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