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
241struct net_device;
242struct scatterlist;
243struct pipe_inode_info;
244struct iov_iter;
245struct napi_struct;
246struct bpf_prog;
247union bpf_attr;
248struct skb_ext;
249
250#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
251struct nf_conntrack {
252 atomic_t use;
253};
254#endif
255
256#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
257struct 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
285struct 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
294struct 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
308extern 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
315typedef struct skb_frag_struct skb_frag_t;
316
317struct 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/**
331 * skb_frag_size - Returns the size of a skb fragment
332 * @frag: skb fragment
333 */
334static inline unsigned int skb_frag_size(const skb_frag_t *frag)
335{
336 return frag->size;
337}
338
339/**
340 * skb_frag_size_set - Sets the size of a skb fragment
341 * @frag: skb fragment
342 * @size: size of fragment
343 */
344static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
345{
346 frag->size = size;
347}
348
349/**
350 * skb_frag_size_add - Incrementes the size of a skb fragment by %delta
351 * @frag: skb fragment
352 * @delta: value to add
353 */
354static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
355{
356 frag->size += delta;
357}
358
359/**
360 * skb_frag_size_sub - Decrements the size of a skb fragment by %delta
361 * @frag: skb fragment
362 * @delta: value to subtract
363 */
364static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
365{
366 frag->size -= delta;
367}
368
369/**
370 * skb_frag_must_loop - Test if %p is a high memory page
371 * @p: fragment's page
372 */
373static inline bool skb_frag_must_loop(struct page *p)
374{
375#if defined(CONFIG_HIGHMEM)
376 if (PageHighMem(p))
377 return true;
378#endif
379 return false;
380}
381
382/**
383 * skb_frag_foreach_page - loop over pages in a fragment
384 *
385 * @f: skb frag to operate on
386 * @f_off: offset from start of f->page.p
387 * @f_len: length from f_off to loop over
388 * @p: (temp var) current page
389 * @p_off: (temp var) offset from start of current page,
390 * non-zero only on first page.
391 * @p_len: (temp var) length in current page,
392 * < PAGE_SIZE only on first and last page.
393 * @copied: (temp var) length so far, excluding current p_len.
394 *
395 * A fragment can hold a compound page, in which case per-page
396 * operations, notably kmap_atomic, must be called for each
397 * regular page.
398 */
399#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
400 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
401 p_off = (f_off) & (PAGE_SIZE - 1), \
402 p_len = skb_frag_must_loop(p) ? \
403 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
404 copied = 0; \
405 copied < f_len; \
406 copied += p_len, p++, p_off = 0, \
407 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
408
409#define HAVE_HW_TIME_STAMP
410
411/**
412 * struct skb_shared_hwtstamps - hardware time stamps
413 * @hwtstamp: hardware time stamp transformed into duration
414 * since arbitrary point in time
415 *
416 * Software time stamps generated by ktime_get_real() are stored in
417 * skb->tstamp.
418 *
419 * hwtstamps can only be compared against other hwtstamps from
420 * the same device.
421 *
422 * This structure is attached to packets as part of the
423 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
424 */
425struct skb_shared_hwtstamps {
426 ktime_t hwtstamp;
427};
428
429/* Definitions for tx_flags in struct skb_shared_info */
430enum {
431 /* generate hardware time stamp */
432 SKBTX_HW_TSTAMP = 1 << 0,
433
434 /* generate software time stamp when queueing packet to NIC */
435 SKBTX_SW_TSTAMP = 1 << 1,
436
437 /* device driver is going to provide hardware time stamp */
438 SKBTX_IN_PROGRESS = 1 << 2,
439
440 /* device driver supports TX zero-copy buffers */
441 SKBTX_DEV_ZEROCOPY = 1 << 3,
442
443 /* generate wifi status information (where possible) */
444 SKBTX_WIFI_STATUS = 1 << 4,
445
446 /* This indicates at least one fragment might be overwritten
447 * (as in vmsplice(), sendfile() ...)
448 * If we need to compute a TX checksum, we'll need to copy
449 * all frags to avoid possible bad checksum
450 */
451 SKBTX_SHARED_FRAG = 1 << 5,
452
453 /* generate software time stamp when entering packet scheduling */
454 SKBTX_SCHED_TSTAMP = 1 << 6,
455};
456
457#define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
458#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
459 SKBTX_SCHED_TSTAMP)
460#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
461
462/*
463 * The callback notifies userspace to release buffers when skb DMA is done in
464 * lower device, the skb last reference should be 0 when calling this.
465 * The zerocopy_success argument is true if zero copy transmit occurred,
466 * false on data copy or out of memory error caused by data copy attempt.
467 * The ctx field is used to track device context.
468 * The desc field is used to track userspace buffer index.
469 */
470struct ubuf_info {
471 void (*callback)(struct ubuf_info *, bool zerocopy_success);
472 union {
473 struct {
474 unsigned long desc;
475 void *ctx;
476 };
477 struct {
478 u32 id;
479 u16 len;
480 u16 zerocopy:1;
481 u32 bytelen;
482 };
483 };
484 refcount_t refcnt;
485
486 struct mmpin {
487 struct user_struct *user;
488 unsigned int num_pg;
489 } mmp;
490};
491
492#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
493
494int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
495void mm_unaccount_pinned_pages(struct mmpin *mmp);
496
497struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
498struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
499 struct ubuf_info *uarg);
500
501static inline void sock_zerocopy_get(struct ubuf_info *uarg)
502{
503 refcount_inc(&uarg->refcnt);
504}
505
506void sock_zerocopy_put(struct ubuf_info *uarg);
507void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
508
509void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
510
511int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
512int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
513 struct msghdr *msg, int len,
514 struct ubuf_info *uarg);
515
516/* This data is invariant across clones and lives at
517 * the end of the header data, ie. at skb->end.
518 */
519struct skb_shared_info {
520 __u8 __unused;
521 __u8 meta_len;
522 __u8 nr_frags;
523 __u8 tx_flags;
524 unsigned short gso_size;
525 /* Warning: this field is not always filled in (UFO)! */
526 unsigned short gso_segs;
527 struct sk_buff *frag_list;
528 struct skb_shared_hwtstamps hwtstamps;
529 unsigned int gso_type;
530 u32 tskey;
531
532 /*
533 * Warning : all fields before dataref are cleared in __alloc_skb()
534 */
535 atomic_t dataref;
536
537 /* Intermediate layers must ensure that destructor_arg
538 * remains valid until skb destructor */
539 void * destructor_arg;
540
541 /* must be last field, see pskb_expand_head() */
542 skb_frag_t frags[MAX_SKB_FRAGS];
543};
544
545/* We divide dataref into two halves. The higher 16 bits hold references
546 * to the payload part of skb->data. The lower 16 bits hold references to
547 * the entire skb->data. A clone of a headerless skb holds the length of
548 * the header in skb->hdr_len.
549 *
550 * All users must obey the rule that the skb->data reference count must be
551 * greater than or equal to the payload reference count.
552 *
553 * Holding a reference to the payload part means that the user does not
554 * care about modifications to the header part of skb->data.
555 */
556#define SKB_DATAREF_SHIFT 16
557#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
558
559
560enum {
561 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
562 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
563 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
564};
565
566enum {
567 SKB_GSO_TCPV4 = 1 << 0,
568
569 /* This indicates the skb is from an untrusted source. */
570 SKB_GSO_DODGY = 1 << 1,
571
572 /* This indicates the tcp segment has CWR set. */
573 SKB_GSO_TCP_ECN = 1 << 2,
574
575 SKB_GSO_TCP_FIXEDID = 1 << 3,
576
577 SKB_GSO_TCPV6 = 1 << 4,
578
579 SKB_GSO_FCOE = 1 << 5,
580
581 SKB_GSO_GRE = 1 << 6,
582
583 SKB_GSO_GRE_CSUM = 1 << 7,
584
585 SKB_GSO_IPXIP4 = 1 << 8,
586
587 SKB_GSO_IPXIP6 = 1 << 9,
588
589 SKB_GSO_UDP_TUNNEL = 1 << 10,
590
591 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
592
593 SKB_GSO_PARTIAL = 1 << 12,
594
595 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
596
597 SKB_GSO_SCTP = 1 << 14,
598
599 SKB_GSO_ESP = 1 << 15,
600
601 SKB_GSO_UDP = 1 << 16,
602
603 SKB_GSO_UDP_L4 = 1 << 17,
604};
605
606#if BITS_PER_LONG > 32
607#define NET_SKBUFF_DATA_USES_OFFSET 1
608#endif
609
610#ifdef NET_SKBUFF_DATA_USES_OFFSET
611typedef unsigned int sk_buff_data_t;
612#else
613typedef unsigned char *sk_buff_data_t;
614#endif
615
616/**
617 * struct sk_buff - socket buffer
618 * @next: Next buffer in list
619 * @prev: Previous buffer in list
620 * @tstamp: Time we arrived/left
621 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
622 * @sk: Socket we are owned by
623 * @dev: Device we arrived on/are leaving by
624 * @cb: Control buffer. Free for use by every layer. Put private vars here
625 * @_skb_refdst: destination entry (with norefcount bit)
626 * @sp: the security path, used for xfrm
627 * @len: Length of actual data
628 * @data_len: Data length
629 * @mac_len: Length of link layer header
630 * @hdr_len: writable header length of cloned skb
631 * @csum: Checksum (must include start/offset pair)
632 * @csum_start: Offset from skb->head where checksumming should start
633 * @csum_offset: Offset from csum_start where checksum should be stored
634 * @priority: Packet queueing priority
635 * @ignore_df: allow local fragmentation
636 * @cloned: Head may be cloned (check refcnt to be sure)
637 * @ip_summed: Driver fed us an IP checksum
638 * @nohdr: Payload reference only, must not modify header
639 * @pkt_type: Packet class
640 * @fclone: skbuff clone status
641 * @ipvs_property: skbuff is owned by ipvs
642 * @offload_fwd_mark: Packet was L2-forwarded in hardware
643 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
644 * @tc_skip_classify: do not classify packet. set by IFB device
645 * @tc_at_ingress: used within tc_classify to distinguish in/egress
646 * @tc_redirected: packet was redirected by a tc action
647 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
648 * @peeked: this packet has been seen already, so stats have been
649 * done for it, don't do them again
650 * @nf_trace: netfilter packet trace flag
651 * @protocol: Packet protocol from driver
652 * @destructor: Destruct function
653 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
654 * @_nfct: Associated connection, if any (with nfctinfo bits)
655 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
656 * @skb_iif: ifindex of device we arrived on
657 * @tc_index: Traffic control index
658 * @hash: the packet hash
659 * @queue_mapping: Queue mapping for multiqueue devices
660 * @xmit_more: More SKBs are pending for this queue
661 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
662 * @active_extensions: active extensions (skb_ext_id types)
663 * @ndisc_nodetype: router type (from link layer)
664 * @ooo_okay: allow the mapping of a socket to a queue to be changed
665 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
666 * ports.
667 * @sw_hash: indicates hash was computed in software stack
668 * @wifi_acked_valid: wifi_acked was set
669 * @wifi_acked: whether frame was acked on wifi or not
670 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
671 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
672 * @dst_pending_confirm: need to confirm neighbour
673 * @decrypted: Decrypted SKB
674 * @napi_id: id of the NAPI struct this skb came from
675 * @secmark: security marking
676 * @mark: Generic packet mark
677 * @vlan_proto: vlan encapsulation protocol
678 * @vlan_tci: vlan tag control information
679 * @inner_protocol: Protocol (encapsulation)
680 * @inner_transport_header: Inner transport layer header (encapsulation)
681 * @inner_network_header: Network layer header (encapsulation)
682 * @inner_mac_header: Link layer header (encapsulation)
683 * @transport_header: Transport layer header
684 * @network_header: Network layer header
685 * @mac_header: Link layer header
686 * @tail: Tail pointer
687 * @end: End pointer
688 * @head: Head of buffer
689 * @data: Data head pointer
690 * @truesize: Buffer size
691 * @users: User count - see {datagram,tcp}.c
692 * @extensions: allocated extensions, valid if active_extensions is nonzero
693 */
694
695struct sk_buff {
696 union {
697 struct {
698 /* These two members must be first. */
699 struct sk_buff *next;
700 struct sk_buff *prev;
701
702 union {
703 struct net_device *dev;
704 /* Some protocols might use this space to store information,
705 * while device pointer would be NULL.
706 * UDP receive path is one user.
707 */
708 unsigned long dev_scratch;
709 };
710 };
711 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
712 struct list_head list;
713 };
714
715 union {
716 struct sock *sk;
717 int ip_defrag_offset;
718 };
719
720 union {
721 ktime_t tstamp;
722 u64 skb_mstamp_ns; /* earliest departure time */
723 };
724 /*
725 * This is the control buffer. It is free to use for every
726 * layer. Please put your private variables there. If you
727 * want to keep them across layers you have to do a skb_clone()
728 * first. This is owned by whoever has the skb queued ATM.
729 */
730 char cb[48] __aligned(8);
731
732 union {
733 struct {
734 unsigned long _skb_refdst;
735 void (*destructor)(struct sk_buff *skb);
736 };
737 struct list_head tcp_tsorted_anchor;
738 };
739
740#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
741 unsigned long _nfct;
742#endif
743 unsigned int len,
744 data_len;
745 __u16 mac_len,
746 hdr_len;
747
748 /* Following fields are _not_ copied in __copy_skb_header()
749 * Note that queue_mapping is here mostly to fill a hole.
750 */
751 __u16 queue_mapping;
752
753/* if you move cloned around you also must adapt those constants */
754#ifdef __BIG_ENDIAN_BITFIELD
755#define CLONED_MASK (1 << 7)
756#else
757#define CLONED_MASK 1
758#endif
759#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
760
761 __u8 __cloned_offset[0];
762 __u8 cloned:1,
763 nohdr:1,
764 fclone:2,
765 peeked:1,
766 head_frag:1,
767 xmit_more:1,
768 pfmemalloc:1;
769#ifdef CONFIG_SKB_EXTENSIONS
770 __u8 active_extensions;
771#endif
772 /* fields enclosed in headers_start/headers_end are copied
773 * using a single memcpy() in __copy_skb_header()
774 */
775 /* private: */
776 __u32 headers_start[0];
777 /* public: */
778
779/* if you move pkt_type around you also must adapt those constants */
780#ifdef __BIG_ENDIAN_BITFIELD
781#define PKT_TYPE_MAX (7 << 5)
782#else
783#define PKT_TYPE_MAX 7
784#endif
785#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
786
787 __u8 __pkt_type_offset[0];
788 __u8 pkt_type:3;
789 __u8 ignore_df:1;
790 __u8 nf_trace:1;
791 __u8 ip_summed:2;
792 __u8 ooo_okay:1;
793
794 __u8 l4_hash:1;
795 __u8 sw_hash:1;
796 __u8 wifi_acked_valid:1;
797 __u8 wifi_acked:1;
798 __u8 no_fcs:1;
799 /* Indicates the inner headers are valid in the skbuff. */
800 __u8 encapsulation:1;
801 __u8 encap_hdr_csum:1;
802 __u8 csum_valid:1;
803
804#ifdef __BIG_ENDIAN_BITFIELD
805#define PKT_VLAN_PRESENT_BIT 7
806#else
807#define PKT_VLAN_PRESENT_BIT 0
808#endif
809#define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
810 __u8 __pkt_vlan_present_offset[0];
811 __u8 vlan_present:1;
812 __u8 csum_complete_sw:1;
813 __u8 csum_level:2;
814 __u8 csum_not_inet:1;
815 __u8 dst_pending_confirm:1;
816#ifdef CONFIG_IPV6_NDISC_NODETYPE
817 __u8 ndisc_nodetype:2;
818#endif
819
820 __u8 ipvs_property:1;
821 __u8 inner_protocol_type:1;
822 __u8 remcsum_offload:1;
823#ifdef CONFIG_NET_SWITCHDEV
824 __u8 offload_fwd_mark:1;
825 __u8 offload_l3_fwd_mark:1;
826#endif
827#ifdef CONFIG_NET_CLS_ACT
828 __u8 tc_skip_classify:1;
829 __u8 tc_at_ingress:1;
830 __u8 tc_redirected:1;
831 __u8 tc_from_ingress:1;
832#endif
833#ifdef CONFIG_TLS_DEVICE
834 __u8 decrypted:1;
835#endif
836
837#ifdef CONFIG_NET_SCHED
838 __u16 tc_index; /* traffic control index */
839#endif
840
841 union {
842 __wsum csum;
843 struct {
844 __u16 csum_start;
845 __u16 csum_offset;
846 };
847 };
848 __u32 priority;
849 int skb_iif;
850 __u32 hash;
851 __be16 vlan_proto;
852 __u16 vlan_tci;
853#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
854 union {
855 unsigned int napi_id;
856 unsigned int sender_cpu;
857 };
858#endif
859#ifdef CONFIG_NETWORK_SECMARK
860 __u32 secmark;
861#endif
862
863 union {
864 __u32 mark;
865 __u32 reserved_tailroom;
866 };
867
868 union {
869 __be16 inner_protocol;
870 __u8 inner_ipproto;
871 };
872
873 __u16 inner_transport_header;
874 __u16 inner_network_header;
875 __u16 inner_mac_header;
876
877 __be16 protocol;
878 __u16 transport_header;
879 __u16 network_header;
880 __u16 mac_header;
881
882 /* private: */
883 __u32 headers_end[0];
884 /* public: */
885
886 /* These elements must be at the end, see alloc_skb() for details. */
887 sk_buff_data_t tail;
888 sk_buff_data_t end;
889 unsigned char *head,
890 *data;
891 unsigned int truesize;
892 refcount_t users;
893
894#ifdef CONFIG_SKB_EXTENSIONS
895 /* only useable after checking ->active_extensions != 0 */
896 struct skb_ext *extensions;
897#endif
898};
899
900#ifdef __KERNEL__
901/*
902 * Handling routines are only of interest to the kernel
903 */
904
905#define SKB_ALLOC_FCLONE 0x01
906#define SKB_ALLOC_RX 0x02
907#define SKB_ALLOC_NAPI 0x04
908
909/**
910 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
911 * @skb: buffer
912 */
913static inline bool skb_pfmemalloc(const struct sk_buff *skb)
914{
915 return unlikely(skb->pfmemalloc);
916}
917
918/*
919 * skb might have a dst pointer attached, refcounted or not.
920 * _skb_refdst low order bit is set if refcount was _not_ taken
921 */
922#define SKB_DST_NOREF 1UL
923#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
924
925#define SKB_NFCT_PTRMASK ~(7UL)
926/**
927 * skb_dst - returns skb dst_entry
928 * @skb: buffer
929 *
930 * Returns skb dst_entry, regardless of reference taken or not.
931 */
932static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
933{
934 /* If refdst was not refcounted, check we still are in a
935 * rcu_read_lock section
936 */
937 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
938 !rcu_read_lock_held() &&
939 !rcu_read_lock_bh_held());
940 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
941}
942
943/**
944 * skb_dst_set - sets skb dst
945 * @skb: buffer
946 * @dst: dst entry
947 *
948 * Sets skb dst, assuming a reference was taken on dst and should
949 * be released by skb_dst_drop()
950 */
951static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
952{
953 skb->_skb_refdst = (unsigned long)dst;
954}
955
956/**
957 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
958 * @skb: buffer
959 * @dst: dst entry
960 *
961 * Sets skb dst, assuming a reference was not taken on dst.
962 * If dst entry is cached, we do not take reference and dst_release
963 * will be avoided by refdst_drop. If dst entry is not cached, we take
964 * reference, so that last dst_release can destroy the dst immediately.
965 */
966static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
967{
968 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
969 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
970}
971
972/**
973 * skb_dst_is_noref - Test if skb dst isn't refcounted
974 * @skb: buffer
975 */
976static inline bool skb_dst_is_noref(const struct sk_buff *skb)
977{
978 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
979}
980
981/**
982 * skb_rtable - Returns the skb &rtable
983 * @skb: buffer
984 */
985static inline struct rtable *skb_rtable(const struct sk_buff *skb)
986{
987 return (struct rtable *)skb_dst(skb);
988}
989
990/* For mangling skb->pkt_type from user space side from applications
991 * such as nft, tc, etc, we only allow a conservative subset of
992 * possible pkt_types to be set.
993*/
994static inline bool skb_pkt_type_ok(u32 ptype)
995{
996 return ptype <= PACKET_OTHERHOST;
997}
998
999/**
1000 * skb_napi_id - Returns the skb's NAPI id
1001 * @skb: buffer
1002 */
1003static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1004{
1005#ifdef CONFIG_NET_RX_BUSY_POLL
1006 return skb->napi_id;
1007#else
1008 return 0;
1009#endif
1010}
1011
1012/**
1013 * skb_unref - decrement the skb's reference count
1014 * @skb: buffer
1015 *
1016 * Returns true if we can free the skb.
1017 */
1018static inline bool skb_unref(struct sk_buff *skb)
1019{
1020 if (unlikely(!skb))
1021 return false;
1022 if (likely(refcount_read(&skb->users) == 1))
1023 smp_rmb();
1024 else if (likely(!refcount_dec_and_test(&skb->users)))
1025 return false;
1026
1027 return true;
1028}
1029
1030void skb_release_head_state(struct sk_buff *skb);
1031void kfree_skb(struct sk_buff *skb);
1032void kfree_skb_list(struct sk_buff *segs);
1033void skb_tx_error(struct sk_buff *skb);
1034void consume_skb(struct sk_buff *skb);
1035void __consume_stateless_skb(struct sk_buff *skb);
1036void __kfree_skb(struct sk_buff *skb);
1037extern struct kmem_cache *skbuff_head_cache;
1038
1039void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1040bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1041 bool *fragstolen, int *delta_truesize);
1042
1043struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1044 int node);
1045struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1046struct sk_buff *build_skb(void *data, unsigned int frag_size);
1047
1048/**
1049 * alloc_skb - allocate a network buffer
1050 * @size: size to allocate
1051 * @priority: allocation mask
1052 *
1053 * This function is a convenient wrapper around __alloc_skb().
1054 */
1055static inline struct sk_buff *alloc_skb(unsigned int size,
1056 gfp_t priority)
1057{
1058 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1059}
1060
1061struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1062 unsigned long data_len,
1063 int max_page_order,
1064 int *errcode,
1065 gfp_t gfp_mask);
1066
1067/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1068struct sk_buff_fclones {
1069 struct sk_buff skb1;
1070
1071 struct sk_buff skb2;
1072
1073 refcount_t fclone_ref;
1074};
1075
1076/**
1077 * skb_fclone_busy - check if fclone is busy
1078 * @sk: socket
1079 * @skb: buffer
1080 *
1081 * Returns true if skb is a fast clone, and its clone is not freed.
1082 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1083 * so we also check that this didnt happen.
1084 */
1085static inline bool skb_fclone_busy(const struct sock *sk,
1086 const struct sk_buff *skb)
1087{
1088 const struct sk_buff_fclones *fclones;
1089
1090 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1091
1092 return skb->fclone == SKB_FCLONE_ORIG &&
1093 refcount_read(&fclones->fclone_ref) > 1 &&
1094 fclones->skb2.sk == sk;
1095}
1096
1097/**
1098 * alloc_skb_fclone - allocate a network buffer from fclone cache
1099 * @size: size to allocate
1100 * @priority: allocation mask
1101 *
1102 * This function is a convenient wrapper around __alloc_skb().
1103 */
1104static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1105 gfp_t priority)
1106{
1107 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1108}
1109
1110struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1111void skb_headers_offset_update(struct sk_buff *skb, int off);
1112int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1113struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1114void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1115struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1116struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1117 gfp_t gfp_mask, bool fclone);
1118static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1119 gfp_t gfp_mask)
1120{
1121 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1122}
1123
1124int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1125struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1126 unsigned int headroom);
1127struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1128 int newtailroom, gfp_t priority);
1129int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1130 int offset, int len);
1131int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1132 int offset, int len);
1133int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1134int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1135
1136/**
1137 * skb_pad - zero pad the tail of an skb
1138 * @skb: buffer to pad
1139 * @pad: space to pad
1140 *
1141 * Ensure that a buffer is followed by a padding area that is zero
1142 * filled. Used by network drivers which may DMA or transfer data
1143 * beyond the buffer end onto the wire.
1144 *
1145 * May return error in out of memory cases. The skb is freed on error.
1146 */
1147static inline int skb_pad(struct sk_buff *skb, int pad)
1148{
1149 return __skb_pad(skb, pad, true);
1150}
1151#define dev_kfree_skb(a) consume_skb(a)
1152
1153int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1154 int offset, size_t size);
1155
1156struct skb_seq_state {
1157 __u32 lower_offset;
1158 __u32 upper_offset;
1159 __u32 frag_idx;
1160 __u32 stepped_offset;
1161 struct sk_buff *root_skb;
1162 struct sk_buff *cur_skb;
1163 __u8 *frag_data;
1164};
1165
1166void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1167 unsigned int to, struct skb_seq_state *st);
1168unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1169 struct skb_seq_state *st);
1170void skb_abort_seq_read(struct skb_seq_state *st);
1171
1172unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1173 unsigned int to, struct ts_config *config);
1174
1175/*
1176 * Packet hash types specify the type of hash in skb_set_hash.
1177 *
1178 * Hash types refer to the protocol layer addresses which are used to
1179 * construct a packet's hash. The hashes are used to differentiate or identify
1180 * flows of the protocol layer for the hash type. Hash types are either
1181 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1182 *
1183 * Properties of hashes:
1184 *
1185 * 1) Two packets in different flows have different hash values
1186 * 2) Two packets in the same flow should have the same hash value
1187 *
1188 * A hash at a higher layer is considered to be more specific. A driver should
1189 * set the most specific hash possible.
1190 *
1191 * A driver cannot indicate a more specific hash than the layer at which a hash
1192 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1193 *
1194 * A driver may indicate a hash level which is less specific than the
1195 * actual layer the hash was computed on. For instance, a hash computed
1196 * at L4 may be considered an L3 hash. This should only be done if the
1197 * driver can't unambiguously determine that the HW computed the hash at
1198 * the higher layer. Note that the "should" in the second property above
1199 * permits this.
1200 */
1201enum pkt_hash_types {
1202 PKT_HASH_TYPE_NONE, /* Undefined type */
1203 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1204 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1205 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1206};
1207
1208static inline void skb_clear_hash(struct sk_buff *skb)
1209{
1210 skb->hash = 0;
1211 skb->sw_hash = 0;
1212 skb->l4_hash = 0;
1213}
1214
1215static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1216{
1217 if (!skb->l4_hash)
1218 skb_clear_hash(skb);
1219}
1220
1221static inline void
1222__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1223{
1224 skb->l4_hash = is_l4;
1225 skb->sw_hash = is_sw;
1226 skb->hash = hash;
1227}
1228
1229static inline void
1230skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1231{
1232 /* Used by drivers to set hash from HW */
1233 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1234}
1235
1236static inline void
1237__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1238{
1239 __skb_set_hash(skb, hash, true, is_l4);
1240}
1241
1242void __skb_get_hash(struct sk_buff *skb);
1243u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1244u32 skb_get_poff(const struct sk_buff *skb);
1245u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1246 const struct flow_keys_basic *keys, int hlen);
1247__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1248 void *data, int hlen_proto);
1249
1250static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1251 int thoff, u8 ip_proto)
1252{
1253 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1254}
1255
1256void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1257 const struct flow_dissector_key *key,
1258 unsigned int key_count);
1259
1260#ifdef CONFIG_NET
1261int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1262 struct bpf_prog *prog);
1263
1264int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1265#else
1266static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1267 struct bpf_prog *prog)
1268{
1269 return -EOPNOTSUPP;
1270}
1271
1272static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1273{
1274 return -EOPNOTSUPP;
1275}
1276#endif
1277
1278struct bpf_flow_keys;
1279bool __skb_flow_bpf_dissect(struct bpf_prog *prog,
1280 const struct sk_buff *skb,
1281 struct flow_dissector *flow_dissector,
1282 struct bpf_flow_keys *flow_keys);
1283bool __skb_flow_dissect(const struct sk_buff *skb,
1284 struct flow_dissector *flow_dissector,
1285 void *target_container,
1286 void *data, __be16 proto, int nhoff, int hlen,
1287 unsigned int flags);
1288
1289static inline bool skb_flow_dissect(const struct sk_buff *skb,
1290 struct flow_dissector *flow_dissector,
1291 void *target_container, unsigned int flags)
1292{
1293 return __skb_flow_dissect(skb, flow_dissector, target_container,
1294 NULL, 0, 0, 0, flags);
1295}
1296
1297static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1298 struct flow_keys *flow,
1299 unsigned int flags)
1300{
1301 memset(flow, 0, sizeof(*flow));
1302 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1303 NULL, 0, 0, 0, flags);
1304}
1305
1306static inline bool
1307skb_flow_dissect_flow_keys_basic(const struct sk_buff *skb,
1308 struct flow_keys_basic *flow, void *data,
1309 __be16 proto, int nhoff, int hlen,
1310 unsigned int flags)
1311{
1312 memset(flow, 0, sizeof(*flow));
1313 return __skb_flow_dissect(skb, &flow_keys_basic_dissector, flow,
1314 data, proto, nhoff, hlen, flags);
1315}
1316
1317void
1318skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1319 struct flow_dissector *flow_dissector,
1320 void *target_container);
1321
1322static inline __u32 skb_get_hash(struct sk_buff *skb)
1323{
1324 if (!skb->l4_hash && !skb->sw_hash)
1325 __skb_get_hash(skb);
1326
1327 return skb->hash;
1328}
1329
1330static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1331{
1332 if (!skb->l4_hash && !skb->sw_hash) {
1333 struct flow_keys keys;
1334 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1335
1336 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1337 }
1338
1339 return skb->hash;
1340}
1341
1342__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1343
1344static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1345{
1346 return skb->hash;
1347}
1348
1349static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1350{
1351 to->hash = from->hash;
1352 to->sw_hash = from->sw_hash;
1353 to->l4_hash = from->l4_hash;
1354};
1355
1356#ifdef NET_SKBUFF_DATA_USES_OFFSET
1357static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1358{
1359 return skb->head + skb->end;
1360}
1361
1362static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1363{
1364 return skb->end;
1365}
1366#else
1367static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1368{
1369 return skb->end;
1370}
1371
1372static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1373{
1374 return skb->end - skb->head;
1375}
1376#endif
1377
1378/* Internal */
1379#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1380
1381static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1382{
1383 return &skb_shinfo(skb)->hwtstamps;
1384}
1385
1386static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1387{
1388 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1389
1390 return is_zcopy ? skb_uarg(skb) : NULL;
1391}
1392
1393static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1394 bool *have_ref)
1395{
1396 if (skb && uarg && !skb_zcopy(skb)) {
1397 if (unlikely(have_ref && *have_ref))
1398 *have_ref = false;
1399 else
1400 sock_zerocopy_get(uarg);
1401 skb_shinfo(skb)->destructor_arg = uarg;
1402 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1403 }
1404}
1405
1406static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1407{
1408 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1409 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1410}
1411
1412static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1413{
1414 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1415}
1416
1417static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1418{
1419 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1420}
1421
1422/* Release a reference on a zerocopy structure */
1423static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1424{
1425 struct ubuf_info *uarg = skb_zcopy(skb);
1426
1427 if (uarg) {
1428 if (uarg->callback == sock_zerocopy_callback) {
1429 uarg->zerocopy = uarg->zerocopy && zerocopy;
1430 sock_zerocopy_put(uarg);
1431 } else if (!skb_zcopy_is_nouarg(skb)) {
1432 uarg->callback(uarg, zerocopy);
1433 }
1434
1435 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1436 }
1437}
1438
1439/* Abort a zerocopy operation and revert zckey on error in send syscall */
1440static inline void skb_zcopy_abort(struct sk_buff *skb)
1441{
1442 struct ubuf_info *uarg = skb_zcopy(skb);
1443
1444 if (uarg) {
1445 sock_zerocopy_put_abort(uarg, false);
1446 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1447 }
1448}
1449
1450static inline void skb_mark_not_on_list(struct sk_buff *skb)
1451{
1452 skb->next = NULL;
1453}
1454
1455static inline void skb_list_del_init(struct sk_buff *skb)
1456{
1457 __list_del_entry(&skb->list);
1458 skb_mark_not_on_list(skb);
1459}
1460
1461/**
1462 * skb_queue_empty - check if a queue is empty
1463 * @list: queue head
1464 *
1465 * Returns true if the queue is empty, false otherwise.
1466 */
1467static inline int skb_queue_empty(const struct sk_buff_head *list)
1468{
1469 return list->next == (const struct sk_buff *) list;
1470}
1471
1472/**
1473 * skb_queue_is_last - check if skb is the last entry in the queue
1474 * @list: queue head
1475 * @skb: buffer
1476 *
1477 * Returns true if @skb is the last buffer on the list.
1478 */
1479static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1480 const struct sk_buff *skb)
1481{
1482 return skb->next == (const struct sk_buff *) list;
1483}
1484
1485/**
1486 * skb_queue_is_first - check if skb is the first entry in the queue
1487 * @list: queue head
1488 * @skb: buffer
1489 *
1490 * Returns true if @skb is the first buffer on the list.
1491 */
1492static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1493 const struct sk_buff *skb)
1494{
1495 return skb->prev == (const struct sk_buff *) list;
1496}
1497
1498/**
1499 * skb_queue_next - return the next packet in the queue
1500 * @list: queue head
1501 * @skb: current buffer
1502 *
1503 * Return the next packet in @list after @skb. It is only valid to
1504 * call this if skb_queue_is_last() evaluates to false.
1505 */
1506static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1507 const struct sk_buff *skb)
1508{
1509 /* This BUG_ON may seem severe, but if we just return then we
1510 * are going to dereference garbage.
1511 */
1512 BUG_ON(skb_queue_is_last(list, skb));
1513 return skb->next;
1514}
1515
1516/**
1517 * skb_queue_prev - return the prev packet in the queue
1518 * @list: queue head
1519 * @skb: current buffer
1520 *
1521 * Return the prev packet in @list before @skb. It is only valid to
1522 * call this if skb_queue_is_first() evaluates to false.
1523 */
1524static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1525 const struct sk_buff *skb)
1526{
1527 /* This BUG_ON may seem severe, but if we just return then we
1528 * are going to dereference garbage.
1529 */
1530 BUG_ON(skb_queue_is_first(list, skb));
1531 return skb->prev;
1532}
1533
1534/**
1535 * skb_get - reference buffer
1536 * @skb: buffer to reference
1537 *
1538 * Makes another reference to a socket buffer and returns a pointer
1539 * to the buffer.
1540 */
1541static inline struct sk_buff *skb_get(struct sk_buff *skb)
1542{
1543 refcount_inc(&skb->users);
1544 return skb;
1545}
1546
1547/*
1548 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1549 */
1550
1551/**
1552 * skb_cloned - is the buffer a clone
1553 * @skb: buffer to check
1554 *
1555 * Returns true if the buffer was generated with skb_clone() and is
1556 * one of multiple shared copies of the buffer. Cloned buffers are
1557 * shared data so must not be written to under normal circumstances.
1558 */
1559static inline int skb_cloned(const struct sk_buff *skb)
1560{
1561 return skb->cloned &&
1562 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1563}
1564
1565static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1566{
1567 might_sleep_if(gfpflags_allow_blocking(pri));
1568
1569 if (skb_cloned(skb))
1570 return pskb_expand_head(skb, 0, 0, pri);
1571
1572 return 0;
1573}
1574
1575/**
1576 * skb_header_cloned - is the header a clone
1577 * @skb: buffer to check
1578 *
1579 * Returns true if modifying the header part of the buffer requires
1580 * the data to be copied.
1581 */
1582static inline int skb_header_cloned(const struct sk_buff *skb)
1583{
1584 int dataref;
1585
1586 if (!skb->cloned)
1587 return 0;
1588
1589 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1590 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1591 return dataref != 1;
1592}
1593
1594static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1595{
1596 might_sleep_if(gfpflags_allow_blocking(pri));
1597
1598 if (skb_header_cloned(skb))
1599 return pskb_expand_head(skb, 0, 0, pri);
1600
1601 return 0;
1602}
1603
1604/**
1605 * __skb_header_release - release reference to header
1606 * @skb: buffer to operate on
1607 */
1608static inline void __skb_header_release(struct sk_buff *skb)
1609{
1610 skb->nohdr = 1;
1611 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1612}
1613
1614
1615/**
1616 * skb_shared - is the buffer shared
1617 * @skb: buffer to check
1618 *
1619 * Returns true if more than one person has a reference to this
1620 * buffer.
1621 */
1622static inline int skb_shared(const struct sk_buff *skb)
1623{
1624 return refcount_read(&skb->users) != 1;
1625}
1626
1627/**
1628 * skb_share_check - check if buffer is shared and if so clone it
1629 * @skb: buffer to check
1630 * @pri: priority for memory allocation
1631 *
1632 * If the buffer is shared the buffer is cloned and the old copy
1633 * drops a reference. A new clone with a single reference is returned.
1634 * If the buffer is not shared the original buffer is returned. When
1635 * being called from interrupt status or with spinlocks held pri must
1636 * be GFP_ATOMIC.
1637 *
1638 * NULL is returned on a memory allocation failure.
1639 */
1640static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1641{
1642 might_sleep_if(gfpflags_allow_blocking(pri));
1643 if (skb_shared(skb)) {
1644 struct sk_buff *nskb = skb_clone(skb, pri);
1645
1646 if (likely(nskb))
1647 consume_skb(skb);
1648 else
1649 kfree_skb(skb);
1650 skb = nskb;
1651 }
1652 return skb;
1653}
1654
1655/*
1656 * Copy shared buffers into a new sk_buff. We effectively do COW on
1657 * packets to handle cases where we have a local reader and forward
1658 * and a couple of other messy ones. The normal one is tcpdumping
1659 * a packet thats being forwarded.
1660 */
1661
1662/**
1663 * skb_unshare - make a copy of a shared buffer
1664 * @skb: buffer to check
1665 * @pri: priority for memory allocation
1666 *
1667 * If the socket buffer is a clone then this function creates a new
1668 * copy of the data, drops a reference count on the old copy and returns
1669 * the new copy with the reference count at 1. If the buffer is not a clone
1670 * the original buffer is returned. When called with a spinlock held or
1671 * from interrupt state @pri must be %GFP_ATOMIC
1672 *
1673 * %NULL is returned on a memory allocation failure.
1674 */
1675static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1676 gfp_t pri)
1677{
1678 might_sleep_if(gfpflags_allow_blocking(pri));
1679 if (skb_cloned(skb)) {
1680 struct sk_buff *nskb = skb_copy(skb, pri);
1681
1682 /* Free our shared copy */
1683 if (likely(nskb))
1684 consume_skb(skb);
1685 else
1686 kfree_skb(skb);
1687 skb = nskb;
1688 }
1689 return skb;
1690}
1691
1692/**
1693 * skb_peek - peek at the head of an &sk_buff_head
1694 * @list_: list to peek at
1695 *
1696 * Peek an &sk_buff. Unlike most other operations you _MUST_
1697 * be careful with this one. A peek leaves the buffer on the
1698 * list and someone else may run off with it. You must hold
1699 * the appropriate locks or have a private queue to do this.
1700 *
1701 * Returns %NULL for an empty list or a pointer to the head element.
1702 * The reference count is not incremented and the reference is therefore
1703 * volatile. Use with caution.
1704 */
1705static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1706{
1707 struct sk_buff *skb = list_->next;
1708
1709 if (skb == (struct sk_buff *)list_)
1710 skb = NULL;
1711 return skb;
1712}
1713
1714/**
1715 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1716 * @list_: list to peek at
1717 *
1718 * Like skb_peek(), but the caller knows that the list is not empty.
1719 */
1720static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1721{
1722 return list_->next;
1723}
1724
1725/**
1726 * skb_peek_next - peek skb following the given one from a queue
1727 * @skb: skb to start from
1728 * @list_: list to peek at
1729 *
1730 * Returns %NULL when the end of the list is met or a pointer to the
1731 * next element. The reference count is not incremented and the
1732 * reference is therefore volatile. Use with caution.
1733 */
1734static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1735 const struct sk_buff_head *list_)
1736{
1737 struct sk_buff *next = skb->next;
1738
1739 if (next == (struct sk_buff *)list_)
1740 next = NULL;
1741 return next;
1742}
1743
1744/**
1745 * skb_peek_tail - peek at the tail of an &sk_buff_head
1746 * @list_: list to peek at
1747 *
1748 * Peek an &sk_buff. Unlike most other operations you _MUST_
1749 * be careful with this one. A peek leaves the buffer on the
1750 * list and someone else may run off with it. You must hold
1751 * the appropriate locks or have a private queue to do this.
1752 *
1753 * Returns %NULL for an empty list or a pointer to the tail element.
1754 * The reference count is not incremented and the reference is therefore
1755 * volatile. Use with caution.
1756 */
1757static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1758{
1759 struct sk_buff *skb = list_->prev;
1760
1761 if (skb == (struct sk_buff *)list_)
1762 skb = NULL;
1763 return skb;
1764
1765}
1766
1767/**
1768 * skb_queue_len - get queue length
1769 * @list_: list to measure
1770 *
1771 * Return the length of an &sk_buff queue.
1772 */
1773static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1774{
1775 return list_->qlen;
1776}
1777
1778/**
1779 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1780 * @list: queue to initialize
1781 *
1782 * This initializes only the list and queue length aspects of
1783 * an sk_buff_head object. This allows to initialize the list
1784 * aspects of an sk_buff_head without reinitializing things like
1785 * the spinlock. It can also be used for on-stack sk_buff_head
1786 * objects where the spinlock is known to not be used.
1787 */
1788static inline void __skb_queue_head_init(struct sk_buff_head *list)
1789{
1790 list->prev = list->next = (struct sk_buff *)list;
1791 list->qlen = 0;
1792}
1793
1794/*
1795 * This function creates a split out lock class for each invocation;
1796 * this is needed for now since a whole lot of users of the skb-queue
1797 * infrastructure in drivers have different locking usage (in hardirq)
1798 * than the networking core (in softirq only). In the long run either the
1799 * network layer or drivers should need annotation to consolidate the
1800 * main types of usage into 3 classes.
1801 */
1802static inline void skb_queue_head_init(struct sk_buff_head *list)
1803{
1804 spin_lock_init(&list->lock);
1805 __skb_queue_head_init(list);
1806}
1807
1808static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1809 struct lock_class_key *class)
1810{
1811 skb_queue_head_init(list);
1812 lockdep_set_class(&list->lock, class);
1813}
1814
1815/*
1816 * Insert an sk_buff on a list.
1817 *
1818 * The "__skb_xxxx()" functions are the non-atomic ones that
1819 * can only be called with interrupts disabled.
1820 */
1821static inline void __skb_insert(struct sk_buff *newsk,
1822 struct sk_buff *prev, struct sk_buff *next,
1823 struct sk_buff_head *list)
1824{
1825 newsk->next = next;
1826 newsk->prev = prev;
1827 next->prev = prev->next = newsk;
1828 list->qlen++;
1829}
1830
1831static inline void __skb_queue_splice(const struct sk_buff_head *list,
1832 struct sk_buff *prev,
1833 struct sk_buff *next)
1834{
1835 struct sk_buff *first = list->next;
1836 struct sk_buff *last = list->prev;
1837
1838 first->prev = prev;
1839 prev->next = first;
1840
1841 last->next = next;
1842 next->prev = last;
1843}
1844
1845/**
1846 * skb_queue_splice - join two skb lists, this is designed for stacks
1847 * @list: the new list to add
1848 * @head: the place to add it in the first list
1849 */
1850static inline void skb_queue_splice(const struct sk_buff_head *list,
1851 struct sk_buff_head *head)
1852{
1853 if (!skb_queue_empty(list)) {
1854 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1855 head->qlen += list->qlen;
1856 }
1857}
1858
1859/**
1860 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1861 * @list: the new list to add
1862 * @head: the place to add it in the first list
1863 *
1864 * The list at @list is reinitialised
1865 */
1866static inline void skb_queue_splice_init(struct sk_buff_head *list,
1867 struct sk_buff_head *head)
1868{
1869 if (!skb_queue_empty(list)) {
1870 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1871 head->qlen += list->qlen;
1872 __skb_queue_head_init(list);
1873 }
1874}
1875
1876/**
1877 * skb_queue_splice_tail - join two skb lists, each list being a queue
1878 * @list: the new list to add
1879 * @head: the place to add it in the first list
1880 */
1881static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1882 struct sk_buff_head *head)
1883{
1884 if (!skb_queue_empty(list)) {
1885 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1886 head->qlen += list->qlen;
1887 }
1888}
1889
1890/**
1891 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1892 * @list: the new list to add
1893 * @head: the place to add it in the first list
1894 *
1895 * Each of the lists is a queue.
1896 * The list at @list is reinitialised
1897 */
1898static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1899 struct sk_buff_head *head)
1900{
1901 if (!skb_queue_empty(list)) {
1902 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1903 head->qlen += list->qlen;
1904 __skb_queue_head_init(list);
1905 }
1906}
1907
1908/**
1909 * __skb_queue_after - queue a buffer at the list head
1910 * @list: list to use
1911 * @prev: place after this buffer
1912 * @newsk: buffer to queue
1913 *
1914 * Queue a buffer int the middle of a list. This function takes no locks
1915 * and you must therefore hold required locks before calling it.
1916 *
1917 * A buffer cannot be placed on two lists at the same time.
1918 */
1919static inline void __skb_queue_after(struct sk_buff_head *list,
1920 struct sk_buff *prev,
1921 struct sk_buff *newsk)
1922{
1923 __skb_insert(newsk, prev, prev->next, list);
1924}
1925
1926void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1927 struct sk_buff_head *list);
1928
1929static inline void __skb_queue_before(struct sk_buff_head *list,
1930 struct sk_buff *next,
1931 struct sk_buff *newsk)
1932{
1933 __skb_insert(newsk, next->prev, next, list);
1934}
1935
1936/**
1937 * __skb_queue_head - queue a buffer at the list head
1938 * @list: list to use
1939 * @newsk: buffer to queue
1940 *
1941 * Queue a buffer at the start of a list. This function takes no locks
1942 * and you must therefore hold required locks before calling it.
1943 *
1944 * A buffer cannot be placed on two lists at the same time.
1945 */
1946static inline void __skb_queue_head(struct sk_buff_head *list,
1947 struct sk_buff *newsk)
1948{
1949 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1950}
1951void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1952
1953/**
1954 * __skb_queue_tail - queue a buffer at the list tail
1955 * @list: list to use
1956 * @newsk: buffer to queue
1957 *
1958 * Queue a buffer at the end of a list. This function takes no locks
1959 * and you must therefore hold required locks before calling it.
1960 *
1961 * A buffer cannot be placed on two lists at the same time.
1962 */
1963static inline void __skb_queue_tail(struct sk_buff_head *list,
1964 struct sk_buff *newsk)
1965{
1966 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1967}
1968void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1969
1970/*
1971 * remove sk_buff from list. _Must_ be called atomically, and with
1972 * the list known..
1973 */
1974void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1975static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1976{
1977 struct sk_buff *next, *prev;
1978
1979 list->qlen--;
1980 next = skb->next;
1981 prev = skb->prev;
1982 skb->next = skb->prev = NULL;
1983 next->prev = prev;
1984 prev->next = next;
1985}
1986
1987/**
1988 * __skb_dequeue - remove from the head of the queue
1989 * @list: list to dequeue from
1990 *
1991 * Remove the head of the list. This function does not take any locks
1992 * so must be used with appropriate locks held only. The head item is
1993 * returned or %NULL if the list is empty.
1994 */
1995static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1996{
1997 struct sk_buff *skb = skb_peek(list);
1998 if (skb)
1999 __skb_unlink(skb, list);
2000 return skb;
2001}
2002struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2003
2004/**
2005 * __skb_dequeue_tail - remove from the tail of the queue
2006 * @list: list to dequeue from
2007 *
2008 * Remove the tail of the list. This function does not take any locks
2009 * so must be used with appropriate locks held only. The tail item is
2010 * returned or %NULL if the list is empty.
2011 */
2012static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2013{
2014 struct sk_buff *skb = skb_peek_tail(list);
2015 if (skb)
2016 __skb_unlink(skb, list);
2017 return skb;
2018}
2019struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2020
2021
2022static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2023{
2024 return skb->data_len;
2025}
2026
2027static inline unsigned int skb_headlen(const struct sk_buff *skb)
2028{
2029 return skb->len - skb->data_len;
2030}
2031
2032static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2033{
2034 unsigned int i, len = 0;
2035
2036 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2037 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2038 return len;
2039}
2040
2041static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2042{
2043 return skb_headlen(skb) + __skb_pagelen(skb);
2044}
2045
2046/**
2047 * __skb_fill_page_desc - initialise a paged fragment in an skb
2048 * @skb: buffer containing fragment to be initialised
2049 * @i: paged fragment index to initialise
2050 * @page: the page to use for this fragment
2051 * @off: the offset to the data with @page
2052 * @size: the length of the data
2053 *
2054 * Initialises the @i'th fragment of @skb to point to &size bytes at
2055 * offset @off within @page.
2056 *
2057 * Does not take any additional reference on the fragment.
2058 */
2059static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2060 struct page *page, int off, int size)
2061{
2062 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2063
2064 /*
2065 * Propagate page pfmemalloc to the skb if we can. The problem is
2066 * that not all callers have unique ownership of the page but rely
2067 * on page_is_pfmemalloc doing the right thing(tm).
2068 */
2069 frag->page.p = page;
2070 frag->page_offset = off;
2071 skb_frag_size_set(frag, size);
2072
2073 page = compound_head(page);
2074 if (page_is_pfmemalloc(page))
2075 skb->pfmemalloc = true;
2076}
2077
2078/**
2079 * skb_fill_page_desc - initialise a paged fragment in an skb
2080 * @skb: buffer containing fragment to be initialised
2081 * @i: paged fragment index to initialise
2082 * @page: the page to use for this fragment
2083 * @off: the offset to the data with @page
2084 * @size: the length of the data
2085 *
2086 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2087 * @skb to point to @size bytes at offset @off within @page. In
2088 * addition updates @skb such that @i is the last fragment.
2089 *
2090 * Does not take any additional reference on the fragment.
2091 */
2092static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2093 struct page *page, int off, int size)
2094{
2095 __skb_fill_page_desc(skb, i, page, off, size);
2096 skb_shinfo(skb)->nr_frags = i + 1;
2097}
2098
2099void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2100 int size, unsigned int truesize);
2101
2102void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2103 unsigned int truesize);
2104
2105#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
2106#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
2107#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2108
2109#ifdef NET_SKBUFF_DATA_USES_OFFSET
2110static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2111{
2112 return skb->head + skb->tail;
2113}
2114
2115static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2116{
2117 skb->tail = skb->data - skb->head;
2118}
2119
2120static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2121{
2122 skb_reset_tail_pointer(skb);
2123 skb->tail += offset;
2124}
2125
2126#else /* NET_SKBUFF_DATA_USES_OFFSET */
2127static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2128{
2129 return skb->tail;
2130}
2131
2132static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2133{
2134 skb->tail = skb->data;
2135}
2136
2137static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2138{
2139 skb->tail = skb->data + offset;
2140}
2141
2142#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2143
2144/*
2145 * Add data to an sk_buff
2146 */
2147void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2148void *skb_put(struct sk_buff *skb, unsigned int len);
2149static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2150{
2151 void *tmp = skb_tail_pointer(skb);
2152 SKB_LINEAR_ASSERT(skb);
2153 skb->tail += len;
2154 skb->len += len;
2155 return tmp;
2156}
2157
2158static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2159{
2160 void *tmp = __skb_put(skb, len);
2161
2162 memset(tmp, 0, len);
2163 return tmp;
2164}
2165
2166static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2167 unsigned int len)
2168{
2169 void *tmp = __skb_put(skb, len);
2170
2171 memcpy(tmp, data, len);
2172 return tmp;
2173}
2174
2175static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2176{
2177 *(u8 *)__skb_put(skb, 1) = val;
2178}
2179
2180static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2181{
2182 void *tmp = skb_put(skb, len);
2183
2184 memset(tmp, 0, len);
2185
2186 return tmp;
2187}
2188
2189static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2190 unsigned int len)
2191{
2192 void *tmp = skb_put(skb, len);
2193
2194 memcpy(tmp, data, len);
2195
2196 return tmp;
2197}
2198
2199static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2200{
2201 *(u8 *)skb_put(skb, 1) = val;
2202}
2203
2204void *skb_push(struct sk_buff *skb, unsigned int len);
2205static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2206{
2207 skb->data -= len;
2208 skb->len += len;
2209 return skb->data;
2210}
2211
2212void *skb_pull(struct sk_buff *skb, unsigned int len);
2213static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2214{
2215 skb->len -= len;
2216 BUG_ON(skb->len < skb->data_len);
2217 return skb->data += len;
2218}
2219
2220static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2221{
2222 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2223}
2224
2225void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2226
2227static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2228{
2229 if (len > skb_headlen(skb) &&
2230 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2231 return NULL;
2232 skb->len -= len;
2233 return skb->data += len;
2234}
2235
2236static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2237{
2238 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2239}
2240
2241static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2242{
2243 if (likely(len <= skb_headlen(skb)))
2244 return 1;
2245 if (unlikely(len > skb->len))
2246 return 0;
2247 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2248}
2249
2250void skb_condense(struct sk_buff *skb);
2251
2252/**
2253 * skb_headroom - bytes at buffer head
2254 * @skb: buffer to check
2255 *
2256 * Return the number of bytes of free space at the head of an &sk_buff.
2257 */
2258static inline unsigned int skb_headroom(const struct sk_buff *skb)
2259{
2260 return skb->data - skb->head;
2261}
2262
2263/**
2264 * skb_tailroom - bytes at buffer end
2265 * @skb: buffer to check
2266 *
2267 * Return the number of bytes of free space at the tail of an sk_buff
2268 */
2269static inline int skb_tailroom(const struct sk_buff *skb)
2270{
2271 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2272}
2273
2274/**
2275 * skb_availroom - bytes at buffer end
2276 * @skb: buffer to check
2277 *
2278 * Return the number of bytes of free space at the tail of an sk_buff
2279 * allocated by sk_stream_alloc()
2280 */
2281static inline int skb_availroom(const struct sk_buff *skb)
2282{
2283 if (skb_is_nonlinear(skb))
2284 return 0;
2285
2286 return skb->end - skb->tail - skb->reserved_tailroom;
2287}
2288
2289/**
2290 * skb_reserve - adjust headroom
2291 * @skb: buffer to alter
2292 * @len: bytes to move
2293 *
2294 * Increase the headroom of an empty &sk_buff by reducing the tail
2295 * room. This is only allowed for an empty buffer.
2296 */
2297static inline void skb_reserve(struct sk_buff *skb, int len)
2298{
2299 skb->data += len;
2300 skb->tail += len;
2301}
2302
2303/**
2304 * skb_tailroom_reserve - adjust reserved_tailroom
2305 * @skb: buffer to alter
2306 * @mtu: maximum amount of headlen permitted
2307 * @needed_tailroom: minimum amount of reserved_tailroom
2308 *
2309 * Set reserved_tailroom so that headlen can be as large as possible but
2310 * not larger than mtu and tailroom cannot be smaller than
2311 * needed_tailroom.
2312 * The required headroom should already have been reserved before using
2313 * this function.
2314 */
2315static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2316 unsigned int needed_tailroom)
2317{
2318 SKB_LINEAR_ASSERT(skb);
2319 if (mtu < skb_tailroom(skb) - needed_tailroom)
2320 /* use at most mtu */
2321 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2322 else
2323 /* use up to all available space */
2324 skb->reserved_tailroom = needed_tailroom;
2325}
2326
2327#define ENCAP_TYPE_ETHER 0
2328#define ENCAP_TYPE_IPPROTO 1
2329
2330static inline void skb_set_inner_protocol(struct sk_buff *skb,
2331 __be16 protocol)
2332{
2333 skb->inner_protocol = protocol;
2334 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2335}
2336
2337static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2338 __u8 ipproto)
2339{
2340 skb->inner_ipproto = ipproto;
2341 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2342}
2343
2344static inline void skb_reset_inner_headers(struct sk_buff *skb)
2345{
2346 skb->inner_mac_header = skb->mac_header;
2347 skb->inner_network_header = skb->network_header;
2348 skb->inner_transport_header = skb->transport_header;
2349}
2350
2351static inline void skb_reset_mac_len(struct sk_buff *skb)
2352{
2353 skb->mac_len = skb->network_header - skb->mac_header;
2354}
2355
2356static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2357 *skb)
2358{
2359 return skb->head + skb->inner_transport_header;
2360}
2361
2362static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2363{
2364 return skb_inner_transport_header(skb) - skb->data;
2365}
2366
2367static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2368{
2369 skb->inner_transport_header = skb->data - skb->head;
2370}
2371
2372static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2373 const int offset)
2374{
2375 skb_reset_inner_transport_header(skb);
2376 skb->inner_transport_header += offset;
2377}
2378
2379static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2380{
2381 return skb->head + skb->inner_network_header;
2382}
2383
2384static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2385{
2386 skb->inner_network_header = skb->data - skb->head;
2387}
2388
2389static inline void skb_set_inner_network_header(struct sk_buff *skb,
2390 const int offset)
2391{
2392 skb_reset_inner_network_header(skb);
2393 skb->inner_network_header += offset;
2394}
2395
2396static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2397{
2398 return skb->head + skb->inner_mac_header;
2399}
2400
2401static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2402{
2403 skb->inner_mac_header = skb->data - skb->head;
2404}
2405
2406static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2407 const int offset)
2408{
2409 skb_reset_inner_mac_header(skb);
2410 skb->inner_mac_header += offset;
2411}
2412static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2413{
2414 return skb->transport_header != (typeof(skb->transport_header))~0U;
2415}
2416
2417static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2418{
2419 return skb->head + skb->transport_header;
2420}
2421
2422static inline void skb_reset_transport_header(struct sk_buff *skb)
2423{
2424 skb->transport_header = skb->data - skb->head;
2425}
2426
2427static inline void skb_set_transport_header(struct sk_buff *skb,
2428 const int offset)
2429{
2430 skb_reset_transport_header(skb);
2431 skb->transport_header += offset;
2432}
2433
2434static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2435{
2436 return skb->head + skb->network_header;
2437}
2438
2439static inline void skb_reset_network_header(struct sk_buff *skb)
2440{
2441 skb->network_header = skb->data - skb->head;
2442}
2443
2444static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2445{
2446 skb_reset_network_header(skb);
2447 skb->network_header += offset;
2448}
2449
2450static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2451{
2452 return skb->head + skb->mac_header;
2453}
2454
2455static inline int skb_mac_offset(const struct sk_buff *skb)
2456{
2457 return skb_mac_header(skb) - skb->data;
2458}
2459
2460static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2461{
2462 return skb->network_header - skb->mac_header;
2463}
2464
2465static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2466{
2467 return skb->mac_header != (typeof(skb->mac_header))~0U;
2468}
2469
2470static inline void skb_reset_mac_header(struct sk_buff *skb)
2471{
2472 skb->mac_header = skb->data - skb->head;
2473}
2474
2475static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2476{
2477 skb_reset_mac_header(skb);
2478 skb->mac_header += offset;
2479}
2480
2481static inline void skb_pop_mac_header(struct sk_buff *skb)
2482{
2483 skb->mac_header = skb->network_header;
2484}
2485
2486static inline void skb_probe_transport_header(struct sk_buff *skb)
2487{
2488 struct flow_keys_basic keys;
2489
2490 if (skb_transport_header_was_set(skb))
2491 return;
2492
2493 if (skb_flow_dissect_flow_keys_basic(skb, &keys, NULL, 0, 0, 0, 0))
2494 skb_set_transport_header(skb, keys.control.thoff);
2495}
2496
2497static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2498{
2499 if (skb_mac_header_was_set(skb)) {
2500 const unsigned char *old_mac = skb_mac_header(skb);
2501
2502 skb_set_mac_header(skb, -skb->mac_len);
2503 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2504 }
2505}
2506
2507static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2508{
2509 return skb->csum_start - skb_headroom(skb);
2510}
2511
2512static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2513{
2514 return skb->head + skb->csum_start;
2515}
2516
2517static inline int skb_transport_offset(const struct sk_buff *skb)
2518{
2519 return skb_transport_header(skb) - skb->data;
2520}
2521
2522static inline u32 skb_network_header_len(const struct sk_buff *skb)
2523{
2524 return skb->transport_header - skb->network_header;
2525}
2526
2527static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2528{
2529 return skb->inner_transport_header - skb->inner_network_header;
2530}
2531
2532static inline int skb_network_offset(const struct sk_buff *skb)
2533{
2534 return skb_network_header(skb) - skb->data;
2535}
2536
2537static inline int skb_inner_network_offset(const struct sk_buff *skb)
2538{
2539 return skb_inner_network_header(skb) - skb->data;
2540}
2541
2542static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2543{
2544 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2545}
2546
2547/*
2548 * CPUs often take a performance hit when accessing unaligned memory
2549 * locations. The actual performance hit varies, it can be small if the
2550 * hardware handles it or large if we have to take an exception and fix it
2551 * in software.
2552 *
2553 * Since an ethernet header is 14 bytes network drivers often end up with
2554 * the IP header at an unaligned offset. The IP header can be aligned by
2555 * shifting the start of the packet by 2 bytes. Drivers should do this
2556 * with:
2557 *
2558 * skb_reserve(skb, NET_IP_ALIGN);
2559 *
2560 * The downside to this alignment of the IP header is that the DMA is now
2561 * unaligned. On some architectures the cost of an unaligned DMA is high
2562 * and this cost outweighs the gains made by aligning the IP header.
2563 *
2564 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2565 * to be overridden.
2566 */
2567#ifndef NET_IP_ALIGN
2568#define NET_IP_ALIGN 2
2569#endif
2570
2571/*
2572 * The networking layer reserves some headroom in skb data (via
2573 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2574 * the header has to grow. In the default case, if the header has to grow
2575 * 32 bytes or less we avoid the reallocation.
2576 *
2577 * Unfortunately this headroom changes the DMA alignment of the resulting
2578 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2579 * on some architectures. An architecture can override this value,
2580 * perhaps setting it to a cacheline in size (since that will maintain
2581 * cacheline alignment of the DMA). It must be a power of 2.
2582 *
2583 * Various parts of the networking layer expect at least 32 bytes of
2584 * headroom, you should not reduce this.
2585 *
2586 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2587 * to reduce average number of cache lines per packet.
2588 * get_rps_cpus() for example only access one 64 bytes aligned block :
2589 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2590 */
2591#ifndef NET_SKB_PAD
2592#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2593#endif
2594
2595int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2596
2597static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2598{
2599 if (WARN_ON(skb_is_nonlinear(skb)))
2600 return;
2601 skb->len = len;
2602 skb_set_tail_pointer(skb, len);
2603}
2604
2605static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2606{
2607 __skb_set_length(skb, len);
2608}
2609
2610void skb_trim(struct sk_buff *skb, unsigned int len);
2611
2612static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2613{
2614 if (skb->data_len)
2615 return ___pskb_trim(skb, len);
2616 __skb_trim(skb, len);
2617 return 0;
2618}
2619
2620static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2621{
2622 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2623}
2624
2625/**
2626 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2627 * @skb: buffer to alter
2628 * @len: new length
2629 *
2630 * This is identical to pskb_trim except that the caller knows that
2631 * the skb is not cloned so we should never get an error due to out-
2632 * of-memory.
2633 */
2634static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2635{
2636 int err = pskb_trim(skb, len);
2637 BUG_ON(err);
2638}
2639
2640static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2641{
2642 unsigned int diff = len - skb->len;
2643
2644 if (skb_tailroom(skb) < diff) {
2645 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2646 GFP_ATOMIC);
2647 if (ret)
2648 return ret;
2649 }
2650 __skb_set_length(skb, len);
2651 return 0;
2652}
2653
2654/**
2655 * skb_orphan - orphan a buffer
2656 * @skb: buffer to orphan
2657 *
2658 * If a buffer currently has an owner then we call the owner's
2659 * destructor function and make the @skb unowned. The buffer continues
2660 * to exist but is no longer charged to its former owner.
2661 */
2662static inline void skb_orphan(struct sk_buff *skb)
2663{
2664 if (skb->destructor) {
2665 skb->destructor(skb);
2666 skb->destructor = NULL;
2667 skb->sk = NULL;
2668 } else {
2669 BUG_ON(skb->sk);
2670 }
2671}
2672
2673/**
2674 * skb_orphan_frags - orphan the frags contained in a buffer
2675 * @skb: buffer to orphan frags from
2676 * @gfp_mask: allocation mask for replacement pages
2677 *
2678 * For each frag in the SKB which needs a destructor (i.e. has an
2679 * owner) create a copy of that frag and release the original
2680 * page by calling the destructor.
2681 */
2682static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2683{
2684 if (likely(!skb_zcopy(skb)))
2685 return 0;
2686 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2687 return 0;
2688 return skb_copy_ubufs(skb, gfp_mask);
2689}
2690
2691/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2692static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2693{
2694 if (likely(!skb_zcopy(skb)))
2695 return 0;
2696 return skb_copy_ubufs(skb, gfp_mask);
2697}
2698
2699/**
2700 * __skb_queue_purge - empty a list
2701 * @list: list to empty
2702 *
2703 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2704 * the list and one reference dropped. This function does not take the
2705 * list lock and the caller must hold the relevant locks to use it.
2706 */
2707static inline void __skb_queue_purge(struct sk_buff_head *list)
2708{
2709 struct sk_buff *skb;
2710 while ((skb = __skb_dequeue(list)) != NULL)
2711 kfree_skb(skb);
2712}
2713void skb_queue_purge(struct sk_buff_head *list);
2714
2715unsigned int skb_rbtree_purge(struct rb_root *root);
2716
2717void *netdev_alloc_frag(unsigned int fragsz);
2718
2719struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2720 gfp_t gfp_mask);
2721
2722/**
2723 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2724 * @dev: network device to receive on
2725 * @length: length to allocate
2726 *
2727 * Allocate a new &sk_buff and assign it a usage count of one. The
2728 * buffer has unspecified headroom built in. Users should allocate
2729 * the headroom they think they need without accounting for the
2730 * built in space. The built in space is used for optimisations.
2731 *
2732 * %NULL is returned if there is no free memory. Although this function
2733 * allocates memory it can be called from an interrupt.
2734 */
2735static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2736 unsigned int length)
2737{
2738 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2739}
2740
2741/* legacy helper around __netdev_alloc_skb() */
2742static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2743 gfp_t gfp_mask)
2744{
2745 return __netdev_alloc_skb(NULL, length, gfp_mask);
2746}
2747
2748/* legacy helper around netdev_alloc_skb() */
2749static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2750{
2751 return netdev_alloc_skb(NULL, length);
2752}
2753
2754
2755static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2756 unsigned int length, gfp_t gfp)
2757{
2758 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2759
2760 if (NET_IP_ALIGN && skb)
2761 skb_reserve(skb, NET_IP_ALIGN);
2762 return skb;
2763}
2764
2765static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2766 unsigned int length)
2767{
2768 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2769}
2770
2771static inline void skb_free_frag(void *addr)
2772{
2773 page_frag_free(addr);
2774}
2775
2776void *napi_alloc_frag(unsigned int fragsz);
2777struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2778 unsigned int length, gfp_t gfp_mask);
2779static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2780 unsigned int length)
2781{
2782 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2783}
2784void napi_consume_skb(struct sk_buff *skb, int budget);
2785
2786void __kfree_skb_flush(void);
2787void __kfree_skb_defer(struct sk_buff *skb);
2788
2789/**
2790 * __dev_alloc_pages - allocate page for network Rx
2791 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2792 * @order: size of the allocation
2793 *
2794 * Allocate a new page.
2795 *
2796 * %NULL is returned if there is no free memory.
2797*/
2798static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2799 unsigned int order)
2800{
2801 /* This piece of code contains several assumptions.
2802 * 1. This is for device Rx, therefor a cold page is preferred.
2803 * 2. The expectation is the user wants a compound page.
2804 * 3. If requesting a order 0 page it will not be compound
2805 * due to the check to see if order has a value in prep_new_page
2806 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2807 * code in gfp_to_alloc_flags that should be enforcing this.
2808 */
2809 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2810
2811 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2812}
2813
2814static inline struct page *dev_alloc_pages(unsigned int order)
2815{
2816 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2817}
2818
2819/**
2820 * __dev_alloc_page - allocate a page for network Rx
2821 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2822 *
2823 * Allocate a new page.
2824 *
2825 * %NULL is returned if there is no free memory.
2826 */
2827static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2828{
2829 return __dev_alloc_pages(gfp_mask, 0);
2830}
2831
2832static inline struct page *dev_alloc_page(void)
2833{
2834 return dev_alloc_pages(0);
2835}
2836
2837/**
2838 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2839 * @page: The page that was allocated from skb_alloc_page
2840 * @skb: The skb that may need pfmemalloc set
2841 */
2842static inline void skb_propagate_pfmemalloc(struct page *page,
2843 struct sk_buff *skb)
2844{
2845 if (page_is_pfmemalloc(page))
2846 skb->pfmemalloc = true;
2847}
2848
2849/**
2850 * skb_frag_page - retrieve the page referred to by a paged fragment
2851 * @frag: the paged fragment
2852 *
2853 * Returns the &struct page associated with @frag.
2854 */
2855static inline struct page *skb_frag_page(const skb_frag_t *frag)
2856{
2857 return frag->page.p;
2858}
2859
2860/**
2861 * __skb_frag_ref - take an addition reference on a paged fragment.
2862 * @frag: the paged fragment
2863 *
2864 * Takes an additional reference on the paged fragment @frag.
2865 */
2866static inline void __skb_frag_ref(skb_frag_t *frag)
2867{
2868 get_page(skb_frag_page(frag));
2869}
2870
2871/**
2872 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2873 * @skb: the buffer
2874 * @f: the fragment offset.
2875 *
2876 * Takes an additional reference on the @f'th paged fragment of @skb.
2877 */
2878static inline void skb_frag_ref(struct sk_buff *skb, int f)
2879{
2880 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2881}
2882
2883/**
2884 * __skb_frag_unref - release a reference on a paged fragment.
2885 * @frag: the paged fragment
2886 *
2887 * Releases a reference on the paged fragment @frag.
2888 */
2889static inline void __skb_frag_unref(skb_frag_t *frag)
2890{
2891 put_page(skb_frag_page(frag));
2892}
2893
2894/**
2895 * skb_frag_unref - release a reference on a paged fragment of an skb.
2896 * @skb: the buffer
2897 * @f: the fragment offset
2898 *
2899 * Releases a reference on the @f'th paged fragment of @skb.
2900 */
2901static inline void skb_frag_unref(struct sk_buff *skb, int f)
2902{
2903 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2904}
2905
2906/**
2907 * skb_frag_address - gets the address of the data contained in a paged fragment
2908 * @frag: the paged fragment buffer
2909 *
2910 * Returns the address of the data within @frag. The page must already
2911 * be mapped.
2912 */
2913static inline void *skb_frag_address(const skb_frag_t *frag)
2914{
2915 return page_address(skb_frag_page(frag)) + frag->page_offset;
2916}
2917
2918/**
2919 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2920 * @frag: the paged fragment buffer
2921 *
2922 * Returns the address of the data within @frag. Checks that the page
2923 * is mapped and returns %NULL otherwise.
2924 */
2925static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2926{
2927 void *ptr = page_address(skb_frag_page(frag));
2928 if (unlikely(!ptr))
2929 return NULL;
2930
2931 return ptr + frag->page_offset;
2932}
2933
2934/**
2935 * __skb_frag_set_page - sets the page contained in a paged fragment
2936 * @frag: the paged fragment
2937 * @page: the page to set
2938 *
2939 * Sets the fragment @frag to contain @page.
2940 */
2941static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2942{
2943 frag->page.p = page;
2944}
2945
2946/**
2947 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2948 * @skb: the buffer
2949 * @f: the fragment offset
2950 * @page: the page to set
2951 *
2952 * Sets the @f'th fragment of @skb to contain @page.
2953 */
2954static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2955 struct page *page)
2956{
2957 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2958}
2959
2960bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2961
2962/**
2963 * skb_frag_dma_map - maps a paged fragment via the DMA API
2964 * @dev: the device to map the fragment to
2965 * @frag: the paged fragment to map
2966 * @offset: the offset within the fragment (starting at the
2967 * fragment's own offset)
2968 * @size: the number of bytes to map
2969 * @dir: the direction of the mapping (``PCI_DMA_*``)
2970 *
2971 * Maps the page associated with @frag to @device.
2972 */
2973static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2974 const skb_frag_t *frag,
2975 size_t offset, size_t size,
2976 enum dma_data_direction dir)
2977{
2978 return dma_map_page(dev, skb_frag_page(frag),
2979 frag->page_offset + offset, size, dir);
2980}
2981
2982static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2983 gfp_t gfp_mask)
2984{
2985 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2986}
2987
2988
2989static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2990 gfp_t gfp_mask)
2991{
2992 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2993}
2994
2995
2996/**
2997 * skb_clone_writable - is the header of a clone writable
2998 * @skb: buffer to check
2999 * @len: length up to which to write
3000 *
3001 * Returns true if modifying the header part of the cloned buffer
3002 * does not requires the data to be copied.
3003 */
3004static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3005{
3006 return !skb_header_cloned(skb) &&
3007 skb_headroom(skb) + len <= skb->hdr_len;
3008}
3009
3010static inline int skb_try_make_writable(struct sk_buff *skb,
3011 unsigned int write_len)
3012{
3013 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3014 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3015}
3016
3017static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3018 int cloned)
3019{
3020 int delta = 0;
3021
3022 if (headroom > skb_headroom(skb))
3023 delta = headroom - skb_headroom(skb);
3024
3025 if (delta || cloned)
3026 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3027 GFP_ATOMIC);
3028 return 0;
3029}
3030
3031/**
3032 * skb_cow - copy header of skb when it is required
3033 * @skb: buffer to cow
3034 * @headroom: needed headroom
3035 *
3036 * If the skb passed lacks sufficient headroom or its data part
3037 * is shared, data is reallocated. If reallocation fails, an error
3038 * is returned and original skb is not changed.
3039 *
3040 * The result is skb with writable area skb->head...skb->tail
3041 * and at least @headroom of space at head.
3042 */
3043static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3044{
3045 return __skb_cow(skb, headroom, skb_cloned(skb));
3046}
3047
3048/**
3049 * skb_cow_head - skb_cow but only making the head writable
3050 * @skb: buffer to cow
3051 * @headroom: needed headroom
3052 *
3053 * This function is identical to skb_cow except that we replace the
3054 * skb_cloned check by skb_header_cloned. It should be used when
3055 * you only need to push on some header and do not need to modify
3056 * the data.
3057 */
3058static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3059{
3060 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3061}
3062
3063/**
3064 * skb_padto - pad an skbuff up to a minimal size
3065 * @skb: buffer to pad
3066 * @len: minimal length
3067 *
3068 * Pads up a buffer to ensure the trailing bytes exist and are
3069 * blanked. If the buffer already contains sufficient data it
3070 * is untouched. Otherwise it is extended. Returns zero on
3071 * success. The skb is freed on error.
3072 */
3073static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3074{
3075 unsigned int size = skb->len;
3076 if (likely(size >= len))
3077 return 0;
3078 return skb_pad(skb, len - size);
3079}
3080
3081/**
3082 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3083 * @skb: buffer to pad
3084 * @len: minimal length
3085 * @free_on_error: free buffer on error
3086 *
3087 * Pads up a buffer to ensure the trailing bytes exist and are
3088 * blanked. If the buffer already contains sufficient data it
3089 * is untouched. Otherwise it is extended. Returns zero on
3090 * success. The skb is freed on error if @free_on_error is true.
3091 */
3092static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3093 bool free_on_error)
3094{
3095 unsigned int size = skb->len;
3096
3097 if (unlikely(size < len)) {
3098 len -= size;
3099 if (__skb_pad(skb, len, free_on_error))
3100 return -ENOMEM;
3101 __skb_put(skb, len);
3102 }
3103 return 0;
3104}
3105
3106/**
3107 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3108 * @skb: buffer to pad
3109 * @len: minimal length
3110 *
3111 * Pads up a buffer to ensure the trailing bytes exist and are
3112 * blanked. If the buffer already contains sufficient data it
3113 * is untouched. Otherwise it is extended. Returns zero on
3114 * success. The skb is freed on error.
3115 */
3116static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3117{
3118 return __skb_put_padto(skb, len, true);
3119}
3120
3121static inline int skb_add_data(struct sk_buff *skb,
3122 struct iov_iter *from, int copy)
3123{
3124 const int off = skb->len;
3125
3126 if (skb->ip_summed == CHECKSUM_NONE) {
3127 __wsum csum = 0;
3128 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3129 &csum, from)) {
3130 skb->csum = csum_block_add(skb->csum, csum, off);
3131 return 0;
3132 }
3133 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3134 return 0;
3135
3136 __skb_trim(skb, off);
3137 return -EFAULT;
3138}
3139
3140static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3141 const struct page *page, int off)
3142{
3143 if (skb_zcopy(skb))
3144 return false;
3145 if (i) {
3146 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3147
3148 return page == skb_frag_page(frag) &&
3149 off == frag->page_offset + skb_frag_size(frag);
3150 }
3151 return false;
3152}
3153
3154static inline int __skb_linearize(struct sk_buff *skb)
3155{
3156 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3157}
3158
3159/**
3160 * skb_linearize - convert paged skb to linear one
3161 * @skb: buffer to linarize
3162 *
3163 * If there is no free memory -ENOMEM is returned, otherwise zero
3164 * is returned and the old skb data released.
3165 */
3166static inline int skb_linearize(struct sk_buff *skb)
3167{
3168 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3169}
3170
3171/**
3172 * skb_has_shared_frag - can any frag be overwritten
3173 * @skb: buffer to test
3174 *
3175 * Return true if the skb has at least one frag that might be modified
3176 * by an external entity (as in vmsplice()/sendfile())
3177 */
3178static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3179{
3180 return skb_is_nonlinear(skb) &&
3181 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3182}
3183
3184/**
3185 * skb_linearize_cow - make sure skb is linear and writable
3186 * @skb: buffer to process
3187 *
3188 * If there is no free memory -ENOMEM is returned, otherwise zero
3189 * is returned and the old skb data released.
3190 */
3191static inline int skb_linearize_cow(struct sk_buff *skb)
3192{
3193 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3194 __skb_linearize(skb) : 0;
3195}
3196
3197static __always_inline void
3198__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3199 unsigned int off)
3200{
3201 if (skb->ip_summed == CHECKSUM_COMPLETE)
3202 skb->csum = csum_block_sub(skb->csum,
3203 csum_partial(start, len, 0), off);
3204 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3205 skb_checksum_start_offset(skb) < 0)
3206 skb->ip_summed = CHECKSUM_NONE;
3207}
3208
3209/**
3210 * skb_postpull_rcsum - update checksum for received skb after pull
3211 * @skb: buffer to update
3212 * @start: start of data before pull
3213 * @len: length of data pulled
3214 *
3215 * After doing a pull on a received packet, you need to call this to
3216 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3217 * CHECKSUM_NONE so that it can be recomputed from scratch.
3218 */
3219static inline void skb_postpull_rcsum(struct sk_buff *skb,
3220 const void *start, unsigned int len)
3221{
3222 __skb_postpull_rcsum(skb, start, len, 0);
3223}
3224
3225static __always_inline void
3226__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3227 unsigned int off)
3228{
3229 if (skb->ip_summed == CHECKSUM_COMPLETE)
3230 skb->csum = csum_block_add(skb->csum,
3231 csum_partial(start, len, 0), off);
3232}
3233
3234/**
3235 * skb_postpush_rcsum - update checksum for received skb after push
3236 * @skb: buffer to update
3237 * @start: start of data after push
3238 * @len: length of data pushed
3239 *
3240 * After doing a push on a received packet, you need to call this to
3241 * update the CHECKSUM_COMPLETE checksum.
3242 */
3243static inline void skb_postpush_rcsum(struct sk_buff *skb,
3244 const void *start, unsigned int len)
3245{
3246 __skb_postpush_rcsum(skb, start, len, 0);
3247}
3248
3249void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3250
3251/**
3252 * skb_push_rcsum - push skb and update receive checksum
3253 * @skb: buffer to update
3254 * @len: length of data pulled
3255 *
3256 * This function performs an skb_push on the packet and updates
3257 * the CHECKSUM_COMPLETE checksum. It should be used on
3258 * receive path processing instead of skb_push unless you know
3259 * that the checksum difference is zero (e.g., a valid IP header)
3260 * or you are setting ip_summed to CHECKSUM_NONE.
3261 */
3262static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3263{
3264 skb_push(skb, len);
3265 skb_postpush_rcsum(skb, skb->data, len);
3266 return skb->data;
3267}
3268
3269int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3270/**
3271 * pskb_trim_rcsum - trim received skb and update checksum
3272 * @skb: buffer to trim
3273 * @len: new length
3274 *
3275 * This is exactly the same as pskb_trim except that it ensures the
3276 * checksum of received packets are still valid after the operation.
3277 * It can change skb pointers.
3278 */
3279
3280static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3281{
3282 if (likely(len >= skb->len))
3283 return 0;
3284 return pskb_trim_rcsum_slow(skb, len);
3285}
3286
3287static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3288{
3289 if (skb->ip_summed == CHECKSUM_COMPLETE)
3290 skb->ip_summed = CHECKSUM_NONE;
3291 __skb_trim(skb, len);
3292 return 0;
3293}
3294
3295static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3296{
3297 if (skb->ip_summed == CHECKSUM_COMPLETE)
3298 skb->ip_summed = CHECKSUM_NONE;
3299 return __skb_grow(skb, len);
3300}
3301
3302#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3303#define skb_rb_first(root) rb_to_skb(rb_first(root))
3304#define skb_rb_last(root) rb_to_skb(rb_last(root))
3305#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3306#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3307
3308#define skb_queue_walk(queue, skb) \
3309 for (skb = (queue)->next; \
3310 skb != (struct sk_buff *)(queue); \
3311 skb = skb->next)
3312
3313#define skb_queue_walk_safe(queue, skb, tmp) \
3314 for (skb = (queue)->next, tmp = skb->next; \
3315 skb != (struct sk_buff *)(queue); \
3316 skb = tmp, tmp = skb->next)
3317
3318#define skb_queue_walk_from(queue, skb) \
3319 for (; skb != (struct sk_buff *)(queue); \
3320 skb = skb->next)
3321
3322#define skb_rbtree_walk(skb, root) \
3323 for (skb = skb_rb_first(root); skb != NULL; \
3324 skb = skb_rb_next(skb))
3325
3326#define skb_rbtree_walk_from(skb) \
3327 for (; skb != NULL; \
3328 skb = skb_rb_next(skb))
3329
3330#define skb_rbtree_walk_from_safe(skb, tmp) \
3331 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3332 skb = tmp)
3333
3334#define skb_queue_walk_from_safe(queue, skb, tmp) \
3335 for (tmp = skb->next; \
3336 skb != (struct sk_buff *)(queue); \
3337 skb = tmp, tmp = skb->next)
3338
3339#define skb_queue_reverse_walk(queue, skb) \
3340 for (skb = (queue)->prev; \
3341 skb != (struct sk_buff *)(queue); \
3342 skb = skb->prev)
3343
3344#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3345 for (skb = (queue)->prev, tmp = skb->prev; \
3346 skb != (struct sk_buff *)(queue); \
3347 skb = tmp, tmp = skb->prev)
3348
3349#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3350 for (tmp = skb->prev; \
3351 skb != (struct sk_buff *)(queue); \
3352 skb = tmp, tmp = skb->prev)
3353
3354static inline bool skb_has_frag_list(const struct sk_buff *skb)
3355{
3356 return skb_shinfo(skb)->frag_list != NULL;
3357}
3358
3359static inline void skb_frag_list_init(struct sk_buff *skb)
3360{
3361 skb_shinfo(skb)->frag_list = NULL;
3362}
3363
3364#define skb_walk_frags(skb, iter) \
3365 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3366
3367
3368int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3369 const struct sk_buff *skb);
3370struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3371 struct sk_buff_head *queue,
3372 unsigned int flags,
3373 void (*destructor)(struct sock *sk,
3374 struct sk_buff *skb),
3375 int *peeked, int *off, int *err,
3376 struct sk_buff **last);
3377struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3378 void (*destructor)(struct sock *sk,
3379 struct sk_buff *skb),
3380 int *peeked, int *off, int *err,
3381 struct sk_buff **last);
3382struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3383 void (*destructor)(struct sock *sk,
3384 struct sk_buff *skb),
3385 int *peeked, int *off, int *err);
3386struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3387 int *err);
3388__poll_t datagram_poll(struct file *file, struct socket *sock,
3389 struct poll_table_struct *wait);
3390int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3391 struct iov_iter *to, int size);
3392static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3393 struct msghdr *msg, int size)
3394{
3395 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3396}
3397int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3398 struct msghdr *msg);
3399int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3400 struct iov_iter *to, int len,
3401 struct ahash_request *hash);
3402int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3403 struct iov_iter *from, int len);
3404int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3405void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3406void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3407static inline void skb_free_datagram_locked(struct sock *sk,
3408 struct sk_buff *skb)
3409{
3410 __skb_free_datagram_locked(sk, skb, 0);
3411}
3412int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3413int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3414int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3415__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3416 int len, __wsum csum);
3417int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3418 struct pipe_inode_info *pipe, unsigned int len,
3419 unsigned int flags);
3420int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3421 int len);
3422void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3423unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3424int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3425 int len, int hlen);
3426void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3427int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3428void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3429bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3430bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3431struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3432struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3433int skb_ensure_writable(struct sk_buff *skb, int write_len);
3434int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3435int skb_vlan_pop(struct sk_buff *skb);
3436int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3437struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3438 gfp_t gfp);
3439
3440static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3441{
3442 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3443}
3444
3445static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3446{
3447 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3448}
3449
3450struct skb_checksum_ops {
3451 __wsum (*update)(const void *mem, int len, __wsum wsum);
3452 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3453};
3454
3455extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3456
3457__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3458 __wsum csum, const struct skb_checksum_ops *ops);
3459__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3460 __wsum csum);
3461
3462static inline void * __must_check
3463__skb_header_pointer(const struct sk_buff *skb, int offset,
3464 int len, void *data, int hlen, void *buffer)
3465{
3466 if (hlen - offset >= len)
3467 return data + offset;
3468
3469 if (!skb ||
3470 skb_copy_bits(skb, offset, buffer, len) < 0)
3471 return NULL;
3472
3473 return buffer;
3474}
3475
3476static inline void * __must_check
3477skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3478{
3479 return __skb_header_pointer(skb, offset, len, skb->data,
3480 skb_headlen(skb), buffer);
3481}
3482
3483/**
3484 * skb_needs_linearize - check if we need to linearize a given skb
3485 * depending on the given device features.
3486 * @skb: socket buffer to check
3487 * @features: net device features
3488 *
3489 * Returns true if either:
3490 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3491 * 2. skb is fragmented and the device does not support SG.
3492 */
3493static inline bool skb_needs_linearize(struct sk_buff *skb,
3494 netdev_features_t features)
3495{
3496 return skb_is_nonlinear(skb) &&
3497 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3498 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3499}
3500
3501static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3502 void *to,
3503 const unsigned int len)
3504{
3505 memcpy(to, skb->data, len);
3506}
3507
3508static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3509 const int offset, void *to,
3510 const unsigned int len)
3511{
3512 memcpy(to, skb->data + offset, len);
3513}
3514
3515static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3516 const void *from,
3517 const unsigned int len)
3518{
3519 memcpy(skb->data, from, len);
3520}
3521
3522static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3523 const int offset,
3524 const void *from,
3525 const unsigned int len)
3526{
3527 memcpy(skb->data + offset, from, len);
3528}
3529
3530void skb_init(void);
3531
3532static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3533{
3534 return skb->tstamp;
3535}
3536
3537/**
3538 * skb_get_timestamp - get timestamp from a skb
3539 * @skb: skb to get stamp from
3540 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3541 *
3542 * Timestamps are stored in the skb as offsets to a base timestamp.
3543 * This function converts the offset back to a struct timeval and stores
3544 * it in stamp.
3545 */
3546static inline void skb_get_timestamp(const struct sk_buff *skb,
3547 struct __kernel_old_timeval *stamp)
3548{
3549 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3550}
3551
3552static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3553 struct __kernel_sock_timeval *stamp)
3554{
3555 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3556
3557 stamp->tv_sec = ts.tv_sec;
3558 stamp->tv_usec = ts.tv_nsec / 1000;
3559}
3560
3561static inline void skb_get_timestampns(const struct sk_buff *skb,
3562 struct timespec *stamp)
3563{
3564 *stamp = ktime_to_timespec(skb->tstamp);
3565}
3566
3567static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3568 struct __kernel_timespec *stamp)
3569{
3570 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3571
3572 stamp->tv_sec = ts.tv_sec;
3573 stamp->tv_nsec = ts.tv_nsec;
3574}
3575
3576static inline void __net_timestamp(struct sk_buff *skb)
3577{
3578 skb->tstamp = ktime_get_real();
3579}
3580
3581static inline ktime_t net_timedelta(ktime_t t)
3582{
3583 return ktime_sub(ktime_get_real(), t);
3584}
3585
3586static inline ktime_t net_invalid_timestamp(void)
3587{
3588 return 0;
3589}
3590
3591static inline u8 skb_metadata_len(const struct sk_buff *skb)
3592{
3593 return skb_shinfo(skb)->meta_len;
3594}
3595
3596static inline void *skb_metadata_end(const struct sk_buff *skb)
3597{
3598 return skb_mac_header(skb);
3599}
3600
3601static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3602 const struct sk_buff *skb_b,
3603 u8 meta_len)
3604{
3605 const void *a = skb_metadata_end(skb_a);
3606 const void *b = skb_metadata_end(skb_b);
3607 /* Using more efficient varaiant than plain call to memcmp(). */
3608#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3609 u64 diffs = 0;
3610
3611 switch (meta_len) {
3612#define __it(x, op) (x -= sizeof(u##op))
3613#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3614 case 32: diffs |= __it_diff(a, b, 64);
3615 /* fall through */
3616 case 24: diffs |= __it_diff(a, b, 64);
3617 /* fall through */
3618 case 16: diffs |= __it_diff(a, b, 64);
3619 /* fall through */
3620 case 8: diffs |= __it_diff(a, b, 64);
3621 break;
3622 case 28: diffs |= __it_diff(a, b, 64);
3623 /* fall through */
3624 case 20: diffs |= __it_diff(a, b, 64);
3625 /* fall through */
3626 case 12: diffs |= __it_diff(a, b, 64);
3627 /* fall through */
3628 case 4: diffs |= __it_diff(a, b, 32);
3629 break;
3630 }
3631 return diffs;
3632#else
3633 return memcmp(a - meta_len, b - meta_len, meta_len);
3634#endif
3635}
3636
3637static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3638 const struct sk_buff *skb_b)
3639{
3640 u8 len_a = skb_metadata_len(skb_a);
3641 u8 len_b = skb_metadata_len(skb_b);
3642
3643 if (!(len_a | len_b))
3644 return false;
3645
3646 return len_a != len_b ?
3647 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3648}
3649
3650static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3651{
3652 skb_shinfo(skb)->meta_len = meta_len;
3653}
3654
3655static inline void skb_metadata_clear(struct sk_buff *skb)
3656{
3657 skb_metadata_set(skb, 0);
3658}
3659
3660struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3661
3662#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3663
3664void skb_clone_tx_timestamp(struct sk_buff *skb);
3665bool skb_defer_rx_timestamp(struct sk_buff *skb);
3666
3667#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3668
3669static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3670{
3671}
3672
3673static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3674{
3675 return false;
3676}
3677
3678#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3679
3680/**
3681 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3682 *
3683 * PHY drivers may accept clones of transmitted packets for
3684 * timestamping via their phy_driver.txtstamp method. These drivers
3685 * must call this function to return the skb back to the stack with a
3686 * timestamp.
3687 *
3688 * @skb: clone of the the original outgoing packet
3689 * @hwtstamps: hardware time stamps
3690 *
3691 */
3692void skb_complete_tx_timestamp(struct sk_buff *skb,
3693 struct skb_shared_hwtstamps *hwtstamps);
3694
3695void __skb_tstamp_tx(struct sk_buff *orig_skb,
3696 struct skb_shared_hwtstamps *hwtstamps,
3697 struct sock *sk, int tstype);
3698
3699/**
3700 * skb_tstamp_tx - queue clone of skb with send time stamps
3701 * @orig_skb: the original outgoing packet
3702 * @hwtstamps: hardware time stamps, may be NULL if not available
3703 *
3704 * If the skb has a socket associated, then this function clones the
3705 * skb (thus sharing the actual data and optional structures), stores
3706 * the optional hardware time stamping information (if non NULL) or
3707 * generates a software time stamp (otherwise), then queues the clone
3708 * to the error queue of the socket. Errors are silently ignored.
3709 */
3710void skb_tstamp_tx(struct sk_buff *orig_skb,
3711 struct skb_shared_hwtstamps *hwtstamps);
3712
3713/**
3714 * skb_tx_timestamp() - Driver hook for transmit timestamping
3715 *
3716 * Ethernet MAC Drivers should call this function in their hard_xmit()
3717 * function immediately before giving the sk_buff to the MAC hardware.
3718 *
3719 * Specifically, one should make absolutely sure that this function is
3720 * called before TX completion of this packet can trigger. Otherwise
3721 * the packet could potentially already be freed.
3722 *
3723 * @skb: A socket buffer.
3724 */
3725static inline void skb_tx_timestamp(struct sk_buff *skb)
3726{
3727 skb_clone_tx_timestamp(skb);
3728 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3729 skb_tstamp_tx(skb, NULL);
3730}
3731
3732/**
3733 * skb_complete_wifi_ack - deliver skb with wifi status
3734 *
3735 * @skb: the original outgoing packet
3736 * @acked: ack status
3737 *
3738 */
3739void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3740
3741__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3742__sum16 __skb_checksum_complete(struct sk_buff *skb);
3743
3744static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3745{
3746 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3747 skb->csum_valid ||
3748 (skb->ip_summed == CHECKSUM_PARTIAL &&
3749 skb_checksum_start_offset(skb) >= 0));
3750}
3751
3752/**
3753 * skb_checksum_complete - Calculate checksum of an entire packet
3754 * @skb: packet to process
3755 *
3756 * This function calculates the checksum over the entire packet plus
3757 * the value of skb->csum. The latter can be used to supply the
3758 * checksum of a pseudo header as used by TCP/UDP. It returns the
3759 * checksum.
3760 *
3761 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3762 * this function can be used to verify that checksum on received
3763 * packets. In that case the function should return zero if the
3764 * checksum is correct. In particular, this function will return zero
3765 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3766 * hardware has already verified the correctness of the checksum.
3767 */
3768static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3769{
3770 return skb_csum_unnecessary(skb) ?
3771 0 : __skb_checksum_complete(skb);
3772}
3773
3774static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3775{
3776 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3777 if (skb->csum_level == 0)
3778 skb->ip_summed = CHECKSUM_NONE;
3779 else
3780 skb->csum_level--;
3781 }
3782}
3783
3784static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3785{
3786 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3787 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3788 skb->csum_level++;
3789 } else if (skb->ip_summed == CHECKSUM_NONE) {
3790 skb->ip_summed = CHECKSUM_UNNECESSARY;
3791 skb->csum_level = 0;
3792 }
3793}
3794
3795/* Check if we need to perform checksum complete validation.
3796 *
3797 * Returns true if checksum complete is needed, false otherwise
3798 * (either checksum is unnecessary or zero checksum is allowed).
3799 */
3800static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3801 bool zero_okay,
3802 __sum16 check)
3803{
3804 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3805 skb->csum_valid = 1;
3806 __skb_decr_checksum_unnecessary(skb);
3807 return false;
3808 }
3809
3810 return true;
3811}
3812
3813/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3814 * in checksum_init.
3815 */
3816#define CHECKSUM_BREAK 76
3817
3818/* Unset checksum-complete
3819 *
3820 * Unset checksum complete can be done when packet is being modified
3821 * (uncompressed for instance) and checksum-complete value is
3822 * invalidated.
3823 */
3824static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3825{
3826 if (skb->ip_summed == CHECKSUM_COMPLETE)
3827 skb->ip_summed = CHECKSUM_NONE;
3828}
3829
3830/* Validate (init) checksum based on checksum complete.
3831 *
3832 * Return values:
3833 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3834 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3835 * checksum is stored in skb->csum for use in __skb_checksum_complete
3836 * non-zero: value of invalid checksum
3837 *
3838 */
3839static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3840 bool complete,
3841 __wsum psum)
3842{
3843 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3844 if (!csum_fold(csum_add(psum, skb->csum))) {
3845 skb->csum_valid = 1;
3846 return 0;
3847 }
3848 }
3849
3850 skb->csum = psum;
3851
3852 if (complete || skb->len <= CHECKSUM_BREAK) {
3853 __sum16 csum;
3854
3855 csum = __skb_checksum_complete(skb);
3856 skb->csum_valid = !csum;
3857 return csum;
3858 }
3859
3860 return 0;
3861}
3862
3863static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3864{
3865 return 0;
3866}
3867
3868/* Perform checksum validate (init). Note that this is a macro since we only
3869 * want to calculate the pseudo header which is an input function if necessary.
3870 * First we try to validate without any computation (checksum unnecessary) and
3871 * then calculate based on checksum complete calling the function to compute
3872 * pseudo header.
3873 *
3874 * Return values:
3875 * 0: checksum is validated or try to in skb_checksum_complete
3876 * non-zero: value of invalid checksum
3877 */
3878#define __skb_checksum_validate(skb, proto, complete, \
3879 zero_okay, check, compute_pseudo) \
3880({ \
3881 __sum16 __ret = 0; \
3882 skb->csum_valid = 0; \
3883 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3884 __ret = __skb_checksum_validate_complete(skb, \
3885 complete, compute_pseudo(skb, proto)); \
3886 __ret; \
3887})
3888
3889#define skb_checksum_init(skb, proto, compute_pseudo) \
3890 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3891
3892#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3893 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3894
3895#define skb_checksum_validate(skb, proto, compute_pseudo) \
3896 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3897
3898#define skb_checksum_validate_zero_check(skb, proto, check, \
3899 compute_pseudo) \
3900 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3901
3902#define skb_checksum_simple_validate(skb) \
3903 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3904
3905static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3906{
3907 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3908}
3909
3910static inline void __skb_checksum_convert(struct sk_buff *skb,
3911 __sum16 check, __wsum pseudo)
3912{
3913 skb->csum = ~pseudo;
3914 skb->ip_summed = CHECKSUM_COMPLETE;
3915}
3916
3917#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3918do { \
3919 if (__skb_checksum_convert_check(skb)) \
3920 __skb_checksum_convert(skb, check, \
3921 compute_pseudo(skb, proto)); \
3922} while (0)
3923
3924static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3925 u16 start, u16 offset)
3926{
3927 skb->ip_summed = CHECKSUM_PARTIAL;
3928 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3929 skb->csum_offset = offset - start;
3930}
3931
3932/* Update skbuf and packet to reflect the remote checksum offload operation.
3933 * When called, ptr indicates the starting point for skb->csum when
3934 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3935 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3936 */
3937static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3938 int start, int offset, bool nopartial)
3939{
3940 __wsum delta;
3941
3942 if (!nopartial) {
3943 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3944 return;
3945 }
3946
3947 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3948 __skb_checksum_complete(skb);
3949 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3950 }
3951
3952 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3953
3954 /* Adjust skb->csum since we changed the packet */
3955 skb->csum = csum_add(skb->csum, delta);
3956}
3957
3958static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3959{
3960#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3961 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3962#else
3963 return NULL;
3964#endif
3965}
3966
3967#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3968void nf_conntrack_destroy(struct nf_conntrack *nfct);
3969static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3970{
3971 if (nfct && atomic_dec_and_test(&nfct->use))
3972 nf_conntrack_destroy(nfct);
3973}
3974static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3975{
3976 if (nfct)
3977 atomic_inc(&nfct->use);
3978}
3979#endif
3980
3981#ifdef CONFIG_SKB_EXTENSIONS
3982enum skb_ext_id {
3983#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3984 SKB_EXT_BRIDGE_NF,
3985#endif
3986#ifdef CONFIG_XFRM
3987 SKB_EXT_SEC_PATH,
3988#endif
3989 SKB_EXT_NUM, /* must be last */
3990};
3991
3992/**
3993 * struct skb_ext - sk_buff extensions
3994 * @refcnt: 1 on allocation, deallocated on 0
3995 * @offset: offset to add to @data to obtain extension address
3996 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
3997 * @data: start of extension data, variable sized
3998 *
3999 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4000 * to use 'u8' types while allowing up to 2kb worth of extension data.
4001 */
4002struct skb_ext {
4003 refcount_t refcnt;
4004 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4005 u8 chunks; /* same */
4006 char data[0] __aligned(8);
4007};
4008
4009void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4010void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4011void __skb_ext_put(struct skb_ext *ext);
4012
4013static inline void skb_ext_put(struct sk_buff *skb)
4014{
4015 if (skb->active_extensions)
4016 __skb_ext_put(skb->extensions);
4017}
4018
4019static inline void __skb_ext_copy(struct sk_buff *dst,
4020 const struct sk_buff *src)
4021{
4022 dst->active_extensions = src->active_extensions;
4023
4024 if (src->active_extensions) {
4025 struct skb_ext *ext = src->extensions;
4026
4027 refcount_inc(&ext->refcnt);
4028 dst->extensions = ext;
4029 }
4030}
4031
4032static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4033{
4034 skb_ext_put(dst);
4035 __skb_ext_copy(dst, src);
4036}
4037
4038static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4039{
4040 return !!ext->offset[i];
4041}
4042
4043static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4044{
4045 return skb->active_extensions & (1 << id);
4046}
4047
4048static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4049{
4050 if (skb_ext_exist(skb, id))
4051 __skb_ext_del(skb, id);
4052}
4053
4054static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4055{
4056 if (skb_ext_exist(skb, id)) {
4057 struct skb_ext *ext = skb->extensions;
4058
4059 return (void *)ext + (ext->offset[id] << 3);
4060 }
4061
4062 return NULL;
4063}
4064#else
4065static inline void skb_ext_put(struct sk_buff *skb) {}
4066static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4067static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4068static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4069#endif /* CONFIG_SKB_EXTENSIONS */
4070
4071static inline void nf_reset(struct sk_buff *skb)
4072{
4073#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4074 nf_conntrack_put(skb_nfct(skb));
4075 skb->_nfct = 0;
4076#endif
4077#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4078 skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
4079#endif
4080}
4081
4082static inline void nf_reset_trace(struct sk_buff *skb)
4083{
4084#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4085 skb->nf_trace = 0;
4086#endif
4087}
4088
4089static inline void ipvs_reset(struct sk_buff *skb)
4090{
4091#if IS_ENABLED(CONFIG_IP_VS)
4092 skb->ipvs_property = 0;
4093#endif
4094}
4095
4096/* Note: This doesn't put any conntrack info in dst. */
4097static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4098 bool copy)
4099{
4100#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4101 dst->_nfct = src->_nfct;
4102 nf_conntrack_get(skb_nfct(src));
4103#endif
4104#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4105 if (copy)
4106 dst->nf_trace = src->nf_trace;
4107#endif
4108}
4109
4110static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4111{
4112#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4113 nf_conntrack_put(skb_nfct(dst));
4114#endif
4115 __nf_copy(dst, src, true);
4116}
4117
4118#ifdef CONFIG_NETWORK_SECMARK
4119static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4120{
4121 to->secmark = from->secmark;
4122}
4123
4124static inline void skb_init_secmark(struct sk_buff *skb)
4125{
4126 skb->secmark = 0;
4127}
4128#else
4129static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4130{ }
4131
4132static inline void skb_init_secmark(struct sk_buff *skb)
4133{ }
4134#endif
4135
4136static inline int secpath_exists(const struct sk_buff *skb)
4137{
4138#ifdef CONFIG_XFRM
4139 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4140#else
4141 return 0;
4142#endif
4143}
4144
4145static inline bool skb_irq_freeable(const struct sk_buff *skb)
4146{
4147 return !skb->destructor &&
4148 !secpath_exists(skb) &&
4149 !skb_nfct(skb) &&
4150 !skb->_skb_refdst &&
4151 !skb_has_frag_list(skb);
4152}
4153
4154static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4155{
4156 skb->queue_mapping = queue_mapping;
4157}
4158
4159static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4160{
4161 return skb->queue_mapping;
4162}
4163
4164static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4165{
4166 to->queue_mapping = from->queue_mapping;
4167}
4168
4169static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4170{
4171 skb->queue_mapping = rx_queue + 1;
4172}
4173
4174static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4175{
4176 return skb->queue_mapping - 1;
4177}
4178
4179static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4180{
4181 return skb->queue_mapping != 0;
4182}
4183
4184static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4185{
4186 skb->dst_pending_confirm = val;
4187}
4188
4189static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4190{
4191 return skb->dst_pending_confirm != 0;
4192}
4193
4194static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4195{
4196#ifdef CONFIG_XFRM
4197 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4198#else
4199 return NULL;
4200#endif
4201}
4202
4203/* Keeps track of mac header offset relative to skb->head.
4204 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4205 * For non-tunnel skb it points to skb_mac_header() and for
4206 * tunnel skb it points to outer mac header.
4207 * Keeps track of level of encapsulation of network headers.
4208 */
4209struct skb_gso_cb {
4210 union {
4211 int mac_offset;
4212 int data_offset;
4213 };
4214 int encap_level;
4215 __wsum csum;
4216 __u16 csum_start;
4217};
4218#define SKB_SGO_CB_OFFSET 32
4219#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4220
4221static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4222{
4223 return (skb_mac_header(inner_skb) - inner_skb->head) -
4224 SKB_GSO_CB(inner_skb)->mac_offset;
4225}
4226
4227static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4228{
4229 int new_headroom, headroom;
4230 int ret;
4231
4232 headroom = skb_headroom(skb);
4233 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4234 if (ret)
4235 return ret;
4236
4237 new_headroom = skb_headroom(skb);
4238 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4239 return 0;
4240}
4241
4242static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4243{
4244 /* Do not update partial checksums if remote checksum is enabled. */
4245 if (skb->remcsum_offload)
4246 return;
4247
4248 SKB_GSO_CB(skb)->csum = res;
4249 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4250}
4251
4252/* Compute the checksum for a gso segment. First compute the checksum value
4253 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4254 * then add in skb->csum (checksum from csum_start to end of packet).
4255 * skb->csum and csum_start are then updated to reflect the checksum of the
4256 * resultant packet starting from the transport header-- the resultant checksum
4257 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4258 * header.
4259 */
4260static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4261{
4262 unsigned char *csum_start = skb_transport_header(skb);
4263 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4264 __wsum partial = SKB_GSO_CB(skb)->csum;
4265
4266 SKB_GSO_CB(skb)->csum = res;
4267 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4268
4269 return csum_fold(csum_partial(csum_start, plen, partial));
4270}
4271
4272static inline bool skb_is_gso(const struct sk_buff *skb)
4273{
4274 return skb_shinfo(skb)->gso_size;
4275}
4276
4277/* Note: Should be called only if skb_is_gso(skb) is true */
4278static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4279{
4280 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4281}
4282
4283/* Note: Should be called only if skb_is_gso(skb) is true */
4284static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4285{
4286 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4287}
4288
4289/* Note: Should be called only if skb_is_gso(skb) is true */
4290static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4291{
4292 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4293}
4294
4295static inline void skb_gso_reset(struct sk_buff *skb)
4296{
4297 skb_shinfo(skb)->gso_size = 0;
4298 skb_shinfo(skb)->gso_segs = 0;
4299 skb_shinfo(skb)->gso_type = 0;
4300}
4301
4302static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4303 u16 increment)
4304{
4305 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4306 return;
4307 shinfo->gso_size += increment;
4308}
4309
4310static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4311 u16 decrement)
4312{
4313 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4314 return;
4315 shinfo->gso_size -= decrement;
4316}
4317
4318void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4319
4320static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4321{
4322 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4323 * wanted then gso_type will be set. */
4324 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4325
4326 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4327 unlikely(shinfo->gso_type == 0)) {
4328 __skb_warn_lro_forwarding(skb);
4329 return true;
4330 }
4331 return false;
4332}
4333
4334static inline void skb_forward_csum(struct sk_buff *skb)
4335{
4336 /* Unfortunately we don't support this one. Any brave souls? */
4337 if (skb->ip_summed == CHECKSUM_COMPLETE)
4338 skb->ip_summed = CHECKSUM_NONE;
4339}
4340
4341/**
4342 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4343 * @skb: skb to check
4344 *
4345 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4346 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4347 * use this helper, to document places where we make this assertion.
4348 */
4349static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4350{
4351#ifdef DEBUG
4352 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4353#endif
4354}
4355
4356bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4357
4358int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4359struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4360 unsigned int transport_len,
4361 __sum16(*skb_chkf)(struct sk_buff *skb));
4362
4363/**
4364 * skb_head_is_locked - Determine if the skb->head is locked down
4365 * @skb: skb to check
4366 *
4367 * The head on skbs build around a head frag can be removed if they are
4368 * not cloned. This function returns true if the skb head is locked down
4369 * due to either being allocated via kmalloc, or by being a clone with
4370 * multiple references to the head.
4371 */
4372static inline bool skb_head_is_locked(const struct sk_buff *skb)
4373{
4374 return !skb->head_frag || skb_cloned(skb);
4375}
4376
4377/* Local Checksum Offload.
4378 * Compute outer checksum based on the assumption that the
4379 * inner checksum will be offloaded later.
4380 * See Documentation/networking/checksum-offloads.rst for
4381 * explanation of how this works.
4382 * Fill in outer checksum adjustment (e.g. with sum of outer
4383 * pseudo-header) before calling.
4384 * Also ensure that inner checksum is in linear data area.
4385 */
4386static inline __wsum lco_csum(struct sk_buff *skb)
4387{
4388 unsigned char *csum_start = skb_checksum_start(skb);
4389 unsigned char *l4_hdr = skb_transport_header(skb);
4390 __wsum partial;
4391
4392 /* Start with complement of inner checksum adjustment */
4393 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4394 skb->csum_offset));
4395
4396 /* Add in checksum of our headers (incl. outer checksum
4397 * adjustment filled in by caller) and return result.
4398 */
4399 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4400}
4401
4402#endif /* __KERNEL__ */
4403#endif /* _LINUX_SKBUFF_H */
4404