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