SHerbert Xu return -ENOMEM; 126927b437c8SHerbert Xu 127027b437c8SHerbert Xu nfrag->next = frag->next; 127185bb2a60SEric Dumazet consume_skb(frag); 127227b437c8SHerbert Xu frag = nfrag; 127327b437c8SHerbert Xu *fragp = frag; 127427b437c8SHerbert Xu } 127527b437c8SHerbert Xu 127627b437c8SHerbert Xu if (end < len) { 127727b437c8SHerbert Xu offset = end; 127827b437c8SHerbert Xu continue; 127927b437c8SHerbert Xu } 128027b437c8SHerbert Xu 128127b437c8SHerbert Xu if (end > len && 128227b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 128327b437c8SHerbert Xu return err; 128427b437c8SHerbert Xu 128527b437c8SHerbert Xu if (frag->next) 128627b437c8SHerbert Xu skb_drop_list(&frag->next); 128727b437c8SHerbert Xu break; 128827b437c8SHerbert Xu } 128927b437c8SHerbert Xu 1290f4d26fb3SHerbert Xu done: 129127b437c8SHerbert Xu if (len > skb_headlen(skb)) { 12921da177e4SLinus Torvalds skb->data_len -= skb->len - len; 12931da177e4SLinus Torvalds skb->len = len; 12941da177e4SLinus Torvalds } else { 12951da177e4SLinus Torvalds skb->len = len; 12961da177e4SLinus Torvalds skb->data_len = 0; 129727a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 12981da177e4SLinus Torvalds } 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds return 0; 13011da177e4SLinus Torvalds } 1302b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 13031da177e4SLinus Torvalds 13041da177e4SLinus Torvalds /** 13051da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 13061da177e4SLinus Torvalds * @skb: buffer to reallocate 13071da177e4SLinus Torvalds * @delta: number of bytes to advance tail 13081da177e4SLinus Torvalds * 13091da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 13101da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 13111da177e4SLinus Torvalds * data from fragmented part. 13121da177e4SLinus Torvalds * 13131da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 13141da177e4SLinus Torvalds * 13151da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 13161da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 13171da177e4SLinus Torvalds * 13181da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 13191da177e4SLinus Torvalds * reloaded after call to this function. 13201da177e4SLinus Torvalds */ 13211da177e4SLinus Torvalds 13221da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 13231da177e4SLinus Torvalds * when it is necessary. 13241da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 13251da177e4SLinus Torvalds * 2. It may change skb pointers. 13261da177e4SLinus Torvalds * 13271da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 13281da177e4SLinus Torvalds */ 13291da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 13301da177e4SLinus Torvalds { 13311da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 13321da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 13331da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 13341da177e4SLinus Torvalds */ 13354305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 13381da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 13391da177e4SLinus Torvalds GFP_ATOMIC)) 13401da177e4SLinus Torvalds return NULL; 13411da177e4SLinus Torvalds } 13421da177e4SLinus Torvalds 134327a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 13441da177e4SLinus Torvalds BUG(); 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 13471da177e4SLinus Torvalds * size of pulled pages. Superb. 13481da177e4SLinus Torvalds */ 134921dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) 13501da177e4SLinus Torvalds goto pull_pages; 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 13531da177e4SLinus Torvalds eat = delta; 13541da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 13559e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 13569e903e08SEric Dumazet 13579e903e08SEric Dumazet if (size >= eat) 13581da177e4SLinus Torvalds goto pull_pages; 13599e903e08SEric Dumazet eat -= size; 13601da177e4SLinus Torvalds } 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 13631da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 13641da177e4SLinus Torvalds * but taking into account that pulling is expected to 13651da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 13661da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 13671da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 13681da177e4SLinus Torvalds */ 13691da177e4SLinus Torvalds if (eat) { 13701da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 13711da177e4SLinus Torvalds struct sk_buff *clone = NULL; 13721da177e4SLinus Torvalds struct sk_buff *insp = NULL; 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds do { 137509a62660SKris Katterjohn BUG_ON(!list); 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds if (list->len <= eat) { 13781da177e4SLinus Torvalds /* Eaten as whole. */ 13791da177e4SLinus Torvalds eat -= list->len; 13801da177e4SLinus Torvalds list = list->next; 13811da177e4SLinus Torvalds insp = list; 13821da177e4SLinus Torvalds } else { 13831da177e4SLinus Torvalds /* Eaten partially. */ 13841da177e4SLinus Torvalds 13851da177e4SLinus Torvalds if (skb_shared(list)) { 13861da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 13871da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 13881da177e4SLinus Torvalds if (!clone) 13891da177e4SLinus Torvalds return NULL; 13901da177e4SLinus Torvalds insp = list->next; 13911da177e4SLinus Torvalds list = clone; 13921da177e4SLinus Torvalds } else { 13931da177e4SLinus Torvalds /* This may be pulled without 13941da177e4SLinus Torvalds * problems. */ 13951da177e4SLinus Torvalds insp = list; 13961da177e4SLinus Torvalds } 13971da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 13981da177e4SLinus Torvalds kfree_skb(clone); 13991da177e4SLinus Torvalds return NULL; 14001da177e4SLinus Torvalds } 14011da177e4SLinus Torvalds break; 14021da177e4SLinus Torvalds } 14031da177e4SLinus Torvalds } while (eat); 14041da177e4SLinus Torvalds 14051da177e4SLinus Torvalds /* Free pulled out fragments. */ 14061da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 14071da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 14081da177e4SLinus Torvalds kfree_skb(list); 14091da177e4SLinus Torvalds } 14101da177e4SLinus Torvalds /* And insert new clone at head. */ 14111da177e4SLinus Torvalds if (clone) { 14121da177e4SLinus Torvalds clone->next = list; 14131da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 14141da177e4SLinus Torvalds } 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 14171da177e4SLinus Torvalds 14181da177e4SLinus Torvalds pull_pages: 14191da177e4SLinus Torvalds eat = delta; 14201da177e4SLinus Torvalds k = 0; 14211da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 14229e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 14239e903e08SEric Dumazet 14249e903e08SEric Dumazet if (size <= eat) { 1425ea2ab693SIan Campbell skb_frag_unref(skb, i); 14269e903e08SEric Dumazet eat -= size; 14271da177e4SLinus Torvalds } else { 14281da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 14291da177e4SLinus Torvalds if (eat) { 14301da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 14319e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 14321da177e4SLinus Torvalds eat = 0; 14331da177e4SLinus Torvalds } 14341da177e4SLinus Torvalds k++; 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 14381da177e4SLinus Torvalds 14391da177e4SLinus Torvalds skb->tail += delta; 14401da177e4SLinus Torvalds skb->data_len -= delta; 14411da177e4SLinus Torvalds 144227a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 14431da177e4SLinus Torvalds } 1444b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 14451da177e4SLinus Torvalds 144622019b17SEric Dumazet /** 144722019b17SEric Dumazet * skb_copy_bits - copy bits from skb to kernel buffer 144822019b17SEric Dumazet * @skb: source skb 144922019b17SEric Dumazet * @offset: offset in source 145022019b17SEric Dumazet * @to: destination buffer 145122019b17SEric Dumazet * @len: number of bytes to copy 145222019b17SEric Dumazet * 145322019b17SEric Dumazet * Copy the specified number of bytes from the source skb to the 145422019b17SEric Dumazet * destination buffer. 145522019b17SEric Dumazet * 145622019b17SEric Dumazet * CAUTION ! : 145722019b17SEric Dumazet * If its prototype is ever changed, 145822019b17SEric Dumazet * check arch/{*}/net/{*}.S files, 145922019b17SEric Dumazet * since it is called from BPF assembly code. 146022019b17SEric Dumazet */ 14611da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 14621da177e4SLinus Torvalds { 14631a028e50SDavid S. Miller int start = skb_headlen(skb); 1464fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1465fbb398a8SDavid S. Miller int i, copy; 14661da177e4SLinus Torvalds 14671da177e4SLinus Torvalds if (offset > (int)skb->len - len) 14681da177e4SLinus Torvalds goto fault; 14691da177e4SLinus Torvalds 14701da177e4SLinus Torvalds /* Copy header. */ 14711a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 14721da177e4SLinus Torvalds if (copy > len) 14731da177e4SLinus Torvalds copy = len; 1474d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 14751da177e4SLinus Torvalds if ((len -= copy) == 0) 14761da177e4SLinus Torvalds return 0; 14771da177e4SLinus Torvalds offset += copy; 14781da177e4SLinus Torvalds to += copy; 14791da177e4SLinus Torvalds } 14801da177e4SLinus Torvalds 14811da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 14821a028e50SDavid S. Miller int end; 148351c56b00SEric Dumazet skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 14841da177e4SLinus Torvalds 1485547b792cSIlpo Järvinen WARN_ON(start > offset + len); 14861a028e50SDavid S. Miller 148751c56b00SEric Dumazet end = start + skb_frag_size(f); 14881da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 14891da177e4SLinus Torvalds u8 *vaddr; 14901da177e4SLinus Torvalds 14911da177e4SLinus Torvalds if (copy > len) 14921da177e4SLinus Torvalds copy = len; 14931da177e4SLinus Torvalds 149451c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(f)); 14951da177e4SLinus Torvalds memcpy(to, 149651c56b00SEric Dumazet vaddr + f->page_offset + offset - start, 149751c56b00SEric Dumazet copy); 149851c56b00SEric Dumazet kunmap_atomic(vaddr); 14991da177e4SLinus Torvalds 15001da177e4SLinus Torvalds if ((len -= copy) == 0) 15011da177e4SLinus Torvalds return 0; 15021da177e4SLinus Torvalds offset += copy; 15031da177e4SLinus Torvalds to += copy; 15041da177e4SLinus Torvalds } 15051a028e50SDavid S. Miller start = end; 15061da177e4SLinus Torvalds } 15071da177e4SLinus Torvalds 1508fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 15091a028e50SDavid S. Miller int end; 15101da177e4SLinus Torvalds 1511547b792cSIlpo Järvinen WARN_ON(start > offset + len); 15121a028e50SDavid S. Miller 1513fbb398a8SDavid S. Miller end = start + frag_iter->len; 15141da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 15151da177e4SLinus Torvalds if (copy > len) 15161da177e4SLinus Torvalds copy = len; 1517fbb398a8SDavid S. Miller if (skb_copy_bits(frag_iter, offset - start, to, copy)) 15181da177e4SLinus Torvalds goto fault; 15191da177e4SLinus Torvalds if ((len -= copy) == 0) 15201da177e4SLinus Torvalds return 0; 15211da177e4SLinus Torvalds offset += copy; 15221da177e4SLinus Torvalds to += copy; 15231da177e4SLinus Torvalds } 15241a028e50SDavid S. Miller start = end; 15251da177e4SLinus Torvalds } 1526a6686f2fSShirley Ma 15271da177e4SLinus Torvalds if (!len) 15281da177e4SLinus Torvalds return 0; 15291da177e4SLinus Torvalds 15301da177e4SLinus Torvalds fault: 15311da177e4SLinus Torvalds return -EFAULT; 15321da177e4SLinus Torvalds } 1533b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 15341da177e4SLinus Torvalds 15359c55e01cSJens Axboe /* 15369c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 15379c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 15389c55e01cSJens Axboe */ 15399c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 15409c55e01cSJens Axboe { 15418b9d3728SJarek Poplawski put_page(spd->pages[i]); 15428b9d3728SJarek Poplawski } 15439c55e01cSJens Axboe 1544a108d5f3SDavid S. Miller static struct page *linear_to_page(struct page *page, unsigned int *len, 15454fb66994SJarek Poplawski unsigned int *offset, 15467a67e56fSJarek Poplawski struct sk_buff *skb, struct sock *sk) 15478b9d3728SJarek Poplawski { 15484fb66994SJarek Poplawski struct page *p = sk->sk_sndmsg_page; 15494fb66994SJarek Poplawski unsigned int off; 15508b9d3728SJarek Poplawski 15514fb66994SJarek Poplawski if (!p) { 15524fb66994SJarek Poplawski new_page: 15534fb66994SJarek Poplawski p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0); 15548b9d3728SJarek Poplawski if (!p) 15558b9d3728SJarek Poplawski return NULL; 15564fb66994SJarek Poplawski 15574fb66994SJarek Poplawski off = sk->sk_sndmsg_off = 0; 15584fb66994SJarek Poplawski /* hold one ref to this page until it's full */ 15594fb66994SJarek Poplawski } else { 15604fb66994SJarek Poplawski unsigned int mlen; 15614fb66994SJarek Poplawski 1562e66e9a31SEric Dumazet /* If we are the only user of the page, we can reset offset */ 1563e66e9a31SEric Dumazet if (page_count(p) == 1) 1564e66e9a31SEric Dumazet sk->sk_sndmsg_off = 0; 15654fb66994SJarek Poplawski off = sk->sk_sndmsg_off; 15664fb66994SJarek Poplawski mlen = PAGE_SIZE - off; 15674fb66994SJarek Poplawski if (mlen < 64 && mlen < *len) { 15684fb66994SJarek Poplawski put_page(p); 15694fb66994SJarek Poplawski goto new_page; 15704fb66994SJarek Poplawski } 15714fb66994SJarek Poplawski 15724fb66994SJarek Poplawski *len = min_t(unsigned int, *len, mlen); 15734fb66994SJarek Poplawski } 15744fb66994SJarek Poplawski 15754fb66994SJarek Poplawski memcpy(page_address(p) + off, page_address(page) + *offset, *len); 15764fb66994SJarek Poplawski sk->sk_sndmsg_off += *len; 15774fb66994SJarek Poplawski *offset = off; 15788b9d3728SJarek Poplawski 15798b9d3728SJarek Poplawski return p; 15809c55e01cSJens Axboe } 15819c55e01cSJens Axboe 158241c73a0dSEric Dumazet static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 158341c73a0dSEric Dumazet struct page *page, 158441c73a0dSEric Dumazet unsigned int offset) 158541c73a0dSEric Dumazet { 158641c73a0dSEric Dumazet return spd->nr_pages && 158741c73a0dSEric Dumazet spd->pages[spd->nr_pages - 1] == page && 158841c73a0dSEric Dumazet (spd->partial[spd->nr_pages - 1].offset + 158941c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len == offset); 159041c73a0dSEric Dumazet } 159141c73a0dSEric Dumazet 15929c55e01cSJens Axboe /* 15939c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 15949c55e01cSJens Axboe */ 1595a108d5f3SDavid S. Miller static bool spd_fill_page(struct splice_pipe_desc *spd, 159635f3d14dSJens Axboe struct pipe_inode_info *pipe, struct page *page, 15974fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 1598d7ccf7c0SEric Dumazet struct sk_buff *skb, bool linear, 15997a67e56fSJarek Poplawski struct sock *sk) 16009c55e01cSJens Axboe { 160141c73a0dSEric Dumazet if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1602a108d5f3SDavid S. Miller return true; 16039c55e01cSJens Axboe 16048b9d3728SJarek Poplawski if (linear) { 16057a67e56fSJarek Poplawski page = linear_to_page(page, len, &offset, skb, sk); 16068b9d3728SJarek Poplawski if (!page) 1607a108d5f3SDavid S. Miller return true; 160841c73a0dSEric Dumazet } 160941c73a0dSEric Dumazet if (spd_can_coalesce(spd, page, offset)) { 161041c73a0dSEric Dumazet spd->partial[spd->nr_pages - 1].len += *len; 1611a108d5f3SDavid S. Miller return false; 161241c73a0dSEric Dumazet } 16138b9d3728SJarek Poplawski get_page(page); 16149c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 16154fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 16169c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 16179c55e01cSJens Axboe spd->nr_pages++; 16188b9d3728SJarek Poplawski 1619a108d5f3SDavid S. Miller return false; 16209c55e01cSJens Axboe } 16219c55e01cSJens Axboe 16222870c43dSOctavian Purdila static inline void __segment_seek(struct page **page, unsigned int *poff, 16232870c43dSOctavian Purdila unsigned int *plen, unsigned int off) 16242870c43dSOctavian Purdila { 1625ce3dd395SJarek Poplawski unsigned long n; 1626ce3dd395SJarek Poplawski 16272870c43dSOctavian Purdila *poff += off; 1628ce3dd395SJarek Poplawski n = *poff / PAGE_SIZE; 1629ce3dd395SJarek Poplawski if (n) 1630ce3dd395SJarek Poplawski *page = nth_page(*page, n); 1631ce3dd395SJarek Poplawski 16322870c43dSOctavian Purdila *poff = *poff % PAGE_SIZE; 16332870c43dSOctavian Purdila *plen -= off; 16342870c43dSOctavian Purdila } 16352870c43dSOctavian Purdila 1636a108d5f3SDavid S. Miller static bool __splice_segment(struct page *page, unsigned int poff, 16372870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 16382870c43dSOctavian Purdila unsigned int *len, struct sk_buff *skb, 1639d7ccf7c0SEric Dumazet struct splice_pipe_desc *spd, bool linear, 164035f3d14dSJens Axboe struct sock *sk, 164135f3d14dSJens Axboe struct pipe_inode_info *pipe) 16429c55e01cSJens Axboe { 16432870c43dSOctavian Purdila if (!*len) 1644a108d5f3SDavid S. Miller return true; 16459c55e01cSJens Axboe 16462870c43dSOctavian Purdila /* skip this segment if already processed */ 16472870c43dSOctavian Purdila if (*off >= plen) { 16482870c43dSOctavian Purdila *off -= plen; 1649a108d5f3SDavid S. Miller return false; 16502870c43dSOctavian Purdila } 16512870c43dSOctavian Purdila 16522870c43dSOctavian Purdila /* ignore any bits we already processed */ 16532870c43dSOctavian Purdila if (*off) { 16542870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, *off); 16552870c43dSOctavian Purdila *off = 0; 16562870c43dSOctavian Purdila } 16572870c43dSOctavian Purdila 16582870c43dSOctavian Purdila do { 16592870c43dSOctavian Purdila unsigned int flen = min(*len, plen); 16602870c43dSOctavian Purdila 16612870c43dSOctavian Purdila /* the linear region may spread across several pages */ 16622870c43dSOctavian Purdila flen = min_t(unsigned int, flen, PAGE_SIZE - poff); 16632870c43dSOctavian Purdila 166435f3d14dSJens Axboe if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk)) 1665a108d5f3SDavid S. Miller return true; 16662870c43dSOctavian Purdila 16672870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, flen); 16682870c43dSOctavian Purdila *len -= flen; 16692870c43dSOctavian Purdila 16702870c43dSOctavian Purdila } while (*len && plen); 16712870c43dSOctavian Purdila 1672a108d5f3SDavid S. Miller return false; 1673db43a282SOctavian Purdila } 16749c55e01cSJens Axboe 16759c55e01cSJens Axboe /* 1676a108d5f3SDavid S. Miller * Map linear and fragment data from the skb to spd. It reports true if the 16772870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 16789c55e01cSJens Axboe */ 1679a108d5f3SDavid S. Miller static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 168035f3d14dSJens Axboe unsigned int *offset, unsigned int *len, 168135f3d14dSJens Axboe struct splice_pipe_desc *spd, struct sock *sk) 16822870c43dSOctavian Purdila { 16832870c43dSOctavian Purdila int seg; 16849c55e01cSJens Axboe 16851d0c0b32SEric Dumazet /* map the linear part : 16862996d31fSAlexander Duyck * If skb->head_frag is set, this 'linear' part is backed by a 16872996d31fSAlexander Duyck * fragment, and if the head is not shared with any clones then 16882996d31fSAlexander Duyck * we can avoid a copy since we own the head portion of this page. 16899c55e01cSJens Axboe */ 16902870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 16912870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 16922870c43dSOctavian Purdila skb_headlen(skb), 16931d0c0b32SEric Dumazet offset, len, skb, spd, 16943a7c1ee4SAlexander Duyck skb_head_is_locked(skb), 16951d0c0b32SEric Dumazet sk, pipe)) 1696a108d5f3SDavid S. Miller return true; 16979c55e01cSJens Axboe 16989c55e01cSJens Axboe /* 16999c55e01cSJens Axboe * then map the fragments 17009c55e01cSJens Axboe */ 17019c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 17029c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 17039c55e01cSJens Axboe 1704ea2ab693SIan Campbell if (__splice_segment(skb_frag_page(f), 17059e903e08SEric Dumazet f->page_offset, skb_frag_size(f), 1706d7ccf7c0SEric Dumazet offset, len, skb, spd, false, sk, pipe)) 1707a108d5f3SDavid S. Miller return true; 17089c55e01cSJens Axboe } 17099c55e01cSJens Axboe 1710a108d5f3SDavid S. Miller return false; 17119c55e01cSJens Axboe } 17129c55e01cSJens Axboe 17139c55e01cSJens Axboe /* 17149c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 17159c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 17169c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 17179c55e01cSJens Axboe * handle that cleanly. 17189c55e01cSJens Axboe */ 17198b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 17209c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 17219c55e01cSJens Axboe unsigned int flags) 17229c55e01cSJens Axboe { 172341c73a0dSEric Dumazet struct partial_page partial[MAX_SKB_FRAGS]; 172441c73a0dSEric Dumazet struct page *pages[MAX_SKB_FRAGS]; 17259c55e01cSJens Axboe struct splice_pipe_desc spd = { 17269c55e01cSJens Axboe .pages = pages, 17279c55e01cSJens Axboe .partial = partial, 17289c55e01cSJens Axboe .flags = flags, 17299c55e01cSJens Axboe .ops = &sock_pipe_buf_ops, 17309c55e01cSJens Axboe .spd_release = sock_spd_release, 17319c55e01cSJens Axboe }; 1732fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 17337a67e56fSJarek Poplawski struct sock *sk = skb->sk; 173435f3d14dSJens Axboe int ret = 0; 173535f3d14dSJens Axboe 17369c55e01cSJens Axboe /* 17379c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 17389c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 17399c55e01cSJens Axboe */ 174035f3d14dSJens Axboe if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 17419c55e01cSJens Axboe goto done; 17429c55e01cSJens Axboe else if (!tlen) 17439c55e01cSJens Axboe goto done; 17449c55e01cSJens Axboe 17459c55e01cSJens Axboe /* 17469c55e01cSJens Axboe * now see if we have a frag_list to map 17479c55e01cSJens Axboe */ 1748fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 1749fbb398a8SDavid S. Miller if (!tlen) 17509c55e01cSJens Axboe break; 175135f3d14dSJens Axboe if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1752fbb398a8SDavid S. Miller break; 17539c55e01cSJens Axboe } 17549c55e01cSJens Axboe 17559c55e01cSJens Axboe done: 17569c55e01cSJens Axboe if (spd.nr_pages) { 17579c55e01cSJens Axboe /* 17589c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 17599c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 17609c55e01cSJens Axboe * here as compared to sendfile(). We enter here 17619c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 17629c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 17639c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 17649c55e01cSJens Axboe * and networking will grab the socket lock. 17659c55e01cSJens Axboe */ 1766293ad604SOctavian Purdila release_sock(sk); 17679c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1768293ad604SOctavian Purdila lock_sock(sk); 17699c55e01cSJens Axboe } 17709c55e01cSJens Axboe 177135f3d14dSJens Axboe return ret; 17729c55e01cSJens Axboe } 17739c55e01cSJens Axboe 1774357b40a1SHerbert Xu /** 1775357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1776357b40a1SHerbert Xu * @skb: destination buffer 1777357b40a1SHerbert Xu * @offset: offset in destination 1778357b40a1SHerbert Xu * @from: source buffer 1779357b40a1SHerbert Xu * @len: number of bytes to copy 1780357b40a1SHerbert Xu * 1781357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1782357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1783357b40a1SHerbert Xu * traversing fragment lists and such. 1784357b40a1SHerbert Xu */ 1785357b40a1SHerbert Xu 17860c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1787357b40a1SHerbert Xu { 17881a028e50SDavid S. Miller int start = skb_headlen(skb); 1789fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 1790fbb398a8SDavid S. Miller int i, copy; 1791357b40a1SHerbert Xu 1792357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1793357b40a1SHerbert Xu goto fault; 1794357b40a1SHerbert Xu 17951a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1796357b40a1SHerbert Xu if (copy > len) 1797357b40a1SHerbert Xu copy = len; 179827d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1799357b40a1SHerbert Xu if ((len -= copy) == 0) 1800357b40a1SHerbert Xu return 0; 1801357b40a1SHerbert Xu offset += copy; 1802357b40a1SHerbert Xu from += copy; 1803357b40a1SHerbert Xu } 1804357b40a1SHerbert Xu 1805357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1806357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 18071a028e50SDavid S. Miller int end; 1808357b40a1SHerbert Xu 1809547b792cSIlpo Järvinen WARN_ON(start > offset + len); 18101a028e50SDavid S. Miller 18119e903e08SEric Dumazet end = start + skb_frag_size(frag); 1812357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1813357b40a1SHerbert Xu u8 *vaddr; 1814357b40a1SHerbert Xu 1815357b40a1SHerbert Xu if (copy > len) 1816357b40a1SHerbert Xu copy = len; 1817357b40a1SHerbert Xu 181851c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 18191a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 18201a028e50SDavid S. Miller from, copy); 182151c56b00SEric Dumazet kunmap_atomic(vaddr); 1822357b40a1SHerbert Xu 1823357b40a1SHerbert Xu if ((len -= copy) == 0) 1824357b40a1SHerbert Xu return 0; 1825357b40a1SHerbert Xu offset += copy; 1826357b40a1SHerbert Xu from += copy; 1827357b40a1SHerbert Xu } 18281a028e50SDavid S. Miller start = end; 1829357b40a1SHerbert Xu } 1830357b40a1SHerbert Xu 1831fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 18321a028e50SDavid S. Miller int end; 1833357b40a1SHerbert Xu 1834547b792cSIlpo Järvinen WARN_ON(start > offset + len); 18351a028e50SDavid S. Miller 1836fbb398a8SDavid S. Miller end = start + frag_iter->len; 1837357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1838357b40a1SHerbert Xu if (copy > len) 1839357b40a1SHerbert Xu copy = len; 1840fbb398a8SDavid S. Miller if (skb_store_bits(frag_iter, offset - start, 18411a028e50SDavid S. Miller from, copy)) 1842357b40a1SHerbert Xu goto fault; 1843357b40a1SHerbert Xu if ((len -= copy) == 0) 1844357b40a1SHerbert Xu return 0; 1845357b40a1SHerbert Xu offset += copy; 1846357b40a1SHerbert Xu from += copy; 1847357b40a1SHerbert Xu } 18481a028e50SDavid S. Miller start = end; 1849357b40a1SHerbert Xu } 1850357b40a1SHerbert Xu if (!len) 1851357b40a1SHerbert Xu return 0; 1852357b40a1SHerbert Xu 1853357b40a1SHerbert Xu fault: 1854357b40a1SHerbert Xu return -EFAULT; 1855357b40a1SHerbert Xu } 1856357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1857357b40a1SHerbert Xu 18581da177e4SLinus Torvalds /* Checksum skb data. */ 18591da177e4SLinus Torvalds 18602bbbc868SAl Viro __wsum skb_checksum(const struct sk_buff *skb, int offset, 18612bbbc868SAl Viro int len, __wsum csum) 18621da177e4SLinus Torvalds { 18631a028e50SDavid S. Miller int start = skb_headlen(skb); 18641a028e50SDavid S. Miller int i, copy = start - offset; 1865fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 18661da177e4SLinus Torvalds int pos = 0; 18671da177e4SLinus Torvalds 18681da177e4SLinus Torvalds /* Checksum header. */ 18691da177e4SLinus Torvalds if (copy > 0) { 18701da177e4SLinus Torvalds if (copy > len) 18711da177e4SLinus Torvalds copy = len; 18721da177e4SLinus Torvalds csum = csum_partial(skb->data + offset, copy, csum); 18731da177e4SLinus Torvalds if ((len -= copy) == 0) 18741da177e4SLinus Torvalds return csum; 18751da177e4SLinus Torvalds offset += copy; 18761da177e4SLinus Torvalds pos = copy; 18771da177e4SLinus Torvalds } 18781da177e4SLinus Torvalds 18791da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 18801a028e50SDavid S. Miller int end; 188151c56b00SEric Dumazet skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 18821da177e4SLinus Torvalds 1883547b792cSIlpo Järvinen WARN_ON(start > offset + len); 18841a028e50SDavid S. Miller 188551c56b00SEric Dumazet end = start + skb_frag_size(frag); 18861da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 188744bb9363SAl Viro __wsum csum2; 18881da177e4SLinus Torvalds u8 *vaddr; 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds if (copy > len) 18911da177e4SLinus Torvalds copy = len; 189251c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 18931a028e50SDavid S. Miller csum2 = csum_partial(vaddr + frag->page_offset + 18941a028e50SDavid S. Miller offset - start, copy, 0); 189551c56b00SEric Dumazet kunmap_atomic(vaddr); 18961da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 18971da177e4SLinus Torvalds if (!(len -= copy)) 18981da177e4SLinus Torvalds return csum; 18991da177e4SLinus Torvalds offset += copy; 19001da177e4SLinus Torvalds pos += copy; 19011da177e4SLinus Torvalds } 19021a028e50SDavid S. Miller start = end; 19031da177e4SLinus Torvalds } 19041da177e4SLinus Torvalds 1905fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 19061a028e50SDavid S. Miller int end; 19071da177e4SLinus Torvalds 1908547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19091a028e50SDavid S. Miller 1910fbb398a8SDavid S. Miller end = start + frag_iter->len; 19111da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 19125f92a738SAl Viro __wsum csum2; 19131da177e4SLinus Torvalds if (copy > len) 19141da177e4SLinus Torvalds copy = len; 1915fbb398a8SDavid S. Miller csum2 = skb_checksum(frag_iter, offset - start, 19161a028e50SDavid S. Miller copy, 0); 19171da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 19181da177e4SLinus Torvalds if ((len -= copy) == 0) 19191da177e4SLinus Torvalds return csum; 19201da177e4SLinus Torvalds offset += copy; 19211da177e4SLinus Torvalds pos += copy; 19221da177e4SLinus Torvalds } 19231a028e50SDavid S. Miller start = end; 19241da177e4SLinus Torvalds } 192509a62660SKris Katterjohn BUG_ON(len); 19261da177e4SLinus Torvalds 19271da177e4SLinus Torvalds return csum; 19281da177e4SLinus Torvalds } 1929b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds /* Both of above in one bottle. */ 19321da177e4SLinus Torvalds 193381d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 193481d77662SAl Viro u8 *to, int len, __wsum csum) 19351da177e4SLinus Torvalds { 19361a028e50SDavid S. Miller int start = skb_headlen(skb); 19371a028e50SDavid S. Miller int i, copy = start - offset; 1938fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 19391da177e4SLinus Torvalds int pos = 0; 19401da177e4SLinus Torvalds 19411da177e4SLinus Torvalds /* Copy header. */ 19421da177e4SLinus Torvalds if (copy > 0) { 19431da177e4SLinus Torvalds if (copy > len) 19441da177e4SLinus Torvalds copy = len; 19451da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 19461da177e4SLinus Torvalds copy, csum); 19471da177e4SLinus Torvalds if ((len -= copy) == 0) 19481da177e4SLinus Torvalds return csum; 19491da177e4SLinus Torvalds offset += copy; 19501da177e4SLinus Torvalds to += copy; 19511da177e4SLinus Torvalds pos = copy; 19521da177e4SLinus Torvalds } 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 19551a028e50SDavid S. Miller int end; 19561da177e4SLinus Torvalds 1957547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19581a028e50SDavid S. Miller 19599e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 19601da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 19615084205fSAl Viro __wsum csum2; 19621da177e4SLinus Torvalds u8 *vaddr; 19631da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 19641da177e4SLinus Torvalds 19651da177e4SLinus Torvalds if (copy > len) 19661da177e4SLinus Torvalds copy = len; 196751c56b00SEric Dumazet vaddr = kmap_atomic(skb_frag_page(frag)); 19681da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 19691a028e50SDavid S. Miller frag->page_offset + 19701a028e50SDavid S. Miller offset - start, to, 19711a028e50SDavid S. Miller copy, 0); 197251c56b00SEric Dumazet kunmap_atomic(vaddr); 19731da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 19741da177e4SLinus Torvalds if (!(len -= copy)) 19751da177e4SLinus Torvalds return csum; 19761da177e4SLinus Torvalds offset += copy; 19771da177e4SLinus Torvalds to += copy; 19781da177e4SLinus Torvalds pos += copy; 19791da177e4SLinus Torvalds } 19801a028e50SDavid S. Miller start = end; 19811da177e4SLinus Torvalds } 19821da177e4SLinus Torvalds 1983fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 198481d77662SAl Viro __wsum csum2; 19851a028e50SDavid S. Miller int end; 19861da177e4SLinus Torvalds 1987547b792cSIlpo Järvinen WARN_ON(start > offset + len); 19881a028e50SDavid S. Miller 1989fbb398a8SDavid S. Miller end = start + frag_iter->len; 19901da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 19911da177e4SLinus Torvalds if (copy > len) 19921da177e4SLinus Torvalds copy = len; 1993fbb398a8SDavid S. Miller csum2 = skb_copy_and_csum_bits(frag_iter, 19941a028e50SDavid S. Miller offset - start, 19951da177e4SLinus Torvalds to, copy, 0); 19961da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 19971da177e4SLinus Torvalds if ((len -= copy) == 0) 19981da177e4SLinus Torvalds return csum; 19991da177e4SLinus Torvalds offset += copy; 20001da177e4SLinus Torvalds to += copy; 20011da177e4SLinus Torvalds pos += copy; 20021da177e4SLinus Torvalds } 20031a028e50SDavid S. Miller start = end; 20041da177e4SLinus Torvalds } 200509a62660SKris Katterjohn BUG_ON(len); 20061da177e4SLinus Torvalds return csum; 20071da177e4SLinus Torvalds } 2008b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 20091da177e4SLinus Torvalds 20101da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 20111da177e4SLinus Torvalds { 2012d3bc23e7SAl Viro __wsum csum; 20131da177e4SLinus Torvalds long csstart; 20141da177e4SLinus Torvalds 201584fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 201655508d60SMichał Mirosław csstart = skb_checksum_start_offset(skb); 20171da177e4SLinus Torvalds else 20181da177e4SLinus Torvalds csstart = skb_headlen(skb); 20191da177e4SLinus Torvalds 202009a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 20211da177e4SLinus Torvalds 2022d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 20231da177e4SLinus Torvalds 20241da177e4SLinus Torvalds csum = 0; 20251da177e4SLinus Torvalds if (csstart != skb->len) 20261da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 20271da177e4SLinus Torvalds skb->len - csstart, 0); 20281da177e4SLinus Torvalds 202984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 2030ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 20311da177e4SLinus Torvalds 2032d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 20331da177e4SLinus Torvalds } 20341da177e4SLinus Torvalds } 2035b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 20361da177e4SLinus Torvalds 20371da177e4SLinus Torvalds /** 20381da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 20391da177e4SLinus Torvalds * @list: list to dequeue from 20401da177e4SLinus Torvalds * 20411da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 20421da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 20431da177e4SLinus Torvalds * returned or %NULL if the list is empty. 20441da177e4SLinus Torvalds */ 20451da177e4SLinus Torvalds 20461da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 20471da177e4SLinus Torvalds { 20481da177e4SLinus Torvalds unsigned long flags; 20491da177e4SLinus Torvalds struct sk_buff *result; 20501da177e4SLinus Torvalds 20511da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 20521da177e4SLinus Torvalds result = __skb_dequeue(list); 20531da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 20541da177e4SLinus Torvalds return result; 20551da177e4SLinus Torvalds } 2056b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 20571da177e4SLinus Torvalds 20581da177e4SLinus Torvalds /** 20591da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 20601da177e4SLinus Torvalds * @list: list to dequeue from 20611da177e4SLinus Torvalds * 20621da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 20631da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 20641da177e4SLinus Torvalds * returned or %NULL if the list is empty. 20651da177e4SLinus Torvalds */ 20661da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 20671da177e4SLinus Torvalds { 20681da177e4SLinus Torvalds unsigned long flags; 20691da177e4SLinus Torvalds struct sk_buff *result; 20701da177e4SLinus Torvalds 20711da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 20721da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 20731da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 20741da177e4SLinus Torvalds return result; 20751da177e4SLinus Torvalds } 2076b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 20771da177e4SLinus Torvalds 20781da177e4SLinus Torvalds /** 20791da177e4SLinus Torvalds * skb_queue_purge - empty a list 20801da177e4SLinus Torvalds * @list: list to empty 20811da177e4SLinus Torvalds * 20821da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 20831da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 20841da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 20851da177e4SLinus Torvalds */ 20861da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 20871da177e4SLinus Torvalds { 20881da177e4SLinus Torvalds struct sk_buff *skb; 20891da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 20901da177e4SLinus Torvalds kfree_skb(skb); 20911da177e4SLinus Torvalds } 2092b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 20931da177e4SLinus Torvalds 20941da177e4SLinus Torvalds /** 20951da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 20961da177e4SLinus Torvalds * @list: list to use 20971da177e4SLinus Torvalds * @newsk: buffer to queue 20981da177e4SLinus Torvalds * 20991da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 21001da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 21011da177e4SLinus Torvalds * safely. 21021da177e4SLinus Torvalds * 21031da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 21041da177e4SLinus Torvalds */ 21051da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 21061da177e4SLinus Torvalds { 21071da177e4SLinus Torvalds unsigned long flags; 21081da177e4SLinus Torvalds 21091da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 21101da177e4SLinus Torvalds __skb_queue_head(list, newsk); 21111da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 21121da177e4SLinus Torvalds } 2113b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 21141da177e4SLinus Torvalds 21151da177e4SLinus Torvalds /** 21161da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 21171da177e4SLinus Torvalds * @list: list to use 21181da177e4SLinus Torvalds * @newsk: buffer to queue 21191da177e4SLinus Torvalds * 21201da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 21211da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 21221da177e4SLinus Torvalds * safely. 21231da177e4SLinus Torvalds * 21241da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 21251da177e4SLinus Torvalds */ 21261da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 21271da177e4SLinus Torvalds { 21281da177e4SLinus Torvalds unsigned long flags; 21291da177e4SLinus Torvalds 21301da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 21311da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 21321da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 21331da177e4SLinus Torvalds } 2134b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 21358728b834SDavid S. Miller 21361da177e4SLinus Torvalds /** 21371da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 21381da177e4SLinus Torvalds * @skb: buffer to remove 21398728b834SDavid S. Miller * @list: list to use 21401da177e4SLinus Torvalds * 21418728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 21428728b834SDavid S. Miller * function is atomic with respect to other list locked calls 21431da177e4SLinus Torvalds * 21448728b834SDavid S. Miller * You must know what list the SKB is on. 21451da177e4SLinus Torvalds */ 21468728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 21471da177e4SLinus Torvalds { 21481da177e4SLinus Torvalds unsigned long flags; 21491da177e4SLinus Torvalds 21501da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 21518728b834SDavid S. Miller __skb_unlink(skb, list); 21521da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 21531da177e4SLinus Torvalds } 2154b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 21551da177e4SLinus Torvalds 21561da177e4SLinus Torvalds /** 21571da177e4SLinus Torvalds * skb_append - append a buffer 21581da177e4SLinus Torvalds * @old: buffer to insert after 21591da177e4SLinus Torvalds * @newsk: buffer to insert 21608728b834SDavid S. Miller * @list: list to use 21611da177e4SLinus Torvalds * 21621da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 21631da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 21641da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 21651da177e4SLinus Torvalds */ 21668728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 21671da177e4SLinus Torvalds { 21681da177e4SLinus Torvalds unsigned long flags; 21691da177e4SLinus Torvalds 21708728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 21717de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 21728728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 21731da177e4SLinus Torvalds } 2174b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 21751da177e4SLinus Torvalds 21761da177e4SLinus Torvalds /** 21771da177e4SLinus Torvalds * skb_insert - insert a buffer 21781da177e4SLinus Torvalds * @old: buffer to insert before 21791da177e4SLinus Torvalds * @newsk: buffer to insert 21808728b834SDavid S. Miller * @list: list to use 21811da177e4SLinus Torvalds * 21828728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 21838728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 21848728b834SDavid S. Miller * calls. 21858728b834SDavid S. Miller * 21861da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 21871da177e4SLinus Torvalds */ 21888728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 21891da177e4SLinus Torvalds { 21901da177e4SLinus Torvalds unsigned long flags; 21911da177e4SLinus Torvalds 21928728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 21938728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 21948728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 21951da177e4SLinus Torvalds } 2196b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 21991da177e4SLinus Torvalds struct sk_buff* skb1, 22001da177e4SLinus Torvalds const u32 len, const int pos) 22011da177e4SLinus Torvalds { 22021da177e4SLinus Torvalds int i; 22031da177e4SLinus Torvalds 2204d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2205d626f62bSArnaldo Carvalho de Melo pos - len); 22061da177e4SLinus Torvalds /* And move data appendix as is. */ 22071da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 22081da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 22091da177e4SLinus Torvalds 22101da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 22111da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 22121da177e4SLinus Torvalds skb1->data_len = skb->data_len; 22131da177e4SLinus Torvalds skb1->len += skb1->data_len; 22141da177e4SLinus Torvalds skb->data_len = 0; 22151da177e4SLinus Torvalds skb->len = len; 221627a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 22171da177e4SLinus Torvalds } 22181da177e4SLinus Torvalds 22191da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 22201da177e4SLinus Torvalds struct sk_buff* skb1, 22211da177e4SLinus Torvalds const u32 len, int pos) 22221da177e4SLinus Torvalds { 22231da177e4SLinus Torvalds int i, k = 0; 22241da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 22251da177e4SLinus Torvalds 22261da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 22271da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 22281da177e4SLinus Torvalds skb->len = len; 22291da177e4SLinus Torvalds skb->data_len = len - pos; 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 22329e903e08SEric Dumazet int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 22331da177e4SLinus Torvalds 22341da177e4SLinus Torvalds if (pos + size > len) { 22351da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 22361da177e4SLinus Torvalds 22371da177e4SLinus Torvalds if (pos < len) { 22381da177e4SLinus Torvalds /* Split frag. 22391da177e4SLinus Torvalds * We have two variants in this case: 22401da177e4SLinus Torvalds * 1. Move all the frag to the second 22411da177e4SLinus Torvalds * part, if it is possible. F.e. 22421da177e4SLinus Torvalds * this approach is mandatory for TUX, 22431da177e4SLinus Torvalds * where splitting is expensive. 22441da177e4SLinus Torvalds * 2. Split is accurately. We make this. 22451da177e4SLinus Torvalds */ 2246ea2ab693SIan Campbell skb_frag_ref(skb, i); 22471da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 22489e903e08SEric Dumazet skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 22499e903e08SEric Dumazet skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 22501da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 22511da177e4SLinus Torvalds } 22521da177e4SLinus Torvalds k++; 22531da177e4SLinus Torvalds } else 22541da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 22551da177e4SLinus Torvalds pos += size; 22561da177e4SLinus Torvalds } 22571da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 22581da177e4SLinus Torvalds } 22591da177e4SLinus Torvalds 22601da177e4SLinus Torvalds /** 22611da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 22621da177e4SLinus Torvalds * @skb: the buffer to split 22631da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 22641da177e4SLinus Torvalds * @len: new length for skb 22651da177e4SLinus Torvalds */ 22661da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 22671da177e4SLinus Torvalds { 22681da177e4SLinus Torvalds int pos = skb_headlen(skb); 22691da177e4SLinus Torvalds 22701da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 22711da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 22721da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 22731da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 22741da177e4SLinus Torvalds } 2275b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 22761da177e4SLinus Torvalds 22779f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 22789f782db3SIlpo Järvinen * 22799f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 22809f782db3SIlpo Järvinen */ 2281832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2282832d11c5SIlpo Järvinen { 22830ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2284832d11c5SIlpo Järvinen } 2285832d11c5SIlpo Järvinen 2286832d11c5SIlpo Järvinen /** 2287832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2288832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2289832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2290832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2291832d11c5SIlpo Järvinen * 2292832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 229320e994a0SFeng King * the length of the skb, from skb to tgt. Returns number bytes shifted. 2294832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2295832d11c5SIlpo Järvinen * 2296832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2297832d11c5SIlpo Järvinen * 2298832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2299832d11c5SIlpo Järvinen * to have non-paged data as well. 2300832d11c5SIlpo Järvinen * 2301832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2302832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2303832d11c5SIlpo Järvinen */ 2304832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2305832d11c5SIlpo Järvinen { 2306832d11c5SIlpo Järvinen int from, to, merge, todo; 2307832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2308832d11c5SIlpo Järvinen 2309832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2310832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2311832d11c5SIlpo Järvinen 2312832d11c5SIlpo Järvinen todo = shiftlen; 2313832d11c5SIlpo Järvinen from = 0; 2314832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2315832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2316832d11c5SIlpo Järvinen 2317832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2318832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2319832d11c5SIlpo Järvinen */ 2320832d11c5SIlpo Järvinen if (!to || 2321ea2ab693SIan Campbell !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2322ea2ab693SIan Campbell fragfrom->page_offset)) { 2323832d11c5SIlpo Järvinen merge = -1; 2324832d11c5SIlpo Järvinen } else { 2325832d11c5SIlpo Järvinen merge = to - 1; 2326832d11c5SIlpo Järvinen 23279e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2328832d11c5SIlpo Järvinen if (todo < 0) { 2329832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2330832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2331832d11c5SIlpo Järvinen return 0; 2332832d11c5SIlpo Järvinen 23339f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 23349f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2335832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2336832d11c5SIlpo Järvinen 23379e903e08SEric Dumazet skb_frag_size_add(fragto, shiftlen); 23389e903e08SEric Dumazet skb_frag_size_sub(fragfrom, shiftlen); 2339832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2340832d11c5SIlpo Järvinen 2341832d11c5SIlpo Järvinen goto onlymerged; 2342832d11c5SIlpo Järvinen } 2343832d11c5SIlpo Järvinen 2344832d11c5SIlpo Järvinen from++; 2345832d11c5SIlpo Järvinen } 2346832d11c5SIlpo Järvinen 2347832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2348832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2349832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2350832d11c5SIlpo Järvinen return 0; 2351832d11c5SIlpo Järvinen 2352832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2353832d11c5SIlpo Järvinen return 0; 2354832d11c5SIlpo Järvinen 2355832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2356832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2357832d11c5SIlpo Järvinen return 0; 2358832d11c5SIlpo Järvinen 2359832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2360832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2361832d11c5SIlpo Järvinen 23629e903e08SEric Dumazet if (todo >= skb_frag_size(fragfrom)) { 2363832d11c5SIlpo Järvinen *fragto = *fragfrom; 23649e903e08SEric Dumazet todo -= skb_frag_size(fragfrom); 2365832d11c5SIlpo Järvinen from++; 2366832d11c5SIlpo Järvinen to++; 2367832d11c5SIlpo Järvinen 2368832d11c5SIlpo Järvinen } else { 2369ea2ab693SIan Campbell __skb_frag_ref(fragfrom); 2370832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2371832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 23729e903e08SEric Dumazet skb_frag_size_set(fragto, todo); 2373832d11c5SIlpo Järvinen 2374832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 23759e903e08SEric Dumazet skb_frag_size_sub(fragfrom, todo); 2376832d11c5SIlpo Järvinen todo = 0; 2377832d11c5SIlpo Järvinen 2378832d11c5SIlpo Järvinen to++; 2379832d11c5SIlpo Järvinen break; 2380832d11c5SIlpo Järvinen } 2381832d11c5SIlpo Järvinen } 2382832d11c5SIlpo Järvinen 2383832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2384832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2385832d11c5SIlpo Järvinen 2386832d11c5SIlpo Järvinen if (merge >= 0) { 2387832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2388832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2389832d11c5SIlpo Järvinen 23909e903e08SEric Dumazet skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2391ea2ab693SIan Campbell __skb_frag_unref(fragfrom); 2392832d11c5SIlpo Järvinen } 2393832d11c5SIlpo Järvinen 2394832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2395832d11c5SIlpo Järvinen to = 0; 2396832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2397832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2398832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2399832d11c5SIlpo Järvinen 2400832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2401832d11c5SIlpo Järvinen 2402832d11c5SIlpo Järvinen onlymerged: 2403832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2404832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2405832d11c5SIlpo Järvinen */ 2406832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2407832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2408832d11c5SIlpo Järvinen 2409832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2410832d11c5SIlpo Järvinen skb->len -= shiftlen; 2411832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2412832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2413832d11c5SIlpo Järvinen tgt->len += shiftlen; 2414832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2415832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2416832d11c5SIlpo Järvinen 2417832d11c5SIlpo Järvinen return shiftlen; 2418832d11c5SIlpo Järvinen } 2419832d11c5SIlpo Järvinen 2420677e90edSThomas Graf /** 2421677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2422677e90edSThomas Graf * @skb: the buffer to read 2423677e90edSThomas Graf * @from: lower offset of data to be read 2424677e90edSThomas Graf * @to: upper offset of data to be read 2425677e90edSThomas Graf * @st: state variable 2426677e90edSThomas Graf * 2427677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2428677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2429677e90edSThomas Graf */ 2430677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2431677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2432677e90edSThomas Graf { 2433677e90edSThomas Graf st->lower_offset = from; 2434677e90edSThomas Graf st->upper_offset = to; 2435677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2436677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2437677e90edSThomas Graf st->frag_data = NULL; 2438677e90edSThomas Graf } 2439b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2440677e90edSThomas Graf 2441677e90edSThomas Graf /** 2442677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2443677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2444677e90edSThomas Graf * @data: destination pointer for data to be returned 2445677e90edSThomas Graf * @st: state variable 2446677e90edSThomas Graf * 2447677e90edSThomas Graf * Reads a block of skb data at &consumed relative to the 2448677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2449677e90edSThomas Graf * the head of the data block to &data and returns the length 2450677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2451677e90edSThomas Graf * offset has been reached. 2452677e90edSThomas Graf * 2453677e90edSThomas Graf * The caller is not required to consume all of the data 2454677e90edSThomas Graf * returned, i.e. &consumed is typically set to the number 2455677e90edSThomas Graf * of bytes already consumed and the next call to 2456677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2457677e90edSThomas Graf * 245825985edcSLucas De Marchi * Note 1: The size of each block of data returned can be arbitrary, 2459677e90edSThomas Graf * this limitation is the cost for zerocopy seqeuental 2460677e90edSThomas Graf * reads of potentially non linear data. 2461677e90edSThomas Graf * 2462bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2463677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2464677e90edSThomas Graf * a stack for this purpose. 2465677e90edSThomas Graf */ 2466677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2467677e90edSThomas Graf struct skb_seq_state *st) 2468677e90edSThomas Graf { 2469677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2470677e90edSThomas Graf skb_frag_t *frag; 2471677e90edSThomas Graf 2472677e90edSThomas Graf if (unlikely(abs_offset >= st->upper_offset)) 2473677e90edSThomas Graf return 0; 2474677e90edSThomas Graf 2475677e90edSThomas Graf next_skb: 247695e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2477677e90edSThomas Graf 2478995b3379SThomas Chenault if (abs_offset < block_limit && !st->frag_data) { 247995e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2480677e90edSThomas Graf return block_limit - abs_offset; 2481677e90edSThomas Graf } 2482677e90edSThomas Graf 2483677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2484677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2485677e90edSThomas Graf 2486677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2487677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 24889e903e08SEric Dumazet block_limit = skb_frag_size(frag) + st->stepped_offset; 2489677e90edSThomas Graf 2490677e90edSThomas Graf if (abs_offset < block_limit) { 2491677e90edSThomas Graf if (!st->frag_data) 249251c56b00SEric Dumazet st->frag_data = kmap_atomic(skb_frag_page(frag)); 2493677e90edSThomas Graf 2494677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2495677e90edSThomas Graf (abs_offset - st->stepped_offset); 2496677e90edSThomas Graf 2497677e90edSThomas Graf return block_limit - abs_offset; 2498677e90edSThomas Graf } 2499677e90edSThomas Graf 2500677e90edSThomas Graf if (st->frag_data) { 250151c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2502677e90edSThomas Graf st->frag_data = NULL; 2503677e90edSThomas Graf } 2504677e90edSThomas Graf 2505677e90edSThomas Graf st->frag_idx++; 25069e903e08SEric Dumazet st->stepped_offset += skb_frag_size(frag); 2507677e90edSThomas Graf } 2508677e90edSThomas Graf 25095b5a60daSOlaf Kirch if (st->frag_data) { 251051c56b00SEric Dumazet kunmap_atomic(st->frag_data); 25115b5a60daSOlaf Kirch st->frag_data = NULL; 25125b5a60daSOlaf Kirch } 25135b5a60daSOlaf Kirch 251421dc3301SDavid S. Miller if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2515677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 251695e3b24cSHerbert Xu st->frag_idx = 0; 2517677e90edSThomas Graf goto next_skb; 251871b3346dSShyam Iyer } else if (st->cur_skb->next) { 251971b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 252071b3346dSShyam Iyer st->frag_idx = 0; 2521677e90edSThomas Graf goto next_skb; 2522677e90edSThomas Graf } 2523677e90edSThomas Graf 2524677e90edSThomas Graf return 0; 2525677e90edSThomas Graf } 2526b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2527677e90edSThomas Graf 2528677e90edSThomas Graf /** 2529677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2530677e90edSThomas Graf * @st: state variable 2531677e90edSThomas Graf * 2532677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2533677e90edSThomas Graf * returned 0. 2534677e90edSThomas Graf */ 2535677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2536677e90edSThomas Graf { 2537677e90edSThomas Graf if (st->frag_data) 253851c56b00SEric Dumazet kunmap_atomic(st->frag_data); 2539677e90edSThomas Graf } 2540b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2541677e90edSThomas Graf 25423fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 25433fc7e8a6SThomas Graf 25443fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 25453fc7e8a6SThomas Graf struct ts_config *conf, 25463fc7e8a6SThomas Graf struct ts_state *state) 25473fc7e8a6SThomas Graf { 25483fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 25493fc7e8a6SThomas Graf } 25503fc7e8a6SThomas Graf 25513fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 25523fc7e8a6SThomas Graf { 25533fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 25543fc7e8a6SThomas Graf } 25553fc7e8a6SThomas Graf 25563fc7e8a6SThomas Graf /** 25573fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 25583fc7e8a6SThomas Graf * @skb: the buffer to look in 25593fc7e8a6SThomas Graf * @from: search offset 25603fc7e8a6SThomas Graf * @to: search limit 25613fc7e8a6SThomas Graf * @config: textsearch configuration 25623fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 25633fc7e8a6SThomas Graf * 25643fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 25653fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 25663fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 25673fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 25683fc7e8a6SThomas Graf */ 25693fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 25703fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 25713fc7e8a6SThomas Graf struct ts_state *state) 25723fc7e8a6SThomas Graf { 2573f72b948dSPhil Oester unsigned int ret; 2574f72b948dSPhil Oester 25753fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 25763fc7e8a6SThomas Graf config->finish = skb_ts_finish; 25773fc7e8a6SThomas Graf 25783fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 25793fc7e8a6SThomas Graf 2580f72b948dSPhil Oester ret = textsearch_find(config, state); 2581f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 25823fc7e8a6SThomas Graf } 2583b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 25843fc7e8a6SThomas Graf 2585e89e9cf5SAnanda Raju /** 2586e89e9cf5SAnanda Raju * skb_append_datato_frags: - append the user data to a skb 2587e89e9cf5SAnanda Raju * @sk: sock structure 2588e89e9cf5SAnanda Raju * @skb: skb structure to be appened with user data. 2589e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2590e89e9cf5SAnanda Raju * @from: pointer to user message iov 2591e89e9cf5SAnanda Raju * @length: length of the iov message 2592e89e9cf5SAnanda Raju * 2593e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2594e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2595e89e9cf5SAnanda Raju */ 2596e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2597dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2598e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2599e89e9cf5SAnanda Raju void *from, int length) 2600e89e9cf5SAnanda Raju { 2601e89e9cf5SAnanda Raju int frg_cnt = 0; 2602e89e9cf5SAnanda Raju skb_frag_t *frag = NULL; 2603e89e9cf5SAnanda Raju struct page *page = NULL; 2604e89e9cf5SAnanda Raju int copy, left; 2605e89e9cf5SAnanda Raju int offset = 0; 2606e89e9cf5SAnanda Raju int ret; 2607e89e9cf5SAnanda Raju 2608e89e9cf5SAnanda Raju do { 2609e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2610e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2611e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2612e89e9cf5SAnanda Raju return -EFAULT; 2613e89e9cf5SAnanda Raju 2614e89e9cf5SAnanda Raju /* allocate a new page for next frag */ 2615e89e9cf5SAnanda Raju page = alloc_pages(sk->sk_allocation, 0); 2616e89e9cf5SAnanda Raju 2617e89e9cf5SAnanda Raju /* If alloc_page fails just return failure and caller will 2618e89e9cf5SAnanda Raju * free previous allocated pages by doing kfree_skb() 2619e89e9cf5SAnanda Raju */ 2620e89e9cf5SAnanda Raju if (page == NULL) 2621e89e9cf5SAnanda Raju return -ENOMEM; 2622e89e9cf5SAnanda Raju 2623e89e9cf5SAnanda Raju /* initialize the next frag */ 2624e89e9cf5SAnanda Raju skb_fill_page_desc(skb, frg_cnt, page, 0, 0); 2625e89e9cf5SAnanda Raju skb->truesize += PAGE_SIZE; 2626e89e9cf5SAnanda Raju atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 2627e89e9cf5SAnanda Raju 2628e89e9cf5SAnanda Raju /* get the new initialized frag */ 2629e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2630e89e9cf5SAnanda Raju frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; 2631e89e9cf5SAnanda Raju 2632e89e9cf5SAnanda Raju /* copy the user data to page */ 2633e89e9cf5SAnanda Raju left = PAGE_SIZE - frag->page_offset; 2634e89e9cf5SAnanda Raju copy = (length > left)? left : length; 2635e89e9cf5SAnanda Raju 26369e903e08SEric Dumazet ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag), 2637e89e9cf5SAnanda Raju offset, copy, 0, skb); 2638e89e9cf5SAnanda Raju if (ret < 0) 2639e89e9cf5SAnanda Raju return -EFAULT; 2640e89e9cf5SAnanda Raju 2641e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 26429e903e08SEric Dumazet skb_frag_size_add(frag, copy); 2643e89e9cf5SAnanda Raju skb->len += copy; 2644e89e9cf5SAnanda Raju skb->data_len += copy; 2645e89e9cf5SAnanda Raju offset += copy; 2646e89e9cf5SAnanda Raju length -= copy; 2647e89e9cf5SAnanda Raju 2648e89e9cf5SAnanda Raju } while (length > 0); 2649e89e9cf5SAnanda Raju 2650e89e9cf5SAnanda Raju return 0; 2651e89e9cf5SAnanda Raju } 2652b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2653e89e9cf5SAnanda Raju 2654cbb042f9SHerbert Xu /** 2655cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2656cbb042f9SHerbert Xu * @skb: buffer to update 2657cbb042f9SHerbert Xu * @len: length of data pulled 2658cbb042f9SHerbert Xu * 2659cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2660fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 266184fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 266284fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 266384fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2664cbb042f9SHerbert Xu */ 2665cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2666cbb042f9SHerbert Xu { 2667cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2668cbb042f9SHerbert Xu skb->len -= len; 2669cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2670cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2671cbb042f9SHerbert Xu return skb->data += len; 2672cbb042f9SHerbert Xu } 2673f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2674f94691acSArnaldo Carvalho de Melo 2675f4c50d99SHerbert Xu /** 2676f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2677f4c50d99SHerbert Xu * @skb: buffer to segment 2678576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2679f4c50d99SHerbert Xu * 2680f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 26814c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 26824c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2683f4c50d99SHerbert Xu */ 2684c8f44affSMichał Mirosław struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features) 2685f4c50d99SHerbert Xu { 2686f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2687f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 268889319d38SHerbert Xu struct sk_buff *fskb = skb_shinfo(skb)->frag_list; 2689f4c50d99SHerbert Xu unsigned int mss = skb_shinfo(skb)->gso_size; 269098e399f8SArnaldo Carvalho de Melo unsigned int doffset = skb->data - skb_mac_header(skb); 2691f4c50d99SHerbert Xu unsigned int offset = doffset; 2692f4c50d99SHerbert Xu unsigned int headroom; 2693f4c50d99SHerbert Xu unsigned int len; 269404ed3e74SMichał Mirosław int sg = !!(features & NETIF_F_SG); 2695f4c50d99SHerbert Xu int nfrags = skb_shinfo(skb)->nr_frags; 2696f4c50d99SHerbert Xu int err = -ENOMEM; 2697f4c50d99SHerbert Xu int i = 0; 2698f4c50d99SHerbert Xu int pos; 2699f4c50d99SHerbert Xu 2700f4c50d99SHerbert Xu __skb_push(skb, doffset); 2701f4c50d99SHerbert Xu headroom = skb_headroom(skb); 2702f4c50d99SHerbert Xu pos = skb_headlen(skb); 2703f4c50d99SHerbert Xu 2704f4c50d99SHerbert Xu do { 2705f4c50d99SHerbert Xu struct sk_buff *nskb; 2706f4c50d99SHerbert Xu skb_frag_t *frag; 2707c8884eddSHerbert Xu int hsize; 2708f4c50d99SHerbert Xu int size; 2709f4c50d99SHerbert Xu 2710f4c50d99SHerbert Xu len = skb->len - offset; 2711f4c50d99SHerbert Xu if (len > mss) 2712f4c50d99SHerbert Xu len = mss; 2713f4c50d99SHerbert Xu 2714f4c50d99SHerbert Xu hsize = skb_headlen(skb) - offset; 2715f4c50d99SHerbert Xu if (hsize < 0) 2716f4c50d99SHerbert Xu hsize = 0; 2717c8884eddSHerbert Xu if (hsize > len || !sg) 2718c8884eddSHerbert Xu hsize = len; 2719f4c50d99SHerbert Xu 272089319d38SHerbert Xu if (!hsize && i >= nfrags) { 272189319d38SHerbert Xu BUG_ON(fskb->len != len); 272289319d38SHerbert Xu 272389319d38SHerbert Xu pos += len; 272489319d38SHerbert Xu nskb = skb_clone(fskb, GFP_ATOMIC); 272589319d38SHerbert Xu fskb = fskb->next; 272689319d38SHerbert Xu 2727f4c50d99SHerbert Xu if (unlikely(!nskb)) 2728f4c50d99SHerbert Xu goto err; 2729f4c50d99SHerbert Xu 2730ec47ea82SAlexander Duyck hsize = skb_end_offset(nskb); 273189319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 273289319d38SHerbert Xu kfree_skb(nskb); 273389319d38SHerbert Xu goto err; 273489319d38SHerbert Xu } 273589319d38SHerbert Xu 2736ec47ea82SAlexander Duyck nskb->truesize += skb_end_offset(nskb) - hsize; 273789319d38SHerbert Xu skb_release_head_state(nskb); 273889319d38SHerbert Xu __skb_push(nskb, doffset); 273989319d38SHerbert Xu } else { 274089319d38SHerbert Xu nskb = alloc_skb(hsize + doffset + headroom, 274189319d38SHerbert Xu GFP_ATOMIC); 274289319d38SHerbert Xu 274389319d38SHerbert Xu if (unlikely(!nskb)) 274489319d38SHerbert Xu goto err; 274589319d38SHerbert Xu 274689319d38SHerbert Xu skb_reserve(nskb, headroom); 274789319d38SHerbert Xu __skb_put(nskb, doffset); 274889319d38SHerbert Xu } 274989319d38SHerbert Xu 2750f4c50d99SHerbert Xu if (segs) 2751f4c50d99SHerbert Xu tail->next = nskb; 2752f4c50d99SHerbert Xu else 2753f4c50d99SHerbert Xu segs = nskb; 2754f4c50d99SHerbert Xu tail = nskb; 2755f4c50d99SHerbert Xu 27566f85a124SHerbert Xu __copy_skb_header(nskb, skb); 2757f4c50d99SHerbert Xu nskb->mac_len = skb->mac_len; 2758f4c50d99SHerbert Xu 27593d3be433SEric Dumazet /* nskb and skb might have different headroom */ 27603d3be433SEric Dumazet if (nskb->ip_summed == CHECKSUM_PARTIAL) 27613d3be433SEric Dumazet nskb->csum_start += skb_headroom(nskb) - headroom; 27623d3be433SEric Dumazet 2763459a98edSArnaldo Carvalho de Melo skb_reset_mac_header(nskb); 2764ddc7b8e3SArnaldo Carvalho de Melo skb_set_network_header(nskb, skb->mac_len); 2765b0e380b1SArnaldo Carvalho de Melo nskb->transport_header = (nskb->network_header + 2766b0e380b1SArnaldo Carvalho de Melo skb_network_header_len(skb)); 276789319d38SHerbert Xu skb_copy_from_linear_data(skb, nskb->data, doffset); 276889319d38SHerbert Xu 27692f181855SHerbert Xu if (fskb != skb_shinfo(skb)->frag_list) 277089319d38SHerbert Xu continue; 277189319d38SHerbert Xu 2772f4c50d99SHerbert Xu if (!sg) { 27736f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 2774f4c50d99SHerbert Xu nskb->csum = skb_copy_and_csum_bits(skb, offset, 2775f4c50d99SHerbert Xu skb_put(nskb, len), 2776f4c50d99SHerbert Xu len, 0); 2777f4c50d99SHerbert Xu continue; 2778f4c50d99SHerbert Xu } 2779f4c50d99SHerbert Xu 2780f4c50d99SHerbert Xu frag = skb_shinfo(nskb)->frags; 2781f4c50d99SHerbert Xu 2782d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, 2783d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 2784f4c50d99SHerbert Xu 278589319d38SHerbert Xu while (pos < offset + len && i < nfrags) { 2786f4c50d99SHerbert Xu *frag = skb_shinfo(skb)->frags[i]; 2787ea2ab693SIan Campbell __skb_frag_ref(frag); 27889e903e08SEric Dumazet size = skb_frag_size(frag); 2789f4c50d99SHerbert Xu 2790f4c50d99SHerbert Xu if (pos < offset) { 2791f4c50d99SHerbert Xu frag->page_offset += offset - pos; 27929e903e08SEric Dumazet skb_frag_size_sub(frag, offset - pos); 2793f4c50d99SHerbert Xu } 2794f4c50d99SHerbert Xu 279589319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 2796f4c50d99SHerbert Xu 2797f4c50d99SHerbert Xu if (pos + size <= offset + len) { 2798f4c50d99SHerbert Xu i++; 2799f4c50d99SHerbert Xu pos += size; 2800f4c50d99SHerbert Xu } else { 28019e903e08SEric Dumazet skb_frag_size_sub(frag, pos + size - (offset + len)); 280289319d38SHerbert Xu goto skip_fraglist; 2803f4c50d99SHerbert Xu } 2804f4c50d99SHerbert Xu 2805f4c50d99SHerbert Xu frag++; 2806f4c50d99SHerbert Xu } 2807f4c50d99SHerbert Xu 280889319d38SHerbert Xu if (pos < offset + len) { 280989319d38SHerbert Xu struct sk_buff *fskb2 = fskb; 281089319d38SHerbert Xu 281189319d38SHerbert Xu BUG_ON(pos + fskb->len != offset + len); 281289319d38SHerbert Xu 281389319d38SHerbert Xu pos += fskb->len; 281489319d38SHerbert Xu fskb = fskb->next; 281589319d38SHerbert Xu 281689319d38SHerbert Xu if (fskb2->next) { 281789319d38SHerbert Xu fskb2 = skb_clone(fskb2, GFP_ATOMIC); 281889319d38SHerbert Xu if (!fskb2) 281989319d38SHerbert Xu goto err; 282089319d38SHerbert Xu } else 282189319d38SHerbert Xu skb_get(fskb2); 282289319d38SHerbert Xu 2823fbb398a8SDavid S. Miller SKB_FRAG_ASSERT(nskb); 282489319d38SHerbert Xu skb_shinfo(nskb)->frag_list = fskb2; 282589319d38SHerbert Xu } 282689319d38SHerbert Xu 282789319d38SHerbert Xu skip_fraglist: 2828f4c50d99SHerbert Xu nskb->data_len = len - hsize; 2829f4c50d99SHerbert Xu nskb->len += nskb->data_len; 2830f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 2831f4c50d99SHerbert Xu } while ((offset += len) < skb->len); 2832f4c50d99SHerbert Xu 2833f4c50d99SHerbert Xu return segs; 2834f4c50d99SHerbert Xu 2835f4c50d99SHerbert Xu err: 2836f4c50d99SHerbert Xu while ((skb = segs)) { 2837f4c50d99SHerbert Xu segs = skb->next; 2838b08d5840SPatrick McHardy kfree_skb(skb); 2839f4c50d99SHerbert Xu } 2840f4c50d99SHerbert Xu return ERR_PTR(err); 2841f4c50d99SHerbert Xu } 2842f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 2843f4c50d99SHerbert Xu 284471d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 284571d93b39SHerbert Xu { 284671d93b39SHerbert Xu struct sk_buff *p = *head; 284771d93b39SHerbert Xu struct sk_buff *nskb; 28489aaa156cSHerbert Xu struct skb_shared_info *skbinfo = skb_shinfo(skb); 28499aaa156cSHerbert Xu struct skb_shared_info *pinfo = skb_shinfo(p); 285071d93b39SHerbert Xu unsigned int headroom; 285186911732SHerbert Xu unsigned int len = skb_gro_len(skb); 285267147ba9SHerbert Xu unsigned int offset = skb_gro_offset(skb); 285367147ba9SHerbert Xu unsigned int headlen = skb_headlen(skb); 2854715dc1f3SEric Dumazet unsigned int delta_truesize; 285571d93b39SHerbert Xu 285686911732SHerbert Xu if (p->len + len >= 65536) 285771d93b39SHerbert Xu return -E2BIG; 285871d93b39SHerbert Xu 28599aaa156cSHerbert Xu if (pinfo->frag_list) 286071d93b39SHerbert Xu goto merge; 286167147ba9SHerbert Xu else if (headlen <= offset) { 286242da6994SHerbert Xu skb_frag_t *frag; 286366e92fcfSHerbert Xu skb_frag_t *frag2; 28649aaa156cSHerbert Xu int i = skbinfo->nr_frags; 28659aaa156cSHerbert Xu int nr_frags = pinfo->nr_frags + i; 286642da6994SHerbert Xu 286766e92fcfSHerbert Xu offset -= headlen; 286866e92fcfSHerbert Xu 286966e92fcfSHerbert Xu if (nr_frags > MAX_SKB_FRAGS) 287081705ad1SHerbert Xu return -E2BIG; 287181705ad1SHerbert Xu 28729aaa156cSHerbert Xu pinfo->nr_frags = nr_frags; 28739aaa156cSHerbert Xu skbinfo->nr_frags = 0; 2874f5572068SHerbert Xu 28759aaa156cSHerbert Xu frag = pinfo->frags + nr_frags; 28769aaa156cSHerbert Xu frag2 = skbinfo->frags + i; 287766e92fcfSHerbert Xu do { 287866e92fcfSHerbert Xu *--frag = *--frag2; 287966e92fcfSHerbert Xu } while (--i); 288066e92fcfSHerbert Xu 288166e92fcfSHerbert Xu frag->page_offset += offset; 28829e903e08SEric Dumazet skb_frag_size_sub(frag, offset); 288366e92fcfSHerbert Xu 2884715dc1f3SEric Dumazet /* all fragments truesize : remove (head size + sk_buff) */ 2885ec47ea82SAlexander Duyck delta_truesize = skb->truesize - 2886ec47ea82SAlexander Duyck SKB_TRUESIZE(skb_end_offset(skb)); 2887715dc1f3SEric Dumazet 2888f5572068SHerbert Xu skb->truesize -= skb->data_len; 2889f5572068SHerbert Xu skb->len -= skb->data_len; 2890f5572068SHerbert Xu skb->data_len = 0; 2891f5572068SHerbert Xu 2892715dc1f3SEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 28935d38a079SHerbert Xu goto done; 2894d7e8883cSEric Dumazet } else if (skb->head_frag) { 2895d7e8883cSEric Dumazet int nr_frags = pinfo->nr_frags; 2896d7e8883cSEric Dumazet skb_frag_t *frag = pinfo->frags + nr_frags; 2897d7e8883cSEric Dumazet struct page *page = virt_to_head_page(skb->head); 2898d7e8883cSEric Dumazet unsigned int first_size = headlen - offset; 2899d7e8883cSEric Dumazet unsigned int first_offset; 2900d7e8883cSEric Dumazet 2901d7e8883cSEric Dumazet if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 2902d7e8883cSEric Dumazet return -E2BIG; 2903d7e8883cSEric Dumazet 2904d7e8883cSEric Dumazet first_offset = skb->data - 2905d7e8883cSEric Dumazet (unsigned char *)page_address(page) + 2906d7e8883cSEric Dumazet offset; 2907d7e8883cSEric Dumazet 2908d7e8883cSEric Dumazet pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 2909d7e8883cSEric Dumazet 2910d7e8883cSEric Dumazet frag->page.p = page; 2911d7e8883cSEric Dumazet frag->page_offset = first_offset; 2912d7e8883cSEric Dumazet skb_frag_size_set(frag, first_size); 2913d7e8883cSEric Dumazet 2914d7e8883cSEric Dumazet memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 2915d7e8883cSEric Dumazet /* We dont need to clear skbinfo->nr_frags here */ 2916d7e8883cSEric Dumazet 2917715dc1f3SEric Dumazet delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 2918d7e8883cSEric Dumazet NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 2919d7e8883cSEric Dumazet goto done; 292069c0cab1SHerbert Xu } else if (skb_gro_len(p) != pinfo->gso_size) 292169c0cab1SHerbert Xu return -E2BIG; 292271d93b39SHerbert Xu 292371d93b39SHerbert Xu headroom = skb_headroom(p); 29243d3be433SEric Dumazet nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); 292571d93b39SHerbert Xu if (unlikely(!nskb)) 292671d93b39SHerbert Xu return -ENOMEM; 292771d93b39SHerbert Xu 292871d93b39SHerbert Xu __copy_skb_header(nskb, p); 292971d93b39SHerbert Xu nskb->mac_len = p->mac_len; 293071d93b39SHerbert Xu 293171d93b39SHerbert Xu skb_reserve(nskb, headroom); 293286911732SHerbert Xu __skb_put(nskb, skb_gro_offset(p)); 293371d93b39SHerbert Xu 293486911732SHerbert Xu skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 293571d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 293671d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 293771d93b39SHerbert Xu 293886911732SHerbert Xu __skb_pull(p, skb_gro_offset(p)); 293986911732SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), 294086911732SHerbert Xu p->data - skb_mac_header(p)); 294171d93b39SHerbert Xu 294271d93b39SHerbert Xu *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p); 294371d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 29449aaa156cSHerbert Xu skb_shinfo(nskb)->gso_size = pinfo->gso_size; 2945622e0ca1SHerbert Xu pinfo->gso_size = 0; 294671d93b39SHerbert Xu skb_header_release(p); 294771d93b39SHerbert Xu nskb->prev = p; 294871d93b39SHerbert Xu 294971d93b39SHerbert Xu nskb->data_len += p->len; 2950de8261c2SEric Dumazet nskb->truesize += p->truesize; 295171d93b39SHerbert Xu nskb->len += p->len; 295271d93b39SHerbert Xu 295371d93b39SHerbert Xu *head = nskb; 295471d93b39SHerbert Xu nskb->next = p->next; 295571d93b39SHerbert Xu p->next = NULL; 295671d93b39SHerbert Xu 295771d93b39SHerbert Xu p = nskb; 295871d93b39SHerbert Xu 295971d93b39SHerbert Xu merge: 2960715dc1f3SEric Dumazet delta_truesize = skb->truesize; 296167147ba9SHerbert Xu if (offset > headlen) { 2962d1dc7abfSMichal Schmidt unsigned int eat = offset - headlen; 2963d1dc7abfSMichal Schmidt 2964d1dc7abfSMichal Schmidt skbinfo->frags[0].page_offset += eat; 29659e903e08SEric Dumazet skb_frag_size_sub(&skbinfo->frags[0], eat); 2966d1dc7abfSMichal Schmidt skb->data_len -= eat; 2967d1dc7abfSMichal Schmidt skb->len -= eat; 296867147ba9SHerbert Xu offset = headlen; 296956035022SHerbert Xu } 297056035022SHerbert Xu 297167147ba9SHerbert Xu __skb_pull(skb, offset); 297256035022SHerbert Xu 297371d93b39SHerbert Xu p->prev->next = skb; 297471d93b39SHerbert Xu p->prev = skb; 297571d93b39SHerbert Xu skb_header_release(skb); 297671d93b39SHerbert Xu 29775d38a079SHerbert Xu done: 29785d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 297937fe4732SHerbert Xu p->data_len += len; 2980715dc1f3SEric Dumazet p->truesize += delta_truesize; 298137fe4732SHerbert Xu p->len += len; 298271d93b39SHerbert Xu 298371d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 298471d93b39SHerbert Xu return 0; 298571d93b39SHerbert Xu } 298671d93b39SHerbert Xu EXPORT_SYMBOL_GPL(skb_gro_receive); 298771d93b39SHerbert Xu 29881da177e4SLinus Torvalds void __init skb_init(void) 29891da177e4SLinus Torvalds { 29901da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 29911da177e4SLinus Torvalds sizeof(struct sk_buff), 29921da177e4SLinus Torvalds 0, 2993e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 299420c2df83SPaul Mundt NULL); 2995d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 2996d179cd12SDavid S. Miller (2*sizeof(struct sk_buff)) + 2997d179cd12SDavid S. Miller sizeof(atomic_t), 2998d179cd12SDavid S. Miller 0, 2999e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 300020c2df83SPaul Mundt NULL); 30011da177e4SLinus Torvalds } 30021da177e4SLinus Torvalds 3003716ea3a7SDavid Howells /** 3004716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3005716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 3006716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 3007716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 3008716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 3009716ea3a7SDavid Howells * 3010716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 3011716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 3012716ea3a7SDavid Howells */ 301351c739d1SDavid S. Miller static int 301451c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3015716ea3a7SDavid Howells { 30161a028e50SDavid S. Miller int start = skb_headlen(skb); 30171a028e50SDavid S. Miller int i, copy = start - offset; 3018fbb398a8SDavid S. Miller struct sk_buff *frag_iter; 3019716ea3a7SDavid Howells int elt = 0; 3020716ea3a7SDavid Howells 3021716ea3a7SDavid Howells if (copy > 0) { 3022716ea3a7SDavid Howells if (copy > len) 3023716ea3a7SDavid Howells copy = len; 3024642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 3025716ea3a7SDavid Howells elt++; 3026716ea3a7SDavid Howells if ((len -= copy) == 0) 3027716ea3a7SDavid Howells return elt; 3028716ea3a7SDavid Howells offset += copy; 3029716ea3a7SDavid Howells } 3030716ea3a7SDavid Howells 3031716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 30321a028e50SDavid S. Miller int end; 3033716ea3a7SDavid Howells 3034547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30351a028e50SDavid S. Miller 30369e903e08SEric Dumazet end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3037716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3038716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3039716ea3a7SDavid Howells 3040716ea3a7SDavid Howells if (copy > len) 3041716ea3a7SDavid Howells copy = len; 3042ea2ab693SIan Campbell sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3043642f1490SJens Axboe frag->page_offset+offset-start); 3044716ea3a7SDavid Howells elt++; 3045716ea3a7SDavid Howells if (!(len -= copy)) 3046716ea3a7SDavid Howells return elt; 3047716ea3a7SDavid Howells offset += copy; 3048716ea3a7SDavid Howells } 30491a028e50SDavid S. Miller start = end; 3050716ea3a7SDavid Howells } 3051716ea3a7SDavid Howells 3052fbb398a8SDavid S. Miller skb_walk_frags(skb, frag_iter) { 30531a028e50SDavid S. Miller int end; 3054716ea3a7SDavid Howells 3055547b792cSIlpo Järvinen WARN_ON(start > offset + len); 30561a028e50SDavid S. Miller 3057fbb398a8SDavid S. Miller end = start + frag_iter->len; 3058716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 3059716ea3a7SDavid Howells if (copy > len) 3060716ea3a7SDavid Howells copy = len; 3061fbb398a8SDavid S. Miller elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 306251c739d1SDavid S. Miller copy); 3063716ea3a7SDavid Howells if ((len -= copy) == 0) 3064716ea3a7SDavid Howells return elt; 3065716ea3a7SDavid Howells offset += copy; 3066716ea3a7SDavid Howells } 30671a028e50SDavid S. Miller start = end; 3068716ea3a7SDavid Howells } 3069716ea3a7SDavid Howells BUG_ON(len); 3070716ea3a7SDavid Howells return elt; 3071716ea3a7SDavid Howells } 3072716ea3a7SDavid Howells 307351c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 307451c739d1SDavid S. Miller { 307551c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 307651c739d1SDavid S. Miller 3077c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 307851c739d1SDavid S. Miller 307951c739d1SDavid S. Miller return nsg; 308051c739d1SDavid S. Miller } 3081b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 308251c739d1SDavid S. Miller 3083716ea3a7SDavid Howells /** 3084716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 3085716ea3a7SDavid Howells * @skb: The socket buffer to check. 3086716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 3087716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 3088716ea3a7SDavid Howells * 3089716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 3090716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 3091716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 3092716ea3a7SDavid Howells * 3093716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 3094716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 3095716ea3a7SDavid Howells * set to point to the skb in which this space begins. 3096716ea3a7SDavid Howells * 3097716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 3098716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 3099716ea3a7SDavid Howells */ 3100716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3101716ea3a7SDavid Howells { 3102716ea3a7SDavid Howells int copyflag; 3103716ea3a7SDavid Howells int elt; 3104716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 3105716ea3a7SDavid Howells 3106716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 3107716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 3108716ea3a7SDavid Howells * at the moment even if they are anonymous). 3109716ea3a7SDavid Howells */ 3110716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3111716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3112716ea3a7SDavid Howells return -ENOMEM; 3113716ea3a7SDavid Howells 3114716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 311521dc3301SDavid S. Miller if (!skb_has_frag_list(skb)) { 3116716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 3117716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 3118716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 3119716ea3a7SDavid Howells * space, 128 bytes is fair. */ 3120716ea3a7SDavid Howells 3121716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 3122716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3123716ea3a7SDavid Howells return -ENOMEM; 3124716ea3a7SDavid Howells 3125716ea3a7SDavid Howells /* Voila! */ 3126716ea3a7SDavid Howells *trailer = skb; 3127716ea3a7SDavid Howells return 1; 3128716ea3a7SDavid Howells } 3129716ea3a7SDavid Howells 3130716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 3131716ea3a7SDavid Howells 3132716ea3a7SDavid Howells elt = 1; 3133716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 3134716ea3a7SDavid Howells copyflag = 0; 3135716ea3a7SDavid Howells 3136716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 3137716ea3a7SDavid Howells int ntail = 0; 3138716ea3a7SDavid Howells 3139716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 3140716ea3a7SDavid Howells * this can happen on input. Copy it and everything 3141716ea3a7SDavid Howells * after it. */ 3142716ea3a7SDavid Howells 3143716ea3a7SDavid Howells if (skb_shared(skb1)) 3144716ea3a7SDavid Howells copyflag = 1; 3145716ea3a7SDavid Howells 3146716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 3147716ea3a7SDavid Howells 3148716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 3149716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 315021dc3301SDavid S. Miller skb_has_frag_list(skb1) || 3151716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 3152716ea3a7SDavid Howells ntail = tailbits + 128; 3153716ea3a7SDavid Howells } 3154716ea3a7SDavid Howells 3155716ea3a7SDavid Howells if (copyflag || 3156716ea3a7SDavid Howells skb_cloned(skb1) || 3157716ea3a7SDavid Howells ntail || 3158716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 315921dc3301SDavid S. Miller skb_has_frag_list(skb1)) { 3160716ea3a7SDavid Howells struct sk_buff *skb2; 3161716ea3a7SDavid Howells 3162716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 3163716ea3a7SDavid Howells if (ntail == 0) 3164716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 3165716ea3a7SDavid Howells else 3166716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 3167716ea3a7SDavid Howells skb_headroom(skb1), 3168716ea3a7SDavid Howells ntail, 3169716ea3a7SDavid Howells GFP_ATOMIC); 3170716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 3171716ea3a7SDavid Howells return -ENOMEM; 3172716ea3a7SDavid Howells 3173716ea3a7SDavid Howells if (skb1->sk) 3174716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 3175716ea3a7SDavid Howells 3176716ea3a7SDavid Howells /* Looking around. Are we still alive? 3177716ea3a7SDavid Howells * OK, link new skb, drop old one */ 3178716ea3a7SDavid Howells 3179716ea3a7SDavid Howells skb2->next = skb1->next; 3180716ea3a7SDavid Howells *skb_p = skb2; 3181716ea3a7SDavid Howells kfree_skb(skb1); 3182716ea3a7SDavid Howells skb1 = skb2; 3183716ea3a7SDavid Howells } 3184716ea3a7SDavid Howells elt++; 3185716ea3a7SDavid Howells *trailer = skb1; 3186716ea3a7SDavid Howells skb_p = &skb1->next; 3187716ea3a7SDavid Howells } 3188716ea3a7SDavid Howells 3189716ea3a7SDavid Howells return elt; 3190716ea3a7SDavid Howells } 3191b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 3192716ea3a7SDavid Howells 3193b1faf566SEric Dumazet static void sock_rmem_free(struct sk_buff *skb) 3194b1faf566SEric Dumazet { 3195b1faf566SEric Dumazet struct sock *sk = skb->sk; 3196b1faf566SEric Dumazet 3197b1faf566SEric Dumazet atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3198b1faf566SEric Dumazet } 3199b1faf566SEric Dumazet 3200b1faf566SEric Dumazet /* 3201b1faf566SEric Dumazet * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3202b1faf566SEric Dumazet */ 3203b1faf566SEric Dumazet int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3204b1faf566SEric Dumazet { 3205110c4330SEric Dumazet int len = skb->len; 3206110c4330SEric Dumazet 3207b1faf566SEric Dumazet if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 320895c96174SEric Dumazet (unsigned int)sk->sk_rcvbuf) 3209b1faf566SEric Dumazet return -ENOMEM; 3210b1faf566SEric Dumazet 3211b1faf566SEric Dumazet skb_orphan(skb); 3212b1faf566SEric Dumazet skb->sk = sk; 3213b1faf566SEric Dumazet skb->destructor = sock_rmem_free; 3214b1faf566SEric Dumazet atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3215b1faf566SEric Dumazet 3216abb57ea4SEric Dumazet /* before exiting rcu section, make sure dst is refcounted */ 3217abb57ea4SEric Dumazet skb_dst_force(skb); 3218abb57ea4SEric Dumazet 3219b1faf566SEric Dumazet skb_queue_tail(&sk->sk_error_queue, skb); 3220b1faf566SEric Dumazet if (!sock_flag(sk, SOCK_DEAD)) 3221110c4330SEric Dumazet sk->sk_data_ready(sk, len); 3222b1faf566SEric Dumazet return 0; 3223b1faf566SEric Dumazet } 3224b1faf566SEric Dumazet EXPORT_SYMBOL(sock_queue_err_skb); 3225b1faf566SEric Dumazet 3226ac45f602SPatrick Ohly void skb_tstamp_tx(struct sk_buff *orig_skb, 3227ac45f602SPatrick Ohly struct skb_shared_hwtstamps *hwtstamps) 3228ac45f602SPatrick Ohly { 3229ac45f602SPatrick Ohly struct sock *sk = orig_skb->sk; 3230ac45f602SPatrick Ohly struct sock_exterr_skb *serr; 3231ac45f602SPatrick Ohly struct sk_buff *skb; 3232ac45f602SPatrick Ohly int err; 3233ac45f602SPatrick Ohly 3234ac45f602SPatrick Ohly if (!sk) 3235ac45f602SPatrick Ohly return; 3236ac45f602SPatrick Ohly 3237ac45f602SPatrick Ohly skb = skb_clone(orig_skb, GFP_ATOMIC); 3238ac45f602SPatrick Ohly if (!skb) 3239ac45f602SPatrick Ohly return; 3240ac45f602SPatrick Ohly 3241ac45f602SPatrick Ohly if (hwtstamps) { 3242ac45f602SPatrick Ohly *skb_hwtstamps(skb) = 3243ac45f602SPatrick Ohly *hwtstamps; 3244ac45f602SPatrick Ohly } else { 3245ac45f602SPatrick Ohly /* 3246ac45f602SPatrick Ohly * no hardware time stamps available, 32472244d07bSOliver Hartkopp * so keep the shared tx_flags and only 3248ac45f602SPatrick Ohly * store software time stamp 3249ac45f602SPatrick Ohly */ 3250ac45f602SPatrick Ohly skb->tstamp = ktime_get_real(); 3251ac45f602SPatrick Ohly } 3252ac45f602SPatrick Ohly 3253ac45f602SPatrick Ohly serr = SKB_EXT_ERR(skb); 3254ac45f602SPatrick Ohly memset(serr, 0, sizeof(*serr)); 3255ac45f602SPatrick Ohly serr->ee.ee_errno = ENOMSG; 3256ac45f602SPatrick Ohly serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 325729030374SEric Dumazet 3258ac45f602SPatrick Ohly err = sock_queue_err_skb(sk, skb); 325929030374SEric Dumazet 3260ac45f602SPatrick Ohly if (err) 3261ac45f602SPatrick Ohly kfree_skb(skb); 3262ac45f602SPatrick Ohly } 3263ac45f602SPatrick Ohly EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3264ac45f602SPatrick Ohly 32656e3e939fSJohannes Berg void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 32666e3e939fSJohannes Berg { 32676e3e939fSJohannes Berg struct sock *sk = skb->sk; 32686e3e939fSJohannes Berg struct sock_exterr_skb *serr; 32696e3e939fSJohannes Berg int err; 32706e3e939fSJohannes Berg 32716e3e939fSJohannes Berg skb->wifi_acked_valid = 1; 32726e3e939fSJohannes Berg skb->wifi_acked = acked; 32736e3e939fSJohannes Berg 32746e3e939fSJohannes Berg serr = SKB_EXT_ERR(skb); 32756e3e939fSJohannes Berg memset(serr, 0, sizeof(*serr)); 32766e3e939fSJohannes Berg serr->ee.ee_errno = ENOMSG; 32776e3e939fSJohannes Berg serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 32786e3e939fSJohannes Berg 32796e3e939fSJohannes Berg err = sock_queue_err_skb(sk, skb); 32806e3e939fSJohannes Berg if (err) 32816e3e939fSJohannes Berg kfree_skb(skb); 32826e3e939fSJohannes Berg } 32836e3e939fSJohannes Berg EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 32846e3e939fSJohannes Berg 3285ac45f602SPatrick Ohly 3286f35d9d8aSRusty Russell /** 3287f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 3288f35d9d8aSRusty Russell * @skb: the skb to set 3289f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 3290f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 3291f35d9d8aSRusty Russell * 3292f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 3293f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3294f35d9d8aSRusty Russell * 3295f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 3296f35d9d8aSRusty Russell * returns false you should drop the packet. 3297f35d9d8aSRusty Russell */ 3298f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3299f35d9d8aSRusty Russell { 33005ff8dda3SHerbert Xu if (unlikely(start > skb_headlen(skb)) || 33015ff8dda3SHerbert Xu unlikely((int)start + off > skb_headlen(skb) - 2)) { 3302e87cc472SJoe Perches net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 33035ff8dda3SHerbert Xu start, off, skb_headlen(skb)); 3304f35d9d8aSRusty Russell return false; 3305f35d9d8aSRusty Russell } 3306f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 3307f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 3308f35d9d8aSRusty Russell skb->csum_offset = off; 3309f35d9d8aSRusty Russell return true; 3310f35d9d8aSRusty Russell } 3311b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3312f35d9d8aSRusty Russell 33134497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 33144497b076SBen Hutchings { 3315e87cc472SJoe Perches net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 3316e87cc472SJoe Perches skb->dev->name); 33174497b076SBen Hutchings } 33184497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 3319