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/tcp.h> 51 #include <linux/udp.h> 52 #include <linux/sctp.h> 53 #include <linux/netdevice.h> 54 #ifdef CONFIG_NET_CLS_ACT 55 #include <net/pkt_sched.h> 56 #endif 57 #include <linux/string.h> 58 #include <linux/skbuff.h> 59 #include <linux/splice.h> 60 #include <linux/cache.h> 61 #include <linux/rtnetlink.h> 62 #include <linux/init.h> 63 #include <linux/scatterlist.h> 64 #include <linux/errqueue.h> 65 #include <linux/prefetch.h> 66 #include <linux/if_vlan.h> 67 68 #include <net/protocol.h> 69 #include <net/dst.h> 70 #include <net/sock.h> 71 #include <net/checksum.h> 72 #include <net/ip6_checksum.h> 73 #include <net/xfrm.h> 74 75 #include <linux/uaccess.h> 76 #include <trace/events/skb.h> 77 #include <linux/highmem.h> 78 #include <linux/capability.h> 79 #include <linux/user_namespace.h> 80 81 struct kmem_cache *skbuff_head_cache __read_mostly; 82 static struct kmem_cache *skbuff_fclone_cache __read_mostly; 83 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS; 84 EXPORT_SYMBOL(sysctl_max_skb_frags); 85 86 /** 87 * skb_panic - private function for out-of-line support 88 * @skb: buffer 89 * @sz: size 90 * @addr: address 91 * @msg: skb_over_panic or skb_under_panic 92 * 93 * Out-of-line support for skb_put() and skb_push(). 94 * Called via the wrapper skb_over_panic() or skb_under_panic(). 95 * Keep out of line to prevent kernel bloat. 96 * __builtin_return_address is not used because it is not always reliable. 97 */ 98 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr, 99 const char msg[]) 100 { 101 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n", 102 msg, addr, skb->len, sz, skb->head, skb->data, 103 (unsigned long)skb->tail, (unsigned long)skb->end, 104 skb->dev ? skb->dev->name : "<NULL>"); 105 BUG(); 106 } 107 108 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr) 109 { 110 skb_panic(skb, sz, addr, __func__); 111 } 112 113 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr) 114 { 115 skb_panic(skb, sz, addr, __func__); 116 } 117 118 /* 119 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells 120 * the caller if emergency pfmemalloc reserves are being used. If it is and 121 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves 122 * may be used. Otherwise, the packet data may be discarded until enough 123 * memory is free 124 */ 125 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \ 126 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc) 127 128 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node, 129 unsigned long ip, bool *pfmemalloc) 130 { 131 void *obj; 132 bool ret_pfmemalloc = false; 133 134 /* 135 * Try a regular allocation, when that fails and we're not entitled 136 * to the reserves, fail. 137 */ 138 obj = kmalloc_node_track_caller(size, 139 flags | __GFP_NOMEMALLOC | __GFP_NOWARN, 140 node); 141 if (obj || !(gfp_pfmemalloc_allowed(flags))) 142 goto out; 143 144 /* Try again but now we are using pfmemalloc reserves */ 145 ret_pfmemalloc = true; 146 obj = kmalloc_node_track_caller(size, flags, node); 147 148 out: 149 if (pfmemalloc) 150 *pfmemalloc = ret_pfmemalloc; 151 152 return obj; 153 } 154 155 /* Allocate a new skbuff. We do this ourselves so we can fill in a few 156 * 'private' fields and also do memory statistics to find all the 157 * [BEEP] leaks. 158 * 159 */ 160 161 /** 162 * __alloc_skb - allocate a network buffer 163 * @size: size to allocate 164 * @gfp_mask: allocation mask 165 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache 166 * instead of head cache and allocate a cloned (child) skb. 167 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for 168 * allocations in case the data is required for writeback 169 * @node: numa node to allocate memory on 170 * 171 * Allocate a new &sk_buff. The returned buffer has no headroom and a 172 * tail room of at least size bytes. The object has a reference count 173 * of one. The return is the buffer. On a failure the return is %NULL. 174 * 175 * Buffers may only be allocated from interrupts using a @gfp_mask of 176 * %GFP_ATOMIC. 177 */ 178 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 179 int flags, int node) 180 { 181 struct kmem_cache *cache; 182 struct skb_shared_info *shinfo; 183 struct sk_buff *skb; 184 u8 *data; 185 bool pfmemalloc; 186 187 cache = (flags & SKB_ALLOC_FCLONE) 188 ? skbuff_fclone_cache : skbuff_head_cache; 189 190 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX)) 191 gfp_mask |= __GFP_MEMALLOC; 192 193 /* Get the HEAD */ 194 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 195 if (!skb) 196 goto out; 197 prefetchw(skb); 198 199 /* We do our best to align skb_shared_info on a separate cache 200 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives 201 * aligned memory blocks, unless SLUB/SLAB debug is enabled. 202 * Both skb->head and skb_shared_info are cache line aligned. 203 */ 204 size = SKB_DATA_ALIGN(size); 205 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 206 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc); 207 if (!data) 208 goto nodata; 209 /* kmalloc(size) might give us more room than requested. 210 * Put skb_shared_info exactly at the end of allocated zone, 211 * to allow max possible filling before reallocation. 212 */ 213 size = SKB_WITH_OVERHEAD(ksize(data)); 214 prefetchw(data + size); 215 216 /* 217 * Only clear those fields we need to clear, not those that we will 218 * actually initialise below. Hence, don't put any more fields after 219 * the tail pointer in struct sk_buff! 220 */ 221 memset(skb, 0, offsetof(struct sk_buff, tail)); 222 /* Account for allocated memory : skb + skb->head */ 223 skb->truesize = SKB_TRUESIZE(size); 224 skb->pfmemalloc = pfmemalloc; 225 refcount_set(&skb->users, 1); 226 skb->head = data; 227 skb->data = data; 228 skb_reset_tail_pointer(skb); 229 skb->end = skb->tail + size; 230 skb->mac_header = (typeof(skb->mac_header))~0U; 231 skb->transport_header = (typeof(skb->transport_header))~0U; 232 233 /* make sure we initialize shinfo sequentially */ 234 shinfo = skb_shinfo(skb); 235 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 236 atomic_set(&shinfo->dataref, 1); 237 kmemcheck_annotate_variable(shinfo->destructor_arg); 238 239 if (flags & SKB_ALLOC_FCLONE) { 240 struct sk_buff_fclones *fclones; 241 242 fclones = container_of(skb, struct sk_buff_fclones, skb1); 243 244 kmemcheck_annotate_bitfield(&fclones->skb2, flags1); 245 skb->fclone = SKB_FCLONE_ORIG; 246 refcount_set(&fclones->fclone_ref, 1); 247 248 fclones->skb2.fclone = SKB_FCLONE_CLONE; 249 } 250 out: 251 return skb; 252 nodata: 253 kmem_cache_free(cache, skb); 254 skb = NULL; 255 goto out; 256 } 257 EXPORT_SYMBOL(__alloc_skb); 258 259 /** 260 * __build_skb - build a network buffer 261 * @data: data buffer provided by caller 262 * @frag_size: size of data, or 0 if head was kmalloced 263 * 264 * Allocate a new &sk_buff. Caller provides space holding head and 265 * skb_shared_info. @data must have been allocated by kmalloc() only if 266 * @frag_size is 0, otherwise data should come from the page allocator 267 * or vmalloc() 268 * The return is the new skb buffer. 269 * On a failure the return is %NULL, and @data is not freed. 270 * Notes : 271 * Before IO, driver allocates only data buffer where NIC put incoming frame 272 * Driver should add room at head (NET_SKB_PAD) and 273 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info)) 274 * After IO, driver calls build_skb(), to allocate sk_buff and populate it 275 * before giving packet to stack. 276 * RX rings only contains data buffers, not full skbs. 277 */ 278 struct sk_buff *__build_skb(void *data, unsigned int frag_size) 279 { 280 struct skb_shared_info *shinfo; 281 struct sk_buff *skb; 282 unsigned int size = frag_size ? : ksize(data); 283 284 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC); 285 if (!skb) 286 return NULL; 287 288 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 289 290 memset(skb, 0, offsetof(struct sk_buff, tail)); 291 skb->truesize = SKB_TRUESIZE(size); 292 refcount_set(&skb->users, 1); 293 skb->head = data; 294 skb->data = data; 295 skb_reset_tail_pointer(skb); 296 skb->end = skb->tail + size; 297 skb->mac_header = (typeof(skb->mac_header))~0U; 298 skb->transport_header = (typeof(skb->transport_header))~0U; 299 300 /* make sure we initialize shinfo sequentially */ 301 shinfo = skb_shinfo(skb); 302 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref)); 303 atomic_set(&shinfo->dataref, 1); 304 kmemcheck_annotate_variable(shinfo->destructor_arg); 305 306 return skb; 307 } 308 309 /* build_skb() is wrapper over __build_skb(), that specifically 310 * takes care of skb->head and skb->pfmemalloc 311 * This means that if @frag_size is not zero, then @data must be backed 312 * by a page fragment, not kmalloc() or vmalloc() 313 */ 314 struct sk_buff *build_skb(void *data, unsigned int frag_size) 315 { 316 struct sk_buff *skb = __build_skb(data, frag_size); 317 318 if (skb && frag_size) { 319 skb->head_frag = 1; 320 if (page_is_pfmemalloc(virt_to_head_page(data))) 321 skb->pfmemalloc = 1; 322 } 323 return skb; 324 } 325 EXPORT_SYMBOL(build_skb); 326 327 #define NAPI_SKB_CACHE_SIZE 64 328 329 struct napi_alloc_cache { 330 struct page_frag_cache page; 331 unsigned int skb_count; 332 void *skb_cache[NAPI_SKB_CACHE_SIZE]; 333 }; 334 335 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache); 336 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache); 337 338 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 339 { 340 struct page_frag_cache *nc; 341 unsigned long flags; 342 void *data; 343 344 local_irq_save(flags); 345 nc = this_cpu_ptr(&netdev_alloc_cache); 346 data = page_frag_alloc(nc, fragsz, gfp_mask); 347 local_irq_restore(flags); 348 return data; 349 } 350 351 /** 352 * netdev_alloc_frag - allocate a page fragment 353 * @fragsz: fragment size 354 * 355 * Allocates a frag from a page for receive buffer. 356 * Uses GFP_ATOMIC allocations. 357 */ 358 void *netdev_alloc_frag(unsigned int fragsz) 359 { 360 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 361 } 362 EXPORT_SYMBOL(netdev_alloc_frag); 363 364 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask) 365 { 366 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 367 368 return page_frag_alloc(&nc->page, fragsz, gfp_mask); 369 } 370 371 void *napi_alloc_frag(unsigned int fragsz) 372 { 373 return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD); 374 } 375 EXPORT_SYMBOL(napi_alloc_frag); 376 377 /** 378 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 379 * @dev: network device to receive on 380 * @len: length to allocate 381 * @gfp_mask: get_free_pages mask, passed to alloc_skb 382 * 383 * Allocate a new &sk_buff and assign it a usage count of one. The 384 * buffer has NET_SKB_PAD headroom built in. Users should allocate 385 * the headroom they think they need without accounting for the 386 * built in space. The built in space is used for optimisations. 387 * 388 * %NULL is returned if there is no free memory. 389 */ 390 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len, 391 gfp_t gfp_mask) 392 { 393 struct page_frag_cache *nc; 394 unsigned long flags; 395 struct sk_buff *skb; 396 bool pfmemalloc; 397 void *data; 398 399 len += NET_SKB_PAD; 400 401 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 402 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 403 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 404 if (!skb) 405 goto skb_fail; 406 goto skb_success; 407 } 408 409 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 410 len = SKB_DATA_ALIGN(len); 411 412 if (sk_memalloc_socks()) 413 gfp_mask |= __GFP_MEMALLOC; 414 415 local_irq_save(flags); 416 417 nc = this_cpu_ptr(&netdev_alloc_cache); 418 data = page_frag_alloc(nc, len, gfp_mask); 419 pfmemalloc = nc->pfmemalloc; 420 421 local_irq_restore(flags); 422 423 if (unlikely(!data)) 424 return NULL; 425 426 skb = __build_skb(data, len); 427 if (unlikely(!skb)) { 428 skb_free_frag(data); 429 return NULL; 430 } 431 432 /* use OR instead of assignment to avoid clearing of bits in mask */ 433 if (pfmemalloc) 434 skb->pfmemalloc = 1; 435 skb->head_frag = 1; 436 437 skb_success: 438 skb_reserve(skb, NET_SKB_PAD); 439 skb->dev = dev; 440 441 skb_fail: 442 return skb; 443 } 444 EXPORT_SYMBOL(__netdev_alloc_skb); 445 446 /** 447 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance 448 * @napi: napi instance this buffer was allocated for 449 * @len: length to allocate 450 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages 451 * 452 * Allocate a new sk_buff for use in NAPI receive. This buffer will 453 * attempt to allocate the head from a special reserved region used 454 * only for NAPI Rx allocation. By doing this we can save several 455 * CPU cycles by avoiding having to disable and re-enable IRQs. 456 * 457 * %NULL is returned if there is no free memory. 458 */ 459 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len, 460 gfp_t gfp_mask) 461 { 462 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 463 struct sk_buff *skb; 464 void *data; 465 466 len += NET_SKB_PAD + NET_IP_ALIGN; 467 468 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) || 469 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) { 470 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE); 471 if (!skb) 472 goto skb_fail; 473 goto skb_success; 474 } 475 476 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 477 len = SKB_DATA_ALIGN(len); 478 479 if (sk_memalloc_socks()) 480 gfp_mask |= __GFP_MEMALLOC; 481 482 data = page_frag_alloc(&nc->page, len, gfp_mask); 483 if (unlikely(!data)) 484 return NULL; 485 486 skb = __build_skb(data, len); 487 if (unlikely(!skb)) { 488 skb_free_frag(data); 489 return NULL; 490 } 491 492 /* use OR instead of assignment to avoid clearing of bits in mask */ 493 if (nc->page.pfmemalloc) 494 skb->pfmemalloc = 1; 495 skb->head_frag = 1; 496 497 skb_success: 498 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 499 skb->dev = napi->dev; 500 501 skb_fail: 502 return skb; 503 } 504 EXPORT_SYMBOL(__napi_alloc_skb); 505 506 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 507 int size, unsigned int truesize) 508 { 509 skb_fill_page_desc(skb, i, page, off, size); 510 skb->len += size; 511 skb->data_len += size; 512 skb->truesize += truesize; 513 } 514 EXPORT_SYMBOL(skb_add_rx_frag); 515 516 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 517 unsigned int truesize) 518 { 519 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 520 521 skb_frag_size_add(frag, size); 522 skb->len += size; 523 skb->data_len += size; 524 skb->truesize += truesize; 525 } 526 EXPORT_SYMBOL(skb_coalesce_rx_frag); 527 528 static void skb_drop_list(struct sk_buff **listp) 529 { 530 kfree_skb_list(*listp); 531 *listp = NULL; 532 } 533 534 static inline void skb_drop_fraglist(struct sk_buff *skb) 535 { 536 skb_drop_list(&skb_shinfo(skb)->frag_list); 537 } 538 539 static void skb_clone_fraglist(struct sk_buff *skb) 540 { 541 struct sk_buff *list; 542 543 skb_walk_frags(skb, list) 544 skb_get(list); 545 } 546 547 static void skb_free_head(struct sk_buff *skb) 548 { 549 unsigned char *head = skb->head; 550 551 if (skb->head_frag) 552 skb_free_frag(head); 553 else 554 kfree(head); 555 } 556 557 static void skb_release_data(struct sk_buff *skb) 558 { 559 struct skb_shared_info *shinfo = skb_shinfo(skb); 560 int i; 561 562 if (skb->cloned && 563 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 564 &shinfo->dataref)) 565 return; 566 567 for (i = 0; i < shinfo->nr_frags; i++) 568 __skb_frag_unref(&shinfo->frags[i]); 569 570 if (shinfo->frag_list) 571 kfree_skb_list(shinfo->frag_list); 572 573 skb_zcopy_clear(skb, true); 574 skb_free_head(skb); 575 } 576 577 /* 578 * Free an skbuff by memory without cleaning the state. 579 */ 580 static void kfree_skbmem(struct sk_buff *skb) 581 { 582 struct sk_buff_fclones *fclones; 583 584 switch (skb->fclone) { 585 case SKB_FCLONE_UNAVAILABLE: 586 kmem_cache_free(skbuff_head_cache, skb); 587 return; 588 589 case SKB_FCLONE_ORIG: 590 fclones = container_of(skb, struct sk_buff_fclones, skb1); 591 592 /* We usually free the clone (TX completion) before original skb 593 * This test would have no chance to be true for the clone, 594 * while here, branch prediction will be good. 595 */ 596 if (refcount_read(&fclones->fclone_ref) == 1) 597 goto fastpath; 598 break; 599 600 default: /* SKB_FCLONE_CLONE */ 601 fclones = container_of(skb, struct sk_buff_fclones, skb2); 602 break; 603 } 604 if (!refcount_dec_and_test(&fclones->fclone_ref)) 605 return; 606 fastpath: 607 kmem_cache_free(skbuff_fclone_cache, fclones); 608 } 609 610 void skb_release_head_state(struct sk_buff *skb) 611 { 612 skb_dst_drop(skb); 613 secpath_reset(skb); 614 if (skb->destructor) { 615 WARN_ON(in_irq()); 616 skb->destructor(skb); 617 } 618 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 619 nf_conntrack_put(skb_nfct(skb)); 620 #endif 621 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 622 nf_bridge_put(skb->nf_bridge); 623 #endif 624 } 625 626 /* Free everything but the sk_buff shell. */ 627 static void skb_release_all(struct sk_buff *skb) 628 { 629 skb_release_head_state(skb); 630 if (likely(skb->head)) 631 skb_release_data(skb); 632 } 633 634 /** 635 * __kfree_skb - private function 636 * @skb: buffer 637 * 638 * Free an sk_buff. Release anything attached to the buffer. 639 * Clean the state. This is an internal helper function. Users should 640 * always call kfree_skb 641 */ 642 643 void __kfree_skb(struct sk_buff *skb) 644 { 645 skb_release_all(skb); 646 kfree_skbmem(skb); 647 } 648 EXPORT_SYMBOL(__kfree_skb); 649 650 /** 651 * kfree_skb - free an sk_buff 652 * @skb: buffer to free 653 * 654 * Drop a reference to the buffer and free it if the usage count has 655 * hit zero. 656 */ 657 void kfree_skb(struct sk_buff *skb) 658 { 659 if (!skb_unref(skb)) 660 return; 661 662 trace_kfree_skb(skb, __builtin_return_address(0)); 663 __kfree_skb(skb); 664 } 665 EXPORT_SYMBOL(kfree_skb); 666 667 void kfree_skb_list(struct sk_buff *segs) 668 { 669 while (segs) { 670 struct sk_buff *next = segs->next; 671 672 kfree_skb(segs); 673 segs = next; 674 } 675 } 676 EXPORT_SYMBOL(kfree_skb_list); 677 678 /** 679 * skb_tx_error - report an sk_buff xmit error 680 * @skb: buffer that triggered an error 681 * 682 * Report xmit error if a device callback is tracking this skb. 683 * skb must be freed afterwards. 684 */ 685 void skb_tx_error(struct sk_buff *skb) 686 { 687 skb_zcopy_clear(skb, true); 688 } 689 EXPORT_SYMBOL(skb_tx_error); 690 691 /** 692 * consume_skb - free an skbuff 693 * @skb: buffer to free 694 * 695 * Drop a ref to the buffer and free it if the usage count has hit zero 696 * Functions identically to kfree_skb, but kfree_skb assumes that the frame 697 * is being dropped after a failure and notes that 698 */ 699 void consume_skb(struct sk_buff *skb) 700 { 701 if (!skb_unref(skb)) 702 return; 703 704 trace_consume_skb(skb); 705 __kfree_skb(skb); 706 } 707 EXPORT_SYMBOL(consume_skb); 708 709 /** 710 * consume_stateless_skb - free an skbuff, assuming it is stateless 711 * @skb: buffer to free 712 * 713 * Alike consume_skb(), but this variant assumes that this is the last 714 * skb reference and all the head states have been already dropped 715 */ 716 void __consume_stateless_skb(struct sk_buff *skb) 717 { 718 trace_consume_skb(skb); 719 skb_release_data(skb); 720 kfree_skbmem(skb); 721 } 722 723 void __kfree_skb_flush(void) 724 { 725 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 726 727 /* flush skb_cache if containing objects */ 728 if (nc->skb_count) { 729 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count, 730 nc->skb_cache); 731 nc->skb_count = 0; 732 } 733 } 734 735 static inline void _kfree_skb_defer(struct sk_buff *skb) 736 { 737 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache); 738 739 /* drop skb->head and call any destructors for packet */ 740 skb_release_all(skb); 741 742 /* record skb to CPU local list */ 743 nc->skb_cache[nc->skb_count++] = skb; 744 745 #ifdef CONFIG_SLUB 746 /* SLUB writes into objects when freeing */ 747 prefetchw(skb); 748 #endif 749 750 /* flush skb_cache if it is filled */ 751 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) { 752 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE, 753 nc->skb_cache); 754 nc->skb_count = 0; 755 } 756 } 757 void __kfree_skb_defer(struct sk_buff *skb) 758 { 759 _kfree_skb_defer(skb); 760 } 761 762 void napi_consume_skb(struct sk_buff *skb, int budget) 763 { 764 if (unlikely(!skb)) 765 return; 766 767 /* Zero budget indicate non-NAPI context called us, like netpoll */ 768 if (unlikely(!budget)) { 769 dev_consume_skb_any(skb); 770 return; 771 } 772 773 if (!skb_unref(skb)) 774 return; 775 776 /* if reaching here SKB is ready to free */ 777 trace_consume_skb(skb); 778 779 /* if SKB is a clone, don't handle this case */ 780 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) { 781 __kfree_skb(skb); 782 return; 783 } 784 785 _kfree_skb_defer(skb); 786 } 787 EXPORT_SYMBOL(napi_consume_skb); 788 789 /* Make sure a field is enclosed inside headers_start/headers_end section */ 790 #define CHECK_SKB_FIELD(field) \ 791 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \ 792 offsetof(struct sk_buff, headers_start)); \ 793 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \ 794 offsetof(struct sk_buff, headers_end)); \ 795 796 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 797 { 798 new->tstamp = old->tstamp; 799 /* We do not copy old->sk */ 800 new->dev = old->dev; 801 memcpy(new->cb, old->cb, sizeof(old->cb)); 802 skb_dst_copy(new, old); 803 #ifdef CONFIG_XFRM 804 new->sp = secpath_get(old->sp); 805 #endif 806 __nf_copy(new, old, false); 807 808 /* Note : this field could be in headers_start/headers_end section 809 * It is not yet because we do not want to have a 16 bit hole 810 */ 811 new->queue_mapping = old->queue_mapping; 812 813 memcpy(&new->headers_start, &old->headers_start, 814 offsetof(struct sk_buff, headers_end) - 815 offsetof(struct sk_buff, headers_start)); 816 CHECK_SKB_FIELD(protocol); 817 CHECK_SKB_FIELD(csum); 818 CHECK_SKB_FIELD(hash); 819 CHECK_SKB_FIELD(priority); 820 CHECK_SKB_FIELD(skb_iif); 821 CHECK_SKB_FIELD(vlan_proto); 822 CHECK_SKB_FIELD(vlan_tci); 823 CHECK_SKB_FIELD(transport_header); 824 CHECK_SKB_FIELD(network_header); 825 CHECK_SKB_FIELD(mac_header); 826 CHECK_SKB_FIELD(inner_protocol); 827 CHECK_SKB_FIELD(inner_transport_header); 828 CHECK_SKB_FIELD(inner_network_header); 829 CHECK_SKB_FIELD(inner_mac_header); 830 CHECK_SKB_FIELD(mark); 831 #ifdef CONFIG_NETWORK_SECMARK 832 CHECK_SKB_FIELD(secmark); 833 #endif 834 #ifdef CONFIG_NET_RX_BUSY_POLL 835 CHECK_SKB_FIELD(napi_id); 836 #endif 837 #ifdef CONFIG_XPS 838 CHECK_SKB_FIELD(sender_cpu); 839 #endif 840 #ifdef CONFIG_NET_SCHED 841 CHECK_SKB_FIELD(tc_index); 842 #endif 843 844 } 845 846 /* 847 * You should not add any new code to this function. Add it to 848 * __copy_skb_header above instead. 849 */ 850 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 851 { 852 #define C(x) n->x = skb->x 853 854 n->next = n->prev = NULL; 855 n->sk = NULL; 856 __copy_skb_header(n, skb); 857 858 C(len); 859 C(data_len); 860 C(mac_len); 861 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 862 n->cloned = 1; 863 n->nohdr = 0; 864 n->destructor = NULL; 865 C(tail); 866 C(end); 867 C(head); 868 C(head_frag); 869 C(data); 870 C(truesize); 871 refcount_set(&n->users, 1); 872 873 atomic_inc(&(skb_shinfo(skb)->dataref)); 874 skb->cloned = 1; 875 876 return n; 877 #undef C 878 } 879 880 /** 881 * skb_morph - morph one skb into another 882 * @dst: the skb to receive the contents 883 * @src: the skb to supply the contents 884 * 885 * This is identical to skb_clone except that the target skb is 886 * supplied by the user. 887 * 888 * The target skb is returned upon exit. 889 */ 890 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 891 { 892 skb_release_all(dst); 893 return __skb_clone(dst, src); 894 } 895 EXPORT_SYMBOL_GPL(skb_morph); 896 897 static int mm_account_pinned_pages(struct mmpin *mmp, size_t size) 898 { 899 unsigned long max_pg, num_pg, new_pg, old_pg; 900 struct user_struct *user; 901 902 if (capable(CAP_IPC_LOCK) || !size) 903 return 0; 904 905 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */ 906 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 907 user = mmp->user ? : current_user(); 908 909 do { 910 old_pg = atomic_long_read(&user->locked_vm); 911 new_pg = old_pg + num_pg; 912 if (new_pg > max_pg) 913 return -ENOBUFS; 914 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) != 915 old_pg); 916 917 if (!mmp->user) { 918 mmp->user = get_uid(user); 919 mmp->num_pg = num_pg; 920 } else { 921 mmp->num_pg += num_pg; 922 } 923 924 return 0; 925 } 926 927 static void mm_unaccount_pinned_pages(struct mmpin *mmp) 928 { 929 if (mmp->user) { 930 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm); 931 free_uid(mmp->user); 932 } 933 } 934 935 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size) 936 { 937 struct ubuf_info *uarg; 938 struct sk_buff *skb; 939 940 WARN_ON_ONCE(!in_task()); 941 942 if (!sock_flag(sk, SOCK_ZEROCOPY)) 943 return NULL; 944 945 skb = sock_omalloc(sk, 0, GFP_KERNEL); 946 if (!skb) 947 return NULL; 948 949 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb)); 950 uarg = (void *)skb->cb; 951 uarg->mmp.user = NULL; 952 953 if (mm_account_pinned_pages(&uarg->mmp, size)) { 954 kfree_skb(skb); 955 return NULL; 956 } 957 958 uarg->callback = sock_zerocopy_callback; 959 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1; 960 uarg->len = 1; 961 uarg->bytelen = size; 962 uarg->zerocopy = 1; 963 refcount_set(&uarg->refcnt, 1); 964 sock_hold(sk); 965 966 return uarg; 967 } 968 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc); 969 970 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg) 971 { 972 return container_of((void *)uarg, struct sk_buff, cb); 973 } 974 975 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 976 struct ubuf_info *uarg) 977 { 978 if (uarg) { 979 const u32 byte_limit = 1 << 19; /* limit to a few TSO */ 980 u32 bytelen, next; 981 982 /* realloc only when socket is locked (TCP, UDP cork), 983 * so uarg->len and sk_zckey access is serialized 984 */ 985 if (!sock_owned_by_user(sk)) { 986 WARN_ON_ONCE(1); 987 return NULL; 988 } 989 990 bytelen = uarg->bytelen + size; 991 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) { 992 /* TCP can create new skb to attach new uarg */ 993 if (sk->sk_type == SOCK_STREAM) 994 goto new_alloc; 995 return NULL; 996 } 997 998 next = (u32)atomic_read(&sk->sk_zckey); 999 if ((u32)(uarg->id + uarg->len) == next) { 1000 if (mm_account_pinned_pages(&uarg->mmp, size)) 1001 return NULL; 1002 uarg->len++; 1003 uarg->bytelen = bytelen; 1004 atomic_set(&sk->sk_zckey, ++next); 1005 sock_zerocopy_get(uarg); 1006 return uarg; 1007 } 1008 } 1009 1010 new_alloc: 1011 return sock_zerocopy_alloc(sk, size); 1012 } 1013 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc); 1014 1015 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len) 1016 { 1017 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 1018 u32 old_lo, old_hi; 1019 u64 sum_len; 1020 1021 old_lo = serr->ee.ee_info; 1022 old_hi = serr->ee.ee_data; 1023 sum_len = old_hi - old_lo + 1ULL + len; 1024 1025 if (sum_len >= (1ULL << 32)) 1026 return false; 1027 1028 if (lo != old_hi + 1) 1029 return false; 1030 1031 serr->ee.ee_data += len; 1032 return true; 1033 } 1034 1035 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) 1036 { 1037 struct sk_buff *tail, *skb = skb_from_uarg(uarg); 1038 struct sock_exterr_skb *serr; 1039 struct sock *sk = skb->sk; 1040 struct sk_buff_head *q; 1041 unsigned long flags; 1042 u32 lo, hi; 1043 u16 len; 1044 1045 mm_unaccount_pinned_pages(&uarg->mmp); 1046 1047 /* if !len, there was only 1 call, and it was aborted 1048 * so do not queue a completion notification 1049 */ 1050 if (!uarg->len || sock_flag(sk, SOCK_DEAD)) 1051 goto release; 1052 1053 len = uarg->len; 1054 lo = uarg->id; 1055 hi = uarg->id + len - 1; 1056 1057 serr = SKB_EXT_ERR(skb); 1058 memset(serr, 0, sizeof(*serr)); 1059 serr->ee.ee_errno = 0; 1060 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY; 1061 serr->ee.ee_data = hi; 1062 serr->ee.ee_info = lo; 1063 if (!success) 1064 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED; 1065 1066 q = &sk->sk_error_queue; 1067 spin_lock_irqsave(&q->lock, flags); 1068 tail = skb_peek_tail(q); 1069 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY || 1070 !skb_zerocopy_notify_extend(tail, lo, len)) { 1071 __skb_queue_tail(q, skb); 1072 skb = NULL; 1073 } 1074 spin_unlock_irqrestore(&q->lock, flags); 1075 1076 sk->sk_error_report(sk); 1077 1078 release: 1079 consume_skb(skb); 1080 sock_put(sk); 1081 } 1082 EXPORT_SYMBOL_GPL(sock_zerocopy_callback); 1083 1084 void sock_zerocopy_put(struct ubuf_info *uarg) 1085 { 1086 if (uarg && refcount_dec_and_test(&uarg->refcnt)) { 1087 if (uarg->callback) 1088 uarg->callback(uarg, uarg->zerocopy); 1089 else 1090 consume_skb(skb_from_uarg(uarg)); 1091 } 1092 } 1093 EXPORT_SYMBOL_GPL(sock_zerocopy_put); 1094 1095 void sock_zerocopy_put_abort(struct ubuf_info *uarg) 1096 { 1097 if (uarg) { 1098 struct sock *sk = skb_from_uarg(uarg)->sk; 1099 1100 atomic_dec(&sk->sk_zckey); 1101 uarg->len--; 1102 1103 sock_zerocopy_put(uarg); 1104 } 1105 } 1106 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); 1107 1108 extern int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb, 1109 struct iov_iter *from, size_t length); 1110 1111 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1112 struct msghdr *msg, int len, 1113 struct ubuf_info *uarg) 1114 { 1115 struct ubuf_info *orig_uarg = skb_zcopy(skb); 1116 struct iov_iter orig_iter = msg->msg_iter; 1117 int err, orig_len = skb->len; 1118 1119 /* An skb can only point to one uarg. This edge case happens when 1120 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc. 1121 */ 1122 if (orig_uarg && uarg != orig_uarg) 1123 return -EEXIST; 1124 1125 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len); 1126 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) { 1127 struct sock *save_sk = skb->sk; 1128 1129 /* Streams do not free skb on error. Reset to prev state. */ 1130 msg->msg_iter = orig_iter; 1131 skb->sk = sk; 1132 ___pskb_trim(skb, orig_len); 1133 skb->sk = save_sk; 1134 return err; 1135 } 1136 1137 skb_zcopy_set(skb, uarg); 1138 return skb->len - orig_len; 1139 } 1140 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); 1141 1142 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, 1143 gfp_t gfp_mask) 1144 { 1145 if (skb_zcopy(orig)) { 1146 if (skb_zcopy(nskb)) { 1147 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */ 1148 if (!gfp_mask) { 1149 WARN_ON_ONCE(1); 1150 return -ENOMEM; 1151 } 1152 if (skb_uarg(nskb) == skb_uarg(orig)) 1153 return 0; 1154 if (skb_copy_ubufs(nskb, GFP_ATOMIC)) 1155 return -EIO; 1156 } 1157 skb_zcopy_set(nskb, skb_uarg(orig)); 1158 } 1159 return 0; 1160 } 1161 1162 /** 1163 * skb_copy_ubufs - copy userspace skb frags buffers to kernel 1164 * @skb: the skb to modify 1165 * @gfp_mask: allocation priority 1166 * 1167 * This must be called on SKBTX_DEV_ZEROCOPY skb. 1168 * It will copy all frags into kernel and drop the reference 1169 * to userspace pages. 1170 * 1171 * If this function is called from an interrupt gfp_mask() must be 1172 * %GFP_ATOMIC. 1173 * 1174 * Returns 0 on success or a negative error code on failure 1175 * to allocate kernel memory to copy to. 1176 */ 1177 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask) 1178 { 1179 int num_frags = skb_shinfo(skb)->nr_frags; 1180 struct page *page, *head = NULL; 1181 int i, new_frags; 1182 u32 d_off; 1183 1184 if (!num_frags) 1185 return 0; 1186 1187 if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) 1188 return -EINVAL; 1189 1190 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT; 1191 for (i = 0; i < new_frags; i++) { 1192 page = alloc_page(gfp_mask); 1193 if (!page) { 1194 while (head) { 1195 struct page *next = (struct page *)page_private(head); 1196 put_page(head); 1197 head = next; 1198 } 1199 return -ENOMEM; 1200 } 1201 set_page_private(page, (unsigned long)head); 1202 head = page; 1203 } 1204 1205 page = head; 1206 d_off = 0; 1207 for (i = 0; i < num_frags; i++) { 1208 skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 1209 u32 p_off, p_len, copied; 1210 struct page *p; 1211 u8 *vaddr; 1212 1213 skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f), 1214 p, p_off, p_len, copied) { 1215 u32 copy, done = 0; 1216 vaddr = kmap_atomic(p); 1217 1218 while (done < p_len) { 1219 if (d_off == PAGE_SIZE) { 1220 d_off = 0; 1221 page = (struct page *)page_private(page); 1222 } 1223 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done); 1224 memcpy(page_address(page) + d_off, 1225 vaddr + p_off + done, copy); 1226 done += copy; 1227 d_off += copy; 1228 } 1229 kunmap_atomic(vaddr); 1230 } 1231 } 1232 1233 /* skb frags release userspace buffers */ 1234 for (i = 0; i < num_frags; i++) 1235 skb_frag_unref(skb, i); 1236 1237 /* skb frags point to kernel buffers */ 1238 for (i = 0; i < new_frags - 1; i++) { 1239 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE); 1240 head = (struct page *)page_private(head); 1241 } 1242 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off); 1243 skb_shinfo(skb)->nr_frags = new_frags; 1244 1245 skb_zcopy_clear(skb, false); 1246 return 0; 1247 } 1248 EXPORT_SYMBOL_GPL(skb_copy_ubufs); 1249 1250 /** 1251 * skb_clone - duplicate an sk_buff 1252 * @skb: buffer to clone 1253 * @gfp_mask: allocation priority 1254 * 1255 * Duplicate an &sk_buff. The new one is not owned by a socket. Both 1256 * copies share the same packet data but not structure. The new 1257 * buffer has a reference count of 1. If the allocation fails the 1258 * function returns %NULL otherwise the new buffer is returned. 1259 * 1260 * If this function is called from an interrupt gfp_mask() must be 1261 * %GFP_ATOMIC. 1262 */ 1263 1264 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 1265 { 1266 struct sk_buff_fclones *fclones = container_of(skb, 1267 struct sk_buff_fclones, 1268 skb1); 1269 struct sk_buff *n; 1270 1271 if (skb_orphan_frags(skb, gfp_mask)) 1272 return NULL; 1273 1274 if (skb->fclone == SKB_FCLONE_ORIG && 1275 refcount_read(&fclones->fclone_ref) == 1) { 1276 n = &fclones->skb2; 1277 refcount_set(&fclones->fclone_ref, 2); 1278 } else { 1279 if (skb_pfmemalloc(skb)) 1280 gfp_mask |= __GFP_MEMALLOC; 1281 1282 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 1283 if (!n) 1284 return NULL; 1285 1286 kmemcheck_annotate_bitfield(n, flags1); 1287 n->fclone = SKB_FCLONE_UNAVAILABLE; 1288 } 1289 1290 return __skb_clone(n, skb); 1291 } 1292 EXPORT_SYMBOL(skb_clone); 1293 1294 static void skb_headers_offset_update(struct sk_buff *skb, int off) 1295 { 1296 /* Only adjust this if it actually is csum_start rather than csum */ 1297 if (skb->ip_summed == CHECKSUM_PARTIAL) 1298 skb->csum_start += off; 1299 /* {transport,network,mac}_header and tail are relative to skb->head */ 1300 skb->transport_header += off; 1301 skb->network_header += off; 1302 if (skb_mac_header_was_set(skb)) 1303 skb->mac_header += off; 1304 skb->inner_transport_header += off; 1305 skb->inner_network_header += off; 1306 skb->inner_mac_header += off; 1307 } 1308 1309 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 1310 { 1311 __copy_skb_header(new, old); 1312 1313 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 1314 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 1315 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 1316 } 1317 1318 static inline int skb_alloc_rx_flag(const struct sk_buff *skb) 1319 { 1320 if (skb_pfmemalloc(skb)) 1321 return SKB_ALLOC_RX; 1322 return 0; 1323 } 1324 1325 /** 1326 * skb_copy - create private copy of an sk_buff 1327 * @skb: buffer to copy 1328 * @gfp_mask: allocation priority 1329 * 1330 * Make a copy of both an &sk_buff and its data. This is used when the 1331 * caller wishes to modify the data and needs a private copy of the 1332 * data to alter. Returns %NULL on failure or the pointer to the buffer 1333 * on success. The returned buffer has a reference count of 1. 1334 * 1335 * As by-product this function converts non-linear &sk_buff to linear 1336 * one, so that &sk_buff becomes completely private and caller is allowed 1337 * to modify all the data of returned buffer. This means that this 1338 * function is not recommended for use in circumstances when only 1339 * header is going to be modified. Use pskb_copy() instead. 1340 */ 1341 1342 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 1343 { 1344 int headerlen = skb_headroom(skb); 1345 unsigned int size = skb_end_offset(skb) + skb->data_len; 1346 struct sk_buff *n = __alloc_skb(size, gfp_mask, 1347 skb_alloc_rx_flag(skb), NUMA_NO_NODE); 1348 1349 if (!n) 1350 return NULL; 1351 1352 /* Set the data pointer */ 1353 skb_reserve(n, headerlen); 1354 /* Set the tail pointer and length */ 1355 skb_put(n, skb->len); 1356 1357 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 1358 BUG(); 1359 1360 copy_skb_header(n, skb); 1361 return n; 1362 } 1363 EXPORT_SYMBOL(skb_copy); 1364 1365 /** 1366 * __pskb_copy_fclone - create copy of an sk_buff with private head. 1367 * @skb: buffer to copy 1368 * @headroom: headroom of new skb 1369 * @gfp_mask: allocation priority 1370 * @fclone: if true allocate the copy of the skb from the fclone 1371 * cache instead of the head cache; it is recommended to set this 1372 * to true for the cases where the copy will likely be cloned 1373 * 1374 * Make a copy of both an &sk_buff and part of its data, located 1375 * in header. Fragmented data remain shared. This is used when 1376 * the caller wishes to modify only header of &sk_buff and needs 1377 * private copy of the header to alter. Returns %NULL on failure 1378 * or the pointer to the buffer on success. 1379 * The returned buffer has a reference count of 1. 1380 */ 1381 1382 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1383 gfp_t gfp_mask, bool fclone) 1384 { 1385 unsigned int size = skb_headlen(skb) + headroom; 1386 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0); 1387 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE); 1388 1389 if (!n) 1390 goto out; 1391 1392 /* Set the data pointer */ 1393 skb_reserve(n, headroom); 1394 /* Set the tail pointer and length */ 1395 skb_put(n, skb_headlen(skb)); 1396 /* Copy the bytes */ 1397 skb_copy_from_linear_data(skb, n->data, n->len); 1398 1399 n->truesize += skb->data_len; 1400 n->data_len = skb->data_len; 1401 n->len = skb->len; 1402 1403 if (skb_shinfo(skb)->nr_frags) { 1404 int i; 1405 1406 if (skb_orphan_frags(skb, gfp_mask) || 1407 skb_zerocopy_clone(n, skb, gfp_mask)) { 1408 kfree_skb(n); 1409 n = NULL; 1410 goto out; 1411 } 1412 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1413 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 1414 skb_frag_ref(skb, i); 1415 } 1416 skb_shinfo(n)->nr_frags = i; 1417 } 1418 1419 if (skb_has_frag_list(skb)) { 1420 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 1421 skb_clone_fraglist(n); 1422 } 1423 1424 copy_skb_header(n, skb); 1425 out: 1426 return n; 1427 } 1428 EXPORT_SYMBOL(__pskb_copy_fclone); 1429 1430 /** 1431 * pskb_expand_head - reallocate header of &sk_buff 1432 * @skb: buffer to reallocate 1433 * @nhead: room to add at head 1434 * @ntail: room to add at tail 1435 * @gfp_mask: allocation priority 1436 * 1437 * Expands (or creates identical copy, if @nhead and @ntail are zero) 1438 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have 1439 * reference count of 1. Returns zero in the case of success or error, 1440 * if expansion failed. In the last case, &sk_buff is not changed. 1441 * 1442 * All the pointers pointing into skb header may change and must be 1443 * reloaded after call to this function. 1444 */ 1445 1446 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 1447 gfp_t gfp_mask) 1448 { 1449 int i, osize = skb_end_offset(skb); 1450 int size = osize + nhead + ntail; 1451 long off; 1452 u8 *data; 1453 1454 BUG_ON(nhead < 0); 1455 1456 if (skb_shared(skb)) 1457 BUG(); 1458 1459 size = SKB_DATA_ALIGN(size); 1460 1461 if (skb_pfmemalloc(skb)) 1462 gfp_mask |= __GFP_MEMALLOC; 1463 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 1464 gfp_mask, NUMA_NO_NODE, NULL); 1465 if (!data) 1466 goto nodata; 1467 size = SKB_WITH_OVERHEAD(ksize(data)); 1468 1469 /* Copy only real data... and, alas, header. This should be 1470 * optimized for the cases when header is void. 1471 */ 1472 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head); 1473 1474 memcpy((struct skb_shared_info *)(data + size), 1475 skb_shinfo(skb), 1476 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags])); 1477 1478 /* 1479 * if shinfo is shared we must drop the old head gracefully, but if it 1480 * is not we can just drop the old head and let the existing refcount 1481 * be since all we did is relocate the values 1482 */ 1483 if (skb_cloned(skb)) { 1484 if (skb_orphan_frags(skb, gfp_mask)) 1485 goto nofrags; 1486 if (skb_zcopy(skb)) 1487 refcount_inc(&skb_uarg(skb)->refcnt); 1488 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 1489 skb_frag_ref(skb, i); 1490 1491 if (skb_has_frag_list(skb)) 1492 skb_clone_fraglist(skb); 1493 1494 skb_release_data(skb); 1495 } else { 1496 skb_free_head(skb); 1497 } 1498 off = (data + nhead) - skb->head; 1499 1500 skb->head = data; 1501 skb->head_frag = 0; 1502 skb->data += off; 1503 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1504 skb->end = size; 1505 off = nhead; 1506 #else 1507 skb->end = skb->head + size; 1508 #endif 1509 skb->tail += off; 1510 skb_headers_offset_update(skb, nhead); 1511 skb->cloned = 0; 1512 skb->hdr_len = 0; 1513 skb->nohdr = 0; 1514 atomic_set(&skb_shinfo(skb)->dataref, 1); 1515 1516 /* It is not generally safe to change skb->truesize. 1517 * For the moment, we really care of rx path, or 1518 * when skb is orphaned (not attached to a socket). 1519 */ 1520 if (!skb->sk || skb->destructor == sock_edemux) 1521 skb->truesize += size - osize; 1522 1523 return 0; 1524 1525 nofrags: 1526 kfree(data); 1527 nodata: 1528 return -ENOMEM; 1529 } 1530 EXPORT_SYMBOL(pskb_expand_head); 1531 1532 /* Make private copy of skb with writable head and some headroom */ 1533 1534 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 1535 { 1536 struct sk_buff *skb2; 1537 int delta = headroom - skb_headroom(skb); 1538 1539 if (delta <= 0) 1540 skb2 = pskb_copy(skb, GFP_ATOMIC); 1541 else { 1542 skb2 = skb_clone(skb, GFP_ATOMIC); 1543 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 1544 GFP_ATOMIC)) { 1545 kfree_skb(skb2); 1546 skb2 = NULL; 1547 } 1548 } 1549 return skb2; 1550 } 1551 EXPORT_SYMBOL(skb_realloc_headroom); 1552 1553 /** 1554 * skb_copy_expand - copy and expand sk_buff 1555 * @skb: buffer to copy 1556 * @newheadroom: new free bytes at head 1557 * @newtailroom: new free bytes at tail 1558 * @gfp_mask: allocation priority 1559 * 1560 * Make a copy of both an &sk_buff and its data and while doing so 1561 * allocate additional space. 1562 * 1563 * This is used when the caller wishes to modify the data and needs a 1564 * private copy of the data to alter as well as more space for new fields. 1565 * Returns %NULL on failure or the pointer to the buffer 1566 * on success. The returned buffer has a reference count of 1. 1567 * 1568 * You must pass %GFP_ATOMIC as the allocation priority if this function 1569 * is called from an interrupt. 1570 */ 1571 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 1572 int newheadroom, int newtailroom, 1573 gfp_t gfp_mask) 1574 { 1575 /* 1576 * Allocate the copy buffer 1577 */ 1578 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom, 1579 gfp_mask, skb_alloc_rx_flag(skb), 1580 NUMA_NO_NODE); 1581 int oldheadroom = skb_headroom(skb); 1582 int head_copy_len, head_copy_off; 1583 1584 if (!n) 1585 return NULL; 1586 1587 skb_reserve(n, newheadroom); 1588 1589 /* Set the tail pointer and length */ 1590 skb_put(n, skb->len); 1591 1592 head_copy_len = oldheadroom; 1593 head_copy_off = 0; 1594 if (newheadroom <= head_copy_len) 1595 head_copy_len = newheadroom; 1596 else 1597 head_copy_off = newheadroom - head_copy_len; 1598 1599 /* Copy the linear header and data. */ 1600 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 1601 skb->len + head_copy_len)) 1602 BUG(); 1603 1604 copy_skb_header(n, skb); 1605 1606 skb_headers_offset_update(n, newheadroom - oldheadroom); 1607 1608 return n; 1609 } 1610 EXPORT_SYMBOL(skb_copy_expand); 1611 1612 /** 1613 * __skb_pad - zero pad the tail of an skb 1614 * @skb: buffer to pad 1615 * @pad: space to pad 1616 * @free_on_error: free buffer on error 1617 * 1618 * Ensure that a buffer is followed by a padding area that is zero 1619 * filled. Used by network drivers which may DMA or transfer data 1620 * beyond the buffer end onto the wire. 1621 * 1622 * May return error in out of memory cases. The skb is freed on error 1623 * if @free_on_error is true. 1624 */ 1625 1626 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error) 1627 { 1628 int err; 1629 int ntail; 1630 1631 /* If the skbuff is non linear tailroom is always zero.. */ 1632 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 1633 memset(skb->data+skb->len, 0, pad); 1634 return 0; 1635 } 1636 1637 ntail = skb->data_len + pad - (skb->end - skb->tail); 1638 if (likely(skb_cloned(skb) || ntail > 0)) { 1639 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 1640 if (unlikely(err)) 1641 goto free_skb; 1642 } 1643 1644 /* FIXME: The use of this function with non-linear skb's really needs 1645 * to be audited. 1646 */ 1647 err = skb_linearize(skb); 1648 if (unlikely(err)) 1649 goto free_skb; 1650 1651 memset(skb->data + skb->len, 0, pad); 1652 return 0; 1653 1654 free_skb: 1655 if (free_on_error) 1656 kfree_skb(skb); 1657 return err; 1658 } 1659 EXPORT_SYMBOL(__skb_pad); 1660 1661 /** 1662 * pskb_put - add data to the tail of a potentially fragmented buffer 1663 * @skb: start of the buffer to use 1664 * @tail: tail fragment of the buffer to use 1665 * @len: amount of data to add 1666 * 1667 * This function extends the used data area of the potentially 1668 * fragmented buffer. @tail must be the last fragment of @skb -- or 1669 * @skb itself. If this would exceed the total buffer size the kernel 1670 * will panic. A pointer to the first byte of the extra data is 1671 * returned. 1672 */ 1673 1674 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len) 1675 { 1676 if (tail != skb) { 1677 skb->data_len += len; 1678 skb->len += len; 1679 } 1680 return skb_put(tail, len); 1681 } 1682 EXPORT_SYMBOL_GPL(pskb_put); 1683 1684 /** 1685 * skb_put - add data to a buffer 1686 * @skb: buffer to use 1687 * @len: amount of data to add 1688 * 1689 * This function extends the used data area of the buffer. If this would 1690 * exceed the total buffer size the kernel will panic. A pointer to the 1691 * first byte of the extra data is returned. 1692 */ 1693 void *skb_put(struct sk_buff *skb, unsigned int len) 1694 { 1695 void *tmp = skb_tail_pointer(skb); 1696 SKB_LINEAR_ASSERT(skb); 1697 skb->tail += len; 1698 skb->len += len; 1699 if (unlikely(skb->tail > skb->end)) 1700 skb_over_panic(skb, len, __builtin_return_address(0)); 1701 return tmp; 1702 } 1703 EXPORT_SYMBOL(skb_put); 1704 1705 /** 1706 * skb_push - add data to the start of a buffer 1707 * @skb: buffer to use 1708 * @len: amount of data to add 1709 * 1710 * This function extends the used data area of the buffer at the buffer 1711 * start. If this would exceed the total buffer headroom the kernel will 1712 * panic. A pointer to the first byte of the extra data is returned. 1713 */ 1714 void *skb_push(struct sk_buff *skb, unsigned int len) 1715 { 1716 skb->data -= len; 1717 skb->len += len; 1718 if (unlikely(skb->data<skb->head)) 1719 skb_under_panic(skb, len, __builtin_return_address(0)); 1720 return skb->data; 1721 } 1722 EXPORT_SYMBOL(skb_push); 1723 1724 /** 1725 * skb_pull - remove data from the start of a buffer 1726 * @skb: buffer to use 1727 * @len: amount of data to remove 1728 * 1729 * This function removes data from the start of a buffer, returning 1730 * the memory to the headroom. A pointer to the next data in the buffer 1731 * is returned. Once the data has been pulled future pushes will overwrite 1732 * the old data. 1733 */ 1734 void *skb_pull(struct sk_buff *skb, unsigned int len) 1735 { 1736 return skb_pull_inline(skb, len); 1737 } 1738 EXPORT_SYMBOL(skb_pull); 1739 1740 /** 1741 * skb_trim - remove end from a buffer 1742 * @skb: buffer to alter 1743 * @len: new length 1744 * 1745 * Cut the length of a buffer down by removing data from the tail. If 1746 * the buffer is already under the length specified it is not modified. 1747 * The skb must be linear. 1748 */ 1749 void skb_trim(struct sk_buff *skb, unsigned int len) 1750 { 1751 if (skb->len > len) 1752 __skb_trim(skb, len); 1753 } 1754 EXPORT_SYMBOL(skb_trim); 1755 1756 /* Trims skb to length len. It can change skb pointers. 1757 */ 1758 1759 int ___pskb_trim(struct sk_buff *skb, unsigned int len) 1760 { 1761 struct sk_buff **fragp; 1762 struct sk_buff *frag; 1763 int offset = skb_headlen(skb); 1764 int nfrags = skb_shinfo(skb)->nr_frags; 1765 int i; 1766 int err; 1767 1768 if (skb_cloned(skb) && 1769 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 1770 return err; 1771 1772 i = 0; 1773 if (offset >= len) 1774 goto drop_pages; 1775 1776 for (; i < nfrags; i++) { 1777 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]); 1778 1779 if (end < len) { 1780 offset = end; 1781 continue; 1782 } 1783 1784 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset); 1785 1786 drop_pages: 1787 skb_shinfo(skb)->nr_frags = i; 1788 1789 for (; i < nfrags; i++) 1790 skb_frag_unref(skb, i); 1791 1792 if (skb_has_frag_list(skb)) 1793 skb_drop_fraglist(skb); 1794 goto done; 1795 } 1796 1797 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 1798 fragp = &frag->next) { 1799 int end = offset + frag->len; 1800 1801 if (skb_shared(frag)) { 1802 struct sk_buff *nfrag; 1803 1804 nfrag = skb_clone(frag, GFP_ATOMIC); 1805 if (unlikely(!nfrag)) 1806 return -ENOMEM; 1807 1808 nfrag->next = frag->next; 1809 consume_skb(frag); 1810 frag = nfrag; 1811 *fragp = frag; 1812 } 1813 1814 if (end < len) { 1815 offset = end; 1816 continue; 1817 } 1818 1819 if (end > len && 1820 unlikely((err = pskb_trim(frag, len - offset)))) 1821 return err; 1822 1823 if (frag->next) 1824 skb_drop_list(&frag->next); 1825 break; 1826 } 1827 1828 done: 1829 if (len > skb_headlen(skb)) { 1830 skb->data_len -= skb->len - len; 1831 skb->len = len; 1832 } else { 1833 skb->len = len; 1834 skb->data_len = 0; 1835 skb_set_tail_pointer(skb, len); 1836 } 1837 1838 if (!skb->sk || skb->destructor == sock_edemux) 1839 skb_condense(skb); 1840 return 0; 1841 } 1842 EXPORT_SYMBOL(___pskb_trim); 1843 1844 /** 1845 * __pskb_pull_tail - advance tail of skb header 1846 * @skb: buffer to reallocate 1847 * @delta: number of bytes to advance tail 1848 * 1849 * The function makes a sense only on a fragmented &sk_buff, 1850 * it expands header moving its tail forward and copying necessary 1851 * data from fragmented part. 1852 * 1853 * &sk_buff MUST have reference count of 1. 1854 * 1855 * Returns %NULL (and &sk_buff does not change) if pull failed 1856 * or value of new tail of skb in the case of success. 1857 * 1858 * All the pointers pointing into skb header may change and must be 1859 * reloaded after call to this function. 1860 */ 1861 1862 /* Moves tail of skb head forward, copying data from fragmented part, 1863 * when it is necessary. 1864 * 1. It may fail due to malloc failure. 1865 * 2. It may change skb pointers. 1866 * 1867 * It is pretty complicated. Luckily, it is called only in exceptional cases. 1868 */ 1869 void *__pskb_pull_tail(struct sk_buff *skb, int delta) 1870 { 1871 /* If skb has not enough free space at tail, get new one 1872 * plus 128 bytes for future expansions. If we have enough 1873 * room at tail, reallocate without expansion only if skb is cloned. 1874 */ 1875 int i, k, eat = (skb->tail + delta) - skb->end; 1876 1877 if (eat > 0 || skb_cloned(skb)) { 1878 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 1879 GFP_ATOMIC)) 1880 return NULL; 1881 } 1882 1883 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 1884 BUG(); 1885 1886 /* Optimization: no fragments, no reasons to preestimate 1887 * size of pulled pages. Superb. 1888 */ 1889 if (!skb_has_frag_list(skb)) 1890 goto pull_pages; 1891 1892 /* Estimate size of pulled pages. */ 1893 eat = delta; 1894 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1895 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 1896 1897 if (size >= eat) 1898 goto pull_pages; 1899 eat -= size; 1900 } 1901 1902 /* If we need update frag list, we are in troubles. 1903 * Certainly, it is possible to add an offset to skb data, 1904 * but taking into account that pulling is expected to 1905 * be very rare operation, it is worth to fight against 1906 * further bloating skb head and crucify ourselves here instead. 1907 * Pure masohism, indeed. 8)8) 1908 */ 1909 if (eat) { 1910 struct sk_buff *list = skb_shinfo(skb)->frag_list; 1911 struct sk_buff *clone = NULL; 1912 struct sk_buff *insp = NULL; 1913 1914 do { 1915 BUG_ON(!list); 1916 1917 if (list->len <= eat) { 1918 /* Eaten as whole. */ 1919 eat -= list->len; 1920 list = list->next; 1921 insp = list; 1922 } else { 1923 /* Eaten partially. */ 1924 1925 if (skb_shared(list)) { 1926 /* Sucks! We need to fork list. :-( */ 1927 clone = skb_clone(list, GFP_ATOMIC); 1928 if (!clone) 1929 return NULL; 1930 insp = list->next; 1931 list = clone; 1932 } else { 1933 /* This may be pulled without 1934 * problems. */ 1935 insp = list; 1936 } 1937 if (!pskb_pull(list, eat)) { 1938 kfree_skb(clone); 1939 return NULL; 1940 } 1941 break; 1942 } 1943 } while (eat); 1944 1945 /* Free pulled out fragments. */ 1946 while ((list = skb_shinfo(skb)->frag_list) != insp) { 1947 skb_shinfo(skb)->frag_list = list->next; 1948 kfree_skb(list); 1949 } 1950 /* And insert new clone at head. */ 1951 if (clone) { 1952 clone->next = list; 1953 skb_shinfo(skb)->frag_list = clone; 1954 } 1955 } 1956 /* Success! Now we may commit changes to skb data. */ 1957 1958 pull_pages: 1959 eat = delta; 1960 k = 0; 1961 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1962 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 1963 1964 if (size <= eat) { 1965 skb_frag_unref(skb, i); 1966 eat -= size; 1967 } else { 1968 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 1969 if (eat) { 1970 skb_shinfo(skb)->frags[k].page_offset += eat; 1971 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat); 1972 if (!i) 1973 goto end; 1974 eat = 0; 1975 } 1976 k++; 1977 } 1978 } 1979 skb_shinfo(skb)->nr_frags = k; 1980 1981 end: 1982 skb->tail += delta; 1983 skb->data_len -= delta; 1984 1985 if (!skb->data_len) 1986 skb_zcopy_clear(skb, false); 1987 1988 return skb_tail_pointer(skb); 1989 } 1990 EXPORT_SYMBOL(__pskb_pull_tail); 1991 1992 /** 1993 * skb_copy_bits - copy bits from skb to kernel buffer 1994 * @skb: source skb 1995 * @offset: offset in source 1996 * @to: destination buffer 1997 * @len: number of bytes to copy 1998 * 1999 * Copy the specified number of bytes from the source skb to the 2000 * destination buffer. 2001 * 2002 * CAUTION ! : 2003 * If its prototype is ever changed, 2004 * check arch/{*}/net/{*}.S files, 2005 * since it is called from BPF assembly code. 2006 */ 2007 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 2008 { 2009 int start = skb_headlen(skb); 2010 struct sk_buff *frag_iter; 2011 int i, copy; 2012 2013 if (offset > (int)skb->len - len) 2014 goto fault; 2015 2016 /* Copy header. */ 2017 if ((copy = start - offset) > 0) { 2018 if (copy > len) 2019 copy = len; 2020 skb_copy_from_linear_data_offset(skb, offset, to, copy); 2021 if ((len -= copy) == 0) 2022 return 0; 2023 offset += copy; 2024 to += copy; 2025 } 2026 2027 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2028 int end; 2029 skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 2030 2031 WARN_ON(start > offset + len); 2032 2033 end = start + skb_frag_size(f); 2034 if ((copy = end - offset) > 0) { 2035 u32 p_off, p_len, copied; 2036 struct page *p; 2037 u8 *vaddr; 2038 2039 if (copy > len) 2040 copy = len; 2041 2042 skb_frag_foreach_page(f, 2043 f->page_offset + offset - start, 2044 copy, p, p_off, p_len, copied) { 2045 vaddr = kmap_atomic(p); 2046 memcpy(to + copied, vaddr + p_off, p_len); 2047 kunmap_atomic(vaddr); 2048 } 2049 2050 if ((len -= copy) == 0) 2051 return 0; 2052 offset += copy; 2053 to += copy; 2054 } 2055 start = end; 2056 } 2057 2058 skb_walk_frags(skb, frag_iter) { 2059 int end; 2060 2061 WARN_ON(start > offset + len); 2062 2063 end = start + frag_iter->len; 2064 if ((copy = end - offset) > 0) { 2065 if (copy > len) 2066 copy = len; 2067 if (skb_copy_bits(frag_iter, offset - start, to, copy)) 2068 goto fault; 2069 if ((len -= copy) == 0) 2070 return 0; 2071 offset += copy; 2072 to += copy; 2073 } 2074 start = end; 2075 } 2076 2077 if (!len) 2078 return 0; 2079 2080 fault: 2081 return -EFAULT; 2082 } 2083 EXPORT_SYMBOL(skb_copy_bits); 2084 2085 /* 2086 * Callback from splice_to_pipe(), if we need to release some pages 2087 * at the end of the spd in case we error'ed out in filling the pipe. 2088 */ 2089 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 2090 { 2091 put_page(spd->pages[i]); 2092 } 2093 2094 static struct page *linear_to_page(struct page *page, unsigned int *len, 2095 unsigned int *offset, 2096 struct sock *sk) 2097 { 2098 struct page_frag *pfrag = sk_page_frag(sk); 2099 2100 if (!sk_page_frag_refill(sk, pfrag)) 2101 return NULL; 2102 2103 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset); 2104 2105 memcpy(page_address(pfrag->page) + pfrag->offset, 2106 page_address(page) + *offset, *len); 2107 *offset = pfrag->offset; 2108 pfrag->offset += *len; 2109 2110 return pfrag->page; 2111 } 2112 2113 static bool spd_can_coalesce(const struct splice_pipe_desc *spd, 2114 struct page *page, 2115 unsigned int offset) 2116 { 2117 return spd->nr_pages && 2118 spd->pages[spd->nr_pages - 1] == page && 2119 (spd->partial[spd->nr_pages - 1].offset + 2120 spd->partial[spd->nr_pages - 1].len == offset); 2121 } 2122 2123 /* 2124 * Fill page/offset/length into spd, if it can hold more pages. 2125 */ 2126 static bool spd_fill_page(struct splice_pipe_desc *spd, 2127 struct pipe_inode_info *pipe, struct page *page, 2128 unsigned int *len, unsigned int offset, 2129 bool linear, 2130 struct sock *sk) 2131 { 2132 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS)) 2133 return true; 2134 2135 if (linear) { 2136 page = linear_to_page(page, len, &offset, sk); 2137 if (!page) 2138 return true; 2139 } 2140 if (spd_can_coalesce(spd, page, offset)) { 2141 spd->partial[spd->nr_pages - 1].len += *len; 2142 return false; 2143 } 2144 get_page(page); 2145 spd->pages[spd->nr_pages] = page; 2146 spd->partial[spd->nr_pages].len = *len; 2147 spd->partial[spd->nr_pages].offset = offset; 2148 spd->nr_pages++; 2149 2150 return false; 2151 } 2152 2153 static bool __splice_segment(struct page *page, unsigned int poff, 2154 unsigned int plen, unsigned int *off, 2155 unsigned int *len, 2156 struct splice_pipe_desc *spd, bool linear, 2157 struct sock *sk, 2158 struct pipe_inode_info *pipe) 2159 { 2160 if (!*len) 2161 return true; 2162 2163 /* skip this segment if already processed */ 2164 if (*off >= plen) { 2165 *off -= plen; 2166 return false; 2167 } 2168 2169 /* ignore any bits we already processed */ 2170 poff += *off; 2171 plen -= *off; 2172 *off = 0; 2173 2174 do { 2175 unsigned int flen = min(*len, plen); 2176 2177 if (spd_fill_page(spd, pipe, page, &flen, poff, 2178 linear, sk)) 2179 return true; 2180 poff += flen; 2181 plen -= flen; 2182 *len -= flen; 2183 } while (*len && plen); 2184 2185 return false; 2186 } 2187 2188 /* 2189 * Map linear and fragment data from the skb to spd. It reports true if the 2190 * pipe is full or if we already spliced the requested length. 2191 */ 2192 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe, 2193 unsigned int *offset, unsigned int *len, 2194 struct splice_pipe_desc *spd, struct sock *sk) 2195 { 2196 int seg; 2197 struct sk_buff *iter; 2198 2199 /* map the linear part : 2200 * If skb->head_frag is set, this 'linear' part is backed by a 2201 * fragment, and if the head is not shared with any clones then 2202 * we can avoid a copy since we own the head portion of this page. 2203 */ 2204 if (__splice_segment(virt_to_page(skb->data), 2205 (unsigned long) skb->data & (PAGE_SIZE - 1), 2206 skb_headlen(skb), 2207 offset, len, spd, 2208 skb_head_is_locked(skb), 2209 sk, pipe)) 2210 return true; 2211 2212 /* 2213 * then map the fragments 2214 */ 2215 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 2216 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 2217 2218 if (__splice_segment(skb_frag_page(f), 2219 f->page_offset, skb_frag_size(f), 2220 offset, len, spd, false, sk, pipe)) 2221 return true; 2222 } 2223 2224 skb_walk_frags(skb, iter) { 2225 if (*offset >= iter->len) { 2226 *offset -= iter->len; 2227 continue; 2228 } 2229 /* __skb_splice_bits() only fails if the output has no room 2230 * left, so no point in going over the frag_list for the error 2231 * case. 2232 */ 2233 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk)) 2234 return true; 2235 } 2236 2237 return false; 2238 } 2239 2240 /* 2241 * Map data from the skb to a pipe. Should handle both the linear part, 2242 * the fragments, and the frag list. 2243 */ 2244 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 2245 struct pipe_inode_info *pipe, unsigned int tlen, 2246 unsigned int flags) 2247 { 2248 struct partial_page partial[MAX_SKB_FRAGS]; 2249 struct page *pages[MAX_SKB_FRAGS]; 2250 struct splice_pipe_desc spd = { 2251 .pages = pages, 2252 .partial = partial, 2253 .nr_pages_max = MAX_SKB_FRAGS, 2254 .ops = &nosteal_pipe_buf_ops, 2255 .spd_release = sock_spd_release, 2256 }; 2257 int ret = 0; 2258 2259 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk); 2260 2261 if (spd.nr_pages) 2262 ret = splice_to_pipe(pipe, &spd); 2263 2264 return ret; 2265 } 2266 EXPORT_SYMBOL_GPL(skb_splice_bits); 2267 2268 /* Send skb data on a socket. Socket must be locked. */ 2269 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 2270 int len) 2271 { 2272 unsigned int orig_len = len; 2273 struct sk_buff *head = skb; 2274 unsigned short fragidx; 2275 int slen, ret; 2276 2277 do_frag_list: 2278 2279 /* Deal with head data */ 2280 while (offset < skb_headlen(skb) && len) { 2281 struct kvec kv; 2282 struct msghdr msg; 2283 2284 slen = min_t(int, len, skb_headlen(skb) - offset); 2285 kv.iov_base = skb->data + offset; 2286 kv.iov_len = slen; 2287 memset(&msg, 0, sizeof(msg)); 2288 2289 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen); 2290 if (ret <= 0) 2291 goto error; 2292 2293 offset += ret; 2294 len -= ret; 2295 } 2296 2297 /* All the data was skb head? */ 2298 if (!len) 2299 goto out; 2300 2301 /* Make offset relative to start of frags */ 2302 offset -= skb_headlen(skb); 2303 2304 /* Find where we are in frag list */ 2305 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 2306 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 2307 2308 if (offset < frag->size) 2309 break; 2310 2311 offset -= frag->size; 2312 } 2313 2314 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) { 2315 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx]; 2316 2317 slen = min_t(size_t, len, frag->size - offset); 2318 2319 while (slen) { 2320 ret = kernel_sendpage_locked(sk, frag->page.p, 2321 frag->page_offset + offset, 2322 slen, MSG_DONTWAIT); 2323 if (ret <= 0) 2324 goto error; 2325 2326 len -= ret; 2327 offset += ret; 2328 slen -= ret; 2329 } 2330 2331 offset = 0; 2332 } 2333 2334 if (len) { 2335 /* Process any frag lists */ 2336 2337 if (skb == head) { 2338 if (skb_has_frag_list(skb)) { 2339 skb = skb_shinfo(skb)->frag_list; 2340 goto do_frag_list; 2341 } 2342 } else if (skb->next) { 2343 skb = skb->next; 2344 goto do_frag_list; 2345 } 2346 } 2347 2348 out: 2349 return orig_len - len; 2350 2351 error: 2352 return orig_len == len ? ret : orig_len - len; 2353 } 2354 EXPORT_SYMBOL_GPL(skb_send_sock_locked); 2355 2356 /* Send skb data on a socket. */ 2357 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len) 2358 { 2359 int ret = 0; 2360 2361 lock_sock(sk); 2362 ret = skb_send_sock_locked(sk, skb, offset, len); 2363 release_sock(sk); 2364 2365 return ret; 2366 } 2367 EXPORT_SYMBOL_GPL(skb_send_sock); 2368 2369 /** 2370 * skb_store_bits - store bits from kernel buffer to skb 2371 * @skb: destination buffer 2372 * @offset: offset in destination 2373 * @from: source buffer 2374 * @len: number of bytes to copy 2375 * 2376 * Copy the specified number of bytes from the source buffer to the 2377 * destination skb. This function handles all the messy bits of 2378 * traversing fragment lists and such. 2379 */ 2380 2381 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 2382 { 2383 int start = skb_headlen(skb); 2384 struct sk_buff *frag_iter; 2385 int i, copy; 2386 2387 if (offset > (int)skb->len - len) 2388 goto fault; 2389 2390 if ((copy = start - offset) > 0) { 2391 if (copy > len) 2392 copy = len; 2393 skb_copy_to_linear_data_offset(skb, offset, from, copy); 2394 if ((len -= copy) == 0) 2395 return 0; 2396 offset += copy; 2397 from += copy; 2398 } 2399 2400 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2401 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2402 int end; 2403 2404 WARN_ON(start > offset + len); 2405 2406 end = start + skb_frag_size(frag); 2407 if ((copy = end - offset) > 0) { 2408 u32 p_off, p_len, copied; 2409 struct page *p; 2410 u8 *vaddr; 2411 2412 if (copy > len) 2413 copy = len; 2414 2415 skb_frag_foreach_page(frag, 2416 frag->page_offset + offset - start, 2417 copy, p, p_off, p_len, copied) { 2418 vaddr = kmap_atomic(p); 2419 memcpy(vaddr + p_off, from + copied, p_len); 2420 kunmap_atomic(vaddr); 2421 } 2422 2423 if ((len -= copy) == 0) 2424 return 0; 2425 offset += copy; 2426 from += copy; 2427 } 2428 start = end; 2429 } 2430 2431 skb_walk_frags(skb, frag_iter) { 2432 int end; 2433 2434 WARN_ON(start > offset + len); 2435 2436 end = start + frag_iter->len; 2437 if ((copy = end - offset) > 0) { 2438 if (copy > len) 2439 copy = len; 2440 if (skb_store_bits(frag_iter, offset - start, 2441 from, copy)) 2442 goto fault; 2443 if ((len -= copy) == 0) 2444 return 0; 2445 offset += copy; 2446 from += copy; 2447 } 2448 start = end; 2449 } 2450 if (!len) 2451 return 0; 2452 2453 fault: 2454 return -EFAULT; 2455 } 2456 EXPORT_SYMBOL(skb_store_bits); 2457 2458 /* Checksum skb data. */ 2459 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2460 __wsum csum, const struct skb_checksum_ops *ops) 2461 { 2462 int start = skb_headlen(skb); 2463 int i, copy = start - offset; 2464 struct sk_buff *frag_iter; 2465 int pos = 0; 2466 2467 /* Checksum header. */ 2468 if (copy > 0) { 2469 if (copy > len) 2470 copy = len; 2471 csum = ops->update(skb->data + offset, copy, csum); 2472 if ((len -= copy) == 0) 2473 return csum; 2474 offset += copy; 2475 pos = copy; 2476 } 2477 2478 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2479 int end; 2480 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2481 2482 WARN_ON(start > offset + len); 2483 2484 end = start + skb_frag_size(frag); 2485 if ((copy = end - offset) > 0) { 2486 u32 p_off, p_len, copied; 2487 struct page *p; 2488 __wsum csum2; 2489 u8 *vaddr; 2490 2491 if (copy > len) 2492 copy = len; 2493 2494 skb_frag_foreach_page(frag, 2495 frag->page_offset + offset - start, 2496 copy, p, p_off, p_len, copied) { 2497 vaddr = kmap_atomic(p); 2498 csum2 = ops->update(vaddr + p_off, p_len, 0); 2499 kunmap_atomic(vaddr); 2500 csum = ops->combine(csum, csum2, pos, p_len); 2501 pos += p_len; 2502 } 2503 2504 if (!(len -= copy)) 2505 return csum; 2506 offset += copy; 2507 } 2508 start = end; 2509 } 2510 2511 skb_walk_frags(skb, frag_iter) { 2512 int end; 2513 2514 WARN_ON(start > offset + len); 2515 2516 end = start + frag_iter->len; 2517 if ((copy = end - offset) > 0) { 2518 __wsum csum2; 2519 if (copy > len) 2520 copy = len; 2521 csum2 = __skb_checksum(frag_iter, offset - start, 2522 copy, 0, ops); 2523 csum = ops->combine(csum, csum2, pos, copy); 2524 if ((len -= copy) == 0) 2525 return csum; 2526 offset += copy; 2527 pos += copy; 2528 } 2529 start = end; 2530 } 2531 BUG_ON(len); 2532 2533 return csum; 2534 } 2535 EXPORT_SYMBOL(__skb_checksum); 2536 2537 __wsum skb_checksum(const struct sk_buff *skb, int offset, 2538 int len, __wsum csum) 2539 { 2540 const struct skb_checksum_ops ops = { 2541 .update = csum_partial_ext, 2542 .combine = csum_block_add_ext, 2543 }; 2544 2545 return __skb_checksum(skb, offset, len, csum, &ops); 2546 } 2547 EXPORT_SYMBOL(skb_checksum); 2548 2549 /* Both of above in one bottle. */ 2550 2551 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 2552 u8 *to, int len, __wsum csum) 2553 { 2554 int start = skb_headlen(skb); 2555 int i, copy = start - offset; 2556 struct sk_buff *frag_iter; 2557 int pos = 0; 2558 2559 /* Copy header. */ 2560 if (copy > 0) { 2561 if (copy > len) 2562 copy = len; 2563 csum = csum_partial_copy_nocheck(skb->data + offset, to, 2564 copy, csum); 2565 if ((len -= copy) == 0) 2566 return csum; 2567 offset += copy; 2568 to += copy; 2569 pos = copy; 2570 } 2571 2572 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2573 int end; 2574 2575 WARN_ON(start > offset + len); 2576 2577 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 2578 if ((copy = end - offset) > 0) { 2579 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2580 u32 p_off, p_len, copied; 2581 struct page *p; 2582 __wsum csum2; 2583 u8 *vaddr; 2584 2585 if (copy > len) 2586 copy = len; 2587 2588 skb_frag_foreach_page(frag, 2589 frag->page_offset + offset - start, 2590 copy, p, p_off, p_len, copied) { 2591 vaddr = kmap_atomic(p); 2592 csum2 = csum_partial_copy_nocheck(vaddr + p_off, 2593 to + copied, 2594 p_len, 0); 2595 kunmap_atomic(vaddr); 2596 csum = csum_block_add(csum, csum2, pos); 2597 pos += p_len; 2598 } 2599 2600 if (!(len -= copy)) 2601 return csum; 2602 offset += copy; 2603 to += copy; 2604 } 2605 start = end; 2606 } 2607 2608 skb_walk_frags(skb, frag_iter) { 2609 __wsum csum2; 2610 int end; 2611 2612 WARN_ON(start > offset + len); 2613 2614 end = start + frag_iter->len; 2615 if ((copy = end - offset) > 0) { 2616 if (copy > len) 2617 copy = len; 2618 csum2 = skb_copy_and_csum_bits(frag_iter, 2619 offset - start, 2620 to, copy, 0); 2621 csum = csum_block_add(csum, csum2, pos); 2622 if ((len -= copy) == 0) 2623 return csum; 2624 offset += copy; 2625 to += copy; 2626 pos += copy; 2627 } 2628 start = end; 2629 } 2630 BUG_ON(len); 2631 return csum; 2632 } 2633 EXPORT_SYMBOL(skb_copy_and_csum_bits); 2634 2635 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum) 2636 { 2637 net_warn_ratelimited( 2638 "%s: attempt to compute crc32c without libcrc32c.ko\n", 2639 __func__); 2640 return 0; 2641 } 2642 2643 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2, 2644 int offset, int len) 2645 { 2646 net_warn_ratelimited( 2647 "%s: attempt to compute crc32c without libcrc32c.ko\n", 2648 __func__); 2649 return 0; 2650 } 2651 2652 static const struct skb_checksum_ops default_crc32c_ops = { 2653 .update = warn_crc32c_csum_update, 2654 .combine = warn_crc32c_csum_combine, 2655 }; 2656 2657 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly = 2658 &default_crc32c_ops; 2659 EXPORT_SYMBOL(crc32c_csum_stub); 2660 2661 /** 2662 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy() 2663 * @from: source buffer 2664 * 2665 * Calculates the amount of linear headroom needed in the 'to' skb passed 2666 * into skb_zerocopy(). 2667 */ 2668 unsigned int 2669 skb_zerocopy_headlen(const struct sk_buff *from) 2670 { 2671 unsigned int hlen = 0; 2672 2673 if (!from->head_frag || 2674 skb_headlen(from) < L1_CACHE_BYTES || 2675 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 2676 hlen = skb_headlen(from); 2677 2678 if (skb_has_frag_list(from)) 2679 hlen = from->len; 2680 2681 return hlen; 2682 } 2683 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen); 2684 2685 /** 2686 * skb_zerocopy - Zero copy skb to skb 2687 * @to: destination buffer 2688 * @from: source buffer 2689 * @len: number of bytes to copy from source buffer 2690 * @hlen: size of linear headroom in destination buffer 2691 * 2692 * Copies up to `len` bytes from `from` to `to` by creating references 2693 * to the frags in the source buffer. 2694 * 2695 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the 2696 * headroom in the `to` buffer. 2697 * 2698 * Return value: 2699 * 0: everything is OK 2700 * -ENOMEM: couldn't orphan frags of @from due to lack of memory 2701 * -EFAULT: skb_copy_bits() found some problem with skb geometry 2702 */ 2703 int 2704 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen) 2705 { 2706 int i, j = 0; 2707 int plen = 0; /* length of skb->head fragment */ 2708 int ret; 2709 struct page *page; 2710 unsigned int offset; 2711 2712 BUG_ON(!from->head_frag && !hlen); 2713 2714 /* dont bother with small payloads */ 2715 if (len <= skb_tailroom(to)) 2716 return skb_copy_bits(from, 0, skb_put(to, len), len); 2717 2718 if (hlen) { 2719 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen); 2720 if (unlikely(ret)) 2721 return ret; 2722 len -= hlen; 2723 } else { 2724 plen = min_t(int, skb_headlen(from), len); 2725 if (plen) { 2726 page = virt_to_head_page(from->head); 2727 offset = from->data - (unsigned char *)page_address(page); 2728 __skb_fill_page_desc(to, 0, page, offset, plen); 2729 get_page(page); 2730 j = 1; 2731 len -= plen; 2732 } 2733 } 2734 2735 to->truesize += len + plen; 2736 to->len += len + plen; 2737 to->data_len += len + plen; 2738 2739 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) { 2740 skb_tx_error(from); 2741 return -ENOMEM; 2742 } 2743 skb_zerocopy_clone(to, from, GFP_ATOMIC); 2744 2745 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) { 2746 if (!len) 2747 break; 2748 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i]; 2749 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len); 2750 len -= skb_shinfo(to)->frags[j].size; 2751 skb_frag_ref(to, j); 2752 j++; 2753 } 2754 skb_shinfo(to)->nr_frags = j; 2755 2756 return 0; 2757 } 2758 EXPORT_SYMBOL_GPL(skb_zerocopy); 2759 2760 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 2761 { 2762 __wsum csum; 2763 long csstart; 2764 2765 if (skb->ip_summed == CHECKSUM_PARTIAL) 2766 csstart = skb_checksum_start_offset(skb); 2767 else 2768 csstart = skb_headlen(skb); 2769 2770 BUG_ON(csstart > skb_headlen(skb)); 2771 2772 skb_copy_from_linear_data(skb, to, csstart); 2773 2774 csum = 0; 2775 if (csstart != skb->len) 2776 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 2777 skb->len - csstart, 0); 2778 2779 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2780 long csstuff = csstart + skb->csum_offset; 2781 2782 *((__sum16 *)(to + csstuff)) = csum_fold(csum); 2783 } 2784 } 2785 EXPORT_SYMBOL(skb_copy_and_csum_dev); 2786 2787 /** 2788 * skb_dequeue - remove from the head of the queue 2789 * @list: list to dequeue from 2790 * 2791 * Remove the head of the list. The list lock is taken so the function 2792 * may be used safely with other locking list functions. The head item is 2793 * returned or %NULL if the list is empty. 2794 */ 2795 2796 struct sk_buff *skb_dequeue(struct sk_buff_head *list) 2797 { 2798 unsigned long flags; 2799 struct sk_buff *result; 2800 2801 spin_lock_irqsave(&list->lock, flags); 2802 result = __skb_dequeue(list); 2803 spin_unlock_irqrestore(&list->lock, flags); 2804 return result; 2805 } 2806 EXPORT_SYMBOL(skb_dequeue); 2807 2808 /** 2809 * skb_dequeue_tail - remove from the tail of the queue 2810 * @list: list to dequeue from 2811 * 2812 * Remove the tail of the list. The list lock is taken so the function 2813 * may be used safely with other locking list functions. The tail item is 2814 * returned or %NULL if the list is empty. 2815 */ 2816 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 2817 { 2818 unsigned long flags; 2819 struct sk_buff *result; 2820 2821 spin_lock_irqsave(&list->lock, flags); 2822 result = __skb_dequeue_tail(list); 2823 spin_unlock_irqrestore(&list->lock, flags); 2824 return result; 2825 } 2826 EXPORT_SYMBOL(skb_dequeue_tail); 2827 2828 /** 2829 * skb_queue_purge - empty a list 2830 * @list: list to empty 2831 * 2832 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2833 * the list and one reference dropped. This function takes the list 2834 * lock and is atomic with respect to other list locking functions. 2835 */ 2836 void skb_queue_purge(struct sk_buff_head *list) 2837 { 2838 struct sk_buff *skb; 2839 while ((skb = skb_dequeue(list)) != NULL) 2840 kfree_skb(skb); 2841 } 2842 EXPORT_SYMBOL(skb_queue_purge); 2843 2844 /** 2845 * skb_rbtree_purge - empty a skb rbtree 2846 * @root: root of the rbtree to empty 2847 * 2848 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from 2849 * the list and one reference dropped. This function does not take 2850 * any lock. Synchronization should be handled by the caller (e.g., TCP 2851 * out-of-order queue is protected by the socket lock). 2852 */ 2853 void skb_rbtree_purge(struct rb_root *root) 2854 { 2855 struct sk_buff *skb, *next; 2856 2857 rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode) 2858 kfree_skb(skb); 2859 2860 *root = RB_ROOT; 2861 } 2862 2863 /** 2864 * skb_queue_head - queue a buffer at the list head 2865 * @list: list to use 2866 * @newsk: buffer to queue 2867 * 2868 * Queue a buffer at the start of the list. This function takes the 2869 * list lock and can be used safely with other locking &sk_buff functions 2870 * safely. 2871 * 2872 * A buffer cannot be placed on two lists at the same time. 2873 */ 2874 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 2875 { 2876 unsigned long flags; 2877 2878 spin_lock_irqsave(&list->lock, flags); 2879 __skb_queue_head(list, newsk); 2880 spin_unlock_irqrestore(&list->lock, flags); 2881 } 2882 EXPORT_SYMBOL(skb_queue_head); 2883 2884 /** 2885 * skb_queue_tail - queue a buffer at the list tail 2886 * @list: list to use 2887 * @newsk: buffer to queue 2888 * 2889 * Queue a buffer at the tail of the list. This function takes the 2890 * list lock and can be used safely with other locking &sk_buff functions 2891 * safely. 2892 * 2893 * A buffer cannot be placed on two lists at the same time. 2894 */ 2895 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 2896 { 2897 unsigned long flags; 2898 2899 spin_lock_irqsave(&list->lock, flags); 2900 __skb_queue_tail(list, newsk); 2901 spin_unlock_irqrestore(&list->lock, flags); 2902 } 2903 EXPORT_SYMBOL(skb_queue_tail); 2904 2905 /** 2906 * skb_unlink - remove a buffer from a list 2907 * @skb: buffer to remove 2908 * @list: list to use 2909 * 2910 * Remove a packet from a list. The list locks are taken and this 2911 * function is atomic with respect to other list locked calls 2912 * 2913 * You must know what list the SKB is on. 2914 */ 2915 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2916 { 2917 unsigned long flags; 2918 2919 spin_lock_irqsave(&list->lock, flags); 2920 __skb_unlink(skb, list); 2921 spin_unlock_irqrestore(&list->lock, flags); 2922 } 2923 EXPORT_SYMBOL(skb_unlink); 2924 2925 /** 2926 * skb_append - append a buffer 2927 * @old: buffer to insert after 2928 * @newsk: buffer to insert 2929 * @list: list to use 2930 * 2931 * Place a packet after a given packet in a list. The list locks are taken 2932 * and this function is atomic with respect to other list locked calls. 2933 * A buffer cannot be placed on two lists at the same time. 2934 */ 2935 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 2936 { 2937 unsigned long flags; 2938 2939 spin_lock_irqsave(&list->lock, flags); 2940 __skb_queue_after(list, old, newsk); 2941 spin_unlock_irqrestore(&list->lock, flags); 2942 } 2943 EXPORT_SYMBOL(skb_append); 2944 2945 /** 2946 * skb_insert - insert a buffer 2947 * @old: buffer to insert before 2948 * @newsk: buffer to insert 2949 * @list: list to use 2950 * 2951 * Place a packet before a given packet in a list. The list locks are 2952 * taken and this function is atomic with respect to other list locked 2953 * calls. 2954 * 2955 * A buffer cannot be placed on two lists at the same time. 2956 */ 2957 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 2958 { 2959 unsigned long flags; 2960 2961 spin_lock_irqsave(&list->lock, flags); 2962 __skb_insert(newsk, old->prev, old, list); 2963 spin_unlock_irqrestore(&list->lock, flags); 2964 } 2965 EXPORT_SYMBOL(skb_insert); 2966 2967 static inline void skb_split_inside_header(struct sk_buff *skb, 2968 struct sk_buff* skb1, 2969 const u32 len, const int pos) 2970 { 2971 int i; 2972 2973 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2974 pos - len); 2975 /* And move data appendix as is. */ 2976 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 2977 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 2978 2979 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 2980 skb_shinfo(skb)->nr_frags = 0; 2981 skb1->data_len = skb->data_len; 2982 skb1->len += skb1->data_len; 2983 skb->data_len = 0; 2984 skb->len = len; 2985 skb_set_tail_pointer(skb, len); 2986 } 2987 2988 static inline void skb_split_no_header(struct sk_buff *skb, 2989 struct sk_buff* skb1, 2990 const u32 len, int pos) 2991 { 2992 int i, k = 0; 2993 const int nfrags = skb_shinfo(skb)->nr_frags; 2994 2995 skb_shinfo(skb)->nr_frags = 0; 2996 skb1->len = skb1->data_len = skb->len - len; 2997 skb->len = len; 2998 skb->data_len = len - pos; 2999 3000 for (i = 0; i < nfrags; i++) { 3001 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]); 3002 3003 if (pos + size > len) { 3004 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 3005 3006 if (pos < len) { 3007 /* Split frag. 3008 * We have two variants in this case: 3009 * 1. Move all the frag to the second 3010 * part, if it is possible. F.e. 3011 * this approach is mandatory for TUX, 3012 * where splitting is expensive. 3013 * 2. Split is accurately. We make this. 3014 */ 3015 skb_frag_ref(skb, i); 3016 skb_shinfo(skb1)->frags[0].page_offset += len - pos; 3017 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos); 3018 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos); 3019 skb_shinfo(skb)->nr_frags++; 3020 } 3021 k++; 3022 } else 3023 skb_shinfo(skb)->nr_frags++; 3024 pos += size; 3025 } 3026 skb_shinfo(skb1)->nr_frags = k; 3027 } 3028 3029 /** 3030 * skb_split - Split fragmented skb to two parts at length len. 3031 * @skb: the buffer to split 3032 * @skb1: the buffer to receive the second part 3033 * @len: new length for skb 3034 */ 3035 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 3036 { 3037 int pos = skb_headlen(skb); 3038 3039 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags & 3040 SKBTX_SHARED_FRAG; 3041 skb_zerocopy_clone(skb1, skb, 0); 3042 if (len < pos) /* Split line is inside header. */ 3043 skb_split_inside_header(skb, skb1, len, pos); 3044 else /* Second chunk has no header, nothing to copy. */ 3045 skb_split_no_header(skb, skb1, len, pos); 3046 } 3047 EXPORT_SYMBOL(skb_split); 3048 3049 /* Shifting from/to a cloned skb is a no-go. 3050 * 3051 * Caller cannot keep skb_shinfo related pointers past calling here! 3052 */ 3053 static int skb_prepare_for_shift(struct sk_buff *skb) 3054 { 3055 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3056 } 3057 3058 /** 3059 * skb_shift - Shifts paged data partially from skb to another 3060 * @tgt: buffer into which tail data gets added 3061 * @skb: buffer from which the paged data comes from 3062 * @shiftlen: shift up to this many bytes 3063 * 3064 * Attempts to shift up to shiftlen worth of bytes, which may be less than 3065 * the length of the skb, from skb to tgt. Returns number bytes shifted. 3066 * It's up to caller to free skb if everything was shifted. 3067 * 3068 * If @tgt runs out of frags, the whole operation is aborted. 3069 * 3070 * Skb cannot include anything else but paged data while tgt is allowed 3071 * to have non-paged data as well. 3072 * 3073 * TODO: full sized shift could be optimized but that would need 3074 * specialized skb free'er to handle frags without up-to-date nr_frags. 3075 */ 3076 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 3077 { 3078 int from, to, merge, todo; 3079 struct skb_frag_struct *fragfrom, *fragto; 3080 3081 BUG_ON(shiftlen > skb->len); 3082 3083 if (skb_headlen(skb)) 3084 return 0; 3085 if (skb_zcopy(tgt) || skb_zcopy(skb)) 3086 return 0; 3087 3088 todo = shiftlen; 3089 from = 0; 3090 to = skb_shinfo(tgt)->nr_frags; 3091 fragfrom = &skb_shinfo(skb)->frags[from]; 3092 3093 /* Actual merge is delayed until the point when we know we can 3094 * commit all, so that we don't have to undo partial changes 3095 */ 3096 if (!to || 3097 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom), 3098 fragfrom->page_offset)) { 3099 merge = -1; 3100 } else { 3101 merge = to - 1; 3102 3103 todo -= skb_frag_size(fragfrom); 3104 if (todo < 0) { 3105 if (skb_prepare_for_shift(skb) || 3106 skb_prepare_for_shift(tgt)) 3107 return 0; 3108 3109 /* All previous frag pointers might be stale! */ 3110 fragfrom = &skb_shinfo(skb)->frags[from]; 3111 fragto = &skb_shinfo(tgt)->frags[merge]; 3112 3113 skb_frag_size_add(fragto, shiftlen); 3114 skb_frag_size_sub(fragfrom, shiftlen); 3115 fragfrom->page_offset += shiftlen; 3116 3117 goto onlymerged; 3118 } 3119 3120 from++; 3121 } 3122 3123 /* Skip full, not-fitting skb to avoid expensive operations */ 3124 if ((shiftlen == skb->len) && 3125 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 3126 return 0; 3127 3128 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 3129 return 0; 3130 3131 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 3132 if (to == MAX_SKB_FRAGS) 3133 return 0; 3134 3135 fragfrom = &skb_shinfo(skb)->frags[from]; 3136 fragto = &skb_shinfo(tgt)->frags[to]; 3137 3138 if (todo >= skb_frag_size(fragfrom)) { 3139 *fragto = *fragfrom; 3140 todo -= skb_frag_size(fragfrom); 3141 from++; 3142 to++; 3143 3144 } else { 3145 __skb_frag_ref(fragfrom); 3146 fragto->page = fragfrom->page; 3147 fragto->page_offset = fragfrom->page_offset; 3148 skb_frag_size_set(fragto, todo); 3149 3150 fragfrom->page_offset += todo; 3151 skb_frag_size_sub(fragfrom, todo); 3152 todo = 0; 3153 3154 to++; 3155 break; 3156 } 3157 } 3158 3159 /* Ready to "commit" this state change to tgt */ 3160 skb_shinfo(tgt)->nr_frags = to; 3161 3162 if (merge >= 0) { 3163 fragfrom = &skb_shinfo(skb)->frags[0]; 3164 fragto = &skb_shinfo(tgt)->frags[merge]; 3165 3166 skb_frag_size_add(fragto, skb_frag_size(fragfrom)); 3167 __skb_frag_unref(fragfrom); 3168 } 3169 3170 /* Reposition in the original skb */ 3171 to = 0; 3172 while (from < skb_shinfo(skb)->nr_frags) 3173 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 3174 skb_shinfo(skb)->nr_frags = to; 3175 3176 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 3177 3178 onlymerged: 3179 /* Most likely the tgt won't ever need its checksum anymore, skb on 3180 * the other hand might need it if it needs to be resent 3181 */ 3182 tgt->ip_summed = CHECKSUM_PARTIAL; 3183 skb->ip_summed = CHECKSUM_PARTIAL; 3184 3185 /* Yak, is it really working this way? Some helper please? */ 3186 skb->len -= shiftlen; 3187 skb->data_len -= shiftlen; 3188 skb->truesize -= shiftlen; 3189 tgt->len += shiftlen; 3190 tgt->data_len += shiftlen; 3191 tgt->truesize += shiftlen; 3192 3193 return shiftlen; 3194 } 3195 3196 /** 3197 * skb_prepare_seq_read - Prepare a sequential read of skb data 3198 * @skb: the buffer to read 3199 * @from: lower offset of data to be read 3200 * @to: upper offset of data to be read 3201 * @st: state variable 3202 * 3203 * Initializes the specified state variable. Must be called before 3204 * invoking skb_seq_read() for the first time. 3205 */ 3206 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 3207 unsigned int to, struct skb_seq_state *st) 3208 { 3209 st->lower_offset = from; 3210 st->upper_offset = to; 3211 st->root_skb = st->cur_skb = skb; 3212 st->frag_idx = st->stepped_offset = 0; 3213 st->frag_data = NULL; 3214 } 3215 EXPORT_SYMBOL(skb_prepare_seq_read); 3216 3217 /** 3218 * skb_seq_read - Sequentially read skb data 3219 * @consumed: number of bytes consumed by the caller so far 3220 * @data: destination pointer for data to be returned 3221 * @st: state variable 3222 * 3223 * Reads a block of skb data at @consumed relative to the 3224 * lower offset specified to skb_prepare_seq_read(). Assigns 3225 * the head of the data block to @data and returns the length 3226 * of the block or 0 if the end of the skb data or the upper 3227 * offset has been reached. 3228 * 3229 * The caller is not required to consume all of the data 3230 * returned, i.e. @consumed is typically set to the number 3231 * of bytes already consumed and the next call to 3232 * skb_seq_read() will return the remaining part of the block. 3233 * 3234 * Note 1: The size of each block of data returned can be arbitrary, 3235 * this limitation is the cost for zerocopy sequential 3236 * reads of potentially non linear data. 3237 * 3238 * Note 2: Fragment lists within fragments are not implemented 3239 * at the moment, state->root_skb could be replaced with 3240 * a stack for this purpose. 3241 */ 3242 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 3243 struct skb_seq_state *st) 3244 { 3245 unsigned int block_limit, abs_offset = consumed + st->lower_offset; 3246 skb_frag_t *frag; 3247 3248 if (unlikely(abs_offset >= st->upper_offset)) { 3249 if (st->frag_data) { 3250 kunmap_atomic(st->frag_data); 3251 st->frag_data = NULL; 3252 } 3253 return 0; 3254 } 3255 3256 next_skb: 3257 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 3258 3259 if (abs_offset < block_limit && !st->frag_data) { 3260 *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 3261 return block_limit - abs_offset; 3262 } 3263 3264 if (st->frag_idx == 0 && !st->frag_data) 3265 st->stepped_offset += skb_headlen(st->cur_skb); 3266 3267 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 3268 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 3269 block_limit = skb_frag_size(frag) + st->stepped_offset; 3270 3271 if (abs_offset < block_limit) { 3272 if (!st->frag_data) 3273 st->frag_data = kmap_atomic(skb_frag_page(frag)); 3274 3275 *data = (u8 *) st->frag_data + frag->page_offset + 3276 (abs_offset - st->stepped_offset); 3277 3278 return block_limit - abs_offset; 3279 } 3280 3281 if (st->frag_data) { 3282 kunmap_atomic(st->frag_data); 3283 st->frag_data = NULL; 3284 } 3285 3286 st->frag_idx++; 3287 st->stepped_offset += skb_frag_size(frag); 3288 } 3289 3290 if (st->frag_data) { 3291 kunmap_atomic(st->frag_data); 3292 st->frag_data = NULL; 3293 } 3294 3295 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) { 3296 st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 3297 st->frag_idx = 0; 3298 goto next_skb; 3299 } else if (st->cur_skb->next) { 3300 st->cur_skb = st->cur_skb->next; 3301 st->frag_idx = 0; 3302 goto next_skb; 3303 } 3304 3305 return 0; 3306 } 3307 EXPORT_SYMBOL(skb_seq_read); 3308 3309 /** 3310 * skb_abort_seq_read - Abort a sequential read of skb data 3311 * @st: state variable 3312 * 3313 * Must be called if skb_seq_read() was not called until it 3314 * returned 0. 3315 */ 3316 void skb_abort_seq_read(struct skb_seq_state *st) 3317 { 3318 if (st->frag_data) 3319 kunmap_atomic(st->frag_data); 3320 } 3321 EXPORT_SYMBOL(skb_abort_seq_read); 3322 3323 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 3324 3325 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 3326 struct ts_config *conf, 3327 struct ts_state *state) 3328 { 3329 return skb_seq_read(offset, text, TS_SKB_CB(state)); 3330 } 3331 3332 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 3333 { 3334 skb_abort_seq_read(TS_SKB_CB(state)); 3335 } 3336 3337 /** 3338 * skb_find_text - Find a text pattern in skb data 3339 * @skb: the buffer to look in 3340 * @from: search offset 3341 * @to: search limit 3342 * @config: textsearch configuration 3343 * 3344 * Finds a pattern in the skb data according to the specified 3345 * textsearch configuration. Use textsearch_next() to retrieve 3346 * subsequent occurrences of the pattern. Returns the offset 3347 * to the first occurrence or UINT_MAX if no match was found. 3348 */ 3349 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 3350 unsigned int to, struct ts_config *config) 3351 { 3352 struct ts_state state; 3353 unsigned int ret; 3354 3355 config->get_next_block = skb_ts_get_next_block; 3356 config->finish = skb_ts_finish; 3357 3358 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state)); 3359 3360 ret = textsearch_find(config, &state); 3361 return (ret <= to - from ? ret : UINT_MAX); 3362 } 3363 EXPORT_SYMBOL(skb_find_text); 3364 3365 /** 3366 * skb_append_datato_frags - append the user data to a skb 3367 * @sk: sock structure 3368 * @skb: skb structure to be appended with user data. 3369 * @getfrag: call back function to be used for getting the user data 3370 * @from: pointer to user message iov 3371 * @length: length of the iov message 3372 * 3373 * Description: This procedure append the user data in the fragment part 3374 * of the skb if any page alloc fails user this procedure returns -ENOMEM 3375 */ 3376 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 3377 int (*getfrag)(void *from, char *to, int offset, 3378 int len, int odd, struct sk_buff *skb), 3379 void *from, int length) 3380 { 3381 int frg_cnt = skb_shinfo(skb)->nr_frags; 3382 int copy; 3383 int offset = 0; 3384 int ret; 3385 struct page_frag *pfrag = ¤t->task_frag; 3386 3387 do { 3388 /* Return error if we don't have space for new frag */ 3389 if (frg_cnt >= MAX_SKB_FRAGS) 3390 return -EMSGSIZE; 3391 3392 if (!sk_page_frag_refill(sk, pfrag)) 3393 return -ENOMEM; 3394 3395 /* copy the user data to page */ 3396 copy = min_t(int, length, pfrag->size - pfrag->offset); 3397 3398 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset, 3399 offset, copy, 0, skb); 3400 if (ret < 0) 3401 return -EFAULT; 3402 3403 /* copy was successful so update the size parameters */ 3404 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset, 3405 copy); 3406 frg_cnt++; 3407 pfrag->offset += copy; 3408 get_page(pfrag->page); 3409 3410 skb->truesize += copy; 3411 refcount_add(copy, &sk->sk_wmem_alloc); 3412 skb->len += copy; 3413 skb->data_len += copy; 3414 offset += copy; 3415 length -= copy; 3416 3417 } while (length > 0); 3418 3419 return 0; 3420 } 3421 EXPORT_SYMBOL(skb_append_datato_frags); 3422 3423 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 3424 int offset, size_t size) 3425 { 3426 int i = skb_shinfo(skb)->nr_frags; 3427 3428 if (skb_can_coalesce(skb, i, page, offset)) { 3429 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size); 3430 } else if (i < MAX_SKB_FRAGS) { 3431 get_page(page); 3432 skb_fill_page_desc(skb, i, page, offset, size); 3433 } else { 3434 return -EMSGSIZE; 3435 } 3436 3437 return 0; 3438 } 3439 EXPORT_SYMBOL_GPL(skb_append_pagefrags); 3440 3441 /** 3442 * skb_pull_rcsum - pull skb and update receive checksum 3443 * @skb: buffer to update 3444 * @len: length of data pulled 3445 * 3446 * This function performs an skb_pull on the packet and updates 3447 * the CHECKSUM_COMPLETE checksum. It should be used on 3448 * receive path processing instead of skb_pull unless you know 3449 * that the checksum difference is zero (e.g., a valid IP header) 3450 * or you are setting ip_summed to CHECKSUM_NONE. 3451 */ 3452 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 3453 { 3454 unsigned char *data = skb->data; 3455 3456 BUG_ON(len > skb->len); 3457 __skb_pull(skb, len); 3458 skb_postpull_rcsum(skb, data, len); 3459 return skb->data; 3460 } 3461 EXPORT_SYMBOL_GPL(skb_pull_rcsum); 3462 3463 /** 3464 * skb_segment - Perform protocol segmentation on skb. 3465 * @head_skb: buffer to segment 3466 * @features: features for the output path (see dev->features) 3467 * 3468 * This function performs segmentation on the given skb. It returns 3469 * a pointer to the first in a list of new skbs for the segments. 3470 * In case of error it returns ERR_PTR(err). 3471 */ 3472 struct sk_buff *skb_segment(struct sk_buff *head_skb, 3473 netdev_features_t features) 3474 { 3475 struct sk_buff *segs = NULL; 3476 struct sk_buff *tail = NULL; 3477 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list; 3478 skb_frag_t *frag = skb_shinfo(head_skb)->frags; 3479 unsigned int mss = skb_shinfo(head_skb)->gso_size; 3480 unsigned int doffset = head_skb->data - skb_mac_header(head_skb); 3481 struct sk_buff *frag_skb = head_skb; 3482 unsigned int offset = doffset; 3483 unsigned int tnl_hlen = skb_tnl_header_len(head_skb); 3484 unsigned int partial_segs = 0; 3485 unsigned int headroom; 3486 unsigned int len = head_skb->len; 3487 __be16 proto; 3488 bool csum, sg; 3489 int nfrags = skb_shinfo(head_skb)->nr_frags; 3490 int err = -ENOMEM; 3491 int i = 0; 3492 int pos; 3493 int dummy; 3494 3495 __skb_push(head_skb, doffset); 3496 proto = skb_network_protocol(head_skb, &dummy); 3497 if (unlikely(!proto)) 3498 return ERR_PTR(-EINVAL); 3499 3500 sg = !!(features & NETIF_F_SG); 3501 csum = !!can_checksum_protocol(features, proto); 3502 3503 if (sg && csum && (mss != GSO_BY_FRAGS)) { 3504 if (!(features & NETIF_F_GSO_PARTIAL)) { 3505 struct sk_buff *iter; 3506 unsigned int frag_len; 3507 3508 if (!list_skb || 3509 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type)) 3510 goto normal; 3511 3512 /* If we get here then all the required 3513 * GSO features except frag_list are supported. 3514 * Try to split the SKB to multiple GSO SKBs 3515 * with no frag_list. 3516 * Currently we can do that only when the buffers don't 3517 * have a linear part and all the buffers except 3518 * the last are of the same length. 3519 */ 3520 frag_len = list_skb->len; 3521 skb_walk_frags(head_skb, iter) { 3522 if (frag_len != iter->len && iter->next) 3523 goto normal; 3524 if (skb_headlen(iter) && !iter->head_frag) 3525 goto normal; 3526 3527 len -= iter->len; 3528 } 3529 3530 if (len != frag_len) 3531 goto normal; 3532 } 3533 3534 /* GSO partial only requires that we trim off any excess that 3535 * doesn't fit into an MSS sized block, so take care of that 3536 * now. 3537 */ 3538 partial_segs = len / mss; 3539 if (partial_segs > 1) 3540 mss *= partial_segs; 3541 else 3542 partial_segs = 0; 3543 } 3544 3545 normal: 3546 headroom = skb_headroom(head_skb); 3547 pos = skb_headlen(head_skb); 3548 3549 do { 3550 struct sk_buff *nskb; 3551 skb_frag_t *nskb_frag; 3552 int hsize; 3553 int size; 3554 3555 if (unlikely(mss == GSO_BY_FRAGS)) { 3556 len = list_skb->len; 3557 } else { 3558 len = head_skb->len - offset; 3559 if (len > mss) 3560 len = mss; 3561 } 3562 3563 hsize = skb_headlen(head_skb) - offset; 3564 if (hsize < 0) 3565 hsize = 0; 3566 if (hsize > len || !sg) 3567 hsize = len; 3568 3569 if (!hsize && i >= nfrags && skb_headlen(list_skb) && 3570 (skb_headlen(list_skb) == len || sg)) { 3571 BUG_ON(skb_headlen(list_skb) > len); 3572 3573 i = 0; 3574 nfrags = skb_shinfo(list_skb)->nr_frags; 3575 frag = skb_shinfo(list_skb)->frags; 3576 frag_skb = list_skb; 3577 pos += skb_headlen(list_skb); 3578 3579 while (pos < offset + len) { 3580 BUG_ON(i >= nfrags); 3581 3582 size = skb_frag_size(frag); 3583 if (pos + size > offset + len) 3584 break; 3585 3586 i++; 3587 pos += size; 3588 frag++; 3589 } 3590 3591 nskb = skb_clone(list_skb, GFP_ATOMIC); 3592 list_skb = list_skb->next; 3593 3594 if (unlikely(!nskb)) 3595 goto err; 3596 3597 if (unlikely(pskb_trim(nskb, len))) { 3598 kfree_skb(nskb); 3599 goto err; 3600 } 3601 3602 hsize = skb_end_offset(nskb); 3603 if (skb_cow_head(nskb, doffset + headroom)) { 3604 kfree_skb(nskb); 3605 goto err; 3606 } 3607 3608 nskb->truesize += skb_end_offset(nskb) - hsize; 3609 skb_release_head_state(nskb); 3610 __skb_push(nskb, doffset); 3611 } else { 3612 nskb = __alloc_skb(hsize + doffset + headroom, 3613 GFP_ATOMIC, skb_alloc_rx_flag(head_skb), 3614 NUMA_NO_NODE); 3615 3616 if (unlikely(!nskb)) 3617 goto err; 3618 3619 skb_reserve(nskb, headroom); 3620 __skb_put(nskb, doffset); 3621 } 3622 3623 if (segs) 3624 tail->next = nskb; 3625 else 3626 segs = nskb; 3627 tail = nskb; 3628 3629 __copy_skb_header(nskb, head_skb); 3630 3631 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom); 3632 skb_reset_mac_len(nskb); 3633 3634 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen, 3635 nskb->data - tnl_hlen, 3636 doffset + tnl_hlen); 3637 3638 if (nskb->len == len + doffset) 3639 goto perform_csum_check; 3640 3641 if (!sg) { 3642 if (!nskb->remcsum_offload) 3643 nskb->ip_summed = CHECKSUM_NONE; 3644 SKB_GSO_CB(nskb)->csum = 3645 skb_copy_and_csum_bits(head_skb, offset, 3646 skb_put(nskb, len), 3647 len, 0); 3648 SKB_GSO_CB(nskb)->csum_start = 3649 skb_headroom(nskb) + doffset; 3650 continue; 3651 } 3652 3653 nskb_frag = skb_shinfo(nskb)->frags; 3654 3655 skb_copy_from_linear_data_offset(head_skb, offset, 3656 skb_put(nskb, hsize), hsize); 3657 3658 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags & 3659 SKBTX_SHARED_FRAG; 3660 if (skb_zerocopy_clone(nskb, head_skb, GFP_ATOMIC)) 3661 goto err; 3662 3663 while (pos < offset + len) { 3664 if (i >= nfrags) { 3665 BUG_ON(skb_headlen(list_skb)); 3666 3667 i = 0; 3668 nfrags = skb_shinfo(list_skb)->nr_frags; 3669 frag = skb_shinfo(list_skb)->frags; 3670 frag_skb = list_skb; 3671 3672 BUG_ON(!nfrags); 3673 3674 list_skb = list_skb->next; 3675 } 3676 3677 if (unlikely(skb_shinfo(nskb)->nr_frags >= 3678 MAX_SKB_FRAGS)) { 3679 net_warn_ratelimited( 3680 "skb_segment: too many frags: %u %u\n", 3681 pos, mss); 3682 goto err; 3683 } 3684 3685 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC))) 3686 goto err; 3687 3688 *nskb_frag = *frag; 3689 __skb_frag_ref(nskb_frag); 3690 size = skb_frag_size(nskb_frag); 3691 3692 if (pos < offset) { 3693 nskb_frag->page_offset += offset - pos; 3694 skb_frag_size_sub(nskb_frag, offset - pos); 3695 } 3696 3697 skb_shinfo(nskb)->nr_frags++; 3698 3699 if (pos + size <= offset + len) { 3700 i++; 3701 frag++; 3702 pos += size; 3703 } else { 3704 skb_frag_size_sub(nskb_frag, pos + size - (offset + len)); 3705 goto skip_fraglist; 3706 } 3707 3708 nskb_frag++; 3709 } 3710 3711 skip_fraglist: 3712 nskb->data_len = len - hsize; 3713 nskb->len += nskb->data_len; 3714 nskb->truesize += nskb->data_len; 3715 3716 perform_csum_check: 3717 if (!csum) { 3718 if (skb_has_shared_frag(nskb)) { 3719 err = __skb_linearize(nskb); 3720 if (err) 3721 goto err; 3722 } 3723 if (!nskb->remcsum_offload) 3724 nskb->ip_summed = CHECKSUM_NONE; 3725 SKB_GSO_CB(nskb)->csum = 3726 skb_checksum(nskb, doffset, 3727 nskb->len - doffset, 0); 3728 SKB_GSO_CB(nskb)->csum_start = 3729 skb_headroom(nskb) + doffset; 3730 } 3731 } while ((offset += len) < head_skb->len); 3732 3733 /* Some callers want to get the end of the list. 3734 * Put it in segs->prev to avoid walking the list. 3735 * (see validate_xmit_skb_list() for example) 3736 */ 3737 segs->prev = tail; 3738 3739 if (partial_segs) { 3740 struct sk_buff *iter; 3741 int type = skb_shinfo(head_skb)->gso_type; 3742 unsigned short gso_size = skb_shinfo(head_skb)->gso_size; 3743 3744 /* Update type to add partial and then remove dodgy if set */ 3745 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL; 3746 type &= ~SKB_GSO_DODGY; 3747 3748 /* Update GSO info and prepare to start updating headers on 3749 * our way back down the stack of protocols. 3750 */ 3751 for (iter = segs; iter; iter = iter->next) { 3752 skb_shinfo(iter)->gso_size = gso_size; 3753 skb_shinfo(iter)->gso_segs = partial_segs; 3754 skb_shinfo(iter)->gso_type = type; 3755 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset; 3756 } 3757 3758 if (tail->len - doffset <= gso_size) 3759 skb_shinfo(tail)->gso_size = 0; 3760 else if (tail != segs) 3761 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size); 3762 } 3763 3764 /* Following permits correct backpressure, for protocols 3765 * using skb_set_owner_w(). 3766 * Idea is to tranfert ownership from head_skb to last segment. 3767 */ 3768 if (head_skb->destructor == sock_wfree) { 3769 swap(tail->truesize, head_skb->truesize); 3770 swap(tail->destructor, head_skb->destructor); 3771 swap(tail->sk, head_skb->sk); 3772 } 3773 return segs; 3774 3775 err: 3776 kfree_skb_list(segs); 3777 return ERR_PTR(err); 3778 } 3779 EXPORT_SYMBOL_GPL(skb_segment); 3780 3781 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 3782 { 3783 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb); 3784 unsigned int offset = skb_gro_offset(skb); 3785 unsigned int headlen = skb_headlen(skb); 3786 unsigned int len = skb_gro_len(skb); 3787 struct sk_buff *lp, *p = *head; 3788 unsigned int delta_truesize; 3789 3790 if (unlikely(p->len + len >= 65536)) 3791 return -E2BIG; 3792 3793 lp = NAPI_GRO_CB(p)->last; 3794 pinfo = skb_shinfo(lp); 3795 3796 if (headlen <= offset) { 3797 skb_frag_t *frag; 3798 skb_frag_t *frag2; 3799 int i = skbinfo->nr_frags; 3800 int nr_frags = pinfo->nr_frags + i; 3801 3802 if (nr_frags > MAX_SKB_FRAGS) 3803 goto merge; 3804 3805 offset -= headlen; 3806 pinfo->nr_frags = nr_frags; 3807 skbinfo->nr_frags = 0; 3808 3809 frag = pinfo->frags + nr_frags; 3810 frag2 = skbinfo->frags + i; 3811 do { 3812 *--frag = *--frag2; 3813 } while (--i); 3814 3815 frag->page_offset += offset; 3816 skb_frag_size_sub(frag, offset); 3817 3818 /* all fragments truesize : remove (head size + sk_buff) */ 3819 delta_truesize = skb->truesize - 3820 SKB_TRUESIZE(skb_end_offset(skb)); 3821 3822 skb->truesize -= skb->data_len; 3823 skb->len -= skb->data_len; 3824 skb->data_len = 0; 3825 3826 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE; 3827 goto done; 3828 } else if (skb->head_frag) { 3829 int nr_frags = pinfo->nr_frags; 3830 skb_frag_t *frag = pinfo->frags + nr_frags; 3831 struct page *page = virt_to_head_page(skb->head); 3832 unsigned int first_size = headlen - offset; 3833 unsigned int first_offset; 3834 3835 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS) 3836 goto merge; 3837 3838 first_offset = skb->data - 3839 (unsigned char *)page_address(page) + 3840 offset; 3841 3842 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags; 3843 3844 frag->page.p = page; 3845 frag->page_offset = first_offset; 3846 skb_frag_size_set(frag, first_size); 3847 3848 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags); 3849 /* We dont need to clear skbinfo->nr_frags here */ 3850 3851 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 3852 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD; 3853 goto done; 3854 } 3855 3856 merge: 3857 delta_truesize = skb->truesize; 3858 if (offset > headlen) { 3859 unsigned int eat = offset - headlen; 3860 3861 skbinfo->frags[0].page_offset += eat; 3862 skb_frag_size_sub(&skbinfo->frags[0], eat); 3863 skb->data_len -= eat; 3864 skb->len -= eat; 3865 offset = headlen; 3866 } 3867 3868 __skb_pull(skb, offset); 3869 3870 if (NAPI_GRO_CB(p)->last == p) 3871 skb_shinfo(p)->frag_list = skb; 3872 else 3873 NAPI_GRO_CB(p)->last->next = skb; 3874 NAPI_GRO_CB(p)->last = skb; 3875 __skb_header_release(skb); 3876 lp = p; 3877 3878 done: 3879 NAPI_GRO_CB(p)->count++; 3880 p->data_len += len; 3881 p->truesize += delta_truesize; 3882 p->len += len; 3883 if (lp != p) { 3884 lp->data_len += len; 3885 lp->truesize += delta_truesize; 3886 lp->len += len; 3887 } 3888 NAPI_GRO_CB(skb)->same_flow = 1; 3889 return 0; 3890 } 3891 EXPORT_SYMBOL_GPL(skb_gro_receive); 3892 3893 void __init skb_init(void) 3894 { 3895 skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 3896 sizeof(struct sk_buff), 3897 0, 3898 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 3899 NULL); 3900 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 3901 sizeof(struct sk_buff_fclones), 3902 0, 3903 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 3904 NULL); 3905 } 3906 3907 static int 3908 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len, 3909 unsigned int recursion_level) 3910 { 3911 int start = skb_headlen(skb); 3912 int i, copy = start - offset; 3913 struct sk_buff *frag_iter; 3914 int elt = 0; 3915 3916 if (unlikely(recursion_level >= 24)) 3917 return -EMSGSIZE; 3918 3919 if (copy > 0) { 3920 if (copy > len) 3921 copy = len; 3922 sg_set_buf(sg, skb->data + offset, copy); 3923 elt++; 3924 if ((len -= copy) == 0) 3925 return elt; 3926 offset += copy; 3927 } 3928 3929 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3930 int end; 3931 3932 WARN_ON(start > offset + len); 3933 3934 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]); 3935 if ((copy = end - offset) > 0) { 3936 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3937 if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 3938 return -EMSGSIZE; 3939 3940 if (copy > len) 3941 copy = len; 3942 sg_set_page(&sg[elt], skb_frag_page(frag), copy, 3943 frag->page_offset+offset-start); 3944 elt++; 3945 if (!(len -= copy)) 3946 return elt; 3947 offset += copy; 3948 } 3949 start = end; 3950 } 3951 3952 skb_walk_frags(skb, frag_iter) { 3953 int end, ret; 3954 3955 WARN_ON(start > offset + len); 3956 3957 end = start + frag_iter->len; 3958 if ((copy = end - offset) > 0) { 3959 if (unlikely(elt && sg_is_last(&sg[elt - 1]))) 3960 return -EMSGSIZE; 3961 3962 if (copy > len) 3963 copy = len; 3964 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start, 3965 copy, recursion_level + 1); 3966 if (unlikely(ret < 0)) 3967 return ret; 3968 elt += ret; 3969 if ((len -= copy) == 0) 3970 return elt; 3971 offset += copy; 3972 } 3973 start = end; 3974 } 3975 BUG_ON(len); 3976 return elt; 3977 } 3978 3979 /** 3980 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 3981 * @skb: Socket buffer containing the buffers to be mapped 3982 * @sg: The scatter-gather list to map into 3983 * @offset: The offset into the buffer's contents to start mapping 3984 * @len: Length of buffer space to be mapped 3985 * 3986 * Fill the specified scatter-gather list with mappings/pointers into a 3987 * region of the buffer space attached to a socket buffer. Returns either 3988 * the number of scatterlist items used, or -EMSGSIZE if the contents 3989 * could not fit. 3990 */ 3991 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 3992 { 3993 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0); 3994 3995 if (nsg <= 0) 3996 return nsg; 3997 3998 sg_mark_end(&sg[nsg - 1]); 3999 4000 return nsg; 4001 } 4002 EXPORT_SYMBOL_GPL(skb_to_sgvec); 4003 4004 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given 4005 * sglist without mark the sg which contain last skb data as the end. 4006 * So the caller can mannipulate sg list as will when padding new data after 4007 * the first call without calling sg_unmark_end to expend sg list. 4008 * 4009 * Scenario to use skb_to_sgvec_nomark: 4010 * 1. sg_init_table 4011 * 2. skb_to_sgvec_nomark(payload1) 4012 * 3. skb_to_sgvec_nomark(payload2) 4013 * 4014 * This is equivalent to: 4015 * 1. sg_init_table 4016 * 2. skb_to_sgvec(payload1) 4017 * 3. sg_unmark_end 4018 * 4. skb_to_sgvec(payload2) 4019 * 4020 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark 4021 * is more preferable. 4022 */ 4023 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 4024 int offset, int len) 4025 { 4026 return __skb_to_sgvec(skb, sg, offset, len, 0); 4027 } 4028 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark); 4029 4030 4031 4032 /** 4033 * skb_cow_data - Check that a socket buffer's data buffers are writable 4034 * @skb: The socket buffer to check. 4035 * @tailbits: Amount of trailing space to be added 4036 * @trailer: Returned pointer to the skb where the @tailbits space begins 4037 * 4038 * Make sure that the data buffers attached to a socket buffer are 4039 * writable. If they are not, private copies are made of the data buffers 4040 * and the socket buffer is set to use these instead. 4041 * 4042 * If @tailbits is given, make sure that there is space to write @tailbits 4043 * bytes of data beyond current end of socket buffer. @trailer will be 4044 * set to point to the skb in which this space begins. 4045 * 4046 * The number of scatterlist elements required to completely map the 4047 * COW'd and extended socket buffer will be returned. 4048 */ 4049 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 4050 { 4051 int copyflag; 4052 int elt; 4053 struct sk_buff *skb1, **skb_p; 4054 4055 /* If skb is cloned or its head is paged, reallocate 4056 * head pulling out all the pages (pages are considered not writable 4057 * at the moment even if they are anonymous). 4058 */ 4059 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 4060 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 4061 return -ENOMEM; 4062 4063 /* Easy case. Most of packets will go this way. */ 4064 if (!skb_has_frag_list(skb)) { 4065 /* A little of trouble, not enough of space for trailer. 4066 * This should not happen, when stack is tuned to generate 4067 * good frames. OK, on miss we reallocate and reserve even more 4068 * space, 128 bytes is fair. */ 4069 4070 if (skb_tailroom(skb) < tailbits && 4071 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 4072 return -ENOMEM; 4073 4074 /* Voila! */ 4075 *trailer = skb; 4076 return 1; 4077 } 4078 4079 /* Misery. We are in troubles, going to mincer fragments... */ 4080 4081 elt = 1; 4082 skb_p = &skb_shinfo(skb)->frag_list; 4083 copyflag = 0; 4084 4085 while ((skb1 = *skb_p) != NULL) { 4086 int ntail = 0; 4087 4088 /* The fragment is partially pulled by someone, 4089 * this can happen on input. Copy it and everything 4090 * after it. */ 4091 4092 if (skb_shared(skb1)) 4093 copyflag = 1; 4094 4095 /* If the skb is the last, worry about trailer. */ 4096 4097 if (skb1->next == NULL && tailbits) { 4098 if (skb_shinfo(skb1)->nr_frags || 4099 skb_has_frag_list(skb1) || 4100 skb_tailroom(skb1) < tailbits) 4101 ntail = tailbits + 128; 4102 } 4103 4104 if (copyflag || 4105 skb_cloned(skb1) || 4106 ntail || 4107 skb_shinfo(skb1)->nr_frags || 4108 skb_has_frag_list(skb1)) { 4109 struct sk_buff *skb2; 4110 4111 /* Fuck, we are miserable poor guys... */ 4112 if (ntail == 0) 4113 skb2 = skb_copy(skb1, GFP_ATOMIC); 4114 else 4115 skb2 = skb_copy_expand(skb1, 4116 skb_headroom(skb1), 4117 ntail, 4118 GFP_ATOMIC); 4119 if (unlikely(skb2 == NULL)) 4120 return -ENOMEM; 4121 4122 if (skb1->sk) 4123 skb_set_owner_w(skb2, skb1->sk); 4124 4125 /* Looking around. Are we still alive? 4126 * OK, link new skb, drop old one */ 4127 4128 skb2->next = skb1->next; 4129 *skb_p = skb2; 4130 kfree_skb(skb1); 4131 skb1 = skb2; 4132 } 4133 elt++; 4134 *trailer = skb1; 4135 skb_p = &skb1->next; 4136 } 4137 4138 return elt; 4139 } 4140 EXPORT_SYMBOL_GPL(skb_cow_data); 4141 4142 static void sock_rmem_free(struct sk_buff *skb) 4143 { 4144 struct sock *sk = skb->sk; 4145 4146 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 4147 } 4148 4149 static void skb_set_err_queue(struct sk_buff *skb) 4150 { 4151 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING. 4152 * So, it is safe to (mis)use it to mark skbs on the error queue. 4153 */ 4154 skb->pkt_type = PACKET_OUTGOING; 4155 BUILD_BUG_ON(PACKET_OUTGOING == 0); 4156 } 4157 4158 /* 4159 * Note: We dont mem charge error packets (no sk_forward_alloc changes) 4160 */ 4161 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 4162 { 4163 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 4164 (unsigned int)sk->sk_rcvbuf) 4165 return -ENOMEM; 4166 4167 skb_orphan(skb); 4168 skb->sk = sk; 4169 skb->destructor = sock_rmem_free; 4170 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 4171 skb_set_err_queue(skb); 4172 4173 /* before exiting rcu section, make sure dst is refcounted */ 4174 skb_dst_force(skb); 4175 4176 skb_queue_tail(&sk->sk_error_queue, skb); 4177 if (!sock_flag(sk, SOCK_DEAD)) 4178 sk->sk_data_ready(sk); 4179 return 0; 4180 } 4181 EXPORT_SYMBOL(sock_queue_err_skb); 4182 4183 static bool is_icmp_err_skb(const struct sk_buff *skb) 4184 { 4185 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP || 4186 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6); 4187 } 4188 4189 struct sk_buff *sock_dequeue_err_skb(struct sock *sk) 4190 { 4191 struct sk_buff_head *q = &sk->sk_error_queue; 4192 struct sk_buff *skb, *skb_next = NULL; 4193 bool icmp_next = false; 4194 unsigned long flags; 4195 4196 spin_lock_irqsave(&q->lock, flags); 4197 skb = __skb_dequeue(q); 4198 if (skb && (skb_next = skb_peek(q))) { 4199 icmp_next = is_icmp_err_skb(skb_next); 4200 if (icmp_next) 4201 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin; 4202 } 4203 spin_unlock_irqrestore(&q->lock, flags); 4204 4205 if (is_icmp_err_skb(skb) && !icmp_next) 4206 sk->sk_err = 0; 4207 4208 if (skb_next) 4209 sk->sk_error_report(sk); 4210 4211 return skb; 4212 } 4213 EXPORT_SYMBOL(sock_dequeue_err_skb); 4214 4215 /** 4216 * skb_clone_sk - create clone of skb, and take reference to socket 4217 * @skb: the skb to clone 4218 * 4219 * This function creates a clone of a buffer that holds a reference on 4220 * sk_refcnt. Buffers created via this function are meant to be 4221 * returned using sock_queue_err_skb, or free via kfree_skb. 4222 * 4223 * When passing buffers allocated with this function to sock_queue_err_skb 4224 * it is necessary to wrap the call with sock_hold/sock_put in order to 4225 * prevent the socket from being released prior to being enqueued on 4226 * the sk_error_queue. 4227 */ 4228 struct sk_buff *skb_clone_sk(struct sk_buff *skb) 4229 { 4230 struct sock *sk = skb->sk; 4231 struct sk_buff *clone; 4232 4233 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 4234 return NULL; 4235 4236 clone = skb_clone(skb, GFP_ATOMIC); 4237 if (!clone) { 4238 sock_put(sk); 4239 return NULL; 4240 } 4241 4242 clone->sk = sk; 4243 clone->destructor = sock_efree; 4244 4245 return clone; 4246 } 4247 EXPORT_SYMBOL(skb_clone_sk); 4248 4249 static void __skb_complete_tx_timestamp(struct sk_buff *skb, 4250 struct sock *sk, 4251 int tstype, 4252 bool opt_stats) 4253 { 4254 struct sock_exterr_skb *serr; 4255 int err; 4256 4257 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb)); 4258 4259 serr = SKB_EXT_ERR(skb); 4260 memset(serr, 0, sizeof(*serr)); 4261 serr->ee.ee_errno = ENOMSG; 4262 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING; 4263 serr->ee.ee_info = tstype; 4264 serr->opt_stats = opt_stats; 4265 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0; 4266 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) { 4267 serr->ee.ee_data = skb_shinfo(skb)->tskey; 4268 if (sk->sk_protocol == IPPROTO_TCP && 4269 sk->sk_type == SOCK_STREAM) 4270 serr->ee.ee_data -= sk->sk_tskey; 4271 } 4272 4273 err = sock_queue_err_skb(sk, skb); 4274 4275 if (err) 4276 kfree_skb(skb); 4277 } 4278 4279 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly) 4280 { 4281 bool ret; 4282 4283 if (likely(sysctl_tstamp_allow_data || tsonly)) 4284 return true; 4285 4286 read_lock_bh(&sk->sk_callback_lock); 4287 ret = sk->sk_socket && sk->sk_socket->file && 4288 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW); 4289 read_unlock_bh(&sk->sk_callback_lock); 4290 return ret; 4291 } 4292 4293 void skb_complete_tx_timestamp(struct sk_buff *skb, 4294 struct skb_shared_hwtstamps *hwtstamps) 4295 { 4296 struct sock *sk = skb->sk; 4297 4298 if (!skb_may_tx_timestamp(sk, false)) 4299 return; 4300 4301 /* Take a reference to prevent skb_orphan() from freeing the socket, 4302 * but only if the socket refcount is not zero. 4303 */ 4304 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 4305 *skb_hwtstamps(skb) = *hwtstamps; 4306 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false); 4307 sock_put(sk); 4308 } 4309 } 4310 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp); 4311 4312 void __skb_tstamp_tx(struct sk_buff *orig_skb, 4313 struct skb_shared_hwtstamps *hwtstamps, 4314 struct sock *sk, int tstype) 4315 { 4316 struct sk_buff *skb; 4317 bool tsonly, opt_stats = false; 4318 4319 if (!sk) 4320 return; 4321 4322 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) && 4323 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS) 4324 return; 4325 4326 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY; 4327 if (!skb_may_tx_timestamp(sk, tsonly)) 4328 return; 4329 4330 if (tsonly) { 4331 #ifdef CONFIG_INET 4332 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) && 4333 sk->sk_protocol == IPPROTO_TCP && 4334 sk->sk_type == SOCK_STREAM) { 4335 skb = tcp_get_timestamping_opt_stats(sk); 4336 opt_stats = true; 4337 } else 4338 #endif 4339 skb = alloc_skb(0, GFP_ATOMIC); 4340 } else { 4341 skb = skb_clone(orig_skb, GFP_ATOMIC); 4342 } 4343 if (!skb) 4344 return; 4345 4346 if (tsonly) { 4347 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags & 4348 SKBTX_ANY_TSTAMP; 4349 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey; 4350 } 4351 4352 if (hwtstamps) 4353 *skb_hwtstamps(skb) = *hwtstamps; 4354 else 4355 skb->tstamp = ktime_get_real(); 4356 4357 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats); 4358 } 4359 EXPORT_SYMBOL_GPL(__skb_tstamp_tx); 4360 4361 void skb_tstamp_tx(struct sk_buff *orig_skb, 4362 struct skb_shared_hwtstamps *hwtstamps) 4363 { 4364 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk, 4365 SCM_TSTAMP_SND); 4366 } 4367 EXPORT_SYMBOL_GPL(skb_tstamp_tx); 4368 4369 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked) 4370 { 4371 struct sock *sk = skb->sk; 4372 struct sock_exterr_skb *serr; 4373 int err = 1; 4374 4375 skb->wifi_acked_valid = 1; 4376 skb->wifi_acked = acked; 4377 4378 serr = SKB_EXT_ERR(skb); 4379 memset(serr, 0, sizeof(*serr)); 4380 serr->ee.ee_errno = ENOMSG; 4381 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS; 4382 4383 /* Take a reference to prevent skb_orphan() from freeing the socket, 4384 * but only if the socket refcount is not zero. 4385 */ 4386 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) { 4387 err = sock_queue_err_skb(sk, skb); 4388 sock_put(sk); 4389 } 4390 if (err) 4391 kfree_skb(skb); 4392 } 4393 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack); 4394 4395 /** 4396 * skb_partial_csum_set - set up and verify partial csum values for packet 4397 * @skb: the skb to set 4398 * @start: the number of bytes after skb->data to start checksumming. 4399 * @off: the offset from start to place the checksum. 4400 * 4401 * For untrusted partially-checksummed packets, we need to make sure the values 4402 * for skb->csum_start and skb->csum_offset are valid so we don't oops. 4403 * 4404 * This function checks and sets those values and skb->ip_summed: if this 4405 * returns false you should drop the packet. 4406 */ 4407 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 4408 { 4409 if (unlikely(start > skb_headlen(skb)) || 4410 unlikely((int)start + off > skb_headlen(skb) - 2)) { 4411 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n", 4412 start, off, skb_headlen(skb)); 4413 return false; 4414 } 4415 skb->ip_summed = CHECKSUM_PARTIAL; 4416 skb->csum_start = skb_headroom(skb) + start; 4417 skb->csum_offset = off; 4418 skb_set_transport_header(skb, start); 4419 return true; 4420 } 4421 EXPORT_SYMBOL_GPL(skb_partial_csum_set); 4422 4423 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len, 4424 unsigned int max) 4425 { 4426 if (skb_headlen(skb) >= len) 4427 return 0; 4428 4429 /* If we need to pullup then pullup to the max, so we 4430 * won't need to do it again. 4431 */ 4432 if (max > skb->len) 4433 max = skb->len; 4434 4435 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) 4436 return -ENOMEM; 4437 4438 if (skb_headlen(skb) < len) 4439 return -EPROTO; 4440 4441 return 0; 4442 } 4443 4444 #define MAX_TCP_HDR_LEN (15 * 4) 4445 4446 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb, 4447 typeof(IPPROTO_IP) proto, 4448 unsigned int off) 4449 { 4450 switch (proto) { 4451 int err; 4452 4453 case IPPROTO_TCP: 4454 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr), 4455 off + MAX_TCP_HDR_LEN); 4456 if (!err && !skb_partial_csum_set(skb, off, 4457 offsetof(struct tcphdr, 4458 check))) 4459 err = -EPROTO; 4460 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check; 4461 4462 case IPPROTO_UDP: 4463 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr), 4464 off + sizeof(struct udphdr)); 4465 if (!err && !skb_partial_csum_set(skb, off, 4466 offsetof(struct udphdr, 4467 check))) 4468 err = -EPROTO; 4469 return err ? ERR_PTR(err) : &udp_hdr(skb)->check; 4470 } 4471 4472 return ERR_PTR(-EPROTO); 4473 } 4474 4475 /* This value should be large enough to cover a tagged ethernet header plus 4476 * maximally sized IP and TCP or UDP headers. 4477 */ 4478 #define MAX_IP_HDR_LEN 128 4479 4480 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate) 4481 { 4482 unsigned int off; 4483 bool fragment; 4484 __sum16 *csum; 4485 int err; 4486 4487 fragment = false; 4488 4489 err = skb_maybe_pull_tail(skb, 4490 sizeof(struct iphdr), 4491 MAX_IP_HDR_LEN); 4492 if (err < 0) 4493 goto out; 4494 4495 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) 4496 fragment = true; 4497 4498 off = ip_hdrlen(skb); 4499 4500 err = -EPROTO; 4501 4502 if (fragment) 4503 goto out; 4504 4505 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off); 4506 if (IS_ERR(csum)) 4507 return PTR_ERR(csum); 4508 4509 if (recalculate) 4510 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 4511 ip_hdr(skb)->daddr, 4512 skb->len - off, 4513 ip_hdr(skb)->protocol, 0); 4514 err = 0; 4515 4516 out: 4517 return err; 4518 } 4519 4520 /* This value should be large enough to cover a tagged ethernet header plus 4521 * an IPv6 header, all options, and a maximal TCP or UDP header. 4522 */ 4523 #define MAX_IPV6_HDR_LEN 256 4524 4525 #define OPT_HDR(type, skb, off) \ 4526 (type *)(skb_network_header(skb) + (off)) 4527 4528 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate) 4529 { 4530 int err; 4531 u8 nexthdr; 4532 unsigned int off; 4533 unsigned int len; 4534 bool fragment; 4535 bool done; 4536 __sum16 *csum; 4537 4538 fragment = false; 4539 done = false; 4540 4541 off = sizeof(struct ipv6hdr); 4542 4543 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); 4544 if (err < 0) 4545 goto out; 4546 4547 nexthdr = ipv6_hdr(skb)->nexthdr; 4548 4549 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); 4550 while (off <= len && !done) { 4551 switch (nexthdr) { 4552 case IPPROTO_DSTOPTS: 4553 case IPPROTO_HOPOPTS: 4554 case IPPROTO_ROUTING: { 4555 struct ipv6_opt_hdr *hp; 4556 4557 err = skb_maybe_pull_tail(skb, 4558 off + 4559 sizeof(struct ipv6_opt_hdr), 4560 MAX_IPV6_HDR_LEN); 4561 if (err < 0) 4562 goto out; 4563 4564 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); 4565 nexthdr = hp->nexthdr; 4566 off += ipv6_optlen(hp); 4567 break; 4568 } 4569 case IPPROTO_AH: { 4570 struct ip_auth_hdr *hp; 4571 4572 err = skb_maybe_pull_tail(skb, 4573 off + 4574 sizeof(struct ip_auth_hdr), 4575 MAX_IPV6_HDR_LEN); 4576 if (err < 0) 4577 goto out; 4578 4579 hp = OPT_HDR(struct ip_auth_hdr, skb, off); 4580 nexthdr = hp->nexthdr; 4581 off += ipv6_authlen(hp); 4582 break; 4583 } 4584 case IPPROTO_FRAGMENT: { 4585 struct frag_hdr *hp; 4586 4587 err = skb_maybe_pull_tail(skb, 4588 off + 4589 sizeof(struct frag_hdr), 4590 MAX_IPV6_HDR_LEN); 4591 if (err < 0) 4592 goto out; 4593 4594 hp = OPT_HDR(struct frag_hdr, skb, off); 4595 4596 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) 4597 fragment = true; 4598 4599 nexthdr = hp->nexthdr; 4600 off += sizeof(struct frag_hdr); 4601 break; 4602 } 4603 default: 4604 done = true; 4605 break; 4606 } 4607 } 4608 4609 err = -EPROTO; 4610 4611 if (!done || fragment) 4612 goto out; 4613 4614 csum = skb_checksum_setup_ip(skb, nexthdr, off); 4615 if (IS_ERR(csum)) 4616 return PTR_ERR(csum); 4617 4618 if (recalculate) 4619 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 4620 &ipv6_hdr(skb)->daddr, 4621 skb->len - off, nexthdr, 0); 4622 err = 0; 4623 4624 out: 4625 return err; 4626 } 4627 4628 /** 4629 * skb_checksum_setup - set up partial checksum offset 4630 * @skb: the skb to set up 4631 * @recalculate: if true the pseudo-header checksum will be recalculated 4632 */ 4633 int skb_checksum_setup(struct sk_buff *skb, bool recalculate) 4634 { 4635 int err; 4636 4637 switch (skb->protocol) { 4638 case htons(ETH_P_IP): 4639 err = skb_checksum_setup_ipv4(skb, recalculate); 4640 break; 4641 4642 case htons(ETH_P_IPV6): 4643 err = skb_checksum_setup_ipv6(skb, recalculate); 4644 break; 4645 4646 default: 4647 err = -EPROTO; 4648 break; 4649 } 4650 4651 return err; 4652 } 4653 EXPORT_SYMBOL(skb_checksum_setup); 4654 4655 /** 4656 * skb_checksum_maybe_trim - maybe trims the given skb 4657 * @skb: the skb to check 4658 * @transport_len: the data length beyond the network header 4659 * 4660 * Checks whether the given skb has data beyond the given transport length. 4661 * If so, returns a cloned skb trimmed to this transport length. 4662 * Otherwise returns the provided skb. Returns NULL in error cases 4663 * (e.g. transport_len exceeds skb length or out-of-memory). 4664 * 4665 * Caller needs to set the skb transport header and free any returned skb if it 4666 * differs from the provided skb. 4667 */ 4668 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb, 4669 unsigned int transport_len) 4670 { 4671 struct sk_buff *skb_chk; 4672 unsigned int len = skb_transport_offset(skb) + transport_len; 4673 int ret; 4674 4675 if (skb->len < len) 4676 return NULL; 4677 else if (skb->len == len) 4678 return skb; 4679 4680 skb_chk = skb_clone(skb, GFP_ATOMIC); 4681 if (!skb_chk) 4682 return NULL; 4683 4684 ret = pskb_trim_rcsum(skb_chk, len); 4685 if (ret) { 4686 kfree_skb(skb_chk); 4687 return NULL; 4688 } 4689 4690 return skb_chk; 4691 } 4692 4693 /** 4694 * skb_checksum_trimmed - validate checksum of an skb 4695 * @skb: the skb to check 4696 * @transport_len: the data length beyond the network header 4697 * @skb_chkf: checksum function to use 4698 * 4699 * Applies the given checksum function skb_chkf to the provided skb. 4700 * Returns a checked and maybe trimmed skb. Returns NULL on error. 4701 * 4702 * If the skb has data beyond the given transport length, then a 4703 * trimmed & cloned skb is checked and returned. 4704 * 4705 * Caller needs to set the skb transport header and free any returned skb if it 4706 * differs from the provided skb. 4707 */ 4708 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4709 unsigned int transport_len, 4710 __sum16(*skb_chkf)(struct sk_buff *skb)) 4711 { 4712 struct sk_buff *skb_chk; 4713 unsigned int offset = skb_transport_offset(skb); 4714 __sum16 ret; 4715 4716 skb_chk = skb_checksum_maybe_trim(skb, transport_len); 4717 if (!skb_chk) 4718 goto err; 4719 4720 if (!pskb_may_pull(skb_chk, offset)) 4721 goto err; 4722 4723 skb_pull_rcsum(skb_chk, offset); 4724 ret = skb_chkf(skb_chk); 4725 skb_push_rcsum(skb_chk, offset); 4726 4727 if (ret) 4728 goto err; 4729 4730 return skb_chk; 4731 4732 err: 4733 if (skb_chk && skb_chk != skb) 4734 kfree_skb(skb_chk); 4735 4736 return NULL; 4737 4738 } 4739 EXPORT_SYMBOL(skb_checksum_trimmed); 4740 4741 void __skb_warn_lro_forwarding(const struct sk_buff *skb) 4742 { 4743 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n", 4744 skb->dev->name); 4745 } 4746 EXPORT_SYMBOL(__skb_warn_lro_forwarding); 4747 4748 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen) 4749 { 4750 if (head_stolen) { 4751 skb_release_head_state(skb); 4752 kmem_cache_free(skbuff_head_cache, skb); 4753 } else { 4754 __kfree_skb(skb); 4755 } 4756 } 4757 EXPORT_SYMBOL(kfree_skb_partial); 4758 4759 /** 4760 * skb_try_coalesce - try to merge skb to prior one 4761 * @to: prior buffer 4762 * @from: buffer to add 4763 * @fragstolen: pointer to boolean 4764 * @delta_truesize: how much more was allocated than was requested 4765 */ 4766 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 4767 bool *fragstolen, int *delta_truesize) 4768 { 4769 int i, delta, len = from->len; 4770 4771 *fragstolen = false; 4772 4773 if (skb_cloned(to)) 4774 return false; 4775 4776 if (len <= skb_tailroom(to)) { 4777 if (len) 4778 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len)); 4779 *delta_truesize = 0; 4780 return true; 4781 } 4782 4783 if (skb_has_frag_list(to) || skb_has_frag_list(from)) 4784 return false; 4785 if (skb_zcopy(to) || skb_zcopy(from)) 4786 return false; 4787 4788 if (skb_headlen(from) != 0) { 4789 struct page *page; 4790 unsigned int offset; 4791 4792 if (skb_shinfo(to)->nr_frags + 4793 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) 4794 return false; 4795 4796 if (skb_head_is_locked(from)) 4797 return false; 4798 4799 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff)); 4800 4801 page = virt_to_head_page(from->head); 4802 offset = from->data - (unsigned char *)page_address(page); 4803 4804 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags, 4805 page, offset, skb_headlen(from)); 4806 *fragstolen = true; 4807 } else { 4808 if (skb_shinfo(to)->nr_frags + 4809 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS) 4810 return false; 4811 4812 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from)); 4813 } 4814 4815 WARN_ON_ONCE(delta < len); 4816 4817 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags, 4818 skb_shinfo(from)->frags, 4819 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t)); 4820 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags; 4821 4822 if (!skb_cloned(from)) 4823 skb_shinfo(from)->nr_frags = 0; 4824 4825 /* if the skb is not cloned this does nothing 4826 * since we set nr_frags to 0. 4827 */ 4828 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) 4829 skb_frag_ref(from, i); 4830 4831 to->truesize += delta; 4832 to->len += len; 4833 to->data_len += len; 4834 4835 *delta_truesize = delta; 4836 return true; 4837 } 4838 EXPORT_SYMBOL(skb_try_coalesce); 4839 4840 /** 4841 * skb_scrub_packet - scrub an skb 4842 * 4843 * @skb: buffer to clean 4844 * @xnet: packet is crossing netns 4845 * 4846 * skb_scrub_packet can be used after encapsulating or decapsulting a packet 4847 * into/from a tunnel. Some information have to be cleared during these 4848 * operations. 4849 * skb_scrub_packet can also be used to clean a skb before injecting it in 4850 * another namespace (@xnet == true). We have to clear all information in the 4851 * skb that could impact namespace isolation. 4852 */ 4853 void skb_scrub_packet(struct sk_buff *skb, bool xnet) 4854 { 4855 skb->tstamp = 0; 4856 skb->pkt_type = PACKET_HOST; 4857 skb->skb_iif = 0; 4858 skb->ignore_df = 0; 4859 skb_dst_drop(skb); 4860 secpath_reset(skb); 4861 nf_reset(skb); 4862 nf_reset_trace(skb); 4863 4864 if (!xnet) 4865 return; 4866 4867 skb_orphan(skb); 4868 skb->mark = 0; 4869 } 4870 EXPORT_SYMBOL_GPL(skb_scrub_packet); 4871 4872 /** 4873 * skb_gso_transport_seglen - Return length of individual segments of a gso packet 4874 * 4875 * @skb: GSO skb 4876 * 4877 * skb_gso_transport_seglen is used to determine the real size of the 4878 * individual segments, including Layer4 headers (TCP/UDP). 4879 * 4880 * The MAC/L2 or network (IP, IPv6) headers are not accounted for. 4881 */ 4882 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb) 4883 { 4884 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4885 unsigned int thlen = 0; 4886 4887 if (skb->encapsulation) { 4888 thlen = skb_inner_transport_header(skb) - 4889 skb_transport_header(skb); 4890 4891 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) 4892 thlen += inner_tcp_hdrlen(skb); 4893 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 4894 thlen = tcp_hdrlen(skb); 4895 } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) { 4896 thlen = sizeof(struct sctphdr); 4897 } 4898 /* UFO sets gso_size to the size of the fragmentation 4899 * payload, i.e. the size of the L4 (UDP) header is already 4900 * accounted for. 4901 */ 4902 return thlen + shinfo->gso_size; 4903 } 4904 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen); 4905 4906 /** 4907 * skb_gso_validate_mtu - Return in case such skb fits a given MTU 4908 * 4909 * @skb: GSO skb 4910 * @mtu: MTU to validate against 4911 * 4912 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU 4913 * once split. 4914 */ 4915 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu) 4916 { 4917 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4918 const struct sk_buff *iter; 4919 unsigned int hlen; 4920 4921 hlen = skb_gso_network_seglen(skb); 4922 4923 if (shinfo->gso_size != GSO_BY_FRAGS) 4924 return hlen <= mtu; 4925 4926 /* Undo this so we can re-use header sizes */ 4927 hlen -= GSO_BY_FRAGS; 4928 4929 skb_walk_frags(skb, iter) { 4930 if (hlen + skb_headlen(iter) > mtu) 4931 return false; 4932 } 4933 4934 return true; 4935 } 4936 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu); 4937 4938 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb) 4939 { 4940 if (skb_cow(skb, skb_headroom(skb)) < 0) { 4941 kfree_skb(skb); 4942 return NULL; 4943 } 4944 4945 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN, 4946 2 * ETH_ALEN); 4947 skb->mac_header += VLAN_HLEN; 4948 return skb; 4949 } 4950 4951 struct sk_buff *skb_vlan_untag(struct sk_buff *skb) 4952 { 4953 struct vlan_hdr *vhdr; 4954 u16 vlan_tci; 4955 4956 if (unlikely(skb_vlan_tag_present(skb))) { 4957 /* vlan_tci is already set-up so leave this for another time */ 4958 return skb; 4959 } 4960 4961 skb = skb_share_check(skb, GFP_ATOMIC); 4962 if (unlikely(!skb)) 4963 goto err_free; 4964 4965 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN))) 4966 goto err_free; 4967 4968 vhdr = (struct vlan_hdr *)skb->data; 4969 vlan_tci = ntohs(vhdr->h_vlan_TCI); 4970 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci); 4971 4972 skb_pull_rcsum(skb, VLAN_HLEN); 4973 vlan_set_encap_proto(skb, vhdr); 4974 4975 skb = skb_reorder_vlan_header(skb); 4976 if (unlikely(!skb)) 4977 goto err_free; 4978 4979 skb_reset_network_header(skb); 4980 skb_reset_transport_header(skb); 4981 skb_reset_mac_len(skb); 4982 4983 return skb; 4984 4985 err_free: 4986 kfree_skb(skb); 4987 return NULL; 4988 } 4989 EXPORT_SYMBOL(skb_vlan_untag); 4990 4991 int skb_ensure_writable(struct sk_buff *skb, int write_len) 4992 { 4993 if (!pskb_may_pull(skb, write_len)) 4994 return -ENOMEM; 4995 4996 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) 4997 return 0; 4998 4999 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 5000 } 5001 EXPORT_SYMBOL(skb_ensure_writable); 5002 5003 /* remove VLAN header from packet and update csum accordingly. 5004 * expects a non skb_vlan_tag_present skb with a vlan tag payload 5005 */ 5006 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci) 5007 { 5008 struct vlan_hdr *vhdr; 5009 int offset = skb->data - skb_mac_header(skb); 5010 int err; 5011 5012 if (WARN_ONCE(offset, 5013 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n", 5014 offset)) { 5015 return -EINVAL; 5016 } 5017 5018 err = skb_ensure_writable(skb, VLAN_ETH_HLEN); 5019 if (unlikely(err)) 5020 return err; 5021 5022 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 5023 5024 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN); 5025 *vlan_tci = ntohs(vhdr->h_vlan_TCI); 5026 5027 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); 5028 __skb_pull(skb, VLAN_HLEN); 5029 5030 vlan_set_encap_proto(skb, vhdr); 5031 skb->mac_header += VLAN_HLEN; 5032 5033 if (skb_network_offset(skb) < ETH_HLEN) 5034 skb_set_network_header(skb, ETH_HLEN); 5035 5036 skb_reset_mac_len(skb); 5037 5038 return err; 5039 } 5040 EXPORT_SYMBOL(__skb_vlan_pop); 5041 5042 /* Pop a vlan tag either from hwaccel or from payload. 5043 * Expects skb->data at mac header. 5044 */ 5045 int skb_vlan_pop(struct sk_buff *skb) 5046 { 5047 u16 vlan_tci; 5048 __be16 vlan_proto; 5049 int err; 5050 5051 if (likely(skb_vlan_tag_present(skb))) { 5052 skb->vlan_tci = 0; 5053 } else { 5054 if (unlikely(!eth_type_vlan(skb->protocol))) 5055 return 0; 5056 5057 err = __skb_vlan_pop(skb, &vlan_tci); 5058 if (err) 5059 return err; 5060 } 5061 /* move next vlan tag to hw accel tag */ 5062 if (likely(!eth_type_vlan(skb->protocol))) 5063 return 0; 5064 5065 vlan_proto = skb->protocol; 5066 err = __skb_vlan_pop(skb, &vlan_tci); 5067 if (unlikely(err)) 5068 return err; 5069 5070 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 5071 return 0; 5072 } 5073 EXPORT_SYMBOL(skb_vlan_pop); 5074 5075 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present). 5076 * Expects skb->data at mac header. 5077 */ 5078 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci) 5079 { 5080 if (skb_vlan_tag_present(skb)) { 5081 int offset = skb->data - skb_mac_header(skb); 5082 int err; 5083 5084 if (WARN_ONCE(offset, 5085 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n", 5086 offset)) { 5087 return -EINVAL; 5088 } 5089 5090 err = __vlan_insert_tag(skb, skb->vlan_proto, 5091 skb_vlan_tag_get(skb)); 5092 if (err) 5093 return err; 5094 5095 skb->protocol = skb->vlan_proto; 5096 skb->mac_len += VLAN_HLEN; 5097 5098 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN); 5099 } 5100 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 5101 return 0; 5102 } 5103 EXPORT_SYMBOL(skb_vlan_push); 5104 5105 /** 5106 * alloc_skb_with_frags - allocate skb with page frags 5107 * 5108 * @header_len: size of linear part 5109 * @data_len: needed length in frags 5110 * @max_page_order: max page order desired. 5111 * @errcode: pointer to error code if any 5112 * @gfp_mask: allocation mask 5113 * 5114 * This can be used to allocate a paged skb, given a maximal order for frags. 5115 */ 5116 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 5117 unsigned long data_len, 5118 int max_page_order, 5119 int *errcode, 5120 gfp_t gfp_mask) 5121 { 5122 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 5123 unsigned long chunk; 5124 struct sk_buff *skb; 5125 struct page *page; 5126 gfp_t gfp_head; 5127 int i; 5128 5129 *errcode = -EMSGSIZE; 5130 /* Note this test could be relaxed, if we succeed to allocate 5131 * high order pages... 5132 */ 5133 if (npages > MAX_SKB_FRAGS) 5134 return NULL; 5135 5136 gfp_head = gfp_mask; 5137 if (gfp_head & __GFP_DIRECT_RECLAIM) 5138 gfp_head |= __GFP_RETRY_MAYFAIL; 5139 5140 *errcode = -ENOBUFS; 5141 skb = alloc_skb(header_len, gfp_head); 5142 if (!skb) 5143 return NULL; 5144 5145 skb->truesize += npages << PAGE_SHIFT; 5146 5147 for (i = 0; npages > 0; i++) { 5148 int order = max_page_order; 5149 5150 while (order) { 5151 if (npages >= 1 << order) { 5152 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) | 5153 __GFP_COMP | 5154 __GFP_NOWARN | 5155 __GFP_NORETRY, 5156 order); 5157 if (page) 5158 goto fill_page; 5159 /* Do not retry other high order allocations */ 5160 order = 1; 5161 max_page_order = 0; 5162 } 5163 order--; 5164 } 5165 page = alloc_page(gfp_mask); 5166 if (!page) 5167 goto failure; 5168 fill_page: 5169 chunk = min_t(unsigned long, data_len, 5170 PAGE_SIZE << order); 5171 skb_fill_page_desc(skb, i, page, 0, chunk); 5172 data_len -= chunk; 5173 npages -= 1 << order; 5174 } 5175 return skb; 5176 5177 failure: 5178 kfree_skb(skb); 5179 return NULL; 5180 } 5181 EXPORT_SYMBOL(alloc_skb_with_frags); 5182 5183 /* carve out the first off bytes from skb when off < headlen */ 5184 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off, 5185 const int headlen, gfp_t gfp_mask) 5186 { 5187 int i; 5188 int size = skb_end_offset(skb); 5189 int new_hlen = headlen - off; 5190 u8 *data; 5191 5192 size = SKB_DATA_ALIGN(size); 5193 5194 if (skb_pfmemalloc(skb)) 5195 gfp_mask |= __GFP_MEMALLOC; 5196 data = kmalloc_reserve(size + 5197 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 5198 gfp_mask, NUMA_NO_NODE, NULL); 5199 if (!data) 5200 return -ENOMEM; 5201 5202 size = SKB_WITH_OVERHEAD(ksize(data)); 5203 5204 /* Copy real data, and all frags */ 5205 skb_copy_from_linear_data_offset(skb, off, data, new_hlen); 5206 skb->len -= off; 5207 5208 memcpy((struct skb_shared_info *)(data + size), 5209 skb_shinfo(skb), 5210 offsetof(struct skb_shared_info, 5211 frags[skb_shinfo(skb)->nr_frags])); 5212 if (skb_cloned(skb)) { 5213 /* drop the old head gracefully */ 5214 if (skb_orphan_frags(skb, gfp_mask)) { 5215 kfree(data); 5216 return -ENOMEM; 5217 } 5218 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 5219 skb_frag_ref(skb, i); 5220 if (skb_has_frag_list(skb)) 5221 skb_clone_fraglist(skb); 5222 skb_release_data(skb); 5223 } else { 5224 /* we can reuse existing recount- all we did was 5225 * relocate values 5226 */ 5227 skb_free_head(skb); 5228 } 5229 5230 skb->head = data; 5231 skb->data = data; 5232 skb->head_frag = 0; 5233 #ifdef NET_SKBUFF_DATA_USES_OFFSET 5234 skb->end = size; 5235 #else 5236 skb->end = skb->head + size; 5237 #endif 5238 skb_set_tail_pointer(skb, skb_headlen(skb)); 5239 skb_headers_offset_update(skb, 0); 5240 skb->cloned = 0; 5241 skb->hdr_len = 0; 5242 skb->nohdr = 0; 5243 atomic_set(&skb_shinfo(skb)->dataref, 1); 5244 5245 return 0; 5246 } 5247 5248 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp); 5249 5250 /* carve out the first eat bytes from skb's frag_list. May recurse into 5251 * pskb_carve() 5252 */ 5253 static int pskb_carve_frag_list(struct sk_buff *skb, 5254 struct skb_shared_info *shinfo, int eat, 5255 gfp_t gfp_mask) 5256 { 5257 struct sk_buff *list = shinfo->frag_list; 5258 struct sk_buff *clone = NULL; 5259 struct sk_buff *insp = NULL; 5260 5261 do { 5262 if (!list) { 5263 pr_err("Not enough bytes to eat. Want %d\n", eat); 5264 return -EFAULT; 5265 } 5266 if (list->len <= eat) { 5267 /* Eaten as whole. */ 5268 eat -= list->len; 5269 list = list->next; 5270 insp = list; 5271 } else { 5272 /* Eaten partially. */ 5273 if (skb_shared(list)) { 5274 clone = skb_clone(list, gfp_mask); 5275 if (!clone) 5276 return -ENOMEM; 5277 insp = list->next; 5278 list = clone; 5279 } else { 5280 /* This may be pulled without problems. */ 5281 insp = list; 5282 } 5283 if (pskb_carve(list, eat, gfp_mask) < 0) { 5284 kfree_skb(clone); 5285 return -ENOMEM; 5286 } 5287 break; 5288 } 5289 } while (eat); 5290 5291 /* Free pulled out fragments. */ 5292 while ((list = shinfo->frag_list) != insp) { 5293 shinfo->frag_list = list->next; 5294 kfree_skb(list); 5295 } 5296 /* And insert new clone at head. */ 5297 if (clone) { 5298 clone->next = list; 5299 shinfo->frag_list = clone; 5300 } 5301 return 0; 5302 } 5303 5304 /* carve off first len bytes from skb. Split line (off) is in the 5305 * non-linear part of skb 5306 */ 5307 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off, 5308 int pos, gfp_t gfp_mask) 5309 { 5310 int i, k = 0; 5311 int size = skb_end_offset(skb); 5312 u8 *data; 5313 const int nfrags = skb_shinfo(skb)->nr_frags; 5314 struct skb_shared_info *shinfo; 5315 5316 size = SKB_DATA_ALIGN(size); 5317 5318 if (skb_pfmemalloc(skb)) 5319 gfp_mask |= __GFP_MEMALLOC; 5320 data = kmalloc_reserve(size + 5321 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)), 5322 gfp_mask, NUMA_NO_NODE, NULL); 5323 if (!data) 5324 return -ENOMEM; 5325 5326 size = SKB_WITH_OVERHEAD(ksize(data)); 5327 5328 memcpy((struct skb_shared_info *)(data + size), 5329 skb_shinfo(skb), offsetof(struct skb_shared_info, 5330 frags[skb_shinfo(skb)->nr_frags])); 5331 if (skb_orphan_frags(skb, gfp_mask)) { 5332 kfree(data); 5333 return -ENOMEM; 5334 } 5335 shinfo = (struct skb_shared_info *)(data + size); 5336 for (i = 0; i < nfrags; i++) { 5337 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]); 5338 5339 if (pos + fsize > off) { 5340 shinfo->frags[k] = skb_shinfo(skb)->frags[i]; 5341 5342 if (pos < off) { 5343 /* Split frag. 5344 * We have two variants in this case: 5345 * 1. Move all the frag to the second 5346 * part, if it is possible. F.e. 5347 * this approach is mandatory for TUX, 5348 * where splitting is expensive. 5349 * 2. Split is accurately. We make this. 5350 */ 5351 shinfo->frags[0].page_offset += off - pos; 5352 skb_frag_size_sub(&shinfo->frags[0], off - pos); 5353 } 5354 skb_frag_ref(skb, i); 5355 k++; 5356 } 5357 pos += fsize; 5358 } 5359 shinfo->nr_frags = k; 5360 if (skb_has_frag_list(skb)) 5361 skb_clone_fraglist(skb); 5362 5363 if (k == 0) { 5364 /* split line is in frag list */ 5365 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask); 5366 } 5367 skb_release_data(skb); 5368 5369 skb->head = data; 5370 skb->head_frag = 0; 5371 skb->data = data; 5372 #ifdef NET_SKBUFF_DATA_USES_OFFSET 5373 skb->end = size; 5374 #else 5375 skb->end = skb->head + size; 5376 #endif 5377 skb_reset_tail_pointer(skb); 5378 skb_headers_offset_update(skb, 0); 5379 skb->cloned = 0; 5380 skb->hdr_len = 0; 5381 skb->nohdr = 0; 5382 skb->len -= off; 5383 skb->data_len = skb->len; 5384 atomic_set(&skb_shinfo(skb)->dataref, 1); 5385 return 0; 5386 } 5387 5388 /* remove len bytes from the beginning of the skb */ 5389 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp) 5390 { 5391 int headlen = skb_headlen(skb); 5392 5393 if (len < headlen) 5394 return pskb_carve_inside_header(skb, len, headlen, gfp); 5395 else 5396 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp); 5397 } 5398 5399 /* Extract to_copy bytes starting at off from skb, and return this in 5400 * a new skb 5401 */ 5402 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, 5403 int to_copy, gfp_t gfp) 5404 { 5405 struct sk_buff *clone = skb_clone(skb, gfp); 5406 5407 if (!clone) 5408 return NULL; 5409 5410 if (pskb_carve(clone, off, gfp) < 0 || 5411 pskb_trim(clone, to_copy)) { 5412 kfree_skb(clone); 5413 return NULL; 5414 } 5415 return clone; 5416 } 5417 EXPORT_SYMBOL(pskb_extract); 5418 5419 /** 5420 * skb_condense - try to get rid of fragments/frag_list if possible 5421 * @skb: buffer 5422 * 5423 * Can be used to save memory before skb is added to a busy queue. 5424 * If packet has bytes in frags and enough tail room in skb->head, 5425 * pull all of them, so that we can free the frags right now and adjust 5426 * truesize. 5427 * Notes: 5428 * We do not reallocate skb->head thus can not fail. 5429 * Caller must re-evaluate skb->truesize if needed. 5430 */ 5431 void skb_condense(struct sk_buff *skb) 5432 { 5433 if (skb->data_len) { 5434 if (skb->data_len > skb->end - skb->tail || 5435 skb_cloned(skb)) 5436 return; 5437 5438 /* Nice, we can free page frag(s) right now */ 5439 __pskb_pull_tail(skb, skb->data_len); 5440 } 5441 /* At this point, skb->truesize might be over estimated, 5442 * because skb had a fragment, and fragments do not tell 5443 * their truesize. 5444 * When we pulled its content into skb->head, fragment 5445 * was freed, but __pskb_pull_tail() could not possibly 5446 * adjust skb->truesize, not knowing the frag truesize. 5447 */ 5448 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 5449 } 5450