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
1214static inline __u32 skb_get_hash(struct sk_buff *skb)
1215{
1216 if (!skb->l4_hash && !skb->sw_hash)
1217 __skb_get_hash(skb);
1218
1219 return skb->hash;
1220}
1221
1222static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1223{
1224 if (!skb->l4_hash && !skb->sw_hash) {
1225 struct flow_keys keys;
1226 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1227
1228 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1229 }
1230
1231 return skb->hash;
1232}
1233
1234__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1235
1236static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1237{
1238 return skb->hash;
1239}
1240
1241static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1242{
1243 to->hash = from->hash;
1244 to->sw_hash = from->sw_hash;
1245 to->l4_hash = from->l4_hash;
1246};
1247
1248#ifdef NET_SKBUFF_DATA_USES_OFFSET
1249static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1250{
1251 return skb->head + skb->end;
1252}
1253
1254static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1255{
1256 return skb->end;
1257}
1258#else
1259static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1260{
1261 return skb->end;
1262}
1263
1264static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1265{
1266 return skb->end - skb->head;
1267}
1268#endif
1269
1270/* Internal */
1271#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1272
1273static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1274{
1275 return &skb_shinfo(skb)->hwtstamps;
1276}
1277
1278static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1279{
1280 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1281
1282 return is_zcopy ? skb_uarg(skb) : NULL;
1283}
1284
1285static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
1286{
1287 if (skb && uarg && !skb_zcopy(skb)) {
1288 sock_zerocopy_get(uarg);
1289 skb_shinfo(skb)->destructor_arg = uarg;
1290 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1291 }
1292}
1293
1294/* Release a reference on a zerocopy structure */
1295static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1296{
1297 struct ubuf_info *uarg = skb_zcopy(skb);
1298
1299 if (uarg) {
1300 if (uarg->callback == sock_zerocopy_callback) {
1301 uarg->zerocopy = uarg->zerocopy && zerocopy;
1302 sock_zerocopy_put(uarg);
1303 } else {
1304 uarg->callback(uarg, zerocopy);
1305 }
1306
1307 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1308 }
1309}
1310
1311/* Abort a zerocopy operation and revert zckey on error in send syscall */
1312static inline void skb_zcopy_abort(struct sk_buff *skb)
1313{
1314 struct ubuf_info *uarg = skb_zcopy(skb);
1315
1316 if (uarg) {
1317 sock_zerocopy_put_abort(uarg);
1318 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1319 }
1320}
1321
1322/**
1323 * skb_queue_empty - check if a queue is empty
1324 * @list: queue head
1325 *
1326 * Returns true if the queue is empty, false otherwise.
1327 */
1328static inline int skb_queue_empty(const struct sk_buff_head *list)
1329{
1330 return list->next == (const struct sk_buff *) list;
1331}
1332
1333/**
1334 * skb_queue_is_last - check if skb is the last entry in the queue
1335 * @list: queue head
1336 * @skb: buffer
1337 *
1338 * Returns true if @skb is the last buffer on the list.
1339 */
1340static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1341 const struct sk_buff *skb)
1342{
1343 return skb->next == (const struct sk_buff *) list;
1344}
1345
1346/**
1347 * skb_queue_is_first - check if skb is the first entry in the queue
1348 * @list: queue head
1349 * @skb: buffer
1350 *
1351 * Returns true if @skb is the first buffer on the list.
1352 */
1353static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1354 const struct sk_buff *skb)
1355{
1356 return skb->prev == (const struct sk_buff *) list;
1357}
1358
1359/**
1360 * skb_queue_next - return the next packet in the queue
1361 * @list: queue head
1362 * @skb: current buffer
1363 *
1364 * Return the next packet in @list after @skb. It is only valid to
1365 * call this if skb_queue_is_last() evaluates to false.
1366 */
1367static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1368 const struct sk_buff *skb)
1369{
1370 /* This BUG_ON may seem severe, but if we just return then we
1371 * are going to dereference garbage.
1372 */
1373 BUG_ON(skb_queue_is_last(list, skb));
1374 return skb->next;
1375}
1376
1377/**
1378 * skb_queue_prev - return the prev packet in the queue
1379 * @list: queue head
1380 * @skb: current buffer
1381 *
1382 * Return the prev packet in @list before @skb. It is only valid to
1383 * call this if skb_queue_is_first() evaluates to false.
1384 */
1385static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1386 const struct sk_buff *skb)
1387{
1388 /* This BUG_ON may seem severe, but if we just return then we
1389 * are going to dereference garbage.
1390 */
1391 BUG_ON(skb_queue_is_first(list, skb));
1392 return skb->prev;
1393}
1394
1395/**
1396 * skb_get - reference buffer
1397 * @skb: buffer to reference
1398 *
1399 * Makes another reference to a socket buffer and returns a pointer
1400 * to the buffer.
1401 */
1402static inline struct sk_buff *skb_get(struct sk_buff *skb)
1403{
1404 refcount_inc(&skb->users);
1405 return skb;
1406}
1407
1408/*
1409 * If users == 1, we are the only owner and are can avoid redundant
1410 * atomic change.
1411 */
1412
1413/**
1414 * skb_cloned - is the buffer a clone
1415 * @skb: buffer to check
1416 *
1417 * Returns true if the buffer was generated with skb_clone() and is
1418 * one of multiple shared copies of the buffer. Cloned buffers are
1419 * shared data so must not be written to under normal circumstances.
1420 */
1421static inline int skb_cloned(const struct sk_buff *skb)
1422{
1423 return skb->cloned &&
1424 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1425}
1426
1427static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1428{
1429 might_sleep_if(gfpflags_allow_blocking(pri));
1430
1431 if (skb_cloned(skb))
1432 return pskb_expand_head(skb, 0, 0, pri);
1433
1434 return 0;
1435}
1436
1437/**
1438 * skb_header_cloned - is the header a clone
1439 * @skb: buffer to check
1440 *
1441 * Returns true if modifying the header part of the buffer requires
1442 * the data to be copied.
1443 */
1444static inline int skb_header_cloned(const struct sk_buff *skb)
1445{
1446 int dataref;
1447
1448 if (!skb->cloned)
1449 return 0;
1450
1451 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1452 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1453 return dataref != 1;
1454}
1455
1456static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1457{
1458 might_sleep_if(gfpflags_allow_blocking(pri));
1459
1460 if (skb_header_cloned(skb))
1461 return pskb_expand_head(skb, 0, 0, pri);
1462
1463 return 0;
1464}
1465
1466/**
1467 * __skb_header_release - release reference to header
1468 * @skb: buffer to operate on
1469 */
1470static inline void __skb_header_release(struct sk_buff *skb)
1471{
1472 skb->nohdr = 1;
1473 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1474}
1475
1476
1477/**
1478 * skb_shared - is the buffer shared
1479 * @skb: buffer to check
1480 *
1481 * Returns true if more than one person has a reference to this
1482 * buffer.
1483 */
1484static inline int skb_shared(const struct sk_buff *skb)
1485{
1486 return refcount_read(&skb->users) != 1;
1487}
1488
1489/**
1490 * skb_share_check - check if buffer is shared and if so clone it
1491 * @skb: buffer to check
1492 * @pri: priority for memory allocation
1493 *
1494 * If the buffer is shared the buffer is cloned and the old copy
1495 * drops a reference. A new clone with a single reference is returned.
1496 * If the buffer is not shared the original buffer is returned. When
1497 * being called from interrupt status or with spinlocks held pri must
1498 * be GFP_ATOMIC.
1499 *
1500 * NULL is returned on a memory allocation failure.
1501 */
1502static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1503{
1504 might_sleep_if(gfpflags_allow_blocking(pri));
1505 if (skb_shared(skb)) {
1506 struct sk_buff *nskb = skb_clone(skb, pri);
1507
1508 if (likely(nskb))
1509 consume_skb(skb);
1510 else
1511 kfree_skb(skb);
1512 skb = nskb;
1513 }
1514 return skb;
1515}
1516
1517/*
1518 * Copy shared buffers into a new sk_buff. We effectively do COW on
1519 * packets to handle cases where we have a local reader and forward
1520 * and a couple of other messy ones. The normal one is tcpdumping
1521 * a packet thats being forwarded.
1522 */
1523
1524/**
1525 * skb_unshare - make a copy of a shared buffer
1526 * @skb: buffer to check
1527 * @pri: priority for memory allocation
1528 *
1529 * If the socket buffer is a clone then this function creates a new
1530 * copy of the data, drops a reference count on the old copy and returns
1531 * the new copy with the reference count at 1. If the buffer is not a clone
1532 * the original buffer is returned. When called with a spinlock held or
1533 * from interrupt state @pri must be %GFP_ATOMIC
1534 *
1535 * %NULL is returned on a memory allocation failure.
1536 */
1537static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1538 gfp_t pri)
1539{
1540 might_sleep_if(gfpflags_allow_blocking(pri));
1541 if (skb_cloned(skb)) {
1542 struct sk_buff *nskb = skb_copy(skb, pri);
1543
1544 /* Free our shared copy */
1545 if (likely(nskb))
1546 consume_skb(skb);
1547 else
1548 kfree_skb(skb);
1549 skb = nskb;
1550 }
1551 return skb;
1552}
1553
1554/**
1555 * skb_peek - peek at the head of an &sk_buff_head
1556 * @list_: list to peek at
1557 *
1558 * Peek an &sk_buff. Unlike most other operations you _MUST_
1559 * be careful with this one. A peek leaves the buffer on the
1560 * list and someone else may run off with it. You must hold
1561 * the appropriate locks or have a private queue to do this.
1562 *
1563 * Returns %NULL for an empty list or a pointer to the head element.
1564 * The reference count is not incremented and the reference is therefore
1565 * volatile. Use with caution.
1566 */
1567static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1568{
1569 struct sk_buff *skb = list_->next;
1570
1571 if (skb == (struct sk_buff *)list_)
1572 skb = NULL;
1573 return skb;
1574}
1575
1576/**
1577 * skb_peek_next - peek skb following the given one from a queue
1578 * @skb: skb to start from
1579 * @list_: list to peek at
1580 *
1581 * Returns %NULL when the end of the list is met or a pointer to the
1582 * next element. The reference count is not incremented and the
1583 * reference is therefore volatile. Use with caution.
1584 */
1585static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1586 const struct sk_buff_head *list_)
1587{
1588 struct sk_buff *next = skb->next;
1589
1590 if (next == (struct sk_buff *)list_)
1591 next = NULL;
1592 return next;
1593}
1594
1595/**
1596 * skb_peek_tail - peek at the tail of an &sk_buff_head
1597 * @list_: list to peek at
1598 *
1599 * Peek an &sk_buff. Unlike most other operations you _MUST_
1600 * be careful with this one. A peek leaves the buffer on the
1601 * list and someone else may run off with it. You must hold
1602 * the appropriate locks or have a private queue to do this.
1603 *
1604 * Returns %NULL for an empty list or a pointer to the tail element.
1605 * The reference count is not incremented and the reference is therefore
1606 * volatile. Use with caution.
1607 */
1608static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1609{
1610 struct sk_buff *skb = list_->prev;
1611
1612 if (skb == (struct sk_buff *)list_)
1613 skb = NULL;
1614 return skb;
1615
1616}
1617
1618/**
1619 * skb_queue_len - get queue length
1620 * @list_: list to measure
1621 *
1622 * Return the length of an &sk_buff queue.
1623 */
1624static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1625{
1626 return list_->qlen;
1627}
1628
1629/**
1630 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1631 * @list: queue to initialize
1632 *
1633 * This initializes only the list and queue length aspects of
1634 * an sk_buff_head object. This allows to initialize the list
1635 * aspects of an sk_buff_head without reinitializing things like
1636 * the spinlock. It can also be used for on-stack sk_buff_head
1637 * objects where the spinlock is known to not be used.
1638 */
1639static inline void __skb_queue_head_init(struct sk_buff_head *list)
1640{
1641 list->prev = list->next = (struct sk_buff *)list;
1642 list->qlen = 0;
1643}
1644
1645/*
1646 * This function creates a split out lock class for each invocation;
1647 * this is needed for now since a whole lot of users of the skb-queue
1648 * infrastructure in drivers have different locking usage (in hardirq)
1649 * than the networking core (in softirq only). In the long run either the
1650 * network layer or drivers should need annotation to consolidate the
1651 * main types of usage into 3 classes.
1652 */
1653static inline void skb_queue_head_init(struct sk_buff_head *list)
1654{
1655 spin_lock_init(&list->lock);
1656 __skb_queue_head_init(list);
1657}
1658
1659static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1660 struct lock_class_key *class)
1661{
1662 skb_queue_head_init(list);
1663 lockdep_set_class(&list->lock, class);
1664}
1665
1666/*
1667 * Insert an sk_buff on a list.
1668 *
1669 * The "__skb_xxxx()" functions are the non-atomic ones that
1670 * can only be called with interrupts disabled.
1671 */
1672void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1673 struct sk_buff_head *list);
1674static inline void __skb_insert(struct sk_buff *newsk,
1675 struct sk_buff *prev, struct sk_buff *next,
1676 struct sk_buff_head *list)
1677{
1678 newsk->next = next;
1679 newsk->prev = prev;
1680 next->prev = prev->next = newsk;
1681 list->qlen++;
1682}
1683
1684static inline void __skb_queue_splice(const struct sk_buff_head *list,
1685 struct sk_buff *prev,
1686 struct sk_buff *next)
1687{
1688 struct sk_buff *first = list->next;
1689 struct sk_buff *last = list->prev;
1690
1691 first->prev = prev;
1692 prev->next = first;
1693
1694 last->next = next;
1695 next->prev = last;
1696}
1697
1698/**
1699 * skb_queue_splice - join two skb lists, this is designed for stacks
1700 * @list: the new list to add
1701 * @head: the place to add it in the first list
1702 */
1703static inline void skb_queue_splice(const struct sk_buff_head *list,
1704 struct sk_buff_head *head)
1705{
1706 if (!skb_queue_empty(list)) {
1707 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1708 head->qlen += list->qlen;
1709 }
1710}
1711
1712/**
1713 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1714 * @list: the new list to add
1715 * @head: the place to add it in the first list
1716 *
1717 * The list at @list is reinitialised
1718 */
1719static inline void skb_queue_splice_init(struct sk_buff_head *list,
1720 struct sk_buff_head *head)
1721{
1722 if (!skb_queue_empty(list)) {
1723 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1724 head->qlen += list->qlen;
1725 __skb_queue_head_init(list);
1726 }
1727}
1728
1729/**
1730 * skb_queue_splice_tail - join two skb lists, each list being a queue
1731 * @list: the new list to add
1732 * @head: the place to add it in the first list
1733 */
1734static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1735 struct sk_buff_head *head)
1736{
1737 if (!skb_queue_empty(list)) {
1738 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1739 head->qlen += list->qlen;
1740 }
1741}
1742
1743/**
1744 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1745 * @list: the new list to add
1746 * @head: the place to add it in the first list
1747 *
1748 * Each of the lists is a queue.
1749 * The list at @list is reinitialised
1750 */
1751static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1752 struct sk_buff_head *head)
1753{
1754 if (!skb_queue_empty(list)) {
1755 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1756 head->qlen += list->qlen;
1757 __skb_queue_head_init(list);
1758 }
1759}
1760
1761/**
1762 * __skb_queue_after - queue a buffer at the list head
1763 * @list: list to use
1764 * @prev: place after this buffer
1765 * @newsk: buffer to queue
1766 *
1767 * Queue a buffer int the middle of a list. This function takes no locks
1768 * and you must therefore hold required locks before calling it.
1769 *
1770 * A buffer cannot be placed on two lists at the same time.
1771 */
1772static inline void __skb_queue_after(struct sk_buff_head *list,
1773 struct sk_buff *prev,
1774 struct sk_buff *newsk)
1775{
1776 __skb_insert(newsk, prev, prev->next, list);
1777}
1778
1779void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1780 struct sk_buff_head *list);
1781
1782static inline void __skb_queue_before(struct sk_buff_head *list,
1783 struct sk_buff *next,
1784 struct sk_buff *newsk)
1785{
1786 __skb_insert(newsk, next->prev, next, list);
1787}
1788
1789/**
1790 * __skb_queue_head - queue a buffer at the list head
1791 * @list: list to use
1792 * @newsk: buffer to queue
1793 *
1794 * Queue a buffer at the start of a list. This function takes no locks
1795 * and you must therefore hold required locks before calling it.
1796 *
1797 * A buffer cannot be placed on two lists at the same time.
1798 */
1799void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1800static inline void __skb_queue_head(struct sk_buff_head *list,
1801 struct sk_buff *newsk)
1802{
1803 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1804}
1805
1806/**
1807 * __skb_queue_tail - queue a buffer at the list tail
1808 * @list: list to use
1809 * @newsk: buffer to queue
1810 *
1811 * Queue a buffer at the end of a list. This function takes no locks
1812 * and you must therefore hold required locks before calling it.
1813 *
1814 * A buffer cannot be placed on two lists at the same time.
1815 */
1816void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1817static inline void __skb_queue_tail(struct sk_buff_head *list,
1818 struct sk_buff *newsk)
1819{
1820 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1821}
1822
1823/*
1824 * remove sk_buff from list. _Must_ be called atomically, and with
1825 * the list known..
1826 */
1827void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1828static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1829{
1830 struct sk_buff *next, *prev;
1831
1832 list->qlen--;
1833 next = skb->next;
1834 prev = skb->prev;
1835 skb->next = skb->prev = NULL;
1836 next->prev = prev;
1837 prev->next = next;
1838}
1839
1840/**
1841 * __skb_dequeue - remove from the head of the queue
1842 * @list: list to dequeue from
1843 *
1844 * Remove the head of the list. This function does not take any locks
1845 * so must be used with appropriate locks held only. The head item is
1846 * returned or %NULL if the list is empty.
1847 */
1848struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1849static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1850{
1851 struct sk_buff *skb = skb_peek(list);
1852 if (skb)
1853 __skb_unlink(skb, list);
1854 return skb;
1855}
1856
1857/**
1858 * __skb_dequeue_tail - remove from the tail of the queue
1859 * @list: list to dequeue from
1860 *
1861 * Remove the tail of the list. This function does not take any locks
1862 * so must be used with appropriate locks held only. The tail item is
1863 * returned or %NULL if the list is empty.
1864 */
1865struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1866static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1867{
1868 struct sk_buff *skb = skb_peek_tail(list);
1869 if (skb)
1870 __skb_unlink(skb, list);
1871 return skb;
1872}
1873
1874
1875static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1876{
1877 return skb->data_len;
1878}
1879
1880static inline unsigned int skb_headlen(const struct sk_buff *skb)
1881{
1882 return skb->len - skb->data_len;
1883}
1884
1885static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
1886{
1887 unsigned int i, len = 0;
1888
1889 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1890 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1891 return len;
1892}
1893
1894static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1895{
1896 return skb_headlen(skb) + __skb_pagelen(skb);
1897}
1898
1899/**
1900 * __skb_fill_page_desc - initialise a paged fragment in an skb
1901 * @skb: buffer containing fragment to be initialised
1902 * @i: paged fragment index to initialise
1903 * @page: the page to use for this fragment
1904 * @off: the offset to the data with @page
1905 * @size: the length of the data
1906 *
1907 * Initialises the @i'th fragment of @skb to point to &size bytes at
1908 * offset @off within @page.
1909 *
1910 * Does not take any additional reference on the fragment.
1911 */
1912static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1913 struct page *page, int off, int size)
1914{
1915 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1916
1917 /*
1918 * Propagate page pfmemalloc to the skb if we can. The problem is
1919 * that not all callers have unique ownership of the page but rely
1920 * on page_is_pfmemalloc doing the right thing(tm).
1921 */
1922 frag->page.p = page;
1923 frag->page_offset = off;
1924 skb_frag_size_set(frag, size);
1925
1926 page = compound_head(page);
1927 if (page_is_pfmemalloc(page))
1928 skb->pfmemalloc = true;
1929}
1930
1931/**
1932 * skb_fill_page_desc - initialise a paged fragment in an skb
1933 * @skb: buffer containing fragment to be initialised
1934 * @i: paged fragment index to initialise
1935 * @page: the page to use for this fragment
1936 * @off: the offset to the data with @page
1937 * @size: the length of the data
1938 *
1939 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1940 * @skb to point to @size bytes at offset @off within @page. In
1941 * addition updates @skb such that @i is the last fragment.
1942 *
1943 * Does not take any additional reference on the fragment.
1944 */
1945static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1946 struct page *page, int off, int size)
1947{
1948 __skb_fill_page_desc(skb, i, page, off, size);
1949 skb_shinfo(skb)->nr_frags = i + 1;
1950}
1951
1952void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1953 int size, unsigned int truesize);
1954
1955void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1956 unsigned int truesize);
1957
1958#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1959#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1960#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1961
1962#ifdef NET_SKBUFF_DATA_USES_OFFSET
1963static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1964{
1965 return skb->head + skb->tail;
1966}
1967
1968static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1969{
1970 skb->tail = skb->data - skb->head;
1971}
1972
1973static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1974{
1975 skb_reset_tail_pointer(skb);
1976 skb->tail += offset;
1977}
1978
1979#else /* NET_SKBUFF_DATA_USES_OFFSET */
1980static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1981{
1982 return skb->tail;
1983}
1984
1985static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1986{
1987 skb->tail = skb->data;
1988}
1989
1990static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1991{
1992 skb->tail = skb->data + offset;
1993}
1994
1995#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1996
1997/*
1998 * Add data to an sk_buff
1999 */
2000void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2001void *skb_put(struct sk_buff *skb, unsigned int len);
2002static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2003{
2004 void *tmp = skb_tail_pointer(skb);
2005 SKB_LINEAR_ASSERT(skb);
2006 skb->tail += len;
2007 skb->len += len;
2008 return tmp;
2009}
2010
2011static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2012{
2013 void *tmp = __skb_put(skb, len);
2014
2015 memset(tmp, 0, len);
2016 return tmp;
2017}
2018
2019static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2020 unsigned int len)
2021{
2022 void *tmp = __skb_put(skb, len);
2023
2024 memcpy(tmp, data, len);
2025 return tmp;
2026}
2027
2028static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2029{
2030 *(u8 *)__skb_put(skb, 1) = val;
2031}
2032
2033static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2034{
2035 void *tmp = skb_put(skb, len);
2036
2037 memset(tmp, 0, len);
2038
2039 return tmp;
2040}
2041
2042static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2043 unsigned int len)
2044{
2045 void *tmp = skb_put(skb, len);
2046
2047 memcpy(tmp, data, len);
2048
2049 return tmp;
2050}
2051
2052static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2053{
2054 *(u8 *)skb_put(skb, 1) = val;
2055}
2056
2057void *skb_push(struct sk_buff *skb, unsigned int len);
2058static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2059{
2060 skb->data -= len;
2061 skb->len += len;
2062 return skb->data;
2063}
2064
2065void *skb_pull(struct sk_buff *skb, unsigned int len);
2066static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2067{
2068 skb->len -= len;
2069 BUG_ON(skb->len < skb->data_len);
2070 return skb->data += len;
2071}
2072
2073static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2074{
2075 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2076}
2077
2078void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2079
2080static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2081{
2082 if (len > skb_headlen(skb) &&
2083 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2084 return NULL;
2085 skb->len -= len;
2086 return skb->data += len;
2087}
2088
2089static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2090{
2091 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2092}
2093
2094static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2095{
2096 if (likely(len <= skb_headlen(skb)))
2097 return 1;
2098 if (unlikely(len > skb->len))
2099 return 0;
2100 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2101}
2102
2103void skb_condense(struct sk_buff *skb);
2104
2105/**
2106 * skb_headroom - bytes at buffer head
2107 * @skb: buffer to check
2108 *
2109 * Return the number of bytes of free space at the head of an &sk_buff.
2110 */
2111static inline unsigned int skb_headroom(const struct sk_buff *skb)
2112{
2113 return skb->data - skb->head;
2114}
2115
2116/**
2117 * skb_tailroom - bytes at buffer end
2118 * @skb: buffer to check
2119 *
2120 * Return the number of bytes of free space at the tail of an sk_buff
2121 */
2122static inline int skb_tailroom(const struct sk_buff *skb)
2123{
2124 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2125}
2126
2127/**
2128 * skb_availroom - bytes at buffer end
2129 * @skb: buffer to check
2130 *
2131 * Return the number of bytes of free space at the tail of an sk_buff
2132 * allocated by sk_stream_alloc()
2133 */
2134static inline int skb_availroom(const struct sk_buff *skb)
2135{
2136 if (skb_is_nonlinear(skb))
2137 return 0;
2138
2139 return skb->end - skb->tail - skb->reserved_tailroom;
2140}
2141
2142/**
2143 * skb_reserve - adjust headroom
2144 * @skb: buffer to alter
2145 * @len: bytes to move
2146 *
2147 * Increase the headroom of an empty &sk_buff by reducing the tail
2148 * room. This is only allowed for an empty buffer.
2149 */
2150static inline void skb_reserve(struct sk_buff *skb, int len)
2151{
2152 skb->data += len;
2153 skb->tail += len;
2154}
2155
2156/**
2157 * skb_tailroom_reserve - adjust reserved_tailroom
2158 * @skb: buffer to alter
2159 * @mtu: maximum amount of headlen permitted
2160 * @needed_tailroom: minimum amount of reserved_tailroom
2161 *
2162 * Set reserved_tailroom so that headlen can be as large as possible but
2163 * not larger than mtu and tailroom cannot be smaller than
2164 * needed_tailroom.
2165 * The required headroom should already have been reserved before using
2166 * this function.
2167 */
2168static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2169 unsigned int needed_tailroom)
2170{
2171 SKB_LINEAR_ASSERT(skb);
2172 if (mtu < skb_tailroom(skb) - needed_tailroom)
2173 /* use at most mtu */
2174 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2175 else
2176 /* use up to all available space */
2177 skb->reserved_tailroom = needed_tailroom;
2178}
2179
2180#define ENCAP_TYPE_ETHER 0
2181#define ENCAP_TYPE_IPPROTO 1
2182
2183static inline void skb_set_inner_protocol(struct sk_buff *skb,
2184 __be16 protocol)
2185{
2186 skb->inner_protocol = protocol;
2187 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2188}
2189
2190static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2191 __u8 ipproto)
2192{
2193 skb->inner_ipproto = ipproto;
2194 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2195}
2196
2197static inline void skb_reset_inner_headers(struct sk_buff *skb)
2198{
2199 skb->inner_mac_header = skb->mac_header;
2200 skb->inner_network_header = skb->network_header;
2201 skb->inner_transport_header = skb->transport_header;
2202}
2203
2204static inline void skb_reset_mac_len(struct sk_buff *skb)
2205{
2206 skb->mac_len = skb->network_header - skb->mac_header;
2207}
2208
2209static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2210 *skb)
2211{
2212 return skb->head + skb->inner_transport_header;
2213}
2214
2215static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2216{
2217 return skb_inner_transport_header(skb) - skb->data;
2218}
2219
2220static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2221{
2222 skb->inner_transport_header = skb->data - skb->head;
2223}
2224
2225static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2226 const int offset)
2227{
2228 skb_reset_inner_transport_header(skb);
2229 skb->inner_transport_header += offset;
2230}
2231
2232static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2233{
2234 return skb->head + skb->inner_network_header;
2235}
2236
2237static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2238{
2239 skb->inner_network_header = skb->data - skb->head;
2240}
2241
2242static inline void skb_set_inner_network_header(struct sk_buff *skb,
2243 const int offset)
2244{
2245 skb_reset_inner_network_header(skb);
2246 skb->inner_network_header += offset;
2247}
2248
2249static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2250{
2251 return skb->head + skb->inner_mac_header;
2252}
2253
2254static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2255{
2256 skb->inner_mac_header = skb->data - skb->head;
2257}
2258
2259static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2260 const int offset)
2261{
2262 skb_reset_inner_mac_header(skb);
2263 skb->inner_mac_header += offset;
2264}
2265static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2266{
2267 return skb->transport_header != (typeof(skb->transport_header))~0U;
2268}
2269
2270static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2271{
2272 return skb->head + skb->transport_header;
2273}
2274
2275static inline void skb_reset_transport_header(struct sk_buff *skb)
2276{
2277 skb->transport_header = skb->data - skb->head;
2278}
2279
2280static inline void skb_set_transport_header(struct sk_buff *skb,
2281 const int offset)
2282{
2283 skb_reset_transport_header(skb);
2284 skb->transport_header += offset;
2285}
2286
2287static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2288{
2289 return skb->head + skb->network_header;
2290}
2291
2292static inline void skb_reset_network_header(struct sk_buff *skb)
2293{
2294 skb->network_header = skb->data - skb->head;
2295}
2296
2297static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2298{
2299 skb_reset_network_header(skb);
2300 skb->network_header += offset;
2301}
2302
2303static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2304{
2305 return skb->head + skb->mac_header;
2306}
2307
2308static inline int skb_mac_offset(const struct sk_buff *skb)
2309{
2310 return skb_mac_header(skb) - skb->data;
2311}
2312
2313static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2314{
2315 return skb->network_header - skb->mac_header;
2316}
2317
2318static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2319{
2320 return skb->mac_header != (typeof(skb->mac_header))~0U;
2321}
2322
2323static inline void skb_reset_mac_header(struct sk_buff *skb)
2324{
2325 skb->mac_header = skb->data - skb->head;
2326}
2327
2328static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2329{
2330 skb_reset_mac_header(skb);
2331 skb->mac_header += offset;
2332}
2333
2334static inline void skb_pop_mac_header(struct sk_buff *skb)
2335{
2336 skb->mac_header = skb->network_header;
2337}
2338
2339static inline void skb_probe_transport_header(struct sk_buff *skb,
2340 const int offset_hint)
2341{
2342 struct flow_keys keys;
2343
2344 if (skb_transport_header_was_set(skb))
2345 return;
2346 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2347 skb_set_transport_header(skb, keys.control.thoff);
2348 else
2349 skb_set_transport_header(skb, offset_hint);
2350}
2351
2352static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2353{
2354 if (skb_mac_header_was_set(skb)) {
2355 const unsigned char *old_mac = skb_mac_header(skb);
2356
2357 skb_set_mac_header(skb, -skb->mac_len);
2358 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2359 }
2360}
2361
2362static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2363{
2364 return skb->csum_start - skb_headroom(skb);
2365}
2366
2367static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2368{
2369 return skb->head + skb->csum_start;
2370}
2371
2372static inline int skb_transport_offset(const struct sk_buff *skb)
2373{
2374 return skb_transport_header(skb) - skb->data;
2375}
2376
2377static inline u32 skb_network_header_len(const struct sk_buff *skb)
2378{
2379 return skb->transport_header - skb->network_header;
2380}
2381
2382static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2383{
2384 return skb->inner_transport_header - skb->inner_network_header;
2385}
2386
2387static inline int skb_network_offset(const struct sk_buff *skb)
2388{
2389 return skb_network_header(skb) - skb->data;
2390}
2391
2392static inline int skb_inner_network_offset(const struct sk_buff *skb)
2393{
2394 return skb_inner_network_header(skb) - skb->data;
2395}
2396
2397static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2398{
2399 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2400}
2401
2402/*
2403 * CPUs often take a performance hit when accessing unaligned memory
2404 * locations. The actual performance hit varies, it can be small if the
2405 * hardware handles it or large if we have to take an exception and fix it
2406 * in software.
2407 *
2408 * Since an ethernet header is 14 bytes network drivers often end up with
2409 * the IP header at an unaligned offset. The IP header can be aligned by
2410 * shifting the start of the packet by 2 bytes. Drivers should do this
2411 * with:
2412 *
2413 * skb_reserve(skb, NET_IP_ALIGN);
2414 *
2415 * The downside to this alignment of the IP header is that the DMA is now
2416 * unaligned. On some architectures the cost of an unaligned DMA is high
2417 * and this cost outweighs the gains made by aligning the IP header.
2418 *
2419 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2420 * to be overridden.
2421 */
2422#ifndef NET_IP_ALIGN
2423#define NET_IP_ALIGN 2
2424#endif
2425
2426/*
2427 * The networking layer reserves some headroom in skb data (via
2428 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2429 * the header has to grow. In the default case, if the header has to grow
2430 * 32 bytes or less we avoid the reallocation.
2431 *
2432 * Unfortunately this headroom changes the DMA alignment of the resulting
2433 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2434 * on some architectures. An architecture can override this value,
2435 * perhaps setting it to a cacheline in size (since that will maintain
2436 * cacheline alignment of the DMA). It must be a power of 2.
2437 *
2438 * Various parts of the networking layer expect at least 32 bytes of
2439 * headroom, you should not reduce this.
2440 *
2441 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2442 * to reduce average number of cache lines per packet.
2443 * get_rps_cpus() for example only access one 64 bytes aligned block :
2444 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2445 */
2446#ifndef NET_SKB_PAD
2447#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2448#endif
2449
2450int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2451
2452static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2453{
2454 if (unlikely(skb_is_nonlinear(skb))) {
2455 WARN_ON(1);
2456 return;
2457 }
2458 skb->len = len;
2459 skb_set_tail_pointer(skb, len);
2460}
2461
2462static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2463{
2464 __skb_set_length(skb, len);
2465}
2466
2467void skb_trim(struct sk_buff *skb, unsigned int len);
2468
2469static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2470{
2471 if (skb->data_len)
2472 return ___pskb_trim(skb, len);
2473 __skb_trim(skb, len);
2474 return 0;
2475}
2476
2477static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2478{
2479 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2480}
2481
2482/**
2483 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2484 * @skb: buffer to alter
2485 * @len: new length
2486 *
2487 * This is identical to pskb_trim except that the caller knows that
2488 * the skb is not cloned so we should never get an error due to out-
2489 * of-memory.
2490 */
2491static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2492{
2493 int err = pskb_trim(skb, len);
2494 BUG_ON(err);
2495}
2496
2497static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2498{
2499 unsigned int diff = len - skb->len;
2500
2501 if (skb_tailroom(skb) < diff) {
2502 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2503 GFP_ATOMIC);
2504 if (ret)
2505 return ret;
2506 }
2507 __skb_set_length(skb, len);
2508 return 0;
2509}
2510
2511/**
2512 * skb_orphan - orphan a buffer
2513 * @skb: buffer to orphan
2514 *
2515 * If a buffer currently has an owner then we call the owner's
2516 * destructor function and make the @skb unowned. The buffer continues
2517 * to exist but is no longer charged to its former owner.
2518 */
2519static inline void skb_orphan(struct sk_buff *skb)
2520{
2521 if (skb->destructor) {
2522 skb->destructor(skb);
2523 skb->destructor = NULL;
2524 skb->sk = NULL;
2525 } else {
2526 BUG_ON(skb->sk);
2527 }
2528}
2529
2530/**
2531 * skb_orphan_frags - orphan the frags contained in a buffer
2532 * @skb: buffer to orphan frags from
2533 * @gfp_mask: allocation mask for replacement pages
2534 *
2535 * For each frag in the SKB which needs a destructor (i.e. has an
2536 * owner) create a copy of that frag and release the original
2537 * page by calling the destructor.
2538 */
2539static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2540{
2541 if (likely(!skb_zcopy(skb)))
2542 return 0;
2543 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2544 return 0;
2545 return skb_copy_ubufs(skb, gfp_mask);
2546}
2547
2548/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2549static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2550{
2551 if (likely(!skb_zcopy(skb)))
2552 return 0;
2553 return skb_copy_ubufs(skb, gfp_mask);
2554}
2555
2556/**
2557 * __skb_queue_purge - empty a list
2558 * @list: list to empty
2559 *
2560 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2561 * the list and one reference dropped. This function does not take the
2562 * list lock and the caller must hold the relevant locks to use it.
2563 */
2564void skb_queue_purge(struct sk_buff_head *list);
2565static inline void __skb_queue_purge(struct sk_buff_head *list)
2566{
2567 struct sk_buff *skb;
2568 while ((skb = __skb_dequeue(list)) != NULL)
2569 kfree_skb(skb);
2570}
2571
2572void skb_rbtree_purge(struct rb_root *root);
2573
2574void *netdev_alloc_frag(unsigned int fragsz);
2575
2576struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2577 gfp_t gfp_mask);
2578
2579/**
2580 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2581 * @dev: network device to receive on
2582 * @length: length to allocate
2583 *
2584 * Allocate a new &sk_buff and assign it a usage count of one. The
2585 * buffer has unspecified headroom built in. Users should allocate
2586 * the headroom they think they need without accounting for the
2587 * built in space. The built in space is used for optimisations.
2588 *
2589 * %NULL is returned if there is no free memory. Although this function
2590 * allocates memory it can be called from an interrupt.
2591 */
2592static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2593 unsigned int length)
2594{
2595 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2596}
2597
2598/* legacy helper around __netdev_alloc_skb() */
2599static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2600 gfp_t gfp_mask)
2601{
2602 return __netdev_alloc_skb(NULL, length, gfp_mask);
2603}
2604
2605/* legacy helper around netdev_alloc_skb() */
2606static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2607{
2608 return netdev_alloc_skb(NULL, length);
2609}
2610
2611
2612static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2613 unsigned int length, gfp_t gfp)
2614{
2615 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2616
2617 if (NET_IP_ALIGN && skb)
2618 skb_reserve(skb, NET_IP_ALIGN);
2619 return skb;
2620}
2621
2622static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2623 unsigned int length)
2624{
2625 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2626}
2627
2628static inline void skb_free_frag(void *addr)
2629{
2630 page_frag_free(addr);
2631}
2632
2633void *napi_alloc_frag(unsigned int fragsz);
2634struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2635 unsigned int length, gfp_t gfp_mask);
2636static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2637 unsigned int length)
2638{
2639 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2640}
2641void napi_consume_skb(struct sk_buff *skb, int budget);
2642
2643void __kfree_skb_flush(void);
2644void __kfree_skb_defer(struct sk_buff *skb);
2645
2646/**
2647 * __dev_alloc_pages - allocate page for network Rx
2648 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2649 * @order: size of the allocation
2650 *
2651 * Allocate a new page.
2652 *
2653 * %NULL is returned if there is no free memory.
2654*/
2655static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2656 unsigned int order)
2657{
2658 /* This piece of code contains several assumptions.
2659 * 1. This is for device Rx, therefor a cold page is preferred.
2660 * 2. The expectation is the user wants a compound page.
2661 * 3. If requesting a order 0 page it will not be compound
2662 * due to the check to see if order has a value in prep_new_page
2663 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2664 * code in gfp_to_alloc_flags that should be enforcing this.
2665 */
2666 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2667
2668 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2669}
2670
2671static inline struct page *dev_alloc_pages(unsigned int order)
2672{
2673 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2674}
2675
2676/**
2677 * __dev_alloc_page - allocate a page for network Rx
2678 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2679 *
2680 * Allocate a new page.
2681 *
2682 * %NULL is returned if there is no free memory.
2683 */
2684static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2685{
2686 return __dev_alloc_pages(gfp_mask, 0);
2687}
2688
2689static inline struct page *dev_alloc_page(void)
2690{
2691 return dev_alloc_pages(0);
2692}
2693
2694/**
2695 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2696 * @page: The page that was allocated from skb_alloc_page
2697 * @skb: The skb that may need pfmemalloc set
2698 */
2699static inline void skb_propagate_pfmemalloc(struct page *page,
2700 struct sk_buff *skb)
2701{
2702 if (page_is_pfmemalloc(page))
2703 skb->pfmemalloc = true;
2704}
2705
2706/**
2707 * skb_frag_page - retrieve the page referred to by a paged fragment
2708 * @frag: the paged fragment
2709 *
2710 * Returns the &struct page associated with @frag.
2711 */
2712static inline struct page *skb_frag_page(const skb_frag_t *frag)
2713{
2714 return frag->page.p;
2715}
2716
2717/**
2718 * __skb_frag_ref - take an addition reference on a paged fragment.
2719 * @frag: the paged fragment
2720 *
2721 * Takes an additional reference on the paged fragment @frag.
2722 */
2723static inline void __skb_frag_ref(skb_frag_t *frag)
2724{
2725 get_page(skb_frag_page(frag));
2726}
2727
2728/**
2729 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2730 * @skb: the buffer
2731 * @f: the fragment offset.
2732 *
2733 * Takes an additional reference on the @f'th paged fragment of @skb.
2734 */
2735static inline void skb_frag_ref(struct sk_buff *skb, int f)
2736{
2737 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2738}
2739
2740/**
2741 * __skb_frag_unref - release a reference on a paged fragment.
2742 * @frag: the paged fragment
2743 *
2744 * Releases a reference on the paged fragment @frag.
2745 */
2746static inline void __skb_frag_unref(skb_frag_t *frag)
2747{
2748 put_page(skb_frag_page(frag));
2749}
2750
2751/**
2752 * skb_frag_unref - release a reference on a paged fragment of an skb.
2753 * @skb: the buffer
2754 * @f: the fragment offset
2755 *
2756 * Releases a reference on the @f'th paged fragment of @skb.
2757 */
2758static inline void skb_frag_unref(struct sk_buff *skb, int f)
2759{
2760 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2761}
2762
2763/**
2764 * skb_frag_address - gets the address of the data contained in a paged fragment
2765 * @frag: the paged fragment buffer
2766 *
2767 * Returns the address of the data within @frag. The page must already
2768 * be mapped.
2769 */
2770static inline void *skb_frag_address(const skb_frag_t *frag)
2771{
2772 return page_address(skb_frag_page(frag)) + frag->page_offset;
2773}
2774
2775/**
2776 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2777 * @frag: the paged fragment buffer
2778 *
2779 * Returns the address of the data within @frag. Checks that the page
2780 * is mapped and returns %NULL otherwise.
2781 */
2782static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2783{
2784 void *ptr = page_address(skb_frag_page(frag));
2785 if (unlikely(!ptr))
2786 return NULL;
2787
2788 return ptr + frag->page_offset;
2789}
2790
2791/**
2792 * __skb_frag_set_page - sets the page contained in a paged fragment
2793 * @frag: the paged fragment
2794 * @page: the page to set
2795 *
2796 * Sets the fragment @frag to contain @page.
2797 */
2798static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2799{
2800 frag->page.p = page;
2801}
2802
2803/**
2804 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2805 * @skb: the buffer
2806 * @f: the fragment offset
2807 * @page: the page to set
2808 *
2809 * Sets the @f'th fragment of @skb to contain @page.
2810 */
2811static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2812 struct page *page)
2813{
2814 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2815}
2816
2817bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2818
2819/**
2820 * skb_frag_dma_map - maps a paged fragment via the DMA API
2821 * @dev: the device to map the fragment to
2822 * @frag: the paged fragment to map
2823 * @offset: the offset within the fragment (starting at the
2824 * fragment's own offset)
2825 * @size: the number of bytes to map
2826 * @dir: the direction of the mapping (``PCI_DMA_*``)
2827 *
2828 * Maps the page associated with @frag to @device.
2829 */
2830static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2831 const skb_frag_t *frag,
2832 size_t offset, size_t size,
2833 enum dma_data_direction dir)
2834{
2835 return dma_map_page(dev, skb_frag_page(frag),
2836 frag->page_offset + offset, size, dir);
2837}
2838
2839static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2840 gfp_t gfp_mask)
2841{
2842 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2843}
2844
2845
2846static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2847 gfp_t gfp_mask)
2848{
2849 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2850}
2851
2852
2853/**
2854 * skb_clone_writable - is the header of a clone writable
2855 * @skb: buffer to check
2856 * @len: length up to which to write
2857 *
2858 * Returns true if modifying the header part of the cloned buffer
2859 * does not requires the data to be copied.
2860 */
2861static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2862{
2863 return !skb_header_cloned(skb) &&
2864 skb_headroom(skb) + len <= skb->hdr_len;
2865}
2866
2867static inline int skb_try_make_writable(struct sk_buff *skb,
2868 unsigned int write_len)
2869{
2870 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2871 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2872}
2873
2874static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2875 int cloned)
2876{
2877 int delta = 0;
2878
2879 if (headroom > skb_headroom(skb))
2880 delta = headroom - skb_headroom(skb);
2881
2882 if (delta || cloned)
2883 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2884 GFP_ATOMIC);
2885 return 0;
2886}
2887
2888/**
2889 * skb_cow - copy header of skb when it is required
2890 * @skb: buffer to cow
2891 * @headroom: needed headroom
2892 *
2893 * If the skb passed lacks sufficient headroom or its data part
2894 * is shared, data is reallocated. If reallocation fails, an error
2895 * is returned and original skb is not changed.
2896 *
2897 * The result is skb with writable area skb->head...skb->tail
2898 * and at least @headroom of space at head.
2899 */
2900static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2901{
2902 return __skb_cow(skb, headroom, skb_cloned(skb));
2903}
2904
2905/**
2906 * skb_cow_head - skb_cow but only making the head writable
2907 * @skb: buffer to cow
2908 * @headroom: needed headroom
2909 *
2910 * This function is identical to skb_cow except that we replace the
2911 * skb_cloned check by skb_header_cloned. It should be used when
2912 * you only need to push on some header and do not need to modify
2913 * the data.
2914 */
2915static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2916{
2917 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2918}
2919
2920/**
2921 * skb_padto - pad an skbuff up to a minimal size
2922 * @skb: buffer to pad
2923 * @len: minimal length
2924 *
2925 * Pads up a buffer to ensure the trailing bytes exist and are
2926 * blanked. If the buffer already contains sufficient data it
2927 * is untouched. Otherwise it is extended. Returns zero on
2928 * success. The skb is freed on error.
2929 */
2930static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2931{
2932 unsigned int size = skb->len;
2933 if (likely(size >= len))
2934 return 0;
2935 return skb_pad(skb, len - size);
2936}
2937
2938/**
2939 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2940 * @skb: buffer to pad
2941 * @len: minimal length
2942 * @free_on_error: free buffer on error
2943 *
2944 * Pads up a buffer to ensure the trailing bytes exist and are
2945 * blanked. If the buffer already contains sufficient data it
2946 * is untouched. Otherwise it is extended. Returns zero on
2947 * success. The skb is freed on error if @free_on_error is true.
2948 */
2949static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
2950 bool free_on_error)
2951{
2952 unsigned int size = skb->len;
2953
2954 if (unlikely(size < len)) {
2955 len -= size;
2956 if (__skb_pad(skb, len, free_on_error))
2957 return -ENOMEM;
2958 __skb_put(skb, len);
2959 }
2960 return 0;
2961}
2962
2963/**
2964 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2965 * @skb: buffer to pad
2966 * @len: minimal length
2967 *
2968 * Pads up a buffer to ensure the trailing bytes exist and are
2969 * blanked. If the buffer already contains sufficient data it
2970 * is untouched. Otherwise it is extended. Returns zero on
2971 * success. The skb is freed on error.
2972 */
2973static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2974{
2975 return __skb_put_padto(skb, len, true);
2976}
2977
2978static inline int skb_add_data(struct sk_buff *skb,
2979 struct iov_iter *from, int copy)
2980{
2981 const int off = skb->len;
2982
2983 if (skb->ip_summed == CHECKSUM_NONE) {
2984 __wsum csum = 0;
2985 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2986 &csum, from)) {
2987 skb->csum = csum_block_add(skb->csum, csum, off);
2988 return 0;
2989 }
2990 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2991 return 0;
2992
2993 __skb_trim(skb, off);
2994 return -EFAULT;
2995}
2996
2997static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2998 const struct page *page, int off)
2999{
3000 if (skb_zcopy(skb))
3001 return false;
3002 if (i) {
3003 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3004
3005 return page == skb_frag_page(frag) &&
3006 off == frag->page_offset + skb_frag_size(frag);
3007 }
3008 return false;
3009}
3010
3011static inline int __skb_linearize(struct sk_buff *skb)
3012{
3013 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3014}
3015
3016/**
3017 * skb_linearize - convert paged skb to linear one
3018 * @skb: buffer to linarize
3019 *
3020 * If there is no free memory -ENOMEM is returned, otherwise zero
3021 * is returned and the old skb data released.
3022 */
3023static inline int skb_linearize(struct sk_buff *skb)
3024{
3025 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3026}
3027
3028/**
3029 * skb_has_shared_frag - can any frag be overwritten
3030 * @skb: buffer to test
3031 *
3032 * Return true if the skb has at least one frag that might be modified
3033 * by an external entity (as in vmsplice()/sendfile())
3034 */
3035static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3036{
3037 return skb_is_nonlinear(skb) &&
3038 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3039}
3040
3041/**
3042 * skb_linearize_cow - make sure skb is linear and writable
3043 * @skb: buffer to process
3044 *
3045 * If there is no free memory -ENOMEM is returned, otherwise zero
3046 * is returned and the old skb data released.
3047 */
3048static inline int skb_linearize_cow(struct sk_buff *skb)
3049{
3050 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3051 __skb_linearize(skb) : 0;
3052}
3053
3054static __always_inline void
3055__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3056 unsigned int off)
3057{
3058 if (skb->ip_summed == CHECKSUM_COMPLETE)
3059 skb->csum = csum_block_sub(skb->csum,
3060 csum_partial(start, len, 0), off);
3061 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3062 skb_checksum_start_offset(skb) < 0)
3063 skb->ip_summed = CHECKSUM_NONE;
3064}
3065
3066/**
3067 * skb_postpull_rcsum - update checksum for received skb after pull
3068 * @skb: buffer to update
3069 * @start: start of data before pull
3070 * @len: length of data pulled
3071 *
3072 * After doing a pull on a received packet, you need to call this to
3073 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3074 * CHECKSUM_NONE so that it can be recomputed from scratch.
3075 */
3076static inline void skb_postpull_rcsum(struct sk_buff *skb,
3077 const void *start, unsigned int len)
3078{
3079 __skb_postpull_rcsum(skb, start, len, 0);
3080}
3081
3082static __always_inline void
3083__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3084 unsigned int off)
3085{
3086 if (skb->ip_summed == CHECKSUM_COMPLETE)
3087 skb->csum = csum_block_add(skb->csum,
3088 csum_partial(start, len, 0), off);
3089}
3090
3091/**
3092 * skb_postpush_rcsum - update checksum for received skb after push
3093 * @skb: buffer to update
3094 * @start: start of data after push
3095 * @len: length of data pushed
3096 *
3097 * After doing a push on a received packet, you need to call this to
3098 * update the CHECKSUM_COMPLETE checksum.
3099 */
3100static inline void skb_postpush_rcsum(struct sk_buff *skb,
3101 const void *start, unsigned int len)
3102{
3103 __skb_postpush_rcsum(skb, start, len, 0);
3104}
3105
3106void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3107
3108/**
3109 * skb_push_rcsum - push skb and update receive checksum
3110 * @skb: buffer to update
3111 * @len: length of data pulled
3112 *
3113 * This function performs an skb_push on the packet and updates
3114 * the CHECKSUM_COMPLETE checksum. It should be used on
3115 * receive path processing instead of skb_push unless you know
3116 * that the checksum difference is zero (e.g., a valid IP header)
3117 * or you are setting ip_summed to CHECKSUM_NONE.
3118 */
3119static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3120{
3121 skb_push(skb, len);
3122 skb_postpush_rcsum(skb, skb->data, len);
3123 return skb->data;
3124}
3125
3126/**
3127 * pskb_trim_rcsum - trim received skb and update checksum
3128 * @skb: buffer to trim
3129 * @len: new length
3130 *
3131 * This is exactly the same as pskb_trim except that it ensures the
3132 * checksum of received packets are still valid after the operation.
3133 */
3134
3135static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3136{
3137 if (likely(len >= skb->len))
3138 return 0;
3139 if (skb->ip_summed == CHECKSUM_COMPLETE)
3140 skb->ip_summed = CHECKSUM_NONE;
3141 return __pskb_trim(skb, len);
3142}
3143
3144static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3145{
3146 if (skb->ip_summed == CHECKSUM_COMPLETE)
3147 skb->ip_summed = CHECKSUM_NONE;
3148 __skb_trim(skb, len);
3149 return 0;
3150}
3151
3152static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3153{
3154 if (skb->ip_summed == CHECKSUM_COMPLETE)
3155 skb->ip_summed = CHECKSUM_NONE;
3156 return __skb_grow(skb, len);
3157}
3158
3159#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3160#define skb_rb_first(root) rb_to_skb(rb_first(root))
3161#define skb_rb_last(root) rb_to_skb(rb_last(root))
3162#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3163#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3164
3165#define skb_queue_walk(queue, skb) \
3166 for (skb = (queue)->next; \
3167 skb != (struct sk_buff *)(queue); \
3168 skb = skb->next)
3169
3170#define skb_queue_walk_safe(queue, skb, tmp) \
3171 for (skb = (queue)->next, tmp = skb->next; \
3172 skb != (struct sk_buff *)(queue); \
3173 skb = tmp, tmp = skb->next)
3174
3175#define skb_queue_walk_from(queue, skb) \
3176 for (; skb != (struct sk_buff *)(queue); \
3177 skb = skb->next)
3178
3179#define skb_rbtree_walk(skb, root) \
3180 for (skb = skb_rb_first(root); skb != NULL; \
3181 skb = skb_rb_next(skb))
3182
3183#define skb_rbtree_walk_from(skb) \
3184 for (; skb != NULL; \
3185 skb = skb_rb_next(skb))
3186
3187#define skb_rbtree_walk_from_safe(skb, tmp) \
3188 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3189 skb = tmp)
3190
3191#define skb_queue_walk_from_safe(queue, skb, tmp) \
3192 for (tmp = skb->next; \
3193 skb != (struct sk_buff *)(queue); \
3194 skb = tmp, tmp = skb->next)
3195
3196#define skb_queue_reverse_walk(queue, skb) \
3197 for (skb = (queue)->prev; \
3198 skb != (struct sk_buff *)(queue); \
3199 skb = skb->prev)
3200
3201#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3202 for (skb = (queue)->prev, tmp = skb->prev; \
3203 skb != (struct sk_buff *)(queue); \
3204 skb = tmp, tmp = skb->prev)
3205
3206#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3207 for (tmp = skb->prev; \
3208 skb != (struct sk_buff *)(queue); \
3209 skb = tmp, tmp = skb->prev)
3210
3211static inline bool skb_has_frag_list(const struct sk_buff *skb)
3212{
3213 return skb_shinfo(skb)->frag_list != NULL;
3214}
3215
3216static inline void skb_frag_list_init(struct sk_buff *skb)
3217{
3218 skb_shinfo(skb)->frag_list = NULL;
3219}
3220
3221#define skb_walk_frags(skb, iter) \
3222 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3223
3224
3225int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3226 const struct sk_buff *skb);
3227struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3228 struct sk_buff_head *queue,
3229 unsigned int flags,
3230 void (*destructor)(struct sock *sk,
3231 struct sk_buff *skb),
3232 int *peeked, int *off, int *err,
3233 struct sk_buff **last);
3234struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3235 void (*destructor)(struct sock *sk,
3236 struct sk_buff *skb),
3237 int *peeked, int *off, int *err,
3238 struct sk_buff **last);
3239struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3240 void (*destructor)(struct sock *sk,
3241 struct sk_buff *skb),
3242 int *peeked, int *off, int *err);
3243struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3244 int *err);
3245unsigned int datagram_poll(struct file *file, struct socket *sock,
3246 struct poll_table_struct *wait);
3247int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3248 struct iov_iter *to, int size);
3249static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3250 struct msghdr *msg, int size)
3251{
3252 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3253}
3254int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3255 struct msghdr *msg);
3256int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3257 struct iov_iter *from, int len);
3258int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3259void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3260void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3261static inline void skb_free_datagram_locked(struct sock *sk,
3262 struct sk_buff *skb)
3263{
3264 __skb_free_datagram_locked(sk, skb, 0);
3265}
3266int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3267int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3268int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3269__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3270 int len, __wsum csum);
3271int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3272 struct pipe_inode_info *pipe, unsigned int len,
3273 unsigned int flags);
3274int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3275 int len);
3276int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3277void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3278unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3279int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3280 int len, int hlen);
3281void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3282int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3283void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3284unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3285bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3286struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3287struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3288int skb_ensure_writable(struct sk_buff *skb, int write_len);
3289int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3290int skb_vlan_pop(struct sk_buff *skb);
3291int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3292struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3293 gfp_t gfp);
3294
3295static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3296{
3297 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3298}
3299
3300static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3301{
3302 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3303}
3304
3305struct skb_checksum_ops {
3306 __wsum (*update)(const void *mem, int len, __wsum wsum);
3307 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3308};
3309
3310extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3311
3312__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3313 __wsum csum, const struct skb_checksum_ops *ops);
3314__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3315 __wsum csum);
3316
3317static inline void * __must_check
3318__skb_header_pointer(const struct sk_buff *skb, int offset,
3319 int len, void *data, int hlen, void *buffer)
3320{
3321 if (hlen - offset >= len)
3322 return data + offset;
3323
3324 if (!skb ||
3325 skb_copy_bits(skb, offset, buffer, len) < 0)
3326 return NULL;
3327
3328 return buffer;
3329}
3330
3331static inline void * __must_check
3332skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3333{
3334 return __skb_header_pointer(skb, offset, len, skb->data,
3335 skb_headlen(skb), buffer);
3336}
3337
3338/**
3339 * skb_needs_linearize - check if we need to linearize a given skb
3340 * depending on the given device features.
3341 * @skb: socket buffer to check
3342 * @features: net device features
3343 *
3344 * Returns true if either:
3345 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3346 * 2. skb is fragmented and the device does not support SG.
3347 */
3348static inline bool skb_needs_linearize(struct sk_buff *skb,
3349 netdev_features_t features)
3350{
3351 return skb_is_nonlinear(skb) &&
3352 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3353 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3354}
3355
3356static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3357 void *to,
3358 const unsigned int len)
3359{
3360 memcpy(to, skb->data, len);
3361}
3362
3363static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3364 const int offset, void *to,
3365 const unsigned int len)
3366{
3367 memcpy(to, skb->data + offset, len);
3368}
3369
3370static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3371 const void *from,
3372 const unsigned int len)
3373{
3374 memcpy(skb->data, from, len);
3375}
3376
3377static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3378 const int offset,
3379 const void *from,
3380 const unsigned int len)
3381{
3382 memcpy(skb->data + offset, from, len);
3383}
3384
3385void skb_init(void);
3386
3387static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3388{
3389 return skb->tstamp;
3390}
3391
3392/**
3393 * skb_get_timestamp - get timestamp from a skb
3394 * @skb: skb to get stamp from
3395 * @stamp: pointer to struct timeval to store stamp in
3396 *
3397 * Timestamps are stored in the skb as offsets to a base timestamp.
3398 * This function converts the offset back to a struct timeval and stores
3399 * it in stamp.
3400 */
3401static inline void skb_get_timestamp(const struct sk_buff *skb,
3402 struct timeval *stamp)
3403{
3404 *stamp = ktime_to_timeval(skb->tstamp);
3405}
3406
3407static inline void skb_get_timestampns(const struct sk_buff *skb,
3408 struct timespec *stamp)
3409{
3410 *stamp = ktime_to_timespec(skb->tstamp);
3411}
3412
3413static inline void __net_timestamp(struct sk_buff *skb)
3414{
3415 skb->tstamp = ktime_get_real();
3416}
3417
3418static inline ktime_t net_timedelta(ktime_t t)
3419{
3420 return ktime_sub(ktime_get_real(), t);
3421}
3422
3423static inline ktime_t net_invalid_timestamp(void)
3424{
3425 return 0;
3426}
3427
3428static inline u8 skb_metadata_len(const struct sk_buff *skb)
3429{
3430 return skb_shinfo(skb)->meta_len;
3431}
3432
3433static inline void *skb_metadata_end(const struct sk_buff *skb)
3434{
3435 return skb_mac_header(skb);
3436}
3437
3438static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3439 const struct sk_buff *skb_b,
3440 u8 meta_len)
3441{
3442 const void *a = skb_metadata_end(skb_a);
3443 const void *b = skb_metadata_end(skb_b);
3444 /* Using more efficient varaiant than plain call to memcmp(). */
3445#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3446 u64 diffs = 0;
3447
3448 switch (meta_len) {
3449#define __it(x, op) (x -= sizeof(u##op))
3450#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3451 case 32: diffs |= __it_diff(a, b, 64);
3452 case 24: diffs |= __it_diff(a, b, 64);
3453 case 16: diffs |= __it_diff(a, b, 64);
3454 case 8: diffs |= __it_diff(a, b, 64);
3455 break;
3456 case 28: diffs |= __it_diff(a, b, 64);
3457 case 20: diffs |= __it_diff(a, b, 64);
3458 case 12: diffs |= __it_diff(a, b, 64);
3459 case 4: diffs |= __it_diff(a, b, 32);
3460 break;
3461 }
3462 return diffs;
3463#else
3464 return memcmp(a - meta_len, b - meta_len, meta_len);
3465#endif
3466}
3467
3468static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3469 const struct sk_buff *skb_b)
3470{
3471 u8 len_a = skb_metadata_len(skb_a);
3472 u8 len_b = skb_metadata_len(skb_b);
3473
3474 if (!(len_a | len_b))
3475 return false;
3476
3477 return len_a != len_b ?
3478 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3479}
3480
3481static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3482{
3483 skb_shinfo(skb)->meta_len = meta_len;
3484}
3485
3486static inline void skb_metadata_clear(struct sk_buff *skb)
3487{
3488 skb_metadata_set(skb, 0);
3489}
3490
3491struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3492
3493#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3494
3495void skb_clone_tx_timestamp(struct sk_buff *skb);
3496bool skb_defer_rx_timestamp(struct sk_buff *skb);
3497
3498#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3499
3500static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3501{
3502}
3503
3504static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3505{
3506 return false;
3507}
3508
3509#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3510
3511/**
3512 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3513 *
3514 * PHY drivers may accept clones of transmitted packets for
3515 * timestamping via their phy_driver.txtstamp method. These drivers
3516 * must call this function to return the skb back to the stack with a
3517 * timestamp.
3518 *
3519 * @skb: clone of the the original outgoing packet
3520 * @hwtstamps: hardware time stamps
3521 *
3522 */
3523void skb_complete_tx_timestamp(struct sk_buff *skb,
3524 struct skb_shared_hwtstamps *hwtstamps);
3525
3526void __skb_tstamp_tx(struct sk_buff *orig_skb,
3527 struct skb_shared_hwtstamps *hwtstamps,
3528 struct sock *sk, int tstype);
3529
3530/**
3531 * skb_tstamp_tx - queue clone of skb with send time stamps
3532 * @orig_skb: the original outgoing packet
3533 * @hwtstamps: hardware time stamps, may be NULL if not available
3534 *
3535 * If the skb has a socket associated, then this function clones the
3536 * skb (thus sharing the actual data and optional structures), stores
3537 * the optional hardware time stamping information (if non NULL) or
3538 * generates a software time stamp (otherwise), then queues the clone
3539 * to the error queue of the socket. Errors are silently ignored.
3540 */
3541void skb_tstamp_tx(struct sk_buff *orig_skb,
3542 struct skb_shared_hwtstamps *hwtstamps);
3543
3544/**
3545 * skb_tx_timestamp() - Driver hook for transmit timestamping
3546 *
3547 * Ethernet MAC Drivers should call this function in their hard_xmit()
3548 * function immediately before giving the sk_buff to the MAC hardware.
3549 *
3550 * Specifically, one should make absolutely sure that this function is
3551 * called before TX completion of this packet can trigger. Otherwise
3552 * the packet could potentially already be freed.
3553 *
3554 * @skb: A socket buffer.
3555 */
3556static inline void skb_tx_timestamp(struct sk_buff *skb)
3557{
3558 skb_clone_tx_timestamp(skb);
3559 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3560 skb_tstamp_tx(skb, NULL);
3561}
3562
3563/**
3564 * skb_complete_wifi_ack - deliver skb with wifi status
3565 *
3566 * @skb: the original outgoing packet
3567 * @acked: ack status
3568 *
3569 */
3570void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3571
3572__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3573__sum16 __skb_checksum_complete(struct sk_buff *skb);
3574
3575static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3576{
3577 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3578 skb->csum_valid ||
3579 (skb->ip_summed == CHECKSUM_PARTIAL &&
3580 skb_checksum_start_offset(skb) >= 0));
3581}
3582
3583/**
3584 * skb_checksum_complete - Calculate checksum of an entire packet
3585 * @skb: packet to process
3586 *
3587 * This function calculates the checksum over the entire packet plus
3588 * the value of skb->csum. The latter can be used to supply the
3589 * checksum of a pseudo header as used by TCP/UDP. It returns the
3590 * checksum.
3591 *
3592 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3593 * this function can be used to verify that checksum on received
3594 * packets. In that case the function should return zero if the
3595 * checksum is correct. In particular, this function will return zero
3596 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3597 * hardware has already verified the correctness of the checksum.
3598 */
3599static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3600{
3601 return skb_csum_unnecessary(skb) ?
3602 0 : __skb_checksum_complete(skb);
3603}
3604
3605static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3606{
3607 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3608 if (skb->csum_level == 0)
3609 skb->ip_summed = CHECKSUM_NONE;
3610 else
3611 skb->csum_level--;
3612 }
3613}
3614
3615static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3616{
3617 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3618 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3619 skb->csum_level++;
3620 } else if (skb->ip_summed == CHECKSUM_NONE) {
3621 skb->ip_summed = CHECKSUM_UNNECESSARY;
3622 skb->csum_level = 0;
3623 }
3624}
3625
3626/* Check if we need to perform checksum complete validation.
3627 *
3628 * Returns true if checksum complete is needed, false otherwise
3629 * (either checksum is unnecessary or zero checksum is allowed).
3630 */
3631static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3632 bool zero_okay,
3633 __sum16 check)
3634{
3635 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3636 skb->csum_valid = 1;
3637 __skb_decr_checksum_unnecessary(skb);
3638 return false;
3639 }
3640
3641 return true;
3642}
3643
3644/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3645 * in checksum_init.
3646 */
3647#define CHECKSUM_BREAK 76
3648
3649/* Unset checksum-complete
3650 *
3651 * Unset checksum complete can be done when packet is being modified
3652 * (uncompressed for instance) and checksum-complete value is
3653 * invalidated.
3654 */
3655static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3656{
3657 if (skb->ip_summed == CHECKSUM_COMPLETE)
3658 skb->ip_summed = CHECKSUM_NONE;
3659}
3660
3661/* Validate (init) checksum based on checksum complete.
3662 *
3663 * Return values:
3664 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3665 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3666 * checksum is stored in skb->csum for use in __skb_checksum_complete
3667 * non-zero: value of invalid checksum
3668 *
3669 */
3670static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3671 bool complete,
3672 __wsum psum)
3673{
3674 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3675 if (!csum_fold(csum_add(psum, skb->csum))) {
3676 skb->csum_valid = 1;
3677 return 0;
3678 }
3679 }
3680
3681 skb->csum = psum;
3682
3683 if (complete || skb->len <= CHECKSUM_BREAK) {
3684 __sum16 csum;
3685
3686 csum = __skb_checksum_complete(skb);
3687 skb->csum_valid = !csum;
3688 return csum;
3689 }
3690
3691 return 0;
3692}
3693
3694static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3695{
3696 return 0;
3697}
3698
3699/* Perform checksum validate (init). Note that this is a macro since we only
3700 * want to calculate the pseudo header which is an input function if necessary.
3701 * First we try to validate without any computation (checksum unnecessary) and
3702 * then calculate based on checksum complete calling the function to compute
3703 * pseudo header.
3704 *
3705 * Return values:
3706 * 0: checksum is validated or try to in skb_checksum_complete
3707 * non-zero: value of invalid checksum
3708 */
3709#define __skb_checksum_validate(skb, proto, complete, \
3710 zero_okay, check, compute_pseudo) \
3711({ \
3712 __sum16 __ret = 0; \
3713 skb->csum_valid = 0; \
3714 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3715 __ret = __skb_checksum_validate_complete(skb, \
3716 complete, compute_pseudo(skb, proto)); \
3717 __ret; \
3718})
3719
3720#define skb_checksum_init(skb, proto, compute_pseudo) \
3721 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3722
3723#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3724 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3725
3726#define skb_checksum_validate(skb, proto, compute_pseudo) \
3727 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3728
3729#define skb_checksum_validate_zero_check(skb, proto, check, \
3730 compute_pseudo) \
3731 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3732
3733#define skb_checksum_simple_validate(skb) \
3734 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3735
3736static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3737{
3738 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3739}
3740
3741static inline void __skb_checksum_convert(struct sk_buff *skb,
3742 __sum16 check, __wsum pseudo)
3743{
3744 skb->csum = ~pseudo;
3745 skb->ip_summed = CHECKSUM_COMPLETE;
3746}
3747
3748#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3749do { \
3750 if (__skb_checksum_convert_check(skb)) \
3751 __skb_checksum_convert(skb, check, \
3752 compute_pseudo(skb, proto)); \
3753} while (0)
3754
3755static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3756 u16 start, u16 offset)
3757{
3758 skb->ip_summed = CHECKSUM_PARTIAL;
3759 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3760 skb->csum_offset = offset - start;
3761}
3762
3763/* Update skbuf and packet to reflect the remote checksum offload operation.
3764 * When called, ptr indicates the starting point for skb->csum when
3765 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3766 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3767 */
3768static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3769 int start, int offset, bool nopartial)
3770{
3771 __wsum delta;
3772
3773 if (!nopartial) {
3774 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3775 return;
3776 }
3777
3778 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3779 __skb_checksum_complete(skb);
3780 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3781 }
3782
3783 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3784
3785 /* Adjust skb->csum since we changed the packet */
3786 skb->csum = csum_add(skb->csum, delta);
3787}
3788
3789static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3790{
3791#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3792 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3793#else
3794 return NULL;
3795#endif
3796}
3797
3798#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3799void nf_conntrack_destroy(struct nf_conntrack *nfct);
3800static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3801{
3802 if (nfct && atomic_dec_and_test(&nfct->use))
3803 nf_conntrack_destroy(nfct);
3804}
3805static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3806{
3807 if (nfct)
3808 atomic_inc(&nfct->use);
3809}
3810#endif
3811#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3812static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3813{
3814 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3815 kfree(nf_bridge);
3816}
3817static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3818{
3819 if (nf_bridge)
3820 refcount_inc(&nf_bridge->use);
3821}
3822#endif /* CONFIG_BRIDGE_NETFILTER */
3823static inline void nf_reset(struct sk_buff *skb)
3824{
3825#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3826 nf_conntrack_put(skb_nfct(skb));
3827 skb->_nfct = 0;
3828#endif
3829#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3830 nf_bridge_put(skb->nf_bridge);
3831 skb->nf_bridge = NULL;
3832#endif
3833}
3834
3835static inline void nf_reset_trace(struct sk_buff *skb)
3836{
3837#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3838 skb->nf_trace = 0;
3839#endif
3840}
3841
3842static inline void ipvs_reset(struct sk_buff *skb)
3843{
3844#if IS_ENABLED(CONFIG_IP_VS)
3845 skb->ipvs_property = 0;
3846#endif
3847}
3848
3849/* Note: This doesn't put any conntrack and bridge info in dst. */
3850static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3851 bool copy)
3852{
3853#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3854 dst->_nfct = src->_nfct;
3855 nf_conntrack_get(skb_nfct(src));
3856#endif
3857#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3858 dst->nf_bridge = src->nf_bridge;
3859 nf_bridge_get(src->nf_bridge);
3860#endif
3861#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3862 if (copy)
3863 dst->nf_trace = src->nf_trace;
3864#endif
3865}
3866
3867static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3868{
3869#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3870 nf_conntrack_put(skb_nfct(dst));
3871#endif
3872#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3873 nf_bridge_put(dst->nf_bridge);
3874#endif
3875 __nf_copy(dst, src, true);
3876}
3877
3878#ifdef CONFIG_NETWORK_SECMARK
3879static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3880{
3881 to->secmark = from->secmark;
3882}
3883
3884static inline void skb_init_secmark(struct sk_buff *skb)
3885{
3886 skb->secmark = 0;
3887}
3888#else
3889static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3890{ }
3891
3892static inline void skb_init_secmark(struct sk_buff *skb)
3893{ }
3894#endif
3895
3896static inline bool skb_irq_freeable(const struct sk_buff *skb)
3897{
3898 return !skb->destructor &&
3899#if IS_ENABLED(CONFIG_XFRM)
3900 !skb->sp &&
3901#endif
3902 !skb_nfct(skb) &&
3903 !skb->_skb_refdst &&
3904 !skb_has_frag_list(skb);
3905}
3906
3907static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3908{
3909 skb->queue_mapping = queue_mapping;
3910}
3911
3912static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3913{
3914 return skb->queue_mapping;
3915}
3916
3917static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3918{
3919 to->queue_mapping = from->queue_mapping;
3920}
3921
3922static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3923{
3924 skb->queue_mapping = rx_queue + 1;
3925}
3926
3927static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3928{
3929 return skb->queue_mapping - 1;
3930}
3931
3932static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3933{
3934 return skb->queue_mapping != 0;
3935}
3936
3937static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3938{
3939 skb->dst_pending_confirm = val;
3940}
3941
3942static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3943{
3944 return skb->dst_pending_confirm != 0;
3945}
3946
3947static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3948{
3949#ifdef CONFIG_XFRM
3950 return skb->sp;
3951#else
3952 return NULL;
3953#endif
3954}
3955
3956/* Keeps track of mac header offset relative to skb->head.
3957 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3958 * For non-tunnel skb it points to skb_mac_header() and for
3959 * tunnel skb it points to outer mac header.
3960 * Keeps track of level of encapsulation of network headers.
3961 */
3962struct skb_gso_cb {
3963 union {
3964 int mac_offset;
3965 int data_offset;
3966 };
3967 int encap_level;
3968 __wsum csum;
3969 __u16 csum_start;
3970};
3971#define SKB_SGO_CB_OFFSET 32
3972#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3973
3974static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3975{
3976 return (skb_mac_header(inner_skb) - inner_skb->head) -
3977 SKB_GSO_CB(inner_skb)->mac_offset;
3978}
3979
3980static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3981{
3982 int new_headroom, headroom;
3983 int ret;
3984
3985 headroom = skb_headroom(skb);
3986 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3987 if (ret)
3988 return ret;
3989
3990 new_headroom = skb_headroom(skb);
3991 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3992 return 0;
3993}
3994
3995static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3996{
3997 /* Do not update partial checksums if remote checksum is enabled. */
3998 if (skb->remcsum_offload)
3999 return;
4000
4001 SKB_GSO_CB(skb)->csum = res;
4002 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4003}
4004
4005/* Compute the checksum for a gso segment. First compute the checksum value
4006 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4007 * then add in skb->csum (checksum from csum_start to end of packet).
4008 * skb->csum and csum_start are then updated to reflect the checksum of the
4009 * resultant packet starting from the transport header-- the resultant checksum
4010 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4011 * header.
4012 */
4013static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4014{
4015 unsigned char *csum_start = skb_transport_header(skb);
4016 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4017 __wsum partial = SKB_GSO_CB(skb)->csum;
4018
4019 SKB_GSO_CB(skb)->csum = res;
4020 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4021
4022 return csum_fold(csum_partial(csum_start, plen, partial));
4023}
4024
4025static inline bool skb_is_gso(const struct sk_buff *skb)
4026{
4027 return skb_shinfo(skb)->gso_size;
4028}
4029
4030/* Note: Should be called only if skb_is_gso(skb) is true */
4031static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4032{
4033 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4034}
4035
4036static inline void skb_gso_reset(struct sk_buff *skb)
4037{
4038 skb_shinfo(skb)->gso_size = 0;
4039 skb_shinfo(skb)->gso_segs = 0;
4040 skb_shinfo(skb)->gso_type = 0;
4041}
4042
4043void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4044
4045static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4046{
4047 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4048 * wanted then gso_type will be set. */
4049 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4050
4051 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4052 unlikely(shinfo->gso_type == 0)) {
4053 __skb_warn_lro_forwarding(skb);
4054 return true;
4055 }
4056 return false;
4057}
4058
4059static inline void skb_forward_csum(struct sk_buff *skb)
4060{
4061 /* Unfortunately we don't support this one. Any brave souls? */
4062 if (skb->ip_summed == CHECKSUM_COMPLETE)
4063 skb->ip_summed = CHECKSUM_NONE;
4064}
4065
4066/**
4067 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4068 * @skb: skb to check
4069 *
4070 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4071 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4072 * use this helper, to document places where we make this assertion.
4073 */
4074static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4075{
4076#ifdef DEBUG
4077 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4078#endif
4079}
4080
4081bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4082
4083int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4084struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4085 unsigned int transport_len,
4086 __sum16(*skb_chkf)(struct sk_buff *skb));
4087
4088/**
4089 * skb_head_is_locked - Determine if the skb->head is locked down
4090 * @skb: skb to check
4091 *
4092 * The head on skbs build around a head frag can be removed if they are
4093 * not cloned. This function returns true if the skb head is locked down
4094 * due to either being allocated via kmalloc, or by being a clone with
4095 * multiple references to the head.
4096 */
4097static inline bool skb_head_is_locked(const struct sk_buff *skb)
4098{
4099 return !skb->head_frag || skb_cloned(skb);
4100}
4101
4102/**
4103 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4104 *
4105 * @skb: GSO skb
4106 *
4107 * skb_gso_network_seglen is used to determine the real size of the
4108 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4109 *
4110 * The MAC/L2 header is not accounted for.
4111 */
4112static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
4113{
4114 unsigned int hdr_len = skb_transport_header(skb) -
4115 skb_network_header(skb);
4116 return hdr_len + skb_gso_transport_seglen(skb);
4117}
4118
4119/* Local Checksum Offload.
4120 * Compute outer checksum based on the assumption that the
4121 * inner checksum will be offloaded later.
4122 * See Documentation/networking/checksum-offloads.txt for
4123 * explanation of how this works.
4124 * Fill in outer checksum adjustment (e.g. with sum of outer
4125 * pseudo-header) before calling.
4126 * Also ensure that inner checksum is in linear data area.
4127 */
4128static inline __wsum lco_csum(struct sk_buff *skb)
4129{
4130 unsigned char *csum_start = skb_checksum_start(skb);
4131 unsigned char *l4_hdr = skb_transport_header(skb);
4132 __wsum partial;
4133
4134 /* Start with complement of inner checksum adjustment */
4135 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4136 skb->csum_offset));
4137
4138 /* Add in checksum of our headers (incl. outer checksum
4139 * adjustment filled in by caller) and return result.
4140 */
4141 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4142}
4143
4144#endif /* __KERNEL__ */
4145#endif /* _LINUX_SKBUFF_H */
4146