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/kmemcheck.h>
19#include <linux/compiler.h>
20#include <linux/time.h>
21#include <linux/bug.h>
22#include <linux/cache.h>
23#include <linux/rbtree.h>
24#include <linux/socket.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 *
113 * skb->csum_level indicates the number of consecutive checksums found in
114 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
115 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
116 * and a device is able to verify the checksums for UDP (possibly zero),
117 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
118 * two. If the device were only able to verify the UDP checksum and not
119 * GRE, either because it doesn't support GRE checksum of because GRE
120 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
121 * not considered in this case).
122 *
123 * CHECKSUM_COMPLETE:
124 *
125 * This is the most generic way. The device supplied checksum of the _whole_
126 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
127 * hardware doesn't need to parse L3/L4 headers to implement this.
128 *
129 * Note: Even if device supports only some protocols, but is able to produce
130 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
131 *
132 * CHECKSUM_PARTIAL:
133 *
134 * A checksum is set up to be offloaded to a device as described in the
135 * output description for CHECKSUM_PARTIAL. This may occur on a packet
136 * received directly from another Linux OS, e.g., a virtualized Linux kernel
137 * on the same host, or it may be set in the input path in GRO or remote
138 * checksum offload. For the purposes of checksum verification, the checksum
139 * referred to by skb->csum_start + skb->csum_offset and any preceding
140 * checksums in the packet are considered verified. Any checksums in the
141 * packet that are after the checksum being offloaded are not considered to
142 * be verified.
143 *
144 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
145 * in the skb->ip_summed for a packet. Values are:
146 *
147 * CHECKSUM_PARTIAL:
148 *
149 * The driver is required to checksum the packet as seen by hard_start_xmit()
150 * from skb->csum_start up to the end, and to record/write the checksum at
151 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
152 * csum_start and csum_offset values are valid values given the length and
153 * offset of the packet, however they should not attempt to validate that the
154 * checksum refers to a legitimate transport layer checksum-- it is the
155 * purview of the stack to validate that csum_start and csum_offset are set
156 * correctly.
157 *
158 * When the stack requests checksum offload for a packet, the driver MUST
159 * ensure that the checksum is set correctly. A driver can either offload the
160 * checksum calculation to the device, or call skb_checksum_help (in the case
161 * that the device does not support offload for a particular checksum).
162 *
163 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
164 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
165 * checksum offload capability. If a device has limited checksum capabilities
166 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
167 * described above) a helper function can be called to resolve
168 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
169 * function takes a spec argument that describes the protocol layer that is
170 * supported for checksum offload and can be called for each packet. If a
171 * packet does not match the specification for offload, skb_checksum_help
172 * 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 ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
193 * accordingly. Note the there is no indication in the skbuff that the
194 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
195 * both IP checksum offload and SCTP CRC offload must verify which offload
196 * is configured for a packet presumably by inspecting packet headers.
197 *
198 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
199 * offloading the FCOE CRC in a packet. To perform this offload the stack
200 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
201 * accordingly. Note the there is no indication in the skbuff that the
202 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
203 * both IP checksum offload and FCOE CRC offload must verify which offload
204 * is configured for a packet presumably by inspecting packet headers.
205 *
206 * E. Checksumming on output with GSO.
207 *
208 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
209 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
210 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
211 * part of the GSO operation is implied. If a checksum is being offloaded
212 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
213 * are set to refer to the outermost checksum being offload (two offloaded
214 * checksums are possible with UDP encapsulation).
215 */
216
217/* Don't change this without changing skb_csum_unnecessary! */
218#define CHECKSUM_NONE 0
219#define CHECKSUM_UNNECESSARY 1
220#define CHECKSUM_COMPLETE 2
221#define CHECKSUM_PARTIAL 3
222
223/* Maximum value in skb->csum_level */
224#define SKB_MAX_CSUM_LEVEL 3
225
226#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
227#define SKB_WITH_OVERHEAD(X) \
228 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
229#define SKB_MAX_ORDER(X, ORDER) \
230 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
231#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
232#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
233
234/* return minimum truesize of one skb containing X bytes of data */
235#define SKB_TRUESIZE(X) ((X) + \
236 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
237 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
238
239struct net_device;
240struct scatterlist;
241struct pipe_inode_info;
242struct iov_iter;
243struct napi_struct;
244
245#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
246struct nf_conntrack {
247 atomic_t use;
248};
249#endif
250
251#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
252struct nf_bridge_info {
253 atomic_t use;
254 enum {
255 BRNF_PROTO_UNCHANGED,
256 BRNF_PROTO_8021Q,
257 BRNF_PROTO_PPPOE
258 } orig_proto:8;
259 u8 pkt_otherhost:1;
260 u8 in_prerouting:1;
261 u8 bridged_dnat:1;
262 __u16 frag_max_size;
263 struct net_device *physindev;
264
265 /* always valid & non-NULL from FORWARD on, for physdev match */
266 struct net_device *physoutdev;
267 union {
268 /* prerouting: detect dnat in orig/reply direction */
269 __be32 ipv4_daddr;
270 struct in6_addr ipv6_daddr;
271
272 /* after prerouting + nat detected: store original source
273 * mac since neigh resolution overwrites it, only used while
274 * skb is out in neigh layer.
275 */
276 char neigh_header[8];
277 };
278};
279#endif
280
281struct sk_buff_head {
282 /* These two members must be first. */
283 struct sk_buff *next;
284 struct sk_buff *prev;
285
286 __u32 qlen;
287 spinlock_t lock;
288};
289
290struct sk_buff;
291
292/* To allow 64K frame to be packed as single skb without frag_list we
293 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
294 * buffers which do not start on a page boundary.
295 *
296 * Since GRO uses frags we allocate at least 16 regardless of page
297 * size.
298 */
299#if (65536/PAGE_SIZE + 1) < 16
300#define MAX_SKB_FRAGS 16UL
301#else
302#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
303#endif
304extern int sysctl_max_skb_frags;
305
306/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
307 * segment using its current segmentation instead.
308 */
309#define GSO_BY_FRAGS 0xFFFF
310
311typedef struct skb_frag_struct skb_frag_t;
312
313struct skb_frag_struct {
314 struct {
315 struct page *p;
316 } page;
317#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
318 __u32 page_offset;
319 __u32 size;
320#else
321 __u16 page_offset;
322 __u16 size;
323#endif
324};
325
326static inline unsigned int skb_frag_size(const skb_frag_t *frag)
327{
328 return frag->size;
329}
330
331static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
332{
333 frag->size = size;
334}
335
336static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
337{
338 frag->size += delta;
339}
340
341static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
342{
343 frag->size -= delta;
344}
345
346#define HAVE_HW_TIME_STAMP
347
348/**
349 * struct skb_shared_hwtstamps - hardware time stamps
350 * @hwtstamp: hardware time stamp transformed into duration
351 * since arbitrary point in time
352 *
353 * Software time stamps generated by ktime_get_real() are stored in
354 * skb->tstamp.
355 *
356 * hwtstamps can only be compared against other hwtstamps from
357 * the same device.
358 *
359 * This structure is attached to packets as part of the
360 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
361 */
362struct skb_shared_hwtstamps {
363 ktime_t hwtstamp;
364};
365
366/* Definitions for tx_flags in struct skb_shared_info */
367enum {
368 /* generate hardware time stamp */
369 SKBTX_HW_TSTAMP = 1 << 0,
370
371 /* generate software time stamp when queueing packet to NIC */
372 SKBTX_SW_TSTAMP = 1 << 1,
373
374 /* device driver is going to provide hardware time stamp */
375 SKBTX_IN_PROGRESS = 1 << 2,
376
377 /* device driver supports TX zero-copy buffers */
378 SKBTX_DEV_ZEROCOPY = 1 << 3,
379
380 /* generate wifi status information (where possible) */
381 SKBTX_WIFI_STATUS = 1 << 4,
382
383 /* This indicates at least one fragment might be overwritten
384 * (as in vmsplice(), sendfile() ...)
385 * If we need to compute a TX checksum, we'll need to copy
386 * all frags to avoid possible bad checksum
387 */
388 SKBTX_SHARED_FRAG = 1 << 5,
389
390 /* generate software time stamp when entering packet scheduling */
391 SKBTX_SCHED_TSTAMP = 1 << 6,
392};
393
394#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
395 SKBTX_SCHED_TSTAMP)
396#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
397
398/*
399 * The callback notifies userspace to release buffers when skb DMA is done in
400 * lower device, the skb last reference should be 0 when calling this.
401 * The zerocopy_success argument is true if zero copy transmit occurred,
402 * false on data copy or out of memory error caused by data copy attempt.
403 * The ctx field is used to track device context.
404 * The desc field is used to track userspace buffer index.
405 */
406struct ubuf_info {
407 void (*callback)(struct ubuf_info *, bool zerocopy_success);
408 void *ctx;
409 unsigned long desc;
410};
411
412/* This data is invariant across clones and lives at
413 * the end of the header data, ie. at skb->end.
414 */
415struct skb_shared_info {
416 unsigned short _unused;
417 unsigned char nr_frags;
418 __u8 tx_flags;
419 unsigned short gso_size;
420 /* Warning: this field is not always filled in (UFO)! */
421 unsigned short gso_segs;
422 struct sk_buff *frag_list;
423 struct skb_shared_hwtstamps hwtstamps;
424 unsigned int gso_type;
425 u32 tskey;
426 __be32 ip6_frag_id;
427
428 /*
429 * Warning : all fields before dataref are cleared in __alloc_skb()
430 */
431 atomic_t dataref;
432
433 /* Intermediate layers must ensure that destructor_arg
434 * remains valid until skb destructor */
435 void * destructor_arg;
436
437 /* must be last field, see pskb_expand_head() */
438 skb_frag_t frags[MAX_SKB_FRAGS];
439};
440
441/* We divide dataref into two halves. The higher 16 bits hold references
442 * to the payload part of skb->data. The lower 16 bits hold references to
443 * the entire skb->data. A clone of a headerless skb holds the length of
444 * the header in skb->hdr_len.
445 *
446 * All users must obey the rule that the skb->data reference count must be
447 * greater than or equal to the payload reference count.
448 *
449 * Holding a reference to the payload part means that the user does not
450 * care about modifications to the header part of skb->data.
451 */
452#define SKB_DATAREF_SHIFT 16
453#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
454
455
456enum {
457 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
458 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
459 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
460};
461
462enum {
463 SKB_GSO_TCPV4 = 1 << 0,
464 SKB_GSO_UDP = 1 << 1,
465
466 /* This indicates the skb is from an untrusted source. */
467 SKB_GSO_DODGY = 1 << 2,
468
469 /* This indicates the tcp segment has CWR set. */
470 SKB_GSO_TCP_ECN = 1 << 3,
471
472 SKB_GSO_TCP_FIXEDID = 1 << 4,
473
474 SKB_GSO_TCPV6 = 1 << 5,
475
476 SKB_GSO_FCOE = 1 << 6,
477
478 SKB_GSO_GRE = 1 << 7,
479
480 SKB_GSO_GRE_CSUM = 1 << 8,
481
482 SKB_GSO_IPXIP4 = 1 << 9,
483
484 SKB_GSO_IPXIP6 = 1 << 10,
485
486 SKB_GSO_UDP_TUNNEL = 1 << 11,
487
488 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
489
490 SKB_GSO_PARTIAL = 1 << 13,
491
492 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
493
494 SKB_GSO_SCTP = 1 << 15,
495
496 SKB_GSO_ESP = 1 << 16,
497};
498
499#if BITS_PER_LONG > 32
500#define NET_SKBUFF_DATA_USES_OFFSET 1
501#endif
502
503#ifdef NET_SKBUFF_DATA_USES_OFFSET
504typedef unsigned int sk_buff_data_t;
505#else
506typedef unsigned char *sk_buff_data_t;
507#endif
508
509/**
510 * struct skb_mstamp - multi resolution time stamps
511 * @stamp_us: timestamp in us resolution
512 * @stamp_jiffies: timestamp in jiffies
513 */
514struct skb_mstamp {
515 union {
516 u64 v64;
517 struct {
518 u32 stamp_us;
519 u32 stamp_jiffies;
520 };
521 };
522};
523
524/**
525 * skb_mstamp_get - get current timestamp
526 * @cl: place to store timestamps
527 */
528static inline void skb_mstamp_get(struct skb_mstamp *cl)
529{
530 u64 val = local_clock();
531
532 do_div(val, NSEC_PER_USEC);
533 cl->stamp_us = (u32)val;
534 cl->stamp_jiffies = (u32)jiffies;
535}
536
537/**
538 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
539 * @t1: pointer to newest sample
540 * @t0: pointer to oldest sample
541 */
542static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
543 const struct skb_mstamp *t0)
544{
545 s32 delta_us = t1->stamp_us - t0->stamp_us;
546 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
547
548 /* If delta_us is negative, this might be because interval is too big,
549 * or local_clock() drift is too big : fallback using jiffies.
550 */
551 if (delta_us <= 0 ||
552 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
553
554 delta_us = jiffies_to_usecs(delta_jiffies);
555
556 return delta_us;
557}
558
559static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
560 const struct skb_mstamp *t0)
561{
562 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
563
564 if (!diff)
565 diff = t1->stamp_us - t0->stamp_us;
566 return diff > 0;
567}
568
569/**
570 * struct sk_buff - socket buffer
571 * @next: Next buffer in list
572 * @prev: Previous buffer in list
573 * @tstamp: Time we arrived/left
574 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
575 * @sk: Socket we are owned by
576 * @dev: Device we arrived on/are leaving by
577 * @cb: Control buffer. Free for use by every layer. Put private vars here
578 * @_skb_refdst: destination entry (with norefcount bit)
579 * @sp: the security path, used for xfrm
580 * @len: Length of actual data
581 * @data_len: Data length
582 * @mac_len: Length of link layer header
583 * @hdr_len: writable header length of cloned skb
584 * @csum: Checksum (must include start/offset pair)
585 * @csum_start: Offset from skb->head where checksumming should start
586 * @csum_offset: Offset from csum_start where checksum should be stored
587 * @priority: Packet queueing priority
588 * @ignore_df: allow local fragmentation
589 * @cloned: Head may be cloned (check refcnt to be sure)
590 * @ip_summed: Driver fed us an IP checksum
591 * @nohdr: Payload reference only, must not modify header
592 * @pkt_type: Packet class
593 * @fclone: skbuff clone status
594 * @ipvs_property: skbuff is owned by ipvs
595 * @tc_skip_classify: do not classify packet. set by IFB device
596 * @tc_at_ingress: used within tc_classify to distinguish in/egress
597 * @tc_redirected: packet was redirected by a tc action
598 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
599 * @peeked: this packet has been seen already, so stats have been
600 * done for it, don't do them again
601 * @nf_trace: netfilter packet trace flag
602 * @protocol: Packet protocol from driver
603 * @destructor: Destruct function
604 * @_nfct: Associated connection, if any (with nfctinfo bits)
605 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
606 * @skb_iif: ifindex of device we arrived on
607 * @tc_index: Traffic control index
608 * @hash: the packet hash
609 * @queue_mapping: Queue mapping for multiqueue devices
610 * @xmit_more: More SKBs are pending for this queue
611 * @ndisc_nodetype: router type (from link layer)
612 * @ooo_okay: allow the mapping of a socket to a queue to be changed
613 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
614 * ports.
615 * @sw_hash: indicates hash was computed in software stack
616 * @wifi_acked_valid: wifi_acked was set
617 * @wifi_acked: whether frame was acked on wifi or not
618 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
619 * @dst_pending_confirm: need to confirm neighbour
620 * @napi_id: id of the NAPI struct this skb came from
621 * @secmark: security marking
622 * @mark: Generic packet mark
623 * @vlan_proto: vlan encapsulation protocol
624 * @vlan_tci: vlan tag control information
625 * @inner_protocol: Protocol (encapsulation)
626 * @inner_transport_header: Inner transport layer header (encapsulation)
627 * @inner_network_header: Network layer header (encapsulation)
628 * @inner_mac_header: Link layer header (encapsulation)
629 * @transport_header: Transport layer header
630 * @network_header: Network layer header
631 * @mac_header: Link layer header
632 * @tail: Tail pointer
633 * @end: End pointer
634 * @head: Head of buffer
635 * @data: Data head pointer
636 * @truesize: Buffer size
637 * @users: User count - see {datagram,tcp}.c
638 */
639
640struct sk_buff {
641 union {
642 struct {
643 /* These two members must be first. */
644 struct sk_buff *next;
645 struct sk_buff *prev;
646
647 union {
648 ktime_t tstamp;
649 struct skb_mstamp skb_mstamp;
650 };
651 };
652 struct rb_node rbnode; /* used in netem & tcp stack */
653 };
654 struct sock *sk;
655
656 union {
657 struct net_device *dev;
658 /* Some protocols might use this space to store information,
659 * while device pointer would be NULL.
660 * UDP receive path is one user.
661 */
662 unsigned long dev_scratch;
663 };
664 /*
665 * This is the control buffer. It is free to use for every
666 * layer. Please put your private variables there. If you
667 * want to keep them across layers you have to do a skb_clone()
668 * first. This is owned by whoever has the skb queued ATM.
669 */
670 char cb[48] __aligned(8);
671
672 unsigned long _skb_refdst;
673 void (*destructor)(struct sk_buff *skb);
674#ifdef CONFIG_XFRM
675 struct sec_path *sp;
676#endif
677#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
678 unsigned long _nfct;
679#endif
680#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
681 struct nf_bridge_info *nf_bridge;
682#endif
683 unsigned int len,
684 data_len;
685 __u16 mac_len,
686 hdr_len;
687
688 /* Following fields are _not_ copied in __copy_skb_header()
689 * Note that queue_mapping is here mostly to fill a hole.
690 */
691 kmemcheck_bitfield_begin(flags1);
692 __u16 queue_mapping;
693
694/* if you move cloned around you also must adapt those constants */
695#ifdef __BIG_ENDIAN_BITFIELD
696#define CLONED_MASK (1 << 7)
697#else
698#define CLONED_MASK 1
699#endif
700#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
701
702 __u8 __cloned_offset[0];
703 __u8 cloned:1,
704 nohdr:1,
705 fclone:2,
706 peeked:1,
707 head_frag:1,
708 xmit_more:1,
709 __unused:1; /* one bit hole */
710 kmemcheck_bitfield_end(flags1);
711
712 /* fields enclosed in headers_start/headers_end are copied
713 * using a single memcpy() in __copy_skb_header()
714 */
715 /* private: */
716 __u32 headers_start[0];
717 /* public: */
718
719/* if you move pkt_type around you also must adapt those constants */
720#ifdef __BIG_ENDIAN_BITFIELD
721#define PKT_TYPE_MAX (7 << 5)
722#else
723#define PKT_TYPE_MAX 7
724#endif
725#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
726
727 __u8 __pkt_type_offset[0];
728 __u8 pkt_type:3;
729 __u8 pfmemalloc:1;
730 __u8 ignore_df:1;
731
732 __u8 nf_trace:1;
733 __u8 ip_summed:2;
734 __u8 ooo_okay:1;
735 __u8 l4_hash:1;
736 __u8 sw_hash:1;
737 __u8 wifi_acked_valid:1;
738 __u8 wifi_acked:1;
739
740 __u8 no_fcs:1;
741 /* Indicates the inner headers are valid in the skbuff. */
742 __u8 encapsulation:1;
743 __u8 encap_hdr_csum:1;
744 __u8 csum_valid:1;
745 __u8 csum_complete_sw:1;
746 __u8 csum_level:2;
747 __u8 csum_bad:1;
748
749 __u8 dst_pending_confirm:1;
750#ifdef CONFIG_IPV6_NDISC_NODETYPE
751 __u8 ndisc_nodetype:2;
752#endif
753 __u8 ipvs_property:1;
754 __u8 inner_protocol_type:1;
755 __u8 remcsum_offload:1;
756#ifdef CONFIG_NET_SWITCHDEV
757 __u8 offload_fwd_mark:1;
758#endif
759#ifdef CONFIG_NET_CLS_ACT
760 __u8 tc_skip_classify:1;
761 __u8 tc_at_ingress:1;
762 __u8 tc_redirected:1;
763 __u8 tc_from_ingress:1;
764#endif
765
766#ifdef CONFIG_NET_SCHED
767 __u16 tc_index; /* traffic control index */
768#endif
769
770 union {
771 __wsum csum;
772 struct {
773 __u16 csum_start;
774 __u16 csum_offset;
775 };
776 };
777 __u32 priority;
778 int skb_iif;
779 __u32 hash;
780 __be16 vlan_proto;
781 __u16 vlan_tci;
782#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
783 union {
784 unsigned int napi_id;
785 unsigned int sender_cpu;
786 };
787#endif
788#ifdef CONFIG_NETWORK_SECMARK
789 __u32 secmark;
790#endif
791
792 union {
793 __u32 mark;
794 __u32 reserved_tailroom;
795 };
796
797 union {
798 __be16 inner_protocol;
799 __u8 inner_ipproto;
800 };
801
802 __u16 inner_transport_header;
803 __u16 inner_network_header;
804 __u16 inner_mac_header;
805
806 __be16 protocol;
807 __u16 transport_header;
808 __u16 network_header;
809 __u16 mac_header;
810
811 /* private: */
812 __u32 headers_end[0];
813 /* public: */
814
815 /* These elements must be at the end, see alloc_skb() for details. */
816 sk_buff_data_t tail;
817 sk_buff_data_t end;
818 unsigned char *head,
819 *data;
820 unsigned int truesize;
821 atomic_t users;
822};
823
824#ifdef __KERNEL__
825/*
826 * Handling routines are only of interest to the kernel
827 */
828#include <linux/slab.h>
829
830
831#define SKB_ALLOC_FCLONE 0x01
832#define SKB_ALLOC_RX 0x02
833#define SKB_ALLOC_NAPI 0x04
834
835/* Returns true if the skb was allocated from PFMEMALLOC reserves */
836static inline bool skb_pfmemalloc(const struct sk_buff *skb)
837{
838 return unlikely(skb->pfmemalloc);
839}
840
841/*
842 * skb might have a dst pointer attached, refcounted or not.
843 * _skb_refdst low order bit is set if refcount was _not_ taken
844 */
845#define SKB_DST_NOREF 1UL
846#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
847
848#define SKB_NFCT_PTRMASK ~(7UL)
849/**
850 * skb_dst - returns skb dst_entry
851 * @skb: buffer
852 *
853 * Returns skb dst_entry, regardless of reference taken or not.
854 */
855static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
856{
857 /* If refdst was not refcounted, check we still are in a
858 * rcu_read_lock section
859 */
860 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
861 !rcu_read_lock_held() &&
862 !rcu_read_lock_bh_held());
863 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
864}
865
866/**
867 * skb_dst_set - sets skb dst
868 * @skb: buffer
869 * @dst: dst entry
870 *
871 * Sets skb dst, assuming a reference was taken on dst and should
872 * be released by skb_dst_drop()
873 */
874static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
875{
876 skb->_skb_refdst = (unsigned long)dst;
877}
878
879/**
880 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
881 * @skb: buffer
882 * @dst: dst entry
883 *
884 * Sets skb dst, assuming a reference was not taken on dst.
885 * If dst entry is cached, we do not take reference and dst_release
886 * will be avoided by refdst_drop. If dst entry is not cached, we take
887 * reference, so that last dst_release can destroy the dst immediately.
888 */
889static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
890{
891 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
892 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
893}
894
895/**
896 * skb_dst_is_noref - Test if skb dst isn't refcounted
897 * @skb: buffer
898 */
899static inline bool skb_dst_is_noref(const struct sk_buff *skb)
900{
901 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
902}
903
904static inline struct rtable *skb_rtable(const struct sk_buff *skb)
905{
906 return (struct rtable *)skb_dst(skb);
907}
908
909/* For mangling skb->pkt_type from user space side from applications
910 * such as nft, tc, etc, we only allow a conservative subset of
911 * possible pkt_types to be set.
912*/
913static inline bool skb_pkt_type_ok(u32 ptype)
914{
915 return ptype <= PACKET_OTHERHOST;
916}
917
918void kfree_skb(struct sk_buff *skb);
919void kfree_skb_list(struct sk_buff *segs);
920void skb_tx_error(struct sk_buff *skb);
921void consume_skb(struct sk_buff *skb);
922void __kfree_skb(struct sk_buff *skb);
923extern struct kmem_cache *skbuff_head_cache;
924
925void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
926bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
927 bool *fragstolen, int *delta_truesize);
928
929struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
930 int node);
931struct sk_buff *__build_skb(void *data, unsigned int frag_size);
932struct sk_buff *build_skb(void *data, unsigned int frag_size);
933static inline struct sk_buff *alloc_skb(unsigned int size,
934 gfp_t priority)
935{
936 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
937}
938
939struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
940 unsigned long data_len,
941 int max_page_order,
942 int *errcode,
943 gfp_t gfp_mask);
944
945/* Layout of fast clones : [skb1][skb2][fclone_ref] */
946struct sk_buff_fclones {
947 struct sk_buff skb1;
948
949 struct sk_buff skb2;
950
951 atomic_t fclone_ref;
952};
953
954/**
955 * skb_fclone_busy - check if fclone is busy
956 * @sk: socket
957 * @skb: buffer
958 *
959 * Returns true if skb is a fast clone, and its clone is not freed.
960 * Some drivers call skb_orphan() in their ndo_start_xmit(),
961 * so we also check that this didnt happen.
962 */
963static inline bool skb_fclone_busy(const struct sock *sk,
964 const struct sk_buff *skb)
965{
966 const struct sk_buff_fclones *fclones;
967
968 fclones = container_of(skb, struct sk_buff_fclones, skb1);
969
970 return skb->fclone == SKB_FCLONE_ORIG &&
971 atomic_read(&fclones->fclone_ref) > 1 &&
972 fclones->skb2.sk == sk;
973}
974
975static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
976 gfp_t priority)
977{
978 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
979}
980
981struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
982static inline struct sk_buff *alloc_skb_head(gfp_t priority)
983{
984 return __alloc_skb_head(priority, -1);
985}
986
987struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
988int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
989struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
990struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
991struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
992 gfp_t gfp_mask, bool fclone);
993static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
994 gfp_t gfp_mask)
995{
996 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
997}
998
999int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1000struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1001 unsigned int headroom);
1002struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1003 int newtailroom, gfp_t priority);
1004int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1005 int offset, int len);
1006int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
1007 int len);
1008int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1009int skb_pad(struct sk_buff *skb, int pad);
1010#define dev_kfree_skb(a) consume_skb(a)
1011
1012int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1013 int getfrag(void *from, char *to, int offset,
1014 int len, int odd, struct sk_buff *skb),
1015 void *from, int length);
1016
1017int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1018 int offset, size_t size);
1019
1020struct skb_seq_state {
1021 __u32 lower_offset;
1022 __u32 upper_offset;
1023 __u32 frag_idx;
1024 __u32 stepped_offset;
1025 struct sk_buff *root_skb;
1026 struct sk_buff *cur_skb;
1027 __u8 *frag_data;
1028};
1029
1030void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1031 unsigned int to, struct skb_seq_state *st);
1032unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1033 struct skb_seq_state *st);
1034void skb_abort_seq_read(struct skb_seq_state *st);
1035
1036unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1037 unsigned int to, struct ts_config *config);
1038
1039/*
1040 * Packet hash types specify the type of hash in skb_set_hash.
1041 *
1042 * Hash types refer to the protocol layer addresses which are used to
1043 * construct a packet's hash. The hashes are used to differentiate or identify
1044 * flows of the protocol layer for the hash type. Hash types are either
1045 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1046 *
1047 * Properties of hashes:
1048 *
1049 * 1) Two packets in different flows have different hash values
1050 * 2) Two packets in the same flow should have the same hash value
1051 *
1052 * A hash at a higher layer is considered to be more specific. A driver should
1053 * set the most specific hash possible.
1054 *
1055 * A driver cannot indicate a more specific hash than the layer at which a hash
1056 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1057 *
1058 * A driver may indicate a hash level which is less specific than the
1059 * actual layer the hash was computed on. For instance, a hash computed
1060 * at L4 may be considered an L3 hash. This should only be done if the
1061 * driver can't unambiguously determine that the HW computed the hash at
1062 * the higher layer. Note that the "should" in the second property above
1063 * permits this.
1064 */
1065enum pkt_hash_types {
1066 PKT_HASH_TYPE_NONE, /* Undefined type */
1067 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1068 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1069 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1070};
1071
1072static inline void skb_clear_hash(struct sk_buff *skb)
1073{
1074 skb->hash = 0;
1075 skb->sw_hash = 0;
1076 skb->l4_hash = 0;
1077}
1078
1079static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1080{
1081 if (!skb->l4_hash)
1082 skb_clear_hash(skb);
1083}
1084
1085static inline void
1086__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1087{
1088 skb->l4_hash = is_l4;
1089 skb->sw_hash = is_sw;
1090 skb->hash = hash;
1091}
1092
1093static inline void
1094skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1095{
1096 /* Used by drivers to set hash from HW */
1097 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1098}
1099
1100static inline void
1101__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1102{
1103 __skb_set_hash(skb, hash, true, is_l4);
1104}
1105
1106void __skb_get_hash(struct sk_buff *skb);
1107u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1108u32 skb_get_poff(const struct sk_buff *skb);
1109u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1110 const struct flow_keys *keys, int hlen);
1111__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1112 void *data, int hlen_proto);
1113
1114static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1115 int thoff, u8 ip_proto)
1116{
1117 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1118}
1119
1120void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1121 const struct flow_dissector_key *key,
1122 unsigned int key_count);
1123
1124bool __skb_flow_dissect(const struct sk_buff *skb,
1125 struct flow_dissector *flow_dissector,
1126 void *target_container,
1127 void *data, __be16 proto, int nhoff, int hlen,
1128 unsigned int flags);
1129
1130static inline bool skb_flow_dissect(const struct sk_buff *skb,
1131 struct flow_dissector *flow_dissector,
1132 void *target_container, unsigned int flags)
1133{
1134 return __skb_flow_dissect(skb, flow_dissector, target_container,
1135 NULL, 0, 0, 0, flags);
1136}
1137
1138static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1139 struct flow_keys *flow,
1140 unsigned int flags)
1141{
1142 memset(flow, 0, sizeof(*flow));
1143 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1144 NULL, 0, 0, 0, flags);
1145}
1146
1147static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1148 void *data, __be16 proto,
1149 int nhoff, int hlen,
1150 unsigned int flags)
1151{
1152 memset(flow, 0, sizeof(*flow));
1153 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1154 data, proto, nhoff, hlen, flags);
1155}
1156
1157static inline __u32 skb_get_hash(struct sk_buff *skb)
1158{
1159 if (!skb->l4_hash && !skb->sw_hash)
1160 __skb_get_hash(skb);
1161
1162 return skb->hash;
1163}
1164
1165__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1166
1167static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1168{
1169 if (!skb->l4_hash && !skb->sw_hash) {
1170 struct flow_keys keys;
1171 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1172
1173 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1174 }
1175
1176 return skb->hash;
1177}
1178
1179__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1180
1181static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1182{
1183 if (!skb->l4_hash && !skb->sw_hash) {
1184 struct flow_keys keys;
1185 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1186
1187 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1188 }
1189
1190 return skb->hash;
1191}
1192
1193__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1194
1195static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1196{
1197 return skb->hash;
1198}
1199
1200static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1201{
1202 to->hash = from->hash;
1203 to->sw_hash = from->sw_hash;
1204 to->l4_hash = from->l4_hash;
1205};
1206
1207#ifdef NET_SKBUFF_DATA_USES_OFFSET
1208static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1209{
1210 return skb->head + skb->end;
1211}
1212
1213static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1214{
1215 return skb->end;
1216}
1217#else
1218static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1219{
1220 return skb->end;
1221}
1222
1223static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1224{
1225 return skb->end - skb->head;
1226}
1227#endif
1228
1229/* Internal */
1230#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1231
1232static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1233{
1234 return &skb_shinfo(skb)->hwtstamps;
1235}
1236
1237/**
1238 * skb_queue_empty - check if a queue is empty
1239 * @list: queue head
1240 *
1241 * Returns true if the queue is empty, false otherwise.
1242 */
1243static inline int skb_queue_empty(const struct sk_buff_head *list)
1244{
1245 return list->next == (const struct sk_buff *) list;
1246}
1247
1248/**
1249 * skb_queue_is_last - check if skb is the last entry in the queue
1250 * @list: queue head
1251 * @skb: buffer
1252 *
1253 * Returns true if @skb is the last buffer on the list.
1254 */
1255static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1256 const struct sk_buff *skb)
1257{
1258 return skb->next == (const struct sk_buff *) list;
1259}
1260
1261/**
1262 * skb_queue_is_first - check if skb is the first entry in the queue
1263 * @list: queue head
1264 * @skb: buffer
1265 *
1266 * Returns true if @skb is the first buffer on the list.
1267 */
1268static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1269 const struct sk_buff *skb)
1270{
1271 return skb->prev == (const struct sk_buff *) list;
1272}
1273
1274/**
1275 * skb_queue_next - return the next packet in the queue
1276 * @list: queue head
1277 * @skb: current buffer
1278 *
1279 * Return the next packet in @list after @skb. It is only valid to
1280 * call this if skb_queue_is_last() evaluates to false.
1281 */
1282static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1283 const struct sk_buff *skb)
1284{
1285 /* This BUG_ON may seem severe, but if we just return then we
1286 * are going to dereference garbage.
1287 */
1288 BUG_ON(skb_queue_is_last(list, skb));
1289 return skb->next;
1290}
1291
1292/**
1293 * skb_queue_prev - return the prev packet in the queue
1294 * @list: queue head
1295 * @skb: current buffer
1296 *
1297 * Return the prev packet in @list before @skb. It is only valid to
1298 * call this if skb_queue_is_first() evaluates to false.
1299 */
1300static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1301 const struct sk_buff *skb)
1302{
1303 /* This BUG_ON may seem severe, but if we just return then we
1304 * are going to dereference garbage.
1305 */
1306 BUG_ON(skb_queue_is_first(list, skb));
1307 return skb->prev;
1308}
1309
1310/**
1311 * skb_get - reference buffer
1312 * @skb: buffer to reference
1313 *
1314 * Makes another reference to a socket buffer and returns a pointer
1315 * to the buffer.
1316 */
1317static inline struct sk_buff *skb_get(struct sk_buff *skb)
1318{
1319 atomic_inc(&skb->users);
1320 return skb;
1321}
1322
1323/*
1324 * If users == 1, we are the only owner and are can avoid redundant
1325 * atomic change.
1326 */
1327
1328/**
1329 * skb_cloned - is the buffer a clone
1330 * @skb: buffer to check
1331 *
1332 * Returns true if the buffer was generated with skb_clone() and is
1333 * one of multiple shared copies of the buffer. Cloned buffers are
1334 * shared data so must not be written to under normal circumstances.
1335 */
1336static inline int skb_cloned(const struct sk_buff *skb)
1337{
1338 return skb->cloned &&
1339 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1340}
1341
1342static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1343{
1344 might_sleep_if(gfpflags_allow_blocking(pri));
1345
1346 if (skb_cloned(skb))
1347 return pskb_expand_head(skb, 0, 0, pri);
1348
1349 return 0;
1350}
1351
1352/**
1353 * skb_header_cloned - is the header a clone
1354 * @skb: buffer to check
1355 *
1356 * Returns true if modifying the header part of the buffer requires
1357 * the data to be copied.
1358 */
1359static inline int skb_header_cloned(const struct sk_buff *skb)
1360{
1361 int dataref;
1362
1363 if (!skb->cloned)
1364 return 0;
1365
1366 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1367 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1368 return dataref != 1;
1369}
1370
1371static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1372{
1373 might_sleep_if(gfpflags_allow_blocking(pri));
1374
1375 if (skb_header_cloned(skb))
1376 return pskb_expand_head(skb, 0, 0, pri);
1377
1378 return 0;
1379}
1380
1381/**
1382 * skb_header_release - release reference to header
1383 * @skb: buffer to operate on
1384 *
1385 * Drop a reference to the header part of the buffer. This is done
1386 * by acquiring a payload reference. You must not read from the header
1387 * part of skb->data after this.
1388 * Note : Check if you can use __skb_header_release() instead.
1389 */
1390static inline void skb_header_release(struct sk_buff *skb)
1391{
1392 BUG_ON(skb->nohdr);
1393 skb->nohdr = 1;
1394 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1395}
1396
1397/**
1398 * __skb_header_release - release reference to header
1399 * @skb: buffer to operate on
1400 *
1401 * Variant of skb_header_release() assuming skb is private to caller.
1402 * We can avoid one atomic operation.
1403 */
1404static inline void __skb_header_release(struct sk_buff *skb)
1405{
1406 skb->nohdr = 1;
1407 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1408}
1409
1410
1411/**
1412 * skb_shared - is the buffer shared
1413 * @skb: buffer to check
1414 *
1415 * Returns true if more than one person has a reference to this
1416 * buffer.
1417 */
1418static inline int skb_shared(const struct sk_buff *skb)
1419{
1420 return atomic_read(&skb->users) != 1;
1421}
1422
1423/**
1424 * skb_share_check - check if buffer is shared and if so clone it
1425 * @skb: buffer to check
1426 * @pri: priority for memory allocation
1427 *
1428 * If the buffer is shared the buffer is cloned and the old copy
1429 * drops a reference. A new clone with a single reference is returned.
1430 * If the buffer is not shared the original buffer is returned. When
1431 * being called from interrupt status or with spinlocks held pri must
1432 * be GFP_ATOMIC.
1433 *
1434 * NULL is returned on a memory allocation failure.
1435 */
1436static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1437{
1438 might_sleep_if(gfpflags_allow_blocking(pri));
1439 if (skb_shared(skb)) {
1440 struct sk_buff *nskb = skb_clone(skb, pri);
1441
1442 if (likely(nskb))
1443 consume_skb(skb);
1444 else
1445 kfree_skb(skb);
1446 skb = nskb;
1447 }
1448 return skb;
1449}
1450
1451/*
1452 * Copy shared buffers into a new sk_buff. We effectively do COW on
1453 * packets to handle cases where we have a local reader and forward
1454 * and a couple of other messy ones. The normal one is tcpdumping
1455 * a packet thats being forwarded.
1456 */
1457
1458/**
1459 * skb_unshare - make a copy of a shared buffer
1460 * @skb: buffer to check
1461 * @pri: priority for memory allocation
1462 *
1463 * If the socket buffer is a clone then this function creates a new
1464 * copy of the data, drops a reference count on the old copy and returns
1465 * the new copy with the reference count at 1. If the buffer is not a clone
1466 * the original buffer is returned. When called with a spinlock held or
1467 * from interrupt state @pri must be %GFP_ATOMIC
1468 *
1469 * %NULL is returned on a memory allocation failure.
1470 */
1471static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1472 gfp_t pri)
1473{
1474 might_sleep_if(gfpflags_allow_blocking(pri));
1475 if (skb_cloned(skb)) {
1476 struct sk_buff *nskb = skb_copy(skb, pri);
1477
1478 /* Free our shared copy */
1479 if (likely(nskb))
1480 consume_skb(skb);
1481 else
1482 kfree_skb(skb);
1483 skb = nskb;
1484 }
1485 return skb;
1486}
1487
1488/**
1489 * skb_peek - peek at the head of an &sk_buff_head
1490 * @list_: list to peek at
1491 *
1492 * Peek an &sk_buff. Unlike most other operations you _MUST_
1493 * be careful with this one. A peek leaves the buffer on the
1494 * list and someone else may run off with it. You must hold
1495 * the appropriate locks or have a private queue to do this.
1496 *
1497 * Returns %NULL for an empty list or a pointer to the head element.
1498 * The reference count is not incremented and the reference is therefore
1499 * volatile. Use with caution.
1500 */
1501static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1502{
1503 struct sk_buff *skb = list_->next;
1504
1505 if (skb == (struct sk_buff *)list_)
1506 skb = NULL;
1507 return skb;
1508}
1509
1510/**
1511 * skb_peek_next - peek skb following the given one from a queue
1512 * @skb: skb to start from
1513 * @list_: list to peek at
1514 *
1515 * Returns %NULL when the end of the list is met or a pointer to the
1516 * next element. The reference count is not incremented and the
1517 * reference is therefore volatile. Use with caution.
1518 */
1519static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1520 const struct sk_buff_head *list_)
1521{
1522 struct sk_buff *next = skb->next;
1523
1524 if (next == (struct sk_buff *)list_)
1525 next = NULL;
1526 return next;
1527}
1528
1529/**
1530 * skb_peek_tail - peek at the tail of an &sk_buff_head
1531 * @list_: list to peek at
1532 *
1533 * Peek an &sk_buff. Unlike most other operations you _MUST_
1534 * be careful with this one. A peek leaves the buffer on the
1535 * list and someone else may run off with it. You must hold
1536 * the appropriate locks or have a private queue to do this.
1537 *
1538 * Returns %NULL for an empty list or a pointer to the tail element.
1539 * The reference count is not incremented and the reference is therefore
1540 * volatile. Use with caution.
1541 */
1542static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1543{
1544 struct sk_buff *skb = list_->prev;
1545
1546 if (skb == (struct sk_buff *)list_)
1547 skb = NULL;
1548 return skb;
1549
1550}
1551
1552/**
1553 * skb_queue_len - get queue length
1554 * @list_: list to measure
1555 *
1556 * Return the length of an &sk_buff queue.
1557 */
1558static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1559{
1560 return list_->qlen;
1561}
1562
1563/**
1564 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1565 * @list: queue to initialize
1566 *
1567 * This initializes only the list and queue length aspects of
1568 * an sk_buff_head object. This allows to initialize the list
1569 * aspects of an sk_buff_head without reinitializing things like
1570 * the spinlock. It can also be used for on-stack sk_buff_head
1571 * objects where the spinlock is known to not be used.
1572 */
1573static inline void __skb_queue_head_init(struct sk_buff_head *list)
1574{
1575 list->prev = list->next = (struct sk_buff *)list;
1576 list->qlen = 0;
1577}
1578
1579/*
1580 * This function creates a split out lock class for each invocation;
1581 * this is needed for now since a whole lot of users of the skb-queue
1582 * infrastructure in drivers have different locking usage (in hardirq)
1583 * than the networking core (in softirq only). In the long run either the
1584 * network layer or drivers should need annotation to consolidate the
1585 * main types of usage into 3 classes.
1586 */
1587static inline void skb_queue_head_init(struct sk_buff_head *list)
1588{
1589 spin_lock_init(&list->lock);
1590 __skb_queue_head_init(list);
1591}
1592
1593static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1594 struct lock_class_key *class)
1595{
1596 skb_queue_head_init(list);
1597 lockdep_set_class(&list->lock, class);
1598}
1599
1600/*
1601 * Insert an sk_buff on a list.
1602 *
1603 * The "__skb_xxxx()" functions are the non-atomic ones that
1604 * can only be called with interrupts disabled.
1605 */
1606void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1607 struct sk_buff_head *list);
1608static inline void __skb_insert(struct sk_buff *newsk,
1609 struct sk_buff *prev, struct sk_buff *next,
1610 struct sk_buff_head *list)
1611{
1612 newsk->next = next;
1613 newsk->prev = prev;
1614 next->prev = prev->next = newsk;
1615 list->qlen++;
1616}
1617
1618static inline void __skb_queue_splice(const struct sk_buff_head *list,
1619 struct sk_buff *prev,
1620 struct sk_buff *next)
1621{
1622 struct sk_buff *first = list->next;
1623 struct sk_buff *last = list->prev;
1624
1625 first->prev = prev;
1626 prev->next = first;
1627
1628 last->next = next;
1629 next->prev = last;
1630}
1631
1632/**
1633 * skb_queue_splice - join two skb lists, this is designed for stacks
1634 * @list: the new list to add
1635 * @head: the place to add it in the first list
1636 */
1637static inline void skb_queue_splice(const struct sk_buff_head *list,
1638 struct sk_buff_head *head)
1639{
1640 if (!skb_queue_empty(list)) {
1641 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1642 head->qlen += list->qlen;
1643 }
1644}
1645
1646/**
1647 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1648 * @list: the new list to add
1649 * @head: the place to add it in the first list
1650 *
1651 * The list at @list is reinitialised
1652 */
1653static inline void skb_queue_splice_init(struct sk_buff_head *list,
1654 struct sk_buff_head *head)
1655{
1656 if (!skb_queue_empty(list)) {
1657 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1658 head->qlen += list->qlen;
1659 __skb_queue_head_init(list);
1660 }
1661}
1662
1663/**
1664 * skb_queue_splice_tail - join two skb lists, each list being a queue
1665 * @list: the new list to add
1666 * @head: the place to add it in the first list
1667 */
1668static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1669 struct sk_buff_head *head)
1670{
1671 if (!skb_queue_empty(list)) {
1672 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1673 head->qlen += list->qlen;
1674 }
1675}
1676
1677/**
1678 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1679 * @list: the new list to add
1680 * @head: the place to add it in the first list
1681 *
1682 * Each of the lists is a queue.
1683 * The list at @list is reinitialised
1684 */
1685static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1686 struct sk_buff_head *head)
1687{
1688 if (!skb_queue_empty(list)) {
1689 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1690 head->qlen += list->qlen;
1691 __skb_queue_head_init(list);
1692 }
1693}
1694
1695/**
1696 * __skb_queue_after - queue a buffer at the list head
1697 * @list: list to use
1698 * @prev: place after this buffer
1699 * @newsk: buffer to queue
1700 *
1701 * Queue a buffer int the middle of a list. This function takes no locks
1702 * and you must therefore hold required locks before calling it.
1703 *
1704 * A buffer cannot be placed on two lists at the same time.
1705 */
1706static inline void __skb_queue_after(struct sk_buff_head *list,
1707 struct sk_buff *prev,
1708 struct sk_buff *newsk)
1709{
1710 __skb_insert(newsk, prev, prev->next, list);
1711}
1712
1713void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1714 struct sk_buff_head *list);
1715
1716static inline void __skb_queue_before(struct sk_buff_head *list,
1717 struct sk_buff *next,
1718 struct sk_buff *newsk)
1719{
1720 __skb_insert(newsk, next->prev, next, list);
1721}
1722
1723/**
1724 * __skb_queue_head - queue a buffer at the list head
1725 * @list: list to use
1726 * @newsk: buffer to queue
1727 *
1728 * Queue a buffer at the start of a list. This function takes no locks
1729 * and you must therefore hold required locks before calling it.
1730 *
1731 * A buffer cannot be placed on two lists at the same time.
1732 */
1733void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1734static inline void __skb_queue_head(struct sk_buff_head *list,
1735 struct sk_buff *newsk)
1736{
1737 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1738}
1739
1740/**
1741 * __skb_queue_tail - queue a buffer at the list tail
1742 * @list: list to use
1743 * @newsk: buffer to queue
1744 *
1745 * Queue a buffer at the end of a list. This function takes no locks
1746 * and you must therefore hold required locks before calling it.
1747 *
1748 * A buffer cannot be placed on two lists at the same time.
1749 */
1750void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1751static inline void __skb_queue_tail(struct sk_buff_head *list,
1752 struct sk_buff *newsk)
1753{
1754 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1755}
1756
1757/*
1758 * remove sk_buff from list. _Must_ be called atomically, and with
1759 * the list known..
1760 */
1761void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1762static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1763{
1764 struct sk_buff *next, *prev;
1765
1766 list->qlen--;
1767 next = skb->next;
1768 prev = skb->prev;
1769 skb->next = skb->prev = NULL;
1770 next->prev = prev;
1771 prev->next = next;
1772}
1773
1774/**
1775 * __skb_dequeue - remove from the head of the queue
1776 * @list: list to dequeue from
1777 *
1778 * Remove the head of the list. This function does not take any locks
1779 * so must be used with appropriate locks held only. The head item is
1780 * returned or %NULL if the list is empty.
1781 */
1782struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1783static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1784{
1785 struct sk_buff *skb = skb_peek(list);
1786 if (skb)
1787 __skb_unlink(skb, list);
1788 return skb;
1789}
1790
1791/**
1792 * __skb_dequeue_tail - remove from the tail of the queue
1793 * @list: list to dequeue from
1794 *
1795 * Remove the tail of the list. This function does not take any locks
1796 * so must be used with appropriate locks held only. The tail item is
1797 * returned or %NULL if the list is empty.
1798 */
1799struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1800static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1801{
1802 struct sk_buff *skb = skb_peek_tail(list);
1803 if (skb)
1804 __skb_unlink(skb, list);
1805 return skb;
1806}
1807
1808
1809static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1810{
1811 return skb->data_len;
1812}
1813
1814static inline unsigned int skb_headlen(const struct sk_buff *skb)
1815{
1816 return skb->len - skb->data_len;
1817}
1818
1819static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1820{
1821 unsigned int i, len = 0;
1822
1823 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1824 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1825 return len + skb_headlen(skb);
1826}
1827
1828/**
1829 * __skb_fill_page_desc - initialise a paged fragment in an skb
1830 * @skb: buffer containing fragment to be initialised
1831 * @i: paged fragment index to initialise
1832 * @page: the page to use for this fragment
1833 * @off: the offset to the data with @page
1834 * @size: the length of the data
1835 *
1836 * Initialises the @i'th fragment of @skb to point to &size bytes at
1837 * offset @off within @page.
1838 *
1839 * Does not take any additional reference on the fragment.
1840 */
1841static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1842 struct page *page, int off, int size)
1843{
1844 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1845
1846 /*
1847 * Propagate page pfmemalloc to the skb if we can. The problem is
1848 * that not all callers have unique ownership of the page but rely
1849 * on page_is_pfmemalloc doing the right thing(tm).
1850 */
1851 frag->page.p = page;
1852 frag->page_offset = off;
1853 skb_frag_size_set(frag, size);
1854
1855 page = compound_head(page);
1856 if (page_is_pfmemalloc(page))
1857 skb->pfmemalloc = true;
1858}
1859
1860/**
1861 * skb_fill_page_desc - initialise a paged fragment in an skb
1862 * @skb: buffer containing fragment to be initialised
1863 * @i: paged fragment index to initialise
1864 * @page: the page to use for this fragment
1865 * @off: the offset to the data with @page
1866 * @size: the length of the data
1867 *
1868 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1869 * @skb to point to @size bytes at offset @off within @page. In
1870 * addition updates @skb such that @i is the last fragment.
1871 *
1872 * Does not take any additional reference on the fragment.
1873 */
1874static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1875 struct page *page, int off, int size)
1876{
1877 __skb_fill_page_desc(skb, i, page, off, size);
1878 skb_shinfo(skb)->nr_frags = i + 1;
1879}
1880
1881void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1882 int size, unsigned int truesize);
1883
1884void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1885 unsigned int truesize);
1886
1887#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1888#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1889#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1890
1891#ifdef NET_SKBUFF_DATA_USES_OFFSET
1892static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1893{
1894 return skb->head + skb->tail;
1895}
1896
1897static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1898{
1899 skb->tail = skb->data - skb->head;
1900}
1901
1902static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1903{
1904 skb_reset_tail_pointer(skb);
1905 skb->tail += offset;
1906}
1907
1908#else /* NET_SKBUFF_DATA_USES_OFFSET */
1909static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1910{
1911 return skb->tail;
1912}
1913
1914static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1915{
1916 skb->tail = skb->data;
1917}
1918
1919static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1920{
1921 skb->tail = skb->data + offset;
1922}
1923
1924#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1925
1926/*
1927 * Add data to an sk_buff
1928 */
1929unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1930unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1931static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1932{
1933 unsigned char *tmp = skb_tail_pointer(skb);
1934 SKB_LINEAR_ASSERT(skb);
1935 skb->tail += len;
1936 skb->len += len;
1937 return tmp;
1938}
1939
1940unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1941static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1942{
1943 skb->data -= len;
1944 skb->len += len;
1945 return skb->data;
1946}
1947
1948unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1949static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1950{
1951 skb->len -= len;
1952 BUG_ON(skb->len < skb->data_len);
1953 return skb->data += len;
1954}
1955
1956static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1957{
1958 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1959}
1960
1961unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1962
1963static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1964{
1965 if (len > skb_headlen(skb) &&
1966 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1967 return NULL;
1968 skb->len -= len;
1969 return skb->data += len;
1970}
1971
1972static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1973{
1974 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1975}
1976
1977static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1978{
1979 if (likely(len <= skb_headlen(skb)))
1980 return 1;
1981 if (unlikely(len > skb->len))
1982 return 0;
1983 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1984}
1985
1986void skb_condense(struct sk_buff *skb);
1987
1988/**
1989 * skb_headroom - bytes at buffer head
1990 * @skb: buffer to check
1991 *
1992 * Return the number of bytes of free space at the head of an &sk_buff.
1993 */
1994static inline unsigned int skb_headroom(const struct sk_buff *skb)
1995{
1996 return skb->data - skb->head;
1997}
1998
1999/**
2000 * skb_tailroom - bytes at buffer end
2001 * @skb: buffer to check
2002 *
2003 * Return the number of bytes of free space at the tail of an sk_buff
2004 */
2005static inline int skb_tailroom(const struct sk_buff *skb)
2006{
2007 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2008}
2009
2010/**
2011 * skb_availroom - bytes at buffer end
2012 * @skb: buffer to check
2013 *
2014 * Return the number of bytes of free space at the tail of an sk_buff
2015 * allocated by sk_stream_alloc()
2016 */
2017static inline int skb_availroom(const struct sk_buff *skb)
2018{
2019 if (skb_is_nonlinear(skb))
2020 return 0;
2021
2022 return skb->end - skb->tail - skb->reserved_tailroom;
2023}
2024
2025/**
2026 * skb_reserve - adjust headroom
2027 * @skb: buffer to alter
2028 * @len: bytes to move
2029 *
2030 * Increase the headroom of an empty &sk_buff by reducing the tail
2031 * room. This is only allowed for an empty buffer.
2032 */
2033static inline void skb_reserve(struct sk_buff *skb, int len)
2034{
2035 skb->data += len;
2036 skb->tail += len;
2037}
2038
2039/**
2040 * skb_tailroom_reserve - adjust reserved_tailroom
2041 * @skb: buffer to alter
2042 * @mtu: maximum amount of headlen permitted
2043 * @needed_tailroom: minimum amount of reserved_tailroom
2044 *
2045 * Set reserved_tailroom so that headlen can be as large as possible but
2046 * not larger than mtu and tailroom cannot be smaller than
2047 * needed_tailroom.
2048 * The required headroom should already have been reserved before using
2049 * this function.
2050 */
2051static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2052 unsigned int needed_tailroom)
2053{
2054 SKB_LINEAR_ASSERT(skb);
2055 if (mtu < skb_tailroom(skb) - needed_tailroom)
2056 /* use at most mtu */
2057 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2058 else
2059 /* use up to all available space */
2060 skb->reserved_tailroom = needed_tailroom;
2061}
2062
2063#define ENCAP_TYPE_ETHER 0
2064#define ENCAP_TYPE_IPPROTO 1
2065
2066static inline void skb_set_inner_protocol(struct sk_buff *skb,
2067 __be16 protocol)
2068{
2069 skb->inner_protocol = protocol;
2070 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2071}
2072
2073static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2074 __u8 ipproto)
2075{
2076 skb->inner_ipproto = ipproto;
2077 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2078}
2079
2080static inline void skb_reset_inner_headers(struct sk_buff *skb)
2081{
2082 skb->inner_mac_header = skb->mac_header;
2083 skb->inner_network_header = skb->network_header;
2084 skb->inner_transport_header = skb->transport_header;
2085}
2086
2087static inline void skb_reset_mac_len(struct sk_buff *skb)
2088{
2089 skb->mac_len = skb->network_header - skb->mac_header;
2090}
2091
2092static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2093 *skb)
2094{
2095 return skb->head + skb->inner_transport_header;
2096}
2097
2098static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2099{
2100 return skb_inner_transport_header(skb) - skb->data;
2101}
2102
2103static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2104{
2105 skb->inner_transport_header = skb->data - skb->head;
2106}
2107
2108static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2109 const int offset)
2110{
2111 skb_reset_inner_transport_header(skb);
2112 skb->inner_transport_header += offset;
2113}
2114
2115static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2116{
2117 return skb->head + skb->inner_network_header;
2118}
2119
2120static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2121{
2122 skb->inner_network_header = skb->data - skb->head;
2123}
2124
2125static inline void skb_set_inner_network_header(struct sk_buff *skb,
2126 const int offset)
2127{
2128 skb_reset_inner_network_header(skb);
2129 skb->inner_network_header += offset;
2130}
2131
2132static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2133{
2134 return skb->head + skb->inner_mac_header;
2135}
2136
2137static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2138{
2139 skb->inner_mac_header = skb->data - skb->head;
2140}
2141
2142static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2143 const int offset)
2144{
2145 skb_reset_inner_mac_header(skb);
2146 skb->inner_mac_header += offset;
2147}
2148static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2149{
2150 return skb->transport_header != (typeof(skb->transport_header))~0U;
2151}
2152
2153static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2154{
2155 return skb->head + skb->transport_header;
2156}
2157
2158static inline void skb_reset_transport_header(struct sk_buff *skb)
2159{
2160 skb->transport_header = skb->data - skb->head;
2161}
2162
2163static inline void skb_set_transport_header(struct sk_buff *skb,
2164 const int offset)
2165{
2166 skb_reset_transport_header(skb);
2167 skb->transport_header += offset;
2168}
2169
2170static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2171{
2172 return skb->head + skb->network_header;
2173}
2174
2175static inline void skb_reset_network_header(struct sk_buff *skb)
2176{
2177 skb->network_header = skb->data - skb->head;
2178}
2179
2180static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2181{
2182 skb_reset_network_header(skb);
2183 skb->network_header += offset;
2184}
2185
2186static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2187{
2188 return skb->head + skb->mac_header;
2189}
2190
2191static inline int skb_mac_offset(const struct sk_buff *skb)
2192{
2193 return skb_mac_header(skb) - skb->data;
2194}
2195
2196static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2197{
2198 return skb->mac_header != (typeof(skb->mac_header))~0U;
2199}
2200
2201static inline void skb_reset_mac_header(struct sk_buff *skb)
2202{
2203 skb->mac_header = skb->data - skb->head;
2204}
2205
2206static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2207{
2208 skb_reset_mac_header(skb);
2209 skb->mac_header += offset;
2210}
2211
2212static inline void skb_pop_mac_header(struct sk_buff *skb)
2213{
2214 skb->mac_header = skb->network_header;
2215}
2216
2217static inline void skb_probe_transport_header(struct sk_buff *skb,
2218 const int offset_hint)
2219{
2220 struct flow_keys keys;
2221
2222 if (skb_transport_header_was_set(skb))
2223 return;
2224 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2225 skb_set_transport_header(skb, keys.control.thoff);
2226 else
2227 skb_set_transport_header(skb, offset_hint);
2228}
2229
2230static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2231{
2232 if (skb_mac_header_was_set(skb)) {
2233 const unsigned char *old_mac = skb_mac_header(skb);
2234
2235 skb_set_mac_header(skb, -skb->mac_len);
2236 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2237 }
2238}
2239
2240static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2241{
2242 return skb->csum_start - skb_headroom(skb);
2243}
2244
2245static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2246{
2247 return skb->head + skb->csum_start;
2248}
2249
2250static inline int skb_transport_offset(const struct sk_buff *skb)
2251{
2252 return skb_transport_header(skb) - skb->data;
2253}
2254
2255static inline u32 skb_network_header_len(const struct sk_buff *skb)
2256{
2257 return skb->transport_header - skb->network_header;
2258}
2259
2260static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2261{
2262 return skb->inner_transport_header - skb->inner_network_header;
2263}
2264
2265static inline int skb_network_offset(const struct sk_buff *skb)
2266{
2267 return skb_network_header(skb) - skb->data;
2268}
2269
2270static inline int skb_inner_network_offset(const struct sk_buff *skb)
2271{
2272 return skb_inner_network_header(skb) - skb->data;
2273}
2274
2275static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2276{
2277 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2278}
2279
2280/*
2281 * CPUs often take a performance hit when accessing unaligned memory
2282 * locations. The actual performance hit varies, it can be small if the
2283 * hardware handles it or large if we have to take an exception and fix it
2284 * in software.
2285 *
2286 * Since an ethernet header is 14 bytes network drivers often end up with
2287 * the IP header at an unaligned offset. The IP header can be aligned by
2288 * shifting the start of the packet by 2 bytes. Drivers should do this
2289 * with:
2290 *
2291 * skb_reserve(skb, NET_IP_ALIGN);
2292 *
2293 * The downside to this alignment of the IP header is that the DMA is now
2294 * unaligned. On some architectures the cost of an unaligned DMA is high
2295 * and this cost outweighs the gains made by aligning the IP header.
2296 *
2297 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2298 * to be overridden.
2299 */
2300#ifndef NET_IP_ALIGN
2301#define NET_IP_ALIGN 2
2302#endif
2303
2304/*
2305 * The networking layer reserves some headroom in skb data (via
2306 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2307 * the header has to grow. In the default case, if the header has to grow
2308 * 32 bytes or less we avoid the reallocation.
2309 *
2310 * Unfortunately this headroom changes the DMA alignment of the resulting
2311 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2312 * on some architectures. An architecture can override this value,
2313 * perhaps setting it to a cacheline in size (since that will maintain
2314 * cacheline alignment of the DMA). It must be a power of 2.
2315 *
2316 * Various parts of the networking layer expect at least 32 bytes of
2317 * headroom, you should not reduce this.
2318 *
2319 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2320 * to reduce average number of cache lines per packet.
2321 * get_rps_cpus() for example only access one 64 bytes aligned block :
2322 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2323 */
2324#ifndef NET_SKB_PAD
2325#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2326#endif
2327
2328int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2329
2330static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2331{
2332 if (unlikely(skb_is_nonlinear(skb))) {
2333 WARN_ON(1);
2334 return;
2335 }
2336 skb->len = len;
2337 skb_set_tail_pointer(skb, len);
2338}
2339
2340static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2341{
2342 __skb_set_length(skb, len);
2343}
2344
2345void skb_trim(struct sk_buff *skb, unsigned int len);
2346
2347static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2348{
2349 if (skb->data_len)
2350 return ___pskb_trim(skb, len);
2351 __skb_trim(skb, len);
2352 return 0;
2353}
2354
2355static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2356{
2357 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2358}
2359
2360/**
2361 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2362 * @skb: buffer to alter
2363 * @len: new length
2364 *
2365 * This is identical to pskb_trim except that the caller knows that
2366 * the skb is not cloned so we should never get an error due to out-
2367 * of-memory.
2368 */
2369static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2370{
2371 int err = pskb_trim(skb, len);
2372 BUG_ON(err);
2373}
2374
2375static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2376{
2377 unsigned int diff = len - skb->len;
2378
2379 if (skb_tailroom(skb) < diff) {
2380 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2381 GFP_ATOMIC);
2382 if (ret)
2383 return ret;
2384 }
2385 __skb_set_length(skb, len);
2386 return 0;
2387}
2388
2389/**
2390 * skb_orphan - orphan a buffer
2391 * @skb: buffer to orphan
2392 *
2393 * If a buffer currently has an owner then we call the owner's
2394 * destructor function and make the @skb unowned. The buffer continues
2395 * to exist but is no longer charged to its former owner.
2396 */
2397static inline void skb_orphan(struct sk_buff *skb)
2398{
2399 if (skb->destructor) {
2400 skb->destructor(skb);
2401 skb->destructor = NULL;
2402 skb->sk = NULL;
2403 } else {
2404 BUG_ON(skb->sk);
2405 }
2406}
2407
2408/**
2409 * skb_orphan_frags - orphan the frags contained in a buffer
2410 * @skb: buffer to orphan frags from
2411 * @gfp_mask: allocation mask for replacement pages
2412 *
2413 * For each frag in the SKB which needs a destructor (i.e. has an
2414 * owner) create a copy of that frag and release the original
2415 * page by calling the destructor.
2416 */
2417static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2418{
2419 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2420 return 0;
2421 return skb_copy_ubufs(skb, gfp_mask);
2422}
2423
2424/**
2425 * __skb_queue_purge - empty a list
2426 * @list: list to empty
2427 *
2428 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2429 * the list and one reference dropped. This function does not take the
2430 * list lock and the caller must hold the relevant locks to use it.
2431 */
2432void skb_queue_purge(struct sk_buff_head *list);
2433static inline void __skb_queue_purge(struct sk_buff_head *list)
2434{
2435 struct sk_buff *skb;
2436 while ((skb = __skb_dequeue(list)) != NULL)
2437 kfree_skb(skb);
2438}
2439
2440void skb_rbtree_purge(struct rb_root *root);
2441
2442void *netdev_alloc_frag(unsigned int fragsz);
2443
2444struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2445 gfp_t gfp_mask);
2446
2447/**
2448 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2449 * @dev: network device to receive on
2450 * @length: length to allocate
2451 *
2452 * Allocate a new &sk_buff and assign it a usage count of one. The
2453 * buffer has unspecified headroom built in. Users should allocate
2454 * the headroom they think they need without accounting for the
2455 * built in space. The built in space is used for optimisations.
2456 *
2457 * %NULL is returned if there is no free memory. Although this function
2458 * allocates memory it can be called from an interrupt.
2459 */
2460static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2461 unsigned int length)
2462{
2463 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2464}
2465
2466/* legacy helper around __netdev_alloc_skb() */
2467static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2468 gfp_t gfp_mask)
2469{
2470 return __netdev_alloc_skb(NULL, length, gfp_mask);
2471}
2472
2473/* legacy helper around netdev_alloc_skb() */
2474static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2475{
2476 return netdev_alloc_skb(NULL, length);
2477}
2478
2479
2480static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2481 unsigned int length, gfp_t gfp)
2482{
2483 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2484
2485 if (NET_IP_ALIGN && skb)
2486 skb_reserve(skb, NET_IP_ALIGN);
2487 return skb;
2488}
2489
2490static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2491 unsigned int length)
2492{
2493 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2494}
2495
2496static inline void skb_free_frag(void *addr)
2497{
2498 page_frag_free(addr);
2499}
2500
2501void *napi_alloc_frag(unsigned int fragsz);
2502struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2503 unsigned int length, gfp_t gfp_mask);
2504static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2505 unsigned int length)
2506{
2507 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2508}
2509void napi_consume_skb(struct sk_buff *skb, int budget);
2510
2511void __kfree_skb_flush(void);
2512void __kfree_skb_defer(struct sk_buff *skb);
2513
2514/**
2515 * __dev_alloc_pages - allocate page for network Rx
2516 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2517 * @order: size of the allocation
2518 *
2519 * Allocate a new page.
2520 *
2521 * %NULL is returned if there is no free memory.
2522*/
2523static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2524 unsigned int order)
2525{
2526 /* This piece of code contains several assumptions.
2527 * 1. This is for device Rx, therefor a cold page is preferred.
2528 * 2. The expectation is the user wants a compound page.
2529 * 3. If requesting a order 0 page it will not be compound
2530 * due to the check to see if order has a value in prep_new_page
2531 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2532 * code in gfp_to_alloc_flags that should be enforcing this.
2533 */
2534 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2535
2536 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2537}
2538
2539static inline struct page *dev_alloc_pages(unsigned int order)
2540{
2541 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2542}
2543
2544/**
2545 * __dev_alloc_page - allocate a page for network Rx
2546 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2547 *
2548 * Allocate a new page.
2549 *
2550 * %NULL is returned if there is no free memory.
2551 */
2552static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2553{
2554 return __dev_alloc_pages(gfp_mask, 0);
2555}
2556
2557static inline struct page *dev_alloc_page(void)
2558{
2559 return dev_alloc_pages(0);
2560}
2561
2562/**
2563 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2564 * @page: The page that was allocated from skb_alloc_page
2565 * @skb: The skb that may need pfmemalloc set
2566 */
2567static inline void skb_propagate_pfmemalloc(struct page *page,
2568 struct sk_buff *skb)
2569{
2570 if (page_is_pfmemalloc(page))
2571 skb->pfmemalloc = true;
2572}
2573
2574/**
2575 * skb_frag_page - retrieve the page referred to by a paged fragment
2576 * @frag: the paged fragment
2577 *
2578 * Returns the &struct page associated with @frag.
2579 */
2580static inline struct page *skb_frag_page(const skb_frag_t *frag)
2581{
2582 return frag->page.p;
2583}
2584
2585/**
2586 * __skb_frag_ref - take an addition reference on a paged fragment.
2587 * @frag: the paged fragment
2588 *
2589 * Takes an additional reference on the paged fragment @frag.
2590 */
2591static inline void __skb_frag_ref(skb_frag_t *frag)
2592{
2593 get_page(skb_frag_page(frag));
2594}
2595
2596/**
2597 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2598 * @skb: the buffer
2599 * @f: the fragment offset.
2600 *
2601 * Takes an additional reference on the @f'th paged fragment of @skb.
2602 */
2603static inline void skb_frag_ref(struct sk_buff *skb, int f)
2604{
2605 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2606}
2607
2608/**
2609 * __skb_frag_unref - release a reference on a paged fragment.
2610 * @frag: the paged fragment
2611 *
2612 * Releases a reference on the paged fragment @frag.
2613 */
2614static inline void __skb_frag_unref(skb_frag_t *frag)
2615{
2616 put_page(skb_frag_page(frag));
2617}
2618
2619/**
2620 * skb_frag_unref - release a reference on a paged fragment of an skb.
2621 * @skb: the buffer
2622 * @f: the fragment offset
2623 *
2624 * Releases a reference on the @f'th paged fragment of @skb.
2625 */
2626static inline void skb_frag_unref(struct sk_buff *skb, int f)
2627{
2628 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2629}
2630
2631/**
2632 * skb_frag_address - gets the address of the data contained in a paged fragment
2633 * @frag: the paged fragment buffer
2634 *
2635 * Returns the address of the data within @frag. The page must already
2636 * be mapped.
2637 */
2638static inline void *skb_frag_address(const skb_frag_t *frag)
2639{
2640 return page_address(skb_frag_page(frag)) + frag->page_offset;
2641}
2642
2643/**
2644 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2645 * @frag: the paged fragment buffer
2646 *
2647 * Returns the address of the data within @frag. Checks that the page
2648 * is mapped and returns %NULL otherwise.
2649 */
2650static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2651{
2652 void *ptr = page_address(skb_frag_page(frag));
2653 if (unlikely(!ptr))
2654 return NULL;
2655
2656 return ptr + frag->page_offset;
2657}
2658
2659/**
2660 * __skb_frag_set_page - sets the page contained in a paged fragment
2661 * @frag: the paged fragment
2662 * @page: the page to set
2663 *
2664 * Sets the fragment @frag to contain @page.
2665 */
2666static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2667{
2668 frag->page.p = page;
2669}
2670
2671/**
2672 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2673 * @skb: the buffer
2674 * @f: the fragment offset
2675 * @page: the page to set
2676 *
2677 * Sets the @f'th fragment of @skb to contain @page.
2678 */
2679static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2680 struct page *page)
2681{
2682 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2683}
2684
2685bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2686
2687/**
2688 * skb_frag_dma_map - maps a paged fragment via the DMA API
2689 * @dev: the device to map the fragment to
2690 * @frag: the paged fragment to map
2691 * @offset: the offset within the fragment (starting at the
2692 * fragment's own offset)
2693 * @size: the number of bytes to map
2694 * @dir: the direction of the mapping (%PCI_DMA_*)
2695 *
2696 * Maps the page associated with @frag to @device.
2697 */
2698static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2699 const skb_frag_t *frag,
2700 size_t offset, size_t size,
2701 enum dma_data_direction dir)
2702{
2703 return dma_map_page(dev, skb_frag_page(frag),
2704 frag->page_offset + offset, size, dir);
2705}
2706
2707static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2708 gfp_t gfp_mask)
2709{
2710 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2711}
2712
2713
2714static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2715 gfp_t gfp_mask)
2716{
2717 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2718}
2719
2720
2721/**
2722 * skb_clone_writable - is the header of a clone writable
2723 * @skb: buffer to check
2724 * @len: length up to which to write
2725 *
2726 * Returns true if modifying the header part of the cloned buffer
2727 * does not requires the data to be copied.
2728 */
2729static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2730{
2731 return !skb_header_cloned(skb) &&
2732 skb_headroom(skb) + len <= skb->hdr_len;
2733}
2734
2735static inline int skb_try_make_writable(struct sk_buff *skb,
2736 unsigned int write_len)
2737{
2738 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2739 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2740}
2741
2742static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2743 int cloned)
2744{
2745 int delta = 0;
2746
2747 if (headroom > skb_headroom(skb))
2748 delta = headroom - skb_headroom(skb);
2749
2750 if (delta || cloned)
2751 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2752 GFP_ATOMIC);
2753 return 0;
2754}
2755
2756/**
2757 * skb_cow - copy header of skb when it is required
2758 * @skb: buffer to cow
2759 * @headroom: needed headroom
2760 *
2761 * If the skb passed lacks sufficient headroom or its data part
2762 * is shared, data is reallocated. If reallocation fails, an error
2763 * is returned and original skb is not changed.
2764 *
2765 * The result is skb with writable area skb->head...skb->tail
2766 * and at least @headroom of space at head.
2767 */
2768static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2769{
2770 return __skb_cow(skb, headroom, skb_cloned(skb));
2771}
2772
2773/**
2774 * skb_cow_head - skb_cow but only making the head writable
2775 * @skb: buffer to cow
2776 * @headroom: needed headroom
2777 *
2778 * This function is identical to skb_cow except that we replace the
2779 * skb_cloned check by skb_header_cloned. It should be used when
2780 * you only need to push on some header and do not need to modify
2781 * the data.
2782 */
2783static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2784{
2785 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2786}
2787
2788/**
2789 * skb_padto - pad an skbuff up to a minimal size
2790 * @skb: buffer to pad
2791 * @len: minimal length
2792 *
2793 * Pads up a buffer to ensure the trailing bytes exist and are
2794 * blanked. If the buffer already contains sufficient data it
2795 * is untouched. Otherwise it is extended. Returns zero on
2796 * success. The skb is freed on error.
2797 */
2798static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2799{
2800 unsigned int size = skb->len;
2801 if (likely(size >= len))
2802 return 0;
2803 return skb_pad(skb, len - size);
2804}
2805
2806/**
2807 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2808 * @skb: buffer to pad
2809 * @len: minimal length
2810 *
2811 * Pads up a buffer to ensure the trailing bytes exist and are
2812 * blanked. If the buffer already contains sufficient data it
2813 * is untouched. Otherwise it is extended. Returns zero on
2814 * success. The skb is freed on error.
2815 */
2816static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2817{
2818 unsigned int size = skb->len;
2819
2820 if (unlikely(size < len)) {
2821 len -= size;
2822 if (skb_pad(skb, len))
2823 return -ENOMEM;
2824 __skb_put(skb, len);
2825 }
2826 return 0;
2827}
2828
2829static inline int skb_add_data(struct sk_buff *skb,
2830 struct iov_iter *from, int copy)
2831{
2832 const int off = skb->len;
2833
2834 if (skb->ip_summed == CHECKSUM_NONE) {
2835 __wsum csum = 0;
2836 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2837 &csum, from)) {
2838 skb->csum = csum_block_add(skb->csum, csum, off);
2839 return 0;
2840 }
2841 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2842 return 0;
2843
2844 __skb_trim(skb, off);
2845 return -EFAULT;
2846}
2847
2848static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2849 const struct page *page, int off)
2850{
2851 if (i) {
2852 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2853
2854 return page == skb_frag_page(frag) &&
2855 off == frag->page_offset + skb_frag_size(frag);
2856 }
2857 return false;
2858}
2859
2860static inline int __skb_linearize(struct sk_buff *skb)
2861{
2862 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2863}
2864
2865/**
2866 * skb_linearize - convert paged skb to linear one
2867 * @skb: buffer to linarize
2868 *
2869 * If there is no free memory -ENOMEM is returned, otherwise zero
2870 * is returned and the old skb data released.
2871 */
2872static inline int skb_linearize(struct sk_buff *skb)
2873{
2874 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2875}
2876
2877/**
2878 * skb_has_shared_frag - can any frag be overwritten
2879 * @skb: buffer to test
2880 *
2881 * Return true if the skb has at least one frag that might be modified
2882 * by an external entity (as in vmsplice()/sendfile())
2883 */
2884static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2885{
2886 return skb_is_nonlinear(skb) &&
2887 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2888}
2889
2890/**
2891 * skb_linearize_cow - make sure skb is linear and writable
2892 * @skb: buffer to process
2893 *
2894 * If there is no free memory -ENOMEM is returned, otherwise zero
2895 * is returned and the old skb data released.
2896 */
2897static inline int skb_linearize_cow(struct sk_buff *skb)
2898{
2899 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2900 __skb_linearize(skb) : 0;
2901}
2902
2903static __always_inline void
2904__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2905 unsigned int off)
2906{
2907 if (skb->ip_summed == CHECKSUM_COMPLETE)
2908 skb->csum = csum_block_sub(skb->csum,
2909 csum_partial(start, len, 0), off);
2910 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2911 skb_checksum_start_offset(skb) < 0)
2912 skb->ip_summed = CHECKSUM_NONE;
2913}
2914
2915/**
2916 * skb_postpull_rcsum - update checksum for received skb after pull
2917 * @skb: buffer to update
2918 * @start: start of data before pull
2919 * @len: length of data pulled
2920 *
2921 * After doing a pull on a received packet, you need to call this to
2922 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2923 * CHECKSUM_NONE so that it can be recomputed from scratch.
2924 */
2925static inline void skb_postpull_rcsum(struct sk_buff *skb,
2926 const void *start, unsigned int len)
2927{
2928 __skb_postpull_rcsum(skb, start, len, 0);
2929}
2930
2931static __always_inline void
2932__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2933 unsigned int off)
2934{
2935 if (skb->ip_summed == CHECKSUM_COMPLETE)
2936 skb->csum = csum_block_add(skb->csum,
2937 csum_partial(start, len, 0), off);
2938}
2939
2940/**
2941 * skb_postpush_rcsum - update checksum for received skb after push
2942 * @skb: buffer to update
2943 * @start: start of data after push
2944 * @len: length of data pushed
2945 *
2946 * After doing a push on a received packet, you need to call this to
2947 * update the CHECKSUM_COMPLETE checksum.
2948 */
2949static inline void skb_postpush_rcsum(struct sk_buff *skb,
2950 const void *start, unsigned int len)
2951{
2952 __skb_postpush_rcsum(skb, start, len, 0);
2953}
2954
2955unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2956
2957/**
2958 * skb_push_rcsum - push skb and update receive checksum
2959 * @skb: buffer to update
2960 * @len: length of data pulled
2961 *
2962 * This function performs an skb_push on the packet and updates
2963 * the CHECKSUM_COMPLETE checksum. It should be used on
2964 * receive path processing instead of skb_push unless you know
2965 * that the checksum difference is zero (e.g., a valid IP header)
2966 * or you are setting ip_summed to CHECKSUM_NONE.
2967 */
2968static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2969 unsigned int len)
2970{
2971 skb_push(skb, len);
2972 skb_postpush_rcsum(skb, skb->data, len);
2973 return skb->data;
2974}
2975
2976/**
2977 * pskb_trim_rcsum - trim received skb and update checksum
2978 * @skb: buffer to trim
2979 * @len: new length
2980 *
2981 * This is exactly the same as pskb_trim except that it ensures the
2982 * checksum of received packets are still valid after the operation.
2983 */
2984
2985static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2986{
2987 if (likely(len >= skb->len))
2988 return 0;
2989 if (skb->ip_summed == CHECKSUM_COMPLETE)
2990 skb->ip_summed = CHECKSUM_NONE;
2991 return __pskb_trim(skb, len);
2992}
2993
2994static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2995{
2996 if (skb->ip_summed == CHECKSUM_COMPLETE)
2997 skb->ip_summed = CHECKSUM_NONE;
2998 __skb_trim(skb, len);
2999 return 0;
3000}
3001
3002static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3003{
3004 if (skb->ip_summed == CHECKSUM_COMPLETE)
3005 skb->ip_summed = CHECKSUM_NONE;
3006 return __skb_grow(skb, len);
3007}
3008
3009#define skb_queue_walk(queue, skb) \
3010 for (skb = (queue)->next; \
3011 skb != (struct sk_buff *)(queue); \
3012 skb = skb->next)
3013
3014#define skb_queue_walk_safe(queue, skb, tmp) \
3015 for (skb = (queue)->next, tmp = skb->next; \
3016 skb != (struct sk_buff *)(queue); \
3017 skb = tmp, tmp = skb->next)
3018
3019#define skb_queue_walk_from(queue, skb) \
3020 for (; skb != (struct sk_buff *)(queue); \
3021 skb = skb->next)
3022
3023#define skb_queue_walk_from_safe(queue, skb, tmp) \
3024 for (tmp = skb->next; \
3025 skb != (struct sk_buff *)(queue); \
3026 skb = tmp, tmp = skb->next)
3027
3028#define skb_queue_reverse_walk(queue, skb) \
3029 for (skb = (queue)->prev; \
3030 skb != (struct sk_buff *)(queue); \
3031 skb = skb->prev)
3032
3033#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3034 for (skb = (queue)->prev, tmp = skb->prev; \
3035 skb != (struct sk_buff *)(queue); \
3036 skb = tmp, tmp = skb->prev)
3037
3038#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3039 for (tmp = skb->prev; \
3040 skb != (struct sk_buff *)(queue); \
3041 skb = tmp, tmp = skb->prev)
3042
3043static inline bool skb_has_frag_list(const struct sk_buff *skb)
3044{
3045 return skb_shinfo(skb)->frag_list != NULL;
3046}
3047
3048static inline void skb_frag_list_init(struct sk_buff *skb)
3049{
3050 skb_shinfo(skb)->frag_list = NULL;
3051}
3052
3053#define skb_walk_frags(skb, iter) \
3054 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3055
3056
3057int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3058 const struct sk_buff *skb);
3059struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3060 void (*destructor)(struct sock *sk,
3061 struct sk_buff *skb),
3062 int *peeked, int *off, int *err,
3063 struct sk_buff **last);
3064struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3065 void (*destructor)(struct sock *sk,
3066 struct sk_buff *skb),
3067 int *peeked, int *off, int *err);
3068struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3069 int *err);
3070unsigned int datagram_poll(struct file *file, struct socket *sock,
3071 struct poll_table_struct *wait);
3072int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3073 struct iov_iter *to, int size);
3074static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3075 struct msghdr *msg, int size)
3076{
3077 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3078}
3079int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3080 struct msghdr *msg);
3081int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3082 struct iov_iter *from, int len);
3083int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3084void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3085void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3086static inline void skb_free_datagram_locked(struct sock *sk,
3087 struct sk_buff *skb)
3088{
3089 __skb_free_datagram_locked(sk, skb, 0);
3090}
3091int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3092int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3093int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3094__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3095 int len, __wsum csum);
3096int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3097 struct pipe_inode_info *pipe, unsigned int len,
3098 unsigned int flags);
3099void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3100unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3101int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3102 int len, int hlen);
3103void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3104int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3105void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3106unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3107bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3108struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3109struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3110int skb_ensure_writable(struct sk_buff *skb, int write_len);
3111int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3112int skb_vlan_pop(struct sk_buff *skb);
3113int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3114struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3115 gfp_t gfp);
3116
3117static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3118{
3119 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3120}
3121
3122static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3123{
3124 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3125}
3126
3127struct skb_checksum_ops {
3128 __wsum (*update)(const void *mem, int len, __wsum wsum);
3129 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3130};
3131
3132__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3133 __wsum csum, const struct skb_checksum_ops *ops);
3134__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3135 __wsum csum);
3136
3137static inline void * __must_check
3138__skb_header_pointer(const struct sk_buff *skb, int offset,
3139 int len, void *data, int hlen, void *buffer)
3140{
3141 if (hlen - offset >= len)
3142 return data + offset;
3143
3144 if (!skb ||
3145 skb_copy_bits(skb, offset, buffer, len) < 0)
3146 return NULL;
3147
3148 return buffer;
3149}
3150
3151static inline void * __must_check
3152skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3153{
3154 return __skb_header_pointer(skb, offset, len, skb->data,
3155 skb_headlen(skb), buffer);
3156}
3157
3158/**
3159 * skb_needs_linearize - check if we need to linearize a given skb
3160 * depending on the given device features.
3161 * @skb: socket buffer to check
3162 * @features: net device features
3163 *
3164 * Returns true if either:
3165 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3166 * 2. skb is fragmented and the device does not support SG.
3167 */
3168static inline bool skb_needs_linearize(struct sk_buff *skb,
3169 netdev_features_t features)
3170{
3171 return skb_is_nonlinear(skb) &&
3172 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3173 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3174}
3175
3176static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3177 void *to,
3178 const unsigned int len)
3179{
3180 memcpy(to, skb->data, len);
3181}
3182
3183static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3184 const int offset, void *to,
3185 const unsigned int len)
3186{
3187 memcpy(to, skb->data + offset, len);
3188}
3189
3190static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3191 const void *from,
3192 const unsigned int len)
3193{
3194 memcpy(skb->data, from, len);
3195}
3196
3197static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3198 const int offset,
3199 const void *from,
3200 const unsigned int len)
3201{
3202 memcpy(skb->data + offset, from, len);
3203}
3204
3205void skb_init(void);
3206
3207static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3208{
3209 return skb->tstamp;
3210}
3211
3212/**
3213 * skb_get_timestamp - get timestamp from a skb
3214 * @skb: skb to get stamp from
3215 * @stamp: pointer to struct timeval to store stamp in
3216 *
3217 * Timestamps are stored in the skb as offsets to a base timestamp.
3218 * This function converts the offset back to a struct timeval and stores
3219 * it in stamp.
3220 */
3221static inline void skb_get_timestamp(const struct sk_buff *skb,
3222 struct timeval *stamp)
3223{
3224 *stamp = ktime_to_timeval(skb->tstamp);
3225}
3226
3227static inline void skb_get_timestampns(const struct sk_buff *skb,
3228 struct timespec *stamp)
3229{
3230 *stamp = ktime_to_timespec(skb->tstamp);
3231}
3232
3233static inline void __net_timestamp(struct sk_buff *skb)
3234{
3235 skb->tstamp = ktime_get_real();
3236}
3237
3238static inline ktime_t net_timedelta(ktime_t t)
3239{
3240 return ktime_sub(ktime_get_real(), t);
3241}
3242
3243static inline ktime_t net_invalid_timestamp(void)
3244{
3245 return 0;
3246}
3247
3248struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3249
3250#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3251
3252void skb_clone_tx_timestamp(struct sk_buff *skb);
3253bool skb_defer_rx_timestamp(struct sk_buff *skb);
3254
3255#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3256
3257static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3258{
3259}
3260
3261static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3262{
3263 return false;
3264}
3265
3266#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3267
3268/**
3269 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3270 *
3271 * PHY drivers may accept clones of transmitted packets for
3272 * timestamping via their phy_driver.txtstamp method. These drivers
3273 * must call this function to return the skb back to the stack with a
3274 * timestamp.
3275 *
3276 * @skb: clone of the the original outgoing packet
3277 * @hwtstamps: hardware time stamps
3278 *
3279 */
3280void skb_complete_tx_timestamp(struct sk_buff *skb,
3281 struct skb_shared_hwtstamps *hwtstamps);
3282
3283void __skb_tstamp_tx(struct sk_buff *orig_skb,
3284 struct skb_shared_hwtstamps *hwtstamps,
3285 struct sock *sk, int tstype);
3286
3287/**
3288 * skb_tstamp_tx - queue clone of skb with send time stamps
3289 * @orig_skb: the original outgoing packet
3290 * @hwtstamps: hardware time stamps, may be NULL if not available
3291 *
3292 * If the skb has a socket associated, then this function clones the
3293 * skb (thus sharing the actual data and optional structures), stores
3294 * the optional hardware time stamping information (if non NULL) or
3295 * generates a software time stamp (otherwise), then queues the clone
3296 * to the error queue of the socket. Errors are silently ignored.
3297 */
3298void skb_tstamp_tx(struct sk_buff *orig_skb,
3299 struct skb_shared_hwtstamps *hwtstamps);
3300
3301static inline void sw_tx_timestamp(struct sk_buff *skb)
3302{
3303 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3304 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3305 skb_tstamp_tx(skb, NULL);
3306}
3307
3308/**
3309 * skb_tx_timestamp() - Driver hook for transmit timestamping
3310 *
3311 * Ethernet MAC Drivers should call this function in their hard_xmit()
3312 * function immediately before giving the sk_buff to the MAC hardware.
3313 *
3314 * Specifically, one should make absolutely sure that this function is
3315 * called before TX completion of this packet can trigger. Otherwise
3316 * the packet could potentially already be freed.
3317 *
3318 * @skb: A socket buffer.
3319 */
3320static inline void skb_tx_timestamp(struct sk_buff *skb)
3321{
3322 skb_clone_tx_timestamp(skb);
3323 sw_tx_timestamp(skb);
3324}
3325
3326/**
3327 * skb_complete_wifi_ack - deliver skb with wifi status
3328 *
3329 * @skb: the original outgoing packet
3330 * @acked: ack status
3331 *
3332 */
3333void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3334
3335__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3336__sum16 __skb_checksum_complete(struct sk_buff *skb);
3337
3338static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3339{
3340 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3341 skb->csum_valid ||
3342 (skb->ip_summed == CHECKSUM_PARTIAL &&
3343 skb_checksum_start_offset(skb) >= 0));
3344}
3345
3346/**
3347 * skb_checksum_complete - Calculate checksum of an entire packet
3348 * @skb: packet to process
3349 *
3350 * This function calculates the checksum over the entire packet plus
3351 * the value of skb->csum. The latter can be used to supply the
3352 * checksum of a pseudo header as used by TCP/UDP. It returns the
3353 * checksum.
3354 *
3355 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3356 * this function can be used to verify that checksum on received
3357 * packets. In that case the function should return zero if the
3358 * checksum is correct. In particular, this function will return zero
3359 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3360 * hardware has already verified the correctness of the checksum.
3361 */
3362static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3363{
3364 return skb_csum_unnecessary(skb) ?
3365 0 : __skb_checksum_complete(skb);
3366}
3367
3368static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3369{
3370 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3371 if (skb->csum_level == 0)
3372 skb->ip_summed = CHECKSUM_NONE;
3373 else
3374 skb->csum_level--;
3375 }
3376}
3377
3378static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3379{
3380 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3381 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3382 skb->csum_level++;
3383 } else if (skb->ip_summed == CHECKSUM_NONE) {
3384 skb->ip_summed = CHECKSUM_UNNECESSARY;
3385 skb->csum_level = 0;
3386 }
3387}
3388
3389static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3390{
3391 /* Mark current checksum as bad (typically called from GRO
3392 * path). In the case that ip_summed is CHECKSUM_NONE
3393 * this must be the first checksum encountered in the packet.
3394 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3395 * checksum after the last one validated. For UDP, a zero
3396 * checksum can not be marked as bad.
3397 */
3398
3399 if (skb->ip_summed == CHECKSUM_NONE ||
3400 skb->ip_summed == CHECKSUM_UNNECESSARY)
3401 skb->csum_bad = 1;
3402}
3403
3404/* Check if we need to perform checksum complete validation.
3405 *
3406 * Returns true if checksum complete is needed, false otherwise
3407 * (either checksum is unnecessary or zero checksum is allowed).
3408 */
3409static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3410 bool zero_okay,
3411 __sum16 check)
3412{
3413 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3414 skb->csum_valid = 1;
3415 __skb_decr_checksum_unnecessary(skb);
3416 return false;
3417 }
3418
3419 return true;
3420}
3421
3422/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3423 * in checksum_init.
3424 */
3425#define CHECKSUM_BREAK 76
3426
3427/* Unset checksum-complete
3428 *
3429 * Unset checksum complete can be done when packet is being modified
3430 * (uncompressed for instance) and checksum-complete value is
3431 * invalidated.
3432 */
3433static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3434{
3435 if (skb->ip_summed == CHECKSUM_COMPLETE)
3436 skb->ip_summed = CHECKSUM_NONE;
3437}
3438
3439/* Validate (init) checksum based on checksum complete.
3440 *
3441 * Return values:
3442 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3443 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3444 * checksum is stored in skb->csum for use in __skb_checksum_complete
3445 * non-zero: value of invalid checksum
3446 *
3447 */
3448static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3449 bool complete,
3450 __wsum psum)
3451{
3452 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3453 if (!csum_fold(csum_add(psum, skb->csum))) {
3454 skb->csum_valid = 1;
3455 return 0;
3456 }
3457 } else if (skb->csum_bad) {
3458 /* ip_summed == CHECKSUM_NONE in this case */
3459 return (__force __sum16)1;
3460 }
3461
3462 skb->csum = psum;
3463
3464 if (complete || skb->len <= CHECKSUM_BREAK) {
3465 __sum16 csum;
3466
3467 csum = __skb_checksum_complete(skb);
3468 skb->csum_valid = !csum;
3469 return csum;
3470 }
3471
3472 return 0;
3473}
3474
3475static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3476{
3477 return 0;
3478}
3479
3480/* Perform checksum validate (init). Note that this is a macro since we only
3481 * want to calculate the pseudo header which is an input function if necessary.
3482 * First we try to validate without any computation (checksum unnecessary) and
3483 * then calculate based on checksum complete calling the function to compute
3484 * pseudo header.
3485 *
3486 * Return values:
3487 * 0: checksum is validated or try to in skb_checksum_complete
3488 * non-zero: value of invalid checksum
3489 */
3490#define __skb_checksum_validate(skb, proto, complete, \
3491 zero_okay, check, compute_pseudo) \
3492({ \
3493 __sum16 __ret = 0; \
3494 skb->csum_valid = 0; \
3495 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3496 __ret = __skb_checksum_validate_complete(skb, \
3497 complete, compute_pseudo(skb, proto)); \
3498 __ret; \
3499})
3500
3501#define skb_checksum_init(skb, proto, compute_pseudo) \
3502 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3503
3504#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3505 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3506
3507#define skb_checksum_validate(skb, proto, compute_pseudo) \
3508 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3509
3510#define skb_checksum_validate_zero_check(skb, proto, check, \
3511 compute_pseudo) \
3512 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3513
3514#define skb_checksum_simple_validate(skb) \
3515 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3516
3517static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3518{
3519 return (skb->ip_summed == CHECKSUM_NONE &&
3520 skb->csum_valid && !skb->csum_bad);
3521}
3522
3523static inline void __skb_checksum_convert(struct sk_buff *skb,
3524 __sum16 check, __wsum pseudo)
3525{
3526 skb->csum = ~pseudo;
3527 skb->ip_summed = CHECKSUM_COMPLETE;
3528}
3529
3530#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3531do { \
3532 if (__skb_checksum_convert_check(skb)) \
3533 __skb_checksum_convert(skb, check, \
3534 compute_pseudo(skb, proto)); \
3535} while (0)
3536
3537static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3538 u16 start, u16 offset)
3539{
3540 skb->ip_summed = CHECKSUM_PARTIAL;
3541 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3542 skb->csum_offset = offset - start;
3543}
3544
3545/* Update skbuf and packet to reflect the remote checksum offload operation.
3546 * When called, ptr indicates the starting point for skb->csum when
3547 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3548 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3549 */
3550static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3551 int start, int offset, bool nopartial)
3552{
3553 __wsum delta;
3554
3555 if (!nopartial) {
3556 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3557 return;
3558 }
3559
3560 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3561 __skb_checksum_complete(skb);
3562 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3563 }
3564
3565 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3566
3567 /* Adjust skb->csum since we changed the packet */
3568 skb->csum = csum_add(skb->csum, delta);
3569}
3570
3571static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3572{
3573#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3574 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3575#else
3576 return NULL;
3577#endif
3578}
3579
3580#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3581void nf_conntrack_destroy(struct nf_conntrack *nfct);
3582static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3583{
3584 if (nfct && atomic_dec_and_test(&nfct->use))
3585 nf_conntrack_destroy(nfct);
3586}
3587static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3588{
3589 if (nfct)
3590 atomic_inc(&nfct->use);
3591}
3592#endif
3593#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3594static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3595{
3596 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3597 kfree(nf_bridge);
3598}
3599static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3600{
3601 if (nf_bridge)
3602 atomic_inc(&nf_bridge->use);
3603}
3604#endif /* CONFIG_BRIDGE_NETFILTER */
3605static inline void nf_reset(struct sk_buff *skb)
3606{
3607#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3608 nf_conntrack_put(skb_nfct(skb));
3609 skb->_nfct = 0;
3610#endif
3611#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3612 nf_bridge_put(skb->nf_bridge);
3613 skb->nf_bridge = NULL;
3614#endif
3615}
3616
3617static inline void nf_reset_trace(struct sk_buff *skb)
3618{
3619#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3620 skb->nf_trace = 0;
3621#endif
3622}
3623
3624/* Note: This doesn't put any conntrack and bridge info in dst. */
3625static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3626 bool copy)
3627{
3628#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3629 dst->_nfct = src->_nfct;
3630 nf_conntrack_get(skb_nfct(src));
3631#endif
3632#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3633 dst->nf_bridge = src->nf_bridge;
3634 nf_bridge_get(src->nf_bridge);
3635#endif
3636#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3637 if (copy)
3638 dst->nf_trace = src->nf_trace;
3639#endif
3640}
3641
3642static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3643{
3644#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3645 nf_conntrack_put(skb_nfct(dst));
3646#endif
3647#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3648 nf_bridge_put(dst->nf_bridge);
3649#endif
3650 __nf_copy(dst, src, true);
3651}
3652
3653#ifdef CONFIG_NETWORK_SECMARK
3654static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3655{
3656 to->secmark = from->secmark;
3657}
3658
3659static inline void skb_init_secmark(struct sk_buff *skb)
3660{
3661 skb->secmark = 0;
3662}
3663#else
3664static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3665{ }
3666
3667static inline void skb_init_secmark(struct sk_buff *skb)
3668{ }
3669#endif
3670
3671static inline bool skb_irq_freeable(const struct sk_buff *skb)
3672{
3673 return !skb->destructor &&
3674#if IS_ENABLED(CONFIG_XFRM)
3675 !skb->sp &&
3676#endif
3677 !skb_nfct(skb) &&
3678 !skb->_skb_refdst &&
3679 !skb_has_frag_list(skb);
3680}
3681
3682static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3683{
3684 skb->queue_mapping = queue_mapping;
3685}
3686
3687static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3688{
3689 return skb->queue_mapping;
3690}
3691
3692static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3693{
3694 to->queue_mapping = from->queue_mapping;
3695}
3696
3697static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3698{
3699 skb->queue_mapping = rx_queue + 1;
3700}
3701
3702static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3703{
3704 return skb->queue_mapping - 1;
3705}
3706
3707static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3708{
3709 return skb->queue_mapping != 0;
3710}
3711
3712static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3713{
3714 skb->dst_pending_confirm = val;
3715}
3716
3717static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3718{
3719 return skb->dst_pending_confirm != 0;
3720}
3721
3722static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3723{
3724#ifdef CONFIG_XFRM
3725 return skb->sp;
3726#else
3727 return NULL;
3728#endif
3729}
3730
3731/* Keeps track of mac header offset relative to skb->head.
3732 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3733 * For non-tunnel skb it points to skb_mac_header() and for
3734 * tunnel skb it points to outer mac header.
3735 * Keeps track of level of encapsulation of network headers.
3736 */
3737struct skb_gso_cb {
3738 union {
3739 int mac_offset;
3740 int data_offset;
3741 };
3742 int encap_level;
3743 __wsum csum;
3744 __u16 csum_start;
3745};
3746#define SKB_SGO_CB_OFFSET 32
3747#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3748
3749static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3750{
3751 return (skb_mac_header(inner_skb) - inner_skb->head) -
3752 SKB_GSO_CB(inner_skb)->mac_offset;
3753}
3754
3755static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3756{
3757 int new_headroom, headroom;
3758 int ret;
3759
3760 headroom = skb_headroom(skb);
3761 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3762 if (ret)
3763 return ret;
3764
3765 new_headroom = skb_headroom(skb);
3766 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3767 return 0;
3768}
3769
3770static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3771{
3772 /* Do not update partial checksums if remote checksum is enabled. */
3773 if (skb->remcsum_offload)
3774 return;
3775
3776 SKB_GSO_CB(skb)->csum = res;
3777 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3778}
3779
3780/* Compute the checksum for a gso segment. First compute the checksum value
3781 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3782 * then add in skb->csum (checksum from csum_start to end of packet).
3783 * skb->csum and csum_start are then updated to reflect the checksum of the
3784 * resultant packet starting from the transport header-- the resultant checksum
3785 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3786 * header.
3787 */
3788static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3789{
3790 unsigned char *csum_start = skb_transport_header(skb);
3791 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3792 __wsum partial = SKB_GSO_CB(skb)->csum;
3793
3794 SKB_GSO_CB(skb)->csum = res;
3795 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3796
3797 return csum_fold(csum_partial(csum_start, plen, partial));
3798}
3799
3800static inline bool skb_is_gso(const struct sk_buff *skb)
3801{
3802 return skb_shinfo(skb)->gso_size;
3803}
3804
3805/* Note: Should be called only if skb_is_gso(skb) is true */
3806static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3807{
3808 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3809}
3810
3811static inline void skb_gso_reset(struct sk_buff *skb)
3812{
3813 skb_shinfo(skb)->gso_size = 0;
3814 skb_shinfo(skb)->gso_segs = 0;
3815 skb_shinfo(skb)->gso_type = 0;
3816}
3817
3818void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3819
3820static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3821{
3822 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3823 * wanted then gso_type will be set. */
3824 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3825
3826 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3827 unlikely(shinfo->gso_type == 0)) {
3828 __skb_warn_lro_forwarding(skb);
3829 return true;
3830 }
3831 return false;
3832}
3833
3834static inline void skb_forward_csum(struct sk_buff *skb)
3835{
3836 /* Unfortunately we don't support this one. Any brave souls? */
3837 if (skb->ip_summed == CHECKSUM_COMPLETE)
3838 skb->ip_summed = CHECKSUM_NONE;
3839}
3840
3841/**
3842 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3843 * @skb: skb to check
3844 *
3845 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3846 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3847 * use this helper, to document places where we make this assertion.
3848 */
3849static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3850{
3851#ifdef DEBUG
3852 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3853#endif
3854}
3855
3856bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3857
3858int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3859struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3860 unsigned int transport_len,
3861 __sum16(*skb_chkf)(struct sk_buff *skb));
3862
3863/**
3864 * skb_head_is_locked - Determine if the skb->head is locked down
3865 * @skb: skb to check
3866 *
3867 * The head on skbs build around a head frag can be removed if they are
3868 * not cloned. This function returns true if the skb head is locked down
3869 * due to either being allocated via kmalloc, or by being a clone with
3870 * multiple references to the head.
3871 */
3872static inline bool skb_head_is_locked(const struct sk_buff *skb)
3873{
3874 return !skb->head_frag || skb_cloned(skb);
3875}
3876
3877/**
3878 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3879 *
3880 * @skb: GSO skb
3881 *
3882 * skb_gso_network_seglen is used to determine the real size of the
3883 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3884 *
3885 * The MAC/L2 header is not accounted for.
3886 */
3887static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3888{
3889 unsigned int hdr_len = skb_transport_header(skb) -
3890 skb_network_header(skb);
3891 return hdr_len + skb_gso_transport_seglen(skb);
3892}
3893
3894/* Local Checksum Offload.
3895 * Compute outer checksum based on the assumption that the
3896 * inner checksum will be offloaded later.
3897 * See Documentation/networking/checksum-offloads.txt for
3898 * explanation of how this works.
3899 * Fill in outer checksum adjustment (e.g. with sum of outer
3900 * pseudo-header) before calling.
3901 * Also ensure that inner checksum is in linear data area.
3902 */
3903static inline __wsum lco_csum(struct sk_buff *skb)
3904{
3905 unsigned char *csum_start = skb_checksum_start(skb);
3906 unsigned char *l4_hdr = skb_transport_header(skb);
3907 __wsum partial;
3908
3909 /* Start with complement of inner checksum adjustment */
3910 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3911 skb->csum_offset));
3912
3913 /* Add in checksum of our headers (incl. outer checksum
3914 * adjustment filled in by caller) and return result.
3915 */
3916 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3917}
3918
3919#endif /* __KERNEL__ */
3920#endif /* _LINUX_SKBUFF_H */
3921