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 <net/flow_dissector.h>
38#include <linux/splice.h>
39#include <linux/in6.h>
40#include <linux/if_packet.h>
41#include <net/flow.h>
42
43/* The interface for checksum offload between the stack and networking drivers
44 * is as follows...
45 *
46 * A. IP checksum related features
47 *
48 * Drivers advertise checksum offload capabilities in the features of a device.
49 * From the stack's point of view these are capabilities offered by the driver,
50 * a driver typically only advertises features that it is capable of offloading
51 * to its device.
52 *
53 * The checksum related features are:
54 *
55 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
56 * IP (one's complement) checksum for any combination
57 * of protocols or protocol layering. The checksum is
58 * computed and set in a packet per the CHECKSUM_PARTIAL
59 * interface (see below).
60 *
61 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
62 * TCP or UDP packets over IPv4. These are specifically
63 * unencapsulated packets of the form IPv4|TCP or
64 * IPv4|UDP where the Protocol field in the IPv4 header
65 * is TCP or UDP. The IPv4 header may contain IP options
66 * This feature cannot be set in features for a device
67 * with NETIF_F_HW_CSUM also set. This feature is being
68 * DEPRECATED (see below).
69 *
70 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
71 * TCP or UDP packets over IPv6. These are specifically
72 * unencapsulated packets of the form IPv6|TCP or
73 * IPv4|UDP where the Next Header field in the IPv6
74 * header is either TCP or UDP. IPv6 extension headers
75 * are not supported with this feature. This feature
76 * cannot be set in features for a device with
77 * NETIF_F_HW_CSUM also set. This feature is being
78 * DEPRECATED (see below).
79 *
80 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
81 * This flag is used only used to disable the RX checksum
82 * feature for a device. The stack will accept receive
83 * checksum indication in packets received on a device
84 * regardless of whether NETIF_F_RXCSUM is set.
85 *
86 * B. Checksumming of received packets by device. Indication of checksum
87 * verification is in set skb->ip_summed. Possible values are:
88 *
89 * CHECKSUM_NONE:
90 *
91 * Device did not checksum this packet e.g. due to lack of capabilities.
92 * The packet contains full (though not verified) checksum in packet but
93 * not in skb->csum. Thus, skb->csum is undefined in this case.
94 *
95 * CHECKSUM_UNNECESSARY:
96 *
97 * The hardware you're dealing with doesn't calculate the full checksum
98 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
99 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
100 * if their checksums are okay. skb->csum is still undefined in this case
101 * though. A driver or device must never modify the checksum field in the
102 * packet even if checksum is verified.
103 *
104 * CHECKSUM_UNNECESSARY is applicable to following protocols:
105 * TCP: IPv6 and IPv4.
106 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
107 * zero UDP checksum for either IPv4 or IPv6, the networking stack
108 * may perform further validation in this case.
109 * GRE: only if the checksum is present in the header.
110 * SCTP: indicates the CRC in SCTP header has been validated.
111 *
112 * skb->csum_level indicates the number of consecutive checksums found in
113 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
114 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
115 * and a device is able to verify the checksums for UDP (possibly zero),
116 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
117 * two. If the device were only able to verify the UDP checksum and not
118 * GRE, either because it doesn't support GRE checksum of because GRE
119 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
120 * not considered in this case).
121 *
122 * CHECKSUM_COMPLETE:
123 *
124 * This is the most generic way. The device supplied checksum of the _whole_
125 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
126 * hardware doesn't need to parse L3/L4 headers to implement this.
127 *
128 * Note: Even if device supports only some protocols, but is able to produce
129 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
130 *
131 * CHECKSUM_PARTIAL:
132 *
133 * A checksum is set up to be offloaded to a device as described in the
134 * output description for CHECKSUM_PARTIAL. This may occur on a packet
135 * received directly from another Linux OS, e.g., a virtualized Linux kernel
136 * on the same host, or it may be set in the input path in GRO or remote
137 * checksum offload. For the purposes of checksum verification, the checksum
138 * referred to by skb->csum_start + skb->csum_offset and any preceding
139 * checksums in the packet are considered verified. Any checksums in the
140 * packet that are after the checksum being offloaded are not considered to
141 * be verified.
142 *
143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
144 * in the skb->ip_summed for a packet. Values are:
145 *
146 * CHECKSUM_PARTIAL:
147 *
148 * The driver is required to checksum the packet as seen by hard_start_xmit()
149 * from skb->csum_start up to the end, and to record/write the checksum at
150 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
151 * csum_start and csum_offset values are valid values given the length and
152 * offset of the packet, however they should not attempt to validate that the
153 * checksum refers to a legitimate transport layer checksum-- it is the
154 * purview of the stack to validate that csum_start and csum_offset are set
155 * correctly.
156 *
157 * When the stack requests checksum offload for a packet, the driver MUST
158 * ensure that the checksum is set correctly. A driver can either offload the
159 * checksum calculation to the device, or call skb_checksum_help (in the case
160 * that the device does not support offload for a particular checksum).
161 *
162 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
163 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
164 * checksum offload capability. If a device has limited checksum capabilities
165 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
166 * described above) a helper function can be called to resolve
167 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
168 * function takes a spec argument that describes the protocol layer that is
169 * supported for checksum offload and can be called for each packet. If a
170 * packet does not match the specification for offload, skb_checksum_help
171 * is called to resolve the checksum.
172 *
173 * CHECKSUM_NONE:
174 *
175 * The skb was already checksummed by the protocol, or a checksum is not
176 * required.
177 *
178 * CHECKSUM_UNNECESSARY:
179 *
180 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
181 * output.
182 *
183 * CHECKSUM_COMPLETE:
184 * Not used in checksum output. If a driver observes a packet with this value
185 * set in skbuff, if should treat as CHECKSUM_NONE being set.
186 *
187 * D. Non-IP checksum (CRC) offloads
188 *
189 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
190 * offloading the SCTP CRC in a packet. To perform this offload the stack
191 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
192 * accordingly. Note the there is no indication in the skbuff that the
193 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
194 * both IP checksum offload and SCTP CRC offload must verify which offload
195 * is configured for a packet presumably by inspecting packet headers.
196 *
197 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
198 * offloading the FCOE CRC in a packet. To perform this offload the stack
199 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
200 * accordingly. Note the there is no indication in the skbuff that the
201 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
202 * both IP checksum offload and FCOE CRC offload must verify which offload
203 * is configured for a packet presumably by inspecting packet headers.
204 *
205 * E. Checksumming on output with GSO.
206 *
207 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
208 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
209 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
210 * part of the GSO operation is implied. If a checksum is being offloaded
211 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
212 * are set to refer to the outermost checksum being offload (two offloaded
213 * checksums are possible with UDP encapsulation).
214 */
215
216/* Don't change this without changing skb_csum_unnecessary! */
217#define CHECKSUM_NONE 0
218#define CHECKSUM_UNNECESSARY 1
219#define CHECKSUM_COMPLETE 2
220#define CHECKSUM_PARTIAL 3
221
222/* Maximum value in skb->csum_level */
223#define SKB_MAX_CSUM_LEVEL 3
224
225#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
226#define SKB_WITH_OVERHEAD(X) \
227 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
228#define SKB_MAX_ORDER(X, ORDER) \
229 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
230#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
231#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
232
233/* return minimum truesize of one skb containing X bytes of data */
234#define SKB_TRUESIZE(X) ((X) + \
235 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
236 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
237
238struct net_device;
239struct scatterlist;
240struct pipe_inode_info;
241struct iov_iter;
242struct napi_struct;
243
244#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
245struct nf_conntrack {
246 atomic_t use;
247};
248#endif
249
250#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
251struct nf_bridge_info {
252 atomic_t use;
253 enum {
254 BRNF_PROTO_UNCHANGED,
255 BRNF_PROTO_8021Q,
256 BRNF_PROTO_PPPOE
257 } orig_proto:8;
258 u8 pkt_otherhost:1;
259 u8 in_prerouting:1;
260 u8 bridged_dnat:1;
261 __u16 frag_max_size;
262 struct net_device *physindev;
263
264 /* always valid & non-NULL from FORWARD on, for physdev match */
265 struct net_device *physoutdev;
266 union {
267 /* prerouting: detect dnat in orig/reply direction */
268 __be32 ipv4_daddr;
269 struct in6_addr ipv6_daddr;
270
271 /* after prerouting + nat detected: store original source
272 * mac since neigh resolution overwrites it, only used while
273 * skb is out in neigh layer.
274 */
275 char neigh_header[8];
276 };
277};
278#endif
279
280struct sk_buff_head {
281 /* These two members must be first. */
282 struct sk_buff *next;
283 struct sk_buff *prev;
284
285 __u32 qlen;
286 spinlock_t lock;
287};
288
289struct sk_buff;
290
291/* To allow 64K frame to be packed as single skb without frag_list we
292 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
293 * buffers which do not start on a page boundary.
294 *
295 * Since GRO uses frags we allocate at least 16 regardless of page
296 * size.
297 */
298#if (65536/PAGE_SIZE + 1) < 16
299#define MAX_SKB_FRAGS 16UL
300#else
301#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
302#endif
303extern int sysctl_max_skb_frags;
304
305/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
306 * segment using its current segmentation instead.
307 */
308#define GSO_BY_FRAGS 0xFFFF
309
310typedef struct skb_frag_struct skb_frag_t;
311
312struct skb_frag_struct {
313 struct {
314 struct page *p;
315 } page;
316#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
317 __u32 page_offset;
318 __u32 size;
319#else
320 __u16 page_offset;
321 __u16 size;
322#endif
323};
324
325static inline unsigned int skb_frag_size(const skb_frag_t *frag)
326{
327 return frag->size;
328}
329
330static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
331{
332 frag->size = size;
333}
334
335static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
336{
337 frag->size += delta;
338}
339
340static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
341{
342 frag->size -= delta;
343}
344
345#define HAVE_HW_TIME_STAMP
346
347/**
348 * struct skb_shared_hwtstamps - hardware time stamps
349 * @hwtstamp: hardware time stamp transformed into duration
350 * since arbitrary point in time
351 *
352 * Software time stamps generated by ktime_get_real() are stored in
353 * skb->tstamp.
354 *
355 * hwtstamps can only be compared against other hwtstamps from
356 * the same device.
357 *
358 * This structure is attached to packets as part of the
359 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
360 */
361struct skb_shared_hwtstamps {
362 ktime_t hwtstamp;
363};
364
365/* Definitions for tx_flags in struct skb_shared_info */
366enum {
367 /* generate hardware time stamp */
368 SKBTX_HW_TSTAMP = 1 << 0,
369
370 /* generate software time stamp when queueing packet to NIC */
371 SKBTX_SW_TSTAMP = 1 << 1,
372
373 /* device driver is going to provide hardware time stamp */
374 SKBTX_IN_PROGRESS = 1 << 2,
375
376 /* device driver supports TX zero-copy buffers */
377 SKBTX_DEV_ZEROCOPY = 1 << 3,
378
379 /* generate wifi status information (where possible) */
380 SKBTX_WIFI_STATUS = 1 << 4,
381
382 /* This indicates at least one fragment might be overwritten
383 * (as in vmsplice(), sendfile() ...)
384 * If we need to compute a TX checksum, we'll need to copy
385 * all frags to avoid possible bad checksum
386 */
387 SKBTX_SHARED_FRAG = 1 << 5,
388
389 /* generate software time stamp when entering packet scheduling */
390 SKBTX_SCHED_TSTAMP = 1 << 6,
391};
392
393#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
394 SKBTX_SCHED_TSTAMP)
395#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
396
397/*
398 * The callback notifies userspace to release buffers when skb DMA is done in
399 * lower device, the skb last reference should be 0 when calling this.
400 * The zerocopy_success argument is true if zero copy transmit occurred,
401 * false on data copy or out of memory error caused by data copy attempt.
402 * The ctx field is used to track device context.
403 * The desc field is used to track userspace buffer index.
404 */
405struct ubuf_info {
406 void (*callback)(struct ubuf_info *, bool zerocopy_success);
407 void *ctx;
408 unsigned long desc;
409};
410
411/* This data is invariant across clones and lives at
412 * the end of the header data, ie. at skb->end.
413 */
414struct skb_shared_info {
415 unsigned char nr_frags;
416 __u8 tx_flags;
417 unsigned short gso_size;
418 /* Warning: this field is not always filled in (UFO)! */
419 unsigned short gso_segs;
420 unsigned short gso_type;
421 struct sk_buff *frag_list;
422 struct skb_shared_hwtstamps hwtstamps;
423 u32 tskey;
424 __be32 ip6_frag_id;
425
426 /*
427 * Warning : all fields before dataref are cleared in __alloc_skb()
428 */
429 atomic_t dataref;
430
431 /* Intermediate layers must ensure that destructor_arg
432 * remains valid until skb destructor */
433 void * destructor_arg;
434
435 /* must be last field, see pskb_expand_head() */
436 skb_frag_t frags[MAX_SKB_FRAGS];
437};
438
439/* We divide dataref into two halves. The higher 16 bits hold references
440 * to the payload part of skb->data. The lower 16 bits hold references to
441 * the entire skb->data. A clone of a headerless skb holds the length of
442 * the header in skb->hdr_len.
443 *
444 * All users must obey the rule that the skb->data reference count must be
445 * greater than or equal to the payload reference count.
446 *
447 * Holding a reference to the payload part means that the user does not
448 * care about modifications to the header part of skb->data.
449 */
450#define SKB_DATAREF_SHIFT 16
451#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
452
453
454enum {
455 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
456 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
457 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
458};
459
460enum {
461 SKB_GSO_TCPV4 = 1 << 0,
462 SKB_GSO_UDP = 1 << 1,
463
464 /* This indicates the skb is from an untrusted source. */
465 SKB_GSO_DODGY = 1 << 2,
466
467 /* This indicates the tcp segment has CWR set. */
468 SKB_GSO_TCP_ECN = 1 << 3,
469
470 SKB_GSO_TCP_FIXEDID = 1 << 4,
471
472 SKB_GSO_TCPV6 = 1 << 5,
473
474 SKB_GSO_FCOE = 1 << 6,
475
476 SKB_GSO_GRE = 1 << 7,
477
478 SKB_GSO_GRE_CSUM = 1 << 8,
479
480 SKB_GSO_IPXIP4 = 1 << 9,
481
482 SKB_GSO_IPXIP6 = 1 << 10,
483
484 SKB_GSO_UDP_TUNNEL = 1 << 11,
485
486 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
487
488 SKB_GSO_PARTIAL = 1 << 13,
489
490 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
491
492 SKB_GSO_SCTP = 1 << 15,
493};
494
495#if BITS_PER_LONG > 32
496#define NET_SKBUFF_DATA_USES_OFFSET 1
497#endif
498
499#ifdef NET_SKBUFF_DATA_USES_OFFSET
500typedef unsigned int sk_buff_data_t;
501#else
502typedef unsigned char *sk_buff_data_t;
503#endif
504
505/**
506 * struct skb_mstamp - multi resolution time stamps
507 * @stamp_us: timestamp in us resolution
508 * @stamp_jiffies: timestamp in jiffies
509 */
510struct skb_mstamp {
511 union {
512 u64 v64;
513 struct {
514 u32 stamp_us;
515 u32 stamp_jiffies;
516 };
517 };
518};
519
520/**
521 * skb_mstamp_get - get current timestamp
522 * @cl: place to store timestamps
523 */
524static inline void skb_mstamp_get(struct skb_mstamp *cl)
525{
526 u64 val = local_clock();
527
528 do_div(val, NSEC_PER_USEC);
529 cl->stamp_us = (u32)val;
530 cl->stamp_jiffies = (u32)jiffies;
531}
532
533/**
534 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
535 * @t1: pointer to newest sample
536 * @t0: pointer to oldest sample
537 */
538static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
539 const struct skb_mstamp *t0)
540{
541 s32 delta_us = t1->stamp_us - t0->stamp_us;
542 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
543
544 /* If delta_us is negative, this might be because interval is too big,
545 * or local_clock() drift is too big : fallback using jiffies.
546 */
547 if (delta_us <= 0 ||
548 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
549
550 delta_us = jiffies_to_usecs(delta_jiffies);
551
552 return delta_us;
553}
554
555static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
556 const struct skb_mstamp *t0)
557{
558 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
559
560 if (!diff)
561 diff = t1->stamp_us - t0->stamp_us;
562 return diff > 0;
563}
564
565/**
566 * struct sk_buff - socket buffer
567 * @next: Next buffer in list
568 * @prev: Previous buffer in list
569 * @tstamp: Time we arrived/left
570 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
571 * @sk: Socket we are owned by
572 * @dev: Device we arrived on/are leaving by
573 * @cb: Control buffer. Free for use by every layer. Put private vars here
574 * @_skb_refdst: destination entry (with norefcount bit)
575 * @sp: the security path, used for xfrm
576 * @len: Length of actual data
577 * @data_len: Data length
578 * @mac_len: Length of link layer header
579 * @hdr_len: writable header length of cloned skb
580 * @csum: Checksum (must include start/offset pair)
581 * @csum_start: Offset from skb->head where checksumming should start
582 * @csum_offset: Offset from csum_start where checksum should be stored
583 * @priority: Packet queueing priority
584 * @ignore_df: allow local fragmentation
585 * @cloned: Head may be cloned (check refcnt to be sure)
586 * @ip_summed: Driver fed us an IP checksum
587 * @nohdr: Payload reference only, must not modify header
588 * @nfctinfo: Relationship of this skb to the connection
589 * @pkt_type: Packet class
590 * @fclone: skbuff clone status
591 * @ipvs_property: skbuff is owned by ipvs
592 * @peeked: this packet has been seen already, so stats have been
593 * done for it, don't do them again
594 * @nf_trace: netfilter packet trace flag
595 * @protocol: Packet protocol from driver
596 * @destructor: Destruct function
597 * @nfct: Associated connection, if any
598 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
599 * @skb_iif: ifindex of device we arrived on
600 * @tc_index: Traffic control index
601 * @tc_verd: traffic control verdict
602 * @hash: the packet hash
603 * @queue_mapping: Queue mapping for multiqueue devices
604 * @xmit_more: More SKBs are pending for this queue
605 * @ndisc_nodetype: router type (from link layer)
606 * @ooo_okay: allow the mapping of a socket to a queue to be changed
607 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
608 * ports.
609 * @sw_hash: indicates hash was computed in software stack
610 * @wifi_acked_valid: wifi_acked was set
611 * @wifi_acked: whether frame was acked on wifi or not
612 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
613 * @napi_id: id of the NAPI struct this skb came from
614 * @secmark: security marking
615 * @mark: Generic packet mark
616 * @vlan_proto: vlan encapsulation protocol
617 * @vlan_tci: vlan tag control information
618 * @inner_protocol: Protocol (encapsulation)
619 * @inner_transport_header: Inner transport layer header (encapsulation)
620 * @inner_network_header: Network layer header (encapsulation)
621 * @inner_mac_header: Link layer header (encapsulation)
622 * @transport_header: Transport layer header
623 * @network_header: Network layer header
624 * @mac_header: Link layer header
625 * @tail: Tail pointer
626 * @end: End pointer
627 * @head: Head of buffer
628 * @data: Data head pointer
629 * @truesize: Buffer size
630 * @users: User count - see {datagram,tcp}.c
631 */
632
633struct sk_buff {
634 union {
635 struct {
636 /* These two members must be first. */
637 struct sk_buff *next;
638 struct sk_buff *prev;
639
640 union {
641 ktime_t tstamp;
642 struct skb_mstamp skb_mstamp;
643 };
644 };
645 struct rb_node rbnode; /* used in netem & tcp stack */
646 };
647 struct sock *sk;
648
649 union {
650 struct net_device *dev;
651 /* Some protocols might use this space to store information,
652 * while device pointer would be NULL.
653 * UDP receive path is one user.
654 */
655 unsigned long dev_scratch;
656 };
657 /*
658 * This is the control buffer. It is free to use for every
659 * layer. Please put your private variables there. If you
660 * want to keep them across layers you have to do a skb_clone()
661 * first. This is owned by whoever has the skb queued ATM.
662 */
663 char cb[48] __aligned(8);
664
665 unsigned long _skb_refdst;
666 void (*destructor)(struct sk_buff *skb);
667#ifdef CONFIG_XFRM
668 struct sec_path *sp;
669#endif
670#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
671 struct nf_conntrack *nfct;
672#endif
673#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
674 struct nf_bridge_info *nf_bridge;
675#endif
676 unsigned int len,
677 data_len;
678 __u16 mac_len,
679 hdr_len;
680
681 /* Following fields are _not_ copied in __copy_skb_header()
682 * Note that queue_mapping is here mostly to fill a hole.
683 */
684 kmemcheck_bitfield_begin(flags1);
685 __u16 queue_mapping;
686
687/* if you move cloned around you also must adapt those constants */
688#ifdef __BIG_ENDIAN_BITFIELD
689#define CLONED_MASK (1 << 7)
690#else
691#define CLONED_MASK 1
692#endif
693#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
694
695 __u8 __cloned_offset[0];
696 __u8 cloned:1,
697 nohdr:1,
698 fclone:2,
699 peeked:1,
700 head_frag:1,
701 xmit_more:1,
702 __unused:1; /* one bit hole */
703 kmemcheck_bitfield_end(flags1);
704
705 /* fields enclosed in headers_start/headers_end are copied
706 * using a single memcpy() in __copy_skb_header()
707 */
708 /* private: */
709 __u32 headers_start[0];
710 /* public: */
711
712/* if you move pkt_type around you also must adapt those constants */
713#ifdef __BIG_ENDIAN_BITFIELD
714#define PKT_TYPE_MAX (7 << 5)
715#else
716#define PKT_TYPE_MAX 7
717#endif
718#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
719
720 __u8 __pkt_type_offset[0];
721 __u8 pkt_type:3;
722 __u8 pfmemalloc:1;
723 __u8 ignore_df:1;
724 __u8 nfctinfo:3;
725
726 __u8 nf_trace:1;
727 __u8 ip_summed:2;
728 __u8 ooo_okay:1;
729 __u8 l4_hash:1;
730 __u8 sw_hash:1;
731 __u8 wifi_acked_valid:1;
732 __u8 wifi_acked:1;
733
734 __u8 no_fcs:1;
735 /* Indicates the inner headers are valid in the skbuff. */
736 __u8 encapsulation:1;
737 __u8 encap_hdr_csum:1;
738 __u8 csum_valid:1;
739 __u8 csum_complete_sw:1;
740 __u8 csum_level:2;
741 __u8 csum_bad:1;
742
743#ifdef CONFIG_IPV6_NDISC_NODETYPE
744 __u8 ndisc_nodetype:2;
745#endif
746 __u8 ipvs_property:1;
747 __u8 inner_protocol_type:1;
748 __u8 remcsum_offload:1;
749#ifdef CONFIG_NET_SWITCHDEV
750 __u8 offload_fwd_mark:1;
751#endif
752 /* 2, 4 or 5 bit hole */
753
754#ifdef CONFIG_NET_SCHED
755 __u16 tc_index; /* traffic control index */
756#ifdef CONFIG_NET_CLS_ACT
757 __u16 tc_verd; /* traffic control verdict */
758#endif
759#endif
760
761 union {
762 __wsum csum;
763 struct {
764 __u16 csum_start;
765 __u16 csum_offset;
766 };
767 };
768 __u32 priority;
769 int skb_iif;
770 __u32 hash;
771 __be16 vlan_proto;
772 __u16 vlan_tci;
773#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
774 union {
775 unsigned int napi_id;
776 unsigned int sender_cpu;
777 };
778#endif
779#ifdef CONFIG_NETWORK_SECMARK
780 __u32 secmark;
781#endif
782
783 union {
784 __u32 mark;
785 __u32 reserved_tailroom;
786 };
787
788 union {
789 __be16 inner_protocol;
790 __u8 inner_ipproto;
791 };
792
793 __u16 inner_transport_header;
794 __u16 inner_network_header;
795 __u16 inner_mac_header;
796
797 __be16 protocol;
798 __u16 transport_header;
799 __u16 network_header;
800 __u16 mac_header;
801
802 /* private: */
803 __u32 headers_end[0];
804 /* public: */
805
806 /* These elements must be at the end, see alloc_skb() for details. */
807 sk_buff_data_t tail;
808 sk_buff_data_t end;
809 unsigned char *head,
810 *data;
811 unsigned int truesize;
812 atomic_t users;
813};
814
815#ifdef __KERNEL__
816/*
817 * Handling routines are only of interest to the kernel
818 */
819#include <linux/slab.h>
820
821
822#define SKB_ALLOC_FCLONE 0x01
823#define SKB_ALLOC_RX 0x02
824#define SKB_ALLOC_NAPI 0x04
825
826/* Returns true if the skb was allocated from PFMEMALLOC reserves */
827static inline bool skb_pfmemalloc(const struct sk_buff *skb)
828{
829 return unlikely(skb->pfmemalloc);
830}
831
832/*
833 * skb might have a dst pointer attached, refcounted or not.
834 * _skb_refdst low order bit is set if refcount was _not_ taken
835 */
836#define SKB_DST_NOREF 1UL
837#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
838
839/**
840 * skb_dst - returns skb dst_entry
841 * @skb: buffer
842 *
843 * Returns skb dst_entry, regardless of reference taken or not.
844 */
845static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
846{
847 /* If refdst was not refcounted, check we still are in a
848 * rcu_read_lock section
849 */
850 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
851 !rcu_read_lock_held() &&
852 !rcu_read_lock_bh_held());
853 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
854}
855
856/**
857 * skb_dst_set - sets skb dst
858 * @skb: buffer
859 * @dst: dst entry
860 *
861 * Sets skb dst, assuming a reference was taken on dst and should
862 * be released by skb_dst_drop()
863 */
864static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
865{
866 skb->_skb_refdst = (unsigned long)dst;
867}
868
869/**
870 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
871 * @skb: buffer
872 * @dst: dst entry
873 *
874 * Sets skb dst, assuming a reference was not taken on dst.
875 * If dst entry is cached, we do not take reference and dst_release
876 * will be avoided by refdst_drop. If dst entry is not cached, we take
877 * reference, so that last dst_release can destroy the dst immediately.
878 */
879static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
880{
881 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
882 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
883}
884
885/**
886 * skb_dst_is_noref - Test if skb dst isn't refcounted
887 * @skb: buffer
888 */
889static inline bool skb_dst_is_noref(const struct sk_buff *skb)
890{
891 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
892}
893
894static inline struct rtable *skb_rtable(const struct sk_buff *skb)
895{
896 return (struct rtable *)skb_dst(skb);
897}
898
899/* For mangling skb->pkt_type from user space side from applications
900 * such as nft, tc, etc, we only allow a conservative subset of
901 * possible pkt_types to be set.
902*/
903static inline bool skb_pkt_type_ok(u32 ptype)
904{
905 return ptype <= PACKET_OTHERHOST;
906}
907
908void kfree_skb(struct sk_buff *skb);
909void kfree_skb_list(struct sk_buff *segs);
910void skb_tx_error(struct sk_buff *skb);
911void consume_skb(struct sk_buff *skb);
912void __kfree_skb(struct sk_buff *skb);
913extern struct kmem_cache *skbuff_head_cache;
914
915void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
916bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
917 bool *fragstolen, int *delta_truesize);
918
919struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
920 int node);
921struct sk_buff *__build_skb(void *data, unsigned int frag_size);
922struct sk_buff *build_skb(void *data, unsigned int frag_size);
923static inline struct sk_buff *alloc_skb(unsigned int size,
924 gfp_t priority)
925{
926 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
927}
928
929struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
930 unsigned long data_len,
931 int max_page_order,
932 int *errcode,
933 gfp_t gfp_mask);
934
935/* Layout of fast clones : [skb1][skb2][fclone_ref] */
936struct sk_buff_fclones {
937 struct sk_buff skb1;
938
939 struct sk_buff skb2;
940
941 atomic_t fclone_ref;
942};
943
944/**
945 * skb_fclone_busy - check if fclone is busy
946 * @sk: socket
947 * @skb: buffer
948 *
949 * Returns true if skb is a fast clone, and its clone is not freed.
950 * Some drivers call skb_orphan() in their ndo_start_xmit(),
951 * so we also check that this didnt happen.
952 */
953static inline bool skb_fclone_busy(const struct sock *sk,
954 const struct sk_buff *skb)
955{
956 const struct sk_buff_fclones *fclones;
957
958 fclones = container_of(skb, struct sk_buff_fclones, skb1);
959
960 return skb->fclone == SKB_FCLONE_ORIG &&
961 atomic_read(&fclones->fclone_ref) > 1 &&
962 fclones->skb2.sk == sk;
963}
964
965static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
966 gfp_t priority)
967{
968 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
969}
970
971struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
972static inline struct sk_buff *alloc_skb_head(gfp_t priority)
973{
974 return __alloc_skb_head(priority, -1);
975}
976
977struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
978int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
979struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
980struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
981struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
982 gfp_t gfp_mask, bool fclone);
983static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
984 gfp_t gfp_mask)
985{
986 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
987}
988
989int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
990struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
991 unsigned int headroom);
992struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
993 int newtailroom, gfp_t priority);
994int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
995 int offset, int len);
996int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
997 int len);
998int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
999int skb_pad(struct sk_buff *skb, int pad);
1000#define dev_kfree_skb(a) consume_skb(a)
1001
1002int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1003 int getfrag(void *from, char *to, int offset,
1004 int len, int odd, struct sk_buff *skb),
1005 void *from, int length);
1006
1007int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1008 int offset, size_t size);
1009
1010struct skb_seq_state {
1011 __u32 lower_offset;
1012 __u32 upper_offset;
1013 __u32 frag_idx;
1014 __u32 stepped_offset;
1015 struct sk_buff *root_skb;
1016 struct sk_buff *cur_skb;
1017 __u8 *frag_data;
1018};
1019
1020void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1021 unsigned int to, struct skb_seq_state *st);
1022unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1023 struct skb_seq_state *st);
1024void skb_abort_seq_read(struct skb_seq_state *st);
1025
1026unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1027 unsigned int to, struct ts_config *config);
1028
1029/*
1030 * Packet hash types specify the type of hash in skb_set_hash.
1031 *
1032 * Hash types refer to the protocol layer addresses which are used to
1033 * construct a packet's hash. The hashes are used to differentiate or identify
1034 * flows of the protocol layer for the hash type. Hash types are either
1035 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1036 *
1037 * Properties of hashes:
1038 *
1039 * 1) Two packets in different flows have different hash values
1040 * 2) Two packets in the same flow should have the same hash value
1041 *
1042 * A hash at a higher layer is considered to be more specific. A driver should
1043 * set the most specific hash possible.
1044 *
1045 * A driver cannot indicate a more specific hash than the layer at which a hash
1046 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1047 *
1048 * A driver may indicate a hash level which is less specific than the
1049 * actual layer the hash was computed on. For instance, a hash computed
1050 * at L4 may be considered an L3 hash. This should only be done if the
1051 * driver can't unambiguously determine that the HW computed the hash at
1052 * the higher layer. Note that the "should" in the second property above
1053 * permits this.
1054 */
1055enum pkt_hash_types {
1056 PKT_HASH_TYPE_NONE, /* Undefined type */
1057 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1058 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1059 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1060};
1061
1062static inline void skb_clear_hash(struct sk_buff *skb)
1063{
1064 skb->hash = 0;
1065 skb->sw_hash = 0;
1066 skb->l4_hash = 0;
1067}
1068
1069static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1070{
1071 if (!skb->l4_hash)
1072 skb_clear_hash(skb);
1073}
1074
1075static inline void
1076__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1077{
1078 skb->l4_hash = is_l4;
1079 skb->sw_hash = is_sw;
1080 skb->hash = hash;
1081}
1082
1083static inline void
1084skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1085{
1086 /* Used by drivers to set hash from HW */
1087 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1088}
1089
1090static inline void
1091__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1092{
1093 __skb_set_hash(skb, hash, true, is_l4);
1094}
1095
1096void __skb_get_hash(struct sk_buff *skb);
1097u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1098u32 skb_get_poff(const struct sk_buff *skb);
1099u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1100 const struct flow_keys *keys, int hlen);
1101__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1102 void *data, int hlen_proto);
1103
1104static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1105 int thoff, u8 ip_proto)
1106{
1107 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1108}
1109
1110void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1111 const struct flow_dissector_key *key,
1112 unsigned int key_count);
1113
1114bool __skb_flow_dissect(const struct sk_buff *skb,
1115 struct flow_dissector *flow_dissector,
1116 void *target_container,
1117 void *data, __be16 proto, int nhoff, int hlen,
1118 unsigned int flags);
1119
1120static inline bool skb_flow_dissect(const struct sk_buff *skb,
1121 struct flow_dissector *flow_dissector,
1122 void *target_container, unsigned int flags)
1123{
1124 return __skb_flow_dissect(skb, flow_dissector, target_container,
1125 NULL, 0, 0, 0, flags);
1126}
1127
1128static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1129 struct flow_keys *flow,
1130 unsigned int flags)
1131{
1132 memset(flow, 0, sizeof(*flow));
1133 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1134 NULL, 0, 0, 0, flags);
1135}
1136
1137static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1138 void *data, __be16 proto,
1139 int nhoff, int hlen,
1140 unsigned int flags)
1141{
1142 memset(flow, 0, sizeof(*flow));
1143 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1144 data, proto, nhoff, hlen, flags);
1145}
1146
1147static inline __u32 skb_get_hash(struct sk_buff *skb)
1148{
1149 if (!skb->l4_hash && !skb->sw_hash)
1150 __skb_get_hash(skb);
1151
1152 return skb->hash;
1153}
1154
1155__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1156
1157static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1158{
1159 if (!skb->l4_hash && !skb->sw_hash) {
1160 struct flow_keys keys;
1161 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1162
1163 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1164 }
1165
1166 return skb->hash;
1167}
1168
1169__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1170
1171static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1172{
1173 if (!skb->l4_hash && !skb->sw_hash) {
1174 struct flow_keys keys;
1175 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1176
1177 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1178 }
1179
1180 return skb->hash;
1181}
1182
1183__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1184
1185static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1186{
1187 return skb->hash;
1188}
1189
1190static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1191{
1192 to->hash = from->hash;
1193 to->sw_hash = from->sw_hash;
1194 to->l4_hash = from->l4_hash;
1195};
1196
1197#ifdef NET_SKBUFF_DATA_USES_OFFSET
1198static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1199{
1200 return skb->head + skb->end;
1201}
1202
1203static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1204{
1205 return skb->end;
1206}
1207#else
1208static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1209{
1210 return skb->end;
1211}
1212
1213static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1214{
1215 return skb->end - skb->head;
1216}
1217#endif
1218
1219/* Internal */
1220#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1221
1222static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1223{
1224 return &skb_shinfo(skb)->hwtstamps;
1225}
1226
1227/**
1228 * skb_queue_empty - check if a queue is empty
1229 * @list: queue head
1230 *
1231 * Returns true if the queue is empty, false otherwise.
1232 */
1233static inline int skb_queue_empty(const struct sk_buff_head *list)
1234{
1235 return list->next == (const struct sk_buff *) list;
1236}
1237
1238/**
1239 * skb_queue_is_last - check if skb is the last entry in the queue
1240 * @list: queue head
1241 * @skb: buffer
1242 *
1243 * Returns true if @skb is the last buffer on the list.
1244 */
1245static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1246 const struct sk_buff *skb)
1247{
1248 return skb->next == (const struct sk_buff *) list;
1249}
1250
1251/**
1252 * skb_queue_is_first - check if skb is the first entry in the queue
1253 * @list: queue head
1254 * @skb: buffer
1255 *
1256 * Returns true if @skb is the first buffer on the list.
1257 */
1258static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1259 const struct sk_buff *skb)
1260{
1261 return skb->prev == (const struct sk_buff *) list;
1262}
1263
1264/**
1265 * skb_queue_next - return the next packet in the queue
1266 * @list: queue head
1267 * @skb: current buffer
1268 *
1269 * Return the next packet in @list after @skb. It is only valid to
1270 * call this if skb_queue_is_last() evaluates to false.
1271 */
1272static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1273 const struct sk_buff *skb)
1274{
1275 /* This BUG_ON may seem severe, but if we just return then we
1276 * are going to dereference garbage.
1277 */
1278 BUG_ON(skb_queue_is_last(list, skb));
1279 return skb->next;
1280}
1281
1282/**
1283 * skb_queue_prev - return the prev packet in the queue
1284 * @list: queue head
1285 * @skb: current buffer
1286 *
1287 * Return the prev packet in @list before @skb. It is only valid to
1288 * call this if skb_queue_is_first() evaluates to false.
1289 */
1290static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1291 const struct sk_buff *skb)
1292{
1293 /* This BUG_ON may seem severe, but if we just return then we
1294 * are going to dereference garbage.
1295 */
1296 BUG_ON(skb_queue_is_first(list, skb));
1297 return skb->prev;
1298}
1299
1300/**
1301 * skb_get - reference buffer
1302 * @skb: buffer to reference
1303 *
1304 * Makes another reference to a socket buffer and returns a pointer
1305 * to the buffer.
1306 */
1307static inline struct sk_buff *skb_get(struct sk_buff *skb)
1308{
1309 atomic_inc(&skb->users);
1310 return skb;
1311}
1312
1313/*
1314 * If users == 1, we are the only owner and are can avoid redundant
1315 * atomic change.
1316 */
1317
1318/**
1319 * skb_cloned - is the buffer a clone
1320 * @skb: buffer to check
1321 *
1322 * Returns true if the buffer was generated with skb_clone() and is
1323 * one of multiple shared copies of the buffer. Cloned buffers are
1324 * shared data so must not be written to under normal circumstances.
1325 */
1326static inline int skb_cloned(const struct sk_buff *skb)
1327{
1328 return skb->cloned &&
1329 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1330}
1331
1332static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1333{
1334 might_sleep_if(gfpflags_allow_blocking(pri));
1335
1336 if (skb_cloned(skb))
1337 return pskb_expand_head(skb, 0, 0, pri);
1338
1339 return 0;
1340}
1341
1342/**
1343 * skb_header_cloned - is the header a clone
1344 * @skb: buffer to check
1345 *
1346 * Returns true if modifying the header part of the buffer requires
1347 * the data to be copied.
1348 */
1349static inline int skb_header_cloned(const struct sk_buff *skb)
1350{
1351 int dataref;
1352
1353 if (!skb->cloned)
1354 return 0;
1355
1356 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1357 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1358 return dataref != 1;
1359}
1360
1361static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1362{
1363 might_sleep_if(gfpflags_allow_blocking(pri));
1364
1365 if (skb_header_cloned(skb))
1366 return pskb_expand_head(skb, 0, 0, pri);
1367
1368 return 0;
1369}
1370
1371/**
1372 * skb_header_release - release reference to header
1373 * @skb: buffer to operate on
1374 *
1375 * Drop a reference to the header part of the buffer. This is done
1376 * by acquiring a payload reference. You must not read from the header
1377 * part of skb->data after this.
1378 * Note : Check if you can use __skb_header_release() instead.
1379 */
1380static inline void skb_header_release(struct sk_buff *skb)
1381{
1382 BUG_ON(skb->nohdr);
1383 skb->nohdr = 1;
1384 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1385}
1386
1387/**
1388 * __skb_header_release - release reference to header
1389 * @skb: buffer to operate on
1390 *
1391 * Variant of skb_header_release() assuming skb is private to caller.
1392 * We can avoid one atomic operation.
1393 */
1394static inline void __skb_header_release(struct sk_buff *skb)
1395{
1396 skb->nohdr = 1;
1397 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1398}
1399
1400
1401/**
1402 * skb_shared - is the buffer shared
1403 * @skb: buffer to check
1404 *
1405 * Returns true if more than one person has a reference to this
1406 * buffer.
1407 */
1408static inline int skb_shared(const struct sk_buff *skb)
1409{
1410 return atomic_read(&skb->users) != 1;
1411}
1412
1413/**
1414 * skb_share_check - check if buffer is shared and if so clone it
1415 * @skb: buffer to check
1416 * @pri: priority for memory allocation
1417 *
1418 * If the buffer is shared the buffer is cloned and the old copy
1419 * drops a reference. A new clone with a single reference is returned.
1420 * If the buffer is not shared the original buffer is returned. When
1421 * being called from interrupt status or with spinlocks held pri must
1422 * be GFP_ATOMIC.
1423 *
1424 * NULL is returned on a memory allocation failure.
1425 */
1426static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1427{
1428 might_sleep_if(gfpflags_allow_blocking(pri));
1429 if (skb_shared(skb)) {
1430 struct sk_buff *nskb = skb_clone(skb, pri);
1431
1432 if (likely(nskb))
1433 consume_skb(skb);
1434 else
1435 kfree_skb(skb);
1436 skb = nskb;
1437 }
1438 return skb;
1439}
1440
1441/*
1442 * Copy shared buffers into a new sk_buff. We effectively do COW on
1443 * packets to handle cases where we have a local reader and forward
1444 * and a couple of other messy ones. The normal one is tcpdumping
1445 * a packet thats being forwarded.
1446 */
1447
1448/**
1449 * skb_unshare - make a copy of a shared buffer
1450 * @skb: buffer to check
1451 * @pri: priority for memory allocation
1452 *
1453 * If the socket buffer is a clone then this function creates a new
1454 * copy of the data, drops a reference count on the old copy and returns
1455 * the new copy with the reference count at 1. If the buffer is not a clone
1456 * the original buffer is returned. When called with a spinlock held or
1457 * from interrupt state @pri must be %GFP_ATOMIC
1458 *
1459 * %NULL is returned on a memory allocation failure.
1460 */
1461static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1462 gfp_t pri)
1463{
1464 might_sleep_if(gfpflags_allow_blocking(pri));
1465 if (skb_cloned(skb)) {
1466 struct sk_buff *nskb = skb_copy(skb, pri);
1467
1468 /* Free our shared copy */
1469 if (likely(nskb))
1470 consume_skb(skb);
1471 else
1472 kfree_skb(skb);
1473 skb = nskb;
1474 }
1475 return skb;
1476}
1477
1478/**
1479 * skb_peek - peek at the head of an &sk_buff_head
1480 * @list_: list to peek at
1481 *
1482 * Peek an &sk_buff. Unlike most other operations you _MUST_
1483 * be careful with this one. A peek leaves the buffer on the
1484 * list and someone else may run off with it. You must hold
1485 * the appropriate locks or have a private queue to do this.
1486 *
1487 * Returns %NULL for an empty list or a pointer to the head element.
1488 * The reference count is not incremented and the reference is therefore
1489 * volatile. Use with caution.
1490 */
1491static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1492{
1493 struct sk_buff *skb = list_->next;
1494
1495 if (skb == (struct sk_buff *)list_)
1496 skb = NULL;
1497 return skb;
1498}
1499
1500/**
1501 * skb_peek_next - peek skb following the given one from a queue
1502 * @skb: skb to start from
1503 * @list_: list to peek at
1504 *
1505 * Returns %NULL when the end of the list is met or a pointer to the
1506 * next element. The reference count is not incremented and the
1507 * reference is therefore volatile. Use with caution.
1508 */
1509static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1510 const struct sk_buff_head *list_)
1511{
1512 struct sk_buff *next = skb->next;
1513
1514 if (next == (struct sk_buff *)list_)
1515 next = NULL;
1516 return next;
1517}
1518
1519/**
1520 * skb_peek_tail - peek at the tail of an &sk_buff_head
1521 * @list_: list to peek at
1522 *
1523 * Peek an &sk_buff. Unlike most other operations you _MUST_
1524 * be careful with this one. A peek leaves the buffer on the
1525 * list and someone else may run off with it. You must hold
1526 * the appropriate locks or have a private queue to do this.
1527 *
1528 * Returns %NULL for an empty list or a pointer to the tail element.
1529 * The reference count is not incremented and the reference is therefore
1530 * volatile. Use with caution.
1531 */
1532static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1533{
1534 struct sk_buff *skb = list_->prev;
1535
1536 if (skb == (struct sk_buff *)list_)
1537 skb = NULL;
1538 return skb;
1539
1540}
1541
1542/**
1543 * skb_queue_len - get queue length
1544 * @list_: list to measure
1545 *
1546 * Return the length of an &sk_buff queue.
1547 */
1548static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1549{
1550 return list_->qlen;
1551}
1552
1553/**
1554 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1555 * @list: queue to initialize
1556 *
1557 * This initializes only the list and queue length aspects of
1558 * an sk_buff_head object. This allows to initialize the list
1559 * aspects of an sk_buff_head without reinitializing things like
1560 * the spinlock. It can also be used for on-stack sk_buff_head
1561 * objects where the spinlock is known to not be used.
1562 */
1563static inline void __skb_queue_head_init(struct sk_buff_head *list)
1564{
1565 list->prev = list->next = (struct sk_buff *)list;
1566 list->qlen = 0;
1567}
1568
1569/*
1570 * This function creates a split out lock class for each invocation;
1571 * this is needed for now since a whole lot of users of the skb-queue
1572 * infrastructure in drivers have different locking usage (in hardirq)
1573 * than the networking core (in softirq only). In the long run either the
1574 * network layer or drivers should need annotation to consolidate the
1575 * main types of usage into 3 classes.
1576 */
1577static inline void skb_queue_head_init(struct sk_buff_head *list)
1578{
1579 spin_lock_init(&list->lock);
1580 __skb_queue_head_init(list);
1581}
1582
1583static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1584 struct lock_class_key *class)
1585{
1586 skb_queue_head_init(list);
1587 lockdep_set_class(&list->lock, class);
1588}
1589
1590/*
1591 * Insert an sk_buff on a list.
1592 *
1593 * The "__skb_xxxx()" functions are the non-atomic ones that
1594 * can only be called with interrupts disabled.
1595 */
1596void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1597 struct sk_buff_head *list);
1598static inline void __skb_insert(struct sk_buff *newsk,
1599 struct sk_buff *prev, struct sk_buff *next,
1600 struct sk_buff_head *list)
1601{
1602 newsk->next = next;
1603 newsk->prev = prev;
1604 next->prev = prev->next = newsk;
1605 list->qlen++;
1606}
1607
1608static inline void __skb_queue_splice(const struct sk_buff_head *list,
1609 struct sk_buff *prev,
1610 struct sk_buff *next)
1611{
1612 struct sk_buff *first = list->next;
1613 struct sk_buff *last = list->prev;
1614
1615 first->prev = prev;
1616 prev->next = first;
1617
1618 last->next = next;
1619 next->prev = last;
1620}
1621
1622/**
1623 * skb_queue_splice - join two skb lists, this is designed for stacks
1624 * @list: the new list to add
1625 * @head: the place to add it in the first list
1626 */
1627static inline void skb_queue_splice(const struct sk_buff_head *list,
1628 struct sk_buff_head *head)
1629{
1630 if (!skb_queue_empty(list)) {
1631 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1632 head->qlen += list->qlen;
1633 }
1634}
1635
1636/**
1637 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1638 * @list: the new list to add
1639 * @head: the place to add it in the first list
1640 *
1641 * The list at @list is reinitialised
1642 */
1643static inline void skb_queue_splice_init(struct sk_buff_head *list,
1644 struct sk_buff_head *head)
1645{
1646 if (!skb_queue_empty(list)) {
1647 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1648 head->qlen += list->qlen;
1649 __skb_queue_head_init(list);
1650 }
1651}
1652
1653/**
1654 * skb_queue_splice_tail - join two skb lists, each list being a queue
1655 * @list: the new list to add
1656 * @head: the place to add it in the first list
1657 */
1658static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1659 struct sk_buff_head *head)
1660{
1661 if (!skb_queue_empty(list)) {
1662 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1663 head->qlen += list->qlen;
1664 }
1665}
1666
1667/**
1668 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1669 * @list: the new list to add
1670 * @head: the place to add it in the first list
1671 *
1672 * Each of the lists is a queue.
1673 * The list at @list is reinitialised
1674 */
1675static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1676 struct sk_buff_head *head)
1677{
1678 if (!skb_queue_empty(list)) {
1679 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1680 head->qlen += list->qlen;
1681 __skb_queue_head_init(list);
1682 }
1683}
1684
1685/**
1686 * __skb_queue_after - queue a buffer at the list head
1687 * @list: list to use
1688 * @prev: place after this buffer
1689 * @newsk: buffer to queue
1690 *
1691 * Queue a buffer int the middle of a list. This function takes no locks
1692 * and you must therefore hold required locks before calling it.
1693 *
1694 * A buffer cannot be placed on two lists at the same time.
1695 */
1696static inline void __skb_queue_after(struct sk_buff_head *list,
1697 struct sk_buff *prev,
1698 struct sk_buff *newsk)
1699{
1700 __skb_insert(newsk, prev, prev->next, list);
1701}
1702
1703void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1704 struct sk_buff_head *list);
1705
1706static inline void __skb_queue_before(struct sk_buff_head *list,
1707 struct sk_buff *next,
1708 struct sk_buff *newsk)
1709{
1710 __skb_insert(newsk, next->prev, next, list);
1711}
1712
1713/**
1714 * __skb_queue_head - queue a buffer at the list head
1715 * @list: list to use
1716 * @newsk: buffer to queue
1717 *
1718 * Queue a buffer at the start of a list. This function takes no locks
1719 * and you must therefore hold required locks before calling it.
1720 *
1721 * A buffer cannot be placed on two lists at the same time.
1722 */
1723void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1724static inline void __skb_queue_head(struct sk_buff_head *list,
1725 struct sk_buff *newsk)
1726{
1727 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1728}
1729
1730/**
1731 * __skb_queue_tail - queue a buffer at the list tail
1732 * @list: list to use
1733 * @newsk: buffer to queue
1734 *
1735 * Queue a buffer at the end of a list. This function takes no locks
1736 * and you must therefore hold required locks before calling it.
1737 *
1738 * A buffer cannot be placed on two lists at the same time.
1739 */
1740void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1741static inline void __skb_queue_tail(struct sk_buff_head *list,
1742 struct sk_buff *newsk)
1743{
1744 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1745}
1746
1747/*
1748 * remove sk_buff from list. _Must_ be called atomically, and with
1749 * the list known..
1750 */
1751void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1752static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1753{
1754 struct sk_buff *next, *prev;
1755
1756 list->qlen--;
1757 next = skb->next;
1758 prev = skb->prev;
1759 skb->next = skb->prev = NULL;
1760 next->prev = prev;
1761 prev->next = next;
1762}
1763
1764/**
1765 * __skb_dequeue - remove from the head of the queue
1766 * @list: list to dequeue from
1767 *
1768 * Remove the head of the list. This function does not take any locks
1769 * so must be used with appropriate locks held only. The head item is
1770 * returned or %NULL if the list is empty.
1771 */
1772struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1773static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1774{
1775 struct sk_buff *skb = skb_peek(list);
1776 if (skb)
1777 __skb_unlink(skb, list);
1778 return skb;
1779}
1780
1781/**
1782 * __skb_dequeue_tail - remove from the tail of the queue
1783 * @list: list to dequeue from
1784 *
1785 * Remove the tail of the list. This function does not take any locks
1786 * so must be used with appropriate locks held only. The tail item is
1787 * returned or %NULL if the list is empty.
1788 */
1789struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1790static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1791{
1792 struct sk_buff *skb = skb_peek_tail(list);
1793 if (skb)
1794 __skb_unlink(skb, list);
1795 return skb;
1796}
1797
1798
1799static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1800{
1801 return skb->data_len;
1802}
1803
1804static inline unsigned int skb_headlen(const struct sk_buff *skb)
1805{
1806 return skb->len - skb->data_len;
1807}
1808
1809static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1810{
1811 unsigned int i, len = 0;
1812
1813 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1814 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1815 return len + skb_headlen(skb);
1816}
1817
1818/**
1819 * __skb_fill_page_desc - initialise a paged fragment in an skb
1820 * @skb: buffer containing fragment to be initialised
1821 * @i: paged fragment index to initialise
1822 * @page: the page to use for this fragment
1823 * @off: the offset to the data with @page
1824 * @size: the length of the data
1825 *
1826 * Initialises the @i'th fragment of @skb to point to &size bytes at
1827 * offset @off within @page.
1828 *
1829 * Does not take any additional reference on the fragment.
1830 */
1831static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1832 struct page *page, int off, int size)
1833{
1834 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1835
1836 /*
1837 * Propagate page pfmemalloc to the skb if we can. The problem is
1838 * that not all callers have unique ownership of the page but rely
1839 * on page_is_pfmemalloc doing the right thing(tm).
1840 */
1841 frag->page.p = page;
1842 frag->page_offset = off;
1843 skb_frag_size_set(frag, size);
1844
1845 page = compound_head(page);
1846 if (page_is_pfmemalloc(page))
1847 skb->pfmemalloc = true;
1848}
1849
1850/**
1851 * skb_fill_page_desc - initialise a paged fragment in an skb
1852 * @skb: buffer containing fragment to be initialised
1853 * @i: paged fragment index to initialise
1854 * @page: the page to use for this fragment
1855 * @off: the offset to the data with @page
1856 * @size: the length of the data
1857 *
1858 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1859 * @skb to point to @size bytes at offset @off within @page. In
1860 * addition updates @skb such that @i is the last fragment.
1861 *
1862 * Does not take any additional reference on the fragment.
1863 */
1864static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1865 struct page *page, int off, int size)
1866{
1867 __skb_fill_page_desc(skb, i, page, off, size);
1868 skb_shinfo(skb)->nr_frags = i + 1;
1869}
1870
1871void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1872 int size, unsigned int truesize);
1873
1874void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1875 unsigned int truesize);
1876
1877#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1878#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1879#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1880
1881#ifdef NET_SKBUFF_DATA_USES_OFFSET
1882static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1883{
1884 return skb->head + skb->tail;
1885}
1886
1887static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1888{
1889 skb->tail = skb->data - skb->head;
1890}
1891
1892static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1893{
1894 skb_reset_tail_pointer(skb);
1895 skb->tail += offset;
1896}
1897
1898#else /* NET_SKBUFF_DATA_USES_OFFSET */
1899static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1900{
1901 return skb->tail;
1902}
1903
1904static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1905{
1906 skb->tail = skb->data;
1907}
1908
1909static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1910{
1911 skb->tail = skb->data + offset;
1912}
1913
1914#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1915
1916/*
1917 * Add data to an sk_buff
1918 */
1919unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1920unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1921static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1922{
1923 unsigned char *tmp = skb_tail_pointer(skb);
1924 SKB_LINEAR_ASSERT(skb);
1925 skb->tail += len;
1926 skb->len += len;
1927 return tmp;
1928}
1929
1930unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1931static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1932{
1933 skb->data -= len;
1934 skb->len += len;
1935 return skb->data;
1936}
1937
1938unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1939static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1940{
1941 skb->len -= len;
1942 BUG_ON(skb->len < skb->data_len);
1943 return skb->data += len;
1944}
1945
1946static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1947{
1948 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1949}
1950
1951unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1952
1953static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1954{
1955 if (len > skb_headlen(skb) &&
1956 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1957 return NULL;
1958 skb->len -= len;
1959 return skb->data += len;
1960}
1961
1962static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1963{
1964 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1965}
1966
1967static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1968{
1969 if (likely(len <= skb_headlen(skb)))
1970 return 1;
1971 if (unlikely(len > skb->len))
1972 return 0;
1973 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1974}
1975
1976void skb_condense(struct sk_buff *skb);
1977
1978/**
1979 * skb_headroom - bytes at buffer head
1980 * @skb: buffer to check
1981 *
1982 * Return the number of bytes of free space at the head of an &sk_buff.
1983 */
1984static inline unsigned int skb_headroom(const struct sk_buff *skb)
1985{
1986 return skb->data - skb->head;
1987}
1988
1989/**
1990 * skb_tailroom - bytes at buffer end
1991 * @skb: buffer to check
1992 *
1993 * Return the number of bytes of free space at the tail of an sk_buff
1994 */
1995static inline int skb_tailroom(const struct sk_buff *skb)
1996{
1997 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1998}
1999
2000/**
2001 * skb_availroom - bytes at buffer end
2002 * @skb: buffer to check
2003 *
2004 * Return the number of bytes of free space at the tail of an sk_buff
2005 * allocated by sk_stream_alloc()
2006 */
2007static inline int skb_availroom(const struct sk_buff *skb)
2008{
2009 if (skb_is_nonlinear(skb))
2010 return 0;
2011
2012 return skb->end - skb->tail - skb->reserved_tailroom;
2013}
2014
2015/**
2016 * skb_reserve - adjust headroom
2017 * @skb: buffer to alter
2018 * @len: bytes to move
2019 *
2020 * Increase the headroom of an empty &sk_buff by reducing the tail
2021 * room. This is only allowed for an empty buffer.
2022 */
2023static inline void skb_reserve(struct sk_buff *skb, int len)
2024{
2025 skb->data += len;
2026 skb->tail += len;
2027}
2028
2029/**
2030 * skb_tailroom_reserve - adjust reserved_tailroom
2031 * @skb: buffer to alter
2032 * @mtu: maximum amount of headlen permitted
2033 * @needed_tailroom: minimum amount of reserved_tailroom
2034 *
2035 * Set reserved_tailroom so that headlen can be as large as possible but
2036 * not larger than mtu and tailroom cannot be smaller than
2037 * needed_tailroom.
2038 * The required headroom should already have been reserved before using
2039 * this function.
2040 */
2041static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2042 unsigned int needed_tailroom)
2043{
2044 SKB_LINEAR_ASSERT(skb);
2045 if (mtu < skb_tailroom(skb) - needed_tailroom)
2046 /* use at most mtu */
2047 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2048 else
2049 /* use up to all available space */
2050 skb->reserved_tailroom = needed_tailroom;
2051}
2052
2053#define ENCAP_TYPE_ETHER 0
2054#define ENCAP_TYPE_IPPROTO 1
2055
2056static inline void skb_set_inner_protocol(struct sk_buff *skb,
2057 __be16 protocol)
2058{
2059 skb->inner_protocol = protocol;
2060 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2061}
2062
2063static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2064 __u8 ipproto)
2065{
2066 skb->inner_ipproto = ipproto;
2067 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2068}
2069
2070static inline void skb_reset_inner_headers(struct sk_buff *skb)
2071{
2072 skb->inner_mac_header = skb->mac_header;
2073 skb->inner_network_header = skb->network_header;
2074 skb->inner_transport_header = skb->transport_header;
2075}
2076
2077static inline void skb_reset_mac_len(struct sk_buff *skb)
2078{
2079 skb->mac_len = skb->network_header - skb->mac_header;
2080}
2081
2082static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2083 *skb)
2084{
2085 return skb->head + skb->inner_transport_header;
2086}
2087
2088static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2089{
2090 return skb_inner_transport_header(skb) - skb->data;
2091}
2092
2093static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2094{
2095 skb->inner_transport_header = skb->data - skb->head;
2096}
2097
2098static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2099 const int offset)
2100{
2101 skb_reset_inner_transport_header(skb);
2102 skb->inner_transport_header += offset;
2103}
2104
2105static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2106{
2107 return skb->head + skb->inner_network_header;
2108}
2109
2110static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2111{
2112 skb->inner_network_header = skb->data - skb->head;
2113}
2114
2115static inline void skb_set_inner_network_header(struct sk_buff *skb,
2116 const int offset)
2117{
2118 skb_reset_inner_network_header(skb);
2119 skb->inner_network_header += offset;
2120}
2121
2122static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2123{
2124 return skb->head + skb->inner_mac_header;
2125}
2126
2127static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2128{
2129 skb->inner_mac_header = skb->data - skb->head;
2130}
2131
2132static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2133 const int offset)
2134{
2135 skb_reset_inner_mac_header(skb);
2136 skb->inner_mac_header += offset;
2137}
2138static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2139{
2140 return skb->transport_header != (typeof(skb->transport_header))~0U;
2141}
2142
2143static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2144{
2145 return skb->head + skb->transport_header;
2146}
2147
2148static inline void skb_reset_transport_header(struct sk_buff *skb)
2149{
2150 skb->transport_header = skb->data - skb->head;
2151}
2152
2153static inline void skb_set_transport_header(struct sk_buff *skb,
2154 const int offset)
2155{
2156 skb_reset_transport_header(skb);
2157 skb->transport_header += offset;
2158}
2159
2160static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2161{
2162 return skb->head + skb->network_header;
2163}
2164
2165static inline void skb_reset_network_header(struct sk_buff *skb)
2166{
2167 skb->network_header = skb->data - skb->head;
2168}
2169
2170static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2171{
2172 skb_reset_network_header(skb);
2173 skb->network_header += offset;
2174}
2175
2176static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2177{
2178 return skb->head + skb->mac_header;
2179}
2180
2181static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2182{
2183 return skb->mac_header != (typeof(skb->mac_header))~0U;
2184}
2185
2186static inline void skb_reset_mac_header(struct sk_buff *skb)
2187{
2188 skb->mac_header = skb->data - skb->head;
2189}
2190
2191static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2192{
2193 skb_reset_mac_header(skb);
2194 skb->mac_header += offset;
2195}
2196
2197static inline void skb_pop_mac_header(struct sk_buff *skb)
2198{
2199 skb->mac_header = skb->network_header;
2200}
2201
2202static inline void skb_probe_transport_header(struct sk_buff *skb,
2203 const int offset_hint)
2204{
2205 struct flow_keys keys;
2206
2207 if (skb_transport_header_was_set(skb))
2208 return;
2209 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2210 skb_set_transport_header(skb, keys.control.thoff);
2211 else
2212 skb_set_transport_header(skb, offset_hint);
2213}
2214
2215static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2216{
2217 if (skb_mac_header_was_set(skb)) {
2218 const unsigned char *old_mac = skb_mac_header(skb);
2219
2220 skb_set_mac_header(skb, -skb->mac_len);
2221 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2222 }
2223}
2224
2225static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2226{
2227 return skb->csum_start - skb_headroom(skb);
2228}
2229
2230static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2231{
2232 return skb->head + skb->csum_start;
2233}
2234
2235static inline int skb_transport_offset(const struct sk_buff *skb)
2236{
2237 return skb_transport_header(skb) - skb->data;
2238}
2239
2240static inline u32 skb_network_header_len(const struct sk_buff *skb)
2241{
2242 return skb->transport_header - skb->network_header;
2243}
2244
2245static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2246{
2247 return skb->inner_transport_header - skb->inner_network_header;
2248}
2249
2250static inline int skb_network_offset(const struct sk_buff *skb)
2251{
2252 return skb_network_header(skb) - skb->data;
2253}
2254
2255static inline int skb_inner_network_offset(const struct sk_buff *skb)
2256{
2257 return skb_inner_network_header(skb) - skb->data;
2258}
2259
2260static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2261{
2262 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2263}
2264
2265/*
2266 * CPUs often take a performance hit when accessing unaligned memory
2267 * locations. The actual performance hit varies, it can be small if the
2268 * hardware handles it or large if we have to take an exception and fix it
2269 * in software.
2270 *
2271 * Since an ethernet header is 14 bytes network drivers often end up with
2272 * the IP header at an unaligned offset. The IP header can be aligned by
2273 * shifting the start of the packet by 2 bytes. Drivers should do this
2274 * with:
2275 *
2276 * skb_reserve(skb, NET_IP_ALIGN);
2277 *
2278 * The downside to this alignment of the IP header is that the DMA is now
2279 * unaligned. On some architectures the cost of an unaligned DMA is high
2280 * and this cost outweighs the gains made by aligning the IP header.
2281 *
2282 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2283 * to be overridden.
2284 */
2285#ifndef NET_IP_ALIGN
2286#define NET_IP_ALIGN 2
2287#endif
2288
2289/*
2290 * The networking layer reserves some headroom in skb data (via
2291 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2292 * the header has to grow. In the default case, if the header has to grow
2293 * 32 bytes or less we avoid the reallocation.
2294 *
2295 * Unfortunately this headroom changes the DMA alignment of the resulting
2296 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2297 * on some architectures. An architecture can override this value,
2298 * perhaps setting it to a cacheline in size (since that will maintain
2299 * cacheline alignment of the DMA). It must be a power of 2.
2300 *
2301 * Various parts of the networking layer expect at least 32 bytes of
2302 * headroom, you should not reduce this.
2303 *
2304 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2305 * to reduce average number of cache lines per packet.
2306 * get_rps_cpus() for example only access one 64 bytes aligned block :
2307 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2308 */
2309#ifndef NET_SKB_PAD
2310#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2311#endif
2312
2313int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2314
2315static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2316{
2317 if (unlikely(skb_is_nonlinear(skb))) {
2318 WARN_ON(1);
2319 return;
2320 }
2321 skb->len = len;
2322 skb_set_tail_pointer(skb, len);
2323}
2324
2325static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2326{
2327 __skb_set_length(skb, len);
2328}
2329
2330void skb_trim(struct sk_buff *skb, unsigned int len);
2331
2332static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2333{
2334 if (skb->data_len)
2335 return ___pskb_trim(skb, len);
2336 __skb_trim(skb, len);
2337 return 0;
2338}
2339
2340static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2341{
2342 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2343}
2344
2345/**
2346 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2347 * @skb: buffer to alter
2348 * @len: new length
2349 *
2350 * This is identical to pskb_trim except that the caller knows that
2351 * the skb is not cloned so we should never get an error due to out-
2352 * of-memory.
2353 */
2354static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2355{
2356 int err = pskb_trim(skb, len);
2357 BUG_ON(err);
2358}
2359
2360static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2361{
2362 unsigned int diff = len - skb->len;
2363
2364 if (skb_tailroom(skb) < diff) {
2365 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2366 GFP_ATOMIC);
2367 if (ret)
2368 return ret;
2369 }
2370 __skb_set_length(skb, len);
2371 return 0;
2372}
2373
2374/**
2375 * skb_orphan - orphan a buffer
2376 * @skb: buffer to orphan
2377 *
2378 * If a buffer currently has an owner then we call the owner's
2379 * destructor function and make the @skb unowned. The buffer continues
2380 * to exist but is no longer charged to its former owner.
2381 */
2382static inline void skb_orphan(struct sk_buff *skb)
2383{
2384 if (skb->destructor) {
2385 skb->destructor(skb);
2386 skb->destructor = NULL;
2387 skb->sk = NULL;
2388 } else {
2389 BUG_ON(skb->sk);
2390 }
2391}
2392
2393/**
2394 * skb_orphan_frags - orphan the frags contained in a buffer
2395 * @skb: buffer to orphan frags from
2396 * @gfp_mask: allocation mask for replacement pages
2397 *
2398 * For each frag in the SKB which needs a destructor (i.e. has an
2399 * owner) create a copy of that frag and release the original
2400 * page by calling the destructor.
2401 */
2402static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2403{
2404 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2405 return 0;
2406 return skb_copy_ubufs(skb, gfp_mask);
2407}
2408
2409/**
2410 * __skb_queue_purge - empty a list
2411 * @list: list to empty
2412 *
2413 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2414 * the list and one reference dropped. This function does not take the
2415 * list lock and the caller must hold the relevant locks to use it.
2416 */
2417void skb_queue_purge(struct sk_buff_head *list);
2418static inline void __skb_queue_purge(struct sk_buff_head *list)
2419{
2420 struct sk_buff *skb;
2421 while ((skb = __skb_dequeue(list)) != NULL)
2422 kfree_skb(skb);
2423}
2424
2425void skb_rbtree_purge(struct rb_root *root);
2426
2427void *netdev_alloc_frag(unsigned int fragsz);
2428
2429struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2430 gfp_t gfp_mask);
2431
2432/**
2433 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2434 * @dev: network device to receive on
2435 * @length: length to allocate
2436 *
2437 * Allocate a new &sk_buff and assign it a usage count of one. The
2438 * buffer has unspecified headroom built in. Users should allocate
2439 * the headroom they think they need without accounting for the
2440 * built in space. The built in space is used for optimisations.
2441 *
2442 * %NULL is returned if there is no free memory. Although this function
2443 * allocates memory it can be called from an interrupt.
2444 */
2445static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2446 unsigned int length)
2447{
2448 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2449}
2450
2451/* legacy helper around __netdev_alloc_skb() */
2452static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2453 gfp_t gfp_mask)
2454{
2455 return __netdev_alloc_skb(NULL, length, gfp_mask);
2456}
2457
2458/* legacy helper around netdev_alloc_skb() */
2459static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2460{
2461 return netdev_alloc_skb(NULL, length);
2462}
2463
2464
2465static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2466 unsigned int length, gfp_t gfp)
2467{
2468 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2469
2470 if (NET_IP_ALIGN && skb)
2471 skb_reserve(skb, NET_IP_ALIGN);
2472 return skb;
2473}
2474
2475static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2476 unsigned int length)
2477{
2478 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2479}
2480
2481static inline void skb_free_frag(void *addr)
2482{
2483 page_frag_free(addr);
2484}
2485
2486void *napi_alloc_frag(unsigned int fragsz);
2487struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2488 unsigned int length, gfp_t gfp_mask);
2489static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2490 unsigned int length)
2491{
2492 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2493}
2494void napi_consume_skb(struct sk_buff *skb, int budget);
2495
2496void __kfree_skb_flush(void);
2497void __kfree_skb_defer(struct sk_buff *skb);
2498
2499/**
2500 * __dev_alloc_pages - allocate page for network Rx
2501 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2502 * @order: size of the allocation
2503 *
2504 * Allocate a new page.
2505 *
2506 * %NULL is returned if there is no free memory.
2507*/
2508static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2509 unsigned int order)
2510{
2511 /* This piece of code contains several assumptions.
2512 * 1. This is for device Rx, therefor a cold page is preferred.
2513 * 2. The expectation is the user wants a compound page.
2514 * 3. If requesting a order 0 page it will not be compound
2515 * due to the check to see if order has a value in prep_new_page
2516 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2517 * code in gfp_to_alloc_flags that should be enforcing this.
2518 */
2519 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2520
2521 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2522}
2523
2524static inline struct page *dev_alloc_pages(unsigned int order)
2525{
2526 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2527}
2528
2529/**
2530 * __dev_alloc_page - allocate a page for network Rx
2531 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2532 *
2533 * Allocate a new page.
2534 *
2535 * %NULL is returned if there is no free memory.
2536 */
2537static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2538{
2539 return __dev_alloc_pages(gfp_mask, 0);
2540}
2541
2542static inline struct page *dev_alloc_page(void)
2543{
2544 return dev_alloc_pages(0);
2545}
2546
2547/**
2548 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2549 * @page: The page that was allocated from skb_alloc_page
2550 * @skb: The skb that may need pfmemalloc set
2551 */
2552static inline void skb_propagate_pfmemalloc(struct page *page,
2553 struct sk_buff *skb)
2554{
2555 if (page_is_pfmemalloc(page))
2556 skb->pfmemalloc = true;
2557}
2558
2559/**
2560 * skb_frag_page - retrieve the page referred to by a paged fragment
2561 * @frag: the paged fragment
2562 *
2563 * Returns the &struct page associated with @frag.
2564 */
2565static inline struct page *skb_frag_page(const skb_frag_t *frag)
2566{
2567 return frag->page.p;
2568}
2569
2570/**
2571 * __skb_frag_ref - take an addition reference on a paged fragment.
2572 * @frag: the paged fragment
2573 *
2574 * Takes an additional reference on the paged fragment @frag.
2575 */
2576static inline void __skb_frag_ref(skb_frag_t *frag)
2577{
2578 get_page(skb_frag_page(frag));
2579}
2580
2581/**
2582 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2583 * @skb: the buffer
2584 * @f: the fragment offset.
2585 *
2586 * Takes an additional reference on the @f'th paged fragment of @skb.
2587 */
2588static inline void skb_frag_ref(struct sk_buff *skb, int f)
2589{
2590 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2591}
2592
2593/**
2594 * __skb_frag_unref - release a reference on a paged fragment.
2595 * @frag: the paged fragment
2596 *
2597 * Releases a reference on the paged fragment @frag.
2598 */
2599static inline void __skb_frag_unref(skb_frag_t *frag)
2600{
2601 put_page(skb_frag_page(frag));
2602}
2603
2604/**
2605 * skb_frag_unref - release a reference on a paged fragment of an skb.
2606 * @skb: the buffer
2607 * @f: the fragment offset
2608 *
2609 * Releases a reference on the @f'th paged fragment of @skb.
2610 */
2611static inline void skb_frag_unref(struct sk_buff *skb, int f)
2612{
2613 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2614}
2615
2616/**
2617 * skb_frag_address - gets the address of the data contained in a paged fragment
2618 * @frag: the paged fragment buffer
2619 *
2620 * Returns the address of the data within @frag. The page must already
2621 * be mapped.
2622 */
2623static inline void *skb_frag_address(const skb_frag_t *frag)
2624{
2625 return page_address(skb_frag_page(frag)) + frag->page_offset;
2626}
2627
2628/**
2629 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2630 * @frag: the paged fragment buffer
2631 *
2632 * Returns the address of the data within @frag. Checks that the page
2633 * is mapped and returns %NULL otherwise.
2634 */
2635static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2636{
2637 void *ptr = page_address(skb_frag_page(frag));
2638 if (unlikely(!ptr))
2639 return NULL;
2640
2641 return ptr + frag->page_offset;
2642}
2643
2644/**
2645 * __skb_frag_set_page - sets the page contained in a paged fragment
2646 * @frag: the paged fragment
2647 * @page: the page to set
2648 *
2649 * Sets the fragment @frag to contain @page.
2650 */
2651static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2652{
2653 frag->page.p = page;
2654}
2655
2656/**
2657 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2658 * @skb: the buffer
2659 * @f: the fragment offset
2660 * @page: the page to set
2661 *
2662 * Sets the @f'th fragment of @skb to contain @page.
2663 */
2664static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2665 struct page *page)
2666{
2667 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2668}
2669
2670bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2671
2672/**
2673 * skb_frag_dma_map - maps a paged fragment via the DMA API
2674 * @dev: the device to map the fragment to
2675 * @frag: the paged fragment to map
2676 * @offset: the offset within the fragment (starting at the
2677 * fragment's own offset)
2678 * @size: the number of bytes to map
2679 * @dir: the direction of the mapping (%PCI_DMA_*)
2680 *
2681 * Maps the page associated with @frag to @device.
2682 */
2683static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2684 const skb_frag_t *frag,
2685 size_t offset, size_t size,
2686 enum dma_data_direction dir)
2687{
2688 return dma_map_page(dev, skb_frag_page(frag),
2689 frag->page_offset + offset, size, dir);
2690}
2691
2692static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2693 gfp_t gfp_mask)
2694{
2695 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2696}
2697
2698
2699static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2700 gfp_t gfp_mask)
2701{
2702 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2703}
2704
2705
2706/**
2707 * skb_clone_writable - is the header of a clone writable
2708 * @skb: buffer to check
2709 * @len: length up to which to write
2710 *
2711 * Returns true if modifying the header part of the cloned buffer
2712 * does not requires the data to be copied.
2713 */
2714static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2715{
2716 return !skb_header_cloned(skb) &&
2717 skb_headroom(skb) + len <= skb->hdr_len;
2718}
2719
2720static inline int skb_try_make_writable(struct sk_buff *skb,
2721 unsigned int write_len)
2722{
2723 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2724 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2725}
2726
2727static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2728 int cloned)
2729{
2730 int delta = 0;
2731
2732 if (headroom > skb_headroom(skb))
2733 delta = headroom - skb_headroom(skb);
2734
2735 if (delta || cloned)
2736 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2737 GFP_ATOMIC);
2738 return 0;
2739}
2740
2741/**
2742 * skb_cow - copy header of skb when it is required
2743 * @skb: buffer to cow
2744 * @headroom: needed headroom
2745 *
2746 * If the skb passed lacks sufficient headroom or its data part
2747 * is shared, data is reallocated. If reallocation fails, an error
2748 * is returned and original skb is not changed.
2749 *
2750 * The result is skb with writable area skb->head...skb->tail
2751 * and at least @headroom of space at head.
2752 */
2753static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2754{
2755 return __skb_cow(skb, headroom, skb_cloned(skb));
2756}
2757
2758/**
2759 * skb_cow_head - skb_cow but only making the head writable
2760 * @skb: buffer to cow
2761 * @headroom: needed headroom
2762 *
2763 * This function is identical to skb_cow except that we replace the
2764 * skb_cloned check by skb_header_cloned. It should be used when
2765 * you only need to push on some header and do not need to modify
2766 * the data.
2767 */
2768static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2769{
2770 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2771}
2772
2773/**
2774 * skb_padto - pad an skbuff up to a minimal size
2775 * @skb: buffer to pad
2776 * @len: minimal length
2777 *
2778 * Pads up a buffer to ensure the trailing bytes exist and are
2779 * blanked. If the buffer already contains sufficient data it
2780 * is untouched. Otherwise it is extended. Returns zero on
2781 * success. The skb is freed on error.
2782 */
2783static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2784{
2785 unsigned int size = skb->len;
2786 if (likely(size >= len))
2787 return 0;
2788 return skb_pad(skb, len - size);
2789}
2790
2791/**
2792 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2793 * @skb: buffer to pad
2794 * @len: minimal length
2795 *
2796 * Pads up a buffer to ensure the trailing bytes exist and are
2797 * blanked. If the buffer already contains sufficient data it
2798 * is untouched. Otherwise it is extended. Returns zero on
2799 * success. The skb is freed on error.
2800 */
2801static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2802{
2803 unsigned int size = skb->len;
2804
2805 if (unlikely(size < len)) {
2806 len -= size;
2807 if (skb_pad(skb, len))
2808 return -ENOMEM;
2809 __skb_put(skb, len);
2810 }
2811 return 0;
2812}
2813
2814static inline int skb_add_data(struct sk_buff *skb,
2815 struct iov_iter *from, int copy)
2816{
2817 const int off = skb->len;
2818
2819 if (skb->ip_summed == CHECKSUM_NONE) {
2820 __wsum csum = 0;
2821 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2822 &csum, from)) {
2823 skb->csum = csum_block_add(skb->csum, csum, off);
2824 return 0;
2825 }
2826 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2827 return 0;
2828
2829 __skb_trim(skb, off);
2830 return -EFAULT;
2831}
2832
2833static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2834 const struct page *page, int off)
2835{
2836 if (i) {
2837 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2838
2839 return page == skb_frag_page(frag) &&
2840 off == frag->page_offset + skb_frag_size(frag);
2841 }
2842 return false;
2843}
2844
2845static inline int __skb_linearize(struct sk_buff *skb)
2846{
2847 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2848}
2849
2850/**
2851 * skb_linearize - convert paged skb to linear one
2852 * @skb: buffer to linarize
2853 *
2854 * If there is no free memory -ENOMEM is returned, otherwise zero
2855 * is returned and the old skb data released.
2856 */
2857static inline int skb_linearize(struct sk_buff *skb)
2858{
2859 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2860}
2861
2862/**
2863 * skb_has_shared_frag - can any frag be overwritten
2864 * @skb: buffer to test
2865 *
2866 * Return true if the skb has at least one frag that might be modified
2867 * by an external entity (as in vmsplice()/sendfile())
2868 */
2869static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2870{
2871 return skb_is_nonlinear(skb) &&
2872 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2873}
2874
2875/**
2876 * skb_linearize_cow - make sure skb is linear and writable
2877 * @skb: buffer to process
2878 *
2879 * If there is no free memory -ENOMEM is returned, otherwise zero
2880 * is returned and the old skb data released.
2881 */
2882static inline int skb_linearize_cow(struct sk_buff *skb)
2883{
2884 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2885 __skb_linearize(skb) : 0;
2886}
2887
2888static __always_inline void
2889__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2890 unsigned int off)
2891{
2892 if (skb->ip_summed == CHECKSUM_COMPLETE)
2893 skb->csum = csum_block_sub(skb->csum,
2894 csum_partial(start, len, 0), off);
2895 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2896 skb_checksum_start_offset(skb) < 0)
2897 skb->ip_summed = CHECKSUM_NONE;
2898}
2899
2900/**
2901 * skb_postpull_rcsum - update checksum for received skb after pull
2902 * @skb: buffer to update
2903 * @start: start of data before pull
2904 * @len: length of data pulled
2905 *
2906 * After doing a pull on a received packet, you need to call this to
2907 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2908 * CHECKSUM_NONE so that it can be recomputed from scratch.
2909 */
2910static inline void skb_postpull_rcsum(struct sk_buff *skb,
2911 const void *start, unsigned int len)
2912{
2913 __skb_postpull_rcsum(skb, start, len, 0);
2914}
2915
2916static __always_inline void
2917__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2918 unsigned int off)
2919{
2920 if (skb->ip_summed == CHECKSUM_COMPLETE)
2921 skb->csum = csum_block_add(skb->csum,
2922 csum_partial(start, len, 0), off);
2923}
2924
2925/**
2926 * skb_postpush_rcsum - update checksum for received skb after push
2927 * @skb: buffer to update
2928 * @start: start of data after push
2929 * @len: length of data pushed
2930 *
2931 * After doing a push on a received packet, you need to call this to
2932 * update the CHECKSUM_COMPLETE checksum.
2933 */
2934static inline void skb_postpush_rcsum(struct sk_buff *skb,
2935 const void *start, unsigned int len)
2936{
2937 __skb_postpush_rcsum(skb, start, len, 0);
2938}
2939
2940unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2941
2942/**
2943 * skb_push_rcsum - push skb and update receive checksum
2944 * @skb: buffer to update
2945 * @len: length of data pulled
2946 *
2947 * This function performs an skb_push on the packet and updates
2948 * the CHECKSUM_COMPLETE checksum. It should be used on
2949 * receive path processing instead of skb_push unless you know
2950 * that the checksum difference is zero (e.g., a valid IP header)
2951 * or you are setting ip_summed to CHECKSUM_NONE.
2952 */
2953static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2954 unsigned int len)
2955{
2956 skb_push(skb, len);
2957 skb_postpush_rcsum(skb, skb->data, len);
2958 return skb->data;
2959}
2960
2961/**
2962 * pskb_trim_rcsum - trim received skb and update checksum
2963 * @skb: buffer to trim
2964 * @len: new length
2965 *
2966 * This is exactly the same as pskb_trim except that it ensures the
2967 * checksum of received packets are still valid after the operation.
2968 */
2969
2970static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2971{
2972 if (likely(len >= skb->len))
2973 return 0;
2974 if (skb->ip_summed == CHECKSUM_COMPLETE)
2975 skb->ip_summed = CHECKSUM_NONE;
2976 return __pskb_trim(skb, len);
2977}
2978
2979static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2980{
2981 if (skb->ip_summed == CHECKSUM_COMPLETE)
2982 skb->ip_summed = CHECKSUM_NONE;
2983 __skb_trim(skb, len);
2984 return 0;
2985}
2986
2987static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
2988{
2989 if (skb->ip_summed == CHECKSUM_COMPLETE)
2990 skb->ip_summed = CHECKSUM_NONE;
2991 return __skb_grow(skb, len);
2992}
2993
2994#define skb_queue_walk(queue, skb) \
2995 for (skb = (queue)->next; \
2996 skb != (struct sk_buff *)(queue); \
2997 skb = skb->next)
2998
2999#define skb_queue_walk_safe(queue, skb, tmp) \
3000 for (skb = (queue)->next, tmp = skb->next; \
3001 skb != (struct sk_buff *)(queue); \
3002 skb = tmp, tmp = skb->next)
3003
3004#define skb_queue_walk_from(queue, skb) \
3005 for (; skb != (struct sk_buff *)(queue); \
3006 skb = skb->next)
3007
3008#define skb_queue_walk_from_safe(queue, skb, tmp) \
3009 for (tmp = skb->next; \
3010 skb != (struct sk_buff *)(queue); \
3011 skb = tmp, tmp = skb->next)
3012
3013#define skb_queue_reverse_walk(queue, skb) \
3014 for (skb = (queue)->prev; \
3015 skb != (struct sk_buff *)(queue); \
3016 skb = skb->prev)
3017
3018#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3019 for (skb = (queue)->prev, tmp = skb->prev; \
3020 skb != (struct sk_buff *)(queue); \
3021 skb = tmp, tmp = skb->prev)
3022
3023#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3024 for (tmp = skb->prev; \
3025 skb != (struct sk_buff *)(queue); \
3026 skb = tmp, tmp = skb->prev)
3027
3028static inline bool skb_has_frag_list(const struct sk_buff *skb)
3029{
3030 return skb_shinfo(skb)->frag_list != NULL;
3031}
3032
3033static inline void skb_frag_list_init(struct sk_buff *skb)
3034{
3035 skb_shinfo(skb)->frag_list = NULL;
3036}
3037
3038#define skb_walk_frags(skb, iter) \
3039 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3040
3041
3042int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3043 const struct sk_buff *skb);
3044struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3045 void (*destructor)(struct sock *sk,
3046 struct sk_buff *skb),
3047 int *peeked, int *off, int *err,
3048 struct sk_buff **last);
3049struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3050 void (*destructor)(struct sock *sk,
3051 struct sk_buff *skb),
3052 int *peeked, int *off, int *err);
3053struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3054 int *err);
3055unsigned int datagram_poll(struct file *file, struct socket *sock,
3056 struct poll_table_struct *wait);
3057int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3058 struct iov_iter *to, int size);
3059static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3060 struct msghdr *msg, int size)
3061{
3062 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3063}
3064int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3065 struct msghdr *msg);
3066int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3067 struct iov_iter *from, int len);
3068int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3069void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3070void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3071static inline void skb_free_datagram_locked(struct sock *sk,
3072 struct sk_buff *skb)
3073{
3074 __skb_free_datagram_locked(sk, skb, 0);
3075}
3076int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3077int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3078int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3079__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3080 int len, __wsum csum);
3081int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3082 struct pipe_inode_info *pipe, unsigned int len,
3083 unsigned int flags);
3084void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3085unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3086int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3087 int len, int hlen);
3088void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3089int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3090void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3091unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3092bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3093struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3094struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3095int skb_ensure_writable(struct sk_buff *skb, int write_len);
3096int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3097int skb_vlan_pop(struct sk_buff *skb);
3098int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3099struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3100 gfp_t gfp);
3101
3102static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3103{
3104 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3105}
3106
3107static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3108{
3109 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3110}
3111
3112struct skb_checksum_ops {
3113 __wsum (*update)(const void *mem, int len, __wsum wsum);
3114 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3115};
3116
3117__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3118 __wsum csum, const struct skb_checksum_ops *ops);
3119__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3120 __wsum csum);
3121
3122static inline void * __must_check
3123__skb_header_pointer(const struct sk_buff *skb, int offset,
3124 int len, void *data, int hlen, void *buffer)
3125{
3126 if (hlen - offset >= len)
3127 return data + offset;
3128
3129 if (!skb ||
3130 skb_copy_bits(skb, offset, buffer, len) < 0)
3131 return NULL;
3132
3133 return buffer;
3134}
3135
3136static inline void * __must_check
3137skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3138{
3139 return __skb_header_pointer(skb, offset, len, skb->data,
3140 skb_headlen(skb), buffer);
3141}
3142
3143/**
3144 * skb_needs_linearize - check if we need to linearize a given skb
3145 * depending on the given device features.
3146 * @skb: socket buffer to check
3147 * @features: net device features
3148 *
3149 * Returns true if either:
3150 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3151 * 2. skb is fragmented and the device does not support SG.
3152 */
3153static inline bool skb_needs_linearize(struct sk_buff *skb,
3154 netdev_features_t features)
3155{
3156 return skb_is_nonlinear(skb) &&
3157 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3158 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3159}
3160
3161static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3162 void *to,
3163 const unsigned int len)
3164{
3165 memcpy(to, skb->data, len);
3166}
3167
3168static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3169 const int offset, void *to,
3170 const unsigned int len)
3171{
3172 memcpy(to, skb->data + offset, len);
3173}
3174
3175static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3176 const void *from,
3177 const unsigned int len)
3178{
3179 memcpy(skb->data, from, len);
3180}
3181
3182static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3183 const int offset,
3184 const void *from,
3185 const unsigned int len)
3186{
3187 memcpy(skb->data + offset, from, len);
3188}
3189
3190void skb_init(void);
3191
3192static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3193{
3194 return skb->tstamp;
3195}
3196
3197/**
3198 * skb_get_timestamp - get timestamp from a skb
3199 * @skb: skb to get stamp from
3200 * @stamp: pointer to struct timeval to store stamp in
3201 *
3202 * Timestamps are stored in the skb as offsets to a base timestamp.
3203 * This function converts the offset back to a struct timeval and stores
3204 * it in stamp.
3205 */
3206static inline void skb_get_timestamp(const struct sk_buff *skb,
3207 struct timeval *stamp)
3208{
3209 *stamp = ktime_to_timeval(skb->tstamp);
3210}
3211
3212static inline void skb_get_timestampns(const struct sk_buff *skb,
3213 struct timespec *stamp)
3214{
3215 *stamp = ktime_to_timespec(skb->tstamp);
3216}
3217
3218static inline void __net_timestamp(struct sk_buff *skb)
3219{
3220 skb->tstamp = ktime_get_real();
3221}
3222
3223static inline ktime_t net_timedelta(ktime_t t)
3224{
3225 return ktime_sub(ktime_get_real(), t);
3226}
3227
3228static inline ktime_t net_invalid_timestamp(void)
3229{
3230 return 0;
3231}
3232
3233struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3234
3235#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3236
3237void skb_clone_tx_timestamp(struct sk_buff *skb);
3238bool skb_defer_rx_timestamp(struct sk_buff *skb);
3239
3240#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3241
3242static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3243{
3244}
3245
3246static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3247{
3248 return false;
3249}
3250
3251#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3252
3253/**
3254 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3255 *
3256 * PHY drivers may accept clones of transmitted packets for
3257 * timestamping via their phy_driver.txtstamp method. These drivers
3258 * must call this function to return the skb back to the stack with a
3259 * timestamp.
3260 *
3261 * @skb: clone of the the original outgoing packet
3262 * @hwtstamps: hardware time stamps
3263 *
3264 */
3265void skb_complete_tx_timestamp(struct sk_buff *skb,
3266 struct skb_shared_hwtstamps *hwtstamps);
3267
3268void __skb_tstamp_tx(struct sk_buff *orig_skb,
3269 struct skb_shared_hwtstamps *hwtstamps,
3270 struct sock *sk, int tstype);
3271
3272/**
3273 * skb_tstamp_tx - queue clone of skb with send time stamps
3274 * @orig_skb: the original outgoing packet
3275 * @hwtstamps: hardware time stamps, may be NULL if not available
3276 *
3277 * If the skb has a socket associated, then this function clones the
3278 * skb (thus sharing the actual data and optional structures), stores
3279 * the optional hardware time stamping information (if non NULL) or
3280 * generates a software time stamp (otherwise), then queues the clone
3281 * to the error queue of the socket. Errors are silently ignored.
3282 */
3283void skb_tstamp_tx(struct sk_buff *orig_skb,
3284 struct skb_shared_hwtstamps *hwtstamps);
3285
3286static inline void sw_tx_timestamp(struct sk_buff *skb)
3287{
3288 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3289 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3290 skb_tstamp_tx(skb, NULL);
3291}
3292
3293/**
3294 * skb_tx_timestamp() - Driver hook for transmit timestamping
3295 *
3296 * Ethernet MAC Drivers should call this function in their hard_xmit()
3297 * function immediately before giving the sk_buff to the MAC hardware.
3298 *
3299 * Specifically, one should make absolutely sure that this function is
3300 * called before TX completion of this packet can trigger. Otherwise
3301 * the packet could potentially already be freed.
3302 *
3303 * @skb: A socket buffer.
3304 */
3305static inline void skb_tx_timestamp(struct sk_buff *skb)
3306{
3307 skb_clone_tx_timestamp(skb);
3308 sw_tx_timestamp(skb);
3309}
3310
3311/**
3312 * skb_complete_wifi_ack - deliver skb with wifi status
3313 *
3314 * @skb: the original outgoing packet
3315 * @acked: ack status
3316 *
3317 */
3318void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3319
3320__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3321__sum16 __skb_checksum_complete(struct sk_buff *skb);
3322
3323static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3324{
3325 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3326 skb->csum_valid ||
3327 (skb->ip_summed == CHECKSUM_PARTIAL &&
3328 skb_checksum_start_offset(skb) >= 0));
3329}
3330
3331/**
3332 * skb_checksum_complete - Calculate checksum of an entire packet
3333 * @skb: packet to process
3334 *
3335 * This function calculates the checksum over the entire packet plus
3336 * the value of skb->csum. The latter can be used to supply the
3337 * checksum of a pseudo header as used by TCP/UDP. It returns the
3338 * checksum.
3339 *
3340 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3341 * this function can be used to verify that checksum on received
3342 * packets. In that case the function should return zero if the
3343 * checksum is correct. In particular, this function will return zero
3344 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3345 * hardware has already verified the correctness of the checksum.
3346 */
3347static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3348{
3349 return skb_csum_unnecessary(skb) ?
3350 0 : __skb_checksum_complete(skb);
3351}
3352
3353static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3354{
3355 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3356 if (skb->csum_level == 0)
3357 skb->ip_summed = CHECKSUM_NONE;
3358 else
3359 skb->csum_level--;
3360 }
3361}
3362
3363static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3364{
3365 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3366 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3367 skb->csum_level++;
3368 } else if (skb->ip_summed == CHECKSUM_NONE) {
3369 skb->ip_summed = CHECKSUM_UNNECESSARY;
3370 skb->csum_level = 0;
3371 }
3372}
3373
3374static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3375{
3376 /* Mark current checksum as bad (typically called from GRO
3377 * path). In the case that ip_summed is CHECKSUM_NONE
3378 * this must be the first checksum encountered in the packet.
3379 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3380 * checksum after the last one validated. For UDP, a zero
3381 * checksum can not be marked as bad.
3382 */
3383
3384 if (skb->ip_summed == CHECKSUM_NONE ||
3385 skb->ip_summed == CHECKSUM_UNNECESSARY)
3386 skb->csum_bad = 1;
3387}
3388
3389/* Check if we need to perform checksum complete validation.
3390 *
3391 * Returns true if checksum complete is needed, false otherwise
3392 * (either checksum is unnecessary or zero checksum is allowed).
3393 */
3394static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3395 bool zero_okay,
3396 __sum16 check)
3397{
3398 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3399 skb->csum_valid = 1;
3400 __skb_decr_checksum_unnecessary(skb);
3401 return false;
3402 }
3403
3404 return true;
3405}
3406
3407/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3408 * in checksum_init.
3409 */
3410#define CHECKSUM_BREAK 76
3411
3412/* Unset checksum-complete
3413 *
3414 * Unset checksum complete can be done when packet is being modified
3415 * (uncompressed for instance) and checksum-complete value is
3416 * invalidated.
3417 */
3418static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3419{
3420 if (skb->ip_summed == CHECKSUM_COMPLETE)
3421 skb->ip_summed = CHECKSUM_NONE;
3422}
3423
3424/* Validate (init) checksum based on checksum complete.
3425 *
3426 * Return values:
3427 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3428 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3429 * checksum is stored in skb->csum for use in __skb_checksum_complete
3430 * non-zero: value of invalid checksum
3431 *
3432 */
3433static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3434 bool complete,
3435 __wsum psum)
3436{
3437 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3438 if (!csum_fold(csum_add(psum, skb->csum))) {
3439 skb->csum_valid = 1;
3440 return 0;
3441 }
3442 } else if (skb->csum_bad) {
3443 /* ip_summed == CHECKSUM_NONE in this case */
3444 return (__force __sum16)1;
3445 }
3446
3447 skb->csum = psum;
3448
3449 if (complete || skb->len <= CHECKSUM_BREAK) {
3450 __sum16 csum;
3451
3452 csum = __skb_checksum_complete(skb);
3453 skb->csum_valid = !csum;
3454 return csum;
3455 }
3456
3457 return 0;
3458}
3459
3460static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3461{
3462 return 0;
3463}
3464
3465/* Perform checksum validate (init). Note that this is a macro since we only
3466 * want to calculate the pseudo header which is an input function if necessary.
3467 * First we try to validate without any computation (checksum unnecessary) and
3468 * then calculate based on checksum complete calling the function to compute
3469 * pseudo header.
3470 *
3471 * Return values:
3472 * 0: checksum is validated or try to in skb_checksum_complete
3473 * non-zero: value of invalid checksum
3474 */
3475#define __skb_checksum_validate(skb, proto, complete, \
3476 zero_okay, check, compute_pseudo) \
3477({ \
3478 __sum16 __ret = 0; \
3479 skb->csum_valid = 0; \
3480 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3481 __ret = __skb_checksum_validate_complete(skb, \
3482 complete, compute_pseudo(skb, proto)); \
3483 __ret; \
3484})
3485
3486#define skb_checksum_init(skb, proto, compute_pseudo) \
3487 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3488
3489#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3490 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3491
3492#define skb_checksum_validate(skb, proto, compute_pseudo) \
3493 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3494
3495#define skb_checksum_validate_zero_check(skb, proto, check, \
3496 compute_pseudo) \
3497 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3498
3499#define skb_checksum_simple_validate(skb) \
3500 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3501
3502static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3503{
3504 return (skb->ip_summed == CHECKSUM_NONE &&
3505 skb->csum_valid && !skb->csum_bad);
3506}
3507
3508static inline void __skb_checksum_convert(struct sk_buff *skb,
3509 __sum16 check, __wsum pseudo)
3510{
3511 skb->csum = ~pseudo;
3512 skb->ip_summed = CHECKSUM_COMPLETE;
3513}
3514
3515#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3516do { \
3517 if (__skb_checksum_convert_check(skb)) \
3518 __skb_checksum_convert(skb, check, \
3519 compute_pseudo(skb, proto)); \
3520} while (0)
3521
3522static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3523 u16 start, u16 offset)
3524{
3525 skb->ip_summed = CHECKSUM_PARTIAL;
3526 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3527 skb->csum_offset = offset - start;
3528}
3529
3530/* Update skbuf and packet to reflect the remote checksum offload operation.
3531 * When called, ptr indicates the starting point for skb->csum when
3532 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3533 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3534 */
3535static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3536 int start, int offset, bool nopartial)
3537{
3538 __wsum delta;
3539
3540 if (!nopartial) {
3541 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3542 return;
3543 }
3544
3545 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3546 __skb_checksum_complete(skb);
3547 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3548 }
3549
3550 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3551
3552 /* Adjust skb->csum since we changed the packet */
3553 skb->csum = csum_add(skb->csum, delta);
3554}
3555
3556#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3557void nf_conntrack_destroy(struct nf_conntrack *nfct);
3558static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3559{
3560 if (nfct && atomic_dec_and_test(&nfct->use))
3561 nf_conntrack_destroy(nfct);
3562}
3563static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3564{
3565 if (nfct)
3566 atomic_inc(&nfct->use);
3567}
3568#endif
3569#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3570static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3571{
3572 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3573 kfree(nf_bridge);
3574}
3575static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3576{
3577 if (nf_bridge)
3578 atomic_inc(&nf_bridge->use);
3579}
3580#endif /* CONFIG_BRIDGE_NETFILTER */
3581static inline void nf_reset(struct sk_buff *skb)
3582{
3583#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3584 nf_conntrack_put(skb->nfct);
3585 skb->nfct = NULL;
3586#endif
3587#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3588 nf_bridge_put(skb->nf_bridge);
3589 skb->nf_bridge = NULL;
3590#endif
3591}
3592
3593static inline void nf_reset_trace(struct sk_buff *skb)
3594{
3595#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3596 skb->nf_trace = 0;
3597#endif
3598}
3599
3600/* Note: This doesn't put any conntrack and bridge info in dst. */
3601static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3602 bool copy)
3603{
3604#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3605 dst->nfct = src->nfct;
3606 nf_conntrack_get(src->nfct);
3607 if (copy)
3608 dst->nfctinfo = src->nfctinfo;
3609#endif
3610#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3611 dst->nf_bridge = src->nf_bridge;
3612 nf_bridge_get(src->nf_bridge);
3613#endif
3614#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3615 if (copy)
3616 dst->nf_trace = src->nf_trace;
3617#endif
3618}
3619
3620static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3621{
3622#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3623 nf_conntrack_put(dst->nfct);
3624#endif
3625#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3626 nf_bridge_put(dst->nf_bridge);
3627#endif
3628 __nf_copy(dst, src, true);
3629}
3630
3631#ifdef CONFIG_NETWORK_SECMARK
3632static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3633{
3634 to->secmark = from->secmark;
3635}
3636
3637static inline void skb_init_secmark(struct sk_buff *skb)
3638{
3639 skb->secmark = 0;
3640}
3641#else
3642static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3643{ }
3644
3645static inline void skb_init_secmark(struct sk_buff *skb)
3646{ }
3647#endif
3648
3649static inline bool skb_irq_freeable(const struct sk_buff *skb)
3650{
3651 return !skb->destructor &&
3652#if IS_ENABLED(CONFIG_XFRM)
3653 !skb->sp &&
3654#endif
3655#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3656 !skb->nfct &&
3657#endif
3658 !skb->_skb_refdst &&
3659 !skb_has_frag_list(skb);
3660}
3661
3662static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3663{
3664 skb->queue_mapping = queue_mapping;
3665}
3666
3667static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3668{
3669 return skb->queue_mapping;
3670}
3671
3672static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3673{
3674 to->queue_mapping = from->queue_mapping;
3675}
3676
3677static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3678{
3679 skb->queue_mapping = rx_queue + 1;
3680}
3681
3682static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3683{
3684 return skb->queue_mapping - 1;
3685}
3686
3687static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3688{
3689 return skb->queue_mapping != 0;
3690}
3691
3692static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3693{
3694#ifdef CONFIG_XFRM
3695 return skb->sp;
3696#else
3697 return NULL;
3698#endif
3699}
3700
3701/* Keeps track of mac header offset relative to skb->head.
3702 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3703 * For non-tunnel skb it points to skb_mac_header() and for
3704 * tunnel skb it points to outer mac header.
3705 * Keeps track of level of encapsulation of network headers.
3706 */
3707struct skb_gso_cb {
3708 union {
3709 int mac_offset;
3710 int data_offset;
3711 };
3712 int encap_level;
3713 __wsum csum;
3714 __u16 csum_start;
3715};
3716#define SKB_SGO_CB_OFFSET 32
3717#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3718
3719static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3720{
3721 return (skb_mac_header(inner_skb) - inner_skb->head) -
3722 SKB_GSO_CB(inner_skb)->mac_offset;
3723}
3724
3725static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3726{
3727 int new_headroom, headroom;
3728 int ret;
3729
3730 headroom = skb_headroom(skb);
3731 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3732 if (ret)
3733 return ret;
3734
3735 new_headroom = skb_headroom(skb);
3736 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3737 return 0;
3738}
3739
3740static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3741{
3742 /* Do not update partial checksums if remote checksum is enabled. */
3743 if (skb->remcsum_offload)
3744 return;
3745
3746 SKB_GSO_CB(skb)->csum = res;
3747 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3748}
3749
3750/* Compute the checksum for a gso segment. First compute the checksum value
3751 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3752 * then add in skb->csum (checksum from csum_start to end of packet).
3753 * skb->csum and csum_start are then updated to reflect the checksum of the
3754 * resultant packet starting from the transport header-- the resultant checksum
3755 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3756 * header.
3757 */
3758static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3759{
3760 unsigned char *csum_start = skb_transport_header(skb);
3761 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3762 __wsum partial = SKB_GSO_CB(skb)->csum;
3763
3764 SKB_GSO_CB(skb)->csum = res;
3765 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3766
3767 return csum_fold(csum_partial(csum_start, plen, partial));
3768}
3769
3770static inline bool skb_is_gso(const struct sk_buff *skb)
3771{
3772 return skb_shinfo(skb)->gso_size;
3773}
3774
3775/* Note: Should be called only if skb_is_gso(skb) is true */
3776static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3777{
3778 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3779}
3780
3781static inline void skb_gso_reset(struct sk_buff *skb)
3782{
3783 skb_shinfo(skb)->gso_size = 0;
3784 skb_shinfo(skb)->gso_segs = 0;
3785 skb_shinfo(skb)->gso_type = 0;
3786}
3787
3788void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3789
3790static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3791{
3792 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3793 * wanted then gso_type will be set. */
3794 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3795
3796 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3797 unlikely(shinfo->gso_type == 0)) {
3798 __skb_warn_lro_forwarding(skb);
3799 return true;
3800 }
3801 return false;
3802}
3803
3804static inline void skb_forward_csum(struct sk_buff *skb)
3805{
3806 /* Unfortunately we don't support this one. Any brave souls? */
3807 if (skb->ip_summed == CHECKSUM_COMPLETE)
3808 skb->ip_summed = CHECKSUM_NONE;
3809}
3810
3811/**
3812 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3813 * @skb: skb to check
3814 *
3815 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3816 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3817 * use this helper, to document places where we make this assertion.
3818 */
3819static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3820{
3821#ifdef DEBUG
3822 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3823#endif
3824}
3825
3826bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3827
3828int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3829struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3830 unsigned int transport_len,
3831 __sum16(*skb_chkf)(struct sk_buff *skb));
3832
3833/**
3834 * skb_head_is_locked - Determine if the skb->head is locked down
3835 * @skb: skb to check
3836 *
3837 * The head on skbs build around a head frag can be removed if they are
3838 * not cloned. This function returns true if the skb head is locked down
3839 * due to either being allocated via kmalloc, or by being a clone with
3840 * multiple references to the head.
3841 */
3842static inline bool skb_head_is_locked(const struct sk_buff *skb)
3843{
3844 return !skb->head_frag || skb_cloned(skb);
3845}
3846
3847/**
3848 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3849 *
3850 * @skb: GSO skb
3851 *
3852 * skb_gso_network_seglen is used to determine the real size of the
3853 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3854 *
3855 * The MAC/L2 header is not accounted for.
3856 */
3857static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3858{
3859 unsigned int hdr_len = skb_transport_header(skb) -
3860 skb_network_header(skb);
3861 return hdr_len + skb_gso_transport_seglen(skb);
3862}
3863
3864/* Local Checksum Offload.
3865 * Compute outer checksum based on the assumption that the
3866 * inner checksum will be offloaded later.
3867 * See Documentation/networking/checksum-offloads.txt for
3868 * explanation of how this works.
3869 * Fill in outer checksum adjustment (e.g. with sum of outer
3870 * pseudo-header) before calling.
3871 * Also ensure that inner checksum is in linear data area.
3872 */
3873static inline __wsum lco_csum(struct sk_buff *skb)
3874{
3875 unsigned char *csum_start = skb_checksum_start(skb);
3876 unsigned char *l4_hdr = skb_transport_header(skb);
3877 __wsum partial;
3878
3879 /* Start with complement of inner checksum adjustment */
3880 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3881 skb->csum_offset));
3882
3883 /* Add in checksum of our headers (incl. outer checksum
3884 * adjustment filled in by caller) and return result.
3885 */
3886 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3887}
3888
3889#endif /* __KERNEL__ */
3890#endif /* _LINUX_SKBUFF_H */
3891