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