]> git.ipfire.org Git - thirdparty/kernel/linux.git/blob - include/linux/skbuff.h
bpf: only adjust gso_size on bytestream protocols
[thirdparty/kernel/linux.git] / include / linux / skbuff.h
1 /*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
16
17 #include <linux/kernel.h>
18 #include <linux/compiler.h>
19 #include <linux/time.h>
20 #include <linux/bug.h>
21 #include <linux/cache.h>
22 #include <linux/rbtree.h>
23 #include <linux/socket.h>
24 #include <linux/refcount.h>
25
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <linux/sched/clock.h>
38 #include <net/flow_dissector.h>
39 #include <linux/splice.h>
40 #include <linux/in6.h>
41 #include <linux/if_packet.h>
42 #include <net/flow.h>
43
44 /* The interface for checksum offload between the stack and networking drivers
45 * is as follows...
46 *
47 * A. IP checksum related features
48 *
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
52 * to its device.
53 *
54 * The checksum related features are:
55 *
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
61 *
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
70 *
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
80 *
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
86 *
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
89 *
90 * CHECKSUM_NONE:
91 *
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
95 *
96 * CHECKSUM_UNNECESSARY:
97 *
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
104 *
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 * FCOE: indicates the CRC in FC frame has been validated.
113 *
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
123 *
124 * CHECKSUM_COMPLETE:
125 *
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
129 *
130 * Notes:
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
134 *
135 * CHECKSUM_PARTIAL:
136 *
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
145 * be verified.
146 *
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
149 *
150 * CHECKSUM_PARTIAL:
151 *
152 * The driver is required to checksum the packet as seen by hard_start_xmit()
153 * from skb->csum_start up to the end, and to record/write the checksum at
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
159 * correctly.
160 *
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
165 *
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
173 *
174 * CHECKSUM_NONE:
175 *
176 * The skb was already checksummed by the protocol, or a checksum is not
177 * required.
178 *
179 * CHECKSUM_UNNECESSARY:
180 *
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
182 * output.
183 *
184 * CHECKSUM_COMPLETE:
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
187 *
188 * D. Non-IP checksum (CRC) offloads
189 *
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
199 *
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
207 *
208 * E. Checksumming on output with GSO.
209 *
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
217 */
218
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
224
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
227
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
235
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
240
241 struct net_device;
242 struct scatterlist;
243 struct pipe_inode_info;
244 struct iov_iter;
245 struct napi_struct;
246 struct bpf_prog;
247 union bpf_attr;
248 struct skb_ext;
249
250 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
251 struct nf_conntrack {
252 atomic_t use;
253 };
254 #endif
255
256 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
257 struct nf_bridge_info {
258 enum {
259 BRNF_PROTO_UNCHANGED,
260 BRNF_PROTO_8021Q,
261 BRNF_PROTO_PPPOE
262 } orig_proto:8;
263 u8 pkt_otherhost:1;
264 u8 in_prerouting:1;
265 u8 bridged_dnat:1;
266 __u16 frag_max_size;
267 struct net_device *physindev;
268
269 /* always valid & non-NULL from FORWARD on, for physdev match */
270 struct net_device *physoutdev;
271 union {
272 /* prerouting: detect dnat in orig/reply direction */
273 __be32 ipv4_daddr;
274 struct in6_addr ipv6_daddr;
275
276 /* after prerouting + nat detected: store original source
277 * mac since neigh resolution overwrites it, only used while
278 * skb is out in neigh layer.
279 */
280 char neigh_header[8];
281 };
282 };
283 #endif
284
285 struct sk_buff_head {
286 /* These two members must be first. */
287 struct sk_buff *next;
288 struct sk_buff *prev;
289
290 __u32 qlen;
291 spinlock_t lock;
292 };
293
294 struct sk_buff;
295
296 /* To allow 64K frame to be packed as single skb without frag_list we
297 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
298 * buffers which do not start on a page boundary.
299 *
300 * Since GRO uses frags we allocate at least 16 regardless of page
301 * size.
302 */
303 #if (65536/PAGE_SIZE + 1) < 16
304 #define MAX_SKB_FRAGS 16UL
305 #else
306 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
307 #endif
308 extern int sysctl_max_skb_frags;
309
310 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
311 * segment using its current segmentation instead.
312 */
313 #define GSO_BY_FRAGS 0xFFFF
314
315 typedef struct skb_frag_struct skb_frag_t;
316
317 struct skb_frag_struct {
318 struct {
319 struct page *p;
320 } page;
321 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
322 __u32 page_offset;
323 __u32 size;
324 #else
325 __u16 page_offset;
326 __u16 size;
327 #endif
328 };
329
330 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
331 {
332 return frag->size;
333 }
334
335 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
336 {
337 frag->size = size;
338 }
339
340 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
341 {
342 frag->size += delta;
343 }
344
345 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
346 {
347 frag->size -= delta;
348 }
349
350 static inline bool skb_frag_must_loop(struct page *p)
351 {
352 #if defined(CONFIG_HIGHMEM)
353 if (PageHighMem(p))
354 return true;
355 #endif
356 return false;
357 }
358
359 /**
360 * skb_frag_foreach_page - loop over pages in a fragment
361 *
362 * @f: skb frag to operate on
363 * @f_off: offset from start of f->page.p
364 * @f_len: length from f_off to loop over
365 * @p: (temp var) current page
366 * @p_off: (temp var) offset from start of current page,
367 * non-zero only on first page.
368 * @p_len: (temp var) length in current page,
369 * < PAGE_SIZE only on first and last page.
370 * @copied: (temp var) length so far, excluding current p_len.
371 *
372 * A fragment can hold a compound page, in which case per-page
373 * operations, notably kmap_atomic, must be called for each
374 * regular page.
375 */
376 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
377 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
378 p_off = (f_off) & (PAGE_SIZE - 1), \
379 p_len = skb_frag_must_loop(p) ? \
380 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
381 copied = 0; \
382 copied < f_len; \
383 copied += p_len, p++, p_off = 0, \
384 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
385
386 #define HAVE_HW_TIME_STAMP
387
388 /**
389 * struct skb_shared_hwtstamps - hardware time stamps
390 * @hwtstamp: hardware time stamp transformed into duration
391 * since arbitrary point in time
392 *
393 * Software time stamps generated by ktime_get_real() are stored in
394 * skb->tstamp.
395 *
396 * hwtstamps can only be compared against other hwtstamps from
397 * the same device.
398 *
399 * This structure is attached to packets as part of the
400 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
401 */
402 struct skb_shared_hwtstamps {
403 ktime_t hwtstamp;
404 };
405
406 /* Definitions for tx_flags in struct skb_shared_info */
407 enum {
408 /* generate hardware time stamp */
409 SKBTX_HW_TSTAMP = 1 << 0,
410
411 /* generate software time stamp when queueing packet to NIC */
412 SKBTX_SW_TSTAMP = 1 << 1,
413
414 /* device driver is going to provide hardware time stamp */
415 SKBTX_IN_PROGRESS = 1 << 2,
416
417 /* device driver supports TX zero-copy buffers */
418 SKBTX_DEV_ZEROCOPY = 1 << 3,
419
420 /* generate wifi status information (where possible) */
421 SKBTX_WIFI_STATUS = 1 << 4,
422
423 /* This indicates at least one fragment might be overwritten
424 * (as in vmsplice(), sendfile() ...)
425 * If we need to compute a TX checksum, we'll need to copy
426 * all frags to avoid possible bad checksum
427 */
428 SKBTX_SHARED_FRAG = 1 << 5,
429
430 /* generate software time stamp when entering packet scheduling */
431 SKBTX_SCHED_TSTAMP = 1 << 6,
432 };
433
434 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
435 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
436 SKBTX_SCHED_TSTAMP)
437 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
438
439 /*
440 * The callback notifies userspace to release buffers when skb DMA is done in
441 * lower device, the skb last reference should be 0 when calling this.
442 * The zerocopy_success argument is true if zero copy transmit occurred,
443 * false on data copy or out of memory error caused by data copy attempt.
444 * The ctx field is used to track device context.
445 * The desc field is used to track userspace buffer index.
446 */
447 struct ubuf_info {
448 void (*callback)(struct ubuf_info *, bool zerocopy_success);
449 union {
450 struct {
451 unsigned long desc;
452 void *ctx;
453 };
454 struct {
455 u32 id;
456 u16 len;
457 u16 zerocopy:1;
458 u32 bytelen;
459 };
460 };
461 refcount_t refcnt;
462
463 struct mmpin {
464 struct user_struct *user;
465 unsigned int num_pg;
466 } mmp;
467 };
468
469 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
470
471 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
472 void mm_unaccount_pinned_pages(struct mmpin *mmp);
473
474 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
475 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
476 struct ubuf_info *uarg);
477
478 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
479 {
480 refcount_inc(&uarg->refcnt);
481 }
482
483 void sock_zerocopy_put(struct ubuf_info *uarg);
484 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
485
486 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
487
488 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
489 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
490 struct msghdr *msg, int len,
491 struct ubuf_info *uarg);
492
493 /* This data is invariant across clones and lives at
494 * the end of the header data, ie. at skb->end.
495 */
496 struct skb_shared_info {
497 __u8 __unused;
498 __u8 meta_len;
499 __u8 nr_frags;
500 __u8 tx_flags;
501 unsigned short gso_size;
502 /* Warning: this field is not always filled in (UFO)! */
503 unsigned short gso_segs;
504 struct sk_buff *frag_list;
505 struct skb_shared_hwtstamps hwtstamps;
506 unsigned int gso_type;
507 u32 tskey;
508
509 /*
510 * Warning : all fields before dataref are cleared in __alloc_skb()
511 */
512 atomic_t dataref;
513
514 /* Intermediate layers must ensure that destructor_arg
515 * remains valid until skb destructor */
516 void * destructor_arg;
517
518 /* must be last field, see pskb_expand_head() */
519 skb_frag_t frags[MAX_SKB_FRAGS];
520 };
521
522 /* We divide dataref into two halves. The higher 16 bits hold references
523 * to the payload part of skb->data. The lower 16 bits hold references to
524 * the entire skb->data. A clone of a headerless skb holds the length of
525 * the header in skb->hdr_len.
526 *
527 * All users must obey the rule that the skb->data reference count must be
528 * greater than or equal to the payload reference count.
529 *
530 * Holding a reference to the payload part means that the user does not
531 * care about modifications to the header part of skb->data.
532 */
533 #define SKB_DATAREF_SHIFT 16
534 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
535
536
537 enum {
538 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
539 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
540 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
541 };
542
543 enum {
544 SKB_GSO_TCPV4 = 1 << 0,
545
546 /* This indicates the skb is from an untrusted source. */
547 SKB_GSO_DODGY = 1 << 1,
548
549 /* This indicates the tcp segment has CWR set. */
550 SKB_GSO_TCP_ECN = 1 << 2,
551
552 SKB_GSO_TCP_FIXEDID = 1 << 3,
553
554 SKB_GSO_TCPV6 = 1 << 4,
555
556 SKB_GSO_FCOE = 1 << 5,
557
558 SKB_GSO_GRE = 1 << 6,
559
560 SKB_GSO_GRE_CSUM = 1 << 7,
561
562 SKB_GSO_IPXIP4 = 1 << 8,
563
564 SKB_GSO_IPXIP6 = 1 << 9,
565
566 SKB_GSO_UDP_TUNNEL = 1 << 10,
567
568 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
569
570 SKB_GSO_PARTIAL = 1 << 12,
571
572 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
573
574 SKB_GSO_SCTP = 1 << 14,
575
576 SKB_GSO_ESP = 1 << 15,
577
578 SKB_GSO_UDP = 1 << 16,
579
580 SKB_GSO_UDP_L4 = 1 << 17,
581 };
582
583 #if BITS_PER_LONG > 32
584 #define NET_SKBUFF_DATA_USES_OFFSET 1
585 #endif
586
587 #ifdef NET_SKBUFF_DATA_USES_OFFSET
588 typedef unsigned int sk_buff_data_t;
589 #else
590 typedef unsigned char *sk_buff_data_t;
591 #endif
592
593 /**
594 * struct sk_buff - socket buffer
595 * @next: Next buffer in list
596 * @prev: Previous buffer in list
597 * @tstamp: Time we arrived/left
598 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
599 * @sk: Socket we are owned by
600 * @dev: Device we arrived on/are leaving by
601 * @cb: Control buffer. Free for use by every layer. Put private vars here
602 * @_skb_refdst: destination entry (with norefcount bit)
603 * @sp: the security path, used for xfrm
604 * @len: Length of actual data
605 * @data_len: Data length
606 * @mac_len: Length of link layer header
607 * @hdr_len: writable header length of cloned skb
608 * @csum: Checksum (must include start/offset pair)
609 * @csum_start: Offset from skb->head where checksumming should start
610 * @csum_offset: Offset from csum_start where checksum should be stored
611 * @priority: Packet queueing priority
612 * @ignore_df: allow local fragmentation
613 * @cloned: Head may be cloned (check refcnt to be sure)
614 * @ip_summed: Driver fed us an IP checksum
615 * @nohdr: Payload reference only, must not modify header
616 * @pkt_type: Packet class
617 * @fclone: skbuff clone status
618 * @ipvs_property: skbuff is owned by ipvs
619 * @offload_fwd_mark: Packet was L2-forwarded in hardware
620 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
621 * @tc_skip_classify: do not classify packet. set by IFB device
622 * @tc_at_ingress: used within tc_classify to distinguish in/egress
623 * @tc_redirected: packet was redirected by a tc action
624 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
625 * @peeked: this packet has been seen already, so stats have been
626 * done for it, don't do them again
627 * @nf_trace: netfilter packet trace flag
628 * @protocol: Packet protocol from driver
629 * @destructor: Destruct function
630 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
631 * @_nfct: Associated connection, if any (with nfctinfo bits)
632 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
633 * @skb_iif: ifindex of device we arrived on
634 * @tc_index: Traffic control index
635 * @hash: the packet hash
636 * @queue_mapping: Queue mapping for multiqueue devices
637 * @xmit_more: More SKBs are pending for this queue
638 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
639 * @active_extensions: active extensions (skb_ext_id types)
640 * @ndisc_nodetype: router type (from link layer)
641 * @ooo_okay: allow the mapping of a socket to a queue to be changed
642 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
643 * ports.
644 * @sw_hash: indicates hash was computed in software stack
645 * @wifi_acked_valid: wifi_acked was set
646 * @wifi_acked: whether frame was acked on wifi or not
647 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
648 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
649 * @dst_pending_confirm: need to confirm neighbour
650 * @decrypted: Decrypted SKB
651 * @napi_id: id of the NAPI struct this skb came from
652 * @secmark: security marking
653 * @mark: Generic packet mark
654 * @vlan_proto: vlan encapsulation protocol
655 * @vlan_tci: vlan tag control information
656 * @inner_protocol: Protocol (encapsulation)
657 * @inner_transport_header: Inner transport layer header (encapsulation)
658 * @inner_network_header: Network layer header (encapsulation)
659 * @inner_mac_header: Link layer header (encapsulation)
660 * @transport_header: Transport layer header
661 * @network_header: Network layer header
662 * @mac_header: Link layer header
663 * @tail: Tail pointer
664 * @end: End pointer
665 * @head: Head of buffer
666 * @data: Data head pointer
667 * @truesize: Buffer size
668 * @users: User count - see {datagram,tcp}.c
669 * @extensions: allocated extensions, valid if active_extensions is nonzero
670 */
671
672 struct sk_buff {
673 union {
674 struct {
675 /* These two members must be first. */
676 struct sk_buff *next;
677 struct sk_buff *prev;
678
679 union {
680 struct net_device *dev;
681 /* Some protocols might use this space to store information,
682 * while device pointer would be NULL.
683 * UDP receive path is one user.
684 */
685 unsigned long dev_scratch;
686 };
687 };
688 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
689 struct list_head list;
690 };
691
692 union {
693 struct sock *sk;
694 int ip_defrag_offset;
695 };
696
697 union {
698 ktime_t tstamp;
699 u64 skb_mstamp_ns; /* earliest departure time */
700 };
701 /*
702 * This is the control buffer. It is free to use for every
703 * layer. Please put your private variables there. If you
704 * want to keep them across layers you have to do a skb_clone()
705 * first. This is owned by whoever has the skb queued ATM.
706 */
707 char cb[48] __aligned(8);
708
709 union {
710 struct {
711 unsigned long _skb_refdst;
712 void (*destructor)(struct sk_buff *skb);
713 };
714 struct list_head tcp_tsorted_anchor;
715 };
716
717 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
718 unsigned long _nfct;
719 #endif
720 unsigned int len,
721 data_len;
722 __u16 mac_len,
723 hdr_len;
724
725 /* Following fields are _not_ copied in __copy_skb_header()
726 * Note that queue_mapping is here mostly to fill a hole.
727 */
728 __u16 queue_mapping;
729
730 /* if you move cloned around you also must adapt those constants */
731 #ifdef __BIG_ENDIAN_BITFIELD
732 #define CLONED_MASK (1 << 7)
733 #else
734 #define CLONED_MASK 1
735 #endif
736 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
737
738 __u8 __cloned_offset[0];
739 __u8 cloned:1,
740 nohdr:1,
741 fclone:2,
742 peeked:1,
743 head_frag:1,
744 xmit_more:1,
745 pfmemalloc:1;
746 #ifdef CONFIG_SKB_EXTENSIONS
747 __u8 active_extensions;
748 #endif
749 /* fields enclosed in headers_start/headers_end are copied
750 * using a single memcpy() in __copy_skb_header()
751 */
752 /* private: */
753 __u32 headers_start[0];
754 /* public: */
755
756 /* if you move pkt_type around you also must adapt those constants */
757 #ifdef __BIG_ENDIAN_BITFIELD
758 #define PKT_TYPE_MAX (7 << 5)
759 #else
760 #define PKT_TYPE_MAX 7
761 #endif
762 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
763
764 __u8 __pkt_type_offset[0];
765 __u8 pkt_type:3;
766 __u8 ignore_df:1;
767 __u8 nf_trace:1;
768 __u8 ip_summed:2;
769 __u8 ooo_okay:1;
770
771 __u8 l4_hash:1;
772 __u8 sw_hash:1;
773 __u8 wifi_acked_valid:1;
774 __u8 wifi_acked:1;
775 __u8 no_fcs:1;
776 /* Indicates the inner headers are valid in the skbuff. */
777 __u8 encapsulation:1;
778 __u8 encap_hdr_csum:1;
779 __u8 csum_valid:1;
780
781 #ifdef __BIG_ENDIAN_BITFIELD
782 #define PKT_VLAN_PRESENT_BIT 7
783 #else
784 #define PKT_VLAN_PRESENT_BIT 0
785 #endif
786 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
787 __u8 __pkt_vlan_present_offset[0];
788 __u8 vlan_present:1;
789 __u8 csum_complete_sw:1;
790 __u8 csum_level:2;
791 __u8 csum_not_inet:1;
792 __u8 dst_pending_confirm:1;
793 #ifdef CONFIG_IPV6_NDISC_NODETYPE
794 __u8 ndisc_nodetype:2;
795 #endif
796
797 __u8 ipvs_property:1;
798 __u8 inner_protocol_type:1;
799 __u8 remcsum_offload:1;
800 #ifdef CONFIG_NET_SWITCHDEV
801 __u8 offload_fwd_mark:1;
802 __u8 offload_l3_fwd_mark:1;
803 #endif
804 #ifdef CONFIG_NET_CLS_ACT
805 __u8 tc_skip_classify:1;
806 __u8 tc_at_ingress:1;
807 __u8 tc_redirected:1;
808 __u8 tc_from_ingress:1;
809 #endif
810 #ifdef CONFIG_TLS_DEVICE
811 __u8 decrypted:1;
812 #endif
813
814 #ifdef CONFIG_NET_SCHED
815 __u16 tc_index; /* traffic control index */
816 #endif
817
818 union {
819 __wsum csum;
820 struct {
821 __u16 csum_start;
822 __u16 csum_offset;
823 };
824 };
825 __u32 priority;
826 int skb_iif;
827 __u32 hash;
828 __be16 vlan_proto;
829 __u16 vlan_tci;
830 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
831 union {
832 unsigned int napi_id;
833 unsigned int sender_cpu;
834 };
835 #endif
836 #ifdef CONFIG_NETWORK_SECMARK
837 __u32 secmark;
838 #endif
839
840 union {
841 __u32 mark;
842 __u32 reserved_tailroom;
843 };
844
845 union {
846 __be16 inner_protocol;
847 __u8 inner_ipproto;
848 };
849
850 __u16 inner_transport_header;
851 __u16 inner_network_header;
852 __u16 inner_mac_header;
853
854 __be16 protocol;
855 __u16 transport_header;
856 __u16 network_header;
857 __u16 mac_header;
858
859 /* private: */
860 __u32 headers_end[0];
861 /* public: */
862
863 /* These elements must be at the end, see alloc_skb() for details. */
864 sk_buff_data_t tail;
865 sk_buff_data_t end;
866 unsigned char *head,
867 *data;
868 unsigned int truesize;
869 refcount_t users;
870
871 #ifdef CONFIG_SKB_EXTENSIONS
872 /* only useable after checking ->active_extensions != 0 */
873 struct skb_ext *extensions;
874 #endif
875 };
876
877 #ifdef __KERNEL__
878 /*
879 * Handling routines are only of interest to the kernel
880 */
881
882 #define SKB_ALLOC_FCLONE 0x01
883 #define SKB_ALLOC_RX 0x02
884 #define SKB_ALLOC_NAPI 0x04
885
886 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
887 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
888 {
889 return unlikely(skb->pfmemalloc);
890 }
891
892 /*
893 * skb might have a dst pointer attached, refcounted or not.
894 * _skb_refdst low order bit is set if refcount was _not_ taken
895 */
896 #define SKB_DST_NOREF 1UL
897 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
898
899 #define SKB_NFCT_PTRMASK ~(7UL)
900 /**
901 * skb_dst - returns skb dst_entry
902 * @skb: buffer
903 *
904 * Returns skb dst_entry, regardless of reference taken or not.
905 */
906 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
907 {
908 /* If refdst was not refcounted, check we still are in a
909 * rcu_read_lock section
910 */
911 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
912 !rcu_read_lock_held() &&
913 !rcu_read_lock_bh_held());
914 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
915 }
916
917 /**
918 * skb_dst_set - sets skb dst
919 * @skb: buffer
920 * @dst: dst entry
921 *
922 * Sets skb dst, assuming a reference was taken on dst and should
923 * be released by skb_dst_drop()
924 */
925 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
926 {
927 skb->_skb_refdst = (unsigned long)dst;
928 }
929
930 /**
931 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
932 * @skb: buffer
933 * @dst: dst entry
934 *
935 * Sets skb dst, assuming a reference was not taken on dst.
936 * If dst entry is cached, we do not take reference and dst_release
937 * will be avoided by refdst_drop. If dst entry is not cached, we take
938 * reference, so that last dst_release can destroy the dst immediately.
939 */
940 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
941 {
942 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
943 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
944 }
945
946 /**
947 * skb_dst_is_noref - Test if skb dst isn't refcounted
948 * @skb: buffer
949 */
950 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
951 {
952 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
953 }
954
955 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
956 {
957 return (struct rtable *)skb_dst(skb);
958 }
959
960 /* For mangling skb->pkt_type from user space side from applications
961 * such as nft, tc, etc, we only allow a conservative subset of
962 * possible pkt_types to be set.
963 */
964 static inline bool skb_pkt_type_ok(u32 ptype)
965 {
966 return ptype <= PACKET_OTHERHOST;
967 }
968
969 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
970 {
971 #ifdef CONFIG_NET_RX_BUSY_POLL
972 return skb->napi_id;
973 #else
974 return 0;
975 #endif
976 }
977
978 /* decrement the reference count and return true if we can free the skb */
979 static inline bool skb_unref(struct sk_buff *skb)
980 {
981 if (unlikely(!skb))
982 return false;
983 if (likely(refcount_read(&skb->users) == 1))
984 smp_rmb();
985 else if (likely(!refcount_dec_and_test(&skb->users)))
986 return false;
987
988 return true;
989 }
990
991 void skb_release_head_state(struct sk_buff *skb);
992 void kfree_skb(struct sk_buff *skb);
993 void kfree_skb_list(struct sk_buff *segs);
994 void skb_tx_error(struct sk_buff *skb);
995 void consume_skb(struct sk_buff *skb);
996 void __consume_stateless_skb(struct sk_buff *skb);
997 void __kfree_skb(struct sk_buff *skb);
998 extern struct kmem_cache *skbuff_head_cache;
999
1000 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1001 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1002 bool *fragstolen, int *delta_truesize);
1003
1004 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1005 int node);
1006 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1007 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1008 static inline struct sk_buff *alloc_skb(unsigned int size,
1009 gfp_t priority)
1010 {
1011 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1012 }
1013
1014 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1015 unsigned long data_len,
1016 int max_page_order,
1017 int *errcode,
1018 gfp_t gfp_mask);
1019
1020 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1021 struct sk_buff_fclones {
1022 struct sk_buff skb1;
1023
1024 struct sk_buff skb2;
1025
1026 refcount_t fclone_ref;
1027 };
1028
1029 /**
1030 * skb_fclone_busy - check if fclone is busy
1031 * @sk: socket
1032 * @skb: buffer
1033 *
1034 * Returns true if skb is a fast clone, and its clone is not freed.
1035 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1036 * so we also check that this didnt happen.
1037 */
1038 static inline bool skb_fclone_busy(const struct sock *sk,
1039 const struct sk_buff *skb)
1040 {
1041 const struct sk_buff_fclones *fclones;
1042
1043 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1044
1045 return skb->fclone == SKB_FCLONE_ORIG &&
1046 refcount_read(&fclones->fclone_ref) > 1 &&
1047 fclones->skb2.sk == sk;
1048 }
1049
1050 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1051 gfp_t priority)
1052 {
1053 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1054 }
1055
1056 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1057 void skb_headers_offset_update(struct sk_buff *skb, int off);
1058 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1059 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1060 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1061 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1062 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1063 gfp_t gfp_mask, bool fclone);
1064 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1065 gfp_t gfp_mask)
1066 {
1067 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1068 }
1069
1070 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1071 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1072 unsigned int headroom);
1073 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1074 int newtailroom, gfp_t priority);
1075 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1076 int offset, int len);
1077 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1078 int offset, int len);
1079 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1080 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1081
1082 /**
1083 * skb_pad - zero pad the tail of an skb
1084 * @skb: buffer to pad
1085 * @pad: space to pad
1086 *
1087 * Ensure that a buffer is followed by a padding area that is zero
1088 * filled. Used by network drivers which may DMA or transfer data
1089 * beyond the buffer end onto the wire.
1090 *
1091 * May return error in out of memory cases. The skb is freed on error.
1092 */
1093 static inline int skb_pad(struct sk_buff *skb, int pad)
1094 {
1095 return __skb_pad(skb, pad, true);
1096 }
1097 #define dev_kfree_skb(a) consume_skb(a)
1098
1099 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1100 int offset, size_t size);
1101
1102 struct skb_seq_state {
1103 __u32 lower_offset;
1104 __u32 upper_offset;
1105 __u32 frag_idx;
1106 __u32 stepped_offset;
1107 struct sk_buff *root_skb;
1108 struct sk_buff *cur_skb;
1109 __u8 *frag_data;
1110 };
1111
1112 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1113 unsigned int to, struct skb_seq_state *st);
1114 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1115 struct skb_seq_state *st);
1116 void skb_abort_seq_read(struct skb_seq_state *st);
1117
1118 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1119 unsigned int to, struct ts_config *config);
1120
1121 /*
1122 * Packet hash types specify the type of hash in skb_set_hash.
1123 *
1124 * Hash types refer to the protocol layer addresses which are used to
1125 * construct a packet's hash. The hashes are used to differentiate or identify
1126 * flows of the protocol layer for the hash type. Hash types are either
1127 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1128 *
1129 * Properties of hashes:
1130 *
1131 * 1) Two packets in different flows have different hash values
1132 * 2) Two packets in the same flow should have the same hash value
1133 *
1134 * A hash at a higher layer is considered to be more specific. A driver should
1135 * set the most specific hash possible.
1136 *
1137 * A driver cannot indicate a more specific hash than the layer at which a hash
1138 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1139 *
1140 * A driver may indicate a hash level which is less specific than the
1141 * actual layer the hash was computed on. For instance, a hash computed
1142 * at L4 may be considered an L3 hash. This should only be done if the
1143 * driver can't unambiguously determine that the HW computed the hash at
1144 * the higher layer. Note that the "should" in the second property above
1145 * permits this.
1146 */
1147 enum pkt_hash_types {
1148 PKT_HASH_TYPE_NONE, /* Undefined type */
1149 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1150 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1151 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1152 };
1153
1154 static inline void skb_clear_hash(struct sk_buff *skb)
1155 {
1156 skb->hash = 0;
1157 skb->sw_hash = 0;
1158 skb->l4_hash = 0;
1159 }
1160
1161 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1162 {
1163 if (!skb->l4_hash)
1164 skb_clear_hash(skb);
1165 }
1166
1167 static inline void
1168 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1169 {
1170 skb->l4_hash = is_l4;
1171 skb->sw_hash = is_sw;
1172 skb->hash = hash;
1173 }
1174
1175 static inline void
1176 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1177 {
1178 /* Used by drivers to set hash from HW */
1179 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1180 }
1181
1182 static inline void
1183 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1184 {
1185 __skb_set_hash(skb, hash, true, is_l4);
1186 }
1187
1188 void __skb_get_hash(struct sk_buff *skb);
1189 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1190 u32 skb_get_poff(const struct sk_buff *skb);
1191 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1192 const struct flow_keys_basic *keys, int hlen);
1193 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1194 void *data, int hlen_proto);
1195
1196 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1197 int thoff, u8 ip_proto)
1198 {
1199 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1200 }
1201
1202 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1203 const struct flow_dissector_key *key,
1204 unsigned int key_count);
1205
1206 #ifdef CONFIG_NET
1207 int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1208 struct bpf_prog *prog);
1209
1210 int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1211 #else
1212 static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1213 struct bpf_prog *prog)
1214 {
1215 return -EOPNOTSUPP;
1216 }
1217
1218 static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1219 {
1220 return -EOPNOTSUPP;
1221 }
1222 #endif
1223
1224 bool __skb_flow_dissect(const struct sk_buff *skb,
1225 struct flow_dissector *flow_dissector,
1226 void *target_container,
1227 void *data, __be16 proto, int nhoff, int hlen,
1228 unsigned int flags);
1229
1230 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1231 struct flow_dissector *flow_dissector,
1232 void *target_container, unsigned int flags)
1233 {
1234 return __skb_flow_dissect(skb, flow_dissector, target_container,
1235 NULL, 0, 0, 0, flags);
1236 }
1237
1238 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1239 struct flow_keys *flow,
1240 unsigned int flags)
1241 {
1242 memset(flow, 0, sizeof(*flow));
1243 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1244 NULL, 0, 0, 0, flags);
1245 }
1246
1247 static inline bool
1248 skb_flow_dissect_flow_keys_basic(const struct sk_buff *skb,
1249 struct flow_keys_basic *flow, void *data,
1250 __be16 proto, int nhoff, int hlen,
1251 unsigned int flags)
1252 {
1253 memset(flow, 0, sizeof(*flow));
1254 return __skb_flow_dissect(skb, &flow_keys_basic_dissector, flow,
1255 data, proto, nhoff, hlen, flags);
1256 }
1257
1258 void
1259 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1260 struct flow_dissector *flow_dissector,
1261 void *target_container);
1262
1263 static inline __u32 skb_get_hash(struct sk_buff *skb)
1264 {
1265 if (!skb->l4_hash && !skb->sw_hash)
1266 __skb_get_hash(skb);
1267
1268 return skb->hash;
1269 }
1270
1271 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1272 {
1273 if (!skb->l4_hash && !skb->sw_hash) {
1274 struct flow_keys keys;
1275 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1276
1277 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1278 }
1279
1280 return skb->hash;
1281 }
1282
1283 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1284
1285 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1286 {
1287 return skb->hash;
1288 }
1289
1290 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1291 {
1292 to->hash = from->hash;
1293 to->sw_hash = from->sw_hash;
1294 to->l4_hash = from->l4_hash;
1295 };
1296
1297 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1298 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1299 {
1300 return skb->head + skb->end;
1301 }
1302
1303 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1304 {
1305 return skb->end;
1306 }
1307 #else
1308 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1309 {
1310 return skb->end;
1311 }
1312
1313 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1314 {
1315 return skb->end - skb->head;
1316 }
1317 #endif
1318
1319 /* Internal */
1320 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1321
1322 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1323 {
1324 return &skb_shinfo(skb)->hwtstamps;
1325 }
1326
1327 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1328 {
1329 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1330
1331 return is_zcopy ? skb_uarg(skb) : NULL;
1332 }
1333
1334 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1335 bool *have_ref)
1336 {
1337 if (skb && uarg && !skb_zcopy(skb)) {
1338 if (unlikely(have_ref && *have_ref))
1339 *have_ref = false;
1340 else
1341 sock_zerocopy_get(uarg);
1342 skb_shinfo(skb)->destructor_arg = uarg;
1343 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1344 }
1345 }
1346
1347 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1348 {
1349 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1350 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1351 }
1352
1353 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1354 {
1355 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1356 }
1357
1358 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1359 {
1360 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1361 }
1362
1363 /* Release a reference on a zerocopy structure */
1364 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1365 {
1366 struct ubuf_info *uarg = skb_zcopy(skb);
1367
1368 if (uarg) {
1369 if (uarg->callback == sock_zerocopy_callback) {
1370 uarg->zerocopy = uarg->zerocopy && zerocopy;
1371 sock_zerocopy_put(uarg);
1372 } else if (!skb_zcopy_is_nouarg(skb)) {
1373 uarg->callback(uarg, zerocopy);
1374 }
1375
1376 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1377 }
1378 }
1379
1380 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1381 static inline void skb_zcopy_abort(struct sk_buff *skb)
1382 {
1383 struct ubuf_info *uarg = skb_zcopy(skb);
1384
1385 if (uarg) {
1386 sock_zerocopy_put_abort(uarg, false);
1387 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1388 }
1389 }
1390
1391 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1392 {
1393 skb->next = NULL;
1394 }
1395
1396 static inline void skb_list_del_init(struct sk_buff *skb)
1397 {
1398 __list_del_entry(&skb->list);
1399 skb_mark_not_on_list(skb);
1400 }
1401
1402 /**
1403 * skb_queue_empty - check if a queue is empty
1404 * @list: queue head
1405 *
1406 * Returns true if the queue is empty, false otherwise.
1407 */
1408 static inline int skb_queue_empty(const struct sk_buff_head *list)
1409 {
1410 return list->next == (const struct sk_buff *) list;
1411 }
1412
1413 /**
1414 * skb_queue_is_last - check if skb is the last entry in the queue
1415 * @list: queue head
1416 * @skb: buffer
1417 *
1418 * Returns true if @skb is the last buffer on the list.
1419 */
1420 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1421 const struct sk_buff *skb)
1422 {
1423 return skb->next == (const struct sk_buff *) list;
1424 }
1425
1426 /**
1427 * skb_queue_is_first - check if skb is the first entry in the queue
1428 * @list: queue head
1429 * @skb: buffer
1430 *
1431 * Returns true if @skb is the first buffer on the list.
1432 */
1433 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1434 const struct sk_buff *skb)
1435 {
1436 return skb->prev == (const struct sk_buff *) list;
1437 }
1438
1439 /**
1440 * skb_queue_next - return the next packet in the queue
1441 * @list: queue head
1442 * @skb: current buffer
1443 *
1444 * Return the next packet in @list after @skb. It is only valid to
1445 * call this if skb_queue_is_last() evaluates to false.
1446 */
1447 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1448 const struct sk_buff *skb)
1449 {
1450 /* This BUG_ON may seem severe, but if we just return then we
1451 * are going to dereference garbage.
1452 */
1453 BUG_ON(skb_queue_is_last(list, skb));
1454 return skb->next;
1455 }
1456
1457 /**
1458 * skb_queue_prev - return the prev packet in the queue
1459 * @list: queue head
1460 * @skb: current buffer
1461 *
1462 * Return the prev packet in @list before @skb. It is only valid to
1463 * call this if skb_queue_is_first() evaluates to false.
1464 */
1465 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1466 const struct sk_buff *skb)
1467 {
1468 /* This BUG_ON may seem severe, but if we just return then we
1469 * are going to dereference garbage.
1470 */
1471 BUG_ON(skb_queue_is_first(list, skb));
1472 return skb->prev;
1473 }
1474
1475 /**
1476 * skb_get - reference buffer
1477 * @skb: buffer to reference
1478 *
1479 * Makes another reference to a socket buffer and returns a pointer
1480 * to the buffer.
1481 */
1482 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1483 {
1484 refcount_inc(&skb->users);
1485 return skb;
1486 }
1487
1488 /*
1489 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1490 */
1491
1492 /**
1493 * skb_cloned - is the buffer a clone
1494 * @skb: buffer to check
1495 *
1496 * Returns true if the buffer was generated with skb_clone() and is
1497 * one of multiple shared copies of the buffer. Cloned buffers are
1498 * shared data so must not be written to under normal circumstances.
1499 */
1500 static inline int skb_cloned(const struct sk_buff *skb)
1501 {
1502 return skb->cloned &&
1503 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1504 }
1505
1506 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1507 {
1508 might_sleep_if(gfpflags_allow_blocking(pri));
1509
1510 if (skb_cloned(skb))
1511 return pskb_expand_head(skb, 0, 0, pri);
1512
1513 return 0;
1514 }
1515
1516 /**
1517 * skb_header_cloned - is the header a clone
1518 * @skb: buffer to check
1519 *
1520 * Returns true if modifying the header part of the buffer requires
1521 * the data to be copied.
1522 */
1523 static inline int skb_header_cloned(const struct sk_buff *skb)
1524 {
1525 int dataref;
1526
1527 if (!skb->cloned)
1528 return 0;
1529
1530 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1531 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1532 return dataref != 1;
1533 }
1534
1535 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1536 {
1537 might_sleep_if(gfpflags_allow_blocking(pri));
1538
1539 if (skb_header_cloned(skb))
1540 return pskb_expand_head(skb, 0, 0, pri);
1541
1542 return 0;
1543 }
1544
1545 /**
1546 * __skb_header_release - release reference to header
1547 * @skb: buffer to operate on
1548 */
1549 static inline void __skb_header_release(struct sk_buff *skb)
1550 {
1551 skb->nohdr = 1;
1552 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1553 }
1554
1555
1556 /**
1557 * skb_shared - is the buffer shared
1558 * @skb: buffer to check
1559 *
1560 * Returns true if more than one person has a reference to this
1561 * buffer.
1562 */
1563 static inline int skb_shared(const struct sk_buff *skb)
1564 {
1565 return refcount_read(&skb->users) != 1;
1566 }
1567
1568 /**
1569 * skb_share_check - check if buffer is shared and if so clone it
1570 * @skb: buffer to check
1571 * @pri: priority for memory allocation
1572 *
1573 * If the buffer is shared the buffer is cloned and the old copy
1574 * drops a reference. A new clone with a single reference is returned.
1575 * If the buffer is not shared the original buffer is returned. When
1576 * being called from interrupt status or with spinlocks held pri must
1577 * be GFP_ATOMIC.
1578 *
1579 * NULL is returned on a memory allocation failure.
1580 */
1581 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1582 {
1583 might_sleep_if(gfpflags_allow_blocking(pri));
1584 if (skb_shared(skb)) {
1585 struct sk_buff *nskb = skb_clone(skb, pri);
1586
1587 if (likely(nskb))
1588 consume_skb(skb);
1589 else
1590 kfree_skb(skb);
1591 skb = nskb;
1592 }
1593 return skb;
1594 }
1595
1596 /*
1597 * Copy shared buffers into a new sk_buff. We effectively do COW on
1598 * packets to handle cases where we have a local reader and forward
1599 * and a couple of other messy ones. The normal one is tcpdumping
1600 * a packet thats being forwarded.
1601 */
1602
1603 /**
1604 * skb_unshare - make a copy of a shared buffer
1605 * @skb: buffer to check
1606 * @pri: priority for memory allocation
1607 *
1608 * If the socket buffer is a clone then this function creates a new
1609 * copy of the data, drops a reference count on the old copy and returns
1610 * the new copy with the reference count at 1. If the buffer is not a clone
1611 * the original buffer is returned. When called with a spinlock held or
1612 * from interrupt state @pri must be %GFP_ATOMIC
1613 *
1614 * %NULL is returned on a memory allocation failure.
1615 */
1616 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1617 gfp_t pri)
1618 {
1619 might_sleep_if(gfpflags_allow_blocking(pri));
1620 if (skb_cloned(skb)) {
1621 struct sk_buff *nskb = skb_copy(skb, pri);
1622
1623 /* Free our shared copy */
1624 if (likely(nskb))
1625 consume_skb(skb);
1626 else
1627 kfree_skb(skb);
1628 skb = nskb;
1629 }
1630 return skb;
1631 }
1632
1633 /**
1634 * skb_peek - peek at the head of an &sk_buff_head
1635 * @list_: list to peek at
1636 *
1637 * Peek an &sk_buff. Unlike most other operations you _MUST_
1638 * be careful with this one. A peek leaves the buffer on the
1639 * list and someone else may run off with it. You must hold
1640 * the appropriate locks or have a private queue to do this.
1641 *
1642 * Returns %NULL for an empty list or a pointer to the head element.
1643 * The reference count is not incremented and the reference is therefore
1644 * volatile. Use with caution.
1645 */
1646 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1647 {
1648 struct sk_buff *skb = list_->next;
1649
1650 if (skb == (struct sk_buff *)list_)
1651 skb = NULL;
1652 return skb;
1653 }
1654
1655 /**
1656 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1657 * @list_: list to peek at
1658 *
1659 * Like skb_peek(), but the caller knows that the list is not empty.
1660 */
1661 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1662 {
1663 return list_->next;
1664 }
1665
1666 /**
1667 * skb_peek_next - peek skb following the given one from a queue
1668 * @skb: skb to start from
1669 * @list_: list to peek at
1670 *
1671 * Returns %NULL when the end of the list is met or a pointer to the
1672 * next element. The reference count is not incremented and the
1673 * reference is therefore volatile. Use with caution.
1674 */
1675 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1676 const struct sk_buff_head *list_)
1677 {
1678 struct sk_buff *next = skb->next;
1679
1680 if (next == (struct sk_buff *)list_)
1681 next = NULL;
1682 return next;
1683 }
1684
1685 /**
1686 * skb_peek_tail - peek at the tail of an &sk_buff_head
1687 * @list_: list to peek at
1688 *
1689 * Peek an &sk_buff. Unlike most other operations you _MUST_
1690 * be careful with this one. A peek leaves the buffer on the
1691 * list and someone else may run off with it. You must hold
1692 * the appropriate locks or have a private queue to do this.
1693 *
1694 * Returns %NULL for an empty list or a pointer to the tail element.
1695 * The reference count is not incremented and the reference is therefore
1696 * volatile. Use with caution.
1697 */
1698 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1699 {
1700 struct sk_buff *skb = list_->prev;
1701
1702 if (skb == (struct sk_buff *)list_)
1703 skb = NULL;
1704 return skb;
1705
1706 }
1707
1708 /**
1709 * skb_queue_len - get queue length
1710 * @list_: list to measure
1711 *
1712 * Return the length of an &sk_buff queue.
1713 */
1714 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1715 {
1716 return list_->qlen;
1717 }
1718
1719 /**
1720 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1721 * @list: queue to initialize
1722 *
1723 * This initializes only the list and queue length aspects of
1724 * an sk_buff_head object. This allows to initialize the list
1725 * aspects of an sk_buff_head without reinitializing things like
1726 * the spinlock. It can also be used for on-stack sk_buff_head
1727 * objects where the spinlock is known to not be used.
1728 */
1729 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1730 {
1731 list->prev = list->next = (struct sk_buff *)list;
1732 list->qlen = 0;
1733 }
1734
1735 /*
1736 * This function creates a split out lock class for each invocation;
1737 * this is needed for now since a whole lot of users of the skb-queue
1738 * infrastructure in drivers have different locking usage (in hardirq)
1739 * than the networking core (in softirq only). In the long run either the
1740 * network layer or drivers should need annotation to consolidate the
1741 * main types of usage into 3 classes.
1742 */
1743 static inline void skb_queue_head_init(struct sk_buff_head *list)
1744 {
1745 spin_lock_init(&list->lock);
1746 __skb_queue_head_init(list);
1747 }
1748
1749 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1750 struct lock_class_key *class)
1751 {
1752 skb_queue_head_init(list);
1753 lockdep_set_class(&list->lock, class);
1754 }
1755
1756 /*
1757 * Insert an sk_buff on a list.
1758 *
1759 * The "__skb_xxxx()" functions are the non-atomic ones that
1760 * can only be called with interrupts disabled.
1761 */
1762 static inline void __skb_insert(struct sk_buff *newsk,
1763 struct sk_buff *prev, struct sk_buff *next,
1764 struct sk_buff_head *list)
1765 {
1766 newsk->next = next;
1767 newsk->prev = prev;
1768 next->prev = prev->next = newsk;
1769 list->qlen++;
1770 }
1771
1772 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1773 struct sk_buff *prev,
1774 struct sk_buff *next)
1775 {
1776 struct sk_buff *first = list->next;
1777 struct sk_buff *last = list->prev;
1778
1779 first->prev = prev;
1780 prev->next = first;
1781
1782 last->next = next;
1783 next->prev = last;
1784 }
1785
1786 /**
1787 * skb_queue_splice - join two skb lists, this is designed for stacks
1788 * @list: the new list to add
1789 * @head: the place to add it in the first list
1790 */
1791 static inline void skb_queue_splice(const struct sk_buff_head *list,
1792 struct sk_buff_head *head)
1793 {
1794 if (!skb_queue_empty(list)) {
1795 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1796 head->qlen += list->qlen;
1797 }
1798 }
1799
1800 /**
1801 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1802 * @list: the new list to add
1803 * @head: the place to add it in the first list
1804 *
1805 * The list at @list is reinitialised
1806 */
1807 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1808 struct sk_buff_head *head)
1809 {
1810 if (!skb_queue_empty(list)) {
1811 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1812 head->qlen += list->qlen;
1813 __skb_queue_head_init(list);
1814 }
1815 }
1816
1817 /**
1818 * skb_queue_splice_tail - join two skb lists, each list being a queue
1819 * @list: the new list to add
1820 * @head: the place to add it in the first list
1821 */
1822 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1823 struct sk_buff_head *head)
1824 {
1825 if (!skb_queue_empty(list)) {
1826 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1827 head->qlen += list->qlen;
1828 }
1829 }
1830
1831 /**
1832 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1833 * @list: the new list to add
1834 * @head: the place to add it in the first list
1835 *
1836 * Each of the lists is a queue.
1837 * The list at @list is reinitialised
1838 */
1839 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1840 struct sk_buff_head *head)
1841 {
1842 if (!skb_queue_empty(list)) {
1843 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1844 head->qlen += list->qlen;
1845 __skb_queue_head_init(list);
1846 }
1847 }
1848
1849 /**
1850 * __skb_queue_after - queue a buffer at the list head
1851 * @list: list to use
1852 * @prev: place after this buffer
1853 * @newsk: buffer to queue
1854 *
1855 * Queue a buffer int the middle of a list. This function takes no locks
1856 * and you must therefore hold required locks before calling it.
1857 *
1858 * A buffer cannot be placed on two lists at the same time.
1859 */
1860 static inline void __skb_queue_after(struct sk_buff_head *list,
1861 struct sk_buff *prev,
1862 struct sk_buff *newsk)
1863 {
1864 __skb_insert(newsk, prev, prev->next, list);
1865 }
1866
1867 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1868 struct sk_buff_head *list);
1869
1870 static inline void __skb_queue_before(struct sk_buff_head *list,
1871 struct sk_buff *next,
1872 struct sk_buff *newsk)
1873 {
1874 __skb_insert(newsk, next->prev, next, list);
1875 }
1876
1877 /**
1878 * __skb_queue_head - queue a buffer at the list head
1879 * @list: list to use
1880 * @newsk: buffer to queue
1881 *
1882 * Queue a buffer at the start of a list. This function takes no locks
1883 * and you must therefore hold required locks before calling it.
1884 *
1885 * A buffer cannot be placed on two lists at the same time.
1886 */
1887 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1888 static inline void __skb_queue_head(struct sk_buff_head *list,
1889 struct sk_buff *newsk)
1890 {
1891 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1892 }
1893
1894 /**
1895 * __skb_queue_tail - queue a buffer at the list tail
1896 * @list: list to use
1897 * @newsk: buffer to queue
1898 *
1899 * Queue a buffer at the end of a list. This function takes no locks
1900 * and you must therefore hold required locks before calling it.
1901 *
1902 * A buffer cannot be placed on two lists at the same time.
1903 */
1904 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1905 static inline void __skb_queue_tail(struct sk_buff_head *list,
1906 struct sk_buff *newsk)
1907 {
1908 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1909 }
1910
1911 /*
1912 * remove sk_buff from list. _Must_ be called atomically, and with
1913 * the list known..
1914 */
1915 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1916 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1917 {
1918 struct sk_buff *next, *prev;
1919
1920 list->qlen--;
1921 next = skb->next;
1922 prev = skb->prev;
1923 skb->next = skb->prev = NULL;
1924 next->prev = prev;
1925 prev->next = next;
1926 }
1927
1928 /**
1929 * __skb_dequeue - remove from the head of the queue
1930 * @list: list to dequeue from
1931 *
1932 * Remove the head of the list. This function does not take any locks
1933 * so must be used with appropriate locks held only. The head item is
1934 * returned or %NULL if the list is empty.
1935 */
1936 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1937 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1938 {
1939 struct sk_buff *skb = skb_peek(list);
1940 if (skb)
1941 __skb_unlink(skb, list);
1942 return skb;
1943 }
1944
1945 /**
1946 * __skb_dequeue_tail - remove from the tail of the queue
1947 * @list: list to dequeue from
1948 *
1949 * Remove the tail of the list. This function does not take any locks
1950 * so must be used with appropriate locks held only. The tail item is
1951 * returned or %NULL if the list is empty.
1952 */
1953 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1954 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1955 {
1956 struct sk_buff *skb = skb_peek_tail(list);
1957 if (skb)
1958 __skb_unlink(skb, list);
1959 return skb;
1960 }
1961
1962
1963 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1964 {
1965 return skb->data_len;
1966 }
1967
1968 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1969 {
1970 return skb->len - skb->data_len;
1971 }
1972
1973 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
1974 {
1975 unsigned int i, len = 0;
1976
1977 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1978 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1979 return len;
1980 }
1981
1982 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1983 {
1984 return skb_headlen(skb) + __skb_pagelen(skb);
1985 }
1986
1987 /**
1988 * __skb_fill_page_desc - initialise a paged fragment in an skb
1989 * @skb: buffer containing fragment to be initialised
1990 * @i: paged fragment index to initialise
1991 * @page: the page to use for this fragment
1992 * @off: the offset to the data with @page
1993 * @size: the length of the data
1994 *
1995 * Initialises the @i'th fragment of @skb to point to &size bytes at
1996 * offset @off within @page.
1997 *
1998 * Does not take any additional reference on the fragment.
1999 */
2000 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2001 struct page *page, int off, int size)
2002 {
2003 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2004
2005 /*
2006 * Propagate page pfmemalloc to the skb if we can. The problem is
2007 * that not all callers have unique ownership of the page but rely
2008 * on page_is_pfmemalloc doing the right thing(tm).
2009 */
2010 frag->page.p = page;
2011 frag->page_offset = off;
2012 skb_frag_size_set(frag, size);
2013
2014 page = compound_head(page);
2015 if (page_is_pfmemalloc(page))
2016 skb->pfmemalloc = true;
2017 }
2018
2019 /**
2020 * skb_fill_page_desc - initialise a paged fragment in an skb
2021 * @skb: buffer containing fragment to be initialised
2022 * @i: paged fragment index to initialise
2023 * @page: the page to use for this fragment
2024 * @off: the offset to the data with @page
2025 * @size: the length of the data
2026 *
2027 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2028 * @skb to point to @size bytes at offset @off within @page. In
2029 * addition updates @skb such that @i is the last fragment.
2030 *
2031 * Does not take any additional reference on the fragment.
2032 */
2033 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2034 struct page *page, int off, int size)
2035 {
2036 __skb_fill_page_desc(skb, i, page, off, size);
2037 skb_shinfo(skb)->nr_frags = i + 1;
2038 }
2039
2040 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2041 int size, unsigned int truesize);
2042
2043 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2044 unsigned int truesize);
2045
2046 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
2047 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
2048 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2049
2050 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2051 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2052 {
2053 return skb->head + skb->tail;
2054 }
2055
2056 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2057 {
2058 skb->tail = skb->data - skb->head;
2059 }
2060
2061 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2062 {
2063 skb_reset_tail_pointer(skb);
2064 skb->tail += offset;
2065 }
2066
2067 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2068 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2069 {
2070 return skb->tail;
2071 }
2072
2073 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2074 {
2075 skb->tail = skb->data;
2076 }
2077
2078 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2079 {
2080 skb->tail = skb->data + offset;
2081 }
2082
2083 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2084
2085 /*
2086 * Add data to an sk_buff
2087 */
2088 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2089 void *skb_put(struct sk_buff *skb, unsigned int len);
2090 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2091 {
2092 void *tmp = skb_tail_pointer(skb);
2093 SKB_LINEAR_ASSERT(skb);
2094 skb->tail += len;
2095 skb->len += len;
2096 return tmp;
2097 }
2098
2099 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2100 {
2101 void *tmp = __skb_put(skb, len);
2102
2103 memset(tmp, 0, len);
2104 return tmp;
2105 }
2106
2107 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2108 unsigned int len)
2109 {
2110 void *tmp = __skb_put(skb, len);
2111
2112 memcpy(tmp, data, len);
2113 return tmp;
2114 }
2115
2116 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2117 {
2118 *(u8 *)__skb_put(skb, 1) = val;
2119 }
2120
2121 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2122 {
2123 void *tmp = skb_put(skb, len);
2124
2125 memset(tmp, 0, len);
2126
2127 return tmp;
2128 }
2129
2130 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2131 unsigned int len)
2132 {
2133 void *tmp = skb_put(skb, len);
2134
2135 memcpy(tmp, data, len);
2136
2137 return tmp;
2138 }
2139
2140 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2141 {
2142 *(u8 *)skb_put(skb, 1) = val;
2143 }
2144
2145 void *skb_push(struct sk_buff *skb, unsigned int len);
2146 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2147 {
2148 skb->data -= len;
2149 skb->len += len;
2150 return skb->data;
2151 }
2152
2153 void *skb_pull(struct sk_buff *skb, unsigned int len);
2154 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2155 {
2156 skb->len -= len;
2157 BUG_ON(skb->len < skb->data_len);
2158 return skb->data += len;
2159 }
2160
2161 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2162 {
2163 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2164 }
2165
2166 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2167
2168 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2169 {
2170 if (len > skb_headlen(skb) &&
2171 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2172 return NULL;
2173 skb->len -= len;
2174 return skb->data += len;
2175 }
2176
2177 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2178 {
2179 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2180 }
2181
2182 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2183 {
2184 if (likely(len <= skb_headlen(skb)))
2185 return 1;
2186 if (unlikely(len > skb->len))
2187 return 0;
2188 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2189 }
2190
2191 void skb_condense(struct sk_buff *skb);
2192
2193 /**
2194 * skb_headroom - bytes at buffer head
2195 * @skb: buffer to check
2196 *
2197 * Return the number of bytes of free space at the head of an &sk_buff.
2198 */
2199 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2200 {
2201 return skb->data - skb->head;
2202 }
2203
2204 /**
2205 * skb_tailroom - bytes at buffer end
2206 * @skb: buffer to check
2207 *
2208 * Return the number of bytes of free space at the tail of an sk_buff
2209 */
2210 static inline int skb_tailroom(const struct sk_buff *skb)
2211 {
2212 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2213 }
2214
2215 /**
2216 * skb_availroom - bytes at buffer end
2217 * @skb: buffer to check
2218 *
2219 * Return the number of bytes of free space at the tail of an sk_buff
2220 * allocated by sk_stream_alloc()
2221 */
2222 static inline int skb_availroom(const struct sk_buff *skb)
2223 {
2224 if (skb_is_nonlinear(skb))
2225 return 0;
2226
2227 return skb->end - skb->tail - skb->reserved_tailroom;
2228 }
2229
2230 /**
2231 * skb_reserve - adjust headroom
2232 * @skb: buffer to alter
2233 * @len: bytes to move
2234 *
2235 * Increase the headroom of an empty &sk_buff by reducing the tail
2236 * room. This is only allowed for an empty buffer.
2237 */
2238 static inline void skb_reserve(struct sk_buff *skb, int len)
2239 {
2240 skb->data += len;
2241 skb->tail += len;
2242 }
2243
2244 /**
2245 * skb_tailroom_reserve - adjust reserved_tailroom
2246 * @skb: buffer to alter
2247 * @mtu: maximum amount of headlen permitted
2248 * @needed_tailroom: minimum amount of reserved_tailroom
2249 *
2250 * Set reserved_tailroom so that headlen can be as large as possible but
2251 * not larger than mtu and tailroom cannot be smaller than
2252 * needed_tailroom.
2253 * The required headroom should already have been reserved before using
2254 * this function.
2255 */
2256 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2257 unsigned int needed_tailroom)
2258 {
2259 SKB_LINEAR_ASSERT(skb);
2260 if (mtu < skb_tailroom(skb) - needed_tailroom)
2261 /* use at most mtu */
2262 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2263 else
2264 /* use up to all available space */
2265 skb->reserved_tailroom = needed_tailroom;
2266 }
2267
2268 #define ENCAP_TYPE_ETHER 0
2269 #define ENCAP_TYPE_IPPROTO 1
2270
2271 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2272 __be16 protocol)
2273 {
2274 skb->inner_protocol = protocol;
2275 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2276 }
2277
2278 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2279 __u8 ipproto)
2280 {
2281 skb->inner_ipproto = ipproto;
2282 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2283 }
2284
2285 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2286 {
2287 skb->inner_mac_header = skb->mac_header;
2288 skb->inner_network_header = skb->network_header;
2289 skb->inner_transport_header = skb->transport_header;
2290 }
2291
2292 static inline void skb_reset_mac_len(struct sk_buff *skb)
2293 {
2294 skb->mac_len = skb->network_header - skb->mac_header;
2295 }
2296
2297 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2298 *skb)
2299 {
2300 return skb->head + skb->inner_transport_header;
2301 }
2302
2303 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2304 {
2305 return skb_inner_transport_header(skb) - skb->data;
2306 }
2307
2308 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2309 {
2310 skb->inner_transport_header = skb->data - skb->head;
2311 }
2312
2313 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2314 const int offset)
2315 {
2316 skb_reset_inner_transport_header(skb);
2317 skb->inner_transport_header += offset;
2318 }
2319
2320 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2321 {
2322 return skb->head + skb->inner_network_header;
2323 }
2324
2325 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2326 {
2327 skb->inner_network_header = skb->data - skb->head;
2328 }
2329
2330 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2331 const int offset)
2332 {
2333 skb_reset_inner_network_header(skb);
2334 skb->inner_network_header += offset;
2335 }
2336
2337 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2338 {
2339 return skb->head + skb->inner_mac_header;
2340 }
2341
2342 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2343 {
2344 skb->inner_mac_header = skb->data - skb->head;
2345 }
2346
2347 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2348 const int offset)
2349 {
2350 skb_reset_inner_mac_header(skb);
2351 skb->inner_mac_header += offset;
2352 }
2353 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2354 {
2355 return skb->transport_header != (typeof(skb->transport_header))~0U;
2356 }
2357
2358 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2359 {
2360 return skb->head + skb->transport_header;
2361 }
2362
2363 static inline void skb_reset_transport_header(struct sk_buff *skb)
2364 {
2365 skb->transport_header = skb->data - skb->head;
2366 }
2367
2368 static inline void skb_set_transport_header(struct sk_buff *skb,
2369 const int offset)
2370 {
2371 skb_reset_transport_header(skb);
2372 skb->transport_header += offset;
2373 }
2374
2375 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2376 {
2377 return skb->head + skb->network_header;
2378 }
2379
2380 static inline void skb_reset_network_header(struct sk_buff *skb)
2381 {
2382 skb->network_header = skb->data - skb->head;
2383 }
2384
2385 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2386 {
2387 skb_reset_network_header(skb);
2388 skb->network_header += offset;
2389 }
2390
2391 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2392 {
2393 return skb->head + skb->mac_header;
2394 }
2395
2396 static inline int skb_mac_offset(const struct sk_buff *skb)
2397 {
2398 return skb_mac_header(skb) - skb->data;
2399 }
2400
2401 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2402 {
2403 return skb->network_header - skb->mac_header;
2404 }
2405
2406 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2407 {
2408 return skb->mac_header != (typeof(skb->mac_header))~0U;
2409 }
2410
2411 static inline void skb_reset_mac_header(struct sk_buff *skb)
2412 {
2413 skb->mac_header = skb->data - skb->head;
2414 }
2415
2416 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2417 {
2418 skb_reset_mac_header(skb);
2419 skb->mac_header += offset;
2420 }
2421
2422 static inline void skb_pop_mac_header(struct sk_buff *skb)
2423 {
2424 skb->mac_header = skb->network_header;
2425 }
2426
2427 static inline void skb_probe_transport_header(struct sk_buff *skb,
2428 const int offset_hint)
2429 {
2430 struct flow_keys_basic keys;
2431
2432 if (skb_transport_header_was_set(skb))
2433 return;
2434
2435 if (skb_flow_dissect_flow_keys_basic(skb, &keys, NULL, 0, 0, 0, 0))
2436 skb_set_transport_header(skb, keys.control.thoff);
2437 else
2438 skb_set_transport_header(skb, offset_hint);
2439 }
2440
2441 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2442 {
2443 if (skb_mac_header_was_set(skb)) {
2444 const unsigned char *old_mac = skb_mac_header(skb);
2445
2446 skb_set_mac_header(skb, -skb->mac_len);
2447 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2448 }
2449 }
2450
2451 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2452 {
2453 return skb->csum_start - skb_headroom(skb);
2454 }
2455
2456 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2457 {
2458 return skb->head + skb->csum_start;
2459 }
2460
2461 static inline int skb_transport_offset(const struct sk_buff *skb)
2462 {
2463 return skb_transport_header(skb) - skb->data;
2464 }
2465
2466 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2467 {
2468 return skb->transport_header - skb->network_header;
2469 }
2470
2471 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2472 {
2473 return skb->inner_transport_header - skb->inner_network_header;
2474 }
2475
2476 static inline int skb_network_offset(const struct sk_buff *skb)
2477 {
2478 return skb_network_header(skb) - skb->data;
2479 }
2480
2481 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2482 {
2483 return skb_inner_network_header(skb) - skb->data;
2484 }
2485
2486 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2487 {
2488 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2489 }
2490
2491 /*
2492 * CPUs often take a performance hit when accessing unaligned memory
2493 * locations. The actual performance hit varies, it can be small if the
2494 * hardware handles it or large if we have to take an exception and fix it
2495 * in software.
2496 *
2497 * Since an ethernet header is 14 bytes network drivers often end up with
2498 * the IP header at an unaligned offset. The IP header can be aligned by
2499 * shifting the start of the packet by 2 bytes. Drivers should do this
2500 * with:
2501 *
2502 * skb_reserve(skb, NET_IP_ALIGN);
2503 *
2504 * The downside to this alignment of the IP header is that the DMA is now
2505 * unaligned. On some architectures the cost of an unaligned DMA is high
2506 * and this cost outweighs the gains made by aligning the IP header.
2507 *
2508 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2509 * to be overridden.
2510 */
2511 #ifndef NET_IP_ALIGN
2512 #define NET_IP_ALIGN 2
2513 #endif
2514
2515 /*
2516 * The networking layer reserves some headroom in skb data (via
2517 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2518 * the header has to grow. In the default case, if the header has to grow
2519 * 32 bytes or less we avoid the reallocation.
2520 *
2521 * Unfortunately this headroom changes the DMA alignment of the resulting
2522 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2523 * on some architectures. An architecture can override this value,
2524 * perhaps setting it to a cacheline in size (since that will maintain
2525 * cacheline alignment of the DMA). It must be a power of 2.
2526 *
2527 * Various parts of the networking layer expect at least 32 bytes of
2528 * headroom, you should not reduce this.
2529 *
2530 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2531 * to reduce average number of cache lines per packet.
2532 * get_rps_cpus() for example only access one 64 bytes aligned block :
2533 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2534 */
2535 #ifndef NET_SKB_PAD
2536 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2537 #endif
2538
2539 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2540
2541 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2542 {
2543 if (WARN_ON(skb_is_nonlinear(skb)))
2544 return;
2545 skb->len = len;
2546 skb_set_tail_pointer(skb, len);
2547 }
2548
2549 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2550 {
2551 __skb_set_length(skb, len);
2552 }
2553
2554 void skb_trim(struct sk_buff *skb, unsigned int len);
2555
2556 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2557 {
2558 if (skb->data_len)
2559 return ___pskb_trim(skb, len);
2560 __skb_trim(skb, len);
2561 return 0;
2562 }
2563
2564 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2565 {
2566 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2567 }
2568
2569 /**
2570 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2571 * @skb: buffer to alter
2572 * @len: new length
2573 *
2574 * This is identical to pskb_trim except that the caller knows that
2575 * the skb is not cloned so we should never get an error due to out-
2576 * of-memory.
2577 */
2578 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2579 {
2580 int err = pskb_trim(skb, len);
2581 BUG_ON(err);
2582 }
2583
2584 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2585 {
2586 unsigned int diff = len - skb->len;
2587
2588 if (skb_tailroom(skb) < diff) {
2589 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2590 GFP_ATOMIC);
2591 if (ret)
2592 return ret;
2593 }
2594 __skb_set_length(skb, len);
2595 return 0;
2596 }
2597
2598 /**
2599 * skb_orphan - orphan a buffer
2600 * @skb: buffer to orphan
2601 *
2602 * If a buffer currently has an owner then we call the owner's
2603 * destructor function and make the @skb unowned. The buffer continues
2604 * to exist but is no longer charged to its former owner.
2605 */
2606 static inline void skb_orphan(struct sk_buff *skb)
2607 {
2608 if (skb->destructor) {
2609 skb->destructor(skb);
2610 skb->destructor = NULL;
2611 skb->sk = NULL;
2612 } else {
2613 BUG_ON(skb->sk);
2614 }
2615 }
2616
2617 /**
2618 * skb_orphan_frags - orphan the frags contained in a buffer
2619 * @skb: buffer to orphan frags from
2620 * @gfp_mask: allocation mask for replacement pages
2621 *
2622 * For each frag in the SKB which needs a destructor (i.e. has an
2623 * owner) create a copy of that frag and release the original
2624 * page by calling the destructor.
2625 */
2626 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2627 {
2628 if (likely(!skb_zcopy(skb)))
2629 return 0;
2630 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2631 return 0;
2632 return skb_copy_ubufs(skb, gfp_mask);
2633 }
2634
2635 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2636 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2637 {
2638 if (likely(!skb_zcopy(skb)))
2639 return 0;
2640 return skb_copy_ubufs(skb, gfp_mask);
2641 }
2642
2643 /**
2644 * __skb_queue_purge - empty a list
2645 * @list: list to empty
2646 *
2647 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2648 * the list and one reference dropped. This function does not take the
2649 * list lock and the caller must hold the relevant locks to use it.
2650 */
2651 void skb_queue_purge(struct sk_buff_head *list);
2652 static inline void __skb_queue_purge(struct sk_buff_head *list)
2653 {
2654 struct sk_buff *skb;
2655 while ((skb = __skb_dequeue(list)) != NULL)
2656 kfree_skb(skb);
2657 }
2658
2659 unsigned int skb_rbtree_purge(struct rb_root *root);
2660
2661 void *netdev_alloc_frag(unsigned int fragsz);
2662
2663 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2664 gfp_t gfp_mask);
2665
2666 /**
2667 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2668 * @dev: network device to receive on
2669 * @length: length to allocate
2670 *
2671 * Allocate a new &sk_buff and assign it a usage count of one. The
2672 * buffer has unspecified headroom built in. Users should allocate
2673 * the headroom they think they need without accounting for the
2674 * built in space. The built in space is used for optimisations.
2675 *
2676 * %NULL is returned if there is no free memory. Although this function
2677 * allocates memory it can be called from an interrupt.
2678 */
2679 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2680 unsigned int length)
2681 {
2682 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2683 }
2684
2685 /* legacy helper around __netdev_alloc_skb() */
2686 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2687 gfp_t gfp_mask)
2688 {
2689 return __netdev_alloc_skb(NULL, length, gfp_mask);
2690 }
2691
2692 /* legacy helper around netdev_alloc_skb() */
2693 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2694 {
2695 return netdev_alloc_skb(NULL, length);
2696 }
2697
2698
2699 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2700 unsigned int length, gfp_t gfp)
2701 {
2702 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2703
2704 if (NET_IP_ALIGN && skb)
2705 skb_reserve(skb, NET_IP_ALIGN);
2706 return skb;
2707 }
2708
2709 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2710 unsigned int length)
2711 {
2712 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2713 }
2714
2715 static inline void skb_free_frag(void *addr)
2716 {
2717 page_frag_free(addr);
2718 }
2719
2720 void *napi_alloc_frag(unsigned int fragsz);
2721 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2722 unsigned int length, gfp_t gfp_mask);
2723 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2724 unsigned int length)
2725 {
2726 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2727 }
2728 void napi_consume_skb(struct sk_buff *skb, int budget);
2729
2730 void __kfree_skb_flush(void);
2731 void __kfree_skb_defer(struct sk_buff *skb);
2732
2733 /**
2734 * __dev_alloc_pages - allocate page for network Rx
2735 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2736 * @order: size of the allocation
2737 *
2738 * Allocate a new page.
2739 *
2740 * %NULL is returned if there is no free memory.
2741 */
2742 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2743 unsigned int order)
2744 {
2745 /* This piece of code contains several assumptions.
2746 * 1. This is for device Rx, therefor a cold page is preferred.
2747 * 2. The expectation is the user wants a compound page.
2748 * 3. If requesting a order 0 page it will not be compound
2749 * due to the check to see if order has a value in prep_new_page
2750 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2751 * code in gfp_to_alloc_flags that should be enforcing this.
2752 */
2753 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2754
2755 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2756 }
2757
2758 static inline struct page *dev_alloc_pages(unsigned int order)
2759 {
2760 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2761 }
2762
2763 /**
2764 * __dev_alloc_page - allocate a page for network Rx
2765 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2766 *
2767 * Allocate a new page.
2768 *
2769 * %NULL is returned if there is no free memory.
2770 */
2771 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2772 {
2773 return __dev_alloc_pages(gfp_mask, 0);
2774 }
2775
2776 static inline struct page *dev_alloc_page(void)
2777 {
2778 return dev_alloc_pages(0);
2779 }
2780
2781 /**
2782 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2783 * @page: The page that was allocated from skb_alloc_page
2784 * @skb: The skb that may need pfmemalloc set
2785 */
2786 static inline void skb_propagate_pfmemalloc(struct page *page,
2787 struct sk_buff *skb)
2788 {
2789 if (page_is_pfmemalloc(page))
2790 skb->pfmemalloc = true;
2791 }
2792
2793 /**
2794 * skb_frag_page - retrieve the page referred to by a paged fragment
2795 * @frag: the paged fragment
2796 *
2797 * Returns the &struct page associated with @frag.
2798 */
2799 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2800 {
2801 return frag->page.p;
2802 }
2803
2804 /**
2805 * __skb_frag_ref - take an addition reference on a paged fragment.
2806 * @frag: the paged fragment
2807 *
2808 * Takes an additional reference on the paged fragment @frag.
2809 */
2810 static inline void __skb_frag_ref(skb_frag_t *frag)
2811 {
2812 get_page(skb_frag_page(frag));
2813 }
2814
2815 /**
2816 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2817 * @skb: the buffer
2818 * @f: the fragment offset.
2819 *
2820 * Takes an additional reference on the @f'th paged fragment of @skb.
2821 */
2822 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2823 {
2824 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2825 }
2826
2827 /**
2828 * __skb_frag_unref - release a reference on a paged fragment.
2829 * @frag: the paged fragment
2830 *
2831 * Releases a reference on the paged fragment @frag.
2832 */
2833 static inline void __skb_frag_unref(skb_frag_t *frag)
2834 {
2835 put_page(skb_frag_page(frag));
2836 }
2837
2838 /**
2839 * skb_frag_unref - release a reference on a paged fragment of an skb.
2840 * @skb: the buffer
2841 * @f: the fragment offset
2842 *
2843 * Releases a reference on the @f'th paged fragment of @skb.
2844 */
2845 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2846 {
2847 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2848 }
2849
2850 /**
2851 * skb_frag_address - gets the address of the data contained in a paged fragment
2852 * @frag: the paged fragment buffer
2853 *
2854 * Returns the address of the data within @frag. The page must already
2855 * be mapped.
2856 */
2857 static inline void *skb_frag_address(const skb_frag_t *frag)
2858 {
2859 return page_address(skb_frag_page(frag)) + frag->page_offset;
2860 }
2861
2862 /**
2863 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2864 * @frag: the paged fragment buffer
2865 *
2866 * Returns the address of the data within @frag. Checks that the page
2867 * is mapped and returns %NULL otherwise.
2868 */
2869 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2870 {
2871 void *ptr = page_address(skb_frag_page(frag));
2872 if (unlikely(!ptr))
2873 return NULL;
2874
2875 return ptr + frag->page_offset;
2876 }
2877
2878 /**
2879 * __skb_frag_set_page - sets the page contained in a paged fragment
2880 * @frag: the paged fragment
2881 * @page: the page to set
2882 *
2883 * Sets the fragment @frag to contain @page.
2884 */
2885 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2886 {
2887 frag->page.p = page;
2888 }
2889
2890 /**
2891 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2892 * @skb: the buffer
2893 * @f: the fragment offset
2894 * @page: the page to set
2895 *
2896 * Sets the @f'th fragment of @skb to contain @page.
2897 */
2898 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2899 struct page *page)
2900 {
2901 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2902 }
2903
2904 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2905
2906 /**
2907 * skb_frag_dma_map - maps a paged fragment via the DMA API
2908 * @dev: the device to map the fragment to
2909 * @frag: the paged fragment to map
2910 * @offset: the offset within the fragment (starting at the
2911 * fragment's own offset)
2912 * @size: the number of bytes to map
2913 * @dir: the direction of the mapping (``PCI_DMA_*``)
2914 *
2915 * Maps the page associated with @frag to @device.
2916 */
2917 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2918 const skb_frag_t *frag,
2919 size_t offset, size_t size,
2920 enum dma_data_direction dir)
2921 {
2922 return dma_map_page(dev, skb_frag_page(frag),
2923 frag->page_offset + offset, size, dir);
2924 }
2925
2926 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2927 gfp_t gfp_mask)
2928 {
2929 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2930 }
2931
2932
2933 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2934 gfp_t gfp_mask)
2935 {
2936 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2937 }
2938
2939
2940 /**
2941 * skb_clone_writable - is the header of a clone writable
2942 * @skb: buffer to check
2943 * @len: length up to which to write
2944 *
2945 * Returns true if modifying the header part of the cloned buffer
2946 * does not requires the data to be copied.
2947 */
2948 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2949 {
2950 return !skb_header_cloned(skb) &&
2951 skb_headroom(skb) + len <= skb->hdr_len;
2952 }
2953
2954 static inline int skb_try_make_writable(struct sk_buff *skb,
2955 unsigned int write_len)
2956 {
2957 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2958 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2959 }
2960
2961 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2962 int cloned)
2963 {
2964 int delta = 0;
2965
2966 if (headroom > skb_headroom(skb))
2967 delta = headroom - skb_headroom(skb);
2968
2969 if (delta || cloned)
2970 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2971 GFP_ATOMIC);
2972 return 0;
2973 }
2974
2975 /**
2976 * skb_cow - copy header of skb when it is required
2977 * @skb: buffer to cow
2978 * @headroom: needed headroom
2979 *
2980 * If the skb passed lacks sufficient headroom or its data part
2981 * is shared, data is reallocated. If reallocation fails, an error
2982 * is returned and original skb is not changed.
2983 *
2984 * The result is skb with writable area skb->head...skb->tail
2985 * and at least @headroom of space at head.
2986 */
2987 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2988 {
2989 return __skb_cow(skb, headroom, skb_cloned(skb));
2990 }
2991
2992 /**
2993 * skb_cow_head - skb_cow but only making the head writable
2994 * @skb: buffer to cow
2995 * @headroom: needed headroom
2996 *
2997 * This function is identical to skb_cow except that we replace the
2998 * skb_cloned check by skb_header_cloned. It should be used when
2999 * you only need to push on some header and do not need to modify
3000 * the data.
3001 */
3002 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3003 {
3004 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3005 }
3006
3007 /**
3008 * skb_padto - pad an skbuff up to a minimal size
3009 * @skb: buffer to pad
3010 * @len: minimal length
3011 *
3012 * Pads up a buffer to ensure the trailing bytes exist and are
3013 * blanked. If the buffer already contains sufficient data it
3014 * is untouched. Otherwise it is extended. Returns zero on
3015 * success. The skb is freed on error.
3016 */
3017 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3018 {
3019 unsigned int size = skb->len;
3020 if (likely(size >= len))
3021 return 0;
3022 return skb_pad(skb, len - size);
3023 }
3024
3025 /**
3026 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3027 * @skb: buffer to pad
3028 * @len: minimal length
3029 * @free_on_error: free buffer on error
3030 *
3031 * Pads up a buffer to ensure the trailing bytes exist and are
3032 * blanked. If the buffer already contains sufficient data it
3033 * is untouched. Otherwise it is extended. Returns zero on
3034 * success. The skb is freed on error if @free_on_error is true.
3035 */
3036 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3037 bool free_on_error)
3038 {
3039 unsigned int size = skb->len;
3040
3041 if (unlikely(size < len)) {
3042 len -= size;
3043 if (__skb_pad(skb, len, free_on_error))
3044 return -ENOMEM;
3045 __skb_put(skb, len);
3046 }
3047 return 0;
3048 }
3049
3050 /**
3051 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3052 * @skb: buffer to pad
3053 * @len: minimal length
3054 *
3055 * Pads up a buffer to ensure the trailing bytes exist and are
3056 * blanked. If the buffer already contains sufficient data it
3057 * is untouched. Otherwise it is extended. Returns zero on
3058 * success. The skb is freed on error.
3059 */
3060 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3061 {
3062 return __skb_put_padto(skb, len, true);
3063 }
3064
3065 static inline int skb_add_data(struct sk_buff *skb,
3066 struct iov_iter *from, int copy)
3067 {
3068 const int off = skb->len;
3069
3070 if (skb->ip_summed == CHECKSUM_NONE) {
3071 __wsum csum = 0;
3072 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3073 &csum, from)) {
3074 skb->csum = csum_block_add(skb->csum, csum, off);
3075 return 0;
3076 }
3077 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3078 return 0;
3079
3080 __skb_trim(skb, off);
3081 return -EFAULT;
3082 }
3083
3084 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3085 const struct page *page, int off)
3086 {
3087 if (skb_zcopy(skb))
3088 return false;
3089 if (i) {
3090 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3091
3092 return page == skb_frag_page(frag) &&
3093 off == frag->page_offset + skb_frag_size(frag);
3094 }
3095 return false;
3096 }
3097
3098 static inline int __skb_linearize(struct sk_buff *skb)
3099 {
3100 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3101 }
3102
3103 /**
3104 * skb_linearize - convert paged skb to linear one
3105 * @skb: buffer to linarize
3106 *
3107 * If there is no free memory -ENOMEM is returned, otherwise zero
3108 * is returned and the old skb data released.
3109 */
3110 static inline int skb_linearize(struct sk_buff *skb)
3111 {
3112 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3113 }
3114
3115 /**
3116 * skb_has_shared_frag - can any frag be overwritten
3117 * @skb: buffer to test
3118 *
3119 * Return true if the skb has at least one frag that might be modified
3120 * by an external entity (as in vmsplice()/sendfile())
3121 */
3122 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3123 {
3124 return skb_is_nonlinear(skb) &&
3125 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3126 }
3127
3128 /**
3129 * skb_linearize_cow - make sure skb is linear and writable
3130 * @skb: buffer to process
3131 *
3132 * If there is no free memory -ENOMEM is returned, otherwise zero
3133 * is returned and the old skb data released.
3134 */
3135 static inline int skb_linearize_cow(struct sk_buff *skb)
3136 {
3137 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3138 __skb_linearize(skb) : 0;
3139 }
3140
3141 static __always_inline void
3142 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3143 unsigned int off)
3144 {
3145 if (skb->ip_summed == CHECKSUM_COMPLETE)
3146 skb->csum = csum_block_sub(skb->csum,
3147 csum_partial(start, len, 0), off);
3148 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3149 skb_checksum_start_offset(skb) < 0)
3150 skb->ip_summed = CHECKSUM_NONE;
3151 }
3152
3153 /**
3154 * skb_postpull_rcsum - update checksum for received skb after pull
3155 * @skb: buffer to update
3156 * @start: start of data before pull
3157 * @len: length of data pulled
3158 *
3159 * After doing a pull on a received packet, you need to call this to
3160 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3161 * CHECKSUM_NONE so that it can be recomputed from scratch.
3162 */
3163 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3164 const void *start, unsigned int len)
3165 {
3166 __skb_postpull_rcsum(skb, start, len, 0);
3167 }
3168
3169 static __always_inline void
3170 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3171 unsigned int off)
3172 {
3173 if (skb->ip_summed == CHECKSUM_COMPLETE)
3174 skb->csum = csum_block_add(skb->csum,
3175 csum_partial(start, len, 0), off);
3176 }
3177
3178 /**
3179 * skb_postpush_rcsum - update checksum for received skb after push
3180 * @skb: buffer to update
3181 * @start: start of data after push
3182 * @len: length of data pushed
3183 *
3184 * After doing a push on a received packet, you need to call this to
3185 * update the CHECKSUM_COMPLETE checksum.
3186 */
3187 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3188 const void *start, unsigned int len)
3189 {
3190 __skb_postpush_rcsum(skb, start, len, 0);
3191 }
3192
3193 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3194
3195 /**
3196 * skb_push_rcsum - push skb and update receive checksum
3197 * @skb: buffer to update
3198 * @len: length of data pulled
3199 *
3200 * This function performs an skb_push on the packet and updates
3201 * the CHECKSUM_COMPLETE checksum. It should be used on
3202 * receive path processing instead of skb_push unless you know
3203 * that the checksum difference is zero (e.g., a valid IP header)
3204 * or you are setting ip_summed to CHECKSUM_NONE.
3205 */
3206 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3207 {
3208 skb_push(skb, len);
3209 skb_postpush_rcsum(skb, skb->data, len);
3210 return skb->data;
3211 }
3212
3213 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3214 /**
3215 * pskb_trim_rcsum - trim received skb and update checksum
3216 * @skb: buffer to trim
3217 * @len: new length
3218 *
3219 * This is exactly the same as pskb_trim except that it ensures the
3220 * checksum of received packets are still valid after the operation.
3221 * It can change skb pointers.
3222 */
3223
3224 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3225 {
3226 if (likely(len >= skb->len))
3227 return 0;
3228 return pskb_trim_rcsum_slow(skb, len);
3229 }
3230
3231 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3232 {
3233 if (skb->ip_summed == CHECKSUM_COMPLETE)
3234 skb->ip_summed = CHECKSUM_NONE;
3235 __skb_trim(skb, len);
3236 return 0;
3237 }
3238
3239 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3240 {
3241 if (skb->ip_summed == CHECKSUM_COMPLETE)
3242 skb->ip_summed = CHECKSUM_NONE;
3243 return __skb_grow(skb, len);
3244 }
3245
3246 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3247 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3248 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3249 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3250 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3251
3252 #define skb_queue_walk(queue, skb) \
3253 for (skb = (queue)->next; \
3254 skb != (struct sk_buff *)(queue); \
3255 skb = skb->next)
3256
3257 #define skb_queue_walk_safe(queue, skb, tmp) \
3258 for (skb = (queue)->next, tmp = skb->next; \
3259 skb != (struct sk_buff *)(queue); \
3260 skb = tmp, tmp = skb->next)
3261
3262 #define skb_queue_walk_from(queue, skb) \
3263 for (; skb != (struct sk_buff *)(queue); \
3264 skb = skb->next)
3265
3266 #define skb_rbtree_walk(skb, root) \
3267 for (skb = skb_rb_first(root); skb != NULL; \
3268 skb = skb_rb_next(skb))
3269
3270 #define skb_rbtree_walk_from(skb) \
3271 for (; skb != NULL; \
3272 skb = skb_rb_next(skb))
3273
3274 #define skb_rbtree_walk_from_safe(skb, tmp) \
3275 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3276 skb = tmp)
3277
3278 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3279 for (tmp = skb->next; \
3280 skb != (struct sk_buff *)(queue); \
3281 skb = tmp, tmp = skb->next)
3282
3283 #define skb_queue_reverse_walk(queue, skb) \
3284 for (skb = (queue)->prev; \
3285 skb != (struct sk_buff *)(queue); \
3286 skb = skb->prev)
3287
3288 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3289 for (skb = (queue)->prev, tmp = skb->prev; \
3290 skb != (struct sk_buff *)(queue); \
3291 skb = tmp, tmp = skb->prev)
3292
3293 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3294 for (tmp = skb->prev; \
3295 skb != (struct sk_buff *)(queue); \
3296 skb = tmp, tmp = skb->prev)
3297
3298 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3299 {
3300 return skb_shinfo(skb)->frag_list != NULL;
3301 }
3302
3303 static inline void skb_frag_list_init(struct sk_buff *skb)
3304 {
3305 skb_shinfo(skb)->frag_list = NULL;
3306 }
3307
3308 #define skb_walk_frags(skb, iter) \
3309 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3310
3311
3312 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3313 const struct sk_buff *skb);
3314 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3315 struct sk_buff_head *queue,
3316 unsigned int flags,
3317 void (*destructor)(struct sock *sk,
3318 struct sk_buff *skb),
3319 int *peeked, int *off, int *err,
3320 struct sk_buff **last);
3321 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3322 void (*destructor)(struct sock *sk,
3323 struct sk_buff *skb),
3324 int *peeked, int *off, int *err,
3325 struct sk_buff **last);
3326 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3327 void (*destructor)(struct sock *sk,
3328 struct sk_buff *skb),
3329 int *peeked, int *off, int *err);
3330 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3331 int *err);
3332 __poll_t datagram_poll(struct file *file, struct socket *sock,
3333 struct poll_table_struct *wait);
3334 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3335 struct iov_iter *to, int size);
3336 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3337 struct msghdr *msg, int size)
3338 {
3339 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3340 }
3341 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3342 struct msghdr *msg);
3343 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3344 struct iov_iter *to, int len,
3345 struct ahash_request *hash);
3346 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3347 struct iov_iter *from, int len);
3348 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3349 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3350 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3351 static inline void skb_free_datagram_locked(struct sock *sk,
3352 struct sk_buff *skb)
3353 {
3354 __skb_free_datagram_locked(sk, skb, 0);
3355 }
3356 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3357 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3358 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3359 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3360 int len, __wsum csum);
3361 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3362 struct pipe_inode_info *pipe, unsigned int len,
3363 unsigned int flags);
3364 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3365 int len);
3366 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3367 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3368 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3369 int len, int hlen);
3370 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3371 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3372 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3373 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3374 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3375 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3376 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3377 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3378 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3379 int skb_vlan_pop(struct sk_buff *skb);
3380 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3381 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3382 gfp_t gfp);
3383
3384 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3385 {
3386 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3387 }
3388
3389 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3390 {
3391 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3392 }
3393
3394 struct skb_checksum_ops {
3395 __wsum (*update)(const void *mem, int len, __wsum wsum);
3396 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3397 };
3398
3399 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3400
3401 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3402 __wsum csum, const struct skb_checksum_ops *ops);
3403 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3404 __wsum csum);
3405
3406 static inline void * __must_check
3407 __skb_header_pointer(const struct sk_buff *skb, int offset,
3408 int len, void *data, int hlen, void *buffer)
3409 {
3410 if (hlen - offset >= len)
3411 return data + offset;
3412
3413 if (!skb ||
3414 skb_copy_bits(skb, offset, buffer, len) < 0)
3415 return NULL;
3416
3417 return buffer;
3418 }
3419
3420 static inline void * __must_check
3421 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3422 {
3423 return __skb_header_pointer(skb, offset, len, skb->data,
3424 skb_headlen(skb), buffer);
3425 }
3426
3427 /**
3428 * skb_needs_linearize - check if we need to linearize a given skb
3429 * depending on the given device features.
3430 * @skb: socket buffer to check
3431 * @features: net device features
3432 *
3433 * Returns true if either:
3434 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3435 * 2. skb is fragmented and the device does not support SG.
3436 */
3437 static inline bool skb_needs_linearize(struct sk_buff *skb,
3438 netdev_features_t features)
3439 {
3440 return skb_is_nonlinear(skb) &&
3441 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3442 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3443 }
3444
3445 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3446 void *to,
3447 const unsigned int len)
3448 {
3449 memcpy(to, skb->data, len);
3450 }
3451
3452 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3453 const int offset, void *to,
3454 const unsigned int len)
3455 {
3456 memcpy(to, skb->data + offset, len);
3457 }
3458
3459 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3460 const void *from,
3461 const unsigned int len)
3462 {
3463 memcpy(skb->data, from, len);
3464 }
3465
3466 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3467 const int offset,
3468 const void *from,
3469 const unsigned int len)
3470 {
3471 memcpy(skb->data + offset, from, len);
3472 }
3473
3474 void skb_init(void);
3475
3476 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3477 {
3478 return skb->tstamp;
3479 }
3480
3481 /**
3482 * skb_get_timestamp - get timestamp from a skb
3483 * @skb: skb to get stamp from
3484 * @stamp: pointer to struct timeval to store stamp in
3485 *
3486 * Timestamps are stored in the skb as offsets to a base timestamp.
3487 * This function converts the offset back to a struct timeval and stores
3488 * it in stamp.
3489 */
3490 static inline void skb_get_timestamp(const struct sk_buff *skb,
3491 struct timeval *stamp)
3492 {
3493 *stamp = ktime_to_timeval(skb->tstamp);
3494 }
3495
3496 static inline void skb_get_timestampns(const struct sk_buff *skb,
3497 struct timespec *stamp)
3498 {
3499 *stamp = ktime_to_timespec(skb->tstamp);
3500 }
3501
3502 static inline void __net_timestamp(struct sk_buff *skb)
3503 {
3504 skb->tstamp = ktime_get_real();
3505 }
3506
3507 static inline ktime_t net_timedelta(ktime_t t)
3508 {
3509 return ktime_sub(ktime_get_real(), t);
3510 }
3511
3512 static inline ktime_t net_invalid_timestamp(void)
3513 {
3514 return 0;
3515 }
3516
3517 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3518 {
3519 return skb_shinfo(skb)->meta_len;
3520 }
3521
3522 static inline void *skb_metadata_end(const struct sk_buff *skb)
3523 {
3524 return skb_mac_header(skb);
3525 }
3526
3527 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3528 const struct sk_buff *skb_b,
3529 u8 meta_len)
3530 {
3531 const void *a = skb_metadata_end(skb_a);
3532 const void *b = skb_metadata_end(skb_b);
3533 /* Using more efficient varaiant than plain call to memcmp(). */
3534 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3535 u64 diffs = 0;
3536
3537 switch (meta_len) {
3538 #define __it(x, op) (x -= sizeof(u##op))
3539 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3540 case 32: diffs |= __it_diff(a, b, 64);
3541 /* fall through */
3542 case 24: diffs |= __it_diff(a, b, 64);
3543 /* fall through */
3544 case 16: diffs |= __it_diff(a, b, 64);
3545 /* fall through */
3546 case 8: diffs |= __it_diff(a, b, 64);
3547 break;
3548 case 28: diffs |= __it_diff(a, b, 64);
3549 /* fall through */
3550 case 20: diffs |= __it_diff(a, b, 64);
3551 /* fall through */
3552 case 12: diffs |= __it_diff(a, b, 64);
3553 /* fall through */
3554 case 4: diffs |= __it_diff(a, b, 32);
3555 break;
3556 }
3557 return diffs;
3558 #else
3559 return memcmp(a - meta_len, b - meta_len, meta_len);
3560 #endif
3561 }
3562
3563 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3564 const struct sk_buff *skb_b)
3565 {
3566 u8 len_a = skb_metadata_len(skb_a);
3567 u8 len_b = skb_metadata_len(skb_b);
3568
3569 if (!(len_a | len_b))
3570 return false;
3571
3572 return len_a != len_b ?
3573 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3574 }
3575
3576 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3577 {
3578 skb_shinfo(skb)->meta_len = meta_len;
3579 }
3580
3581 static inline void skb_metadata_clear(struct sk_buff *skb)
3582 {
3583 skb_metadata_set(skb, 0);
3584 }
3585
3586 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3587
3588 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3589
3590 void skb_clone_tx_timestamp(struct sk_buff *skb);
3591 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3592
3593 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3594
3595 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3596 {
3597 }
3598
3599 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3600 {
3601 return false;
3602 }
3603
3604 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3605
3606 /**
3607 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3608 *
3609 * PHY drivers may accept clones of transmitted packets for
3610 * timestamping via their phy_driver.txtstamp method. These drivers
3611 * must call this function to return the skb back to the stack with a
3612 * timestamp.
3613 *
3614 * @skb: clone of the the original outgoing packet
3615 * @hwtstamps: hardware time stamps
3616 *
3617 */
3618 void skb_complete_tx_timestamp(struct sk_buff *skb,
3619 struct skb_shared_hwtstamps *hwtstamps);
3620
3621 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3622 struct skb_shared_hwtstamps *hwtstamps,
3623 struct sock *sk, int tstype);
3624
3625 /**
3626 * skb_tstamp_tx - queue clone of skb with send time stamps
3627 * @orig_skb: the original outgoing packet
3628 * @hwtstamps: hardware time stamps, may be NULL if not available
3629 *
3630 * If the skb has a socket associated, then this function clones the
3631 * skb (thus sharing the actual data and optional structures), stores
3632 * the optional hardware time stamping information (if non NULL) or
3633 * generates a software time stamp (otherwise), then queues the clone
3634 * to the error queue of the socket. Errors are silently ignored.
3635 */
3636 void skb_tstamp_tx(struct sk_buff *orig_skb,
3637 struct skb_shared_hwtstamps *hwtstamps);
3638
3639 /**
3640 * skb_tx_timestamp() - Driver hook for transmit timestamping
3641 *
3642 * Ethernet MAC Drivers should call this function in their hard_xmit()
3643 * function immediately before giving the sk_buff to the MAC hardware.
3644 *
3645 * Specifically, one should make absolutely sure that this function is
3646 * called before TX completion of this packet can trigger. Otherwise
3647 * the packet could potentially already be freed.
3648 *
3649 * @skb: A socket buffer.
3650 */
3651 static inline void skb_tx_timestamp(struct sk_buff *skb)
3652 {
3653 skb_clone_tx_timestamp(skb);
3654 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3655 skb_tstamp_tx(skb, NULL);
3656 }
3657
3658 /**
3659 * skb_complete_wifi_ack - deliver skb with wifi status
3660 *
3661 * @skb: the original outgoing packet
3662 * @acked: ack status
3663 *
3664 */
3665 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3666
3667 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3668 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3669
3670 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3671 {
3672 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3673 skb->csum_valid ||
3674 (skb->ip_summed == CHECKSUM_PARTIAL &&
3675 skb_checksum_start_offset(skb) >= 0));
3676 }
3677
3678 /**
3679 * skb_checksum_complete - Calculate checksum of an entire packet
3680 * @skb: packet to process
3681 *
3682 * This function calculates the checksum over the entire packet plus
3683 * the value of skb->csum. The latter can be used to supply the
3684 * checksum of a pseudo header as used by TCP/UDP. It returns the
3685 * checksum.
3686 *
3687 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3688 * this function can be used to verify that checksum on received
3689 * packets. In that case the function should return zero if the
3690 * checksum is correct. In particular, this function will return zero
3691 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3692 * hardware has already verified the correctness of the checksum.
3693 */
3694 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3695 {
3696 return skb_csum_unnecessary(skb) ?
3697 0 : __skb_checksum_complete(skb);
3698 }
3699
3700 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3701 {
3702 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3703 if (skb->csum_level == 0)
3704 skb->ip_summed = CHECKSUM_NONE;
3705 else
3706 skb->csum_level--;
3707 }
3708 }
3709
3710 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3711 {
3712 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3713 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3714 skb->csum_level++;
3715 } else if (skb->ip_summed == CHECKSUM_NONE) {
3716 skb->ip_summed = CHECKSUM_UNNECESSARY;
3717 skb->csum_level = 0;
3718 }
3719 }
3720
3721 /* Check if we need to perform checksum complete validation.
3722 *
3723 * Returns true if checksum complete is needed, false otherwise
3724 * (either checksum is unnecessary or zero checksum is allowed).
3725 */
3726 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3727 bool zero_okay,
3728 __sum16 check)
3729 {
3730 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3731 skb->csum_valid = 1;
3732 __skb_decr_checksum_unnecessary(skb);
3733 return false;
3734 }
3735
3736 return true;
3737 }
3738
3739 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3740 * in checksum_init.
3741 */
3742 #define CHECKSUM_BREAK 76
3743
3744 /* Unset checksum-complete
3745 *
3746 * Unset checksum complete can be done when packet is being modified
3747 * (uncompressed for instance) and checksum-complete value is
3748 * invalidated.
3749 */
3750 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3751 {
3752 if (skb->ip_summed == CHECKSUM_COMPLETE)
3753 skb->ip_summed = CHECKSUM_NONE;
3754 }
3755
3756 /* Validate (init) checksum based on checksum complete.
3757 *
3758 * Return values:
3759 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3760 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3761 * checksum is stored in skb->csum for use in __skb_checksum_complete
3762 * non-zero: value of invalid checksum
3763 *
3764 */
3765 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3766 bool complete,
3767 __wsum psum)
3768 {
3769 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3770 if (!csum_fold(csum_add(psum, skb->csum))) {
3771 skb->csum_valid = 1;
3772 return 0;
3773 }
3774 }
3775
3776 skb->csum = psum;
3777
3778 if (complete || skb->len <= CHECKSUM_BREAK) {
3779 __sum16 csum;
3780
3781 csum = __skb_checksum_complete(skb);
3782 skb->csum_valid = !csum;
3783 return csum;
3784 }
3785
3786 return 0;
3787 }
3788
3789 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3790 {
3791 return 0;
3792 }
3793
3794 /* Perform checksum validate (init). Note that this is a macro since we only
3795 * want to calculate the pseudo header which is an input function if necessary.
3796 * First we try to validate without any computation (checksum unnecessary) and
3797 * then calculate based on checksum complete calling the function to compute
3798 * pseudo header.
3799 *
3800 * Return values:
3801 * 0: checksum is validated or try to in skb_checksum_complete
3802 * non-zero: value of invalid checksum
3803 */
3804 #define __skb_checksum_validate(skb, proto, complete, \
3805 zero_okay, check, compute_pseudo) \
3806 ({ \
3807 __sum16 __ret = 0; \
3808 skb->csum_valid = 0; \
3809 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3810 __ret = __skb_checksum_validate_complete(skb, \
3811 complete, compute_pseudo(skb, proto)); \
3812 __ret; \
3813 })
3814
3815 #define skb_checksum_init(skb, proto, compute_pseudo) \
3816 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3817
3818 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3819 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3820
3821 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3822 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3823
3824 #define skb_checksum_validate_zero_check(skb, proto, check, \
3825 compute_pseudo) \
3826 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3827
3828 #define skb_checksum_simple_validate(skb) \
3829 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3830
3831 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3832 {
3833 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3834 }
3835
3836 static inline void __skb_checksum_convert(struct sk_buff *skb,
3837 __sum16 check, __wsum pseudo)
3838 {
3839 skb->csum = ~pseudo;
3840 skb->ip_summed = CHECKSUM_COMPLETE;
3841 }
3842
3843 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3844 do { \
3845 if (__skb_checksum_convert_check(skb)) \
3846 __skb_checksum_convert(skb, check, \
3847 compute_pseudo(skb, proto)); \
3848 } while (0)
3849
3850 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3851 u16 start, u16 offset)
3852 {
3853 skb->ip_summed = CHECKSUM_PARTIAL;
3854 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3855 skb->csum_offset = offset - start;
3856 }
3857
3858 /* Update skbuf and packet to reflect the remote checksum offload operation.
3859 * When called, ptr indicates the starting point for skb->csum when
3860 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3861 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3862 */
3863 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3864 int start, int offset, bool nopartial)
3865 {
3866 __wsum delta;
3867
3868 if (!nopartial) {
3869 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3870 return;
3871 }
3872
3873 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3874 __skb_checksum_complete(skb);
3875 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3876 }
3877
3878 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3879
3880 /* Adjust skb->csum since we changed the packet */
3881 skb->csum = csum_add(skb->csum, delta);
3882 }
3883
3884 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3885 {
3886 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3887 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3888 #else
3889 return NULL;
3890 #endif
3891 }
3892
3893 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3894 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3895 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3896 {
3897 if (nfct && atomic_dec_and_test(&nfct->use))
3898 nf_conntrack_destroy(nfct);
3899 }
3900 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3901 {
3902 if (nfct)
3903 atomic_inc(&nfct->use);
3904 }
3905 #endif
3906
3907 #ifdef CONFIG_SKB_EXTENSIONS
3908 enum skb_ext_id {
3909 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3910 SKB_EXT_BRIDGE_NF,
3911 #endif
3912 #ifdef CONFIG_XFRM
3913 SKB_EXT_SEC_PATH,
3914 #endif
3915 SKB_EXT_NUM, /* must be last */
3916 };
3917
3918 /**
3919 * struct skb_ext - sk_buff extensions
3920 * @refcnt: 1 on allocation, deallocated on 0
3921 * @offset: offset to add to @data to obtain extension address
3922 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
3923 * @data: start of extension data, variable sized
3924 *
3925 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
3926 * to use 'u8' types while allowing up to 2kb worth of extension data.
3927 */
3928 struct skb_ext {
3929 refcount_t refcnt;
3930 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
3931 u8 chunks; /* same */
3932 char data[0] __aligned(8);
3933 };
3934
3935 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
3936 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
3937 void __skb_ext_put(struct skb_ext *ext);
3938
3939 static inline void skb_ext_put(struct sk_buff *skb)
3940 {
3941 if (skb->active_extensions)
3942 __skb_ext_put(skb->extensions);
3943 }
3944
3945 static inline void __skb_ext_copy(struct sk_buff *dst,
3946 const struct sk_buff *src)
3947 {
3948 dst->active_extensions = src->active_extensions;
3949
3950 if (src->active_extensions) {
3951 struct skb_ext *ext = src->extensions;
3952
3953 refcount_inc(&ext->refcnt);
3954 dst->extensions = ext;
3955 }
3956 }
3957
3958 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
3959 {
3960 skb_ext_put(dst);
3961 __skb_ext_copy(dst, src);
3962 }
3963
3964 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
3965 {
3966 return !!ext->offset[i];
3967 }
3968
3969 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
3970 {
3971 return skb->active_extensions & (1 << id);
3972 }
3973
3974 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
3975 {
3976 if (skb_ext_exist(skb, id))
3977 __skb_ext_del(skb, id);
3978 }
3979
3980 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
3981 {
3982 if (skb_ext_exist(skb, id)) {
3983 struct skb_ext *ext = skb->extensions;
3984
3985 return (void *)ext + (ext->offset[id] << 3);
3986 }
3987
3988 return NULL;
3989 }
3990 #else
3991 static inline void skb_ext_put(struct sk_buff *skb) {}
3992 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
3993 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
3994 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
3995 #endif /* CONFIG_SKB_EXTENSIONS */
3996
3997 static inline void nf_reset(struct sk_buff *skb)
3998 {
3999 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4000 nf_conntrack_put(skb_nfct(skb));
4001 skb->_nfct = 0;
4002 #endif
4003 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4004 skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
4005 #endif
4006 }
4007
4008 static inline void nf_reset_trace(struct sk_buff *skb)
4009 {
4010 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4011 skb->nf_trace = 0;
4012 #endif
4013 }
4014
4015 static inline void ipvs_reset(struct sk_buff *skb)
4016 {
4017 #if IS_ENABLED(CONFIG_IP_VS)
4018 skb->ipvs_property = 0;
4019 #endif
4020 }
4021
4022 /* Note: This doesn't put any conntrack info in dst. */
4023 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4024 bool copy)
4025 {
4026 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4027 dst->_nfct = src->_nfct;
4028 nf_conntrack_get(skb_nfct(src));
4029 #endif
4030 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4031 if (copy)
4032 dst->nf_trace = src->nf_trace;
4033 #endif
4034 }
4035
4036 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4037 {
4038 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4039 nf_conntrack_put(skb_nfct(dst));
4040 #endif
4041 __nf_copy(dst, src, true);
4042 }
4043
4044 #ifdef CONFIG_NETWORK_SECMARK
4045 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4046 {
4047 to->secmark = from->secmark;
4048 }
4049
4050 static inline void skb_init_secmark(struct sk_buff *skb)
4051 {
4052 skb->secmark = 0;
4053 }
4054 #else
4055 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4056 { }
4057
4058 static inline void skb_init_secmark(struct sk_buff *skb)
4059 { }
4060 #endif
4061
4062 static inline int secpath_exists(const struct sk_buff *skb)
4063 {
4064 #ifdef CONFIG_XFRM
4065 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4066 #else
4067 return 0;
4068 #endif
4069 }
4070
4071 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4072 {
4073 return !skb->destructor &&
4074 !secpath_exists(skb) &&
4075 !skb_nfct(skb) &&
4076 !skb->_skb_refdst &&
4077 !skb_has_frag_list(skb);
4078 }
4079
4080 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4081 {
4082 skb->queue_mapping = queue_mapping;
4083 }
4084
4085 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4086 {
4087 return skb->queue_mapping;
4088 }
4089
4090 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4091 {
4092 to->queue_mapping = from->queue_mapping;
4093 }
4094
4095 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4096 {
4097 skb->queue_mapping = rx_queue + 1;
4098 }
4099
4100 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4101 {
4102 return skb->queue_mapping - 1;
4103 }
4104
4105 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4106 {
4107 return skb->queue_mapping != 0;
4108 }
4109
4110 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4111 {
4112 skb->dst_pending_confirm = val;
4113 }
4114
4115 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4116 {
4117 return skb->dst_pending_confirm != 0;
4118 }
4119
4120 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4121 {
4122 #ifdef CONFIG_XFRM
4123 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4124 #else
4125 return NULL;
4126 #endif
4127 }
4128
4129 /* Keeps track of mac header offset relative to skb->head.
4130 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4131 * For non-tunnel skb it points to skb_mac_header() and for
4132 * tunnel skb it points to outer mac header.
4133 * Keeps track of level of encapsulation of network headers.
4134 */
4135 struct skb_gso_cb {
4136 union {
4137 int mac_offset;
4138 int data_offset;
4139 };
4140 int encap_level;
4141 __wsum csum;
4142 __u16 csum_start;
4143 };
4144 #define SKB_SGO_CB_OFFSET 32
4145 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4146
4147 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4148 {
4149 return (skb_mac_header(inner_skb) - inner_skb->head) -
4150 SKB_GSO_CB(inner_skb)->mac_offset;
4151 }
4152
4153 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4154 {
4155 int new_headroom, headroom;
4156 int ret;
4157
4158 headroom = skb_headroom(skb);
4159 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4160 if (ret)
4161 return ret;
4162
4163 new_headroom = skb_headroom(skb);
4164 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4165 return 0;
4166 }
4167
4168 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4169 {
4170 /* Do not update partial checksums if remote checksum is enabled. */
4171 if (skb->remcsum_offload)
4172 return;
4173
4174 SKB_GSO_CB(skb)->csum = res;
4175 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4176 }
4177
4178 /* Compute the checksum for a gso segment. First compute the checksum value
4179 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4180 * then add in skb->csum (checksum from csum_start to end of packet).
4181 * skb->csum and csum_start are then updated to reflect the checksum of the
4182 * resultant packet starting from the transport header-- the resultant checksum
4183 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4184 * header.
4185 */
4186 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4187 {
4188 unsigned char *csum_start = skb_transport_header(skb);
4189 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4190 __wsum partial = SKB_GSO_CB(skb)->csum;
4191
4192 SKB_GSO_CB(skb)->csum = res;
4193 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4194
4195 return csum_fold(csum_partial(csum_start, plen, partial));
4196 }
4197
4198 static inline bool skb_is_gso(const struct sk_buff *skb)
4199 {
4200 return skb_shinfo(skb)->gso_size;
4201 }
4202
4203 /* Note: Should be called only if skb_is_gso(skb) is true */
4204 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4205 {
4206 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4207 }
4208
4209 /* Note: Should be called only if skb_is_gso(skb) is true */
4210 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4211 {
4212 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4213 }
4214
4215 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4216 {
4217 return skb_is_gso(skb) &&
4218 skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4219 }
4220
4221 static inline void skb_gso_reset(struct sk_buff *skb)
4222 {
4223 skb_shinfo(skb)->gso_size = 0;
4224 skb_shinfo(skb)->gso_segs = 0;
4225 skb_shinfo(skb)->gso_type = 0;
4226 }
4227
4228 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4229 u16 increment)
4230 {
4231 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4232 return;
4233 shinfo->gso_size += increment;
4234 }
4235
4236 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4237 u16 decrement)
4238 {
4239 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4240 return;
4241 shinfo->gso_size -= decrement;
4242 }
4243
4244 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4245
4246 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4247 {
4248 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4249 * wanted then gso_type will be set. */
4250 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4251
4252 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4253 unlikely(shinfo->gso_type == 0)) {
4254 __skb_warn_lro_forwarding(skb);
4255 return true;
4256 }
4257 return false;
4258 }
4259
4260 static inline void skb_forward_csum(struct sk_buff *skb)
4261 {
4262 /* Unfortunately we don't support this one. Any brave souls? */
4263 if (skb->ip_summed == CHECKSUM_COMPLETE)
4264 skb->ip_summed = CHECKSUM_NONE;
4265 }
4266
4267 /**
4268 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4269 * @skb: skb to check
4270 *
4271 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4272 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4273 * use this helper, to document places where we make this assertion.
4274 */
4275 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4276 {
4277 #ifdef DEBUG
4278 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4279 #endif
4280 }
4281
4282 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4283
4284 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4285 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4286 unsigned int transport_len,
4287 __sum16(*skb_chkf)(struct sk_buff *skb));
4288
4289 /**
4290 * skb_head_is_locked - Determine if the skb->head is locked down
4291 * @skb: skb to check
4292 *
4293 * The head on skbs build around a head frag can be removed if they are
4294 * not cloned. This function returns true if the skb head is locked down
4295 * due to either being allocated via kmalloc, or by being a clone with
4296 * multiple references to the head.
4297 */
4298 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4299 {
4300 return !skb->head_frag || skb_cloned(skb);
4301 }
4302
4303 /* Local Checksum Offload.
4304 * Compute outer checksum based on the assumption that the
4305 * inner checksum will be offloaded later.
4306 * See Documentation/networking/checksum-offloads.txt for
4307 * explanation of how this works.
4308 * Fill in outer checksum adjustment (e.g. with sum of outer
4309 * pseudo-header) before calling.
4310 * Also ensure that inner checksum is in linear data area.
4311 */
4312 static inline __wsum lco_csum(struct sk_buff *skb)
4313 {
4314 unsigned char *csum_start = skb_checksum_start(skb);
4315 unsigned char *l4_hdr = skb_transport_header(skb);
4316 __wsum partial;
4317
4318 /* Start with complement of inner checksum adjustment */
4319 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4320 skb->csum_offset));
4321
4322 /* Add in checksum of our headers (incl. outer checksum
4323 * adjustment filled in by caller) and return result.
4324 */
4325 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4326 }
4327
4328 #endif /* __KERNEL__ */
4329 #endif /* _LINUX_SKBUFF_H */