Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * Definitions for the 'struct sk_buff' memory handlers.
4 *
5 * Authors:
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 */
9
10#ifndef _LINUX_SKBUFF_H
11#define _LINUX_SKBUFF_H
12
13#include <linux/kernel.h>
14#include <linux/compiler.h>
15#include <linux/time.h>
16#include <linux/bug.h>
17#include <linux/bvec.h>
18#include <linux/cache.h>
19#include <linux/rbtree.h>
20#include <linux/socket.h>
21#include <linux/refcount.h>
22
23#include <linux/atomic.h>
24#include <asm/types.h>
25#include <linux/spinlock.h>
26#include <linux/net.h>
27#include <linux/textsearch.h>
28#include <net/checksum.h>
29#include <linux/rcupdate.h>
30#include <linux/hrtimer.h>
31#include <linux/dma-mapping.h>
32#include <linux/netdev_features.h>
33#include <linux/sched.h>
34#include <linux/sched/clock.h>
35#include <net/flow_dissector.h>
36#include <linux/splice.h>
37#include <linux/in6.h>
38#include <linux/if_packet.h>
39#include <net/flow.h>
40#include <net/page_pool.h>
41#if IS_ENABLED(CONFIG_NF_CONNTRACK)
42#include <linux/netfilter/nf_conntrack_common.h>
43#endif
44
45/* The interface for checksum offload between the stack and networking drivers
46 * is as follows...
47 *
48 * A. IP checksum related features
49 *
50 * Drivers advertise checksum offload capabilities in the features of a device.
51 * From the stack's point of view these are capabilities offered by the driver.
52 * A driver typically only advertises features that it is capable of offloading
53 * to its device.
54 *
55 * The checksum related features are:
56 *
57 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
58 * IP (one's complement) checksum for any combination
59 * of protocols or protocol layering. The checksum is
60 * computed and set in a packet per the CHECKSUM_PARTIAL
61 * interface (see below).
62 *
63 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
64 * TCP or UDP packets over IPv4. These are specifically
65 * unencapsulated packets of the form IPv4|TCP or
66 * IPv4|UDP where the Protocol field in the IPv4 header
67 * is TCP or UDP. The IPv4 header may contain IP options.
68 * This feature cannot be set in features for a device
69 * with NETIF_F_HW_CSUM also set. This feature is being
70 * DEPRECATED (see below).
71 *
72 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
73 * TCP or UDP packets over IPv6. These are specifically
74 * unencapsulated packets of the form IPv6|TCP or
75 * IPv6|UDP where the Next Header field in the IPv6
76 * header is either TCP or UDP. IPv6 extension headers
77 * are not supported with this feature. This feature
78 * cannot be set in features for a device with
79 * NETIF_F_HW_CSUM also set. This feature is being
80 * DEPRECATED (see below).
81 *
82 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
83 * This flag is only used to disable the RX checksum
84 * feature for a device. The stack will accept receive
85 * checksum indication in packets received on a device
86 * regardless of whether NETIF_F_RXCSUM is set.
87 *
88 * B. Checksumming of received packets by device. Indication of checksum
89 * verification is set in skb->ip_summed. Possible values are:
90 *
91 * CHECKSUM_NONE:
92 *
93 * Device did not checksum this packet e.g. due to lack of capabilities.
94 * The packet contains full (though not verified) checksum in packet but
95 * not in skb->csum. Thus, skb->csum is undefined in this case.
96 *
97 * CHECKSUM_UNNECESSARY:
98 *
99 * The hardware you're dealing with doesn't calculate the full checksum
100 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
101 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
102 * if their checksums are okay. skb->csum is still undefined in this case
103 * though. A driver or device must never modify the checksum field in the
104 * packet even if checksum is verified.
105 *
106 * CHECKSUM_UNNECESSARY is applicable to following protocols:
107 * TCP: IPv6 and IPv4.
108 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
109 * zero UDP checksum for either IPv4 or IPv6, the networking stack
110 * may perform further validation in this case.
111 * GRE: only if the checksum is present in the header.
112 * SCTP: indicates the CRC in SCTP header has been validated.
113 * FCOE: indicates the CRC in FC frame has been validated.
114 *
115 * skb->csum_level indicates the number of consecutive checksums found in
116 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
117 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
118 * and a device is able to verify the checksums for UDP (possibly zero),
119 * GRE (checksum flag is set) and TCP, skb->csum_level would be set to
120 * two. If the device were only able to verify the UDP checksum and not
121 * GRE, either because it doesn't support GRE checksum or because GRE
122 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
123 * not considered in this case).
124 *
125 * CHECKSUM_COMPLETE:
126 *
127 * This is the most generic way. The device supplied checksum of the _whole_
128 * packet as seen by netif_rx() and fills in skb->csum. This means the
129 * hardware doesn't need to parse L3/L4 headers to implement this.
130 *
131 * Notes:
132 * - Even if device supports only some protocols, but is able to produce
133 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
134 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
135 *
136 * CHECKSUM_PARTIAL:
137 *
138 * A checksum is set up to be offloaded to a device as described in the
139 * output description for CHECKSUM_PARTIAL. This may occur on a packet
140 * received directly from another Linux OS, e.g., a virtualized Linux kernel
141 * on the same host, or it may be set in the input path in GRO or remote
142 * checksum offload. For the purposes of checksum verification, the checksum
143 * referred to by skb->csum_start + skb->csum_offset and any preceding
144 * checksums in the packet are considered verified. Any checksums in the
145 * packet that are after the checksum being offloaded are not considered to
146 * be verified.
147 *
148 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
149 * in the skb->ip_summed for a packet. Values are:
150 *
151 * CHECKSUM_PARTIAL:
152 *
153 * The driver is required to checksum the packet as seen by hard_start_xmit()
154 * from skb->csum_start up to the end, and to record/write the checksum at
155 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
156 * csum_start and csum_offset values are valid values given the length and
157 * offset of the packet, but it should not attempt to validate that the
158 * checksum refers to a legitimate transport layer checksum -- it is the
159 * purview of the stack to validate that csum_start and csum_offset are set
160 * correctly.
161 *
162 * When the stack requests checksum offload for a packet, the driver MUST
163 * ensure that the checksum is set correctly. A driver can either offload the
164 * checksum calculation to the device, or call skb_checksum_help (in the case
165 * that the device does not support offload for a particular checksum).
166 *
167 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
168 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
169 * checksum offload capability.
170 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
171 * on network device checksumming capabilities: if a packet does not match
172 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
173 * csum_not_inet, see item D.) is called to resolve the checksum.
174 *
175 * CHECKSUM_NONE:
176 *
177 * The skb was already checksummed by the protocol, or a checksum is not
178 * required.
179 *
180 * CHECKSUM_UNNECESSARY:
181 *
182 * This has the same meaning as CHECKSUM_NONE for checksum offload on
183 * output.
184 *
185 * CHECKSUM_COMPLETE:
186 * Not used in checksum output. If a driver observes a packet with this value
187 * set in skbuff, it should treat the packet as if CHECKSUM_NONE were set.
188 *
189 * D. Non-IP checksum (CRC) offloads
190 *
191 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
192 * offloading the SCTP CRC in a packet. To perform this offload the stack
193 * will set csum_start and csum_offset accordingly, set ip_summed to
194 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
195 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
196 * A driver that supports both IP checksum offload and SCTP CRC32c offload
197 * must verify which offload is configured for a packet by testing the
198 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
199 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
200 *
201 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
202 * offloading the FCOE CRC in a packet. To perform this offload the stack
203 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
204 * accordingly. Note that there is no indication in the skbuff that the
205 * CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
206 * both IP checksum offload and FCOE CRC offload must verify which offload
207 * is configured for a packet, presumably by inspecting packet headers.
208 *
209 * E. Checksumming on output with GSO.
210 *
211 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
212 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
213 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
214 * part of the GSO operation is implied. If a checksum is being offloaded
215 * with GSO then ip_summed is CHECKSUM_PARTIAL, and both csum_start and
216 * csum_offset are set to refer to the outermost checksum being offloaded
217 * (two offloaded checksums are possible with UDP encapsulation).
218 */
219
220/* Don't change this without changing skb_csum_unnecessary! */
221#define CHECKSUM_NONE 0
222#define CHECKSUM_UNNECESSARY 1
223#define CHECKSUM_COMPLETE 2
224#define CHECKSUM_PARTIAL 3
225
226/* Maximum value in skb->csum_level */
227#define SKB_MAX_CSUM_LEVEL 3
228
229#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
230#define SKB_WITH_OVERHEAD(X) \
231 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
232#define SKB_MAX_ORDER(X, ORDER) \
233 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
234#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
235#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236
237/* return minimum truesize of one skb containing X bytes of data */
238#define SKB_TRUESIZE(X) ((X) + \
239 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
240 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
241
242struct ahash_request;
243struct net_device;
244struct scatterlist;
245struct pipe_inode_info;
246struct iov_iter;
247struct napi_struct;
248struct bpf_prog;
249union bpf_attr;
250struct skb_ext;
251
252#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
253struct nf_bridge_info {
254 enum {
255 BRNF_PROTO_UNCHANGED,
256 BRNF_PROTO_8021Q,
257 BRNF_PROTO_PPPOE
258 } orig_proto:8;
259 u8 pkt_otherhost:1;
260 u8 in_prerouting:1;
261 u8 bridged_dnat:1;
262 __u16 frag_max_size;
263 struct net_device *physindev;
264
265 /* always valid & non-NULL from FORWARD on, for physdev match */
266 struct net_device *physoutdev;
267 union {
268 /* prerouting: detect dnat in orig/reply direction */
269 __be32 ipv4_daddr;
270 struct in6_addr ipv6_daddr;
271
272 /* after prerouting + nat detected: store original source
273 * mac since neigh resolution overwrites it, only used while
274 * skb is out in neigh layer.
275 */
276 char neigh_header[8];
277 };
278};
279#endif
280
281#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
282/* Chain in tc_skb_ext will be used to share the tc chain with
283 * ovs recirc_id. It will be set to the current chain by tc
284 * and read by ovs to recirc_id.
285 */
286struct tc_skb_ext {
287 __u32 chain;
288 __u16 mru;
289 bool post_ct;
290};
291#endif
292
293struct sk_buff_head {
294 /* These two members must be first. */
295 struct sk_buff *next;
296 struct sk_buff *prev;
297
298 __u32 qlen;
299 spinlock_t lock;
300};
301
302struct sk_buff;
303
304/* To allow 64K frame to be packed as single skb without frag_list we
305 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
306 * buffers which do not start on a page boundary.
307 *
308 * Since GRO uses frags we allocate at least 16 regardless of page
309 * size.
310 */
311#if (65536/PAGE_SIZE + 1) < 16
312#define MAX_SKB_FRAGS 16UL
313#else
314#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
315#endif
316extern int sysctl_max_skb_frags;
317
318/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
319 * segment using its current segmentation instead.
320 */
321#define GSO_BY_FRAGS 0xFFFF
322
323typedef struct bio_vec skb_frag_t;
324
325/**
326 * skb_frag_size() - Returns the size of a skb fragment
327 * @frag: skb fragment
328 */
329static inline unsigned int skb_frag_size(const skb_frag_t *frag)
330{
331 return frag->bv_len;
332}
333
334/**
335 * skb_frag_size_set() - Sets the size of a skb fragment
336 * @frag: skb fragment
337 * @size: size of fragment
338 */
339static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
340{
341 frag->bv_len = size;
342}
343
344/**
345 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
346 * @frag: skb fragment
347 * @delta: value to add
348 */
349static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
350{
351 frag->bv_len += delta;
352}
353
354/**
355 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
356 * @frag: skb fragment
357 * @delta: value to subtract
358 */
359static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
360{
361 frag->bv_len -= delta;
362}
363
364/**
365 * skb_frag_must_loop - Test if %p is a high memory page
366 * @p: fragment's page
367 */
368static inline bool skb_frag_must_loop(struct page *p)
369{
370#if defined(CONFIG_HIGHMEM)
371 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
372 return true;
373#endif
374 return false;
375}
376
377/**
378 * skb_frag_foreach_page - loop over pages in a fragment
379 *
380 * @f: skb frag to operate on
381 * @f_off: offset from start of f->bv_page
382 * @f_len: length from f_off to loop over
383 * @p: (temp var) current page
384 * @p_off: (temp var) offset from start of current page,
385 * non-zero only on first page.
386 * @p_len: (temp var) length in current page,
387 * < PAGE_SIZE only on first and last page.
388 * @copied: (temp var) length so far, excluding current p_len.
389 *
390 * A fragment can hold a compound page, in which case per-page
391 * operations, notably kmap_atomic, must be called for each
392 * regular page.
393 */
394#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
395 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
396 p_off = (f_off) & (PAGE_SIZE - 1), \
397 p_len = skb_frag_must_loop(p) ? \
398 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
399 copied = 0; \
400 copied < f_len; \
401 copied += p_len, p++, p_off = 0, \
402 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
403
404#define HAVE_HW_TIME_STAMP
405
406/**
407 * struct skb_shared_hwtstamps - hardware time stamps
408 * @hwtstamp: hardware time stamp transformed into duration
409 * since arbitrary point in time
410 *
411 * Software time stamps generated by ktime_get_real() are stored in
412 * skb->tstamp.
413 *
414 * hwtstamps can only be compared against other hwtstamps from
415 * the same device.
416 *
417 * This structure is attached to packets as part of the
418 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
419 */
420struct skb_shared_hwtstamps {
421 ktime_t hwtstamp;
422};
423
424/* Definitions for tx_flags in struct skb_shared_info */
425enum {
426 /* generate hardware time stamp */
427 SKBTX_HW_TSTAMP = 1 << 0,
428
429 /* generate software time stamp when queueing packet to NIC */
430 SKBTX_SW_TSTAMP = 1 << 1,
431
432 /* device driver is going to provide hardware time stamp */
433 SKBTX_IN_PROGRESS = 1 << 2,
434
435 /* generate wifi status information (where possible) */
436 SKBTX_WIFI_STATUS = 1 << 4,
437
438 /* generate software time stamp when entering packet scheduling */
439 SKBTX_SCHED_TSTAMP = 1 << 6,
440};
441
442#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
443 SKBTX_SCHED_TSTAMP)
444#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
445
446/* Definitions for flags in struct skb_shared_info */
447enum {
448 /* use zcopy routines */
449 SKBFL_ZEROCOPY_ENABLE = BIT(0),
450
451 /* This indicates at least one fragment might be overwritten
452 * (as in vmsplice(), sendfile() ...)
453 * If we need to compute a TX checksum, we'll need to copy
454 * all frags to avoid possible bad checksum
455 */
456 SKBFL_SHARED_FRAG = BIT(1),
457
458 /* segment contains only zerocopy data and should not be
459 * charged to the kernel memory.
460 */
461 SKBFL_PURE_ZEROCOPY = BIT(2),
462};
463
464#define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
465#define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY)
466
467/*
468 * The callback notifies userspace to release buffers when skb DMA is done in
469 * lower device, the skb last reference should be 0 when calling this.
470 * The zerocopy_success argument is true if zero copy transmit occurred,
471 * false on data copy or out of memory error caused by data copy attempt.
472 * The ctx field is used to track device context.
473 * The desc field is used to track userspace buffer index.
474 */
475struct ubuf_info {
476 void (*callback)(struct sk_buff *, struct ubuf_info *,
477 bool zerocopy_success);
478 union {
479 struct {
480 unsigned long desc;
481 void *ctx;
482 };
483 struct {
484 u32 id;
485 u16 len;
486 u16 zerocopy:1;
487 u32 bytelen;
488 };
489 };
490 refcount_t refcnt;
491 u8 flags;
492
493 struct mmpin {
494 struct user_struct *user;
495 unsigned int num_pg;
496 } mmp;
497};
498
499#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
500
501int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
502void mm_unaccount_pinned_pages(struct mmpin *mmp);
503
504struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size);
505struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
506 struct ubuf_info *uarg);
507
508void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
509
510void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
511 bool success);
512
513int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
514int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
515 struct msghdr *msg, int len,
516 struct ubuf_info *uarg);
517
518/* This data is invariant across clones and lives at
519 * the end of the header data, ie. at skb->end.
520 */
521struct skb_shared_info {
522 __u8 flags;
523 __u8 meta_len;
524 __u8 nr_frags;
525 __u8 tx_flags;
526 unsigned short gso_size;
527 /* Warning: this field is not always filled in (UFO)! */
528 unsigned short gso_segs;
529 struct sk_buff *frag_list;
530 struct skb_shared_hwtstamps hwtstamps;
531 unsigned int gso_type;
532 u32 tskey;
533
534 /*
535 * Warning : all fields before dataref are cleared in __alloc_skb()
536 */
537 atomic_t dataref;
538
539 /* Intermediate layers must ensure that destructor_arg
540 * remains valid until skb destructor */
541 void * destructor_arg;
542
543 /* must be last field, see pskb_expand_head() */
544 skb_frag_t frags[MAX_SKB_FRAGS];
545};
546
547/* We divide dataref into two halves. The higher 16 bits hold references
548 * to the payload part of skb->data. The lower 16 bits hold references to
549 * the entire skb->data. A clone of a headerless skb holds the length of
550 * the header in skb->hdr_len.
551 *
552 * All users must obey the rule that the skb->data reference count must be
553 * greater than or equal to the payload reference count.
554 *
555 * Holding a reference to the payload part means that the user does not
556 * care about modifications to the header part of skb->data.
557 */
558#define SKB_DATAREF_SHIFT 16
559#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
560
561
562enum {
563 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
564 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
565 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
566};
567
568enum {
569 SKB_GSO_TCPV4 = 1 << 0,
570
571 /* This indicates the skb is from an untrusted source. */
572 SKB_GSO_DODGY = 1 << 1,
573
574 /* This indicates the tcp segment has CWR set. */
575 SKB_GSO_TCP_ECN = 1 << 2,
576
577 SKB_GSO_TCP_FIXEDID = 1 << 3,
578
579 SKB_GSO_TCPV6 = 1 << 4,
580
581 SKB_GSO_FCOE = 1 << 5,
582
583 SKB_GSO_GRE = 1 << 6,
584
585 SKB_GSO_GRE_CSUM = 1 << 7,
586
587 SKB_GSO_IPXIP4 = 1 << 8,
588
589 SKB_GSO_IPXIP6 = 1 << 9,
590
591 SKB_GSO_UDP_TUNNEL = 1 << 10,
592
593 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
594
595 SKB_GSO_PARTIAL = 1 << 12,
596
597 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
598
599 SKB_GSO_SCTP = 1 << 14,
600
601 SKB_GSO_ESP = 1 << 15,
602
603 SKB_GSO_UDP = 1 << 16,
604
605 SKB_GSO_UDP_L4 = 1 << 17,
606
607 SKB_GSO_FRAGLIST = 1 << 18,
608};
609
610#if BITS_PER_LONG > 32
611#define NET_SKBUFF_DATA_USES_OFFSET 1
612#endif
613
614#ifdef NET_SKBUFF_DATA_USES_OFFSET
615typedef unsigned int sk_buff_data_t;
616#else
617typedef unsigned char *sk_buff_data_t;
618#endif
619
620/**
621 * struct sk_buff - socket buffer
622 * @next: Next buffer in list
623 * @prev: Previous buffer in list
624 * @tstamp: Time we arrived/left
625 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
626 * for retransmit timer
627 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
628 * @list: queue head
629 * @sk: Socket we are owned by
630 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
631 * fragmentation management
632 * @dev: Device we arrived on/are leaving by
633 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
634 * @cb: Control buffer. Free for use by every layer. Put private vars here
635 * @_skb_refdst: destination entry (with norefcount bit)
636 * @sp: the security path, used for xfrm
637 * @len: Length of actual data
638 * @data_len: Data length
639 * @mac_len: Length of link layer header
640 * @hdr_len: writable header length of cloned skb
641 * @csum: Checksum (must include start/offset pair)
642 * @csum_start: Offset from skb->head where checksumming should start
643 * @csum_offset: Offset from csum_start where checksum should be stored
644 * @priority: Packet queueing priority
645 * @ignore_df: allow local fragmentation
646 * @cloned: Head may be cloned (check refcnt to be sure)
647 * @ip_summed: Driver fed us an IP checksum
648 * @nohdr: Payload reference only, must not modify header
649 * @pkt_type: Packet class
650 * @fclone: skbuff clone status
651 * @ipvs_property: skbuff is owned by ipvs
652 * @inner_protocol_type: whether the inner protocol is
653 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
654 * @remcsum_offload: remote checksum offload is enabled
655 * @offload_fwd_mark: Packet was L2-forwarded in hardware
656 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
657 * @tc_skip_classify: do not classify packet. set by IFB device
658 * @tc_at_ingress: used within tc_classify to distinguish in/egress
659 * @redirected: packet was redirected by packet classifier
660 * @from_ingress: packet was redirected from the ingress path
661 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
662 * @peeked: this packet has been seen already, so stats have been
663 * done for it, don't do them again
664 * @nf_trace: netfilter packet trace flag
665 * @protocol: Packet protocol from driver
666 * @destructor: Destruct function
667 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
668 * @_sk_redir: socket redirection information for skmsg
669 * @_nfct: Associated connection, if any (with nfctinfo bits)
670 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
671 * @skb_iif: ifindex of device we arrived on
672 * @tc_index: Traffic control index
673 * @hash: the packet hash
674 * @queue_mapping: Queue mapping for multiqueue devices
675 * @head_frag: skb was allocated from page fragments,
676 * not allocated by kmalloc() or vmalloc().
677 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
678 * @pp_recycle: mark the packet for recycling instead of freeing (implies
679 * page_pool support on driver)
680 * @active_extensions: active extensions (skb_ext_id types)
681 * @ndisc_nodetype: router type (from link layer)
682 * @ooo_okay: allow the mapping of a socket to a queue to be changed
683 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
684 * ports.
685 * @sw_hash: indicates hash was computed in software stack
686 * @wifi_acked_valid: wifi_acked was set
687 * @wifi_acked: whether frame was acked on wifi or not
688 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
689 * @encapsulation: indicates the inner headers in the skbuff are valid
690 * @encap_hdr_csum: software checksum is needed
691 * @csum_valid: checksum is already valid
692 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
693 * @csum_complete_sw: checksum was completed by software
694 * @csum_level: indicates the number of consecutive checksums found in
695 * the packet minus one that have been verified as
696 * CHECKSUM_UNNECESSARY (max 3)
697 * @dst_pending_confirm: need to confirm neighbour
698 * @decrypted: Decrypted SKB
699 * @slow_gro: state present at GRO time, slower prepare step required
700 * @napi_id: id of the NAPI struct this skb came from
701 * @sender_cpu: (aka @napi_id) source CPU in XPS
702 * @secmark: security marking
703 * @mark: Generic packet mark
704 * @reserved_tailroom: (aka @mark) number of bytes of free space available
705 * at the tail of an sk_buff
706 * @vlan_present: VLAN tag is present
707 * @vlan_proto: vlan encapsulation protocol
708 * @vlan_tci: vlan tag control information
709 * @inner_protocol: Protocol (encapsulation)
710 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
711 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
712 * @inner_transport_header: Inner transport layer header (encapsulation)
713 * @inner_network_header: Network layer header (encapsulation)
714 * @inner_mac_header: Link layer header (encapsulation)
715 * @transport_header: Transport layer header
716 * @network_header: Network layer header
717 * @mac_header: Link layer header
718 * @kcov_handle: KCOV remote handle for remote coverage collection
719 * @tail: Tail pointer
720 * @end: End pointer
721 * @head: Head of buffer
722 * @data: Data head pointer
723 * @truesize: Buffer size
724 * @users: User count - see {datagram,tcp}.c
725 * @extensions: allocated extensions, valid if active_extensions is nonzero
726 */
727
728struct sk_buff {
729 union {
730 struct {
731 /* These two members must be first. */
732 struct sk_buff *next;
733 struct sk_buff *prev;
734
735 union {
736 struct net_device *dev;
737 /* Some protocols might use this space to store information,
738 * while device pointer would be NULL.
739 * UDP receive path is one user.
740 */
741 unsigned long dev_scratch;
742 };
743 };
744 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
745 struct list_head list;
746 };
747
748 union {
749 struct sock *sk;
750 int ip_defrag_offset;
751 };
752
753 union {
754 ktime_t tstamp;
755 u64 skb_mstamp_ns; /* earliest departure time */
756 };
757 /*
758 * This is the control buffer. It is free to use for every
759 * layer. Please put your private variables there. If you
760 * want to keep them across layers you have to do a skb_clone()
761 * first. This is owned by whoever has the skb queued ATM.
762 */
763 char cb[48] __aligned(8);
764
765 union {
766 struct {
767 unsigned long _skb_refdst;
768 void (*destructor)(struct sk_buff *skb);
769 };
770 struct list_head tcp_tsorted_anchor;
771#ifdef CONFIG_NET_SOCK_MSG
772 unsigned long _sk_redir;
773#endif
774 };
775
776#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
777 unsigned long _nfct;
778#endif
779 unsigned int len,
780 data_len;
781 __u16 mac_len,
782 hdr_len;
783
784 /* Following fields are _not_ copied in __copy_skb_header()
785 * Note that queue_mapping is here mostly to fill a hole.
786 */
787 __u16 queue_mapping;
788
789/* if you move cloned around you also must adapt those constants */
790#ifdef __BIG_ENDIAN_BITFIELD
791#define CLONED_MASK (1 << 7)
792#else
793#define CLONED_MASK 1
794#endif
795#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
796
797 /* private: */
798 __u8 __cloned_offset[0];
799 /* public: */
800 __u8 cloned:1,
801 nohdr:1,
802 fclone:2,
803 peeked:1,
804 head_frag:1,
805 pfmemalloc:1,
806 pp_recycle:1; /* page_pool recycle indicator */
807#ifdef CONFIG_SKB_EXTENSIONS
808 __u8 active_extensions;
809#endif
810
811 /* fields enclosed in headers_start/headers_end are copied
812 * using a single memcpy() in __copy_skb_header()
813 */
814 /* private: */
815 __u32 headers_start[0];
816 /* public: */
817
818/* if you move pkt_type around you also must adapt those constants */
819#ifdef __BIG_ENDIAN_BITFIELD
820#define PKT_TYPE_MAX (7 << 5)
821#else
822#define PKT_TYPE_MAX 7
823#endif
824#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
825
826 /* private: */
827 __u8 __pkt_type_offset[0];
828 /* public: */
829 __u8 pkt_type:3;
830 __u8 ignore_df:1;
831 __u8 nf_trace:1;
832 __u8 ip_summed:2;
833 __u8 ooo_okay:1;
834
835 __u8 l4_hash:1;
836 __u8 sw_hash:1;
837 __u8 wifi_acked_valid:1;
838 __u8 wifi_acked:1;
839 __u8 no_fcs:1;
840 /* Indicates the inner headers are valid in the skbuff. */
841 __u8 encapsulation:1;
842 __u8 encap_hdr_csum:1;
843 __u8 csum_valid:1;
844
845#ifdef __BIG_ENDIAN_BITFIELD
846#define PKT_VLAN_PRESENT_BIT 7
847#else
848#define PKT_VLAN_PRESENT_BIT 0
849#endif
850#define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
851 /* private: */
852 __u8 __pkt_vlan_present_offset[0];
853 /* public: */
854 __u8 vlan_present:1;
855 __u8 csum_complete_sw:1;
856 __u8 csum_level:2;
857 __u8 csum_not_inet:1;
858 __u8 dst_pending_confirm:1;
859#ifdef CONFIG_IPV6_NDISC_NODETYPE
860 __u8 ndisc_nodetype:2;
861#endif
862
863 __u8 ipvs_property:1;
864 __u8 inner_protocol_type:1;
865 __u8 remcsum_offload:1;
866#ifdef CONFIG_NET_SWITCHDEV
867 __u8 offload_fwd_mark:1;
868 __u8 offload_l3_fwd_mark:1;
869#endif
870#ifdef CONFIG_NET_CLS_ACT
871 __u8 tc_skip_classify:1;
872 __u8 tc_at_ingress:1;
873#endif
874 __u8 redirected:1;
875#ifdef CONFIG_NET_REDIRECT
876 __u8 from_ingress:1;
877#endif
878#ifdef CONFIG_NETFILTER_SKIP_EGRESS
879 __u8 nf_skip_egress:1;
880#endif
881#ifdef CONFIG_TLS_DEVICE
882 __u8 decrypted:1;
883#endif
884 __u8 slow_gro:1;
885
886#ifdef CONFIG_NET_SCHED
887 __u16 tc_index; /* traffic control index */
888#endif
889
890 union {
891 __wsum csum;
892 struct {
893 __u16 csum_start;
894 __u16 csum_offset;
895 };
896 };
897 __u32 priority;
898 int skb_iif;
899 __u32 hash;
900 __be16 vlan_proto;
901 __u16 vlan_tci;
902#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
903 union {
904 unsigned int napi_id;
905 unsigned int sender_cpu;
906 };
907#endif
908#ifdef CONFIG_NETWORK_SECMARK
909 __u32 secmark;
910#endif
911
912 union {
913 __u32 mark;
914 __u32 reserved_tailroom;
915 };
916
917 union {
918 __be16 inner_protocol;
919 __u8 inner_ipproto;
920 };
921
922 __u16 inner_transport_header;
923 __u16 inner_network_header;
924 __u16 inner_mac_header;
925
926 __be16 protocol;
927 __u16 transport_header;
928 __u16 network_header;
929 __u16 mac_header;
930
931#ifdef CONFIG_KCOV
932 u64 kcov_handle;
933#endif
934
935 /* private: */
936 __u32 headers_end[0];
937 /* public: */
938
939 /* These elements must be at the end, see alloc_skb() for details. */
940 sk_buff_data_t tail;
941 sk_buff_data_t end;
942 unsigned char *head,
943 *data;
944 unsigned int truesize;
945 refcount_t users;
946
947#ifdef CONFIG_SKB_EXTENSIONS
948 /* only useable after checking ->active_extensions != 0 */
949 struct skb_ext *extensions;
950#endif
951};
952
953#ifdef __KERNEL__
954/*
955 * Handling routines are only of interest to the kernel
956 */
957
958#define SKB_ALLOC_FCLONE 0x01
959#define SKB_ALLOC_RX 0x02
960#define SKB_ALLOC_NAPI 0x04
961
962/**
963 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
964 * @skb: buffer
965 */
966static inline bool skb_pfmemalloc(const struct sk_buff *skb)
967{
968 return unlikely(skb->pfmemalloc);
969}
970
971/*
972 * skb might have a dst pointer attached, refcounted or not.
973 * _skb_refdst low order bit is set if refcount was _not_ taken
974 */
975#define SKB_DST_NOREF 1UL
976#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
977
978/**
979 * skb_dst - returns skb dst_entry
980 * @skb: buffer
981 *
982 * Returns skb dst_entry, regardless of reference taken or not.
983 */
984static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
985{
986 /* If refdst was not refcounted, check we still are in a
987 * rcu_read_lock section
988 */
989 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
990 !rcu_read_lock_held() &&
991 !rcu_read_lock_bh_held());
992 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
993}
994
995/**
996 * skb_dst_set - sets skb dst
997 * @skb: buffer
998 * @dst: dst entry
999 *
1000 * Sets skb dst, assuming a reference was taken on dst and should
1001 * be released by skb_dst_drop()
1002 */
1003static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1004{
1005 skb->slow_gro |= !!dst;
1006 skb->_skb_refdst = (unsigned long)dst;
1007}
1008
1009/**
1010 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1011 * @skb: buffer
1012 * @dst: dst entry
1013 *
1014 * Sets skb dst, assuming a reference was not taken on dst.
1015 * If dst entry is cached, we do not take reference and dst_release
1016 * will be avoided by refdst_drop. If dst entry is not cached, we take
1017 * reference, so that last dst_release can destroy the dst immediately.
1018 */
1019static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1020{
1021 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1022 skb->slow_gro |= !!dst;
1023 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1024}
1025
1026/**
1027 * skb_dst_is_noref - Test if skb dst isn't refcounted
1028 * @skb: buffer
1029 */
1030static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1031{
1032 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1033}
1034
1035/**
1036 * skb_rtable - Returns the skb &rtable
1037 * @skb: buffer
1038 */
1039static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1040{
1041 return (struct rtable *)skb_dst(skb);
1042}
1043
1044/* For mangling skb->pkt_type from user space side from applications
1045 * such as nft, tc, etc, we only allow a conservative subset of
1046 * possible pkt_types to be set.
1047*/
1048static inline bool skb_pkt_type_ok(u32 ptype)
1049{
1050 return ptype <= PACKET_OTHERHOST;
1051}
1052
1053/**
1054 * skb_napi_id - Returns the skb's NAPI id
1055 * @skb: buffer
1056 */
1057static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1058{
1059#ifdef CONFIG_NET_RX_BUSY_POLL
1060 return skb->napi_id;
1061#else
1062 return 0;
1063#endif
1064}
1065
1066/**
1067 * skb_unref - decrement the skb's reference count
1068 * @skb: buffer
1069 *
1070 * Returns true if we can free the skb.
1071 */
1072static inline bool skb_unref(struct sk_buff *skb)
1073{
1074 if (unlikely(!skb))
1075 return false;
1076 if (likely(refcount_read(&skb->users) == 1))
1077 smp_rmb();
1078 else if (likely(!refcount_dec_and_test(&skb->users)))
1079 return false;
1080
1081 return true;
1082}
1083
1084void skb_release_head_state(struct sk_buff *skb);
1085void kfree_skb(struct sk_buff *skb);
1086void kfree_skb_list(struct sk_buff *segs);
1087void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1088void skb_tx_error(struct sk_buff *skb);
1089
1090#ifdef CONFIG_TRACEPOINTS
1091void consume_skb(struct sk_buff *skb);
1092#else
1093static inline void consume_skb(struct sk_buff *skb)
1094{
1095 return kfree_skb(skb);
1096}
1097#endif
1098
1099void __consume_stateless_skb(struct sk_buff *skb);
1100void __kfree_skb(struct sk_buff *skb);
1101extern struct kmem_cache *skbuff_head_cache;
1102
1103void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1104bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1105 bool *fragstolen, int *delta_truesize);
1106
1107struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1108 int node);
1109struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1110struct sk_buff *build_skb(void *data, unsigned int frag_size);
1111struct sk_buff *build_skb_around(struct sk_buff *skb,
1112 void *data, unsigned int frag_size);
1113
1114struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1115
1116/**
1117 * alloc_skb - allocate a network buffer
1118 * @size: size to allocate
1119 * @priority: allocation mask
1120 *
1121 * This function is a convenient wrapper around __alloc_skb().
1122 */
1123static inline struct sk_buff *alloc_skb(unsigned int size,
1124 gfp_t priority)
1125{
1126 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1127}
1128
1129struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1130 unsigned long data_len,
1131 int max_page_order,
1132 int *errcode,
1133 gfp_t gfp_mask);
1134struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1135
1136/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1137struct sk_buff_fclones {
1138 struct sk_buff skb1;
1139
1140 struct sk_buff skb2;
1141
1142 refcount_t fclone_ref;
1143};
1144
1145/**
1146 * skb_fclone_busy - check if fclone is busy
1147 * @sk: socket
1148 * @skb: buffer
1149 *
1150 * Returns true if skb is a fast clone, and its clone is not freed.
1151 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1152 * so we also check that this didnt happen.
1153 */
1154static inline bool skb_fclone_busy(const struct sock *sk,
1155 const struct sk_buff *skb)
1156{
1157 const struct sk_buff_fclones *fclones;
1158
1159 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1160
1161 return skb->fclone == SKB_FCLONE_ORIG &&
1162 refcount_read(&fclones->fclone_ref) > 1 &&
1163 READ_ONCE(fclones->skb2.sk) == sk;
1164}
1165
1166/**
1167 * alloc_skb_fclone - allocate a network buffer from fclone cache
1168 * @size: size to allocate
1169 * @priority: allocation mask
1170 *
1171 * This function is a convenient wrapper around __alloc_skb().
1172 */
1173static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1174 gfp_t priority)
1175{
1176 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1177}
1178
1179struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1180void skb_headers_offset_update(struct sk_buff *skb, int off);
1181int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1182struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1183void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1184struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1185struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1186 gfp_t gfp_mask, bool fclone);
1187static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1188 gfp_t gfp_mask)
1189{
1190 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1191}
1192
1193int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1194struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1195 unsigned int headroom);
1196struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1197struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1198 int newtailroom, gfp_t priority);
1199int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1200 int offset, int len);
1201int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1202 int offset, int len);
1203int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1204int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1205
1206/**
1207 * skb_pad - zero pad the tail of an skb
1208 * @skb: buffer to pad
1209 * @pad: space to pad
1210 *
1211 * Ensure that a buffer is followed by a padding area that is zero
1212 * filled. Used by network drivers which may DMA or transfer data
1213 * beyond the buffer end onto the wire.
1214 *
1215 * May return error in out of memory cases. The skb is freed on error.
1216 */
1217static inline int skb_pad(struct sk_buff *skb, int pad)
1218{
1219 return __skb_pad(skb, pad, true);
1220}
1221#define dev_kfree_skb(a) consume_skb(a)
1222
1223int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1224 int offset, size_t size);
1225
1226struct skb_seq_state {
1227 __u32 lower_offset;
1228 __u32 upper_offset;
1229 __u32 frag_idx;
1230 __u32 stepped_offset;
1231 struct sk_buff *root_skb;
1232 struct sk_buff *cur_skb;
1233 __u8 *frag_data;
1234 __u32 frag_off;
1235};
1236
1237void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1238 unsigned int to, struct skb_seq_state *st);
1239unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1240 struct skb_seq_state *st);
1241void skb_abort_seq_read(struct skb_seq_state *st);
1242
1243unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1244 unsigned int to, struct ts_config *config);
1245
1246/*
1247 * Packet hash types specify the type of hash in skb_set_hash.
1248 *
1249 * Hash types refer to the protocol layer addresses which are used to
1250 * construct a packet's hash. The hashes are used to differentiate or identify
1251 * flows of the protocol layer for the hash type. Hash types are either
1252 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1253 *
1254 * Properties of hashes:
1255 *
1256 * 1) Two packets in different flows have different hash values
1257 * 2) Two packets in the same flow should have the same hash value
1258 *
1259 * A hash at a higher layer is considered to be more specific. A driver should
1260 * set the most specific hash possible.
1261 *
1262 * A driver cannot indicate a more specific hash than the layer at which a hash
1263 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1264 *
1265 * A driver may indicate a hash level which is less specific than the
1266 * actual layer the hash was computed on. For instance, a hash computed
1267 * at L4 may be considered an L3 hash. This should only be done if the
1268 * driver can't unambiguously determine that the HW computed the hash at
1269 * the higher layer. Note that the "should" in the second property above
1270 * permits this.
1271 */
1272enum pkt_hash_types {
1273 PKT_HASH_TYPE_NONE, /* Undefined type */
1274 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1275 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1276 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1277};
1278
1279static inline void skb_clear_hash(struct sk_buff *skb)
1280{
1281 skb->hash = 0;
1282 skb->sw_hash = 0;
1283 skb->l4_hash = 0;
1284}
1285
1286static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1287{
1288 if (!skb->l4_hash)
1289 skb_clear_hash(skb);
1290}
1291
1292static inline void
1293__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1294{
1295 skb->l4_hash = is_l4;
1296 skb->sw_hash = is_sw;
1297 skb->hash = hash;
1298}
1299
1300static inline void
1301skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1302{
1303 /* Used by drivers to set hash from HW */
1304 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1305}
1306
1307static inline void
1308__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1309{
1310 __skb_set_hash(skb, hash, true, is_l4);
1311}
1312
1313void __skb_get_hash(struct sk_buff *skb);
1314u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1315u32 skb_get_poff(const struct sk_buff *skb);
1316u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1317 const struct flow_keys_basic *keys, int hlen);
1318__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1319 const void *data, int hlen_proto);
1320
1321static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1322 int thoff, u8 ip_proto)
1323{
1324 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1325}
1326
1327void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1328 const struct flow_dissector_key *key,
1329 unsigned int key_count);
1330
1331struct bpf_flow_dissector;
1332bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1333 __be16 proto, int nhoff, int hlen, unsigned int flags);
1334
1335bool __skb_flow_dissect(const struct net *net,
1336 const struct sk_buff *skb,
1337 struct flow_dissector *flow_dissector,
1338 void *target_container, const void *data,
1339 __be16 proto, int nhoff, int hlen, unsigned int flags);
1340
1341static inline bool skb_flow_dissect(const struct sk_buff *skb,
1342 struct flow_dissector *flow_dissector,
1343 void *target_container, unsigned int flags)
1344{
1345 return __skb_flow_dissect(NULL, skb, flow_dissector,
1346 target_container, NULL, 0, 0, 0, flags);
1347}
1348
1349static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1350 struct flow_keys *flow,
1351 unsigned int flags)
1352{
1353 memset(flow, 0, sizeof(*flow));
1354 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1355 flow, NULL, 0, 0, 0, flags);
1356}
1357
1358static inline bool
1359skb_flow_dissect_flow_keys_basic(const struct net *net,
1360 const struct sk_buff *skb,
1361 struct flow_keys_basic *flow,
1362 const void *data, __be16 proto,
1363 int nhoff, int hlen, unsigned int flags)
1364{
1365 memset(flow, 0, sizeof(*flow));
1366 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1367 data, proto, nhoff, hlen, flags);
1368}
1369
1370void skb_flow_dissect_meta(const struct sk_buff *skb,
1371 struct flow_dissector *flow_dissector,
1372 void *target_container);
1373
1374/* Gets a skb connection tracking info, ctinfo map should be a
1375 * map of mapsize to translate enum ip_conntrack_info states
1376 * to user states.
1377 */
1378void
1379skb_flow_dissect_ct(const struct sk_buff *skb,
1380 struct flow_dissector *flow_dissector,
1381 void *target_container,
1382 u16 *ctinfo_map, size_t mapsize,
1383 bool post_ct);
1384void
1385skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1386 struct flow_dissector *flow_dissector,
1387 void *target_container);
1388
1389void skb_flow_dissect_hash(const struct sk_buff *skb,
1390 struct flow_dissector *flow_dissector,
1391 void *target_container);
1392
1393static inline __u32 skb_get_hash(struct sk_buff *skb)
1394{
1395 if (!skb->l4_hash && !skb->sw_hash)
1396 __skb_get_hash(skb);
1397
1398 return skb->hash;
1399}
1400
1401static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1402{
1403 if (!skb->l4_hash && !skb->sw_hash) {
1404 struct flow_keys keys;
1405 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1406
1407 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1408 }
1409
1410 return skb->hash;
1411}
1412
1413__u32 skb_get_hash_perturb(const struct sk_buff *skb,
1414 const siphash_key_t *perturb);
1415
1416static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1417{
1418 return skb->hash;
1419}
1420
1421static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1422{
1423 to->hash = from->hash;
1424 to->sw_hash = from->sw_hash;
1425 to->l4_hash = from->l4_hash;
1426};
1427
1428static inline void skb_copy_decrypted(struct sk_buff *to,
1429 const struct sk_buff *from)
1430{
1431#ifdef CONFIG_TLS_DEVICE
1432 to->decrypted = from->decrypted;
1433#endif
1434}
1435
1436#ifdef NET_SKBUFF_DATA_USES_OFFSET
1437static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1438{
1439 return skb->head + skb->end;
1440}
1441
1442static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1443{
1444 return skb->end;
1445}
1446#else
1447static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1448{
1449 return skb->end;
1450}
1451
1452static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1453{
1454 return skb->end - skb->head;
1455}
1456#endif
1457
1458/* Internal */
1459#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1460
1461static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1462{
1463 return &skb_shinfo(skb)->hwtstamps;
1464}
1465
1466static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1467{
1468 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1469
1470 return is_zcopy ? skb_uarg(skb) : NULL;
1471}
1472
1473static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1474{
1475 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1476}
1477
1478static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1479 const struct sk_buff *skb2)
1480{
1481 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1482}
1483
1484static inline void net_zcopy_get(struct ubuf_info *uarg)
1485{
1486 refcount_inc(&uarg->refcnt);
1487}
1488
1489static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1490{
1491 skb_shinfo(skb)->destructor_arg = uarg;
1492 skb_shinfo(skb)->flags |= uarg->flags;
1493}
1494
1495static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1496 bool *have_ref)
1497{
1498 if (skb && uarg && !skb_zcopy(skb)) {
1499 if (unlikely(have_ref && *have_ref))
1500 *have_ref = false;
1501 else
1502 net_zcopy_get(uarg);
1503 skb_zcopy_init(skb, uarg);
1504 }
1505}
1506
1507static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1508{
1509 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1510 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1511}
1512
1513static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1514{
1515 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1516}
1517
1518static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1519{
1520 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1521}
1522
1523static inline void net_zcopy_put(struct ubuf_info *uarg)
1524{
1525 if (uarg)
1526 uarg->callback(NULL, uarg, true);
1527}
1528
1529static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1530{
1531 if (uarg) {
1532 if (uarg->callback == msg_zerocopy_callback)
1533 msg_zerocopy_put_abort(uarg, have_uref);
1534 else if (have_uref)
1535 net_zcopy_put(uarg);
1536 }
1537}
1538
1539/* Release a reference on a zerocopy structure */
1540static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1541{
1542 struct ubuf_info *uarg = skb_zcopy(skb);
1543
1544 if (uarg) {
1545 if (!skb_zcopy_is_nouarg(skb))
1546 uarg->callback(skb, uarg, zerocopy_success);
1547
1548 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1549 }
1550}
1551
1552static inline void skb_mark_not_on_list(struct sk_buff *skb)
1553{
1554 skb->next = NULL;
1555}
1556
1557/* Iterate through singly-linked GSO fragments of an skb. */
1558#define skb_list_walk_safe(first, skb, next_skb) \
1559 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1560 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1561
1562static inline void skb_list_del_init(struct sk_buff *skb)
1563{
1564 __list_del_entry(&skb->list);
1565 skb_mark_not_on_list(skb);
1566}
1567
1568/**
1569 * skb_queue_empty - check if a queue is empty
1570 * @list: queue head
1571 *
1572 * Returns true if the queue is empty, false otherwise.
1573 */
1574static inline int skb_queue_empty(const struct sk_buff_head *list)
1575{
1576 return list->next == (const struct sk_buff *) list;
1577}
1578
1579/**
1580 * skb_queue_empty_lockless - check if a queue is empty
1581 * @list: queue head
1582 *
1583 * Returns true if the queue is empty, false otherwise.
1584 * This variant can be used in lockless contexts.
1585 */
1586static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1587{
1588 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1589}
1590
1591
1592/**
1593 * skb_queue_is_last - check if skb is the last entry in the queue
1594 * @list: queue head
1595 * @skb: buffer
1596 *
1597 * Returns true if @skb is the last buffer on the list.
1598 */
1599static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1600 const struct sk_buff *skb)
1601{
1602 return skb->next == (const struct sk_buff *) list;
1603}
1604
1605/**
1606 * skb_queue_is_first - check if skb is the first entry in the queue
1607 * @list: queue head
1608 * @skb: buffer
1609 *
1610 * Returns true if @skb is the first buffer on the list.
1611 */
1612static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1613 const struct sk_buff *skb)
1614{
1615 return skb->prev == (const struct sk_buff *) list;
1616}
1617
1618/**
1619 * skb_queue_next - return the next packet in the queue
1620 * @list: queue head
1621 * @skb: current buffer
1622 *
1623 * Return the next packet in @list after @skb. It is only valid to
1624 * call this if skb_queue_is_last() evaluates to false.
1625 */
1626static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1627 const struct sk_buff *skb)
1628{
1629 /* This BUG_ON may seem severe, but if we just return then we
1630 * are going to dereference garbage.
1631 */
1632 BUG_ON(skb_queue_is_last(list, skb));
1633 return skb->next;
1634}
1635
1636/**
1637 * skb_queue_prev - return the prev packet in the queue
1638 * @list: queue head
1639 * @skb: current buffer
1640 *
1641 * Return the prev packet in @list before @skb. It is only valid to
1642 * call this if skb_queue_is_first() evaluates to false.
1643 */
1644static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1645 const struct sk_buff *skb)
1646{
1647 /* This BUG_ON may seem severe, but if we just return then we
1648 * are going to dereference garbage.
1649 */
1650 BUG_ON(skb_queue_is_first(list, skb));
1651 return skb->prev;
1652}
1653
1654/**
1655 * skb_get - reference buffer
1656 * @skb: buffer to reference
1657 *
1658 * Makes another reference to a socket buffer and returns a pointer
1659 * to the buffer.
1660 */
1661static inline struct sk_buff *skb_get(struct sk_buff *skb)
1662{
1663 refcount_inc(&skb->users);
1664 return skb;
1665}
1666
1667/*
1668 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1669 */
1670
1671/**
1672 * skb_cloned - is the buffer a clone
1673 * @skb: buffer to check
1674 *
1675 * Returns true if the buffer was generated with skb_clone() and is
1676 * one of multiple shared copies of the buffer. Cloned buffers are
1677 * shared data so must not be written to under normal circumstances.
1678 */
1679static inline int skb_cloned(const struct sk_buff *skb)
1680{
1681 return skb->cloned &&
1682 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1683}
1684
1685static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1686{
1687 might_sleep_if(gfpflags_allow_blocking(pri));
1688
1689 if (skb_cloned(skb))
1690 return pskb_expand_head(skb, 0, 0, pri);
1691
1692 return 0;
1693}
1694
1695/* This variant of skb_unclone() makes sure skb->truesize is not changed */
1696static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1697{
1698 might_sleep_if(gfpflags_allow_blocking(pri));
1699
1700 if (skb_cloned(skb)) {
1701 unsigned int save = skb->truesize;
1702 int res;
1703
1704 res = pskb_expand_head(skb, 0, 0, pri);
1705 skb->truesize = save;
1706 return res;
1707 }
1708 return 0;
1709}
1710
1711/**
1712 * skb_header_cloned - is the header a clone
1713 * @skb: buffer to check
1714 *
1715 * Returns true if modifying the header part of the buffer requires
1716 * the data to be copied.
1717 */
1718static inline int skb_header_cloned(const struct sk_buff *skb)
1719{
1720 int dataref;
1721
1722 if (!skb->cloned)
1723 return 0;
1724
1725 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1726 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1727 return dataref != 1;
1728}
1729
1730static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1731{
1732 might_sleep_if(gfpflags_allow_blocking(pri));
1733
1734 if (skb_header_cloned(skb))
1735 return pskb_expand_head(skb, 0, 0, pri);
1736
1737 return 0;
1738}
1739
1740/**
1741 * __skb_header_release - release reference to header
1742 * @skb: buffer to operate on
1743 */
1744static inline void __skb_header_release(struct sk_buff *skb)
1745{
1746 skb->nohdr = 1;
1747 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1748}
1749
1750
1751/**
1752 * skb_shared - is the buffer shared
1753 * @skb: buffer to check
1754 *
1755 * Returns true if more than one person has a reference to this
1756 * buffer.
1757 */
1758static inline int skb_shared(const struct sk_buff *skb)
1759{
1760 return refcount_read(&skb->users) != 1;
1761}
1762
1763/**
1764 * skb_share_check - check if buffer is shared and if so clone it
1765 * @skb: buffer to check
1766 * @pri: priority for memory allocation
1767 *
1768 * If the buffer is shared the buffer is cloned and the old copy
1769 * drops a reference. A new clone with a single reference is returned.
1770 * If the buffer is not shared the original buffer is returned. When
1771 * being called from interrupt status or with spinlocks held pri must
1772 * be GFP_ATOMIC.
1773 *
1774 * NULL is returned on a memory allocation failure.
1775 */
1776static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1777{
1778 might_sleep_if(gfpflags_allow_blocking(pri));
1779 if (skb_shared(skb)) {
1780 struct sk_buff *nskb = skb_clone(skb, pri);
1781
1782 if (likely(nskb))
1783 consume_skb(skb);
1784 else
1785 kfree_skb(skb);
1786 skb = nskb;
1787 }
1788 return skb;
1789}
1790
1791/*
1792 * Copy shared buffers into a new sk_buff. We effectively do COW on
1793 * packets to handle cases where we have a local reader and forward
1794 * and a couple of other messy ones. The normal one is tcpdumping
1795 * a packet thats being forwarded.
1796 */
1797
1798/**
1799 * skb_unshare - make a copy of a shared buffer
1800 * @skb: buffer to check
1801 * @pri: priority for memory allocation
1802 *
1803 * If the socket buffer is a clone then this function creates a new
1804 * copy of the data, drops a reference count on the old copy and returns
1805 * the new copy with the reference count at 1. If the buffer is not a clone
1806 * the original buffer is returned. When called with a spinlock held or
1807 * from interrupt state @pri must be %GFP_ATOMIC
1808 *
1809 * %NULL is returned on a memory allocation failure.
1810 */
1811static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1812 gfp_t pri)
1813{
1814 might_sleep_if(gfpflags_allow_blocking(pri));
1815 if (skb_cloned(skb)) {
1816 struct sk_buff *nskb = skb_copy(skb, pri);
1817
1818 /* Free our shared copy */
1819 if (likely(nskb))
1820 consume_skb(skb);
1821 else
1822 kfree_skb(skb);
1823 skb = nskb;
1824 }
1825 return skb;
1826}
1827
1828/**
1829 * skb_peek - peek at the head of an &sk_buff_head
1830 * @list_: list to peek at
1831 *
1832 * Peek an &sk_buff. Unlike most other operations you _MUST_
1833 * be careful with this one. A peek leaves the buffer on the
1834 * list and someone else may run off with it. You must hold
1835 * the appropriate locks or have a private queue to do this.
1836 *
1837 * Returns %NULL for an empty list or a pointer to the head element.
1838 * The reference count is not incremented and the reference is therefore
1839 * volatile. Use with caution.
1840 */
1841static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1842{
1843 struct sk_buff *skb = list_->next;
1844
1845 if (skb == (struct sk_buff *)list_)
1846 skb = NULL;
1847 return skb;
1848}
1849
1850/**
1851 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1852 * @list_: list to peek at
1853 *
1854 * Like skb_peek(), but the caller knows that the list is not empty.
1855 */
1856static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1857{
1858 return list_->next;
1859}
1860
1861/**
1862 * skb_peek_next - peek skb following the given one from a queue
1863 * @skb: skb to start from
1864 * @list_: list to peek at
1865 *
1866 * Returns %NULL when the end of the list is met or a pointer to the
1867 * next element. The reference count is not incremented and the
1868 * reference is therefore volatile. Use with caution.
1869 */
1870static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1871 const struct sk_buff_head *list_)
1872{
1873 struct sk_buff *next = skb->next;
1874
1875 if (next == (struct sk_buff *)list_)
1876 next = NULL;
1877 return next;
1878}
1879
1880/**
1881 * skb_peek_tail - peek at the tail of an &sk_buff_head
1882 * @list_: list to peek at
1883 *
1884 * Peek an &sk_buff. Unlike most other operations you _MUST_
1885 * be careful with this one. A peek leaves the buffer on the
1886 * list and someone else may run off with it. You must hold
1887 * the appropriate locks or have a private queue to do this.
1888 *
1889 * Returns %NULL for an empty list or a pointer to the tail element.
1890 * The reference count is not incremented and the reference is therefore
1891 * volatile. Use with caution.
1892 */
1893static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1894{
1895 struct sk_buff *skb = READ_ONCE(list_->prev);
1896
1897 if (skb == (struct sk_buff *)list_)
1898 skb = NULL;
1899 return skb;
1900
1901}
1902
1903/**
1904 * skb_queue_len - get queue length
1905 * @list_: list to measure
1906 *
1907 * Return the length of an &sk_buff queue.
1908 */
1909static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1910{
1911 return list_->qlen;
1912}
1913
1914/**
1915 * skb_queue_len_lockless - get queue length
1916 * @list_: list to measure
1917 *
1918 * Return the length of an &sk_buff queue.
1919 * This variant can be used in lockless contexts.
1920 */
1921static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
1922{
1923 return READ_ONCE(list_->qlen);
1924}
1925
1926/**
1927 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1928 * @list: queue to initialize
1929 *
1930 * This initializes only the list and queue length aspects of
1931 * an sk_buff_head object. This allows to initialize the list
1932 * aspects of an sk_buff_head without reinitializing things like
1933 * the spinlock. It can also be used for on-stack sk_buff_head
1934 * objects where the spinlock is known to not be used.
1935 */
1936static inline void __skb_queue_head_init(struct sk_buff_head *list)
1937{
1938 list->prev = list->next = (struct sk_buff *)list;
1939 list->qlen = 0;
1940}
1941
1942/*
1943 * This function creates a split out lock class for each invocation;
1944 * this is needed for now since a whole lot of users of the skb-queue
1945 * infrastructure in drivers have different locking usage (in hardirq)
1946 * than the networking core (in softirq only). In the long run either the
1947 * network layer or drivers should need annotation to consolidate the
1948 * main types of usage into 3 classes.
1949 */
1950static inline void skb_queue_head_init(struct sk_buff_head *list)
1951{
1952 spin_lock_init(&list->lock);
1953 __skb_queue_head_init(list);
1954}
1955
1956static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1957 struct lock_class_key *class)
1958{
1959 skb_queue_head_init(list);
1960 lockdep_set_class(&list->lock, class);
1961}
1962
1963/*
1964 * Insert an sk_buff on a list.
1965 *
1966 * The "__skb_xxxx()" functions are the non-atomic ones that
1967 * can only be called with interrupts disabled.
1968 */
1969static inline void __skb_insert(struct sk_buff *newsk,
1970 struct sk_buff *prev, struct sk_buff *next,
1971 struct sk_buff_head *list)
1972{
1973 /* See skb_queue_empty_lockless() and skb_peek_tail()
1974 * for the opposite READ_ONCE()
1975 */
1976 WRITE_ONCE(newsk->next, next);
1977 WRITE_ONCE(newsk->prev, prev);
1978 WRITE_ONCE(next->prev, newsk);
1979 WRITE_ONCE(prev->next, newsk);
1980 WRITE_ONCE(list->qlen, list->qlen + 1);
1981}
1982
1983static inline void __skb_queue_splice(const struct sk_buff_head *list,
1984 struct sk_buff *prev,
1985 struct sk_buff *next)
1986{
1987 struct sk_buff *first = list->next;
1988 struct sk_buff *last = list->prev;
1989
1990 WRITE_ONCE(first->prev, prev);
1991 WRITE_ONCE(prev->next, first);
1992
1993 WRITE_ONCE(last->next, next);
1994 WRITE_ONCE(next->prev, last);
1995}
1996
1997/**
1998 * skb_queue_splice - join two skb lists, this is designed for stacks
1999 * @list: the new list to add
2000 * @head: the place to add it in the first list
2001 */
2002static inline void skb_queue_splice(const struct sk_buff_head *list,
2003 struct sk_buff_head *head)
2004{
2005 if (!skb_queue_empty(list)) {
2006 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2007 head->qlen += list->qlen;
2008 }
2009}
2010
2011/**
2012 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2013 * @list: the new list to add
2014 * @head: the place to add it in the first list
2015 *
2016 * The list at @list is reinitialised
2017 */
2018static inline void skb_queue_splice_init(struct sk_buff_head *list,
2019 struct sk_buff_head *head)
2020{
2021 if (!skb_queue_empty(list)) {
2022 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2023 head->qlen += list->qlen;
2024 __skb_queue_head_init(list);
2025 }
2026}
2027
2028/**
2029 * skb_queue_splice_tail - join two skb lists, each list being a queue
2030 * @list: the new list to add
2031 * @head: the place to add it in the first list
2032 */
2033static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2034 struct sk_buff_head *head)
2035{
2036 if (!skb_queue_empty(list)) {
2037 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2038 head->qlen += list->qlen;
2039 }
2040}
2041
2042/**
2043 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2044 * @list: the new list to add
2045 * @head: the place to add it in the first list
2046 *
2047 * Each of the lists is a queue.
2048 * The list at @list is reinitialised
2049 */
2050static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2051 struct sk_buff_head *head)
2052{
2053 if (!skb_queue_empty(list)) {
2054 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2055 head->qlen += list->qlen;
2056 __skb_queue_head_init(list);
2057 }
2058}
2059
2060/**
2061 * __skb_queue_after - queue a buffer at the list head
2062 * @list: list to use
2063 * @prev: place after this buffer
2064 * @newsk: buffer to queue
2065 *
2066 * Queue a buffer int the middle of a list. This function takes no locks
2067 * and you must therefore hold required locks before calling it.
2068 *
2069 * A buffer cannot be placed on two lists at the same time.
2070 */
2071static inline void __skb_queue_after(struct sk_buff_head *list,
2072 struct sk_buff *prev,
2073 struct sk_buff *newsk)
2074{
2075 __skb_insert(newsk, prev, prev->next, list);
2076}
2077
2078void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2079 struct sk_buff_head *list);
2080
2081static inline void __skb_queue_before(struct sk_buff_head *list,
2082 struct sk_buff *next,
2083 struct sk_buff *newsk)
2084{
2085 __skb_insert(newsk, next->prev, next, list);
2086}
2087
2088/**
2089 * __skb_queue_head - queue a buffer at the list head
2090 * @list: list to use
2091 * @newsk: buffer to queue
2092 *
2093 * Queue a buffer at the start of a list. This function takes no locks
2094 * and you must therefore hold required locks before calling it.
2095 *
2096 * A buffer cannot be placed on two lists at the same time.
2097 */
2098static inline void __skb_queue_head(struct sk_buff_head *list,
2099 struct sk_buff *newsk)
2100{
2101 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2102}
2103void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2104
2105/**
2106 * __skb_queue_tail - queue a buffer at the list tail
2107 * @list: list to use
2108 * @newsk: buffer to queue
2109 *
2110 * Queue a buffer at the end of a list. This function takes no locks
2111 * and you must therefore hold required locks before calling it.
2112 *
2113 * A buffer cannot be placed on two lists at the same time.
2114 */
2115static inline void __skb_queue_tail(struct sk_buff_head *list,
2116 struct sk_buff *newsk)
2117{
2118 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2119}
2120void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2121
2122/*
2123 * remove sk_buff from list. _Must_ be called atomically, and with
2124 * the list known..
2125 */
2126void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2127static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2128{
2129 struct sk_buff *next, *prev;
2130
2131 WRITE_ONCE(list->qlen, list->qlen - 1);
2132 next = skb->next;
2133 prev = skb->prev;
2134 skb->next = skb->prev = NULL;
2135 WRITE_ONCE(next->prev, prev);
2136 WRITE_ONCE(prev->next, next);
2137}
2138
2139/**
2140 * __skb_dequeue - remove from the head of the queue
2141 * @list: list to dequeue from
2142 *
2143 * Remove the head of the list. This function does not take any locks
2144 * so must be used with appropriate locks held only. The head item is
2145 * returned or %NULL if the list is empty.
2146 */
2147static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2148{
2149 struct sk_buff *skb = skb_peek(list);
2150 if (skb)
2151 __skb_unlink(skb, list);
2152 return skb;
2153}
2154struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2155
2156/**
2157 * __skb_dequeue_tail - remove from the tail of the queue
2158 * @list: list to dequeue from
2159 *
2160 * Remove the tail of the list. This function does not take any locks
2161 * so must be used with appropriate locks held only. The tail item is
2162 * returned or %NULL if the list is empty.
2163 */
2164static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2165{
2166 struct sk_buff *skb = skb_peek_tail(list);
2167 if (skb)
2168 __skb_unlink(skb, list);
2169 return skb;
2170}
2171struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2172
2173
2174static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2175{
2176 return skb->data_len;
2177}
2178
2179static inline unsigned int skb_headlen(const struct sk_buff *skb)
2180{
2181 return skb->len - skb->data_len;
2182}
2183
2184static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2185{
2186 unsigned int i, len = 0;
2187
2188 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2189 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2190 return len;
2191}
2192
2193static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2194{
2195 return skb_headlen(skb) + __skb_pagelen(skb);
2196}
2197
2198/**
2199 * __skb_fill_page_desc - initialise a paged fragment in an skb
2200 * @skb: buffer containing fragment to be initialised
2201 * @i: paged fragment index to initialise
2202 * @page: the page to use for this fragment
2203 * @off: the offset to the data with @page
2204 * @size: the length of the data
2205 *
2206 * Initialises the @i'th fragment of @skb to point to &size bytes at
2207 * offset @off within @page.
2208 *
2209 * Does not take any additional reference on the fragment.
2210 */
2211static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2212 struct page *page, int off, int size)
2213{
2214 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2215
2216 /*
2217 * Propagate page pfmemalloc to the skb if we can. The problem is
2218 * that not all callers have unique ownership of the page but rely
2219 * on page_is_pfmemalloc doing the right thing(tm).
2220 */
2221 frag->bv_page = page;
2222 frag->bv_offset = off;
2223 skb_frag_size_set(frag, size);
2224
2225 page = compound_head(page);
2226 if (page_is_pfmemalloc(page))
2227 skb->pfmemalloc = true;
2228}
2229
2230/**
2231 * skb_fill_page_desc - initialise a paged fragment in an skb
2232 * @skb: buffer containing fragment to be initialised
2233 * @i: paged fragment index to initialise
2234 * @page: the page to use for this fragment
2235 * @off: the offset to the data with @page
2236 * @size: the length of the data
2237 *
2238 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2239 * @skb to point to @size bytes at offset @off within @page. In
2240 * addition updates @skb such that @i is the last fragment.
2241 *
2242 * Does not take any additional reference on the fragment.
2243 */
2244static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2245 struct page *page, int off, int size)
2246{
2247 __skb_fill_page_desc(skb, i, page, off, size);
2248 skb_shinfo(skb)->nr_frags = i + 1;
2249}
2250
2251void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2252 int size, unsigned int truesize);
2253
2254void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2255 unsigned int truesize);
2256
2257#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2258
2259#ifdef NET_SKBUFF_DATA_USES_OFFSET
2260static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2261{
2262 return skb->head + skb->tail;
2263}
2264
2265static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2266{
2267 skb->tail = skb->data - skb->head;
2268}
2269
2270static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2271{
2272 skb_reset_tail_pointer(skb);
2273 skb->tail += offset;
2274}
2275
2276#else /* NET_SKBUFF_DATA_USES_OFFSET */
2277static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2278{
2279 return skb->tail;
2280}
2281
2282static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2283{
2284 skb->tail = skb->data;
2285}
2286
2287static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2288{
2289 skb->tail = skb->data + offset;
2290}
2291
2292#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2293
2294/*
2295 * Add data to an sk_buff
2296 */
2297void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2298void *skb_put(struct sk_buff *skb, unsigned int len);
2299static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2300{
2301 void *tmp = skb_tail_pointer(skb);
2302 SKB_LINEAR_ASSERT(skb);
2303 skb->tail += len;
2304 skb->len += len;
2305 return tmp;
2306}
2307
2308static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2309{
2310 void *tmp = __skb_put(skb, len);
2311
2312 memset(tmp, 0, len);
2313 return tmp;
2314}
2315
2316static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2317 unsigned int len)
2318{
2319 void *tmp = __skb_put(skb, len);
2320
2321 memcpy(tmp, data, len);
2322 return tmp;
2323}
2324
2325static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2326{
2327 *(u8 *)__skb_put(skb, 1) = val;
2328}
2329
2330static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2331{
2332 void *tmp = skb_put(skb, len);
2333
2334 memset(tmp, 0, len);
2335
2336 return tmp;
2337}
2338
2339static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2340 unsigned int len)
2341{
2342 void *tmp = skb_put(skb, len);
2343
2344 memcpy(tmp, data, len);
2345
2346 return tmp;
2347}
2348
2349static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2350{
2351 *(u8 *)skb_put(skb, 1) = val;
2352}
2353
2354void *skb_push(struct sk_buff *skb, unsigned int len);
2355static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2356{
2357 skb->data -= len;
2358 skb->len += len;
2359 return skb->data;
2360}
2361
2362void *skb_pull(struct sk_buff *skb, unsigned int len);
2363static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2364{
2365 skb->len -= len;
2366 BUG_ON(skb->len < skb->data_len);
2367 return skb->data += len;
2368}
2369
2370static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2371{
2372 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2373}
2374
2375void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2376
2377static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2378{
2379 if (len > skb_headlen(skb) &&
2380 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2381 return NULL;
2382 skb->len -= len;
2383 return skb->data += len;
2384}
2385
2386static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2387{
2388 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2389}
2390
2391static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2392{
2393 if (likely(len <= skb_headlen(skb)))
2394 return true;
2395 if (unlikely(len > skb->len))
2396 return false;
2397 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2398}
2399
2400void skb_condense(struct sk_buff *skb);
2401
2402/**
2403 * skb_headroom - bytes at buffer head
2404 * @skb: buffer to check
2405 *
2406 * Return the number of bytes of free space at the head of an &sk_buff.
2407 */
2408static inline unsigned int skb_headroom(const struct sk_buff *skb)
2409{
2410 return skb->data - skb->head;
2411}
2412
2413/**
2414 * skb_tailroom - bytes at buffer end
2415 * @skb: buffer to check
2416 *
2417 * Return the number of bytes of free space at the tail of an sk_buff
2418 */
2419static inline int skb_tailroom(const struct sk_buff *skb)
2420{
2421 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2422}
2423
2424/**
2425 * skb_availroom - bytes at buffer end
2426 * @skb: buffer to check
2427 *
2428 * Return the number of bytes of free space at the tail of an sk_buff
2429 * allocated by sk_stream_alloc()
2430 */
2431static inline int skb_availroom(const struct sk_buff *skb)
2432{
2433 if (skb_is_nonlinear(skb))
2434 return 0;
2435
2436 return skb->end - skb->tail - skb->reserved_tailroom;
2437}
2438
2439/**
2440 * skb_reserve - adjust headroom
2441 * @skb: buffer to alter
2442 * @len: bytes to move
2443 *
2444 * Increase the headroom of an empty &sk_buff by reducing the tail
2445 * room. This is only allowed for an empty buffer.
2446 */
2447static inline void skb_reserve(struct sk_buff *skb, int len)
2448{
2449 skb->data += len;
2450 skb->tail += len;
2451}
2452
2453/**
2454 * skb_tailroom_reserve - adjust reserved_tailroom
2455 * @skb: buffer to alter
2456 * @mtu: maximum amount of headlen permitted
2457 * @needed_tailroom: minimum amount of reserved_tailroom
2458 *
2459 * Set reserved_tailroom so that headlen can be as large as possible but
2460 * not larger than mtu and tailroom cannot be smaller than
2461 * needed_tailroom.
2462 * The required headroom should already have been reserved before using
2463 * this function.
2464 */
2465static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2466 unsigned int needed_tailroom)
2467{
2468 SKB_LINEAR_ASSERT(skb);
2469 if (mtu < skb_tailroom(skb) - needed_tailroom)
2470 /* use at most mtu */
2471 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2472 else
2473 /* use up to all available space */
2474 skb->reserved_tailroom = needed_tailroom;
2475}
2476
2477#define ENCAP_TYPE_ETHER 0
2478#define ENCAP_TYPE_IPPROTO 1
2479
2480static inline void skb_set_inner_protocol(struct sk_buff *skb,
2481 __be16 protocol)
2482{
2483 skb->inner_protocol = protocol;
2484 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2485}
2486
2487static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2488 __u8 ipproto)
2489{
2490 skb->inner_ipproto = ipproto;
2491 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2492}
2493
2494static inline void skb_reset_inner_headers(struct sk_buff *skb)
2495{
2496 skb->inner_mac_header = skb->mac_header;
2497 skb->inner_network_header = skb->network_header;
2498 skb->inner_transport_header = skb->transport_header;
2499}
2500
2501static inline void skb_reset_mac_len(struct sk_buff *skb)
2502{
2503 skb->mac_len = skb->network_header - skb->mac_header;
2504}
2505
2506static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2507 *skb)
2508{
2509 return skb->head + skb->inner_transport_header;
2510}
2511
2512static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2513{
2514 return skb_inner_transport_header(skb) - skb->data;
2515}
2516
2517static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2518{
2519 skb->inner_transport_header = skb->data - skb->head;
2520}
2521
2522static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2523 const int offset)
2524{
2525 skb_reset_inner_transport_header(skb);
2526 skb->inner_transport_header += offset;
2527}
2528
2529static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2530{
2531 return skb->head + skb->inner_network_header;
2532}
2533
2534static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2535{
2536 skb->inner_network_header = skb->data - skb->head;
2537}
2538
2539static inline void skb_set_inner_network_header(struct sk_buff *skb,
2540 const int offset)
2541{
2542 skb_reset_inner_network_header(skb);
2543 skb->inner_network_header += offset;
2544}
2545
2546static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2547{
2548 return skb->head + skb->inner_mac_header;
2549}
2550
2551static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2552{
2553 skb->inner_mac_header = skb->data - skb->head;
2554}
2555
2556static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2557 const int offset)
2558{
2559 skb_reset_inner_mac_header(skb);
2560 skb->inner_mac_header += offset;
2561}
2562static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2563{
2564 return skb->transport_header != (typeof(skb->transport_header))~0U;
2565}
2566
2567static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2568{
2569 return skb->head + skb->transport_header;
2570}
2571
2572static inline void skb_reset_transport_header(struct sk_buff *skb)
2573{
2574 skb->transport_header = skb->data - skb->head;
2575}
2576
2577static inline void skb_set_transport_header(struct sk_buff *skb,
2578 const int offset)
2579{
2580 skb_reset_transport_header(skb);
2581 skb->transport_header += offset;
2582}
2583
2584static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2585{
2586 return skb->head + skb->network_header;
2587}
2588
2589static inline void skb_reset_network_header(struct sk_buff *skb)
2590{
2591 skb->network_header = skb->data - skb->head;
2592}
2593
2594static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2595{
2596 skb_reset_network_header(skb);
2597 skb->network_header += offset;
2598}
2599
2600static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2601{
2602 return skb->head + skb->mac_header;
2603}
2604
2605static inline int skb_mac_offset(const struct sk_buff *skb)
2606{
2607 return skb_mac_header(skb) - skb->data;
2608}
2609
2610static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2611{
2612 return skb->network_header - skb->mac_header;
2613}
2614
2615static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2616{
2617 return skb->mac_header != (typeof(skb->mac_header))~0U;
2618}
2619
2620static inline void skb_unset_mac_header(struct sk_buff *skb)
2621{
2622 skb->mac_header = (typeof(skb->mac_header))~0U;
2623}
2624
2625static inline void skb_reset_mac_header(struct sk_buff *skb)
2626{
2627 skb->mac_header = skb->data - skb->head;
2628}
2629
2630static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2631{
2632 skb_reset_mac_header(skb);
2633 skb->mac_header += offset;
2634}
2635
2636static inline void skb_pop_mac_header(struct sk_buff *skb)
2637{
2638 skb->mac_header = skb->network_header;
2639}
2640
2641static inline void skb_probe_transport_header(struct sk_buff *skb)
2642{
2643 struct flow_keys_basic keys;
2644
2645 if (skb_transport_header_was_set(skb))
2646 return;
2647
2648 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2649 NULL, 0, 0, 0, 0))
2650 skb_set_transport_header(skb, keys.control.thoff);
2651}
2652
2653static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2654{
2655 if (skb_mac_header_was_set(skb)) {
2656 const unsigned char *old_mac = skb_mac_header(skb);
2657
2658 skb_set_mac_header(skb, -skb->mac_len);
2659 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2660 }
2661}
2662
2663static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2664{
2665 return skb->csum_start - skb_headroom(skb);
2666}
2667
2668static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2669{
2670 return skb->head + skb->csum_start;
2671}
2672
2673static inline int skb_transport_offset(const struct sk_buff *skb)
2674{
2675 return skb_transport_header(skb) - skb->data;
2676}
2677
2678static inline u32 skb_network_header_len(const struct sk_buff *skb)
2679{
2680 return skb->transport_header - skb->network_header;
2681}
2682
2683static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2684{
2685 return skb->inner_transport_header - skb->inner_network_header;
2686}
2687
2688static inline int skb_network_offset(const struct sk_buff *skb)
2689{
2690 return skb_network_header(skb) - skb->data;
2691}
2692
2693static inline int skb_inner_network_offset(const struct sk_buff *skb)
2694{
2695 return skb_inner_network_header(skb) - skb->data;
2696}
2697
2698static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2699{
2700 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2701}
2702
2703/*
2704 * CPUs often take a performance hit when accessing unaligned memory
2705 * locations. The actual performance hit varies, it can be small if the
2706 * hardware handles it or large if we have to take an exception and fix it
2707 * in software.
2708 *
2709 * Since an ethernet header is 14 bytes network drivers often end up with
2710 * the IP header at an unaligned offset. The IP header can be aligned by
2711 * shifting the start of the packet by 2 bytes. Drivers should do this
2712 * with:
2713 *
2714 * skb_reserve(skb, NET_IP_ALIGN);
2715 *
2716 * The downside to this alignment of the IP header is that the DMA is now
2717 * unaligned. On some architectures the cost of an unaligned DMA is high
2718 * and this cost outweighs the gains made by aligning the IP header.
2719 *
2720 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2721 * to be overridden.
2722 */
2723#ifndef NET_IP_ALIGN
2724#define NET_IP_ALIGN 2
2725#endif
2726
2727/*
2728 * The networking layer reserves some headroom in skb data (via
2729 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2730 * the header has to grow. In the default case, if the header has to grow
2731 * 32 bytes or less we avoid the reallocation.
2732 *
2733 * Unfortunately this headroom changes the DMA alignment of the resulting
2734 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2735 * on some architectures. An architecture can override this value,
2736 * perhaps setting it to a cacheline in size (since that will maintain
2737 * cacheline alignment of the DMA). It must be a power of 2.
2738 *
2739 * Various parts of the networking layer expect at least 32 bytes of
2740 * headroom, you should not reduce this.
2741 *
2742 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2743 * to reduce average number of cache lines per packet.
2744 * get_rps_cpu() for example only access one 64 bytes aligned block :
2745 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2746 */
2747#ifndef NET_SKB_PAD
2748#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2749#endif
2750
2751int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2752
2753static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2754{
2755 if (WARN_ON(skb_is_nonlinear(skb)))
2756 return;
2757 skb->len = len;
2758 skb_set_tail_pointer(skb, len);
2759}
2760
2761static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2762{
2763 __skb_set_length(skb, len);
2764}
2765
2766void skb_trim(struct sk_buff *skb, unsigned int len);
2767
2768static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2769{
2770 if (skb->data_len)
2771 return ___pskb_trim(skb, len);
2772 __skb_trim(skb, len);
2773 return 0;
2774}
2775
2776static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2777{
2778 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2779}
2780
2781/**
2782 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2783 * @skb: buffer to alter
2784 * @len: new length
2785 *
2786 * This is identical to pskb_trim except that the caller knows that
2787 * the skb is not cloned so we should never get an error due to out-
2788 * of-memory.
2789 */
2790static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2791{
2792 int err = pskb_trim(skb, len);
2793 BUG_ON(err);
2794}
2795
2796static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2797{
2798 unsigned int diff = len - skb->len;
2799
2800 if (skb_tailroom(skb) < diff) {
2801 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2802 GFP_ATOMIC);
2803 if (ret)
2804 return ret;
2805 }
2806 __skb_set_length(skb, len);
2807 return 0;
2808}
2809
2810/**
2811 * skb_orphan - orphan a buffer
2812 * @skb: buffer to orphan
2813 *
2814 * If a buffer currently has an owner then we call the owner's
2815 * destructor function and make the @skb unowned. The buffer continues
2816 * to exist but is no longer charged to its former owner.
2817 */
2818static inline void skb_orphan(struct sk_buff *skb)
2819{
2820 if (skb->destructor) {
2821 skb->destructor(skb);
2822 skb->destructor = NULL;
2823 skb->sk = NULL;
2824 } else {
2825 BUG_ON(skb->sk);
2826 }
2827}
2828
2829/**
2830 * skb_orphan_frags - orphan the frags contained in a buffer
2831 * @skb: buffer to orphan frags from
2832 * @gfp_mask: allocation mask for replacement pages
2833 *
2834 * For each frag in the SKB which needs a destructor (i.e. has an
2835 * owner) create a copy of that frag and release the original
2836 * page by calling the destructor.
2837 */
2838static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2839{
2840 if (likely(!skb_zcopy(skb)))
2841 return 0;
2842 if (!skb_zcopy_is_nouarg(skb) &&
2843 skb_uarg(skb)->callback == msg_zerocopy_callback)
2844 return 0;
2845 return skb_copy_ubufs(skb, gfp_mask);
2846}
2847
2848/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2849static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2850{
2851 if (likely(!skb_zcopy(skb)))
2852 return 0;
2853 return skb_copy_ubufs(skb, gfp_mask);
2854}
2855
2856/**
2857 * __skb_queue_purge - empty a list
2858 * @list: list to empty
2859 *
2860 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2861 * the list and one reference dropped. This function does not take the
2862 * list lock and the caller must hold the relevant locks to use it.
2863 */
2864static inline void __skb_queue_purge(struct sk_buff_head *list)
2865{
2866 struct sk_buff *skb;
2867 while ((skb = __skb_dequeue(list)) != NULL)
2868 kfree_skb(skb);
2869}
2870void skb_queue_purge(struct sk_buff_head *list);
2871
2872unsigned int skb_rbtree_purge(struct rb_root *root);
2873
2874void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
2875
2876/**
2877 * netdev_alloc_frag - allocate a page fragment
2878 * @fragsz: fragment size
2879 *
2880 * Allocates a frag from a page for receive buffer.
2881 * Uses GFP_ATOMIC allocations.
2882 */
2883static inline void *netdev_alloc_frag(unsigned int fragsz)
2884{
2885 return __netdev_alloc_frag_align(fragsz, ~0u);
2886}
2887
2888static inline void *netdev_alloc_frag_align(unsigned int fragsz,
2889 unsigned int align)
2890{
2891 WARN_ON_ONCE(!is_power_of_2(align));
2892 return __netdev_alloc_frag_align(fragsz, -align);
2893}
2894
2895struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2896 gfp_t gfp_mask);
2897
2898/**
2899 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2900 * @dev: network device to receive on
2901 * @length: length to allocate
2902 *
2903 * Allocate a new &sk_buff and assign it a usage count of one. The
2904 * buffer has unspecified headroom built in. Users should allocate
2905 * the headroom they think they need without accounting for the
2906 * built in space. The built in space is used for optimisations.
2907 *
2908 * %NULL is returned if there is no free memory. Although this function
2909 * allocates memory it can be called from an interrupt.
2910 */
2911static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2912 unsigned int length)
2913{
2914 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2915}
2916
2917/* legacy helper around __netdev_alloc_skb() */
2918static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2919 gfp_t gfp_mask)
2920{
2921 return __netdev_alloc_skb(NULL, length, gfp_mask);
2922}
2923
2924/* legacy helper around netdev_alloc_skb() */
2925static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2926{
2927 return netdev_alloc_skb(NULL, length);
2928}
2929
2930
2931static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2932 unsigned int length, gfp_t gfp)
2933{
2934 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2935
2936 if (NET_IP_ALIGN && skb)
2937 skb_reserve(skb, NET_IP_ALIGN);
2938 return skb;
2939}
2940
2941static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2942 unsigned int length)
2943{
2944 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2945}
2946
2947static inline void skb_free_frag(void *addr)
2948{
2949 page_frag_free(addr);
2950}
2951
2952void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
2953
2954static inline void *napi_alloc_frag(unsigned int fragsz)
2955{
2956 return __napi_alloc_frag_align(fragsz, ~0u);
2957}
2958
2959static inline void *napi_alloc_frag_align(unsigned int fragsz,
2960 unsigned int align)
2961{
2962 WARN_ON_ONCE(!is_power_of_2(align));
2963 return __napi_alloc_frag_align(fragsz, -align);
2964}
2965
2966struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2967 unsigned int length, gfp_t gfp_mask);
2968static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2969 unsigned int length)
2970{
2971 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2972}
2973void napi_consume_skb(struct sk_buff *skb, int budget);
2974
2975void napi_skb_free_stolen_head(struct sk_buff *skb);
2976void __kfree_skb_defer(struct sk_buff *skb);
2977
2978/**
2979 * __dev_alloc_pages - allocate page for network Rx
2980 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2981 * @order: size of the allocation
2982 *
2983 * Allocate a new page.
2984 *
2985 * %NULL is returned if there is no free memory.
2986*/
2987static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2988 unsigned int order)
2989{
2990 /* This piece of code contains several assumptions.
2991 * 1. This is for device Rx, therefor a cold page is preferred.
2992 * 2. The expectation is the user wants a compound page.
2993 * 3. If requesting a order 0 page it will not be compound
2994 * due to the check to see if order has a value in prep_new_page
2995 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2996 * code in gfp_to_alloc_flags that should be enforcing this.
2997 */
2998 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2999
3000 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3001}
3002
3003static inline struct page *dev_alloc_pages(unsigned int order)
3004{
3005 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3006}
3007
3008/**
3009 * __dev_alloc_page - allocate a page for network Rx
3010 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3011 *
3012 * Allocate a new page.
3013 *
3014 * %NULL is returned if there is no free memory.
3015 */
3016static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3017{
3018 return __dev_alloc_pages(gfp_mask, 0);
3019}
3020
3021static inline struct page *dev_alloc_page(void)
3022{
3023 return dev_alloc_pages(0);
3024}
3025
3026/**
3027 * dev_page_is_reusable - check whether a page can be reused for network Rx
3028 * @page: the page to test
3029 *
3030 * A page shouldn't be considered for reusing/recycling if it was allocated
3031 * under memory pressure or at a distant memory node.
3032 *
3033 * Returns false if this page should be returned to page allocator, true
3034 * otherwise.
3035 */
3036static inline bool dev_page_is_reusable(const struct page *page)
3037{
3038 return likely(page_to_nid(page) == numa_mem_id() &&
3039 !page_is_pfmemalloc(page));
3040}
3041
3042/**
3043 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3044 * @page: The page that was allocated from skb_alloc_page
3045 * @skb: The skb that may need pfmemalloc set
3046 */
3047static inline void skb_propagate_pfmemalloc(const struct page *page,
3048 struct sk_buff *skb)
3049{
3050 if (page_is_pfmemalloc(page))
3051 skb->pfmemalloc = true;
3052}
3053
3054/**
3055 * skb_frag_off() - Returns the offset of a skb fragment
3056 * @frag: the paged fragment
3057 */
3058static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3059{
3060 return frag->bv_offset;
3061}
3062
3063/**
3064 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3065 * @frag: skb fragment
3066 * @delta: value to add
3067 */
3068static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3069{
3070 frag->bv_offset += delta;
3071}
3072
3073/**
3074 * skb_frag_off_set() - Sets the offset of a skb fragment
3075 * @frag: skb fragment
3076 * @offset: offset of fragment
3077 */
3078static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3079{
3080 frag->bv_offset = offset;
3081}
3082
3083/**
3084 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3085 * @fragto: skb fragment where offset is set
3086 * @fragfrom: skb fragment offset is copied from
3087 */
3088static inline void skb_frag_off_copy(skb_frag_t *fragto,
3089 const skb_frag_t *fragfrom)
3090{
3091 fragto->bv_offset = fragfrom->bv_offset;
3092}
3093
3094/**
3095 * skb_frag_page - retrieve the page referred to by a paged fragment
3096 * @frag: the paged fragment
3097 *
3098 * Returns the &struct page associated with @frag.
3099 */
3100static inline struct page *skb_frag_page(const skb_frag_t *frag)
3101{
3102 return frag->bv_page;
3103}
3104
3105/**
3106 * __skb_frag_ref - take an addition reference on a paged fragment.
3107 * @frag: the paged fragment
3108 *
3109 * Takes an additional reference on the paged fragment @frag.
3110 */
3111static inline void __skb_frag_ref(skb_frag_t *frag)
3112{
3113 get_page(skb_frag_page(frag));
3114}
3115
3116/**
3117 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3118 * @skb: the buffer
3119 * @f: the fragment offset.
3120 *
3121 * Takes an additional reference on the @f'th paged fragment of @skb.
3122 */
3123static inline void skb_frag_ref(struct sk_buff *skb, int f)
3124{
3125 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3126}
3127
3128/**
3129 * __skb_frag_unref - release a reference on a paged fragment.
3130 * @frag: the paged fragment
3131 * @recycle: recycle the page if allocated via page_pool
3132 *
3133 * Releases a reference on the paged fragment @frag
3134 * or recycles the page via the page_pool API.
3135 */
3136static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3137{
3138 struct page *page = skb_frag_page(frag);
3139
3140#ifdef CONFIG_PAGE_POOL
3141 if (recycle && page_pool_return_skb_page(page))
3142 return;
3143#endif
3144 put_page(page);
3145}
3146
3147/**
3148 * skb_frag_unref - release a reference on a paged fragment of an skb.
3149 * @skb: the buffer
3150 * @f: the fragment offset
3151 *
3152 * Releases a reference on the @f'th paged fragment of @skb.
3153 */
3154static inline void skb_frag_unref(struct sk_buff *skb, int f)
3155{
3156 __skb_frag_unref(&skb_shinfo(skb)->frags[f], skb->pp_recycle);
3157}
3158
3159/**
3160 * skb_frag_address - gets the address of the data contained in a paged fragment
3161 * @frag: the paged fragment buffer
3162 *
3163 * Returns the address of the data within @frag. The page must already
3164 * be mapped.
3165 */
3166static inline void *skb_frag_address(const skb_frag_t *frag)
3167{
3168 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3169}
3170
3171/**
3172 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3173 * @frag: the paged fragment buffer
3174 *
3175 * Returns the address of the data within @frag. Checks that the page
3176 * is mapped and returns %NULL otherwise.
3177 */
3178static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3179{
3180 void *ptr = page_address(skb_frag_page(frag));
3181 if (unlikely(!ptr))
3182 return NULL;
3183
3184 return ptr + skb_frag_off(frag);
3185}
3186
3187/**
3188 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3189 * @fragto: skb fragment where page is set
3190 * @fragfrom: skb fragment page is copied from
3191 */
3192static inline void skb_frag_page_copy(skb_frag_t *fragto,
3193 const skb_frag_t *fragfrom)
3194{
3195 fragto->bv_page = fragfrom->bv_page;
3196}
3197
3198/**
3199 * __skb_frag_set_page - sets the page contained in a paged fragment
3200 * @frag: the paged fragment
3201 * @page: the page to set
3202 *
3203 * Sets the fragment @frag to contain @page.
3204 */
3205static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3206{
3207 frag->bv_page = page;
3208}
3209
3210/**
3211 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3212 * @skb: the buffer
3213 * @f: the fragment offset
3214 * @page: the page to set
3215 *
3216 * Sets the @f'th fragment of @skb to contain @page.
3217 */
3218static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3219 struct page *page)
3220{
3221 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3222}
3223
3224bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3225
3226/**
3227 * skb_frag_dma_map - maps a paged fragment via the DMA API
3228 * @dev: the device to map the fragment to
3229 * @frag: the paged fragment to map
3230 * @offset: the offset within the fragment (starting at the
3231 * fragment's own offset)
3232 * @size: the number of bytes to map
3233 * @dir: the direction of the mapping (``PCI_DMA_*``)
3234 *
3235 * Maps the page associated with @frag to @device.
3236 */
3237static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3238 const skb_frag_t *frag,
3239 size_t offset, size_t size,
3240 enum dma_data_direction dir)
3241{
3242 return dma_map_page(dev, skb_frag_page(frag),
3243 skb_frag_off(frag) + offset, size, dir);
3244}
3245
3246static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3247 gfp_t gfp_mask)
3248{
3249 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3250}
3251
3252
3253static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3254 gfp_t gfp_mask)
3255{
3256 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3257}
3258
3259
3260/**
3261 * skb_clone_writable - is the header of a clone writable
3262 * @skb: buffer to check
3263 * @len: length up to which to write
3264 *
3265 * Returns true if modifying the header part of the cloned buffer
3266 * does not requires the data to be copied.
3267 */
3268static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3269{
3270 return !skb_header_cloned(skb) &&
3271 skb_headroom(skb) + len <= skb->hdr_len;
3272}
3273
3274static inline int skb_try_make_writable(struct sk_buff *skb,
3275 unsigned int write_len)
3276{
3277 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3278 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3279}
3280
3281static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3282 int cloned)
3283{
3284 int delta = 0;
3285
3286 if (headroom > skb_headroom(skb))
3287 delta = headroom - skb_headroom(skb);
3288
3289 if (delta || cloned)
3290 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3291 GFP_ATOMIC);
3292 return 0;
3293}
3294
3295/**
3296 * skb_cow - copy header of skb when it is required
3297 * @skb: buffer to cow
3298 * @headroom: needed headroom
3299 *
3300 * If the skb passed lacks sufficient headroom or its data part
3301 * is shared, data is reallocated. If reallocation fails, an error
3302 * is returned and original skb is not changed.
3303 *
3304 * The result is skb with writable area skb->head...skb->tail
3305 * and at least @headroom of space at head.
3306 */
3307static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3308{
3309 return __skb_cow(skb, headroom, skb_cloned(skb));
3310}
3311
3312/**
3313 * skb_cow_head - skb_cow but only making the head writable
3314 * @skb: buffer to cow
3315 * @headroom: needed headroom
3316 *
3317 * This function is identical to skb_cow except that we replace the
3318 * skb_cloned check by skb_header_cloned. It should be used when
3319 * you only need to push on some header and do not need to modify
3320 * the data.
3321 */
3322static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3323{
3324 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3325}
3326
3327/**
3328 * skb_padto - pad an skbuff up to a minimal size
3329 * @skb: buffer to pad
3330 * @len: minimal length
3331 *
3332 * Pads up a buffer to ensure the trailing bytes exist and are
3333 * blanked. If the buffer already contains sufficient data it
3334 * is untouched. Otherwise it is extended. Returns zero on
3335 * success. The skb is freed on error.
3336 */
3337static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3338{
3339 unsigned int size = skb->len;
3340 if (likely(size >= len))
3341 return 0;
3342 return skb_pad(skb, len - size);
3343}
3344
3345/**
3346 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3347 * @skb: buffer to pad
3348 * @len: minimal length
3349 * @free_on_error: free buffer on error
3350 *
3351 * Pads up a buffer to ensure the trailing bytes exist and are
3352 * blanked. If the buffer already contains sufficient data it
3353 * is untouched. Otherwise it is extended. Returns zero on
3354 * success. The skb is freed on error if @free_on_error is true.
3355 */
3356static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3357 unsigned int len,
3358 bool free_on_error)
3359{
3360 unsigned int size = skb->len;
3361
3362 if (unlikely(size < len)) {
3363 len -= size;
3364 if (__skb_pad(skb, len, free_on_error))
3365 return -ENOMEM;
3366 __skb_put(skb, len);
3367 }
3368 return 0;
3369}
3370
3371/**
3372 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3373 * @skb: buffer to pad
3374 * @len: minimal length
3375 *
3376 * Pads up a buffer to ensure the trailing bytes exist and are
3377 * blanked. If the buffer already contains sufficient data it
3378 * is untouched. Otherwise it is extended. Returns zero on
3379 * success. The skb is freed on error.
3380 */
3381static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3382{
3383 return __skb_put_padto(skb, len, true);
3384}
3385
3386static inline int skb_add_data(struct sk_buff *skb,
3387 struct iov_iter *from, int copy)
3388{
3389 const int off = skb->len;
3390
3391 if (skb->ip_summed == CHECKSUM_NONE) {
3392 __wsum csum = 0;
3393 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3394 &csum, from)) {
3395 skb->csum = csum_block_add(skb->csum, csum, off);
3396 return 0;
3397 }
3398 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3399 return 0;
3400
3401 __skb_trim(skb, off);
3402 return -EFAULT;
3403}
3404
3405static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3406 const struct page *page, int off)
3407{
3408 if (skb_zcopy(skb))
3409 return false;
3410 if (i) {
3411 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3412
3413 return page == skb_frag_page(frag) &&
3414 off == skb_frag_off(frag) + skb_frag_size(frag);
3415 }
3416 return false;
3417}
3418
3419static inline int __skb_linearize(struct sk_buff *skb)
3420{
3421 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3422}
3423
3424/**
3425 * skb_linearize - convert paged skb to linear one
3426 * @skb: buffer to linarize
3427 *
3428 * If there is no free memory -ENOMEM is returned, otherwise zero
3429 * is returned and the old skb data released.
3430 */
3431static inline int skb_linearize(struct sk_buff *skb)
3432{
3433 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3434}
3435
3436/**
3437 * skb_has_shared_frag - can any frag be overwritten
3438 * @skb: buffer to test
3439 *
3440 * Return true if the skb has at least one frag that might be modified
3441 * by an external entity (as in vmsplice()/sendfile())
3442 */
3443static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3444{
3445 return skb_is_nonlinear(skb) &&
3446 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3447}
3448
3449/**
3450 * skb_linearize_cow - make sure skb is linear and writable
3451 * @skb: buffer to process
3452 *
3453 * If there is no free memory -ENOMEM is returned, otherwise zero
3454 * is returned and the old skb data released.
3455 */
3456static inline int skb_linearize_cow(struct sk_buff *skb)
3457{
3458 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3459 __skb_linearize(skb) : 0;
3460}
3461
3462static __always_inline void
3463__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3464 unsigned int off)
3465{
3466 if (skb->ip_summed == CHECKSUM_COMPLETE)
3467 skb->csum = csum_block_sub(skb->csum,
3468 csum_partial(start, len, 0), off);
3469 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3470 skb_checksum_start_offset(skb) < 0)
3471 skb->ip_summed = CHECKSUM_NONE;
3472}
3473
3474/**
3475 * skb_postpull_rcsum - update checksum for received skb after pull
3476 * @skb: buffer to update
3477 * @start: start of data before pull
3478 * @len: length of data pulled
3479 *
3480 * After doing a pull on a received packet, you need to call this to
3481 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3482 * CHECKSUM_NONE so that it can be recomputed from scratch.
3483 */
3484static inline void skb_postpull_rcsum(struct sk_buff *skb,
3485 const void *start, unsigned int len)
3486{
3487 __skb_postpull_rcsum(skb, start, len, 0);
3488}
3489
3490static __always_inline void
3491__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3492 unsigned int off)
3493{
3494 if (skb->ip_summed == CHECKSUM_COMPLETE)
3495 skb->csum = csum_block_add(skb->csum,
3496 csum_partial(start, len, 0), off);
3497}
3498
3499/**
3500 * skb_postpush_rcsum - update checksum for received skb after push
3501 * @skb: buffer to update
3502 * @start: start of data after push
3503 * @len: length of data pushed
3504 *
3505 * After doing a push on a received packet, you need to call this to
3506 * update the CHECKSUM_COMPLETE checksum.
3507 */
3508static inline void skb_postpush_rcsum(struct sk_buff *skb,
3509 const void *start, unsigned int len)
3510{
3511 __skb_postpush_rcsum(skb, start, len, 0);
3512}
3513
3514void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3515
3516/**
3517 * skb_push_rcsum - push skb and update receive checksum
3518 * @skb: buffer to update
3519 * @len: length of data pulled
3520 *
3521 * This function performs an skb_push on the packet and updates
3522 * the CHECKSUM_COMPLETE checksum. It should be used on
3523 * receive path processing instead of skb_push unless you know
3524 * that the checksum difference is zero (e.g., a valid IP header)
3525 * or you are setting ip_summed to CHECKSUM_NONE.
3526 */
3527static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3528{
3529 skb_push(skb, len);
3530 skb_postpush_rcsum(skb, skb->data, len);
3531 return skb->data;
3532}
3533
3534int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3535/**
3536 * pskb_trim_rcsum - trim received skb and update checksum
3537 * @skb: buffer to trim
3538 * @len: new length
3539 *
3540 * This is exactly the same as pskb_trim except that it ensures the
3541 * checksum of received packets are still valid after the operation.
3542 * It can change skb pointers.
3543 */
3544
3545static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3546{
3547 if (likely(len >= skb->len))
3548 return 0;
3549 return pskb_trim_rcsum_slow(skb, len);
3550}
3551
3552static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3553{
3554 if (skb->ip_summed == CHECKSUM_COMPLETE)
3555 skb->ip_summed = CHECKSUM_NONE;
3556 __skb_trim(skb, len);
3557 return 0;
3558}
3559
3560static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3561{
3562 if (skb->ip_summed == CHECKSUM_COMPLETE)
3563 skb->ip_summed = CHECKSUM_NONE;
3564 return __skb_grow(skb, len);
3565}
3566
3567#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3568#define skb_rb_first(root) rb_to_skb(rb_first(root))
3569#define skb_rb_last(root) rb_to_skb(rb_last(root))
3570#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3571#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3572
3573#define skb_queue_walk(queue, skb) \
3574 for (skb = (queue)->next; \
3575 skb != (struct sk_buff *)(queue); \
3576 skb = skb->next)
3577
3578#define skb_queue_walk_safe(queue, skb, tmp) \
3579 for (skb = (queue)->next, tmp = skb->next; \
3580 skb != (struct sk_buff *)(queue); \
3581 skb = tmp, tmp = skb->next)
3582
3583#define skb_queue_walk_from(queue, skb) \
3584 for (; skb != (struct sk_buff *)(queue); \
3585 skb = skb->next)
3586
3587#define skb_rbtree_walk(skb, root) \
3588 for (skb = skb_rb_first(root); skb != NULL; \
3589 skb = skb_rb_next(skb))
3590
3591#define skb_rbtree_walk_from(skb) \
3592 for (; skb != NULL; \
3593 skb = skb_rb_next(skb))
3594
3595#define skb_rbtree_walk_from_safe(skb, tmp) \
3596 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3597 skb = tmp)
3598
3599#define skb_queue_walk_from_safe(queue, skb, tmp) \
3600 for (tmp = skb->next; \
3601 skb != (struct sk_buff *)(queue); \
3602 skb = tmp, tmp = skb->next)
3603
3604#define skb_queue_reverse_walk(queue, skb) \
3605 for (skb = (queue)->prev; \
3606 skb != (struct sk_buff *)(queue); \
3607 skb = skb->prev)
3608
3609#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3610 for (skb = (queue)->prev, tmp = skb->prev; \
3611 skb != (struct sk_buff *)(queue); \
3612 skb = tmp, tmp = skb->prev)
3613
3614#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3615 for (tmp = skb->prev; \
3616 skb != (struct sk_buff *)(queue); \
3617 skb = tmp, tmp = skb->prev)
3618
3619static inline bool skb_has_frag_list(const struct sk_buff *skb)
3620{
3621 return skb_shinfo(skb)->frag_list != NULL;
3622}
3623
3624static inline void skb_frag_list_init(struct sk_buff *skb)
3625{
3626 skb_shinfo(skb)->frag_list = NULL;
3627}
3628
3629#define skb_walk_frags(skb, iter) \
3630 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3631
3632
3633int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3634 int *err, long *timeo_p,
3635 const struct sk_buff *skb);
3636struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3637 struct sk_buff_head *queue,
3638 unsigned int flags,
3639 int *off, int *err,
3640 struct sk_buff **last);
3641struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3642 struct sk_buff_head *queue,
3643 unsigned int flags, int *off, int *err,
3644 struct sk_buff **last);
3645struct sk_buff *__skb_recv_datagram(struct sock *sk,
3646 struct sk_buff_head *sk_queue,
3647 unsigned int flags, int *off, int *err);
3648struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3649 int *err);
3650__poll_t datagram_poll(struct file *file, struct socket *sock,
3651 struct poll_table_struct *wait);
3652int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3653 struct iov_iter *to, int size);
3654static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3655 struct msghdr *msg, int size)
3656{
3657 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3658}
3659int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3660 struct msghdr *msg);
3661int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3662 struct iov_iter *to, int len,
3663 struct ahash_request *hash);
3664int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3665 struct iov_iter *from, int len);
3666int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3667void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3668void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3669static inline void skb_free_datagram_locked(struct sock *sk,
3670 struct sk_buff *skb)
3671{
3672 __skb_free_datagram_locked(sk, skb, 0);
3673}
3674int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3675int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3676int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3677__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3678 int len);
3679int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3680 struct pipe_inode_info *pipe, unsigned int len,
3681 unsigned int flags);
3682int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3683 int len);
3684int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3685void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3686unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3687int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3688 int len, int hlen);
3689void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3690int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3691void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3692bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3693bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3694struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3695struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3696 unsigned int offset);
3697struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3698int skb_ensure_writable(struct sk_buff *skb, int write_len);
3699int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3700int skb_vlan_pop(struct sk_buff *skb);
3701int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3702int skb_eth_pop(struct sk_buff *skb);
3703int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3704 const unsigned char *src);
3705int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3706 int mac_len, bool ethernet);
3707int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3708 bool ethernet);
3709int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3710int skb_mpls_dec_ttl(struct sk_buff *skb);
3711struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3712 gfp_t gfp);
3713
3714static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3715{
3716 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3717}
3718
3719static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3720{
3721 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3722}
3723
3724struct skb_checksum_ops {
3725 __wsum (*update)(const void *mem, int len, __wsum wsum);
3726 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3727};
3728
3729extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3730
3731__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3732 __wsum csum, const struct skb_checksum_ops *ops);
3733__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3734 __wsum csum);
3735
3736static inline void * __must_check
3737__skb_header_pointer(const struct sk_buff *skb, int offset, int len,
3738 const void *data, int hlen, void *buffer)
3739{
3740 if (likely(hlen - offset >= len))
3741 return (void *)data + offset;
3742
3743 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
3744 return NULL;
3745
3746 return buffer;
3747}
3748
3749static inline void * __must_check
3750skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3751{
3752 return __skb_header_pointer(skb, offset, len, skb->data,
3753 skb_headlen(skb), buffer);
3754}
3755
3756/**
3757 * skb_needs_linearize - check if we need to linearize a given skb
3758 * depending on the given device features.
3759 * @skb: socket buffer to check
3760 * @features: net device features
3761 *
3762 * Returns true if either:
3763 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3764 * 2. skb is fragmented and the device does not support SG.
3765 */
3766static inline bool skb_needs_linearize(struct sk_buff *skb,
3767 netdev_features_t features)
3768{
3769 return skb_is_nonlinear(skb) &&
3770 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3771 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3772}
3773
3774static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3775 void *to,
3776 const unsigned int len)
3777{
3778 memcpy(to, skb->data, len);
3779}
3780
3781static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3782 const int offset, void *to,
3783 const unsigned int len)
3784{
3785 memcpy(to, skb->data + offset, len);
3786}
3787
3788static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3789 const void *from,
3790 const unsigned int len)
3791{
3792 memcpy(skb->data, from, len);
3793}
3794
3795static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3796 const int offset,
3797 const void *from,
3798 const unsigned int len)
3799{
3800 memcpy(skb->data + offset, from, len);
3801}
3802
3803void skb_init(void);
3804
3805static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3806{
3807 return skb->tstamp;
3808}
3809
3810/**
3811 * skb_get_timestamp - get timestamp from a skb
3812 * @skb: skb to get stamp from
3813 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3814 *
3815 * Timestamps are stored in the skb as offsets to a base timestamp.
3816 * This function converts the offset back to a struct timeval and stores
3817 * it in stamp.
3818 */
3819static inline void skb_get_timestamp(const struct sk_buff *skb,
3820 struct __kernel_old_timeval *stamp)
3821{
3822 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3823}
3824
3825static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3826 struct __kernel_sock_timeval *stamp)
3827{
3828 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3829
3830 stamp->tv_sec = ts.tv_sec;
3831 stamp->tv_usec = ts.tv_nsec / 1000;
3832}
3833
3834static inline void skb_get_timestampns(const struct sk_buff *skb,
3835 struct __kernel_old_timespec *stamp)
3836{
3837 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3838
3839 stamp->tv_sec = ts.tv_sec;
3840 stamp->tv_nsec = ts.tv_nsec;
3841}
3842
3843static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3844 struct __kernel_timespec *stamp)
3845{
3846 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3847
3848 stamp->tv_sec = ts.tv_sec;
3849 stamp->tv_nsec = ts.tv_nsec;
3850}
3851
3852static inline void __net_timestamp(struct sk_buff *skb)
3853{
3854 skb->tstamp = ktime_get_real();
3855}
3856
3857static inline ktime_t net_timedelta(ktime_t t)
3858{
3859 return ktime_sub(ktime_get_real(), t);
3860}
3861
3862static inline ktime_t net_invalid_timestamp(void)
3863{
3864 return 0;
3865}
3866
3867static inline u8 skb_metadata_len(const struct sk_buff *skb)
3868{
3869 return skb_shinfo(skb)->meta_len;
3870}
3871
3872static inline void *skb_metadata_end(const struct sk_buff *skb)
3873{
3874 return skb_mac_header(skb);
3875}
3876
3877static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3878 const struct sk_buff *skb_b,
3879 u8 meta_len)
3880{
3881 const void *a = skb_metadata_end(skb_a);
3882 const void *b = skb_metadata_end(skb_b);
3883 /* Using more efficient varaiant than plain call to memcmp(). */
3884#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3885 u64 diffs = 0;
3886
3887 switch (meta_len) {
3888#define __it(x, op) (x -= sizeof(u##op))
3889#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3890 case 32: diffs |= __it_diff(a, b, 64);
3891 fallthrough;
3892 case 24: diffs |= __it_diff(a, b, 64);
3893 fallthrough;
3894 case 16: diffs |= __it_diff(a, b, 64);
3895 fallthrough;
3896 case 8: diffs |= __it_diff(a, b, 64);
3897 break;
3898 case 28: diffs |= __it_diff(a, b, 64);
3899 fallthrough;
3900 case 20: diffs |= __it_diff(a, b, 64);
3901 fallthrough;
3902 case 12: diffs |= __it_diff(a, b, 64);
3903 fallthrough;
3904 case 4: diffs |= __it_diff(a, b, 32);
3905 break;
3906 }
3907 return diffs;
3908#else
3909 return memcmp(a - meta_len, b - meta_len, meta_len);
3910#endif
3911}
3912
3913static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3914 const struct sk_buff *skb_b)
3915{
3916 u8 len_a = skb_metadata_len(skb_a);
3917 u8 len_b = skb_metadata_len(skb_b);
3918
3919 if (!(len_a | len_b))
3920 return false;
3921
3922 return len_a != len_b ?
3923 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3924}
3925
3926static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3927{
3928 skb_shinfo(skb)->meta_len = meta_len;
3929}
3930
3931static inline void skb_metadata_clear(struct sk_buff *skb)
3932{
3933 skb_metadata_set(skb, 0);
3934}
3935
3936struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3937
3938#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3939
3940void skb_clone_tx_timestamp(struct sk_buff *skb);
3941bool skb_defer_rx_timestamp(struct sk_buff *skb);
3942
3943#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3944
3945static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3946{
3947}
3948
3949static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3950{
3951 return false;
3952}
3953
3954#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3955
3956/**
3957 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3958 *
3959 * PHY drivers may accept clones of transmitted packets for
3960 * timestamping via their phy_driver.txtstamp method. These drivers
3961 * must call this function to return the skb back to the stack with a
3962 * timestamp.
3963 *
3964 * @skb: clone of the original outgoing packet
3965 * @hwtstamps: hardware time stamps
3966 *
3967 */
3968void skb_complete_tx_timestamp(struct sk_buff *skb,
3969 struct skb_shared_hwtstamps *hwtstamps);
3970
3971void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
3972 struct skb_shared_hwtstamps *hwtstamps,
3973 struct sock *sk, int tstype);
3974
3975/**
3976 * skb_tstamp_tx - queue clone of skb with send time stamps
3977 * @orig_skb: the original outgoing packet
3978 * @hwtstamps: hardware time stamps, may be NULL if not available
3979 *
3980 * If the skb has a socket associated, then this function clones the
3981 * skb (thus sharing the actual data and optional structures), stores
3982 * the optional hardware time stamping information (if non NULL) or
3983 * generates a software time stamp (otherwise), then queues the clone
3984 * to the error queue of the socket. Errors are silently ignored.
3985 */
3986void skb_tstamp_tx(struct sk_buff *orig_skb,
3987 struct skb_shared_hwtstamps *hwtstamps);
3988
3989/**
3990 * skb_tx_timestamp() - Driver hook for transmit timestamping
3991 *
3992 * Ethernet MAC Drivers should call this function in their hard_xmit()
3993 * function immediately before giving the sk_buff to the MAC hardware.
3994 *
3995 * Specifically, one should make absolutely sure that this function is
3996 * called before TX completion of this packet can trigger. Otherwise
3997 * the packet could potentially already be freed.
3998 *
3999 * @skb: A socket buffer.
4000 */
4001static inline void skb_tx_timestamp(struct sk_buff *skb)
4002{
4003 skb_clone_tx_timestamp(skb);
4004 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4005 skb_tstamp_tx(skb, NULL);
4006}
4007
4008/**
4009 * skb_complete_wifi_ack - deliver skb with wifi status
4010 *
4011 * @skb: the original outgoing packet
4012 * @acked: ack status
4013 *
4014 */
4015void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4016
4017__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4018__sum16 __skb_checksum_complete(struct sk_buff *skb);
4019
4020static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4021{
4022 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4023 skb->csum_valid ||
4024 (skb->ip_summed == CHECKSUM_PARTIAL &&
4025 skb_checksum_start_offset(skb) >= 0));
4026}
4027
4028/**
4029 * skb_checksum_complete - Calculate checksum of an entire packet
4030 * @skb: packet to process
4031 *
4032 * This function calculates the checksum over the entire packet plus
4033 * the value of skb->csum. The latter can be used to supply the
4034 * checksum of a pseudo header as used by TCP/UDP. It returns the
4035 * checksum.
4036 *
4037 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4038 * this function can be used to verify that checksum on received
4039 * packets. In that case the function should return zero if the
4040 * checksum is correct. In particular, this function will return zero
4041 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4042 * hardware has already verified the correctness of the checksum.
4043 */
4044static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4045{
4046 return skb_csum_unnecessary(skb) ?
4047 0 : __skb_checksum_complete(skb);
4048}
4049
4050static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4051{
4052 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4053 if (skb->csum_level == 0)
4054 skb->ip_summed = CHECKSUM_NONE;
4055 else
4056 skb->csum_level--;
4057 }
4058}
4059
4060static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4061{
4062 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4063 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4064 skb->csum_level++;
4065 } else if (skb->ip_summed == CHECKSUM_NONE) {
4066 skb->ip_summed = CHECKSUM_UNNECESSARY;
4067 skb->csum_level = 0;
4068 }
4069}
4070
4071static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4072{
4073 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4074 skb->ip_summed = CHECKSUM_NONE;
4075 skb->csum_level = 0;
4076 }
4077}
4078
4079/* Check if we need to perform checksum complete validation.
4080 *
4081 * Returns true if checksum complete is needed, false otherwise
4082 * (either checksum is unnecessary or zero checksum is allowed).
4083 */
4084static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4085 bool zero_okay,
4086 __sum16 check)
4087{
4088 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4089 skb->csum_valid = 1;
4090 __skb_decr_checksum_unnecessary(skb);
4091 return false;
4092 }
4093
4094 return true;
4095}
4096
4097/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4098 * in checksum_init.
4099 */
4100#define CHECKSUM_BREAK 76
4101
4102/* Unset checksum-complete
4103 *
4104 * Unset checksum complete can be done when packet is being modified
4105 * (uncompressed for instance) and checksum-complete value is
4106 * invalidated.
4107 */
4108static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4109{
4110 if (skb->ip_summed == CHECKSUM_COMPLETE)
4111 skb->ip_summed = CHECKSUM_NONE;
4112}
4113
4114/* Validate (init) checksum based on checksum complete.
4115 *
4116 * Return values:
4117 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4118 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4119 * checksum is stored in skb->csum for use in __skb_checksum_complete
4120 * non-zero: value of invalid checksum
4121 *
4122 */
4123static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4124 bool complete,
4125 __wsum psum)
4126{
4127 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4128 if (!csum_fold(csum_add(psum, skb->csum))) {
4129 skb->csum_valid = 1;
4130 return 0;
4131 }
4132 }
4133
4134 skb->csum = psum;
4135
4136 if (complete || skb->len <= CHECKSUM_BREAK) {
4137 __sum16 csum;
4138
4139 csum = __skb_checksum_complete(skb);
4140 skb->csum_valid = !csum;
4141 return csum;
4142 }
4143
4144 return 0;
4145}
4146
4147static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4148{
4149 return 0;
4150}
4151
4152/* Perform checksum validate (init). Note that this is a macro since we only
4153 * want to calculate the pseudo header which is an input function if necessary.
4154 * First we try to validate without any computation (checksum unnecessary) and
4155 * then calculate based on checksum complete calling the function to compute
4156 * pseudo header.
4157 *
4158 * Return values:
4159 * 0: checksum is validated or try to in skb_checksum_complete
4160 * non-zero: value of invalid checksum
4161 */
4162#define __skb_checksum_validate(skb, proto, complete, \
4163 zero_okay, check, compute_pseudo) \
4164({ \
4165 __sum16 __ret = 0; \
4166 skb->csum_valid = 0; \
4167 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4168 __ret = __skb_checksum_validate_complete(skb, \
4169 complete, compute_pseudo(skb, proto)); \
4170 __ret; \
4171})
4172
4173#define skb_checksum_init(skb, proto, compute_pseudo) \
4174 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4175
4176#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4177 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4178
4179#define skb_checksum_validate(skb, proto, compute_pseudo) \
4180 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4181
4182#define skb_checksum_validate_zero_check(skb, proto, check, \
4183 compute_pseudo) \
4184 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4185
4186#define skb_checksum_simple_validate(skb) \
4187 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4188
4189static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4190{
4191 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4192}
4193
4194static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4195{
4196 skb->csum = ~pseudo;
4197 skb->ip_summed = CHECKSUM_COMPLETE;
4198}
4199
4200#define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4201do { \
4202 if (__skb_checksum_convert_check(skb)) \
4203 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4204} while (0)
4205
4206static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4207 u16 start, u16 offset)
4208{
4209 skb->ip_summed = CHECKSUM_PARTIAL;
4210 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4211 skb->csum_offset = offset - start;
4212}
4213
4214/* Update skbuf and packet to reflect the remote checksum offload operation.
4215 * When called, ptr indicates the starting point for skb->csum when
4216 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4217 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4218 */
4219static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4220 int start, int offset, bool nopartial)
4221{
4222 __wsum delta;
4223
4224 if (!nopartial) {
4225 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4226 return;
4227 }
4228
4229 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4230 __skb_checksum_complete(skb);
4231 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4232 }
4233
4234 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4235
4236 /* Adjust skb->csum since we changed the packet */
4237 skb->csum = csum_add(skb->csum, delta);
4238}
4239
4240static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4241{
4242#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4243 return (void *)(skb->_nfct & NFCT_PTRMASK);
4244#else
4245 return NULL;
4246#endif
4247}
4248
4249static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4250{
4251#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4252 return skb->_nfct;
4253#else
4254 return 0UL;
4255#endif
4256}
4257
4258static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4259{
4260#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4261 skb->slow_gro |= !!nfct;
4262 skb->_nfct = nfct;
4263#endif
4264}
4265
4266#ifdef CONFIG_SKB_EXTENSIONS
4267enum skb_ext_id {
4268#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4269 SKB_EXT_BRIDGE_NF,
4270#endif
4271#ifdef CONFIG_XFRM
4272 SKB_EXT_SEC_PATH,
4273#endif
4274#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4275 TC_SKB_EXT,
4276#endif
4277#if IS_ENABLED(CONFIG_MPTCP)
4278 SKB_EXT_MPTCP,
4279#endif
4280#if IS_ENABLED(CONFIG_MCTP_FLOWS)
4281 SKB_EXT_MCTP,
4282#endif
4283 SKB_EXT_NUM, /* must be last */
4284};
4285
4286/**
4287 * struct skb_ext - sk_buff extensions
4288 * @refcnt: 1 on allocation, deallocated on 0
4289 * @offset: offset to add to @data to obtain extension address
4290 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4291 * @data: start of extension data, variable sized
4292 *
4293 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4294 * to use 'u8' types while allowing up to 2kb worth of extension data.
4295 */
4296struct skb_ext {
4297 refcount_t refcnt;
4298 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4299 u8 chunks; /* same */
4300 char data[] __aligned(8);
4301};
4302
4303struct skb_ext *__skb_ext_alloc(gfp_t flags);
4304void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4305 struct skb_ext *ext);
4306void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4307void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4308void __skb_ext_put(struct skb_ext *ext);
4309
4310static inline void skb_ext_put(struct sk_buff *skb)
4311{
4312 if (skb->active_extensions)
4313 __skb_ext_put(skb->extensions);
4314}
4315
4316static inline void __skb_ext_copy(struct sk_buff *dst,
4317 const struct sk_buff *src)
4318{
4319 dst->active_extensions = src->active_extensions;
4320
4321 if (src->active_extensions) {
4322 struct skb_ext *ext = src->extensions;
4323
4324 refcount_inc(&ext->refcnt);
4325 dst->extensions = ext;
4326 }
4327}
4328
4329static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4330{
4331 skb_ext_put(dst);
4332 __skb_ext_copy(dst, src);
4333}
4334
4335static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4336{
4337 return !!ext->offset[i];
4338}
4339
4340static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4341{
4342 return skb->active_extensions & (1 << id);
4343}
4344
4345static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4346{
4347 if (skb_ext_exist(skb, id))
4348 __skb_ext_del(skb, id);
4349}
4350
4351static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4352{
4353 if (skb_ext_exist(skb, id)) {
4354 struct skb_ext *ext = skb->extensions;
4355
4356 return (void *)ext + (ext->offset[id] << 3);
4357 }
4358
4359 return NULL;
4360}
4361
4362static inline void skb_ext_reset(struct sk_buff *skb)
4363{
4364 if (unlikely(skb->active_extensions)) {
4365 __skb_ext_put(skb->extensions);
4366 skb->active_extensions = 0;
4367 }
4368}
4369
4370static inline bool skb_has_extensions(struct sk_buff *skb)
4371{
4372 return unlikely(skb->active_extensions);
4373}
4374#else
4375static inline void skb_ext_put(struct sk_buff *skb) {}
4376static inline void skb_ext_reset(struct sk_buff *skb) {}
4377static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4378static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4379static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4380static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4381#endif /* CONFIG_SKB_EXTENSIONS */
4382
4383static inline void nf_reset_ct(struct sk_buff *skb)
4384{
4385#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4386 nf_conntrack_put(skb_nfct(skb));
4387 skb->_nfct = 0;
4388#endif
4389}
4390
4391static inline void nf_reset_trace(struct sk_buff *skb)
4392{
4393#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4394 skb->nf_trace = 0;
4395#endif
4396}
4397
4398static inline void ipvs_reset(struct sk_buff *skb)
4399{
4400#if IS_ENABLED(CONFIG_IP_VS)
4401 skb->ipvs_property = 0;
4402#endif
4403}
4404
4405/* Note: This doesn't put any conntrack info in dst. */
4406static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4407 bool copy)
4408{
4409#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4410 dst->_nfct = src->_nfct;
4411 nf_conntrack_get(skb_nfct(src));
4412#endif
4413#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4414 if (copy)
4415 dst->nf_trace = src->nf_trace;
4416#endif
4417}
4418
4419static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4420{
4421#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4422 nf_conntrack_put(skb_nfct(dst));
4423#endif
4424 dst->slow_gro = src->slow_gro;
4425 __nf_copy(dst, src, true);
4426}
4427
4428#ifdef CONFIG_NETWORK_SECMARK
4429static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4430{
4431 to->secmark = from->secmark;
4432}
4433
4434static inline void skb_init_secmark(struct sk_buff *skb)
4435{
4436 skb->secmark = 0;
4437}
4438#else
4439static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4440{ }
4441
4442static inline void skb_init_secmark(struct sk_buff *skb)
4443{ }
4444#endif
4445
4446static inline int secpath_exists(const struct sk_buff *skb)
4447{
4448#ifdef CONFIG_XFRM
4449 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4450#else
4451 return 0;
4452#endif
4453}
4454
4455static inline bool skb_irq_freeable(const struct sk_buff *skb)
4456{
4457 return !skb->destructor &&
4458 !secpath_exists(skb) &&
4459 !skb_nfct(skb) &&
4460 !skb->_skb_refdst &&
4461 !skb_has_frag_list(skb);
4462}
4463
4464static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4465{
4466 skb->queue_mapping = queue_mapping;
4467}
4468
4469static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4470{
4471 return skb->queue_mapping;
4472}
4473
4474static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4475{
4476 to->queue_mapping = from->queue_mapping;
4477}
4478
4479static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4480{
4481 skb->queue_mapping = rx_queue + 1;
4482}
4483
4484static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4485{
4486 return skb->queue_mapping - 1;
4487}
4488
4489static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4490{
4491 return skb->queue_mapping != 0;
4492}
4493
4494static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4495{
4496 skb->dst_pending_confirm = val;
4497}
4498
4499static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4500{
4501 return skb->dst_pending_confirm != 0;
4502}
4503
4504static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4505{
4506#ifdef CONFIG_XFRM
4507 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4508#else
4509 return NULL;
4510#endif
4511}
4512
4513/* Keeps track of mac header offset relative to skb->head.
4514 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4515 * For non-tunnel skb it points to skb_mac_header() and for
4516 * tunnel skb it points to outer mac header.
4517 * Keeps track of level of encapsulation of network headers.
4518 */
4519struct skb_gso_cb {
4520 union {
4521 int mac_offset;
4522 int data_offset;
4523 };
4524 int encap_level;
4525 __wsum csum;
4526 __u16 csum_start;
4527};
4528#define SKB_GSO_CB_OFFSET 32
4529#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4530
4531static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4532{
4533 return (skb_mac_header(inner_skb) - inner_skb->head) -
4534 SKB_GSO_CB(inner_skb)->mac_offset;
4535}
4536
4537static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4538{
4539 int new_headroom, headroom;
4540 int ret;
4541
4542 headroom = skb_headroom(skb);
4543 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4544 if (ret)
4545 return ret;
4546
4547 new_headroom = skb_headroom(skb);
4548 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4549 return 0;
4550}
4551
4552static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4553{
4554 /* Do not update partial checksums if remote checksum is enabled. */
4555 if (skb->remcsum_offload)
4556 return;
4557
4558 SKB_GSO_CB(skb)->csum = res;
4559 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4560}
4561
4562/* Compute the checksum for a gso segment. First compute the checksum value
4563 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4564 * then add in skb->csum (checksum from csum_start to end of packet).
4565 * skb->csum and csum_start are then updated to reflect the checksum of the
4566 * resultant packet starting from the transport header-- the resultant checksum
4567 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4568 * header.
4569 */
4570static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4571{
4572 unsigned char *csum_start = skb_transport_header(skb);
4573 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4574 __wsum partial = SKB_GSO_CB(skb)->csum;
4575
4576 SKB_GSO_CB(skb)->csum = res;
4577 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4578
4579 return csum_fold(csum_partial(csum_start, plen, partial));
4580}
4581
4582static inline bool skb_is_gso(const struct sk_buff *skb)
4583{
4584 return skb_shinfo(skb)->gso_size;
4585}
4586
4587/* Note: Should be called only if skb_is_gso(skb) is true */
4588static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4589{
4590 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4591}
4592
4593/* Note: Should be called only if skb_is_gso(skb) is true */
4594static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4595{
4596 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4597}
4598
4599/* Note: Should be called only if skb_is_gso(skb) is true */
4600static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4601{
4602 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4603}
4604
4605static inline void skb_gso_reset(struct sk_buff *skb)
4606{
4607 skb_shinfo(skb)->gso_size = 0;
4608 skb_shinfo(skb)->gso_segs = 0;
4609 skb_shinfo(skb)->gso_type = 0;
4610}
4611
4612static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4613 u16 increment)
4614{
4615 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4616 return;
4617 shinfo->gso_size += increment;
4618}
4619
4620static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4621 u16 decrement)
4622{
4623 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4624 return;
4625 shinfo->gso_size -= decrement;
4626}
4627
4628void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4629
4630static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4631{
4632 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4633 * wanted then gso_type will be set. */
4634 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4635
4636 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4637 unlikely(shinfo->gso_type == 0)) {
4638 __skb_warn_lro_forwarding(skb);
4639 return true;
4640 }
4641 return false;
4642}
4643
4644static inline void skb_forward_csum(struct sk_buff *skb)
4645{
4646 /* Unfortunately we don't support this one. Any brave souls? */
4647 if (skb->ip_summed == CHECKSUM_COMPLETE)
4648 skb->ip_summed = CHECKSUM_NONE;
4649}
4650
4651/**
4652 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4653 * @skb: skb to check
4654 *
4655 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4656 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4657 * use this helper, to document places where we make this assertion.
4658 */
4659static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4660{
4661#ifdef DEBUG
4662 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4663#endif
4664}
4665
4666bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4667
4668int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4669struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4670 unsigned int transport_len,
4671 __sum16(*skb_chkf)(struct sk_buff *skb));
4672
4673/**
4674 * skb_head_is_locked - Determine if the skb->head is locked down
4675 * @skb: skb to check
4676 *
4677 * The head on skbs build around a head frag can be removed if they are
4678 * not cloned. This function returns true if the skb head is locked down
4679 * due to either being allocated via kmalloc, or by being a clone with
4680 * multiple references to the head.
4681 */
4682static inline bool skb_head_is_locked(const struct sk_buff *skb)
4683{
4684 return !skb->head_frag || skb_cloned(skb);
4685}
4686
4687/* Local Checksum Offload.
4688 * Compute outer checksum based on the assumption that the
4689 * inner checksum will be offloaded later.
4690 * See Documentation/networking/checksum-offloads.rst for
4691 * explanation of how this works.
4692 * Fill in outer checksum adjustment (e.g. with sum of outer
4693 * pseudo-header) before calling.
4694 * Also ensure that inner checksum is in linear data area.
4695 */
4696static inline __wsum lco_csum(struct sk_buff *skb)
4697{
4698 unsigned char *csum_start = skb_checksum_start(skb);
4699 unsigned char *l4_hdr = skb_transport_header(skb);
4700 __wsum partial;
4701
4702 /* Start with complement of inner checksum adjustment */
4703 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4704 skb->csum_offset));
4705
4706 /* Add in checksum of our headers (incl. outer checksum
4707 * adjustment filled in by caller) and return result.
4708 */
4709 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4710}
4711
4712static inline bool skb_is_redirected(const struct sk_buff *skb)
4713{
4714 return skb->redirected;
4715}
4716
4717static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
4718{
4719 skb->redirected = 1;
4720#ifdef CONFIG_NET_REDIRECT
4721 skb->from_ingress = from_ingress;
4722 if (skb->from_ingress)
4723 skb->tstamp = 0;
4724#endif
4725}
4726
4727static inline void skb_reset_redirect(struct sk_buff *skb)
4728{
4729 skb->redirected = 0;
4730}
4731
4732static inline bool skb_csum_is_sctp(struct sk_buff *skb)
4733{
4734 return skb->csum_not_inet;
4735}
4736
4737static inline void skb_set_kcov_handle(struct sk_buff *skb,
4738 const u64 kcov_handle)
4739{
4740#ifdef CONFIG_KCOV
4741 skb->kcov_handle = kcov_handle;
4742#endif
4743}
4744
4745static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
4746{
4747#ifdef CONFIG_KCOV
4748 return skb->kcov_handle;
4749#else
4750 return 0;
4751#endif
4752}
4753
4754#ifdef CONFIG_PAGE_POOL
4755static inline void skb_mark_for_recycle(struct sk_buff *skb)
4756{
4757 skb->pp_recycle = 1;
4758}
4759#endif
4760
4761static inline bool skb_pp_recycle(struct sk_buff *skb, void *data)
4762{
4763 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
4764 return false;
4765 return page_pool_return_skb_page(virt_to_page(data));
4766}
4767
4768#endif /* __KERNEL__ */
4769#endif /* _LINUX_SKBUFF_H */