]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/linux/skbuff.h
tcp: switch TCP TS option (RFC 7323) to 1ms clock
[mirror_ubuntu-artful-kernel.git] / include / linux / skbuff.h
1 /*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
16
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
25
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <linux/sched/clock.h>
38 #include <net/flow_dissector.h>
39 #include <linux/splice.h>
40 #include <linux/in6.h>
41 #include <linux/if_packet.h>
42 #include <net/flow.h>
43
44 /* The interface for checksum offload between the stack and networking drivers
45 * is as follows...
46 *
47 * A. IP checksum related features
48 *
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
52 * to its device.
53 *
54 * The checksum related features are:
55 *
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
61 *
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
70 *
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
80 *
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
86 *
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
89 *
90 * CHECKSUM_NONE:
91 *
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
95 *
96 * CHECKSUM_UNNECESSARY:
97 *
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
104 *
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 *
113 * skb->csum_level indicates the number of consecutive checksums found in
114 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
115 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
116 * and a device is able to verify the checksums for UDP (possibly zero),
117 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
118 * two. If the device were only able to verify the UDP checksum and not
119 * GRE, either because it doesn't support GRE checksum of because GRE
120 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
121 * not considered in this case).
122 *
123 * CHECKSUM_COMPLETE:
124 *
125 * This is the most generic way. The device supplied checksum of the _whole_
126 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
127 * hardware doesn't need to parse L3/L4 headers to implement this.
128 *
129 * Note: Even if device supports only some protocols, but is able to produce
130 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
131 *
132 * CHECKSUM_PARTIAL:
133 *
134 * A checksum is set up to be offloaded to a device as described in the
135 * output description for CHECKSUM_PARTIAL. This may occur on a packet
136 * received directly from another Linux OS, e.g., a virtualized Linux kernel
137 * on the same host, or it may be set in the input path in GRO or remote
138 * checksum offload. For the purposes of checksum verification, the checksum
139 * referred to by skb->csum_start + skb->csum_offset and any preceding
140 * checksums in the packet are considered verified. Any checksums in the
141 * packet that are after the checksum being offloaded are not considered to
142 * be verified.
143 *
144 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
145 * in the skb->ip_summed for a packet. Values are:
146 *
147 * CHECKSUM_PARTIAL:
148 *
149 * The driver is required to checksum the packet as seen by hard_start_xmit()
150 * from skb->csum_start up to the end, and to record/write the checksum at
151 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
152 * csum_start and csum_offset values are valid values given the length and
153 * offset of the packet, however they should not attempt to validate that the
154 * checksum refers to a legitimate transport layer checksum-- it is the
155 * purview of the stack to validate that csum_start and csum_offset are set
156 * correctly.
157 *
158 * When the stack requests checksum offload for a packet, the driver MUST
159 * ensure that the checksum is set correctly. A driver can either offload the
160 * checksum calculation to the device, or call skb_checksum_help (in the case
161 * that the device does not support offload for a particular checksum).
162 *
163 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
164 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
165 * checksum offload capability. If a device has limited checksum capabilities
166 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
167 * described above) a helper function can be called to resolve
168 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
169 * function takes a spec argument that describes the protocol layer that is
170 * supported for checksum offload and can be called for each packet. If a
171 * packet does not match the specification for offload, skb_checksum_help
172 * is called to resolve the checksum.
173 *
174 * CHECKSUM_NONE:
175 *
176 * The skb was already checksummed by the protocol, or a checksum is not
177 * required.
178 *
179 * CHECKSUM_UNNECESSARY:
180 *
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
182 * output.
183 *
184 * CHECKSUM_COMPLETE:
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
187 *
188 * D. Non-IP checksum (CRC) offloads
189 *
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
193 * accordingly. Note the there is no indication in the skbuff that the
194 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
195 * both IP checksum offload and SCTP CRC offload must verify which offload
196 * is configured for a packet presumably by inspecting packet headers.
197 *
198 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
199 * offloading the FCOE CRC in a packet. To perform this offload the stack
200 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
201 * accordingly. Note the there is no indication in the skbuff that the
202 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
203 * both IP checksum offload and FCOE CRC offload must verify which offload
204 * is configured for a packet presumably by inspecting packet headers.
205 *
206 * E. Checksumming on output with GSO.
207 *
208 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
209 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
210 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
211 * part of the GSO operation is implied. If a checksum is being offloaded
212 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
213 * are set to refer to the outermost checksum being offload (two offloaded
214 * checksums are possible with UDP encapsulation).
215 */
216
217 /* Don't change this without changing skb_csum_unnecessary! */
218 #define CHECKSUM_NONE 0
219 #define CHECKSUM_UNNECESSARY 1
220 #define CHECKSUM_COMPLETE 2
221 #define CHECKSUM_PARTIAL 3
222
223 /* Maximum value in skb->csum_level */
224 #define SKB_MAX_CSUM_LEVEL 3
225
226 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
227 #define SKB_WITH_OVERHEAD(X) \
228 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
229 #define SKB_MAX_ORDER(X, ORDER) \
230 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
231 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
232 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
233
234 /* return minimum truesize of one skb containing X bytes of data */
235 #define SKB_TRUESIZE(X) ((X) + \
236 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
237 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
238
239 struct net_device;
240 struct scatterlist;
241 struct pipe_inode_info;
242 struct iov_iter;
243 struct napi_struct;
244
245 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
246 struct nf_conntrack {
247 atomic_t use;
248 };
249 #endif
250
251 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
252 struct nf_bridge_info {
253 atomic_t use;
254 enum {
255 BRNF_PROTO_UNCHANGED,
256 BRNF_PROTO_8021Q,
257 BRNF_PROTO_PPPOE
258 } orig_proto:8;
259 u8 pkt_otherhost:1;
260 u8 in_prerouting:1;
261 u8 bridged_dnat:1;
262 __u16 frag_max_size;
263 struct net_device *physindev;
264
265 /* always valid & non-NULL from FORWARD on, for physdev match */
266 struct net_device *physoutdev;
267 union {
268 /* prerouting: detect dnat in orig/reply direction */
269 __be32 ipv4_daddr;
270 struct in6_addr ipv6_daddr;
271
272 /* after prerouting + nat detected: store original source
273 * mac since neigh resolution overwrites it, only used while
274 * skb is out in neigh layer.
275 */
276 char neigh_header[8];
277 };
278 };
279 #endif
280
281 struct sk_buff_head {
282 /* These two members must be first. */
283 struct sk_buff *next;
284 struct sk_buff *prev;
285
286 __u32 qlen;
287 spinlock_t lock;
288 };
289
290 struct sk_buff;
291
292 /* To allow 64K frame to be packed as single skb without frag_list we
293 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
294 * buffers which do not start on a page boundary.
295 *
296 * Since GRO uses frags we allocate at least 16 regardless of page
297 * size.
298 */
299 #if (65536/PAGE_SIZE + 1) < 16
300 #define MAX_SKB_FRAGS 16UL
301 #else
302 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
303 #endif
304 extern int sysctl_max_skb_frags;
305
306 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
307 * segment using its current segmentation instead.
308 */
309 #define GSO_BY_FRAGS 0xFFFF
310
311 typedef struct skb_frag_struct skb_frag_t;
312
313 struct skb_frag_struct {
314 struct {
315 struct page *p;
316 } page;
317 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
318 __u32 page_offset;
319 __u32 size;
320 #else
321 __u16 page_offset;
322 __u16 size;
323 #endif
324 };
325
326 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
327 {
328 return frag->size;
329 }
330
331 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
332 {
333 frag->size = size;
334 }
335
336 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
337 {
338 frag->size += delta;
339 }
340
341 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
342 {
343 frag->size -= delta;
344 }
345
346 #define HAVE_HW_TIME_STAMP
347
348 /**
349 * struct skb_shared_hwtstamps - hardware time stamps
350 * @hwtstamp: hardware time stamp transformed into duration
351 * since arbitrary point in time
352 *
353 * Software time stamps generated by ktime_get_real() are stored in
354 * skb->tstamp.
355 *
356 * hwtstamps can only be compared against other hwtstamps from
357 * the same device.
358 *
359 * This structure is attached to packets as part of the
360 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
361 */
362 struct skb_shared_hwtstamps {
363 ktime_t hwtstamp;
364 };
365
366 /* Definitions for tx_flags in struct skb_shared_info */
367 enum {
368 /* generate hardware time stamp */
369 SKBTX_HW_TSTAMP = 1 << 0,
370
371 /* generate software time stamp when queueing packet to NIC */
372 SKBTX_SW_TSTAMP = 1 << 1,
373
374 /* device driver is going to provide hardware time stamp */
375 SKBTX_IN_PROGRESS = 1 << 2,
376
377 /* device driver supports TX zero-copy buffers */
378 SKBTX_DEV_ZEROCOPY = 1 << 3,
379
380 /* generate wifi status information (where possible) */
381 SKBTX_WIFI_STATUS = 1 << 4,
382
383 /* This indicates at least one fragment might be overwritten
384 * (as in vmsplice(), sendfile() ...)
385 * If we need to compute a TX checksum, we'll need to copy
386 * all frags to avoid possible bad checksum
387 */
388 SKBTX_SHARED_FRAG = 1 << 5,
389
390 /* generate software time stamp when entering packet scheduling */
391 SKBTX_SCHED_TSTAMP = 1 << 6,
392 };
393
394 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
395 SKBTX_SCHED_TSTAMP)
396 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
397
398 /*
399 * The callback notifies userspace to release buffers when skb DMA is done in
400 * lower device, the skb last reference should be 0 when calling this.
401 * The zerocopy_success argument is true if zero copy transmit occurred,
402 * false on data copy or out of memory error caused by data copy attempt.
403 * The ctx field is used to track device context.
404 * The desc field is used to track userspace buffer index.
405 */
406 struct ubuf_info {
407 void (*callback)(struct ubuf_info *, bool zerocopy_success);
408 void *ctx;
409 unsigned long desc;
410 };
411
412 /* This data is invariant across clones and lives at
413 * the end of the header data, ie. at skb->end.
414 */
415 struct skb_shared_info {
416 unsigned short _unused;
417 unsigned char nr_frags;
418 __u8 tx_flags;
419 unsigned short gso_size;
420 /* Warning: this field is not always filled in (UFO)! */
421 unsigned short gso_segs;
422 struct sk_buff *frag_list;
423 struct skb_shared_hwtstamps hwtstamps;
424 unsigned int gso_type;
425 u32 tskey;
426 __be32 ip6_frag_id;
427
428 /*
429 * Warning : all fields before dataref are cleared in __alloc_skb()
430 */
431 atomic_t dataref;
432
433 /* Intermediate layers must ensure that destructor_arg
434 * remains valid until skb destructor */
435 void * destructor_arg;
436
437 /* must be last field, see pskb_expand_head() */
438 skb_frag_t frags[MAX_SKB_FRAGS];
439 };
440
441 /* We divide dataref into two halves. The higher 16 bits hold references
442 * to the payload part of skb->data. The lower 16 bits hold references to
443 * the entire skb->data. A clone of a headerless skb holds the length of
444 * the header in skb->hdr_len.
445 *
446 * All users must obey the rule that the skb->data reference count must be
447 * greater than or equal to the payload reference count.
448 *
449 * Holding a reference to the payload part means that the user does not
450 * care about modifications to the header part of skb->data.
451 */
452 #define SKB_DATAREF_SHIFT 16
453 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
454
455
456 enum {
457 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
458 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
459 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
460 };
461
462 enum {
463 SKB_GSO_TCPV4 = 1 << 0,
464 SKB_GSO_UDP = 1 << 1,
465
466 /* This indicates the skb is from an untrusted source. */
467 SKB_GSO_DODGY = 1 << 2,
468
469 /* This indicates the tcp segment has CWR set. */
470 SKB_GSO_TCP_ECN = 1 << 3,
471
472 SKB_GSO_TCP_FIXEDID = 1 << 4,
473
474 SKB_GSO_TCPV6 = 1 << 5,
475
476 SKB_GSO_FCOE = 1 << 6,
477
478 SKB_GSO_GRE = 1 << 7,
479
480 SKB_GSO_GRE_CSUM = 1 << 8,
481
482 SKB_GSO_IPXIP4 = 1 << 9,
483
484 SKB_GSO_IPXIP6 = 1 << 10,
485
486 SKB_GSO_UDP_TUNNEL = 1 << 11,
487
488 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
489
490 SKB_GSO_PARTIAL = 1 << 13,
491
492 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
493
494 SKB_GSO_SCTP = 1 << 15,
495
496 SKB_GSO_ESP = 1 << 16,
497 };
498
499 #if BITS_PER_LONG > 32
500 #define NET_SKBUFF_DATA_USES_OFFSET 1
501 #endif
502
503 #ifdef NET_SKBUFF_DATA_USES_OFFSET
504 typedef unsigned int sk_buff_data_t;
505 #else
506 typedef unsigned char *sk_buff_data_t;
507 #endif
508
509 /**
510 * struct sk_buff - socket buffer
511 * @next: Next buffer in list
512 * @prev: Previous buffer in list
513 * @tstamp: Time we arrived/left
514 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
515 * @sk: Socket we are owned by
516 * @dev: Device we arrived on/are leaving by
517 * @cb: Control buffer. Free for use by every layer. Put private vars here
518 * @_skb_refdst: destination entry (with norefcount bit)
519 * @sp: the security path, used for xfrm
520 * @len: Length of actual data
521 * @data_len: Data length
522 * @mac_len: Length of link layer header
523 * @hdr_len: writable header length of cloned skb
524 * @csum: Checksum (must include start/offset pair)
525 * @csum_start: Offset from skb->head where checksumming should start
526 * @csum_offset: Offset from csum_start where checksum should be stored
527 * @priority: Packet queueing priority
528 * @ignore_df: allow local fragmentation
529 * @cloned: Head may be cloned (check refcnt to be sure)
530 * @ip_summed: Driver fed us an IP checksum
531 * @nohdr: Payload reference only, must not modify header
532 * @pkt_type: Packet class
533 * @fclone: skbuff clone status
534 * @ipvs_property: skbuff is owned by ipvs
535 * @tc_skip_classify: do not classify packet. set by IFB device
536 * @tc_at_ingress: used within tc_classify to distinguish in/egress
537 * @tc_redirected: packet was redirected by a tc action
538 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
539 * @peeked: this packet has been seen already, so stats have been
540 * done for it, don't do them again
541 * @nf_trace: netfilter packet trace flag
542 * @protocol: Packet protocol from driver
543 * @destructor: Destruct function
544 * @_nfct: Associated connection, if any (with nfctinfo bits)
545 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
546 * @skb_iif: ifindex of device we arrived on
547 * @tc_index: Traffic control index
548 * @hash: the packet hash
549 * @queue_mapping: Queue mapping for multiqueue devices
550 * @xmit_more: More SKBs are pending for this queue
551 * @ndisc_nodetype: router type (from link layer)
552 * @ooo_okay: allow the mapping of a socket to a queue to be changed
553 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
554 * ports.
555 * @sw_hash: indicates hash was computed in software stack
556 * @wifi_acked_valid: wifi_acked was set
557 * @wifi_acked: whether frame was acked on wifi or not
558 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
559 * @dst_pending_confirm: need to confirm neighbour
560 * @napi_id: id of the NAPI struct this skb came from
561 * @secmark: security marking
562 * @mark: Generic packet mark
563 * @vlan_proto: vlan encapsulation protocol
564 * @vlan_tci: vlan tag control information
565 * @inner_protocol: Protocol (encapsulation)
566 * @inner_transport_header: Inner transport layer header (encapsulation)
567 * @inner_network_header: Network layer header (encapsulation)
568 * @inner_mac_header: Link layer header (encapsulation)
569 * @transport_header: Transport layer header
570 * @network_header: Network layer header
571 * @mac_header: Link layer header
572 * @tail: Tail pointer
573 * @end: End pointer
574 * @head: Head of buffer
575 * @data: Data head pointer
576 * @truesize: Buffer size
577 * @users: User count - see {datagram,tcp}.c
578 */
579
580 struct sk_buff {
581 union {
582 struct {
583 /* These two members must be first. */
584 struct sk_buff *next;
585 struct sk_buff *prev;
586
587 union {
588 ktime_t tstamp;
589 u64 skb_mstamp;
590 };
591 };
592 struct rb_node rbnode; /* used in netem & tcp stack */
593 };
594 struct sock *sk;
595
596 union {
597 struct net_device *dev;
598 /* Some protocols might use this space to store information,
599 * while device pointer would be NULL.
600 * UDP receive path is one user.
601 */
602 unsigned long dev_scratch;
603 };
604 /*
605 * This is the control buffer. It is free to use for every
606 * layer. Please put your private variables there. If you
607 * want to keep them across layers you have to do a skb_clone()
608 * first. This is owned by whoever has the skb queued ATM.
609 */
610 char cb[48] __aligned(8);
611
612 unsigned long _skb_refdst;
613 void (*destructor)(struct sk_buff *skb);
614 #ifdef CONFIG_XFRM
615 struct sec_path *sp;
616 #endif
617 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
618 unsigned long _nfct;
619 #endif
620 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
621 struct nf_bridge_info *nf_bridge;
622 #endif
623 unsigned int len,
624 data_len;
625 __u16 mac_len,
626 hdr_len;
627
628 /* Following fields are _not_ copied in __copy_skb_header()
629 * Note that queue_mapping is here mostly to fill a hole.
630 */
631 kmemcheck_bitfield_begin(flags1);
632 __u16 queue_mapping;
633
634 /* if you move cloned around you also must adapt those constants */
635 #ifdef __BIG_ENDIAN_BITFIELD
636 #define CLONED_MASK (1 << 7)
637 #else
638 #define CLONED_MASK 1
639 #endif
640 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
641
642 __u8 __cloned_offset[0];
643 __u8 cloned:1,
644 nohdr:1,
645 fclone:2,
646 peeked:1,
647 head_frag:1,
648 xmit_more:1,
649 __unused:1; /* one bit hole */
650 kmemcheck_bitfield_end(flags1);
651
652 /* fields enclosed in headers_start/headers_end are copied
653 * using a single memcpy() in __copy_skb_header()
654 */
655 /* private: */
656 __u32 headers_start[0];
657 /* public: */
658
659 /* if you move pkt_type around you also must adapt those constants */
660 #ifdef __BIG_ENDIAN_BITFIELD
661 #define PKT_TYPE_MAX (7 << 5)
662 #else
663 #define PKT_TYPE_MAX 7
664 #endif
665 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
666
667 __u8 __pkt_type_offset[0];
668 __u8 pkt_type:3;
669 __u8 pfmemalloc:1;
670 __u8 ignore_df:1;
671
672 __u8 nf_trace:1;
673 __u8 ip_summed:2;
674 __u8 ooo_okay:1;
675 __u8 l4_hash:1;
676 __u8 sw_hash:1;
677 __u8 wifi_acked_valid:1;
678 __u8 wifi_acked:1;
679
680 __u8 no_fcs:1;
681 /* Indicates the inner headers are valid in the skbuff. */
682 __u8 encapsulation:1;
683 __u8 encap_hdr_csum:1;
684 __u8 csum_valid:1;
685 __u8 csum_complete_sw:1;
686 __u8 csum_level:2;
687 __u8 csum_bad:1;
688
689 __u8 dst_pending_confirm:1;
690 #ifdef CONFIG_IPV6_NDISC_NODETYPE
691 __u8 ndisc_nodetype:2;
692 #endif
693 __u8 ipvs_property:1;
694 __u8 inner_protocol_type:1;
695 __u8 remcsum_offload:1;
696 #ifdef CONFIG_NET_SWITCHDEV
697 __u8 offload_fwd_mark:1;
698 #endif
699 #ifdef CONFIG_NET_CLS_ACT
700 __u8 tc_skip_classify:1;
701 __u8 tc_at_ingress:1;
702 __u8 tc_redirected:1;
703 __u8 tc_from_ingress:1;
704 #endif
705
706 #ifdef CONFIG_NET_SCHED
707 __u16 tc_index; /* traffic control index */
708 #endif
709
710 union {
711 __wsum csum;
712 struct {
713 __u16 csum_start;
714 __u16 csum_offset;
715 };
716 };
717 __u32 priority;
718 int skb_iif;
719 __u32 hash;
720 __be16 vlan_proto;
721 __u16 vlan_tci;
722 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
723 union {
724 unsigned int napi_id;
725 unsigned int sender_cpu;
726 };
727 #endif
728 #ifdef CONFIG_NETWORK_SECMARK
729 __u32 secmark;
730 #endif
731
732 union {
733 __u32 mark;
734 __u32 reserved_tailroom;
735 };
736
737 union {
738 __be16 inner_protocol;
739 __u8 inner_ipproto;
740 };
741
742 __u16 inner_transport_header;
743 __u16 inner_network_header;
744 __u16 inner_mac_header;
745
746 __be16 protocol;
747 __u16 transport_header;
748 __u16 network_header;
749 __u16 mac_header;
750
751 /* private: */
752 __u32 headers_end[0];
753 /* public: */
754
755 /* These elements must be at the end, see alloc_skb() for details. */
756 sk_buff_data_t tail;
757 sk_buff_data_t end;
758 unsigned char *head,
759 *data;
760 unsigned int truesize;
761 atomic_t users;
762 };
763
764 #ifdef __KERNEL__
765 /*
766 * Handling routines are only of interest to the kernel
767 */
768 #include <linux/slab.h>
769
770
771 #define SKB_ALLOC_FCLONE 0x01
772 #define SKB_ALLOC_RX 0x02
773 #define SKB_ALLOC_NAPI 0x04
774
775 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
776 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
777 {
778 return unlikely(skb->pfmemalloc);
779 }
780
781 /*
782 * skb might have a dst pointer attached, refcounted or not.
783 * _skb_refdst low order bit is set if refcount was _not_ taken
784 */
785 #define SKB_DST_NOREF 1UL
786 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
787
788 #define SKB_NFCT_PTRMASK ~(7UL)
789 /**
790 * skb_dst - returns skb dst_entry
791 * @skb: buffer
792 *
793 * Returns skb dst_entry, regardless of reference taken or not.
794 */
795 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
796 {
797 /* If refdst was not refcounted, check we still are in a
798 * rcu_read_lock section
799 */
800 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
801 !rcu_read_lock_held() &&
802 !rcu_read_lock_bh_held());
803 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
804 }
805
806 /**
807 * skb_dst_set - sets skb dst
808 * @skb: buffer
809 * @dst: dst entry
810 *
811 * Sets skb dst, assuming a reference was taken on dst and should
812 * be released by skb_dst_drop()
813 */
814 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
815 {
816 skb->_skb_refdst = (unsigned long)dst;
817 }
818
819 /**
820 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
821 * @skb: buffer
822 * @dst: dst entry
823 *
824 * Sets skb dst, assuming a reference was not taken on dst.
825 * If dst entry is cached, we do not take reference and dst_release
826 * will be avoided by refdst_drop. If dst entry is not cached, we take
827 * reference, so that last dst_release can destroy the dst immediately.
828 */
829 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
830 {
831 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
832 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
833 }
834
835 /**
836 * skb_dst_is_noref - Test if skb dst isn't refcounted
837 * @skb: buffer
838 */
839 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
840 {
841 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
842 }
843
844 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
845 {
846 return (struct rtable *)skb_dst(skb);
847 }
848
849 /* For mangling skb->pkt_type from user space side from applications
850 * such as nft, tc, etc, we only allow a conservative subset of
851 * possible pkt_types to be set.
852 */
853 static inline bool skb_pkt_type_ok(u32 ptype)
854 {
855 return ptype <= PACKET_OTHERHOST;
856 }
857
858 void kfree_skb(struct sk_buff *skb);
859 void kfree_skb_list(struct sk_buff *segs);
860 void skb_tx_error(struct sk_buff *skb);
861 void consume_skb(struct sk_buff *skb);
862 void __kfree_skb(struct sk_buff *skb);
863 extern struct kmem_cache *skbuff_head_cache;
864
865 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
866 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
867 bool *fragstolen, int *delta_truesize);
868
869 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
870 int node);
871 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
872 struct sk_buff *build_skb(void *data, unsigned int frag_size);
873 static inline struct sk_buff *alloc_skb(unsigned int size,
874 gfp_t priority)
875 {
876 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
877 }
878
879 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
880 unsigned long data_len,
881 int max_page_order,
882 int *errcode,
883 gfp_t gfp_mask);
884
885 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
886 struct sk_buff_fclones {
887 struct sk_buff skb1;
888
889 struct sk_buff skb2;
890
891 atomic_t fclone_ref;
892 };
893
894 /**
895 * skb_fclone_busy - check if fclone is busy
896 * @sk: socket
897 * @skb: buffer
898 *
899 * Returns true if skb is a fast clone, and its clone is not freed.
900 * Some drivers call skb_orphan() in their ndo_start_xmit(),
901 * so we also check that this didnt happen.
902 */
903 static inline bool skb_fclone_busy(const struct sock *sk,
904 const struct sk_buff *skb)
905 {
906 const struct sk_buff_fclones *fclones;
907
908 fclones = container_of(skb, struct sk_buff_fclones, skb1);
909
910 return skb->fclone == SKB_FCLONE_ORIG &&
911 atomic_read(&fclones->fclone_ref) > 1 &&
912 fclones->skb2.sk == sk;
913 }
914
915 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
916 gfp_t priority)
917 {
918 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
919 }
920
921 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
922 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
923 {
924 return __alloc_skb_head(priority, -1);
925 }
926
927 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
928 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
929 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
930 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
931 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
932 gfp_t gfp_mask, bool fclone);
933 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
934 gfp_t gfp_mask)
935 {
936 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
937 }
938
939 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
940 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
941 unsigned int headroom);
942 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
943 int newtailroom, gfp_t priority);
944 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
945 int offset, int len);
946 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
947 int len);
948 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
949 int skb_pad(struct sk_buff *skb, int pad);
950 #define dev_kfree_skb(a) consume_skb(a)
951
952 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
953 int getfrag(void *from, char *to, int offset,
954 int len, int odd, struct sk_buff *skb),
955 void *from, int length);
956
957 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
958 int offset, size_t size);
959
960 struct skb_seq_state {
961 __u32 lower_offset;
962 __u32 upper_offset;
963 __u32 frag_idx;
964 __u32 stepped_offset;
965 struct sk_buff *root_skb;
966 struct sk_buff *cur_skb;
967 __u8 *frag_data;
968 };
969
970 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
971 unsigned int to, struct skb_seq_state *st);
972 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
973 struct skb_seq_state *st);
974 void skb_abort_seq_read(struct skb_seq_state *st);
975
976 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
977 unsigned int to, struct ts_config *config);
978
979 /*
980 * Packet hash types specify the type of hash in skb_set_hash.
981 *
982 * Hash types refer to the protocol layer addresses which are used to
983 * construct a packet's hash. The hashes are used to differentiate or identify
984 * flows of the protocol layer for the hash type. Hash types are either
985 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
986 *
987 * Properties of hashes:
988 *
989 * 1) Two packets in different flows have different hash values
990 * 2) Two packets in the same flow should have the same hash value
991 *
992 * A hash at a higher layer is considered to be more specific. A driver should
993 * set the most specific hash possible.
994 *
995 * A driver cannot indicate a more specific hash than the layer at which a hash
996 * was computed. For instance an L3 hash cannot be set as an L4 hash.
997 *
998 * A driver may indicate a hash level which is less specific than the
999 * actual layer the hash was computed on. For instance, a hash computed
1000 * at L4 may be considered an L3 hash. This should only be done if the
1001 * driver can't unambiguously determine that the HW computed the hash at
1002 * the higher layer. Note that the "should" in the second property above
1003 * permits this.
1004 */
1005 enum pkt_hash_types {
1006 PKT_HASH_TYPE_NONE, /* Undefined type */
1007 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1008 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1009 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1010 };
1011
1012 static inline void skb_clear_hash(struct sk_buff *skb)
1013 {
1014 skb->hash = 0;
1015 skb->sw_hash = 0;
1016 skb->l4_hash = 0;
1017 }
1018
1019 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1020 {
1021 if (!skb->l4_hash)
1022 skb_clear_hash(skb);
1023 }
1024
1025 static inline void
1026 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1027 {
1028 skb->l4_hash = is_l4;
1029 skb->sw_hash = is_sw;
1030 skb->hash = hash;
1031 }
1032
1033 static inline void
1034 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1035 {
1036 /* Used by drivers to set hash from HW */
1037 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1038 }
1039
1040 static inline void
1041 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1042 {
1043 __skb_set_hash(skb, hash, true, is_l4);
1044 }
1045
1046 void __skb_get_hash(struct sk_buff *skb);
1047 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1048 u32 skb_get_poff(const struct sk_buff *skb);
1049 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1050 const struct flow_keys *keys, int hlen);
1051 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1052 void *data, int hlen_proto);
1053
1054 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1055 int thoff, u8 ip_proto)
1056 {
1057 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1058 }
1059
1060 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1061 const struct flow_dissector_key *key,
1062 unsigned int key_count);
1063
1064 bool __skb_flow_dissect(const struct sk_buff *skb,
1065 struct flow_dissector *flow_dissector,
1066 void *target_container,
1067 void *data, __be16 proto, int nhoff, int hlen,
1068 unsigned int flags);
1069
1070 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1071 struct flow_dissector *flow_dissector,
1072 void *target_container, unsigned int flags)
1073 {
1074 return __skb_flow_dissect(skb, flow_dissector, target_container,
1075 NULL, 0, 0, 0, flags);
1076 }
1077
1078 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1079 struct flow_keys *flow,
1080 unsigned int flags)
1081 {
1082 memset(flow, 0, sizeof(*flow));
1083 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1084 NULL, 0, 0, 0, flags);
1085 }
1086
1087 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1088 void *data, __be16 proto,
1089 int nhoff, int hlen,
1090 unsigned int flags)
1091 {
1092 memset(flow, 0, sizeof(*flow));
1093 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1094 data, proto, nhoff, hlen, flags);
1095 }
1096
1097 static inline __u32 skb_get_hash(struct sk_buff *skb)
1098 {
1099 if (!skb->l4_hash && !skb->sw_hash)
1100 __skb_get_hash(skb);
1101
1102 return skb->hash;
1103 }
1104
1105 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1106
1107 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1108 {
1109 if (!skb->l4_hash && !skb->sw_hash) {
1110 struct flow_keys keys;
1111 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1112
1113 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1114 }
1115
1116 return skb->hash;
1117 }
1118
1119 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1120
1121 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1122 {
1123 if (!skb->l4_hash && !skb->sw_hash) {
1124 struct flow_keys keys;
1125 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1126
1127 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1128 }
1129
1130 return skb->hash;
1131 }
1132
1133 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1134
1135 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1136 {
1137 return skb->hash;
1138 }
1139
1140 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1141 {
1142 to->hash = from->hash;
1143 to->sw_hash = from->sw_hash;
1144 to->l4_hash = from->l4_hash;
1145 };
1146
1147 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1148 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1149 {
1150 return skb->head + skb->end;
1151 }
1152
1153 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1154 {
1155 return skb->end;
1156 }
1157 #else
1158 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1159 {
1160 return skb->end;
1161 }
1162
1163 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1164 {
1165 return skb->end - skb->head;
1166 }
1167 #endif
1168
1169 /* Internal */
1170 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1171
1172 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1173 {
1174 return &skb_shinfo(skb)->hwtstamps;
1175 }
1176
1177 /**
1178 * skb_queue_empty - check if a queue is empty
1179 * @list: queue head
1180 *
1181 * Returns true if the queue is empty, false otherwise.
1182 */
1183 static inline int skb_queue_empty(const struct sk_buff_head *list)
1184 {
1185 return list->next == (const struct sk_buff *) list;
1186 }
1187
1188 /**
1189 * skb_queue_is_last - check if skb is the last entry in the queue
1190 * @list: queue head
1191 * @skb: buffer
1192 *
1193 * Returns true if @skb is the last buffer on the list.
1194 */
1195 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1196 const struct sk_buff *skb)
1197 {
1198 return skb->next == (const struct sk_buff *) list;
1199 }
1200
1201 /**
1202 * skb_queue_is_first - check if skb is the first entry in the queue
1203 * @list: queue head
1204 * @skb: buffer
1205 *
1206 * Returns true if @skb is the first buffer on the list.
1207 */
1208 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1209 const struct sk_buff *skb)
1210 {
1211 return skb->prev == (const struct sk_buff *) list;
1212 }
1213
1214 /**
1215 * skb_queue_next - return the next packet in the queue
1216 * @list: queue head
1217 * @skb: current buffer
1218 *
1219 * Return the next packet in @list after @skb. It is only valid to
1220 * call this if skb_queue_is_last() evaluates to false.
1221 */
1222 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1223 const struct sk_buff *skb)
1224 {
1225 /* This BUG_ON may seem severe, but if we just return then we
1226 * are going to dereference garbage.
1227 */
1228 BUG_ON(skb_queue_is_last(list, skb));
1229 return skb->next;
1230 }
1231
1232 /**
1233 * skb_queue_prev - return the prev packet in the queue
1234 * @list: queue head
1235 * @skb: current buffer
1236 *
1237 * Return the prev packet in @list before @skb. It is only valid to
1238 * call this if skb_queue_is_first() evaluates to false.
1239 */
1240 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1241 const struct sk_buff *skb)
1242 {
1243 /* This BUG_ON may seem severe, but if we just return then we
1244 * are going to dereference garbage.
1245 */
1246 BUG_ON(skb_queue_is_first(list, skb));
1247 return skb->prev;
1248 }
1249
1250 /**
1251 * skb_get - reference buffer
1252 * @skb: buffer to reference
1253 *
1254 * Makes another reference to a socket buffer and returns a pointer
1255 * to the buffer.
1256 */
1257 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1258 {
1259 atomic_inc(&skb->users);
1260 return skb;
1261 }
1262
1263 /*
1264 * If users == 1, we are the only owner and are can avoid redundant
1265 * atomic change.
1266 */
1267
1268 /**
1269 * skb_cloned - is the buffer a clone
1270 * @skb: buffer to check
1271 *
1272 * Returns true if the buffer was generated with skb_clone() and is
1273 * one of multiple shared copies of the buffer. Cloned buffers are
1274 * shared data so must not be written to under normal circumstances.
1275 */
1276 static inline int skb_cloned(const struct sk_buff *skb)
1277 {
1278 return skb->cloned &&
1279 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1280 }
1281
1282 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1283 {
1284 might_sleep_if(gfpflags_allow_blocking(pri));
1285
1286 if (skb_cloned(skb))
1287 return pskb_expand_head(skb, 0, 0, pri);
1288
1289 return 0;
1290 }
1291
1292 /**
1293 * skb_header_cloned - is the header a clone
1294 * @skb: buffer to check
1295 *
1296 * Returns true if modifying the header part of the buffer requires
1297 * the data to be copied.
1298 */
1299 static inline int skb_header_cloned(const struct sk_buff *skb)
1300 {
1301 int dataref;
1302
1303 if (!skb->cloned)
1304 return 0;
1305
1306 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1307 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1308 return dataref != 1;
1309 }
1310
1311 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1312 {
1313 might_sleep_if(gfpflags_allow_blocking(pri));
1314
1315 if (skb_header_cloned(skb))
1316 return pskb_expand_head(skb, 0, 0, pri);
1317
1318 return 0;
1319 }
1320
1321 /**
1322 * skb_header_release - release reference to header
1323 * @skb: buffer to operate on
1324 *
1325 * Drop a reference to the header part of the buffer. This is done
1326 * by acquiring a payload reference. You must not read from the header
1327 * part of skb->data after this.
1328 * Note : Check if you can use __skb_header_release() instead.
1329 */
1330 static inline void skb_header_release(struct sk_buff *skb)
1331 {
1332 BUG_ON(skb->nohdr);
1333 skb->nohdr = 1;
1334 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1335 }
1336
1337 /**
1338 * __skb_header_release - release reference to header
1339 * @skb: buffer to operate on
1340 *
1341 * Variant of skb_header_release() assuming skb is private to caller.
1342 * We can avoid one atomic operation.
1343 */
1344 static inline void __skb_header_release(struct sk_buff *skb)
1345 {
1346 skb->nohdr = 1;
1347 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1348 }
1349
1350
1351 /**
1352 * skb_shared - is the buffer shared
1353 * @skb: buffer to check
1354 *
1355 * Returns true if more than one person has a reference to this
1356 * buffer.
1357 */
1358 static inline int skb_shared(const struct sk_buff *skb)
1359 {
1360 return atomic_read(&skb->users) != 1;
1361 }
1362
1363 /**
1364 * skb_share_check - check if buffer is shared and if so clone it
1365 * @skb: buffer to check
1366 * @pri: priority for memory allocation
1367 *
1368 * If the buffer is shared the buffer is cloned and the old copy
1369 * drops a reference. A new clone with a single reference is returned.
1370 * If the buffer is not shared the original buffer is returned. When
1371 * being called from interrupt status or with spinlocks held pri must
1372 * be GFP_ATOMIC.
1373 *
1374 * NULL is returned on a memory allocation failure.
1375 */
1376 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1377 {
1378 might_sleep_if(gfpflags_allow_blocking(pri));
1379 if (skb_shared(skb)) {
1380 struct sk_buff *nskb = skb_clone(skb, pri);
1381
1382 if (likely(nskb))
1383 consume_skb(skb);
1384 else
1385 kfree_skb(skb);
1386 skb = nskb;
1387 }
1388 return skb;
1389 }
1390
1391 /*
1392 * Copy shared buffers into a new sk_buff. We effectively do COW on
1393 * packets to handle cases where we have a local reader and forward
1394 * and a couple of other messy ones. The normal one is tcpdumping
1395 * a packet thats being forwarded.
1396 */
1397
1398 /**
1399 * skb_unshare - make a copy of a shared buffer
1400 * @skb: buffer to check
1401 * @pri: priority for memory allocation
1402 *
1403 * If the socket buffer is a clone then this function creates a new
1404 * copy of the data, drops a reference count on the old copy and returns
1405 * the new copy with the reference count at 1. If the buffer is not a clone
1406 * the original buffer is returned. When called with a spinlock held or
1407 * from interrupt state @pri must be %GFP_ATOMIC
1408 *
1409 * %NULL is returned on a memory allocation failure.
1410 */
1411 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1412 gfp_t pri)
1413 {
1414 might_sleep_if(gfpflags_allow_blocking(pri));
1415 if (skb_cloned(skb)) {
1416 struct sk_buff *nskb = skb_copy(skb, pri);
1417
1418 /* Free our shared copy */
1419 if (likely(nskb))
1420 consume_skb(skb);
1421 else
1422 kfree_skb(skb);
1423 skb = nskb;
1424 }
1425 return skb;
1426 }
1427
1428 /**
1429 * skb_peek - peek at the head of an &sk_buff_head
1430 * @list_: list to peek at
1431 *
1432 * Peek an &sk_buff. Unlike most other operations you _MUST_
1433 * be careful with this one. A peek leaves the buffer on the
1434 * list and someone else may run off with it. You must hold
1435 * the appropriate locks or have a private queue to do this.
1436 *
1437 * Returns %NULL for an empty list or a pointer to the head element.
1438 * The reference count is not incremented and the reference is therefore
1439 * volatile. Use with caution.
1440 */
1441 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1442 {
1443 struct sk_buff *skb = list_->next;
1444
1445 if (skb == (struct sk_buff *)list_)
1446 skb = NULL;
1447 return skb;
1448 }
1449
1450 /**
1451 * skb_peek_next - peek skb following the given one from a queue
1452 * @skb: skb to start from
1453 * @list_: list to peek at
1454 *
1455 * Returns %NULL when the end of the list is met or a pointer to the
1456 * next element. The reference count is not incremented and the
1457 * reference is therefore volatile. Use with caution.
1458 */
1459 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1460 const struct sk_buff_head *list_)
1461 {
1462 struct sk_buff *next = skb->next;
1463
1464 if (next == (struct sk_buff *)list_)
1465 next = NULL;
1466 return next;
1467 }
1468
1469 /**
1470 * skb_peek_tail - peek at the tail of an &sk_buff_head
1471 * @list_: list to peek at
1472 *
1473 * Peek an &sk_buff. Unlike most other operations you _MUST_
1474 * be careful with this one. A peek leaves the buffer on the
1475 * list and someone else may run off with it. You must hold
1476 * the appropriate locks or have a private queue to do this.
1477 *
1478 * Returns %NULL for an empty list or a pointer to the tail element.
1479 * The reference count is not incremented and the reference is therefore
1480 * volatile. Use with caution.
1481 */
1482 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1483 {
1484 struct sk_buff *skb = list_->prev;
1485
1486 if (skb == (struct sk_buff *)list_)
1487 skb = NULL;
1488 return skb;
1489
1490 }
1491
1492 /**
1493 * skb_queue_len - get queue length
1494 * @list_: list to measure
1495 *
1496 * Return the length of an &sk_buff queue.
1497 */
1498 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1499 {
1500 return list_->qlen;
1501 }
1502
1503 /**
1504 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1505 * @list: queue to initialize
1506 *
1507 * This initializes only the list and queue length aspects of
1508 * an sk_buff_head object. This allows to initialize the list
1509 * aspects of an sk_buff_head without reinitializing things like
1510 * the spinlock. It can also be used for on-stack sk_buff_head
1511 * objects where the spinlock is known to not be used.
1512 */
1513 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1514 {
1515 list->prev = list->next = (struct sk_buff *)list;
1516 list->qlen = 0;
1517 }
1518
1519 /*
1520 * This function creates a split out lock class for each invocation;
1521 * this is needed for now since a whole lot of users of the skb-queue
1522 * infrastructure in drivers have different locking usage (in hardirq)
1523 * than the networking core (in softirq only). In the long run either the
1524 * network layer or drivers should need annotation to consolidate the
1525 * main types of usage into 3 classes.
1526 */
1527 static inline void skb_queue_head_init(struct sk_buff_head *list)
1528 {
1529 spin_lock_init(&list->lock);
1530 __skb_queue_head_init(list);
1531 }
1532
1533 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1534 struct lock_class_key *class)
1535 {
1536 skb_queue_head_init(list);
1537 lockdep_set_class(&list->lock, class);
1538 }
1539
1540 /*
1541 * Insert an sk_buff on a list.
1542 *
1543 * The "__skb_xxxx()" functions are the non-atomic ones that
1544 * can only be called with interrupts disabled.
1545 */
1546 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1547 struct sk_buff_head *list);
1548 static inline void __skb_insert(struct sk_buff *newsk,
1549 struct sk_buff *prev, struct sk_buff *next,
1550 struct sk_buff_head *list)
1551 {
1552 newsk->next = next;
1553 newsk->prev = prev;
1554 next->prev = prev->next = newsk;
1555 list->qlen++;
1556 }
1557
1558 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1559 struct sk_buff *prev,
1560 struct sk_buff *next)
1561 {
1562 struct sk_buff *first = list->next;
1563 struct sk_buff *last = list->prev;
1564
1565 first->prev = prev;
1566 prev->next = first;
1567
1568 last->next = next;
1569 next->prev = last;
1570 }
1571
1572 /**
1573 * skb_queue_splice - join two skb lists, this is designed for stacks
1574 * @list: the new list to add
1575 * @head: the place to add it in the first list
1576 */
1577 static inline void skb_queue_splice(const struct sk_buff_head *list,
1578 struct sk_buff_head *head)
1579 {
1580 if (!skb_queue_empty(list)) {
1581 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1582 head->qlen += list->qlen;
1583 }
1584 }
1585
1586 /**
1587 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1588 * @list: the new list to add
1589 * @head: the place to add it in the first list
1590 *
1591 * The list at @list is reinitialised
1592 */
1593 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1594 struct sk_buff_head *head)
1595 {
1596 if (!skb_queue_empty(list)) {
1597 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1598 head->qlen += list->qlen;
1599 __skb_queue_head_init(list);
1600 }
1601 }
1602
1603 /**
1604 * skb_queue_splice_tail - join two skb lists, each list being a queue
1605 * @list: the new list to add
1606 * @head: the place to add it in the first list
1607 */
1608 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1609 struct sk_buff_head *head)
1610 {
1611 if (!skb_queue_empty(list)) {
1612 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1613 head->qlen += list->qlen;
1614 }
1615 }
1616
1617 /**
1618 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1619 * @list: the new list to add
1620 * @head: the place to add it in the first list
1621 *
1622 * Each of the lists is a queue.
1623 * The list at @list is reinitialised
1624 */
1625 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1626 struct sk_buff_head *head)
1627 {
1628 if (!skb_queue_empty(list)) {
1629 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1630 head->qlen += list->qlen;
1631 __skb_queue_head_init(list);
1632 }
1633 }
1634
1635 /**
1636 * __skb_queue_after - queue a buffer at the list head
1637 * @list: list to use
1638 * @prev: place after this buffer
1639 * @newsk: buffer to queue
1640 *
1641 * Queue a buffer int the middle of a list. This function takes no locks
1642 * and you must therefore hold required locks before calling it.
1643 *
1644 * A buffer cannot be placed on two lists at the same time.
1645 */
1646 static inline void __skb_queue_after(struct sk_buff_head *list,
1647 struct sk_buff *prev,
1648 struct sk_buff *newsk)
1649 {
1650 __skb_insert(newsk, prev, prev->next, list);
1651 }
1652
1653 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1654 struct sk_buff_head *list);
1655
1656 static inline void __skb_queue_before(struct sk_buff_head *list,
1657 struct sk_buff *next,
1658 struct sk_buff *newsk)
1659 {
1660 __skb_insert(newsk, next->prev, next, list);
1661 }
1662
1663 /**
1664 * __skb_queue_head - queue a buffer at the list head
1665 * @list: list to use
1666 * @newsk: buffer to queue
1667 *
1668 * Queue a buffer at the start of a list. This function takes no locks
1669 * and you must therefore hold required locks before calling it.
1670 *
1671 * A buffer cannot be placed on two lists at the same time.
1672 */
1673 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1674 static inline void __skb_queue_head(struct sk_buff_head *list,
1675 struct sk_buff *newsk)
1676 {
1677 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1678 }
1679
1680 /**
1681 * __skb_queue_tail - queue a buffer at the list tail
1682 * @list: list to use
1683 * @newsk: buffer to queue
1684 *
1685 * Queue a buffer at the end of a list. This function takes no locks
1686 * and you must therefore hold required locks before calling it.
1687 *
1688 * A buffer cannot be placed on two lists at the same time.
1689 */
1690 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1691 static inline void __skb_queue_tail(struct sk_buff_head *list,
1692 struct sk_buff *newsk)
1693 {
1694 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1695 }
1696
1697 /*
1698 * remove sk_buff from list. _Must_ be called atomically, and with
1699 * the list known..
1700 */
1701 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1702 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1703 {
1704 struct sk_buff *next, *prev;
1705
1706 list->qlen--;
1707 next = skb->next;
1708 prev = skb->prev;
1709 skb->next = skb->prev = NULL;
1710 next->prev = prev;
1711 prev->next = next;
1712 }
1713
1714 /**
1715 * __skb_dequeue - remove from the head of the queue
1716 * @list: list to dequeue from
1717 *
1718 * Remove the head of the list. This function does not take any locks
1719 * so must be used with appropriate locks held only. The head item is
1720 * returned or %NULL if the list is empty.
1721 */
1722 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1723 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1724 {
1725 struct sk_buff *skb = skb_peek(list);
1726 if (skb)
1727 __skb_unlink(skb, list);
1728 return skb;
1729 }
1730
1731 /**
1732 * __skb_dequeue_tail - remove from the tail of the queue
1733 * @list: list to dequeue from
1734 *
1735 * Remove the tail of the list. This function does not take any locks
1736 * so must be used with appropriate locks held only. The tail item is
1737 * returned or %NULL if the list is empty.
1738 */
1739 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1740 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1741 {
1742 struct sk_buff *skb = skb_peek_tail(list);
1743 if (skb)
1744 __skb_unlink(skb, list);
1745 return skb;
1746 }
1747
1748
1749 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1750 {
1751 return skb->data_len;
1752 }
1753
1754 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1755 {
1756 return skb->len - skb->data_len;
1757 }
1758
1759 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1760 {
1761 unsigned int i, len = 0;
1762
1763 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1764 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1765 return len + skb_headlen(skb);
1766 }
1767
1768 /**
1769 * __skb_fill_page_desc - initialise a paged fragment in an skb
1770 * @skb: buffer containing fragment to be initialised
1771 * @i: paged fragment index to initialise
1772 * @page: the page to use for this fragment
1773 * @off: the offset to the data with @page
1774 * @size: the length of the data
1775 *
1776 * Initialises the @i'th fragment of @skb to point to &size bytes at
1777 * offset @off within @page.
1778 *
1779 * Does not take any additional reference on the fragment.
1780 */
1781 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1782 struct page *page, int off, int size)
1783 {
1784 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1785
1786 /*
1787 * Propagate page pfmemalloc to the skb if we can. The problem is
1788 * that not all callers have unique ownership of the page but rely
1789 * on page_is_pfmemalloc doing the right thing(tm).
1790 */
1791 frag->page.p = page;
1792 frag->page_offset = off;
1793 skb_frag_size_set(frag, size);
1794
1795 page = compound_head(page);
1796 if (page_is_pfmemalloc(page))
1797 skb->pfmemalloc = true;
1798 }
1799
1800 /**
1801 * skb_fill_page_desc - initialise a paged fragment in an skb
1802 * @skb: buffer containing fragment to be initialised
1803 * @i: paged fragment index to initialise
1804 * @page: the page to use for this fragment
1805 * @off: the offset to the data with @page
1806 * @size: the length of the data
1807 *
1808 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1809 * @skb to point to @size bytes at offset @off within @page. In
1810 * addition updates @skb such that @i is the last fragment.
1811 *
1812 * Does not take any additional reference on the fragment.
1813 */
1814 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1815 struct page *page, int off, int size)
1816 {
1817 __skb_fill_page_desc(skb, i, page, off, size);
1818 skb_shinfo(skb)->nr_frags = i + 1;
1819 }
1820
1821 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1822 int size, unsigned int truesize);
1823
1824 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1825 unsigned int truesize);
1826
1827 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1828 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1829 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1830
1831 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1832 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1833 {
1834 return skb->head + skb->tail;
1835 }
1836
1837 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1838 {
1839 skb->tail = skb->data - skb->head;
1840 }
1841
1842 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1843 {
1844 skb_reset_tail_pointer(skb);
1845 skb->tail += offset;
1846 }
1847
1848 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1849 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1850 {
1851 return skb->tail;
1852 }
1853
1854 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1855 {
1856 skb->tail = skb->data;
1857 }
1858
1859 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1860 {
1861 skb->tail = skb->data + offset;
1862 }
1863
1864 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1865
1866 /*
1867 * Add data to an sk_buff
1868 */
1869 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1870 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1871 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1872 {
1873 unsigned char *tmp = skb_tail_pointer(skb);
1874 SKB_LINEAR_ASSERT(skb);
1875 skb->tail += len;
1876 skb->len += len;
1877 return tmp;
1878 }
1879
1880 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1881 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1882 {
1883 skb->data -= len;
1884 skb->len += len;
1885 return skb->data;
1886 }
1887
1888 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1889 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1890 {
1891 skb->len -= len;
1892 BUG_ON(skb->len < skb->data_len);
1893 return skb->data += len;
1894 }
1895
1896 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1897 {
1898 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1899 }
1900
1901 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1902
1903 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1904 {
1905 if (len > skb_headlen(skb) &&
1906 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1907 return NULL;
1908 skb->len -= len;
1909 return skb->data += len;
1910 }
1911
1912 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1913 {
1914 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1915 }
1916
1917 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1918 {
1919 if (likely(len <= skb_headlen(skb)))
1920 return 1;
1921 if (unlikely(len > skb->len))
1922 return 0;
1923 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1924 }
1925
1926 void skb_condense(struct sk_buff *skb);
1927
1928 /**
1929 * skb_headroom - bytes at buffer head
1930 * @skb: buffer to check
1931 *
1932 * Return the number of bytes of free space at the head of an &sk_buff.
1933 */
1934 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1935 {
1936 return skb->data - skb->head;
1937 }
1938
1939 /**
1940 * skb_tailroom - bytes at buffer end
1941 * @skb: buffer to check
1942 *
1943 * Return the number of bytes of free space at the tail of an sk_buff
1944 */
1945 static inline int skb_tailroom(const struct sk_buff *skb)
1946 {
1947 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1948 }
1949
1950 /**
1951 * skb_availroom - bytes at buffer end
1952 * @skb: buffer to check
1953 *
1954 * Return the number of bytes of free space at the tail of an sk_buff
1955 * allocated by sk_stream_alloc()
1956 */
1957 static inline int skb_availroom(const struct sk_buff *skb)
1958 {
1959 if (skb_is_nonlinear(skb))
1960 return 0;
1961
1962 return skb->end - skb->tail - skb->reserved_tailroom;
1963 }
1964
1965 /**
1966 * skb_reserve - adjust headroom
1967 * @skb: buffer to alter
1968 * @len: bytes to move
1969 *
1970 * Increase the headroom of an empty &sk_buff by reducing the tail
1971 * room. This is only allowed for an empty buffer.
1972 */
1973 static inline void skb_reserve(struct sk_buff *skb, int len)
1974 {
1975 skb->data += len;
1976 skb->tail += len;
1977 }
1978
1979 /**
1980 * skb_tailroom_reserve - adjust reserved_tailroom
1981 * @skb: buffer to alter
1982 * @mtu: maximum amount of headlen permitted
1983 * @needed_tailroom: minimum amount of reserved_tailroom
1984 *
1985 * Set reserved_tailroom so that headlen can be as large as possible but
1986 * not larger than mtu and tailroom cannot be smaller than
1987 * needed_tailroom.
1988 * The required headroom should already have been reserved before using
1989 * this function.
1990 */
1991 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
1992 unsigned int needed_tailroom)
1993 {
1994 SKB_LINEAR_ASSERT(skb);
1995 if (mtu < skb_tailroom(skb) - needed_tailroom)
1996 /* use at most mtu */
1997 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
1998 else
1999 /* use up to all available space */
2000 skb->reserved_tailroom = needed_tailroom;
2001 }
2002
2003 #define ENCAP_TYPE_ETHER 0
2004 #define ENCAP_TYPE_IPPROTO 1
2005
2006 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2007 __be16 protocol)
2008 {
2009 skb->inner_protocol = protocol;
2010 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2011 }
2012
2013 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2014 __u8 ipproto)
2015 {
2016 skb->inner_ipproto = ipproto;
2017 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2018 }
2019
2020 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2021 {
2022 skb->inner_mac_header = skb->mac_header;
2023 skb->inner_network_header = skb->network_header;
2024 skb->inner_transport_header = skb->transport_header;
2025 }
2026
2027 static inline void skb_reset_mac_len(struct sk_buff *skb)
2028 {
2029 skb->mac_len = skb->network_header - skb->mac_header;
2030 }
2031
2032 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2033 *skb)
2034 {
2035 return skb->head + skb->inner_transport_header;
2036 }
2037
2038 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2039 {
2040 return skb_inner_transport_header(skb) - skb->data;
2041 }
2042
2043 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2044 {
2045 skb->inner_transport_header = skb->data - skb->head;
2046 }
2047
2048 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2049 const int offset)
2050 {
2051 skb_reset_inner_transport_header(skb);
2052 skb->inner_transport_header += offset;
2053 }
2054
2055 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2056 {
2057 return skb->head + skb->inner_network_header;
2058 }
2059
2060 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2061 {
2062 skb->inner_network_header = skb->data - skb->head;
2063 }
2064
2065 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2066 const int offset)
2067 {
2068 skb_reset_inner_network_header(skb);
2069 skb->inner_network_header += offset;
2070 }
2071
2072 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2073 {
2074 return skb->head + skb->inner_mac_header;
2075 }
2076
2077 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2078 {
2079 skb->inner_mac_header = skb->data - skb->head;
2080 }
2081
2082 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2083 const int offset)
2084 {
2085 skb_reset_inner_mac_header(skb);
2086 skb->inner_mac_header += offset;
2087 }
2088 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2089 {
2090 return skb->transport_header != (typeof(skb->transport_header))~0U;
2091 }
2092
2093 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2094 {
2095 return skb->head + skb->transport_header;
2096 }
2097
2098 static inline void skb_reset_transport_header(struct sk_buff *skb)
2099 {
2100 skb->transport_header = skb->data - skb->head;
2101 }
2102
2103 static inline void skb_set_transport_header(struct sk_buff *skb,
2104 const int offset)
2105 {
2106 skb_reset_transport_header(skb);
2107 skb->transport_header += offset;
2108 }
2109
2110 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2111 {
2112 return skb->head + skb->network_header;
2113 }
2114
2115 static inline void skb_reset_network_header(struct sk_buff *skb)
2116 {
2117 skb->network_header = skb->data - skb->head;
2118 }
2119
2120 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2121 {
2122 skb_reset_network_header(skb);
2123 skb->network_header += offset;
2124 }
2125
2126 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2127 {
2128 return skb->head + skb->mac_header;
2129 }
2130
2131 static inline int skb_mac_offset(const struct sk_buff *skb)
2132 {
2133 return skb_mac_header(skb) - skb->data;
2134 }
2135
2136 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2137 {
2138 return skb->mac_header != (typeof(skb->mac_header))~0U;
2139 }
2140
2141 static inline void skb_reset_mac_header(struct sk_buff *skb)
2142 {
2143 skb->mac_header = skb->data - skb->head;
2144 }
2145
2146 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2147 {
2148 skb_reset_mac_header(skb);
2149 skb->mac_header += offset;
2150 }
2151
2152 static inline void skb_pop_mac_header(struct sk_buff *skb)
2153 {
2154 skb->mac_header = skb->network_header;
2155 }
2156
2157 static inline void skb_probe_transport_header(struct sk_buff *skb,
2158 const int offset_hint)
2159 {
2160 struct flow_keys keys;
2161
2162 if (skb_transport_header_was_set(skb))
2163 return;
2164 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2165 skb_set_transport_header(skb, keys.control.thoff);
2166 else
2167 skb_set_transport_header(skb, offset_hint);
2168 }
2169
2170 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2171 {
2172 if (skb_mac_header_was_set(skb)) {
2173 const unsigned char *old_mac = skb_mac_header(skb);
2174
2175 skb_set_mac_header(skb, -skb->mac_len);
2176 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2177 }
2178 }
2179
2180 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2181 {
2182 return skb->csum_start - skb_headroom(skb);
2183 }
2184
2185 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2186 {
2187 return skb->head + skb->csum_start;
2188 }
2189
2190 static inline int skb_transport_offset(const struct sk_buff *skb)
2191 {
2192 return skb_transport_header(skb) - skb->data;
2193 }
2194
2195 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2196 {
2197 return skb->transport_header - skb->network_header;
2198 }
2199
2200 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2201 {
2202 return skb->inner_transport_header - skb->inner_network_header;
2203 }
2204
2205 static inline int skb_network_offset(const struct sk_buff *skb)
2206 {
2207 return skb_network_header(skb) - skb->data;
2208 }
2209
2210 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2211 {
2212 return skb_inner_network_header(skb) - skb->data;
2213 }
2214
2215 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2216 {
2217 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2218 }
2219
2220 /*
2221 * CPUs often take a performance hit when accessing unaligned memory
2222 * locations. The actual performance hit varies, it can be small if the
2223 * hardware handles it or large if we have to take an exception and fix it
2224 * in software.
2225 *
2226 * Since an ethernet header is 14 bytes network drivers often end up with
2227 * the IP header at an unaligned offset. The IP header can be aligned by
2228 * shifting the start of the packet by 2 bytes. Drivers should do this
2229 * with:
2230 *
2231 * skb_reserve(skb, NET_IP_ALIGN);
2232 *
2233 * The downside to this alignment of the IP header is that the DMA is now
2234 * unaligned. On some architectures the cost of an unaligned DMA is high
2235 * and this cost outweighs the gains made by aligning the IP header.
2236 *
2237 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2238 * to be overridden.
2239 */
2240 #ifndef NET_IP_ALIGN
2241 #define NET_IP_ALIGN 2
2242 #endif
2243
2244 /*
2245 * The networking layer reserves some headroom in skb data (via
2246 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2247 * the header has to grow. In the default case, if the header has to grow
2248 * 32 bytes or less we avoid the reallocation.
2249 *
2250 * Unfortunately this headroom changes the DMA alignment of the resulting
2251 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2252 * on some architectures. An architecture can override this value,
2253 * perhaps setting it to a cacheline in size (since that will maintain
2254 * cacheline alignment of the DMA). It must be a power of 2.
2255 *
2256 * Various parts of the networking layer expect at least 32 bytes of
2257 * headroom, you should not reduce this.
2258 *
2259 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2260 * to reduce average number of cache lines per packet.
2261 * get_rps_cpus() for example only access one 64 bytes aligned block :
2262 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2263 */
2264 #ifndef NET_SKB_PAD
2265 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2266 #endif
2267
2268 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2269
2270 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2271 {
2272 if (unlikely(skb_is_nonlinear(skb))) {
2273 WARN_ON(1);
2274 return;
2275 }
2276 skb->len = len;
2277 skb_set_tail_pointer(skb, len);
2278 }
2279
2280 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2281 {
2282 __skb_set_length(skb, len);
2283 }
2284
2285 void skb_trim(struct sk_buff *skb, unsigned int len);
2286
2287 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2288 {
2289 if (skb->data_len)
2290 return ___pskb_trim(skb, len);
2291 __skb_trim(skb, len);
2292 return 0;
2293 }
2294
2295 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2296 {
2297 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2298 }
2299
2300 /**
2301 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2302 * @skb: buffer to alter
2303 * @len: new length
2304 *
2305 * This is identical to pskb_trim except that the caller knows that
2306 * the skb is not cloned so we should never get an error due to out-
2307 * of-memory.
2308 */
2309 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2310 {
2311 int err = pskb_trim(skb, len);
2312 BUG_ON(err);
2313 }
2314
2315 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2316 {
2317 unsigned int diff = len - skb->len;
2318
2319 if (skb_tailroom(skb) < diff) {
2320 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2321 GFP_ATOMIC);
2322 if (ret)
2323 return ret;
2324 }
2325 __skb_set_length(skb, len);
2326 return 0;
2327 }
2328
2329 /**
2330 * skb_orphan - orphan a buffer
2331 * @skb: buffer to orphan
2332 *
2333 * If a buffer currently has an owner then we call the owner's
2334 * destructor function and make the @skb unowned. The buffer continues
2335 * to exist but is no longer charged to its former owner.
2336 */
2337 static inline void skb_orphan(struct sk_buff *skb)
2338 {
2339 if (skb->destructor) {
2340 skb->destructor(skb);
2341 skb->destructor = NULL;
2342 skb->sk = NULL;
2343 } else {
2344 BUG_ON(skb->sk);
2345 }
2346 }
2347
2348 /**
2349 * skb_orphan_frags - orphan the frags contained in a buffer
2350 * @skb: buffer to orphan frags from
2351 * @gfp_mask: allocation mask for replacement pages
2352 *
2353 * For each frag in the SKB which needs a destructor (i.e. has an
2354 * owner) create a copy of that frag and release the original
2355 * page by calling the destructor.
2356 */
2357 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2358 {
2359 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2360 return 0;
2361 return skb_copy_ubufs(skb, gfp_mask);
2362 }
2363
2364 /**
2365 * __skb_queue_purge - empty a list
2366 * @list: list to empty
2367 *
2368 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2369 * the list and one reference dropped. This function does not take the
2370 * list lock and the caller must hold the relevant locks to use it.
2371 */
2372 void skb_queue_purge(struct sk_buff_head *list);
2373 static inline void __skb_queue_purge(struct sk_buff_head *list)
2374 {
2375 struct sk_buff *skb;
2376 while ((skb = __skb_dequeue(list)) != NULL)
2377 kfree_skb(skb);
2378 }
2379
2380 void skb_rbtree_purge(struct rb_root *root);
2381
2382 void *netdev_alloc_frag(unsigned int fragsz);
2383
2384 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2385 gfp_t gfp_mask);
2386
2387 /**
2388 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2389 * @dev: network device to receive on
2390 * @length: length to allocate
2391 *
2392 * Allocate a new &sk_buff and assign it a usage count of one. The
2393 * buffer has unspecified headroom built in. Users should allocate
2394 * the headroom they think they need without accounting for the
2395 * built in space. The built in space is used for optimisations.
2396 *
2397 * %NULL is returned if there is no free memory. Although this function
2398 * allocates memory it can be called from an interrupt.
2399 */
2400 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2401 unsigned int length)
2402 {
2403 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2404 }
2405
2406 /* legacy helper around __netdev_alloc_skb() */
2407 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2408 gfp_t gfp_mask)
2409 {
2410 return __netdev_alloc_skb(NULL, length, gfp_mask);
2411 }
2412
2413 /* legacy helper around netdev_alloc_skb() */
2414 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2415 {
2416 return netdev_alloc_skb(NULL, length);
2417 }
2418
2419
2420 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2421 unsigned int length, gfp_t gfp)
2422 {
2423 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2424
2425 if (NET_IP_ALIGN && skb)
2426 skb_reserve(skb, NET_IP_ALIGN);
2427 return skb;
2428 }
2429
2430 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2431 unsigned int length)
2432 {
2433 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2434 }
2435
2436 static inline void skb_free_frag(void *addr)
2437 {
2438 page_frag_free(addr);
2439 }
2440
2441 void *napi_alloc_frag(unsigned int fragsz);
2442 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2443 unsigned int length, gfp_t gfp_mask);
2444 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2445 unsigned int length)
2446 {
2447 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2448 }
2449 void napi_consume_skb(struct sk_buff *skb, int budget);
2450
2451 void __kfree_skb_flush(void);
2452 void __kfree_skb_defer(struct sk_buff *skb);
2453
2454 /**
2455 * __dev_alloc_pages - allocate page for network Rx
2456 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2457 * @order: size of the allocation
2458 *
2459 * Allocate a new page.
2460 *
2461 * %NULL is returned if there is no free memory.
2462 */
2463 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2464 unsigned int order)
2465 {
2466 /* This piece of code contains several assumptions.
2467 * 1. This is for device Rx, therefor a cold page is preferred.
2468 * 2. The expectation is the user wants a compound page.
2469 * 3. If requesting a order 0 page it will not be compound
2470 * due to the check to see if order has a value in prep_new_page
2471 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2472 * code in gfp_to_alloc_flags that should be enforcing this.
2473 */
2474 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2475
2476 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2477 }
2478
2479 static inline struct page *dev_alloc_pages(unsigned int order)
2480 {
2481 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2482 }
2483
2484 /**
2485 * __dev_alloc_page - allocate a page for network Rx
2486 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2487 *
2488 * Allocate a new page.
2489 *
2490 * %NULL is returned if there is no free memory.
2491 */
2492 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2493 {
2494 return __dev_alloc_pages(gfp_mask, 0);
2495 }
2496
2497 static inline struct page *dev_alloc_page(void)
2498 {
2499 return dev_alloc_pages(0);
2500 }
2501
2502 /**
2503 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2504 * @page: The page that was allocated from skb_alloc_page
2505 * @skb: The skb that may need pfmemalloc set
2506 */
2507 static inline void skb_propagate_pfmemalloc(struct page *page,
2508 struct sk_buff *skb)
2509 {
2510 if (page_is_pfmemalloc(page))
2511 skb->pfmemalloc = true;
2512 }
2513
2514 /**
2515 * skb_frag_page - retrieve the page referred to by a paged fragment
2516 * @frag: the paged fragment
2517 *
2518 * Returns the &struct page associated with @frag.
2519 */
2520 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2521 {
2522 return frag->page.p;
2523 }
2524
2525 /**
2526 * __skb_frag_ref - take an addition reference on a paged fragment.
2527 * @frag: the paged fragment
2528 *
2529 * Takes an additional reference on the paged fragment @frag.
2530 */
2531 static inline void __skb_frag_ref(skb_frag_t *frag)
2532 {
2533 get_page(skb_frag_page(frag));
2534 }
2535
2536 /**
2537 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2538 * @skb: the buffer
2539 * @f: the fragment offset.
2540 *
2541 * Takes an additional reference on the @f'th paged fragment of @skb.
2542 */
2543 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2544 {
2545 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2546 }
2547
2548 /**
2549 * __skb_frag_unref - release a reference on a paged fragment.
2550 * @frag: the paged fragment
2551 *
2552 * Releases a reference on the paged fragment @frag.
2553 */
2554 static inline void __skb_frag_unref(skb_frag_t *frag)
2555 {
2556 put_page(skb_frag_page(frag));
2557 }
2558
2559 /**
2560 * skb_frag_unref - release a reference on a paged fragment of an skb.
2561 * @skb: the buffer
2562 * @f: the fragment offset
2563 *
2564 * Releases a reference on the @f'th paged fragment of @skb.
2565 */
2566 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2567 {
2568 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2569 }
2570
2571 /**
2572 * skb_frag_address - gets the address of the data contained in a paged fragment
2573 * @frag: the paged fragment buffer
2574 *
2575 * Returns the address of the data within @frag. The page must already
2576 * be mapped.
2577 */
2578 static inline void *skb_frag_address(const skb_frag_t *frag)
2579 {
2580 return page_address(skb_frag_page(frag)) + frag->page_offset;
2581 }
2582
2583 /**
2584 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2585 * @frag: the paged fragment buffer
2586 *
2587 * Returns the address of the data within @frag. Checks that the page
2588 * is mapped and returns %NULL otherwise.
2589 */
2590 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2591 {
2592 void *ptr = page_address(skb_frag_page(frag));
2593 if (unlikely(!ptr))
2594 return NULL;
2595
2596 return ptr + frag->page_offset;
2597 }
2598
2599 /**
2600 * __skb_frag_set_page - sets the page contained in a paged fragment
2601 * @frag: the paged fragment
2602 * @page: the page to set
2603 *
2604 * Sets the fragment @frag to contain @page.
2605 */
2606 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2607 {
2608 frag->page.p = page;
2609 }
2610
2611 /**
2612 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2613 * @skb: the buffer
2614 * @f: the fragment offset
2615 * @page: the page to set
2616 *
2617 * Sets the @f'th fragment of @skb to contain @page.
2618 */
2619 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2620 struct page *page)
2621 {
2622 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2623 }
2624
2625 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2626
2627 /**
2628 * skb_frag_dma_map - maps a paged fragment via the DMA API
2629 * @dev: the device to map the fragment to
2630 * @frag: the paged fragment to map
2631 * @offset: the offset within the fragment (starting at the
2632 * fragment's own offset)
2633 * @size: the number of bytes to map
2634 * @dir: the direction of the mapping (%PCI_DMA_*)
2635 *
2636 * Maps the page associated with @frag to @device.
2637 */
2638 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2639 const skb_frag_t *frag,
2640 size_t offset, size_t size,
2641 enum dma_data_direction dir)
2642 {
2643 return dma_map_page(dev, skb_frag_page(frag),
2644 frag->page_offset + offset, size, dir);
2645 }
2646
2647 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2648 gfp_t gfp_mask)
2649 {
2650 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2651 }
2652
2653
2654 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2655 gfp_t gfp_mask)
2656 {
2657 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2658 }
2659
2660
2661 /**
2662 * skb_clone_writable - is the header of a clone writable
2663 * @skb: buffer to check
2664 * @len: length up to which to write
2665 *
2666 * Returns true if modifying the header part of the cloned buffer
2667 * does not requires the data to be copied.
2668 */
2669 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2670 {
2671 return !skb_header_cloned(skb) &&
2672 skb_headroom(skb) + len <= skb->hdr_len;
2673 }
2674
2675 static inline int skb_try_make_writable(struct sk_buff *skb,
2676 unsigned int write_len)
2677 {
2678 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2679 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2680 }
2681
2682 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2683 int cloned)
2684 {
2685 int delta = 0;
2686
2687 if (headroom > skb_headroom(skb))
2688 delta = headroom - skb_headroom(skb);
2689
2690 if (delta || cloned)
2691 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2692 GFP_ATOMIC);
2693 return 0;
2694 }
2695
2696 /**
2697 * skb_cow - copy header of skb when it is required
2698 * @skb: buffer to cow
2699 * @headroom: needed headroom
2700 *
2701 * If the skb passed lacks sufficient headroom or its data part
2702 * is shared, data is reallocated. If reallocation fails, an error
2703 * is returned and original skb is not changed.
2704 *
2705 * The result is skb with writable area skb->head...skb->tail
2706 * and at least @headroom of space at head.
2707 */
2708 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2709 {
2710 return __skb_cow(skb, headroom, skb_cloned(skb));
2711 }
2712
2713 /**
2714 * skb_cow_head - skb_cow but only making the head writable
2715 * @skb: buffer to cow
2716 * @headroom: needed headroom
2717 *
2718 * This function is identical to skb_cow except that we replace the
2719 * skb_cloned check by skb_header_cloned. It should be used when
2720 * you only need to push on some header and do not need to modify
2721 * the data.
2722 */
2723 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2724 {
2725 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2726 }
2727
2728 /**
2729 * skb_padto - pad an skbuff up to a minimal size
2730 * @skb: buffer to pad
2731 * @len: minimal length
2732 *
2733 * Pads up a buffer to ensure the trailing bytes exist and are
2734 * blanked. If the buffer already contains sufficient data it
2735 * is untouched. Otherwise it is extended. Returns zero on
2736 * success. The skb is freed on error.
2737 */
2738 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2739 {
2740 unsigned int size = skb->len;
2741 if (likely(size >= len))
2742 return 0;
2743 return skb_pad(skb, len - size);
2744 }
2745
2746 /**
2747 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2748 * @skb: buffer to pad
2749 * @len: minimal length
2750 *
2751 * Pads up a buffer to ensure the trailing bytes exist and are
2752 * blanked. If the buffer already contains sufficient data it
2753 * is untouched. Otherwise it is extended. Returns zero on
2754 * success. The skb is freed on error.
2755 */
2756 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2757 {
2758 unsigned int size = skb->len;
2759
2760 if (unlikely(size < len)) {
2761 len -= size;
2762 if (skb_pad(skb, len))
2763 return -ENOMEM;
2764 __skb_put(skb, len);
2765 }
2766 return 0;
2767 }
2768
2769 static inline int skb_add_data(struct sk_buff *skb,
2770 struct iov_iter *from, int copy)
2771 {
2772 const int off = skb->len;
2773
2774 if (skb->ip_summed == CHECKSUM_NONE) {
2775 __wsum csum = 0;
2776 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2777 &csum, from)) {
2778 skb->csum = csum_block_add(skb->csum, csum, off);
2779 return 0;
2780 }
2781 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2782 return 0;
2783
2784 __skb_trim(skb, off);
2785 return -EFAULT;
2786 }
2787
2788 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2789 const struct page *page, int off)
2790 {
2791 if (i) {
2792 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2793
2794 return page == skb_frag_page(frag) &&
2795 off == frag->page_offset + skb_frag_size(frag);
2796 }
2797 return false;
2798 }
2799
2800 static inline int __skb_linearize(struct sk_buff *skb)
2801 {
2802 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2803 }
2804
2805 /**
2806 * skb_linearize - convert paged skb to linear one
2807 * @skb: buffer to linarize
2808 *
2809 * If there is no free memory -ENOMEM is returned, otherwise zero
2810 * is returned and the old skb data released.
2811 */
2812 static inline int skb_linearize(struct sk_buff *skb)
2813 {
2814 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2815 }
2816
2817 /**
2818 * skb_has_shared_frag - can any frag be overwritten
2819 * @skb: buffer to test
2820 *
2821 * Return true if the skb has at least one frag that might be modified
2822 * by an external entity (as in vmsplice()/sendfile())
2823 */
2824 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2825 {
2826 return skb_is_nonlinear(skb) &&
2827 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2828 }
2829
2830 /**
2831 * skb_linearize_cow - make sure skb is linear and writable
2832 * @skb: buffer to process
2833 *
2834 * If there is no free memory -ENOMEM is returned, otherwise zero
2835 * is returned and the old skb data released.
2836 */
2837 static inline int skb_linearize_cow(struct sk_buff *skb)
2838 {
2839 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2840 __skb_linearize(skb) : 0;
2841 }
2842
2843 static __always_inline void
2844 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2845 unsigned int off)
2846 {
2847 if (skb->ip_summed == CHECKSUM_COMPLETE)
2848 skb->csum = csum_block_sub(skb->csum,
2849 csum_partial(start, len, 0), off);
2850 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2851 skb_checksum_start_offset(skb) < 0)
2852 skb->ip_summed = CHECKSUM_NONE;
2853 }
2854
2855 /**
2856 * skb_postpull_rcsum - update checksum for received skb after pull
2857 * @skb: buffer to update
2858 * @start: start of data before pull
2859 * @len: length of data pulled
2860 *
2861 * After doing a pull on a received packet, you need to call this to
2862 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2863 * CHECKSUM_NONE so that it can be recomputed from scratch.
2864 */
2865 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2866 const void *start, unsigned int len)
2867 {
2868 __skb_postpull_rcsum(skb, start, len, 0);
2869 }
2870
2871 static __always_inline void
2872 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2873 unsigned int off)
2874 {
2875 if (skb->ip_summed == CHECKSUM_COMPLETE)
2876 skb->csum = csum_block_add(skb->csum,
2877 csum_partial(start, len, 0), off);
2878 }
2879
2880 /**
2881 * skb_postpush_rcsum - update checksum for received skb after push
2882 * @skb: buffer to update
2883 * @start: start of data after push
2884 * @len: length of data pushed
2885 *
2886 * After doing a push on a received packet, you need to call this to
2887 * update the CHECKSUM_COMPLETE checksum.
2888 */
2889 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2890 const void *start, unsigned int len)
2891 {
2892 __skb_postpush_rcsum(skb, start, len, 0);
2893 }
2894
2895 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2896
2897 /**
2898 * skb_push_rcsum - push skb and update receive checksum
2899 * @skb: buffer to update
2900 * @len: length of data pulled
2901 *
2902 * This function performs an skb_push on the packet and updates
2903 * the CHECKSUM_COMPLETE checksum. It should be used on
2904 * receive path processing instead of skb_push unless you know
2905 * that the checksum difference is zero (e.g., a valid IP header)
2906 * or you are setting ip_summed to CHECKSUM_NONE.
2907 */
2908 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2909 unsigned int len)
2910 {
2911 skb_push(skb, len);
2912 skb_postpush_rcsum(skb, skb->data, len);
2913 return skb->data;
2914 }
2915
2916 /**
2917 * pskb_trim_rcsum - trim received skb and update checksum
2918 * @skb: buffer to trim
2919 * @len: new length
2920 *
2921 * This is exactly the same as pskb_trim except that it ensures the
2922 * checksum of received packets are still valid after the operation.
2923 */
2924
2925 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2926 {
2927 if (likely(len >= skb->len))
2928 return 0;
2929 if (skb->ip_summed == CHECKSUM_COMPLETE)
2930 skb->ip_summed = CHECKSUM_NONE;
2931 return __pskb_trim(skb, len);
2932 }
2933
2934 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2935 {
2936 if (skb->ip_summed == CHECKSUM_COMPLETE)
2937 skb->ip_summed = CHECKSUM_NONE;
2938 __skb_trim(skb, len);
2939 return 0;
2940 }
2941
2942 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
2943 {
2944 if (skb->ip_summed == CHECKSUM_COMPLETE)
2945 skb->ip_summed = CHECKSUM_NONE;
2946 return __skb_grow(skb, len);
2947 }
2948
2949 #define skb_queue_walk(queue, skb) \
2950 for (skb = (queue)->next; \
2951 skb != (struct sk_buff *)(queue); \
2952 skb = skb->next)
2953
2954 #define skb_queue_walk_safe(queue, skb, tmp) \
2955 for (skb = (queue)->next, tmp = skb->next; \
2956 skb != (struct sk_buff *)(queue); \
2957 skb = tmp, tmp = skb->next)
2958
2959 #define skb_queue_walk_from(queue, skb) \
2960 for (; skb != (struct sk_buff *)(queue); \
2961 skb = skb->next)
2962
2963 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2964 for (tmp = skb->next; \
2965 skb != (struct sk_buff *)(queue); \
2966 skb = tmp, tmp = skb->next)
2967
2968 #define skb_queue_reverse_walk(queue, skb) \
2969 for (skb = (queue)->prev; \
2970 skb != (struct sk_buff *)(queue); \
2971 skb = skb->prev)
2972
2973 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2974 for (skb = (queue)->prev, tmp = skb->prev; \
2975 skb != (struct sk_buff *)(queue); \
2976 skb = tmp, tmp = skb->prev)
2977
2978 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2979 for (tmp = skb->prev; \
2980 skb != (struct sk_buff *)(queue); \
2981 skb = tmp, tmp = skb->prev)
2982
2983 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2984 {
2985 return skb_shinfo(skb)->frag_list != NULL;
2986 }
2987
2988 static inline void skb_frag_list_init(struct sk_buff *skb)
2989 {
2990 skb_shinfo(skb)->frag_list = NULL;
2991 }
2992
2993 #define skb_walk_frags(skb, iter) \
2994 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2995
2996
2997 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2998 const struct sk_buff *skb);
2999 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3000 struct sk_buff_head *queue,
3001 unsigned int flags,
3002 void (*destructor)(struct sock *sk,
3003 struct sk_buff *skb),
3004 int *peeked, int *off, int *err,
3005 struct sk_buff **last);
3006 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3007 void (*destructor)(struct sock *sk,
3008 struct sk_buff *skb),
3009 int *peeked, int *off, int *err,
3010 struct sk_buff **last);
3011 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3012 void (*destructor)(struct sock *sk,
3013 struct sk_buff *skb),
3014 int *peeked, int *off, int *err);
3015 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3016 int *err);
3017 unsigned int datagram_poll(struct file *file, struct socket *sock,
3018 struct poll_table_struct *wait);
3019 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3020 struct iov_iter *to, int size);
3021 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3022 struct msghdr *msg, int size)
3023 {
3024 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3025 }
3026 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3027 struct msghdr *msg);
3028 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3029 struct iov_iter *from, int len);
3030 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3031 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3032 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3033 static inline void skb_free_datagram_locked(struct sock *sk,
3034 struct sk_buff *skb)
3035 {
3036 __skb_free_datagram_locked(sk, skb, 0);
3037 }
3038 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3039 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3040 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3041 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3042 int len, __wsum csum);
3043 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3044 struct pipe_inode_info *pipe, unsigned int len,
3045 unsigned int flags);
3046 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3047 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3048 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3049 int len, int hlen);
3050 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3051 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3052 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3053 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3054 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3055 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3056 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3057 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3058 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3059 int skb_vlan_pop(struct sk_buff *skb);
3060 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3061 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3062 gfp_t gfp);
3063
3064 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3065 {
3066 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3067 }
3068
3069 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3070 {
3071 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3072 }
3073
3074 struct skb_checksum_ops {
3075 __wsum (*update)(const void *mem, int len, __wsum wsum);
3076 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3077 };
3078
3079 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3080 __wsum csum, const struct skb_checksum_ops *ops);
3081 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3082 __wsum csum);
3083
3084 static inline void * __must_check
3085 __skb_header_pointer(const struct sk_buff *skb, int offset,
3086 int len, void *data, int hlen, void *buffer)
3087 {
3088 if (hlen - offset >= len)
3089 return data + offset;
3090
3091 if (!skb ||
3092 skb_copy_bits(skb, offset, buffer, len) < 0)
3093 return NULL;
3094
3095 return buffer;
3096 }
3097
3098 static inline void * __must_check
3099 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3100 {
3101 return __skb_header_pointer(skb, offset, len, skb->data,
3102 skb_headlen(skb), buffer);
3103 }
3104
3105 /**
3106 * skb_needs_linearize - check if we need to linearize a given skb
3107 * depending on the given device features.
3108 * @skb: socket buffer to check
3109 * @features: net device features
3110 *
3111 * Returns true if either:
3112 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3113 * 2. skb is fragmented and the device does not support SG.
3114 */
3115 static inline bool skb_needs_linearize(struct sk_buff *skb,
3116 netdev_features_t features)
3117 {
3118 return skb_is_nonlinear(skb) &&
3119 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3120 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3121 }
3122
3123 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3124 void *to,
3125 const unsigned int len)
3126 {
3127 memcpy(to, skb->data, len);
3128 }
3129
3130 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3131 const int offset, void *to,
3132 const unsigned int len)
3133 {
3134 memcpy(to, skb->data + offset, len);
3135 }
3136
3137 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3138 const void *from,
3139 const unsigned int len)
3140 {
3141 memcpy(skb->data, from, len);
3142 }
3143
3144 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3145 const int offset,
3146 const void *from,
3147 const unsigned int len)
3148 {
3149 memcpy(skb->data + offset, from, len);
3150 }
3151
3152 void skb_init(void);
3153
3154 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3155 {
3156 return skb->tstamp;
3157 }
3158
3159 /**
3160 * skb_get_timestamp - get timestamp from a skb
3161 * @skb: skb to get stamp from
3162 * @stamp: pointer to struct timeval to store stamp in
3163 *
3164 * Timestamps are stored in the skb as offsets to a base timestamp.
3165 * This function converts the offset back to a struct timeval and stores
3166 * it in stamp.
3167 */
3168 static inline void skb_get_timestamp(const struct sk_buff *skb,
3169 struct timeval *stamp)
3170 {
3171 *stamp = ktime_to_timeval(skb->tstamp);
3172 }
3173
3174 static inline void skb_get_timestampns(const struct sk_buff *skb,
3175 struct timespec *stamp)
3176 {
3177 *stamp = ktime_to_timespec(skb->tstamp);
3178 }
3179
3180 static inline void __net_timestamp(struct sk_buff *skb)
3181 {
3182 skb->tstamp = ktime_get_real();
3183 }
3184
3185 static inline ktime_t net_timedelta(ktime_t t)
3186 {
3187 return ktime_sub(ktime_get_real(), t);
3188 }
3189
3190 static inline ktime_t net_invalid_timestamp(void)
3191 {
3192 return 0;
3193 }
3194
3195 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3196
3197 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3198
3199 void skb_clone_tx_timestamp(struct sk_buff *skb);
3200 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3201
3202 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3203
3204 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3205 {
3206 }
3207
3208 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3209 {
3210 return false;
3211 }
3212
3213 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3214
3215 /**
3216 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3217 *
3218 * PHY drivers may accept clones of transmitted packets for
3219 * timestamping via their phy_driver.txtstamp method. These drivers
3220 * must call this function to return the skb back to the stack with a
3221 * timestamp.
3222 *
3223 * @skb: clone of the the original outgoing packet
3224 * @hwtstamps: hardware time stamps
3225 *
3226 */
3227 void skb_complete_tx_timestamp(struct sk_buff *skb,
3228 struct skb_shared_hwtstamps *hwtstamps);
3229
3230 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3231 struct skb_shared_hwtstamps *hwtstamps,
3232 struct sock *sk, int tstype);
3233
3234 /**
3235 * skb_tstamp_tx - queue clone of skb with send time stamps
3236 * @orig_skb: the original outgoing packet
3237 * @hwtstamps: hardware time stamps, may be NULL if not available
3238 *
3239 * If the skb has a socket associated, then this function clones the
3240 * skb (thus sharing the actual data and optional structures), stores
3241 * the optional hardware time stamping information (if non NULL) or
3242 * generates a software time stamp (otherwise), then queues the clone
3243 * to the error queue of the socket. Errors are silently ignored.
3244 */
3245 void skb_tstamp_tx(struct sk_buff *orig_skb,
3246 struct skb_shared_hwtstamps *hwtstamps);
3247
3248 static inline void sw_tx_timestamp(struct sk_buff *skb)
3249 {
3250 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3251 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3252 skb_tstamp_tx(skb, NULL);
3253 }
3254
3255 /**
3256 * skb_tx_timestamp() - Driver hook for transmit timestamping
3257 *
3258 * Ethernet MAC Drivers should call this function in their hard_xmit()
3259 * function immediately before giving the sk_buff to the MAC hardware.
3260 *
3261 * Specifically, one should make absolutely sure that this function is
3262 * called before TX completion of this packet can trigger. Otherwise
3263 * the packet could potentially already be freed.
3264 *
3265 * @skb: A socket buffer.
3266 */
3267 static inline void skb_tx_timestamp(struct sk_buff *skb)
3268 {
3269 skb_clone_tx_timestamp(skb);
3270 sw_tx_timestamp(skb);
3271 }
3272
3273 /**
3274 * skb_complete_wifi_ack - deliver skb with wifi status
3275 *
3276 * @skb: the original outgoing packet
3277 * @acked: ack status
3278 *
3279 */
3280 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3281
3282 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3283 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3284
3285 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3286 {
3287 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3288 skb->csum_valid ||
3289 (skb->ip_summed == CHECKSUM_PARTIAL &&
3290 skb_checksum_start_offset(skb) >= 0));
3291 }
3292
3293 /**
3294 * skb_checksum_complete - Calculate checksum of an entire packet
3295 * @skb: packet to process
3296 *
3297 * This function calculates the checksum over the entire packet plus
3298 * the value of skb->csum. The latter can be used to supply the
3299 * checksum of a pseudo header as used by TCP/UDP. It returns the
3300 * checksum.
3301 *
3302 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3303 * this function can be used to verify that checksum on received
3304 * packets. In that case the function should return zero if the
3305 * checksum is correct. In particular, this function will return zero
3306 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3307 * hardware has already verified the correctness of the checksum.
3308 */
3309 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3310 {
3311 return skb_csum_unnecessary(skb) ?
3312 0 : __skb_checksum_complete(skb);
3313 }
3314
3315 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3316 {
3317 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3318 if (skb->csum_level == 0)
3319 skb->ip_summed = CHECKSUM_NONE;
3320 else
3321 skb->csum_level--;
3322 }
3323 }
3324
3325 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3326 {
3327 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3328 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3329 skb->csum_level++;
3330 } else if (skb->ip_summed == CHECKSUM_NONE) {
3331 skb->ip_summed = CHECKSUM_UNNECESSARY;
3332 skb->csum_level = 0;
3333 }
3334 }
3335
3336 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3337 {
3338 /* Mark current checksum as bad (typically called from GRO
3339 * path). In the case that ip_summed is CHECKSUM_NONE
3340 * this must be the first checksum encountered in the packet.
3341 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3342 * checksum after the last one validated. For UDP, a zero
3343 * checksum can not be marked as bad.
3344 */
3345
3346 if (skb->ip_summed == CHECKSUM_NONE ||
3347 skb->ip_summed == CHECKSUM_UNNECESSARY)
3348 skb->csum_bad = 1;
3349 }
3350
3351 /* Check if we need to perform checksum complete validation.
3352 *
3353 * Returns true if checksum complete is needed, false otherwise
3354 * (either checksum is unnecessary or zero checksum is allowed).
3355 */
3356 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3357 bool zero_okay,
3358 __sum16 check)
3359 {
3360 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3361 skb->csum_valid = 1;
3362 __skb_decr_checksum_unnecessary(skb);
3363 return false;
3364 }
3365
3366 return true;
3367 }
3368
3369 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3370 * in checksum_init.
3371 */
3372 #define CHECKSUM_BREAK 76
3373
3374 /* Unset checksum-complete
3375 *
3376 * Unset checksum complete can be done when packet is being modified
3377 * (uncompressed for instance) and checksum-complete value is
3378 * invalidated.
3379 */
3380 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3381 {
3382 if (skb->ip_summed == CHECKSUM_COMPLETE)
3383 skb->ip_summed = CHECKSUM_NONE;
3384 }
3385
3386 /* Validate (init) checksum based on checksum complete.
3387 *
3388 * Return values:
3389 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3390 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3391 * checksum is stored in skb->csum for use in __skb_checksum_complete
3392 * non-zero: value of invalid checksum
3393 *
3394 */
3395 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3396 bool complete,
3397 __wsum psum)
3398 {
3399 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3400 if (!csum_fold(csum_add(psum, skb->csum))) {
3401 skb->csum_valid = 1;
3402 return 0;
3403 }
3404 } else if (skb->csum_bad) {
3405 /* ip_summed == CHECKSUM_NONE in this case */
3406 return (__force __sum16)1;
3407 }
3408
3409 skb->csum = psum;
3410
3411 if (complete || skb->len <= CHECKSUM_BREAK) {
3412 __sum16 csum;
3413
3414 csum = __skb_checksum_complete(skb);
3415 skb->csum_valid = !csum;
3416 return csum;
3417 }
3418
3419 return 0;
3420 }
3421
3422 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3423 {
3424 return 0;
3425 }
3426
3427 /* Perform checksum validate (init). Note that this is a macro since we only
3428 * want to calculate the pseudo header which is an input function if necessary.
3429 * First we try to validate without any computation (checksum unnecessary) and
3430 * then calculate based on checksum complete calling the function to compute
3431 * pseudo header.
3432 *
3433 * Return values:
3434 * 0: checksum is validated or try to in skb_checksum_complete
3435 * non-zero: value of invalid checksum
3436 */
3437 #define __skb_checksum_validate(skb, proto, complete, \
3438 zero_okay, check, compute_pseudo) \
3439 ({ \
3440 __sum16 __ret = 0; \
3441 skb->csum_valid = 0; \
3442 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3443 __ret = __skb_checksum_validate_complete(skb, \
3444 complete, compute_pseudo(skb, proto)); \
3445 __ret; \
3446 })
3447
3448 #define skb_checksum_init(skb, proto, compute_pseudo) \
3449 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3450
3451 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3452 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3453
3454 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3455 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3456
3457 #define skb_checksum_validate_zero_check(skb, proto, check, \
3458 compute_pseudo) \
3459 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3460
3461 #define skb_checksum_simple_validate(skb) \
3462 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3463
3464 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3465 {
3466 return (skb->ip_summed == CHECKSUM_NONE &&
3467 skb->csum_valid && !skb->csum_bad);
3468 }
3469
3470 static inline void __skb_checksum_convert(struct sk_buff *skb,
3471 __sum16 check, __wsum pseudo)
3472 {
3473 skb->csum = ~pseudo;
3474 skb->ip_summed = CHECKSUM_COMPLETE;
3475 }
3476
3477 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3478 do { \
3479 if (__skb_checksum_convert_check(skb)) \
3480 __skb_checksum_convert(skb, check, \
3481 compute_pseudo(skb, proto)); \
3482 } while (0)
3483
3484 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3485 u16 start, u16 offset)
3486 {
3487 skb->ip_summed = CHECKSUM_PARTIAL;
3488 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3489 skb->csum_offset = offset - start;
3490 }
3491
3492 /* Update skbuf and packet to reflect the remote checksum offload operation.
3493 * When called, ptr indicates the starting point for skb->csum when
3494 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3495 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3496 */
3497 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3498 int start, int offset, bool nopartial)
3499 {
3500 __wsum delta;
3501
3502 if (!nopartial) {
3503 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3504 return;
3505 }
3506
3507 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3508 __skb_checksum_complete(skb);
3509 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3510 }
3511
3512 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3513
3514 /* Adjust skb->csum since we changed the packet */
3515 skb->csum = csum_add(skb->csum, delta);
3516 }
3517
3518 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3519 {
3520 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3521 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3522 #else
3523 return NULL;
3524 #endif
3525 }
3526
3527 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3528 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3529 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3530 {
3531 if (nfct && atomic_dec_and_test(&nfct->use))
3532 nf_conntrack_destroy(nfct);
3533 }
3534 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3535 {
3536 if (nfct)
3537 atomic_inc(&nfct->use);
3538 }
3539 #endif
3540 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3541 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3542 {
3543 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3544 kfree(nf_bridge);
3545 }
3546 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3547 {
3548 if (nf_bridge)
3549 atomic_inc(&nf_bridge->use);
3550 }
3551 #endif /* CONFIG_BRIDGE_NETFILTER */
3552 static inline void nf_reset(struct sk_buff *skb)
3553 {
3554 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3555 nf_conntrack_put(skb_nfct(skb));
3556 skb->_nfct = 0;
3557 #endif
3558 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3559 nf_bridge_put(skb->nf_bridge);
3560 skb->nf_bridge = NULL;
3561 #endif
3562 }
3563
3564 static inline void nf_reset_trace(struct sk_buff *skb)
3565 {
3566 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3567 skb->nf_trace = 0;
3568 #endif
3569 }
3570
3571 /* Note: This doesn't put any conntrack and bridge info in dst. */
3572 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3573 bool copy)
3574 {
3575 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3576 dst->_nfct = src->_nfct;
3577 nf_conntrack_get(skb_nfct(src));
3578 #endif
3579 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3580 dst->nf_bridge = src->nf_bridge;
3581 nf_bridge_get(src->nf_bridge);
3582 #endif
3583 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3584 if (copy)
3585 dst->nf_trace = src->nf_trace;
3586 #endif
3587 }
3588
3589 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3590 {
3591 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3592 nf_conntrack_put(skb_nfct(dst));
3593 #endif
3594 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3595 nf_bridge_put(dst->nf_bridge);
3596 #endif
3597 __nf_copy(dst, src, true);
3598 }
3599
3600 #ifdef CONFIG_NETWORK_SECMARK
3601 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3602 {
3603 to->secmark = from->secmark;
3604 }
3605
3606 static inline void skb_init_secmark(struct sk_buff *skb)
3607 {
3608 skb->secmark = 0;
3609 }
3610 #else
3611 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3612 { }
3613
3614 static inline void skb_init_secmark(struct sk_buff *skb)
3615 { }
3616 #endif
3617
3618 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3619 {
3620 return !skb->destructor &&
3621 #if IS_ENABLED(CONFIG_XFRM)
3622 !skb->sp &&
3623 #endif
3624 !skb_nfct(skb) &&
3625 !skb->_skb_refdst &&
3626 !skb_has_frag_list(skb);
3627 }
3628
3629 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3630 {
3631 skb->queue_mapping = queue_mapping;
3632 }
3633
3634 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3635 {
3636 return skb->queue_mapping;
3637 }
3638
3639 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3640 {
3641 to->queue_mapping = from->queue_mapping;
3642 }
3643
3644 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3645 {
3646 skb->queue_mapping = rx_queue + 1;
3647 }
3648
3649 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3650 {
3651 return skb->queue_mapping - 1;
3652 }
3653
3654 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3655 {
3656 return skb->queue_mapping != 0;
3657 }
3658
3659 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3660 {
3661 skb->dst_pending_confirm = val;
3662 }
3663
3664 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3665 {
3666 return skb->dst_pending_confirm != 0;
3667 }
3668
3669 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3670 {
3671 #ifdef CONFIG_XFRM
3672 return skb->sp;
3673 #else
3674 return NULL;
3675 #endif
3676 }
3677
3678 /* Keeps track of mac header offset relative to skb->head.
3679 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3680 * For non-tunnel skb it points to skb_mac_header() and for
3681 * tunnel skb it points to outer mac header.
3682 * Keeps track of level of encapsulation of network headers.
3683 */
3684 struct skb_gso_cb {
3685 union {
3686 int mac_offset;
3687 int data_offset;
3688 };
3689 int encap_level;
3690 __wsum csum;
3691 __u16 csum_start;
3692 };
3693 #define SKB_SGO_CB_OFFSET 32
3694 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3695
3696 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3697 {
3698 return (skb_mac_header(inner_skb) - inner_skb->head) -
3699 SKB_GSO_CB(inner_skb)->mac_offset;
3700 }
3701
3702 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3703 {
3704 int new_headroom, headroom;
3705 int ret;
3706
3707 headroom = skb_headroom(skb);
3708 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3709 if (ret)
3710 return ret;
3711
3712 new_headroom = skb_headroom(skb);
3713 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3714 return 0;
3715 }
3716
3717 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3718 {
3719 /* Do not update partial checksums if remote checksum is enabled. */
3720 if (skb->remcsum_offload)
3721 return;
3722
3723 SKB_GSO_CB(skb)->csum = res;
3724 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3725 }
3726
3727 /* Compute the checksum for a gso segment. First compute the checksum value
3728 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3729 * then add in skb->csum (checksum from csum_start to end of packet).
3730 * skb->csum and csum_start are then updated to reflect the checksum of the
3731 * resultant packet starting from the transport header-- the resultant checksum
3732 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3733 * header.
3734 */
3735 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3736 {
3737 unsigned char *csum_start = skb_transport_header(skb);
3738 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3739 __wsum partial = SKB_GSO_CB(skb)->csum;
3740
3741 SKB_GSO_CB(skb)->csum = res;
3742 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3743
3744 return csum_fold(csum_partial(csum_start, plen, partial));
3745 }
3746
3747 static inline bool skb_is_gso(const struct sk_buff *skb)
3748 {
3749 return skb_shinfo(skb)->gso_size;
3750 }
3751
3752 /* Note: Should be called only if skb_is_gso(skb) is true */
3753 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3754 {
3755 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3756 }
3757
3758 static inline void skb_gso_reset(struct sk_buff *skb)
3759 {
3760 skb_shinfo(skb)->gso_size = 0;
3761 skb_shinfo(skb)->gso_segs = 0;
3762 skb_shinfo(skb)->gso_type = 0;
3763 }
3764
3765 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3766
3767 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3768 {
3769 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3770 * wanted then gso_type will be set. */
3771 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3772
3773 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3774 unlikely(shinfo->gso_type == 0)) {
3775 __skb_warn_lro_forwarding(skb);
3776 return true;
3777 }
3778 return false;
3779 }
3780
3781 static inline void skb_forward_csum(struct sk_buff *skb)
3782 {
3783 /* Unfortunately we don't support this one. Any brave souls? */
3784 if (skb->ip_summed == CHECKSUM_COMPLETE)
3785 skb->ip_summed = CHECKSUM_NONE;
3786 }
3787
3788 /**
3789 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3790 * @skb: skb to check
3791 *
3792 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3793 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3794 * use this helper, to document places where we make this assertion.
3795 */
3796 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3797 {
3798 #ifdef DEBUG
3799 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3800 #endif
3801 }
3802
3803 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3804
3805 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3806 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3807 unsigned int transport_len,
3808 __sum16(*skb_chkf)(struct sk_buff *skb));
3809
3810 /**
3811 * skb_head_is_locked - Determine if the skb->head is locked down
3812 * @skb: skb to check
3813 *
3814 * The head on skbs build around a head frag can be removed if they are
3815 * not cloned. This function returns true if the skb head is locked down
3816 * due to either being allocated via kmalloc, or by being a clone with
3817 * multiple references to the head.
3818 */
3819 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3820 {
3821 return !skb->head_frag || skb_cloned(skb);
3822 }
3823
3824 /**
3825 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3826 *
3827 * @skb: GSO skb
3828 *
3829 * skb_gso_network_seglen is used to determine the real size of the
3830 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3831 *
3832 * The MAC/L2 header is not accounted for.
3833 */
3834 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3835 {
3836 unsigned int hdr_len = skb_transport_header(skb) -
3837 skb_network_header(skb);
3838 return hdr_len + skb_gso_transport_seglen(skb);
3839 }
3840
3841 /* Local Checksum Offload.
3842 * Compute outer checksum based on the assumption that the
3843 * inner checksum will be offloaded later.
3844 * See Documentation/networking/checksum-offloads.txt for
3845 * explanation of how this works.
3846 * Fill in outer checksum adjustment (e.g. with sum of outer
3847 * pseudo-header) before calling.
3848 * Also ensure that inner checksum is in linear data area.
3849 */
3850 static inline __wsum lco_csum(struct sk_buff *skb)
3851 {
3852 unsigned char *csum_start = skb_checksum_start(skb);
3853 unsigned char *l4_hdr = skb_transport_header(skb);
3854 __wsum partial;
3855
3856 /* Start with complement of inner checksum adjustment */
3857 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3858 skb->csum_offset));
3859
3860 /* Add in checksum of our headers (incl. outer checksum
3861 * adjustment filled in by caller) and return result.
3862 */
3863 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3864 }
3865
3866 #endif /* __KERNEL__ */
3867 #endif /* _LINUX_SKBUFF_H */