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