1 /* 2 * Routines having to do with the 'struct sk_buff' memory handlers. 3 * 4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 5 * Florian La Roche <rzsfl@rz.uni-sb.de> 6 * 7 * Fixes: 8 * Alan Cox : Fixed the worst of the load 9 * balancer bugs. 10 * Dave Platt : Interrupt stacking fix. 11 * Richard Kooijman : Timestamp fixes. 12 * Alan Cox : Changed buffer format. 13 * Alan Cox : destructor hook for AF_UNIX etc. 14 * Linus Torvalds : Better skb_clone. 15 * Alan Cox : Added skb_copy. 16 * Alan Cox : Added all the changed routines Linus 17 * only put in the headers 18 * Ray VanTassle : Fixed --skb->lock in free 19 * Alan Cox : skb_copy copy arp field 20 * Andi Kleen : slabified it. 21 * Robert Olsson : Removed skb_head_pool 22 * 23 * NOTE: 24 * The __skb_ routines should be called with interrupts 25 * disabled, or you better be *real* sure that the operation is atomic 26 * with respect to whatever list is being frobbed (e.g. via lock_sock() 27 * or via disabling bottom half handlers, etc). 28 * 29 * This program is free software; you can redistribute it and/or 30 * modify it under the terms of the GNU General Public License 31 * as published by the Free Software Foundation; either version 32 * 2 of the License, or (at your option) any later version. 33 */ 34 35 /* 36 * The functions in this file will not compile correctly with gcc 2.4.x 37 */ 38 39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 40 41 #include <linux/module.h> 42 #include <linux/types.h> 43 #include <linux/kernel.h> 44 #include <linux/kmemcheck.h> 45 #include <linux/mm.h> 46 #include <linux/interrupt.h> 47 #include <linux/in.h> 48 #include <linux/inet.h> 49 #include <linux/slab.h> 50 #include <linux/netdevice.h> 51 #ifdef CONFIG_NET_CLS_ACT 52 #include <net/pkt_sched.h> 53 #endif 54 #include <linux/string.h> 55 #include <linux/skbuff.h> 56 #include <linux/splice.h> 57 #include <linux/cache.h> 58 #include <linux/rtnetlink.h> 59 #include <linux/init.h> 60 #include <linux/scatterlist.h> 61 #include <linux/errqueue.h> 62 #include <linux/prefetch.h> 63 64 #include <net/protocol.h> 65 #include <net/dst.h> 66 #include <net/sock.h> 67 #include <net/checksum.h> 68 #include <net/xfrm.h> 69 70 #include <asm/uaccess.h> 71 #include <trace/events/skb.h> 72 #include <linux/highmem.h> 73 74 struct kmem_cache *skbuff_head_cache __read_mostly; 75 static struct kmem_cache *skbuff_fclone_cache __read_mostly; 76 77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe, 78 struct pipe_buffer *buf) 79 { 80 put_page(buf->page); 81 } 82 83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe, 84 struct pipe_buffer *buf) 85 { 86 get_page(buf->page); 87 } 88 89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, 90 struct pipe_buffer *buf) 91 { 92 return 1; 93 } 94 95 96 /* Pipe buffer operations for a socket. */ 97 static const struct pipe_buf_operations sock_pipe_buf_ops = { 98 .can_merge = 0, 99 .map = generic_pipe_buf_map, 100 .unmap = generic_pipe_buf_unmap, 101 .confirm = generic_pipe_buf_confirm, 102 .release = sock_pipe_buf_release, 103 .steal = sock_pipe_buf_steal, 104 .get = sock_pipe_buf_get, 105 }; 106 107 /* 108 * Keep out-of-line to prevent kernel bloat. 109 * __builtin_return_address is not used because it is not always 110 * reliable. 111 */ 112 113 /** 114 * skb_over_panic - private function 115 * @skb: buffer 116 * @sz: size 117 * @here: address 118 * 119 * Out of line support code for skb_put(). Not user callable. 120 */ 121 static void skb_over_panic(struct sk_buff *skb, int sz, void *here) 122 { 123 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 124 __func__, here, skb->len, sz, skb->head, skb->data, 125 (unsigned long)skb->tail, (unsigned long)skb->end, 126 skb->dev ? skb->dev->name : "<NULL>"); 127 BUG(); 128 } 129 130 /** 131 * skb_under_panic - private function 132 * @skb: buffer 133 * @sz: size 134 * @here: address 135 * 136 * Out of line support code for skb_push(). Not user callable. 137 */ 138 139 static void skb_under_panic(struct sk_buff *skb, int sz, void *here) 140 { 141 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 142 __func__, here, skb->len, sz, skb->head, skb->data, 143 (unsigned long)skb->tail, (unsigned long)skb->end, 144 skb->dev ? skb->dev->name : "<NULL>"); 145 BUG(); 146 } 147 148 149 /* 150 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 151 * the caller if emergency pfmemalloc reserves are being used. If it is and 152 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 153 * may be used. Otherwise, the packet data may be discarded until enough 154 * memory is free 155 */ 156 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 157 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 158 void *__kmalloc_reserve(size_t size, gfp_t flags, int node, unsigned long ip, 159 bool *pfmemalloc) 160 { 161 void *obj; 162 bool ret_pfmemalloc = false; 163 164 /* 165 * Try a regular allocation, when that fails and we're not entitled 166 * to the reserves, fail. 167 */ 168 obj = kmalloc_node_track_caller(size, 169 flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 170 node); 171 if (obj || !(gfp_pfmemalloc_allowed(flags))) 172 goto out; 173 174 /* Try again but now we are using pfmemalloc reserves */ 175 ret_pfmemalloc = true; 176 obj = kmalloc_node_track_caller(size, flags, node); 177 178 out: 179 if (pfmemalloc) 180 *pfmemalloc = ret_pfmemalloc; 181 182 return obj; 183 } 184 185 /* Allocate a new skbuff. We do this ourselves so we can fill in a few 186 * 'private' fields and also do memory statistics to find all the 187 * [BEEP] leaks. 188 * 189 */ 190 191 /** 192 * __alloc_skb - allocate a network buffer 193 * @size: size to allocate 194 * @gfp_mask: allocation mask 195 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 196 * instead of head cache and allocate a cloned (child) skb. 197 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 198 * allocations in case the data is required for writeback 199 * @node: numa node to allocate memory on 200 * 201 * Allocate a new &sk_buff. The returned buffer has no headroom and a 202 * tail room of at least size bytes. The object has a reference count 203 * of one. The return is the buffer. On a failure the return is %NULL. 204 * 205 * Buffers may only be allocated from interrupts using a @gfp_mask of 206 * %GFP_ATOMIC. 207 */ 208 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 209 int flags, int node) 210 { 211 struct kmem_cache *cache; 212 struct skb_shared_info *shinfo; 213 struct sk_buff *skb; 214 u8 *data; 215 bool pfmemalloc; 216 217 cache = (flags & SKB_ALLOC_FCLONE) 218 ? skbuff_fclone_cache : skbuff_head_cache; 219 220 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 221 gfp_mask |= __GFP_MEMALLOC; 222 223 /* Get the HEAD */ 224 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 225 if (!skb) 226 goto out; 227 prefetchw(skb); 228 229 /* We do our best to align skb_shared_info on a separate cache 230 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 231 * aligned memory blocks, unless SLUB/SLAB debug is enabled. 232 * Both skb->head and skb_shared_info are cache line aligned. 233 */ 234 size = SKB_DATA_ALIGN(size); 235 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 236 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 237 if (!data) 238 goto nodata; 239 /* kmalloc(size) might give us more room than requested. 240 * Put skb_shared_info exactly at the end of allocated zone, 241 * to allow max possible filling before reallocation. 242 */ 243 size = SKB_WITH_OVERHEAD(ksize(data)); 244 prefetchw(data + size); 245 246 /* 247 * Only clear those fields we need to clear, not those that we will 248 * actually initialise below. Hence, don't put any more fields after 249 * the tail pointer in struct sk_buff! 250 */ 251 memset(skb, 0, offsetof(struct sk_buff, tail)); 252 /* Account for allocated memory : skb + skb->head */ 253 skb->truesize = SKB_TRUESIZE(size); 254 skb->pfmemalloc = pfmemalloc; 255 atomic_set(&skb->users, 1); 256 skb->head = data; 257 skb->data = data; 258 skb_reset_tail_pointer(skb); 259 skb->end = skb->tail + size; 260 #ifdef NET_SKBUFF_DATA_USES_OFFSET 261 skb->mac_header = ~0U; 262 #endif 263 264 /* make sure we initialize shinfo sequentially */ 265 shinfo = skb_shinfo(skb); 266 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 267 atomic_set(&shinfo->dataref, 1); 268 kmemcheck_annotate_variable(shinfo->destructor_arg); 269 270 if (flags & SKB_ALLOC_FCLONE) { 271 struct sk_buff *child = skb + 1; 272 atomic_t *fclone_ref = (atomic_t *) (child + 1); 273 274 kmemcheck_annotate_bitfield(child, flags1); 275 kmemcheck_annotate_bitfield(child, flags2); 276 skb->fclone = SKB_FCLONE_ORIG; 277 atomic_set(fclone_ref, 1); 278 279 child->fclone = SKB_FCLONE_UNAVAILABLE; 280 child->pfmemalloc = pfmemalloc; 281 } 282 out: 283 return skb; 284 nodata: 285 kmem_cache_free(cache, skb); 286 skb = NULL; 287 goto out; 288 } 289 EXPORT_SYMBOL(__alloc_skb); 290 291 /** 292 * build_skb - build a network buffer 293 * @data: data buffer provided by caller 294 * @frag_size: size of fragment, or 0 if head was kmalloced 295 * 296 * Allocate a new &sk_buff. Caller provides space holding head and 297 * skb_shared_info. @data must have been allocated by kmalloc() 298 * The return is the new skb buffer. 299 * On a failure the return is %NULL, and @data is not freed. 300 * Notes : 301 * Before IO, driver allocates only data buffer where NIC put incoming frame 302 * Driver should add room at head (NET_SKB_PAD) and 303 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 304 * After IO, driver calls build_skb(), to allocate sk_buff and populate it 305 * before giving packet to stack. 306 * RX rings only contains data buffers, not full skbs. 307 */ 308 struct sk_buff *build_skb(void *data, unsigned int frag_size) 309 { 310 struct skb_shared_info *shinfo; 311 struct sk_buff *skb; 312 unsigned int size = frag_size ? : ksize(data); 313 314 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 315 if (!skb) 316 return NULL; 317 318 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 319 320 memset(skb, 0, offsetof(struct sk_buff, tail)); 321 skb->truesize = SKB_TRUESIZE(size); 322 skb->head_frag = frag_size != 0; 323 atomic_set(&skb->users, 1); 324 skb->head = data; 325 skb->data = data; 326 skb_reset_tail_pointer(skb); 327 skb->end = skb->tail + size; 328 #ifdef NET_SKBUFF_DATA_USES_OFFSET 329 skb->mac_header = ~0U; 330 #endif 331 332 /* make sure we initialize shinfo sequentially */ 333 shinfo = skb_shinfo(skb); 334 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 335 atomic_set(&shinfo->dataref, 1); 336 kmemcheck_annotate_variable(shinfo->destructor_arg); 337 338 return skb; 339 } 340 EXPORT_SYMBOL(build_skb); 341 342 struct netdev_alloc_cache { 343 struct page_frag frag; 344 /* we maintain a pagecount bias, so that we dont dirty cache line 345 * containing page->_count every time we allocate a fragment. 346 */ 347 unsigned int pagecnt_bias; 348 }; 349 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache); 350 351 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768) 352 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER) 353 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE 354 355 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 356 { 357 struct netdev_alloc_cache *nc; 358 void *data = NULL; 359 int order; 360 unsigned long flags; 361 362 local_irq_save(flags); 363 nc = &__get_cpu_var(netdev_alloc_cache); 364 if (unlikely(!nc->frag.page)) { 365 refill: 366 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) { 367 gfp_t gfp = gfp_mask; 368 369 if (order) 370 gfp |= __GFP_COMP | __GFP_NOWARN; 371 nc->frag.page = alloc_pages(gfp, order); 372 if (likely(nc->frag.page)) 373 break; 374 if (--order < 0) 375 goto end; 376 } 377 nc->frag.size = PAGE_SIZE << order; 378 recycle: 379 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS); 380 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS; 381 nc->frag.offset = 0; 382 } 383 384 if (nc->frag.offset + fragsz > nc->frag.size) { 385 /* avoid unnecessary locked operations if possible */ 386 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) || 387 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count)) 388 goto recycle; 389 goto refill; 390 } 391 392 data = page_address(nc->frag.page) + nc->frag.offset; 393 nc->frag.offset += fragsz; 394 nc->pagecnt_bias--; 395 end: 396 local_irq_restore(flags); 397 return data; 398 } 399 400 /** 401 * netdev_alloc_frag - allocate a page fragment 402 * @fragsz: fragment size 403 * 404 * Allocates a frag from a page for receive buffer. 405 * Uses GFP_ATOMIC allocations. 406 */ 407 void *netdev_alloc_frag(unsigned int fragsz) 408 { 409 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 410 } 411 EXPORT_SYMBOL(netdev_alloc_frag); 412 413 /** 414 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 415 * @dev: network device to receive on 416 * @length: length to allocate 417 * @gfp_mask: get_free_pages mask, passed to alloc_skb 418 * 419 * Allocate a new &sk_buff and assign it a usage count of one. The 420 * buffer has unspecified headroom built in. Users should allocate 421 * the headroom they think they need without accounting for the 422 * built in space. The built in space is used for optimisations. 423 * 424 * %NULL is returned if there is no free memory. 425 */ 426 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 427 unsigned int length, gfp_t gfp_mask) 428 { 429 struct sk_buff *skb = NULL; 430 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) + 431 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 432 433 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) { 434 void *data; 435 436 if (sk_memalloc_socks()) 437 gfp_mask |= __GFP_MEMALLOC; 438 439 data = __netdev_alloc_frag(fragsz, gfp_mask); 440 441 if (likely(data)) { 442 skb = build_skb(data, fragsz); 443 if (unlikely(!skb)) 444 put_page(virt_to_head_page(data)); 445 } 446 } else { 447 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 448 SKB_ALLOC_RX, NUMA_NO_NODE); 449 } 450 if (likely(skb)) { 451 skb_reserve(skb, NET_SKB_PAD); 452 skb->dev = dev; 453 } 454 return skb; 455 } 456 EXPORT_SYMBOL(__netdev_alloc_skb); 457 458 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 459 int size, unsigned int truesize) 460 { 461 skb_fill_page_desc(skb, i, page, off, size); 462 skb->len += size; 463 skb->data_len += size; 464 skb->truesize += truesize; 465 } 466 EXPORT_SYMBOL(skb_add_rx_frag); 467 468 static void skb_drop_list(struct sk_buff **listp) 469 { 470 struct sk_buff *list = *listp; 471 472 *listp = NULL; 473 474 do { 475 struct sk_buff *this = list; 476 list = list->next; 477 kfree_skb(this); 478 } while (list); 479 } 480 481 static inline void skb_drop_fraglist(struct sk_buff *skb) 482 { 483 skb_drop_list(&skb_shinfo(skb)->frag_list); 484 } 485 486 static void skb_clone_fraglist(struct sk_buff *skb) 487 { 488 struct sk_buff *list; 489 490 skb_walk_frags(skb, list) 491 skb_get(list); 492 } 493 494 static void skb_free_head(struct sk_buff *skb) 495 { 496 if (skb->head_frag) 497 put_page(virt_to_head_page(skb->head)); 498 else 499 kfree(skb->head); 500 } 501 502 static void skb_release_data(struct sk_buff *skb) 503 { 504 if (!skb->cloned || 505 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 506 &skb_shinfo(skb)->dataref)) { 507 if (skb_shinfo(skb)->nr_frags) { 508 int i; 509 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 510 skb_frag_unref(skb, i); 511 } 512 513 /* 514 * If skb buf is from userspace, we need to notify the caller 515 * the lower device DMA has done; 516 */ 517 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 518 struct ubuf_info *uarg; 519 520 uarg = skb_shinfo(skb)->destructor_arg; 521 if (uarg->callback) 522 uarg->callback(uarg, true); 523 } 524 525 if (skb_has_frag_list(skb)) 526 skb_drop_fraglist(skb); 527 528 skb_free_head(skb); 529 } 530 } 531 532 /* 533 * Free an skbuff by memory without cleaning the state. 534 */ 535 static void kfree_skbmem(struct sk_buff *skb) 536 { 537 struct sk_buff *other; 538 atomic_t *fclone_ref; 539 540 switch (skb->fclone) { 541 case SKB_FCLONE_UNAVAILABLE: 542 kmem_cache_free(skbuff_head_cache, skb); 543 break; 544 545 case SKB_FCLONE_ORIG: 546 fclone_ref = (atomic_t *) (skb + 2); 547 if (atomic_dec_and_test(fclone_ref)) 548 kmem_cache_free(skbuff_fclone_cache, skb); 549 break; 550 551 case SKB_FCLONE_CLONE: 552 fclone_ref = (atomic_t *) (skb + 1); 553 other = skb - 1; 554 555 /* The clone portion is available for 556 * fast-cloning again. 557 */ 558 skb->fclone = SKB_FCLONE_UNAVAILABLE; 559 560 if (atomic_dec_and_test(fclone_ref)) 561 kmem_cache_free(skbuff_fclone_cache, other); 562 break; 563 } 564 } 565 566 static void skb_release_head_state(struct sk_buff *skb) 567 { 568 skb_dst_drop(skb); 569 #ifdef CONFIG_XFRM 570 secpath_put(skb->sp); 571 #endif 572 if (skb->destructor) { 573 WARN_ON(in_irq()); 574 skb->destructor(skb); 575 } 576 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 577 nf_conntrack_put(skb->nfct); 578 #endif 579 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 580 nf_conntrack_put_reasm(skb->nfct_reasm); 581 #endif 582 #ifdef CONFIG_BRIDGE_NETFILTER 583 nf_bridge_put(skb->nf_bridge); 584 #endif 585 /* XXX: IS this still necessary? - JHS */ 586 #ifdef CONFIG_NET_SCHED 587 skb->tc_index = 0; 588 #ifdef CONFIG_NET_CLS_ACT 589 skb->tc_verd = 0; 590 #endif 591 #endif 592 } 593 594 /* Free everything but the sk_buff shell. */ 595 static void skb_release_all(struct sk_buff *skb) 596 { 597 skb_release_head_state(skb); 598 skb_release_data(skb); 599 } 600 601 /** 602 * __kfree_skb - private function 603 * @skb: buffer 604 * 605 * Free an sk_buff. Release anything attached to the buffer. 606 * Clean the state. This is an internal helper function. Users should 607 * always call kfree_skb 608 */ 609 610 void __kfree_skb(struct sk_buff *skb) 611 { 612 skb_release_all(skb); 613 kfree_skbmem(skb); 614 } 615 EXPORT_SYMBOL(__kfree_skb); 616 617 /** 618 * kfree_skb - free an sk_buff 619 * @skb: buffer to free 620 * 621 * Drop a reference to the buffer and free it if the usage count has 622 * hit zero. 623 */ 624 void kfree_skb(struct sk_buff *skb) 625 { 626 if (unlikely(!skb)) 627 return; 628 if (likely(atomic_read(&skb->users) == 1)) 629 smp_rmb(); 630 else if (likely(!atomic_dec_and_test(&skb->users))) 631 return; 632 trace_kfree_skb(skb, __builtin_return_address(0)); 633 __kfree_skb(skb); 634 } 635 EXPORT_SYMBOL(kfree_skb); 636 637 /** 638 * skb_tx_error - report an sk_buff xmit error 639 * @skb: buffer that triggered an error 640 * 641 * Report xmit error if a device callback is tracking this skb. 642 * skb must be freed afterwards. 643 */ 644 void skb_tx_error(struct sk_buff *skb) 645 { 646 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) { 647 struct ubuf_info *uarg; 648 649 uarg = skb_shinfo(skb)->destructor_arg; 650 if (uarg->callback) 651 uarg->callback(uarg, false); 652 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 653 } 654 } 655 EXPORT_SYMBOL(skb_tx_error); 656 657 /** 658 * consume_skb - free an skbuff 659 * @skb: buffer to free 660 * 661 * Drop a ref to the buffer and free it if the usage count has hit zero 662 * Functions identically to kfree_skb, but kfree_skb assumes that the frame 663 * is being dropped after a failure and notes that 664 */ 665 void consume_skb(struct sk_buff *skb) 666 { 667 if (unlikely(!skb)) 668 return; 669 if (likely(atomic_read(&skb->users) == 1)) 670 smp_rmb(); 671 else if (likely(!atomic_dec_and_test(&skb->users))) 672 return; 673 trace_consume_skb(skb); 674 __kfree_skb(skb); 675 } 676 EXPORT_SYMBOL(consume_skb); 677 678 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 679 { 680 new->tstamp = old->tstamp; 681 new->dev = old->dev; 682 new->transport_header = old->transport_header; 683 new->network_header = old->network_header; 684 new->mac_header = old->mac_header; 685 new->inner_transport_header = old->inner_transport_header; 686 new->inner_network_header = old->inner_network_header; 687 skb_dst_copy(new, old); 688 new->rxhash = old->rxhash; 689 new->ooo_okay = old->ooo_okay; 690 new->l4_rxhash = old->l4_rxhash; 691 new->no_fcs = old->no_fcs; 692 new->encapsulation = old->encapsulation; 693 #ifdef CONFIG_XFRM 694 new->sp = secpath_get(old->sp); 695 #endif 696 memcpy(new->cb, old->cb, sizeof(old->cb)); 697 new->csum = old->csum; 698 new->local_df = old->local_df; 699 new->pkt_type = old->pkt_type; 700 new->ip_summed = old->ip_summed; 701 skb_copy_queue_mapping(new, old); 702 new->priority = old->priority; 703 #if IS_ENABLED(CONFIG_IP_VS) 704 new->ipvs_property = old->ipvs_property; 705 #endif 706 new->pfmemalloc = old->pfmemalloc; 707 new->protocol = old->protocol; 708 new->mark = old->mark; 709 new->skb_iif = old->skb_iif; 710 __nf_copy(new, old); 711 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) 712 new->nf_trace = old->nf_trace; 713 #endif 714 #ifdef CONFIG_NET_SCHED 715 new->tc_index = old->tc_index; 716 #ifdef CONFIG_NET_CLS_ACT 717 new->tc_verd = old->tc_verd; 718 #endif 719 #endif 720 new->vlan_tci = old->vlan_tci; 721 722 skb_copy_secmark(new, old); 723 } 724 725 /* 726 * You should not add any new code to this function. Add it to 727 * __copy_skb_header above instead. 728 */ 729 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 730 { 731 #define C(x) n->x = skb->x 732 733 n->next = n->prev = NULL; 734 n->sk = NULL; 735 __copy_skb_header(n, skb); 736 737 C(len); 738 C(data_len); 739 C(mac_len); 740 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 741 n->cloned = 1; 742 n->nohdr = 0; 743 n->destructor = NULL; 744 C(tail); 745 C(end); 746 C(head); 747 C(head_frag); 748 C(data); 749 C(truesize); 750 atomic_set(&n->users, 1); 751 752 atomic_inc(&(skb_shinfo(skb)->dataref)); 753 skb->cloned = 1; 754 755 return n; 756 #undef C 757 } 758 759 /** 760 * skb_morph - morph one skb into another 761 * @dst: the skb to receive the contents 762 * @src: the skb to supply the contents 763 * 764 * This is identical to skb_clone except that the target skb is 765 * supplied by the user. 766 * 767 * The target skb is returned upon exit. 768 */ 769 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 770 { 771 skb_release_all(dst); 772 return __skb_clone(dst, src); 773 } 774 EXPORT_SYMBOL_GPL(skb_morph); 775 776 /** 777 * skb_copy_ubufs - copy userspace skb frags buffers to kernel 778 * @skb: the skb to modify 779 * @gfp_mask: allocation priority 780 * 781 * This must be called on SKBTX_DEV_ZEROCOPY skb. 782 * It will copy all frags into kernel and drop the reference 783 * to userspace pages. 784 * 785 * If this function is called from an interrupt gfp_mask() must be 786 * %GFP_ATOMIC. 787 * 788 * Returns 0 on success or a negative error code on failure 789 * to allocate kernel memory to copy to. 790 */ 791 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 792 { 793 int i; 794 int num_frags = skb_shinfo(skb)->nr_frags; 795 struct page *page, *head = NULL; 796 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg; 797 798 for (i = 0; i < num_frags; i++) { 799 u8 *vaddr; 800 skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 801 802 page = alloc_page(gfp_mask); 803 if (!page) { 804 while (head) { 805 struct page *next = (struct page *)head->private; 806 put_page(head); 807 head = next; 808 } 809 return -ENOMEM; 810 } 811 vaddr = kmap_atomic(skb_frag_page(f)); 812 memcpy(page_address(page), 813 vaddr + f->page_offset, skb_frag_size(f)); 814 kunmap_atomic(vaddr); 815 page->private = (unsigned long)head; 816 head = page; 817 } 818 819 /* skb frags release userspace buffers */ 820 for (i = 0; i < num_frags; i++) 821 skb_frag_unref(skb, i); 822 823 uarg->callback(uarg, false); 824 825 /* skb frags point to kernel buffers */ 826 for (i = num_frags - 1; i >= 0; i--) { 827 __skb_fill_page_desc(skb, i, head, 0, 828 skb_shinfo(skb)->frags[i].size); 829 head = (struct page *)head->private; 830 } 831 832 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY; 833 return 0; 834 } 835 EXPORT_SYMBOL_GPL(skb_copy_ubufs); 836 837 /** 838 * skb_clone - duplicate an sk_buff 839 * @skb: buffer to clone 840 * @gfp_mask: allocation priority 841 * 842 * Duplicate an &sk_buff. The new one is not owned by a socket. Both 843 * copies share the same packet data but not structure. The new 844 * buffer has a reference count of 1. If the allocation fails the 845 * function returns %NULL otherwise the new buffer is returned. 846 * 847 * If this function is called from an interrupt gfp_mask() must be 848 * %GFP_ATOMIC. 849 */ 850 851 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 852 { 853 struct sk_buff *n; 854 855 if (skb_orphan_frags(skb, gfp_mask)) 856 return NULL; 857 858 n = skb + 1; 859 if (skb->fclone == SKB_FCLONE_ORIG && 860 n->fclone == SKB_FCLONE_UNAVAILABLE) { 861 atomic_t *fclone_ref = (atomic_t *) (n + 1); 862 n->fclone = SKB_FCLONE_CLONE; 863 atomic_inc(fclone_ref); 864 } else { 865 if (skb_pfmemalloc(skb)) 866 gfp_mask |= __GFP_MEMALLOC; 867 868 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 869 if (!n) 870 return NULL; 871 872 kmemcheck_annotate_bitfield(n, flags1); 873 kmemcheck_annotate_bitfield(n, flags2); 874 n->fclone = SKB_FCLONE_UNAVAILABLE; 875 } 876 877 return __skb_clone(n, skb); 878 } 879 EXPORT_SYMBOL(skb_clone); 880 881 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 882 { 883 #ifndef NET_SKBUFF_DATA_USES_OFFSET 884 /* 885 * Shift between the two data areas in bytes 886 */ 887 unsigned long offset = new->data - old->data; 888 #endif 889 890 __copy_skb_header(new, old); 891 892 #ifndef NET_SKBUFF_DATA_USES_OFFSET 893 /* {transport,network,mac}_header are relative to skb->head */ 894 new->transport_header += offset; 895 new->network_header += offset; 896 if (skb_mac_header_was_set(new)) 897 new->mac_header += offset; 898 new->inner_transport_header += offset; 899 new->inner_network_header += offset; 900 #endif 901 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 902 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 903 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 904 } 905 906 static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 907 { 908 if (skb_pfmemalloc(skb)) 909 return SKB_ALLOC_RX; 910 return 0; 911 } 912 913 /** 914 * skb_copy - create private copy of an sk_buff 915 * @skb: buffer to copy 916 * @gfp_mask: allocation priority 917 * 918 * Make a copy of both an &sk_buff and its data. This is used when the 919 * caller wishes to modify the data and needs a private copy of the 920 * data to alter. Returns %NULL on failure or the pointer to the buffer 921 * on success. The returned buffer has a reference count of 1. 922 * 923 * As by-product this function converts non-linear &sk_buff to linear 924 * one, so that &sk_buff becomes completely private and caller is allowed 925 * to modify all the data of returned buffer. This means that this 926 * function is not recommended for use in circumstances when only 927 * header is going to be modified. Use pskb_copy() instead. 928 */ 929 930 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 931 { 932 int headerlen = skb_headroom(skb); 933 unsigned int size = skb_end_offset(skb) + skb->data_len; 934 struct sk_buff *n = __alloc_skb(size, gfp_mask, 935 skb_alloc_rx_flag(skb), NUMA_NO_NODE); 936 937 if (!n) 938 return NULL; 939 940 /* Set the data pointer */ 941 skb_reserve(n, headerlen); 942 /* Set the tail pointer and length */ 943 skb_put(n, skb->len); 944 945 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 946 BUG(); 947 948 copy_skb_header(n, skb); 949 return n; 950 } 951 EXPORT_SYMBOL(skb_copy); 952 953 /** 954 * __pskb_copy - create copy of an sk_buff with private head. 955 * @skb: buffer to copy 956 * @headroom: headroom of new skb 957 * @gfp_mask: allocation priority 958 * 959 * Make a copy of both an &sk_buff and part of its data, located 960 * in header. Fragmented data remain shared. This is used when 961 * the caller wishes to modify only header of &sk_buff and needs 962 * private copy of the header to alter. Returns %NULL on failure 963 * or the pointer to the buffer on success. 964 * The returned buffer has a reference count of 1. 965 */ 966 967 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask) 968 { 969 unsigned int size = skb_headlen(skb) + headroom; 970 struct sk_buff *n = __alloc_skb(size, gfp_mask, 971 skb_alloc_rx_flag(skb), NUMA_NO_NODE); 972 973 if (!n) 974 goto out; 975 976 /* Set the data pointer */ 977 skb_reserve(n, headroom); 978 /* Set the tail pointer and length */ 979 skb_put(n, skb_headlen(skb)); 980 /* Copy the bytes */ 981 skb_copy_from_linear_data(skb, n->data, n->len); 982 983 n->truesize += skb->data_len; 984 n->data_len = skb->data_len; 985 n->len = skb->len; 986 987 if (skb_shinfo(skb)->nr_frags) { 988 int i; 989 990 if (skb_orphan_frags(skb, gfp_mask)) { 991 kfree_skb(n); 992 n = NULL; 993 goto out; 994 } 995 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 996 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 997 skb_frag_ref(skb, i); 998 } 999 skb_shinfo(n)->nr_frags = i; 1000 } 1001 1002 if (skb_has_frag_list(skb)) { 1003 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 1004 skb_clone_fraglist(n); 1005 } 1006 1007 copy_skb_header(n, skb); 1008 out: 1009 return n; 1010 } 1011 EXPORT_SYMBOL(__pskb_copy); 1012 1013 /** 1014 * pskb_expand_head - reallocate header of &sk_buff 1015 * @skb: buffer to reallocate 1016 * @nhead: room to add at head 1017 * @ntail: room to add at tail 1018 * @gfp_mask: allocation priority 1019 * 1020 * Expands (or creates identical copy, if &nhead and &ntail are zero) 1021 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have 1022 * reference count of 1. Returns zero in the case of success or error, 1023 * if expansion failed. In the last case, &sk_buff is not changed. 1024 * 1025 * All the pointers pointing into skb header may change and must be 1026 * reloaded after call to this function. 1027 */ 1028 1029 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1030 gfp_t gfp_mask) 1031 { 1032 int i; 1033 u8 *data; 1034 int size = nhead + skb_end_offset(skb) + ntail; 1035 long off; 1036 1037 BUG_ON(nhead < 0); 1038 1039 if (skb_shared(skb)) 1040 BUG(); 1041 1042 size = SKB_DATA_ALIGN(size); 1043 1044 if (skb_pfmemalloc(skb)) 1045 gfp_mask |= __GFP_MEMALLOC; 1046 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1047 gfp_mask, NUMA_NO_NODE, NULL); 1048 if (!data) 1049 goto nodata; 1050 size = SKB_WITH_OVERHEAD(ksize(data)); 1051 1052 /* Copy only real data... and, alas, header. This should be 1053 * optimized for the cases when header is void. 1054 */ 1055 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 1056 1057 memcpy((struct skb_shared_info *)(data + size), 1058 skb_shinfo(skb), 1059 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 1060 1061 /* 1062 * if shinfo is shared we must drop the old head gracefully, but if it 1063 * is not we can just drop the old head and let the existing refcount 1064 * be since all we did is relocate the values 1065 */ 1066 if (skb_cloned(skb)) { 1067 /* copy this zero copy skb frags */ 1068 if (skb_orphan_frags(skb, gfp_mask)) 1069 goto nofrags; 1070 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1071 skb_frag_ref(skb, i); 1072 1073 if (skb_has_frag_list(skb)) 1074 skb_clone_fraglist(skb); 1075 1076 skb_release_data(skb); 1077 } else { 1078 skb_free_head(skb); 1079 } 1080 off = (data + nhead) - skb->head; 1081 1082 skb->head = data; 1083 skb->head_frag = 0; 1084 skb->data += off; 1085 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1086 skb->end = size; 1087 off = nhead; 1088 #else 1089 skb->end = skb->head + size; 1090 #endif 1091 /* {transport,network,mac}_header and tail are relative to skb->head */ 1092 skb->tail += off; 1093 skb->transport_header += off; 1094 skb->network_header += off; 1095 if (skb_mac_header_was_set(skb)) 1096 skb->mac_header += off; 1097 skb->inner_transport_header += off; 1098 skb->inner_network_header += off; 1099 /* Only adjust this if it actually is csum_start rather than csum */ 1100 if (skb->ip_summed == CHECKSUM_PARTIAL) 1101 skb->csum_start += nhead; 1102 skb->cloned = 0; 1103 skb->hdr_len = 0; 1104 skb->nohdr = 0; 1105 atomic_set(&skb_shinfo(skb)->dataref, 1); 1106 return 0; 1107 1108 nofrags: 1109 kfree(data); 1110 nodata: 1111 return -ENOMEM; 1112 } 1113 EXPORT_SYMBOL(pskb_expand_head); 1114 1115 /* Make private copy of skb with writable head and some headroom */ 1116 1117 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 1118 { 1119 struct sk_buff *skb2; 1120 int delta = headroom - skb_headroom(skb); 1121 1122 if (delta <= 0) 1123 skb2 = pskb_copy(skb, GFP_ATOMIC); 1124 else { 1125 skb2 = skb_clone(skb, GFP_ATOMIC); 1126 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 1127 GFP_ATOMIC)) { 1128 kfree_skb(skb2); 1129 skb2 = NULL; 1130 } 1131 } 1132 return skb2; 1133 } 1134 EXPORT_SYMBOL(skb_realloc_headroom); 1135 1136 /** 1137 * skb_copy_expand - copy and expand sk_buff 1138 * @skb: buffer to copy 1139 * @newheadroom: new free bytes at head 1140 * @newtailroom: new free bytes at tail 1141 * @gfp_mask: allocation priority 1142 * 1143 * Make a copy of both an &sk_buff and its data and while doing so 1144 * allocate additional space. 1145 * 1146 * This is used when the caller wishes to modify the data and needs a 1147 * private copy of the data to alter as well as more space for new fields. 1148 * Returns %NULL on failure or the pointer to the buffer 1149 * on success. The returned buffer has a reference count of 1. 1150 * 1151 * You must pass %GFP_ATOMIC as the allocation priority if this function 1152 * is called from an interrupt. 1153 */ 1154 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 1155 int newheadroom, int newtailroom, 1156 gfp_t gfp_mask) 1157 { 1158 /* 1159 * Allocate the copy buffer 1160 */ 1161 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1162 gfp_mask, skb_alloc_rx_flag(skb), 1163 NUMA_NO_NODE); 1164 int oldheadroom = skb_headroom(skb); 1165 int head_copy_len, head_copy_off; 1166 int off; 1167 1168 if (!n) 1169 return NULL; 1170 1171 skb_reserve(n, newheadroom); 1172 1173 /* Set the tail pointer and length */ 1174 skb_put(n, skb->len); 1175 1176 head_copy_len = oldheadroom; 1177 head_copy_off = 0; 1178 if (newheadroom <= head_copy_len) 1179 head_copy_len = newheadroom; 1180 else 1181 head_copy_off = newheadroom - head_copy_len; 1182 1183 /* Copy the linear header and data. */ 1184 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 1185 skb->len + head_copy_len)) 1186 BUG(); 1187 1188 copy_skb_header(n, skb); 1189 1190 off = newheadroom - oldheadroom; 1191 if (n->ip_summed == CHECKSUM_PARTIAL) 1192 n->csum_start += off; 1193 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1194 n->transport_header += off; 1195 n->network_header += off; 1196 if (skb_mac_header_was_set(skb)) 1197 n->mac_header += off; 1198 n->inner_transport_header += off; 1199 n->inner_network_header += off; 1200 #endif 1201 1202 return n; 1203 } 1204 EXPORT_SYMBOL(skb_copy_expand); 1205 1206 /** 1207 * skb_pad - zero pad the tail of an skb 1208 * @skb: buffer to pad 1209 * @pad: space to pad 1210 * 1211 * Ensure that a buffer is followed by a padding area that is zero 1212 * filled. Used by network drivers which may DMA or transfer data 1213 * beyond the buffer end onto the wire. 1214 * 1215 * May return error in out of memory cases. The skb is freed on error. 1216 */ 1217 1218 int skb_pad(struct sk_buff *skb, int pad) 1219 { 1220 int err; 1221 int ntail; 1222 1223 /* If the skbuff is non linear tailroom is always zero.. */ 1224 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 1225 memset(skb->data+skb->len, 0, pad); 1226 return 0; 1227 } 1228 1229 ntail = skb->data_len + pad - (skb->end - skb->tail); 1230 if (likely(skb_cloned(skb) || ntail > 0)) { 1231 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 1232 if (unlikely(err)) 1233 goto free_skb; 1234 } 1235 1236 /* FIXME: The use of this function with non-linear skb's really needs 1237 * to be audited. 1238 */ 1239 err = skb_linearize(skb); 1240 if (unlikely(err)) 1241 goto free_skb; 1242 1243 memset(skb->data + skb->len, 0, pad); 1244 return 0; 1245 1246 free_skb: 1247 kfree_skb(skb); 1248 return err; 1249 } 1250 EXPORT_SYMBOL(skb_pad); 1251 1252 /** 1253 * skb_put - add data to a buffer 1254 * @skb: buffer to use 1255 * @len: amount of data to add 1256 * 1257 * This function extends the used data area of the buffer. If this would 1258 * exceed the total buffer size the kernel will panic. A pointer to the 1259 * first byte of the extra data is returned. 1260 */ 1261 unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 1262 { 1263 unsigned char *tmp = skb_tail_pointer(skb); 1264 SKB_LINEAR_ASSERT(skb); 1265 skb->tail += len; 1266 skb->len += len; 1267 if (unlikely(skb->tail > skb->end)) 1268 skb_over_panic(skb, len, __builtin_return_address(0)); 1269 return tmp; 1270 } 1271 EXPORT_SYMBOL(skb_put); 1272 1273 /** 1274 * skb_push - add data to the start of a buffer 1275 * @skb: buffer to use 1276 * @len: amount of data to add 1277 * 1278 * This function extends the used data area of the buffer at the buffer 1279 * start. If this would exceed the total buffer headroom the kernel will 1280 * panic. A pointer to the first byte of the extra data is returned. 1281 */ 1282 unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 1283 { 1284 skb->data -= len; 1285 skb->len += len; 1286 if (unlikely(skb->data<skb->head)) 1287 skb_under_panic(skb, len, __builtin_return_address(0)); 1288 return skb->data; 1289 } 1290 EXPORT_SYMBOL(skb_push); 1291 1292 /** 1293 * skb_pull - remove data from the start of a buffer 1294 * @skb: buffer to use 1295 * @len: amount of data to remove 1296 * 1297 * This function removes data from the start of a buffer, returning 1298 * the memory to the headroom. A pointer to the next data in the buffer 1299 * is returned. Once the data has been pulled future pushes will overwrite 1300 * the old data. 1301 */ 1302 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 1303 { 1304 return skb_pull_inline(skb, len); 1305 } 1306 EXPORT_SYMBOL(skb_pull); 1307 1308 /** 1309 * skb_trim - remove end from a buffer 1310 * @skb: buffer to alter 1311 * @len: new length 1312 * 1313 * Cut the length of a buffer down by removing data from the tail. If 1314 * the buffer is already under the length specified it is not modified. 1315 * The skb must be linear. 1316 */ 1317 void skb_trim(struct sk_buff *skb, unsigned int len) 1318 { 1319 if (skb->len > len) 1320 __skb_trim(skb, len); 1321 } 1322 EXPORT_SYMBOL(skb_trim); 1323 1324 /* Trims skb to length len. It can change skb pointers. 1325 */ 1326 1327 int ___pskb_trim(struct sk_buff *skb, unsigned int len) 1328 { 1329 struct sk_buff **fragp; 1330 struct sk_buff *frag; 1331 int offset = skb_headlen(skb); 1332 int nfrags = skb_shinfo(skb)->nr_frags; 1333 int i; 1334 int err; 1335 1336 if (skb_cloned(skb) && 1337 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 1338 return err; 1339 1340 i = 0; 1341 if (offset >= len) 1342 goto drop_pages; 1343 1344 for (; i < nfrags; i++) { 1345 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 1346 1347 if (end < len) { 1348 offset = end; 1349 continue; 1350 } 1351 1352 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 1353 1354 drop_pages: 1355 skb_shinfo(skb)->nr_frags = i; 1356 1357 for (; i < nfrags; i++) 1358 skb_frag_unref(skb, i); 1359 1360 if (skb_has_frag_list(skb)) 1361 skb_drop_fraglist(skb); 1362 goto done; 1363 } 1364 1365 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 1366 fragp = &frag->next) { 1367 int end = offset + frag->len; 1368 1369 if (skb_shared(frag)) { 1370 struct sk_buff *nfrag; 1371 1372 nfrag = skb_clone(frag, GFP_ATOMIC); 1373 if (unlikely(!nfrag)) 1374 return -ENOMEM; 1375 1376 nfrag->next = frag->next; 1377 consume_skb(frag); 1378 frag = nfrag; 1379 *fragp = frag; 1380 } 1381 1382 if (end < len) { 1383 offset = end; 1384 continue; 1385 } 1386 1387 if (end > len && 1388 unlikely((err = pskb_trim(frag, len - offset)))) 1389 return err; 1390 1391 if (frag->next) 1392 skb_drop_list(&frag->next); 1393 break; 1394 } 1395 1396 done: 1397 if (len > skb_headlen(skb)) { 1398 skb->data_len -= skb->len - len; 1399 skb->len = len; 1400 } else { 1401 skb->len = len; 1402 skb->data_len = 0; 1403 skb_set_tail_pointer(skb, len); 1404 } 1405 1406 return 0; 1407 } 1408 EXPORT_SYMBOL(___pskb_trim); 1409 1410 /** 1411 * __pskb_pull_tail - advance tail of skb header 1412 * @skb: buffer to reallocate 1413 * @delta: number of bytes to advance tail 1414 * 1415 * The function makes a sense only on a fragmented &sk_buff, 1416 * it expands header moving its tail forward and copying necessary 1417 * data from fragmented part. 1418 * 1419 * &sk_buff MUST have reference count of 1. 1420 * 1421 * Returns %NULL (and &sk_buff does not change) if pull failed 1422 * or value of new tail of skb in the case of success. 1423 * 1424 * All the pointers pointing into skb header may change and must be 1425 * reloaded after call to this function. 1426 */ 1427 1428 /* Moves tail of skb head forward, copying data from fragmented part, 1429 * when it is necessary. 1430 * 1. It may fail due to malloc failure. 1431 * 2. It may change skb pointers. 1432 * 1433 * It is pretty complicated. Luckily, it is called only in exceptional cases. 1434 */ 1435 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 1436 { 1437 /* If skb has not enough free space at tail, get new one 1438 * plus 128 bytes for future expansions. If we have enough 1439 * room at tail, reallocate without expansion only if skb is cloned. 1440 */ 1441 int i, k, eat = (skb->tail + delta) - skb->end; 1442 1443 if (eat > 0 || skb_cloned(skb)) { 1444 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 1445 GFP_ATOMIC)) 1446 return NULL; 1447 } 1448 1449 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 1450 BUG(); 1451 1452 /* Optimization: no fragments, no reasons to preestimate 1453 * size of pulled pages. Superb. 1454 */ 1455 if (!skb_has_frag_list(skb)) 1456 goto pull_pages; 1457 1458 /* Estimate size of pulled pages. */ 1459 eat = delta; 1460 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1461 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 1462 1463 if (size >= eat) 1464 goto pull_pages; 1465 eat -= size; 1466 } 1467 1468 /* If we need update frag list, we are in troubles. 1469 * Certainly, it possible to add an offset to skb data, 1470 * but taking into account that pulling is expected to 1471 * be very rare operation, it is worth to fight against 1472 * further bloating skb head and crucify ourselves here instead. 1473 * Pure masohism, indeed. 8)8) 1474 */ 1475 if (eat) { 1476 struct sk_buff *list = skb_shinfo(skb)->frag_list; 1477 struct sk_buff *clone = NULL; 1478 struct sk_buff *insp = NULL; 1479 1480 do { 1481 BUG_ON(!list); 1482 1483 if (list->len <= eat) { 1484 /* Eaten as whole. */ 1485 eat -= list->len; 1486 list = list->next; 1487 insp = list; 1488 } else { 1489 /* Eaten partially. */ 1490 1491 if (skb_shared(list)) { 1492 /* Sucks! We need to fork list. :-( */ 1493 clone = skb_clone(list, GFP_ATOMIC); 1494 if (!clone) 1495 return NULL; 1496 insp = list->next; 1497 list = clone; 1498 } else { 1499 /* This may be pulled without 1500 * problems. */ 1501 insp = list; 1502 } 1503 if (!pskb_pull(list, eat)) { 1504 kfree_skb(clone); 1505 return NULL; 1506 } 1507 break; 1508 } 1509 } while (eat); 1510 1511 /* Free pulled out fragments. */ 1512 while ((list = skb_shinfo(skb)->frag_list) != insp) { 1513 skb_shinfo(skb)->frag_list = list->next; 1514 kfree_skb(list); 1515 } 1516 /* And insert new clone at head. */ 1517 if (clone) { 1518 clone->next = list; 1519 skb_shinfo(skb)->frag_list = clone; 1520 } 1521 } 1522 /* Success! Now we may commit changes to skb data. */ 1523 1524 pull_pages: 1525 eat = delta; 1526 k = 0; 1527 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1528 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 1529 1530 if (size <= eat) { 1531 skb_frag_unref(skb, i); 1532 eat -= size; 1533 } else { 1534 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 1535 if (eat) { 1536 skb_shinfo(skb)->frags[k].page_offset += eat; 1537 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 1538 eat = 0; 1539 } 1540 k++; 1541 } 1542 } 1543 skb_shinfo(skb)->nr_frags = k; 1544 1545 skb->tail += delta; 1546 skb->data_len -= delta; 1547 1548 return skb_tail_pointer(skb); 1549 } 1550 EXPORT_SYMBOL(__pskb_pull_tail); 1551 1552 /** 1553 * skb_copy_bits - copy bits from skb to kernel buffer 1554 * @skb: source skb 1555 * @offset: offset in source 1556 * @to: destination buffer 1557 * @len: number of bytes to copy 1558 * 1559 * Copy the specified number of bytes from the source skb to the 1560 * destination buffer. 1561 * 1562 * CAUTION ! : 1563 * If its prototype is ever changed, 1564 * check arch/{*}/net/{*}.S files, 1565 * since it is called from BPF assembly code. 1566 */ 1567 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 1568 { 1569 int start = skb_headlen(skb); 1570 struct sk_buff *frag_iter; 1571 int i, copy; 1572 1573 if (offset > (int)skb->len - len) 1574 goto fault; 1575 1576 /* Copy header. */ 1577 if ((copy = start - offset) > 0) { 1578 if (copy > len) 1579 copy = len; 1580 skb_copy_from_linear_data_offset(skb, offset, to, copy); 1581 if ((len -= copy) == 0) 1582 return 0; 1583 offset += copy; 1584 to += copy; 1585 } 1586 1587 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1588 int end; 1589 skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 1590 1591 WARN_ON(start > offset + len); 1592 1593 end = start + skb_frag_size(f); 1594 if ((copy = end - offset) > 0) { 1595 u8 *vaddr; 1596 1597 if (copy > len) 1598 copy = len; 1599 1600 vaddr = kmap_atomic(skb_frag_page(f)); 1601 memcpy(to, 1602 vaddr + f->page_offset + offset - start, 1603 copy); 1604 kunmap_atomic(vaddr); 1605 1606 if ((len -= copy) == 0) 1607 return 0; 1608 offset += copy; 1609 to += copy; 1610 } 1611 start = end; 1612 } 1613 1614 skb_walk_frags(skb, frag_iter) { 1615 int end; 1616 1617 WARN_ON(start > offset + len); 1618 1619 end = start + frag_iter->len; 1620 if ((copy = end - offset) > 0) { 1621 if (copy > len) 1622 copy = len; 1623 if (skb_copy_bits(frag_iter, offset - start, to, copy)) 1624 goto fault; 1625 if ((len -= copy) == 0) 1626 return 0; 1627 offset += copy; 1628 to += copy; 1629 } 1630 start = end; 1631 } 1632 1633 if (!len) 1634 return 0; 1635 1636 fault: 1637 return -EFAULT; 1638 } 1639 EXPORT_SYMBOL(skb_copy_bits); 1640 1641 /* 1642 * Callback from splice_to_pipe(), if we need to release some pages 1643 * at the end of the spd in case we error'ed out in filling the pipe. 1644 */ 1645 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 1646 { 1647 put_page(spd->pages[i]); 1648 } 1649 1650 static struct page *linear_to_page(struct page *page, unsigned int *len, 1651 unsigned int *offset, 1652 struct sock *sk) 1653 { 1654 struct page_frag *pfrag = sk_page_frag(sk); 1655 1656 if (!sk_page_frag_refill(sk, pfrag)) 1657 return NULL; 1658 1659 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 1660 1661 memcpy(page_address(pfrag->page) + pfrag->offset, 1662 page_address(page) + *offset, *len); 1663 *offset = pfrag->offset; 1664 pfrag->offset += *len; 1665 1666 return pfrag->page; 1667 } 1668 1669 static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 1670 struct page *page, 1671 unsigned int offset) 1672 { 1673 return spd->nr_pages && 1674 spd->pages[spd->nr_pages - 1] == page && 1675 (spd->partial[spd->nr_pages - 1].offset + 1676 spd->partial[spd->nr_pages - 1].len == offset); 1677 } 1678 1679 /* 1680 * Fill page/offset/length into spd, if it can hold more pages. 1681 */ 1682 static bool spd_fill_page(struct splice_pipe_desc *spd, 1683 struct pipe_inode_info *pipe, struct page *page, 1684 unsigned int *len, unsigned int offset, 1685 bool linear, 1686 struct sock *sk) 1687 { 1688 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 1689 return true; 1690 1691 if (linear) { 1692 page = linear_to_page(page, len, &offset, sk); 1693 if (!page) 1694 return true; 1695 } 1696 if (spd_can_coalesce(spd, page, offset)) { 1697 spd->partial[spd->nr_pages - 1].len += *len; 1698 return false; 1699 } 1700 get_page(page); 1701 spd->pages[spd->nr_pages] = page; 1702 spd->partial[spd->nr_pages].len = *len; 1703 spd->partial[spd->nr_pages].offset = offset; 1704 spd->nr_pages++; 1705 1706 return false; 1707 } 1708 1709 static bool __splice_segment(struct page *page, unsigned int poff, 1710 unsigned int plen, unsigned int *off, 1711 unsigned int *len, 1712 struct splice_pipe_desc *spd, bool linear, 1713 struct sock *sk, 1714 struct pipe_inode_info *pipe) 1715 { 1716 if (!*len) 1717 return true; 1718 1719 /* skip this segment if already processed */ 1720 if (*off >= plen) { 1721 *off -= plen; 1722 return false; 1723 } 1724 1725 /* ignore any bits we already processed */ 1726 poff += *off; 1727 plen -= *off; 1728 *off = 0; 1729 1730 do { 1731 unsigned int flen = min(*len, plen); 1732 1733 if (spd_fill_page(spd, pipe, page, &flen, poff, 1734 linear, sk)) 1735 return true; 1736 poff += flen; 1737 plen -= flen; 1738 *len -= flen; 1739 } while (*len && plen); 1740 1741 return false; 1742 } 1743 1744 /* 1745 * Map linear and fragment data from the skb to spd. It reports true if the 1746 * pipe is full or if we already spliced the requested length. 1747 */ 1748 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 1749 unsigned int *offset, unsigned int *len, 1750 struct splice_pipe_desc *spd, struct sock *sk) 1751 { 1752 int seg; 1753 1754 /* map the linear part : 1755 * If skb->head_frag is set, this 'linear' part is backed by a 1756 * fragment, and if the head is not shared with any clones then 1757 * we can avoid a copy since we own the head portion of this page. 1758 */ 1759 if (__splice_segment(virt_to_page(skb->data), 1760 (unsigned long) skb->data & (PAGE_SIZE - 1), 1761 skb_headlen(skb), 1762 offset, len, spd, 1763 skb_head_is_locked(skb), 1764 sk, pipe)) 1765 return true; 1766 1767 /* 1768 * then map the fragments 1769 */ 1770 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 1771 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 1772 1773 if (__splice_segment(skb_frag_page(f), 1774 f->page_offset, skb_frag_size(f), 1775 offset, len, spd, false, sk, pipe)) 1776 return true; 1777 } 1778 1779 return false; 1780 } 1781 1782 /* 1783 * Map data from the skb to a pipe. Should handle both the linear part, 1784 * the fragments, and the frag list. It does NOT handle frag lists within 1785 * the frag list, if such a thing exists. We'd probably need to recurse to 1786 * handle that cleanly. 1787 */ 1788 int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 1789 struct pipe_inode_info *pipe, unsigned int tlen, 1790 unsigned int flags) 1791 { 1792 struct partial_page partial[MAX_SKB_FRAGS]; 1793 struct page *pages[MAX_SKB_FRAGS]; 1794 struct splice_pipe_desc spd = { 1795 .pages = pages, 1796 .partial = partial, 1797 .nr_pages_max = MAX_SKB_FRAGS, 1798 .flags = flags, 1799 .ops = &sock_pipe_buf_ops, 1800 .spd_release = sock_spd_release, 1801 }; 1802 struct sk_buff *frag_iter; 1803 struct sock *sk = skb->sk; 1804 int ret = 0; 1805 1806 /* 1807 * __skb_splice_bits() only fails if the output has no room left, 1808 * so no point in going over the frag_list for the error case. 1809 */ 1810 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk)) 1811 goto done; 1812 else if (!tlen) 1813 goto done; 1814 1815 /* 1816 * now see if we have a frag_list to map 1817 */ 1818 skb_walk_frags(skb, frag_iter) { 1819 if (!tlen) 1820 break; 1821 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk)) 1822 break; 1823 } 1824 1825 done: 1826 if (spd.nr_pages) { 1827 /* 1828 * Drop the socket lock, otherwise we have reverse 1829 * locking dependencies between sk_lock and i_mutex 1830 * here as compared to sendfile(). We enter here 1831 * with the socket lock held, and splice_to_pipe() will 1832 * grab the pipe inode lock. For sendfile() emulation, 1833 * we call into ->sendpage() with the i_mutex lock held 1834 * and networking will grab the socket lock. 1835 */ 1836 release_sock(sk); 1837 ret = splice_to_pipe(pipe, &spd); 1838 lock_sock(sk); 1839 } 1840 1841 return ret; 1842 } 1843 1844 /** 1845 * skb_store_bits - store bits from kernel buffer to skb 1846 * @skb: destination buffer 1847 * @offset: offset in destination 1848 * @from: source buffer 1849 * @len: number of bytes to copy 1850 * 1851 * Copy the specified number of bytes from the source buffer to the 1852 * destination skb. This function handles all the messy bits of 1853 * traversing fragment lists and such. 1854 */ 1855 1856 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1857 { 1858 int start = skb_headlen(skb); 1859 struct sk_buff *frag_iter; 1860 int i, copy; 1861 1862 if (offset > (int)skb->len - len) 1863 goto fault; 1864 1865 if ((copy = start - offset) > 0) { 1866 if (copy > len) 1867 copy = len; 1868 skb_copy_to_linear_data_offset(skb, offset, from, copy); 1869 if ((len -= copy) == 0) 1870 return 0; 1871 offset += copy; 1872 from += copy; 1873 } 1874 1875 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1876 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1877 int end; 1878 1879 WARN_ON(start > offset + len); 1880 1881 end = start + skb_frag_size(frag); 1882 if ((copy = end - offset) > 0) { 1883 u8 *vaddr; 1884 1885 if (copy > len) 1886 copy = len; 1887 1888 vaddr = kmap_atomic(skb_frag_page(frag)); 1889 memcpy(vaddr + frag->page_offset + offset - start, 1890 from, copy); 1891 kunmap_atomic(vaddr); 1892 1893 if ((len -= copy) == 0) 1894 return 0; 1895 offset += copy; 1896 from += copy; 1897 } 1898 start = end; 1899 } 1900 1901 skb_walk_frags(skb, frag_iter) { 1902 int end; 1903 1904 WARN_ON(start > offset + len); 1905 1906 end = start + frag_iter->len; 1907 if ((copy = end - offset) > 0) { 1908 if (copy > len) 1909 copy = len; 1910 if (skb_store_bits(frag_iter, offset - start, 1911 from, copy)) 1912 goto fault; 1913 if ((len -= copy) == 0) 1914 return 0; 1915 offset += copy; 1916 from += copy; 1917 } 1918 start = end; 1919 } 1920 if (!len) 1921 return 0; 1922 1923 fault: 1924 return -EFAULT; 1925 } 1926 EXPORT_SYMBOL(skb_store_bits); 1927 1928 /* Checksum skb data. */ 1929 1930 __wsum skb_checksum(const struct sk_buff *skb, int offset, 1931 int len, __wsum csum) 1932 { 1933 int start = skb_headlen(skb); 1934 int i, copy = start - offset; 1935 struct sk_buff *frag_iter; 1936 int pos = 0; 1937 1938 /* Checksum header. */ 1939 if (copy > 0) { 1940 if (copy > len) 1941 copy = len; 1942 csum = csum_partial(skb->data + offset, copy, csum); 1943 if ((len -= copy) == 0) 1944 return csum; 1945 offset += copy; 1946 pos = copy; 1947 } 1948 1949 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1950 int end; 1951 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1952 1953 WARN_ON(start > offset + len); 1954 1955 end = start + skb_frag_size(frag); 1956 if ((copy = end - offset) > 0) { 1957 __wsum csum2; 1958 u8 *vaddr; 1959 1960 if (copy > len) 1961 copy = len; 1962 vaddr = kmap_atomic(skb_frag_page(frag)); 1963 csum2 = csum_partial(vaddr + frag->page_offset + 1964 offset - start, copy, 0); 1965 kunmap_atomic(vaddr); 1966 csum = csum_block_add(csum, csum2, pos); 1967 if (!(len -= copy)) 1968 return csum; 1969 offset += copy; 1970 pos += copy; 1971 } 1972 start = end; 1973 } 1974 1975 skb_walk_frags(skb, frag_iter) { 1976 int end; 1977 1978 WARN_ON(start > offset + len); 1979 1980 end = start + frag_iter->len; 1981 if ((copy = end - offset) > 0) { 1982 __wsum csum2; 1983 if (copy > len) 1984 copy = len; 1985 csum2 = skb_checksum(frag_iter, offset - start, 1986 copy, 0); 1987 csum = csum_block_add(csum, csum2, pos); 1988 if ((len -= copy) == 0) 1989 return csum; 1990 offset += copy; 1991 pos += copy; 1992 } 1993 start = end; 1994 } 1995 BUG_ON(len); 1996 1997 return csum; 1998 } 1999 EXPORT_SYMBOL(skb_checksum); 2000 2001 /* Both of above in one bottle. */ 2002 2003 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 2004 u8 *to, int len, __wsum csum) 2005 { 2006 int start = skb_headlen(skb); 2007 int i, copy = start - offset; 2008 struct sk_buff *frag_iter; 2009 int pos = 0; 2010 2011 /* Copy header. */ 2012 if (copy > 0) { 2013 if (copy > len) 2014 copy = len; 2015 csum = csum_partial_copy_nocheck(skb->data + offset, to, 2016 copy, csum); 2017 if ((len -= copy) == 0) 2018 return csum; 2019 offset += copy; 2020 to += copy; 2021 pos = copy; 2022 } 2023 2024 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2025 int end; 2026 2027 WARN_ON(start > offset + len); 2028 2029 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 2030 if ((copy = end - offset) > 0) { 2031 __wsum csum2; 2032 u8 *vaddr; 2033 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2034 2035 if (copy > len) 2036 copy = len; 2037 vaddr = kmap_atomic(skb_frag_page(frag)); 2038 csum2 = csum_partial_copy_nocheck(vaddr + 2039 frag->page_offset + 2040 offset - start, to, 2041 copy, 0); 2042 kunmap_atomic(vaddr); 2043 csum = csum_block_add(csum, csum2, pos); 2044 if (!(len -= copy)) 2045 return csum; 2046 offset += copy; 2047 to += copy; 2048 pos += copy; 2049 } 2050 start = end; 2051 } 2052 2053 skb_walk_frags(skb, frag_iter) { 2054 __wsum csum2; 2055 int end; 2056 2057 WARN_ON(start > offset + len); 2058 2059 end = start + frag_iter->len; 2060 if ((copy = end - offset) > 0) { 2061 if (copy > len) 2062 copy = len; 2063 csum2 = skb_copy_and_csum_bits(frag_iter, 2064 offset - start, 2065 to, copy, 0); 2066 csum = csum_block_add(csum, csum2, pos); 2067 if ((len -= copy) == 0) 2068 return csum; 2069 offset += copy; 2070 to += copy; 2071 pos += copy; 2072 } 2073 start = end; 2074 } 2075 BUG_ON(len); 2076 return csum; 2077 } 2078 EXPORT_SYMBOL(skb_copy_and_csum_bits); 2079 2080 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 2081 { 2082 __wsum csum; 2083 long csstart; 2084 2085 if (skb->ip_summed == CHECKSUM_PARTIAL) 2086 csstart = skb_checksum_start_offset(skb); 2087 else 2088 csstart = skb_headlen(skb); 2089 2090 BUG_ON(csstart > skb_headlen(skb)); 2091 2092 skb_copy_from_linear_data(skb, to, csstart); 2093 2094 csum = 0; 2095 if (csstart != skb->len) 2096 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 2097 skb->len - csstart, 0); 2098 2099 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2100 long csstuff = csstart + skb->csum_offset; 2101 2102 *((__sum16 *)(to + csstuff)) = csum_fold(csum); 2103 } 2104 } 2105 EXPORT_SYMBOL(skb_copy_and_csum_dev); 2106 2107 /** 2108 * skb_dequeue - remove from the head of the queue 2109 * @list: list to dequeue from 2110 * 2111 * Remove the head of the list. The list lock is taken so the function 2112 * may be used safely with other locking list functions. The head item is 2113 * returned or %NULL if the list is empty. 2114 */ 2115 2116 struct sk_buff *skb_dequeue(struct sk_buff_head *list) 2117 { 2118 unsigned long flags; 2119 struct sk_buff *result; 2120 2121 spin_lock_irqsave(&list->lock, flags); 2122 result = __skb_dequeue(list); 2123 spin_unlock_irqrestore(&list->lock, flags); 2124 return result; 2125 } 2126 EXPORT_SYMBOL(skb_dequeue); 2127 2128 /** 2129 * skb_dequeue_tail - remove from the tail of the queue 2130 * @list: list to dequeue from 2131 * 2132 * Remove the tail of the list. The list lock is taken so the function 2133 * may be used safely with other locking list functions. The tail item is 2134 * returned or %NULL if the list is empty. 2135 */ 2136 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 2137 { 2138 unsigned long flags; 2139 struct sk_buff *result; 2140 2141 spin_lock_irqsave(&list->lock, flags); 2142 result = __skb_dequeue_tail(list); 2143 spin_unlock_irqrestore(&list->lock, flags); 2144 return result; 2145 } 2146 EXPORT_SYMBOL(skb_dequeue_tail); 2147 2148 /** 2149 * skb_queue_purge - empty a list 2150 * @list: list to empty 2151 * 2152 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2153 * the list and one reference dropped. This function takes the list 2154 * lock and is atomic with respect to other list locking functions. 2155 */ 2156 void skb_queue_purge(struct sk_buff_head *list) 2157 { 2158 struct sk_buff *skb; 2159 while ((skb = skb_dequeue(list)) != NULL) 2160 kfree_skb(skb); 2161 } 2162 EXPORT_SYMBOL(skb_queue_purge); 2163 2164 /** 2165 * skb_queue_head - queue a buffer at the list head 2166 * @list: list to use 2167 * @newsk: buffer to queue 2168 * 2169 * Queue a buffer at the start of the list. This function takes the 2170 * list lock and can be used safely with other locking &sk_buff functions 2171 * safely. 2172 * 2173 * A buffer cannot be placed on two lists at the same time. 2174 */ 2175 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 2176 { 2177 unsigned long flags; 2178 2179 spin_lock_irqsave(&list->lock, flags); 2180 __skb_queue_head(list, newsk); 2181 spin_unlock_irqrestore(&list->lock, flags); 2182 } 2183 EXPORT_SYMBOL(skb_queue_head); 2184 2185 /** 2186 * skb_queue_tail - queue a buffer at the list tail 2187 * @list: list to use 2188 * @newsk: buffer to queue 2189 * 2190 * Queue a buffer at the tail of the list. This function takes the 2191 * list lock and can be used safely with other locking &sk_buff functions 2192 * safely. 2193 * 2194 * A buffer cannot be placed on two lists at the same time. 2195 */ 2196 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 2197 { 2198 unsigned long flags; 2199 2200 spin_lock_irqsave(&list->lock, flags); 2201 __skb_queue_tail(list, newsk); 2202 spin_unlock_irqrestore(&list->lock, flags); 2203 } 2204 EXPORT_SYMBOL(skb_queue_tail); 2205 2206 /** 2207 * skb_unlink - remove a buffer from a list 2208 * @skb: buffer to remove 2209 * @list: list to use 2210 * 2211 * Remove a packet from a list. The list locks are taken and this 2212 * function is atomic with respect to other list locked calls 2213 * 2214 * You must know what list the SKB is on. 2215 */ 2216 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2217 { 2218 unsigned long flags; 2219 2220 spin_lock_irqsave(&list->lock, flags); 2221 __skb_unlink(skb, list); 2222 spin_unlock_irqrestore(&list->lock, flags); 2223 } 2224 EXPORT_SYMBOL(skb_unlink); 2225 2226 /** 2227 * skb_append - append a buffer 2228 * @old: buffer to insert after 2229 * @newsk: buffer to insert 2230 * @list: list to use 2231 * 2232 * Place a packet after a given packet in a list. The list locks are taken 2233 * and this function is atomic with respect to other list locked calls. 2234 * A buffer cannot be placed on two lists at the same time. 2235 */ 2236 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 2237 { 2238 unsigned long flags; 2239 2240 spin_lock_irqsave(&list->lock, flags); 2241 __skb_queue_after(list, old, newsk); 2242 spin_unlock_irqrestore(&list->lock, flags); 2243 } 2244 EXPORT_SYMBOL(skb_append); 2245 2246 /** 2247 * skb_insert - insert a buffer 2248 * @old: buffer to insert before 2249 * @newsk: buffer to insert 2250 * @list: list to use 2251 * 2252 * Place a packet before a given packet in a list. The list locks are 2253 * taken and this function is atomic with respect to other list locked 2254 * calls. 2255 * 2256 * A buffer cannot be placed on two lists at the same time. 2257 */ 2258 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 2259 { 2260 unsigned long flags; 2261 2262 spin_lock_irqsave(&list->lock, flags); 2263 __skb_insert(newsk, old->prev, old, list); 2264 spin_unlock_irqrestore(&list->lock, flags); 2265 } 2266 EXPORT_SYMBOL(skb_insert); 2267 2268 static inline void skb_split_inside_header(struct sk_buff *skb, 2269 struct sk_buff* skb1, 2270 const u32 len, const int pos) 2271 { 2272 int i; 2273 2274 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2275 pos - len); 2276 /* And move data appendix as is. */ 2277 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 2278 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 2279 2280 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 2281 skb_shinfo(skb)->nr_frags = 0; 2282 skb1->data_len = skb->data_len; 2283 skb1->len += skb1->data_len; 2284 skb->data_len = 0; 2285 skb->len = len; 2286 skb_set_tail_pointer(skb, len); 2287 } 2288 2289 static inline void skb_split_no_header(struct sk_buff *skb, 2290 struct sk_buff* skb1, 2291 const u32 len, int pos) 2292 { 2293 int i, k = 0; 2294 const int nfrags = skb_shinfo(skb)->nr_frags; 2295 2296 skb_shinfo(skb)->nr_frags = 0; 2297 skb1->len = skb1->data_len = skb->len - len; 2298 skb->len = len; 2299 skb->data_len = len - pos; 2300 2301 for (i = 0; i < nfrags; i++) { 2302 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 2303 2304 if (pos + size > len) { 2305 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 2306 2307 if (pos < len) { 2308 /* Split frag. 2309 * We have two variants in this case: 2310 * 1. Move all the frag to the second 2311 * part, if it is possible. F.e. 2312 * this approach is mandatory for TUX, 2313 * where splitting is expensive. 2314 * 2. Split is accurately. We make this. 2315 */ 2316 skb_frag_ref(skb, i); 2317 skb_shinfo(skb1)->frags[0].page_offset += len - pos; 2318 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 2319 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 2320 skb_shinfo(skb)->nr_frags++; 2321 } 2322 k++; 2323 } else 2324 skb_shinfo(skb)->nr_frags++; 2325 pos += size; 2326 } 2327 skb_shinfo(skb1)->nr_frags = k; 2328 } 2329 2330 /** 2331 * skb_split - Split fragmented skb to two parts at length len. 2332 * @skb: the buffer to split 2333 * @skb1: the buffer to receive the second part 2334 * @len: new length for skb 2335 */ 2336 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 2337 { 2338 int pos = skb_headlen(skb); 2339 2340 if (len < pos) /* Split line is inside header. */ 2341 skb_split_inside_header(skb, skb1, len, pos); 2342 else /* Second chunk has no header, nothing to copy. */ 2343 skb_split_no_header(skb, skb1, len, pos); 2344 } 2345 EXPORT_SYMBOL(skb_split); 2346 2347 /* Shifting from/to a cloned skb is a no-go. 2348 * 2349 * Caller cannot keep skb_shinfo related pointers past calling here! 2350 */ 2351 static int skb_prepare_for_shift(struct sk_buff *skb) 2352 { 2353 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2354 } 2355 2356 /** 2357 * skb_shift - Shifts paged data partially from skb to another 2358 * @tgt: buffer into which tail data gets added 2359 * @skb: buffer from which the paged data comes from 2360 * @shiftlen: shift up to this many bytes 2361 * 2362 * Attempts to shift up to shiftlen worth of bytes, which may be less than 2363 * the length of the skb, from skb to tgt. Returns number bytes shifted. 2364 * It's up to caller to free skb if everything was shifted. 2365 * 2366 * If @tgt runs out of frags, the whole operation is aborted. 2367 * 2368 * Skb cannot include anything else but paged data while tgt is allowed 2369 * to have non-paged data as well. 2370 * 2371 * TODO: full sized shift could be optimized but that would need 2372 * specialized skb free'er to handle frags without up-to-date nr_frags. 2373 */ 2374 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2375 { 2376 int from, to, merge, todo; 2377 struct skb_frag_struct *fragfrom, *fragto; 2378 2379 BUG_ON(shiftlen > skb->len); 2380 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2381 2382 todo = shiftlen; 2383 from = 0; 2384 to = skb_shinfo(tgt)->nr_frags; 2385 fragfrom = &skb_shinfo(skb)->frags[from]; 2386 2387 /* Actual merge is delayed until the point when we know we can 2388 * commit all, so that we don't have to undo partial changes 2389 */ 2390 if (!to || 2391 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 2392 fragfrom->page_offset)) { 2393 merge = -1; 2394 } else { 2395 merge = to - 1; 2396 2397 todo -= skb_frag_size(fragfrom); 2398 if (todo < 0) { 2399 if (skb_prepare_for_shift(skb) || 2400 skb_prepare_for_shift(tgt)) 2401 return 0; 2402 2403 /* All previous frag pointers might be stale! */ 2404 fragfrom = &skb_shinfo(skb)->frags[from]; 2405 fragto = &skb_shinfo(tgt)->frags[merge]; 2406 2407 skb_frag_size_add(fragto, shiftlen); 2408 skb_frag_size_sub(fragfrom, shiftlen); 2409 fragfrom->page_offset += shiftlen; 2410 2411 goto onlymerged; 2412 } 2413 2414 from++; 2415 } 2416 2417 /* Skip full, not-fitting skb to avoid expensive operations */ 2418 if ((shiftlen == skb->len) && 2419 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2420 return 0; 2421 2422 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2423 return 0; 2424 2425 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2426 if (to == MAX_SKB_FRAGS) 2427 return 0; 2428 2429 fragfrom = &skb_shinfo(skb)->frags[from]; 2430 fragto = &skb_shinfo(tgt)->frags[to]; 2431 2432 if (todo >= skb_frag_size(fragfrom)) { 2433 *fragto = *fragfrom; 2434 todo -= skb_frag_size(fragfrom); 2435 from++; 2436 to++; 2437 2438 } else { 2439 __skb_frag_ref(fragfrom); 2440 fragto->page = fragfrom->page; 2441 fragto->page_offset = fragfrom->page_offset; 2442 skb_frag_size_set(fragto, todo); 2443 2444 fragfrom->page_offset += todo; 2445 skb_frag_size_sub(fragfrom, todo); 2446 todo = 0; 2447 2448 to++; 2449 break; 2450 } 2451 } 2452 2453 /* Ready to "commit" this state change to tgt */ 2454 skb_shinfo(tgt)->nr_frags = to; 2455 2456 if (merge >= 0) { 2457 fragfrom = &skb_shinfo(skb)->frags[0]; 2458 fragto = &skb_shinfo(tgt)->frags[merge]; 2459 2460 skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 2461 __skb_frag_unref(fragfrom); 2462 } 2463 2464 /* Reposition in the original skb */ 2465 to = 0; 2466 while (from < skb_shinfo(skb)->nr_frags) 2467 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2468 skb_shinfo(skb)->nr_frags = to; 2469 2470 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2471 2472 onlymerged: 2473 /* Most likely the tgt won't ever need its checksum anymore, skb on 2474 * the other hand might need it if it needs to be resent 2475 */ 2476 tgt->ip_summed = CHECKSUM_PARTIAL; 2477 skb->ip_summed = CHECKSUM_PARTIAL; 2478 2479 /* Yak, is it really working this way? Some helper please? */ 2480 skb->len -= shiftlen; 2481 skb->data_len -= shiftlen; 2482 skb->truesize -= shiftlen; 2483 tgt->len += shiftlen; 2484 tgt->data_len += shiftlen; 2485 tgt->truesize += shiftlen; 2486 2487 return shiftlen; 2488 } 2489 2490 /** 2491 * skb_prepare_seq_read - Prepare a sequential read of skb data 2492 * @skb: the buffer to read 2493 * @from: lower offset of data to be read 2494 * @to: upper offset of data to be read 2495 * @st: state variable 2496 * 2497 * Initializes the specified state variable. Must be called before 2498 * invoking skb_seq_read() for the first time. 2499 */ 2500 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2501 unsigned int to, struct skb_seq_state *st) 2502 { 2503 st->lower_offset = from; 2504 st->upper_offset = to; 2505 st->root_skb = st->cur_skb = skb; 2506 st->frag_idx = st->stepped_offset = 0; 2507 st->frag_data = NULL; 2508 } 2509 EXPORT_SYMBOL(skb_prepare_seq_read); 2510 2511 /** 2512 * skb_seq_read - Sequentially read skb data 2513 * @consumed: number of bytes consumed by the caller so far 2514 * @data: destination pointer for data to be returned 2515 * @st: state variable 2516 * 2517 * Reads a block of skb data at &consumed relative to the 2518 * lower offset specified to skb_prepare_seq_read(). Assigns 2519 * the head of the data block to &data and returns the length 2520 * of the block or 0 if the end of the skb data or the upper 2521 * offset has been reached. 2522 * 2523 * The caller is not required to consume all of the data 2524 * returned, i.e. &consumed is typically set to the number 2525 * of bytes already consumed and the next call to 2526 * skb_seq_read() will return the remaining part of the block. 2527 * 2528 * Note 1: The size of each block of data returned can be arbitrary, 2529 * this limitation is the cost for zerocopy seqeuental 2530 * reads of potentially non linear data. 2531 * 2532 * Note 2: Fragment lists within fragments are not implemented 2533 * at the moment, state->root_skb could be replaced with 2534 * a stack for this purpose. 2535 */ 2536 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2537 struct skb_seq_state *st) 2538 { 2539 unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2540 skb_frag_t *frag; 2541 2542 if (unlikely(abs_offset >= st->upper_offset)) 2543 return 0; 2544 2545 next_skb: 2546 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2547 2548 if (abs_offset < block_limit && !st->frag_data) { 2549 *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2550 return block_limit - abs_offset; 2551 } 2552 2553 if (st->frag_idx == 0 && !st->frag_data) 2554 st->stepped_offset += skb_headlen(st->cur_skb); 2555 2556 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2557 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 2558 block_limit = skb_frag_size(frag) + st->stepped_offset; 2559 2560 if (abs_offset < block_limit) { 2561 if (!st->frag_data) 2562 st->frag_data = kmap_atomic(skb_frag_page(frag)); 2563 2564 *data = (u8 *) st->frag_data + frag->page_offset + 2565 (abs_offset - st->stepped_offset); 2566 2567 return block_limit - abs_offset; 2568 } 2569 2570 if (st->frag_data) { 2571 kunmap_atomic(st->frag_data); 2572 st->frag_data = NULL; 2573 } 2574 2575 st->frag_idx++; 2576 st->stepped_offset += skb_frag_size(frag); 2577 } 2578 2579 if (st->frag_data) { 2580 kunmap_atomic(st->frag_data); 2581 st->frag_data = NULL; 2582 } 2583 2584 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 2585 st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 2586 st->frag_idx = 0; 2587 goto next_skb; 2588 } else if (st->cur_skb->next) { 2589 st->cur_skb = st->cur_skb->next; 2590 st->frag_idx = 0; 2591 goto next_skb; 2592 } 2593 2594 return 0; 2595 } 2596 EXPORT_SYMBOL(skb_seq_read); 2597 2598 /** 2599 * skb_abort_seq_read - Abort a sequential read of skb data 2600 * @st: state variable 2601 * 2602 * Must be called if skb_seq_read() was not called until it 2603 * returned 0. 2604 */ 2605 void skb_abort_seq_read(struct skb_seq_state *st) 2606 { 2607 if (st->frag_data) 2608 kunmap_atomic(st->frag_data); 2609 } 2610 EXPORT_SYMBOL(skb_abort_seq_read); 2611 2612 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 2613 2614 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 2615 struct ts_config *conf, 2616 struct ts_state *state) 2617 { 2618 return skb_seq_read(offset, text, TS_SKB_CB(state)); 2619 } 2620 2621 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 2622 { 2623 skb_abort_seq_read(TS_SKB_CB(state)); 2624 } 2625 2626 /** 2627 * skb_find_text - Find a text pattern in skb data 2628 * @skb: the buffer to look in 2629 * @from: search offset 2630 * @to: search limit 2631 * @config: textsearch configuration 2632 * @state: uninitialized textsearch state variable 2633 * 2634 * Finds a pattern in the skb data according to the specified 2635 * textsearch configuration. Use textsearch_next() to retrieve 2636 * subsequent occurrences of the pattern. Returns the offset 2637 * to the first occurrence or UINT_MAX if no match was found. 2638 */ 2639 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 2640 unsigned int to, struct ts_config *config, 2641 struct ts_state *state) 2642 { 2643 unsigned int ret; 2644 2645 config->get_next_block = skb_ts_get_next_block; 2646 config->finish = skb_ts_finish; 2647 2648 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 2649 2650 ret = textsearch_find(config, state); 2651 return (ret <= to - from ? ret : UINT_MAX); 2652 } 2653 EXPORT_SYMBOL(skb_find_text); 2654 2655 /** 2656 * skb_append_datato_frags - append the user data to a skb 2657 * @sk: sock structure 2658 * @skb: skb structure to be appened with user data. 2659 * @getfrag: call back function to be used for getting the user data 2660 * @from: pointer to user message iov 2661 * @length: length of the iov message 2662 * 2663 * Description: This procedure append the user data in the fragment part 2664 * of the skb if any page alloc fails user this procedure returns -ENOMEM 2665 */ 2666 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2667 int (*getfrag)(void *from, char *to, int offset, 2668 int len, int odd, struct sk_buff *skb), 2669 void *from, int length) 2670 { 2671 int frg_cnt = 0; 2672 skb_frag_t *frag = NULL; 2673 struct page *page = NULL; 2674 int copy, left; 2675 int offset = 0; 2676 int ret; 2677 2678 do { 2679 /* Return error if we don't have space for new frag */ 2680 frg_cnt = skb_shinfo(skb)->nr_frags; 2681 if (frg_cnt >= MAX_SKB_FRAGS) 2682 return -EFAULT; 2683 2684 /* allocate a new page for next frag */ 2685 page = alloc_pages(sk->sk_allocation, 0); 2686 2687 /* If alloc_page fails just return failure and caller will 2688 * free previous allocated pages by doing kfree_skb() 2689 */ 2690 if (page == NULL) 2691 return -ENOMEM; 2692 2693 /* initialize the next frag */ 2694 skb_fill_page_desc(skb, frg_cnt, page, 0, 0); 2695 skb->truesize += PAGE_SIZE; 2696 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 2697 2698 /* get the new initialized frag */ 2699 frg_cnt = skb_shinfo(skb)->nr_frags; 2700 frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; 2701 2702 /* copy the user data to page */ 2703 left = PAGE_SIZE - frag->page_offset; 2704 copy = (length > left)? left : length; 2705 2706 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag), 2707 offset, copy, 0, skb); 2708 if (ret < 0) 2709 return -EFAULT; 2710 2711 /* copy was successful so update the size parameters */ 2712 skb_frag_size_add(frag, copy); 2713 skb->len += copy; 2714 skb->data_len += copy; 2715 offset += copy; 2716 length -= copy; 2717 2718 } while (length > 0); 2719 2720 return 0; 2721 } 2722 EXPORT_SYMBOL(skb_append_datato_frags); 2723 2724 /** 2725 * skb_pull_rcsum - pull skb and update receive checksum 2726 * @skb: buffer to update 2727 * @len: length of data pulled 2728 * 2729 * This function performs an skb_pull on the packet and updates 2730 * the CHECKSUM_COMPLETE checksum. It should be used on 2731 * receive path processing instead of skb_pull unless you know 2732 * that the checksum difference is zero (e.g., a valid IP header) 2733 * or you are setting ip_summed to CHECKSUM_NONE. 2734 */ 2735 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2736 { 2737 BUG_ON(len > skb->len); 2738 skb->len -= len; 2739 BUG_ON(skb->len < skb->data_len); 2740 skb_postpull_rcsum(skb, skb->data, len); 2741 return skb->data += len; 2742 } 2743 EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2744 2745 /** 2746 * skb_segment - Perform protocol segmentation on skb. 2747 * @skb: buffer to segment 2748 * @features: features for the output path (see dev->features) 2749 * 2750 * This function performs segmentation on the given skb. It returns 2751 * a pointer to the first in a list of new skbs for the segments. 2752 * In case of error it returns ERR_PTR(err). 2753 */ 2754 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features) 2755 { 2756 struct sk_buff *segs = NULL; 2757 struct sk_buff *tail = NULL; 2758 struct sk_buff *fskb = skb_shinfo(skb)->frag_list; 2759 unsigned int mss = skb_shinfo(skb)->gso_size; 2760 unsigned int doffset = skb->data - skb_mac_header(skb); 2761 unsigned int offset = doffset; 2762 unsigned int headroom; 2763 unsigned int len; 2764 int sg = !!(features & NETIF_F_SG); 2765 int nfrags = skb_shinfo(skb)->nr_frags; 2766 int err = -ENOMEM; 2767 int i = 0; 2768 int pos; 2769 2770 __skb_push(skb, doffset); 2771 headroom = skb_headroom(skb); 2772 pos = skb_headlen(skb); 2773 2774 do { 2775 struct sk_buff *nskb; 2776 skb_frag_t *frag; 2777 int hsize; 2778 int size; 2779 2780 len = skb->len - offset; 2781 if (len > mss) 2782 len = mss; 2783 2784 hsize = skb_headlen(skb) - offset; 2785 if (hsize < 0) 2786 hsize = 0; 2787 if (hsize > len || !sg) 2788 hsize = len; 2789 2790 if (!hsize && i >= nfrags) { 2791 BUG_ON(fskb->len != len); 2792 2793 pos += len; 2794 nskb = skb_clone(fskb, GFP_ATOMIC); 2795 fskb = fskb->next; 2796 2797 if (unlikely(!nskb)) 2798 goto err; 2799 2800 hsize = skb_end_offset(nskb); 2801 if (skb_cow_head(nskb, doffset + headroom)) { 2802 kfree_skb(nskb); 2803 goto err; 2804 } 2805 2806 nskb->truesize += skb_end_offset(nskb) - hsize; 2807 skb_release_head_state(nskb); 2808 __skb_push(nskb, doffset); 2809 } else { 2810 nskb = __alloc_skb(hsize + doffset + headroom, 2811 GFP_ATOMIC, skb_alloc_rx_flag(skb), 2812 NUMA_NO_NODE); 2813 2814 if (unlikely(!nskb)) 2815 goto err; 2816 2817 skb_reserve(nskb, headroom); 2818 __skb_put(nskb, doffset); 2819 } 2820 2821 if (segs) 2822 tail->next = nskb; 2823 else 2824 segs = nskb; 2825 tail = nskb; 2826 2827 __copy_skb_header(nskb, skb); 2828 nskb->mac_len = skb->mac_len; 2829 2830 /* nskb and skb might have different headroom */ 2831 if (nskb->ip_summed == CHECKSUM_PARTIAL) 2832 nskb->csum_start += skb_headroom(nskb) - headroom; 2833 2834 skb_reset_mac_header(nskb); 2835 skb_set_network_header(nskb, skb->mac_len); 2836 nskb->transport_header = (nskb->network_header + 2837 skb_network_header_len(skb)); 2838 skb_copy_from_linear_data(skb, nskb->data, doffset); 2839 2840 if (fskb != skb_shinfo(skb)->frag_list) 2841 continue; 2842 2843 if (!sg) { 2844 nskb->ip_summed = CHECKSUM_NONE; 2845 nskb->csum = skb_copy_and_csum_bits(skb, offset, 2846 skb_put(nskb, len), 2847 len, 0); 2848 continue; 2849 } 2850 2851 frag = skb_shinfo(nskb)->frags; 2852 2853 skb_copy_from_linear_data_offset(skb, offset, 2854 skb_put(nskb, hsize), hsize); 2855 2856 while (pos < offset + len && i < nfrags) { 2857 *frag = skb_shinfo(skb)->frags[i]; 2858 __skb_frag_ref(frag); 2859 size = skb_frag_size(frag); 2860 2861 if (pos < offset) { 2862 frag->page_offset += offset - pos; 2863 skb_frag_size_sub(frag, offset - pos); 2864 } 2865 2866 skb_shinfo(nskb)->nr_frags++; 2867 2868 if (pos + size <= offset + len) { 2869 i++; 2870 pos += size; 2871 } else { 2872 skb_frag_size_sub(frag, pos + size - (offset + len)); 2873 goto skip_fraglist; 2874 } 2875 2876 frag++; 2877 } 2878 2879 if (pos < offset + len) { 2880 struct sk_buff *fskb2 = fskb; 2881 2882 BUG_ON(pos + fskb->len != offset + len); 2883 2884 pos += fskb->len; 2885 fskb = fskb->next; 2886 2887 if (fskb2->next) { 2888 fskb2 = skb_clone(fskb2, GFP_ATOMIC); 2889 if (!fskb2) 2890 goto err; 2891 } else 2892 skb_get(fskb2); 2893 2894 SKB_FRAG_ASSERT(nskb); 2895 skb_shinfo(nskb)->frag_list = fskb2; 2896 } 2897 2898 skip_fraglist: 2899 nskb->data_len = len - hsize; 2900 nskb->len += nskb->data_len; 2901 nskb->truesize += nskb->data_len; 2902 } while ((offset += len) < skb->len); 2903 2904 return segs; 2905 2906 err: 2907 while ((skb = segs)) { 2908 segs = skb->next; 2909 kfree_skb(skb); 2910 } 2911 return ERR_PTR(err); 2912 } 2913 EXPORT_SYMBOL_GPL(skb_segment); 2914 2915 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 2916 { 2917 struct sk_buff *p = *head; 2918 struct sk_buff *nskb; 2919 struct skb_shared_info *skbinfo = skb_shinfo(skb); 2920 struct skb_shared_info *pinfo = skb_shinfo(p); 2921 unsigned int headroom; 2922 unsigned int len = skb_gro_len(skb); 2923 unsigned int offset = skb_gro_offset(skb); 2924 unsigned int headlen = skb_headlen(skb); 2925 unsigned int delta_truesize; 2926 2927 if (p->len + len >= 65536) 2928 return -E2BIG; 2929 2930 if (pinfo->frag_list) 2931 goto merge; 2932 else if (headlen <= offset) { 2933 skb_frag_t *frag; 2934 skb_frag_t *frag2; 2935 int i = skbinfo->nr_frags; 2936 int nr_frags = pinfo->nr_frags + i; 2937 2938 offset -= headlen; 2939 2940 if (nr_frags > MAX_SKB_FRAGS) 2941 return -E2BIG; 2942 2943 pinfo->nr_frags = nr_frags; 2944 skbinfo->nr_frags = 0; 2945 2946 frag = pinfo->frags + nr_frags; 2947 frag2 = skbinfo->frags + i; 2948 do { 2949 *--frag = *--frag2; 2950 } while (--i); 2951 2952 frag->page_offset += offset; 2953 skb_frag_size_sub(frag, offset); 2954 2955 /* all fragments truesize : remove (head size + sk_buff) */ 2956 delta_truesize = skb->truesize - 2957 SKB_TRUESIZE(skb_end_offset(skb)); 2958 2959 skb->truesize -= skb->data_len; 2960 skb->len -= skb->data_len; 2961 skb->data_len = 0; 2962 2963 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 2964 goto done; 2965 } else if (skb->head_frag) { 2966 int nr_frags = pinfo->nr_frags; 2967 skb_frag_t *frag = pinfo->frags + nr_frags; 2968 struct page *page = virt_to_head_page(skb->head); 2969 unsigned int first_size = headlen - offset; 2970 unsigned int first_offset; 2971 2972 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 2973 return -E2BIG; 2974 2975 first_offset = skb->data - 2976 (unsigned char *)page_address(page) + 2977 offset; 2978 2979 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 2980 2981 frag->page.p = page; 2982 frag->page_offset = first_offset; 2983 skb_frag_size_set(frag, first_size); 2984 2985 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 2986 /* We dont need to clear skbinfo->nr_frags here */ 2987 2988 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 2989 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 2990 goto done; 2991 } else if (skb_gro_len(p) != pinfo->gso_size) 2992 return -E2BIG; 2993 2994 headroom = skb_headroom(p); 2995 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC); 2996 if (unlikely(!nskb)) 2997 return -ENOMEM; 2998 2999 __copy_skb_header(nskb, p); 3000 nskb->mac_len = p->mac_len; 3001 3002 skb_reserve(nskb, headroom); 3003 __skb_put(nskb, skb_gro_offset(p)); 3004 3005 skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 3006 skb_set_network_header(nskb, skb_network_offset(p)); 3007 skb_set_transport_header(nskb, skb_transport_offset(p)); 3008 3009 __skb_pull(p, skb_gro_offset(p)); 3010 memcpy(skb_mac_header(nskb), skb_mac_header(p), 3011 p->data - skb_mac_header(p)); 3012 3013 skb_shinfo(nskb)->frag_list = p; 3014 skb_shinfo(nskb)->gso_size = pinfo->gso_size; 3015 pinfo->gso_size = 0; 3016 skb_header_release(p); 3017 NAPI_GRO_CB(nskb)->last = p; 3018 3019 nskb->data_len += p->len; 3020 nskb->truesize += p->truesize; 3021 nskb->len += p->len; 3022 3023 *head = nskb; 3024 nskb->next = p->next; 3025 p->next = NULL; 3026 3027 p = nskb; 3028 3029 merge: 3030 delta_truesize = skb->truesize; 3031 if (offset > headlen) { 3032 unsigned int eat = offset - headlen; 3033 3034 skbinfo->frags[0].page_offset += eat; 3035 skb_frag_size_sub(&skbinfo->frags[0], eat); 3036 skb->data_len -= eat; 3037 skb->len -= eat; 3038 offset = headlen; 3039 } 3040 3041 __skb_pull(skb, offset); 3042 3043 NAPI_GRO_CB(p)->last->next = skb; 3044 NAPI_GRO_CB(p)->last = skb; 3045 skb_header_release(skb); 3046 3047 done: 3048 NAPI_GRO_CB(p)->count++; 3049 p->data_len += len; 3050 p->truesize += delta_truesize; 3051 p->len += len; 3052 3053 NAPI_GRO_CB(skb)->same_flow = 1; 3054 return 0; 3055 } 3056 EXPORT_SYMBOL_GPL(skb_gro_receive); 3057 3058 void __init skb_init(void) 3059 { 3060 skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 3061 sizeof(struct sk_buff), 3062 0, 3063 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 3064 NULL); 3065 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3066 (2*sizeof(struct sk_buff)) + 3067 sizeof(atomic_t), 3068 0, 3069 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 3070 NULL); 3071 } 3072 3073 /** 3074 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3075 * @skb: Socket buffer containing the buffers to be mapped 3076 * @sg: The scatter-gather list to map into 3077 * @offset: The offset into the buffer's contents to start mapping 3078 * @len: Length of buffer space to be mapped 3079 * 3080 * Fill the specified scatter-gather list with mappings/pointers into a 3081 * region of the buffer space attached to a socket buffer. 3082 */ 3083 static int 3084 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3085 { 3086 int start = skb_headlen(skb); 3087 int i, copy = start - offset; 3088 struct sk_buff *frag_iter; 3089 int elt = 0; 3090 3091 if (copy > 0) { 3092 if (copy > len) 3093 copy = len; 3094 sg_set_buf(sg, skb->data + offset, copy); 3095 elt++; 3096 if ((len -= copy) == 0) 3097 return elt; 3098 offset += copy; 3099 } 3100 3101 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3102 int end; 3103 3104 WARN_ON(start > offset + len); 3105 3106 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3107 if ((copy = end - offset) > 0) { 3108 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3109 3110 if (copy > len) 3111 copy = len; 3112 sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3113 frag->page_offset+offset-start); 3114 elt++; 3115 if (!(len -= copy)) 3116 return elt; 3117 offset += copy; 3118 } 3119 start = end; 3120 } 3121 3122 skb_walk_frags(skb, frag_iter) { 3123 int end; 3124 3125 WARN_ON(start > offset + len); 3126 3127 end = start + frag_iter->len; 3128 if ((copy = end - offset) > 0) { 3129 if (copy > len) 3130 copy = len; 3131 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start, 3132 copy); 3133 if ((len -= copy) == 0) 3134 return elt; 3135 offset += copy; 3136 } 3137 start = end; 3138 } 3139 BUG_ON(len); 3140 return elt; 3141 } 3142 3143 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3144 { 3145 int nsg = __skb_to_sgvec(skb, sg, offset, len); 3146 3147 sg_mark_end(&sg[nsg - 1]); 3148 3149 return nsg; 3150 } 3151 EXPORT_SYMBOL_GPL(skb_to_sgvec); 3152 3153 /** 3154 * skb_cow_data - Check that a socket buffer's data buffers are writable 3155 * @skb: The socket buffer to check. 3156 * @tailbits: Amount of trailing space to be added 3157 * @trailer: Returned pointer to the skb where the @tailbits space begins 3158 * 3159 * Make sure that the data buffers attached to a socket buffer are 3160 * writable. If they are not, private copies are made of the data buffers 3161 * and the socket buffer is set to use these instead. 3162 * 3163 * If @tailbits is given, make sure that there is space to write @tailbits 3164 * bytes of data beyond current end of socket buffer. @trailer will be 3165 * set to point to the skb in which this space begins. 3166 * 3167 * The number of scatterlist elements required to completely map the 3168 * COW'd and extended socket buffer will be returned. 3169 */ 3170 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 3171 { 3172 int copyflag; 3173 int elt; 3174 struct sk_buff *skb1, **skb_p; 3175 3176 /* If skb is cloned or its head is paged, reallocate 3177 * head pulling out all the pages (pages are considered not writable 3178 * at the moment even if they are anonymous). 3179 */ 3180 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 3181 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 3182 return -ENOMEM; 3183 3184 /* Easy case. Most of packets will go this way. */ 3185 if (!skb_has_frag_list(skb)) { 3186 /* A little of trouble, not enough of space for trailer. 3187 * This should not happen, when stack is tuned to generate 3188 * good frames. OK, on miss we reallocate and reserve even more 3189 * space, 128 bytes is fair. */ 3190 3191 if (skb_tailroom(skb) < tailbits && 3192 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 3193 return -ENOMEM; 3194 3195 /* Voila! */ 3196 *trailer = skb; 3197 return 1; 3198 } 3199 3200 /* Misery. We are in troubles, going to mincer fragments... */ 3201 3202 elt = 1; 3203 skb_p = &skb_shinfo(skb)->frag_list; 3204 copyflag = 0; 3205 3206 while ((skb1 = *skb_p) != NULL) { 3207 int ntail = 0; 3208 3209 /* The fragment is partially pulled by someone, 3210 * this can happen on input. Copy it and everything 3211 * after it. */ 3212 3213 if (skb_shared(skb1)) 3214 copyflag = 1; 3215 3216 /* If the skb is the last, worry about trailer. */ 3217 3218 if (skb1->next == NULL && tailbits) { 3219 if (skb_shinfo(skb1)->nr_frags || 3220 skb_has_frag_list(skb1) || 3221 skb_tailroom(skb1) < tailbits) 3222 ntail = tailbits + 128; 3223 } 3224 3225 if (copyflag || 3226 skb_cloned(skb1) || 3227 ntail || 3228 skb_shinfo(skb1)->nr_frags || 3229 skb_has_frag_list(skb1)) { 3230 struct sk_buff *skb2; 3231 3232 /* Fuck, we are miserable poor guys... */ 3233 if (ntail == 0) 3234 skb2 = skb_copy(skb1, GFP_ATOMIC); 3235 else 3236 skb2 = skb_copy_expand(skb1, 3237 skb_headroom(skb1), 3238 ntail, 3239 GFP_ATOMIC); 3240 if (unlikely(skb2 == NULL)) 3241 return -ENOMEM; 3242 3243 if (skb1->sk) 3244 skb_set_owner_w(skb2, skb1->sk); 3245 3246 /* Looking around. Are we still alive? 3247 * OK, link new skb, drop old one */ 3248 3249 skb2->next = skb1->next; 3250 *skb_p = skb2; 3251 kfree_skb(skb1); 3252 skb1 = skb2; 3253 } 3254 elt++; 3255 *trailer = skb1; 3256 skb_p = &skb1->next; 3257 } 3258 3259 return elt; 3260 } 3261 EXPORT_SYMBOL_GPL(skb_cow_data); 3262 3263 static void sock_rmem_free(struct sk_buff *skb) 3264 { 3265 struct sock *sk = skb->sk; 3266 3267 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 3268 } 3269 3270 /* 3271 * Note: We dont mem charge error packets (no sk_forward_alloc changes) 3272 */ 3273 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 3274 { 3275 int len = skb->len; 3276 3277 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 3278 (unsigned int)sk->sk_rcvbuf) 3279 return -ENOMEM; 3280 3281 skb_orphan(skb); 3282 skb->sk = sk; 3283 skb->destructor = sock_rmem_free; 3284 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 3285 3286 /* before exiting rcu section, make sure dst is refcounted */ 3287 skb_dst_force(skb); 3288 3289 skb_queue_tail(&sk->sk_error_queue, skb); 3290 if (!sock_flag(sk, SOCK_DEAD)) 3291 sk->sk_data_ready(sk, len); 3292 return 0; 3293 } 3294 EXPORT_SYMBOL(sock_queue_err_skb); 3295 3296 void skb_tstamp_tx(struct sk_buff *orig_skb, 3297 struct skb_shared_hwtstamps *hwtstamps) 3298 { 3299 struct sock *sk = orig_skb->sk; 3300 struct sock_exterr_skb *serr; 3301 struct sk_buff *skb; 3302 int err; 3303 3304 if (!sk) 3305 return; 3306 3307 skb = skb_clone(orig_skb, GFP_ATOMIC); 3308 if (!skb) 3309 return; 3310 3311 if (hwtstamps) { 3312 *skb_hwtstamps(skb) = 3313 *hwtstamps; 3314 } else { 3315 /* 3316 * no hardware time stamps available, 3317 * so keep the shared tx_flags and only 3318 * store software time stamp 3319 */ 3320 skb->tstamp = ktime_get_real(); 3321 } 3322 3323 serr = SKB_EXT_ERR(skb); 3324 memset(serr, 0, sizeof(*serr)); 3325 serr->ee.ee_errno = ENOMSG; 3326 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 3327 3328 err = sock_queue_err_skb(sk, skb); 3329 3330 if (err) 3331 kfree_skb(skb); 3332 } 3333 EXPORT_SYMBOL_GPL(skb_tstamp_tx); 3334 3335 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 3336 { 3337 struct sock *sk = skb->sk; 3338 struct sock_exterr_skb *serr; 3339 int err; 3340 3341 skb->wifi_acked_valid = 1; 3342 skb->wifi_acked = acked; 3343 3344 serr = SKB_EXT_ERR(skb); 3345 memset(serr, 0, sizeof(*serr)); 3346 serr->ee.ee_errno = ENOMSG; 3347 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 3348 3349 err = sock_queue_err_skb(sk, skb); 3350 if (err) 3351 kfree_skb(skb); 3352 } 3353 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 3354 3355 3356 /** 3357 * skb_partial_csum_set - set up and verify partial csum values for packet 3358 * @skb: the skb to set 3359 * @start: the number of bytes after skb->data to start checksumming. 3360 * @off: the offset from start to place the checksum. 3361 * 3362 * For untrusted partially-checksummed packets, we need to make sure the values 3363 * for skb->csum_start and skb->csum_offset are valid so we don't oops. 3364 * 3365 * This function checks and sets those values and skb->ip_summed: if this 3366 * returns false you should drop the packet. 3367 */ 3368 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 3369 { 3370 if (unlikely(start > skb_headlen(skb)) || 3371 unlikely((int)start + off > skb_headlen(skb) - 2)) { 3372 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 3373 start, off, skb_headlen(skb)); 3374 return false; 3375 } 3376 skb->ip_summed = CHECKSUM_PARTIAL; 3377 skb->csum_start = skb_headroom(skb) + start; 3378 skb->csum_offset = off; 3379 return true; 3380 } 3381 EXPORT_SYMBOL_GPL(skb_partial_csum_set); 3382 3383 void __skb_warn_lro_forwarding(const struct sk_buff *skb) 3384 { 3385 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 3386 skb->dev->name); 3387 } 3388 EXPORT_SYMBOL(__skb_warn_lro_forwarding); 3389 3390 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 3391 { 3392 if (head_stolen) { 3393 skb_release_head_state(skb); 3394 kmem_cache_free(skbuff_head_cache, skb); 3395 } else { 3396 __kfree_skb(skb); 3397 } 3398 } 3399 EXPORT_SYMBOL(kfree_skb_partial); 3400 3401 /** 3402 * skb_try_coalesce - try to merge skb to prior one 3403 * @to: prior buffer 3404 * @from: buffer to add 3405 * @fragstolen: pointer to boolean 3406 * @delta_truesize: how much more was allocated than was requested 3407 */ 3408 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 3409 bool *fragstolen, int *delta_truesize) 3410 { 3411 int i, delta, len = from->len; 3412 3413 *fragstolen = false; 3414 3415 if (skb_cloned(to)) 3416 return false; 3417 3418 if (len <= skb_tailroom(to)) { 3419 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 3420 *delta_truesize = 0; 3421 return true; 3422 } 3423 3424 if (skb_has_frag_list(to) || skb_has_frag_list(from)) 3425 return false; 3426 3427 if (skb_headlen(from) != 0) { 3428 struct page *page; 3429 unsigned int offset; 3430 3431 if (skb_shinfo(to)->nr_frags + 3432 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 3433 return false; 3434 3435 if (skb_head_is_locked(from)) 3436 return false; 3437 3438 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3439 3440 page = virt_to_head_page(from->head); 3441 offset = from->data - (unsigned char *)page_address(page); 3442 3443 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 3444 page, offset, skb_headlen(from)); 3445 *fragstolen = true; 3446 } else { 3447 if (skb_shinfo(to)->nr_frags + 3448 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 3449 return false; 3450 3451 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 3452 } 3453 3454 WARN_ON_ONCE(delta < len); 3455 3456 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 3457 skb_shinfo(from)->frags, 3458 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 3459 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 3460 3461 if (!skb_cloned(from)) 3462 skb_shinfo(from)->nr_frags = 0; 3463 3464 /* if the skb is not cloned this does nothing 3465 * since we set nr_frags to 0. 3466 */ 3467 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 3468 skb_frag_ref(from, i); 3469 3470 to->truesize += delta; 3471 to->len += len; 3472 to->data_len += len; 3473 3474 *delta_truesize = delta; 3475 return true; 3476 } 3477 EXPORT_SYMBOL(skb_try_coalesce); 3478