]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/linux/skbuff.h
net: use skb->csum_not_inet to identify packets needing crc32c
[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 set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
199 *
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
207 *
208 * E. Checksumming on output with GSO.
209 *
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
217 */
218
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
224
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
227
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
235
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
240
241 struct net_device;
242 struct scatterlist;
243 struct pipe_inode_info;
244 struct iov_iter;
245 struct napi_struct;
246
247 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
248 struct nf_conntrack {
249 atomic_t use;
250 };
251 #endif
252
253 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
254 struct nf_bridge_info {
255 atomic_t use;
256 enum {
257 BRNF_PROTO_UNCHANGED,
258 BRNF_PROTO_8021Q,
259 BRNF_PROTO_PPPOE
260 } orig_proto:8;
261 u8 pkt_otherhost:1;
262 u8 in_prerouting:1;
263 u8 bridged_dnat:1;
264 __u16 frag_max_size;
265 struct net_device *physindev;
266
267 /* always valid & non-NULL from FORWARD on, for physdev match */
268 struct net_device *physoutdev;
269 union {
270 /* prerouting: detect dnat in orig/reply direction */
271 __be32 ipv4_daddr;
272 struct in6_addr ipv6_daddr;
273
274 /* after prerouting + nat detected: store original source
275 * mac since neigh resolution overwrites it, only used while
276 * skb is out in neigh layer.
277 */
278 char neigh_header[8];
279 };
280 };
281 #endif
282
283 struct sk_buff_head {
284 /* These two members must be first. */
285 struct sk_buff *next;
286 struct sk_buff *prev;
287
288 __u32 qlen;
289 spinlock_t lock;
290 };
291
292 struct sk_buff;
293
294 /* To allow 64K frame to be packed as single skb without frag_list we
295 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
296 * buffers which do not start on a page boundary.
297 *
298 * Since GRO uses frags we allocate at least 16 regardless of page
299 * size.
300 */
301 #if (65536/PAGE_SIZE + 1) < 16
302 #define MAX_SKB_FRAGS 16UL
303 #else
304 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
305 #endif
306 extern int sysctl_max_skb_frags;
307
308 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
309 * segment using its current segmentation instead.
310 */
311 #define GSO_BY_FRAGS 0xFFFF
312
313 typedef struct skb_frag_struct skb_frag_t;
314
315 struct skb_frag_struct {
316 struct {
317 struct page *p;
318 } page;
319 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
320 __u32 page_offset;
321 __u32 size;
322 #else
323 __u16 page_offset;
324 __u16 size;
325 #endif
326 };
327
328 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
329 {
330 return frag->size;
331 }
332
333 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
334 {
335 frag->size = size;
336 }
337
338 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
339 {
340 frag->size += delta;
341 }
342
343 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
344 {
345 frag->size -= delta;
346 }
347
348 #define HAVE_HW_TIME_STAMP
349
350 /**
351 * struct skb_shared_hwtstamps - hardware time stamps
352 * @hwtstamp: hardware time stamp transformed into duration
353 * since arbitrary point in time
354 *
355 * Software time stamps generated by ktime_get_real() are stored in
356 * skb->tstamp.
357 *
358 * hwtstamps can only be compared against other hwtstamps from
359 * the same device.
360 *
361 * This structure is attached to packets as part of the
362 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
363 */
364 struct skb_shared_hwtstamps {
365 ktime_t hwtstamp;
366 };
367
368 /* Definitions for tx_flags in struct skb_shared_info */
369 enum {
370 /* generate hardware time stamp */
371 SKBTX_HW_TSTAMP = 1 << 0,
372
373 /* generate software time stamp when queueing packet to NIC */
374 SKBTX_SW_TSTAMP = 1 << 1,
375
376 /* device driver is going to provide hardware time stamp */
377 SKBTX_IN_PROGRESS = 1 << 2,
378
379 /* device driver supports TX zero-copy buffers */
380 SKBTX_DEV_ZEROCOPY = 1 << 3,
381
382 /* generate wifi status information (where possible) */
383 SKBTX_WIFI_STATUS = 1 << 4,
384
385 /* This indicates at least one fragment might be overwritten
386 * (as in vmsplice(), sendfile() ...)
387 * If we need to compute a TX checksum, we'll need to copy
388 * all frags to avoid possible bad checksum
389 */
390 SKBTX_SHARED_FRAG = 1 << 5,
391
392 /* generate software time stamp when entering packet scheduling */
393 SKBTX_SCHED_TSTAMP = 1 << 6,
394 };
395
396 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
397 SKBTX_SCHED_TSTAMP)
398 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
399
400 /*
401 * The callback notifies userspace to release buffers when skb DMA is done in
402 * lower device, the skb last reference should be 0 when calling this.
403 * The zerocopy_success argument is true if zero copy transmit occurred,
404 * false on data copy or out of memory error caused by data copy attempt.
405 * The ctx field is used to track device context.
406 * The desc field is used to track userspace buffer index.
407 */
408 struct ubuf_info {
409 void (*callback)(struct ubuf_info *, bool zerocopy_success);
410 void *ctx;
411 unsigned long desc;
412 };
413
414 /* This data is invariant across clones and lives at
415 * the end of the header data, ie. at skb->end.
416 */
417 struct skb_shared_info {
418 unsigned short _unused;
419 unsigned char nr_frags;
420 __u8 tx_flags;
421 unsigned short gso_size;
422 /* Warning: this field is not always filled in (UFO)! */
423 unsigned short gso_segs;
424 struct sk_buff *frag_list;
425 struct skb_shared_hwtstamps hwtstamps;
426 unsigned int gso_type;
427 u32 tskey;
428 __be32 ip6_frag_id;
429
430 /*
431 * Warning : all fields before dataref are cleared in __alloc_skb()
432 */
433 atomic_t dataref;
434
435 /* Intermediate layers must ensure that destructor_arg
436 * remains valid until skb destructor */
437 void * destructor_arg;
438
439 /* must be last field, see pskb_expand_head() */
440 skb_frag_t frags[MAX_SKB_FRAGS];
441 };
442
443 /* We divide dataref into two halves. The higher 16 bits hold references
444 * to the payload part of skb->data. The lower 16 bits hold references to
445 * the entire skb->data. A clone of a headerless skb holds the length of
446 * the header in skb->hdr_len.
447 *
448 * All users must obey the rule that the skb->data reference count must be
449 * greater than or equal to the payload reference count.
450 *
451 * Holding a reference to the payload part means that the user does not
452 * care about modifications to the header part of skb->data.
453 */
454 #define SKB_DATAREF_SHIFT 16
455 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
456
457
458 enum {
459 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
460 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
461 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
462 };
463
464 enum {
465 SKB_GSO_TCPV4 = 1 << 0,
466 SKB_GSO_UDP = 1 << 1,
467
468 /* This indicates the skb is from an untrusted source. */
469 SKB_GSO_DODGY = 1 << 2,
470
471 /* This indicates the tcp segment has CWR set. */
472 SKB_GSO_TCP_ECN = 1 << 3,
473
474 SKB_GSO_TCP_FIXEDID = 1 << 4,
475
476 SKB_GSO_TCPV6 = 1 << 5,
477
478 SKB_GSO_FCOE = 1 << 6,
479
480 SKB_GSO_GRE = 1 << 7,
481
482 SKB_GSO_GRE_CSUM = 1 << 8,
483
484 SKB_GSO_IPXIP4 = 1 << 9,
485
486 SKB_GSO_IPXIP6 = 1 << 10,
487
488 SKB_GSO_UDP_TUNNEL = 1 << 11,
489
490 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
491
492 SKB_GSO_PARTIAL = 1 << 13,
493
494 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
495
496 SKB_GSO_SCTP = 1 << 15,
497
498 SKB_GSO_ESP = 1 << 16,
499 };
500
501 #if BITS_PER_LONG > 32
502 #define NET_SKBUFF_DATA_USES_OFFSET 1
503 #endif
504
505 #ifdef NET_SKBUFF_DATA_USES_OFFSET
506 typedef unsigned int sk_buff_data_t;
507 #else
508 typedef unsigned char *sk_buff_data_t;
509 #endif
510
511 /**
512 * struct sk_buff - socket buffer
513 * @next: Next buffer in list
514 * @prev: Previous buffer in list
515 * @tstamp: Time we arrived/left
516 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
517 * @sk: Socket we are owned by
518 * @dev: Device we arrived on/are leaving by
519 * @cb: Control buffer. Free for use by every layer. Put private vars here
520 * @_skb_refdst: destination entry (with norefcount bit)
521 * @sp: the security path, used for xfrm
522 * @len: Length of actual data
523 * @data_len: Data length
524 * @mac_len: Length of link layer header
525 * @hdr_len: writable header length of cloned skb
526 * @csum: Checksum (must include start/offset pair)
527 * @csum_start: Offset from skb->head where checksumming should start
528 * @csum_offset: Offset from csum_start where checksum should be stored
529 * @priority: Packet queueing priority
530 * @ignore_df: allow local fragmentation
531 * @cloned: Head may be cloned (check refcnt to be sure)
532 * @ip_summed: Driver fed us an IP checksum
533 * @nohdr: Payload reference only, must not modify header
534 * @pkt_type: Packet class
535 * @fclone: skbuff clone status
536 * @ipvs_property: skbuff is owned by ipvs
537 * @tc_skip_classify: do not classify packet. set by IFB device
538 * @tc_at_ingress: used within tc_classify to distinguish in/egress
539 * @tc_redirected: packet was redirected by a tc action
540 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
541 * @peeked: this packet has been seen already, so stats have been
542 * done for it, don't do them again
543 * @nf_trace: netfilter packet trace flag
544 * @protocol: Packet protocol from driver
545 * @destructor: Destruct function
546 * @_nfct: Associated connection, if any (with nfctinfo bits)
547 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
548 * @skb_iif: ifindex of device we arrived on
549 * @tc_index: Traffic control index
550 * @hash: the packet hash
551 * @queue_mapping: Queue mapping for multiqueue devices
552 * @xmit_more: More SKBs are pending for this queue
553 * @ndisc_nodetype: router type (from link layer)
554 * @ooo_okay: allow the mapping of a socket to a queue to be changed
555 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
556 * ports.
557 * @sw_hash: indicates hash was computed in software stack
558 * @wifi_acked_valid: wifi_acked was set
559 * @wifi_acked: whether frame was acked on wifi or not
560 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
561 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
562 * @dst_pending_confirm: need to confirm neighbour
563 * @napi_id: id of the NAPI struct this skb came from
564 * @secmark: security marking
565 * @mark: Generic packet mark
566 * @vlan_proto: vlan encapsulation protocol
567 * @vlan_tci: vlan tag control information
568 * @inner_protocol: Protocol (encapsulation)
569 * @inner_transport_header: Inner transport layer header (encapsulation)
570 * @inner_network_header: Network layer header (encapsulation)
571 * @inner_mac_header: Link layer header (encapsulation)
572 * @transport_header: Transport layer header
573 * @network_header: Network layer header
574 * @mac_header: Link layer header
575 * @tail: Tail pointer
576 * @end: End pointer
577 * @head: Head of buffer
578 * @data: Data head pointer
579 * @truesize: Buffer size
580 * @users: User count - see {datagram,tcp}.c
581 */
582
583 struct sk_buff {
584 union {
585 struct {
586 /* These two members must be first. */
587 struct sk_buff *next;
588 struct sk_buff *prev;
589
590 union {
591 ktime_t tstamp;
592 u64 skb_mstamp;
593 };
594 };
595 struct rb_node rbnode; /* used in netem & tcp stack */
596 };
597 struct sock *sk;
598
599 union {
600 struct net_device *dev;
601 /* Some protocols might use this space to store information,
602 * while device pointer would be NULL.
603 * UDP receive path is one user.
604 */
605 unsigned long dev_scratch;
606 };
607 /*
608 * This is the control buffer. It is free to use for every
609 * layer. Please put your private variables there. If you
610 * want to keep them across layers you have to do a skb_clone()
611 * first. This is owned by whoever has the skb queued ATM.
612 */
613 char cb[48] __aligned(8);
614
615 unsigned long _skb_refdst;
616 void (*destructor)(struct sk_buff *skb);
617 #ifdef CONFIG_XFRM
618 struct sec_path *sp;
619 #endif
620 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
621 unsigned long _nfct;
622 #endif
623 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
624 struct nf_bridge_info *nf_bridge;
625 #endif
626 unsigned int len,
627 data_len;
628 __u16 mac_len,
629 hdr_len;
630
631 /* Following fields are _not_ copied in __copy_skb_header()
632 * Note that queue_mapping is here mostly to fill a hole.
633 */
634 kmemcheck_bitfield_begin(flags1);
635 __u16 queue_mapping;
636
637 /* if you move cloned around you also must adapt those constants */
638 #ifdef __BIG_ENDIAN_BITFIELD
639 #define CLONED_MASK (1 << 7)
640 #else
641 #define CLONED_MASK 1
642 #endif
643 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
644
645 __u8 __cloned_offset[0];
646 __u8 cloned:1,
647 nohdr:1,
648 fclone:2,
649 peeked:1,
650 head_frag:1,
651 xmit_more:1,
652 __unused:1; /* one bit hole */
653 kmemcheck_bitfield_end(flags1);
654
655 /* fields enclosed in headers_start/headers_end are copied
656 * using a single memcpy() in __copy_skb_header()
657 */
658 /* private: */
659 __u32 headers_start[0];
660 /* public: */
661
662 /* if you move pkt_type around you also must adapt those constants */
663 #ifdef __BIG_ENDIAN_BITFIELD
664 #define PKT_TYPE_MAX (7 << 5)
665 #else
666 #define PKT_TYPE_MAX 7
667 #endif
668 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
669
670 __u8 __pkt_type_offset[0];
671 __u8 pkt_type:3;
672 __u8 pfmemalloc:1;
673 __u8 ignore_df:1;
674
675 __u8 nf_trace:1;
676 __u8 ip_summed:2;
677 __u8 ooo_okay:1;
678 __u8 l4_hash:1;
679 __u8 sw_hash:1;
680 __u8 wifi_acked_valid:1;
681 __u8 wifi_acked:1;
682
683 __u8 no_fcs:1;
684 /* Indicates the inner headers are valid in the skbuff. */
685 __u8 encapsulation:1;
686 __u8 encap_hdr_csum:1;
687 __u8 csum_valid:1;
688 __u8 csum_complete_sw:1;
689 __u8 csum_level:2;
690 __u8 csum_not_inet:1;
691
692 __u8 dst_pending_confirm:1;
693 #ifdef CONFIG_IPV6_NDISC_NODETYPE
694 __u8 ndisc_nodetype:2;
695 #endif
696 __u8 ipvs_property:1;
697 __u8 inner_protocol_type:1;
698 __u8 remcsum_offload:1;
699 #ifdef CONFIG_NET_SWITCHDEV
700 __u8 offload_fwd_mark:1;
701 #endif
702 #ifdef CONFIG_NET_CLS_ACT
703 __u8 tc_skip_classify:1;
704 __u8 tc_at_ingress:1;
705 __u8 tc_redirected:1;
706 __u8 tc_from_ingress:1;
707 #endif
708
709 #ifdef CONFIG_NET_SCHED
710 __u16 tc_index; /* traffic control index */
711 #endif
712
713 union {
714 __wsum csum;
715 struct {
716 __u16 csum_start;
717 __u16 csum_offset;
718 };
719 };
720 __u32 priority;
721 int skb_iif;
722 __u32 hash;
723 __be16 vlan_proto;
724 __u16 vlan_tci;
725 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
726 union {
727 unsigned int napi_id;
728 unsigned int sender_cpu;
729 };
730 #endif
731 #ifdef CONFIG_NETWORK_SECMARK
732 __u32 secmark;
733 #endif
734
735 union {
736 __u32 mark;
737 __u32 reserved_tailroom;
738 };
739
740 union {
741 __be16 inner_protocol;
742 __u8 inner_ipproto;
743 };
744
745 __u16 inner_transport_header;
746 __u16 inner_network_header;
747 __u16 inner_mac_header;
748
749 __be16 protocol;
750 __u16 transport_header;
751 __u16 network_header;
752 __u16 mac_header;
753
754 /* private: */
755 __u32 headers_end[0];
756 /* public: */
757
758 /* These elements must be at the end, see alloc_skb() for details. */
759 sk_buff_data_t tail;
760 sk_buff_data_t end;
761 unsigned char *head,
762 *data;
763 unsigned int truesize;
764 atomic_t users;
765 };
766
767 #ifdef __KERNEL__
768 /*
769 * Handling routines are only of interest to the kernel
770 */
771 #include <linux/slab.h>
772
773
774 #define SKB_ALLOC_FCLONE 0x01
775 #define SKB_ALLOC_RX 0x02
776 #define SKB_ALLOC_NAPI 0x04
777
778 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
779 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
780 {
781 return unlikely(skb->pfmemalloc);
782 }
783
784 /*
785 * skb might have a dst pointer attached, refcounted or not.
786 * _skb_refdst low order bit is set if refcount was _not_ taken
787 */
788 #define SKB_DST_NOREF 1UL
789 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
790
791 #define SKB_NFCT_PTRMASK ~(7UL)
792 /**
793 * skb_dst - returns skb dst_entry
794 * @skb: buffer
795 *
796 * Returns skb dst_entry, regardless of reference taken or not.
797 */
798 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
799 {
800 /* If refdst was not refcounted, check we still are in a
801 * rcu_read_lock section
802 */
803 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
804 !rcu_read_lock_held() &&
805 !rcu_read_lock_bh_held());
806 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
807 }
808
809 /**
810 * skb_dst_set - sets skb dst
811 * @skb: buffer
812 * @dst: dst entry
813 *
814 * Sets skb dst, assuming a reference was taken on dst and should
815 * be released by skb_dst_drop()
816 */
817 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
818 {
819 skb->_skb_refdst = (unsigned long)dst;
820 }
821
822 /**
823 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
824 * @skb: buffer
825 * @dst: dst entry
826 *
827 * Sets skb dst, assuming a reference was not taken on dst.
828 * If dst entry is cached, we do not take reference and dst_release
829 * will be avoided by refdst_drop. If dst entry is not cached, we take
830 * reference, so that last dst_release can destroy the dst immediately.
831 */
832 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
833 {
834 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
835 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
836 }
837
838 /**
839 * skb_dst_is_noref - Test if skb dst isn't refcounted
840 * @skb: buffer
841 */
842 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
843 {
844 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
845 }
846
847 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
848 {
849 return (struct rtable *)skb_dst(skb);
850 }
851
852 /* For mangling skb->pkt_type from user space side from applications
853 * such as nft, tc, etc, we only allow a conservative subset of
854 * possible pkt_types to be set.
855 */
856 static inline bool skb_pkt_type_ok(u32 ptype)
857 {
858 return ptype <= PACKET_OTHERHOST;
859 }
860
861 void kfree_skb(struct sk_buff *skb);
862 void kfree_skb_list(struct sk_buff *segs);
863 void skb_tx_error(struct sk_buff *skb);
864 void consume_skb(struct sk_buff *skb);
865 void __kfree_skb(struct sk_buff *skb);
866 extern struct kmem_cache *skbuff_head_cache;
867
868 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
869 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
870 bool *fragstolen, int *delta_truesize);
871
872 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
873 int node);
874 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
875 struct sk_buff *build_skb(void *data, unsigned int frag_size);
876 static inline struct sk_buff *alloc_skb(unsigned int size,
877 gfp_t priority)
878 {
879 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
880 }
881
882 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
883 unsigned long data_len,
884 int max_page_order,
885 int *errcode,
886 gfp_t gfp_mask);
887
888 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
889 struct sk_buff_fclones {
890 struct sk_buff skb1;
891
892 struct sk_buff skb2;
893
894 atomic_t fclone_ref;
895 };
896
897 /**
898 * skb_fclone_busy - check if fclone is busy
899 * @sk: socket
900 * @skb: buffer
901 *
902 * Returns true if skb is a fast clone, and its clone is not freed.
903 * Some drivers call skb_orphan() in their ndo_start_xmit(),
904 * so we also check that this didnt happen.
905 */
906 static inline bool skb_fclone_busy(const struct sock *sk,
907 const struct sk_buff *skb)
908 {
909 const struct sk_buff_fclones *fclones;
910
911 fclones = container_of(skb, struct sk_buff_fclones, skb1);
912
913 return skb->fclone == SKB_FCLONE_ORIG &&
914 atomic_read(&fclones->fclone_ref) > 1 &&
915 fclones->skb2.sk == sk;
916 }
917
918 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
919 gfp_t priority)
920 {
921 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
922 }
923
924 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
925 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
926 {
927 return __alloc_skb_head(priority, -1);
928 }
929
930 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
931 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
932 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
933 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
934 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
935 gfp_t gfp_mask, bool fclone);
936 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
937 gfp_t gfp_mask)
938 {
939 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
940 }
941
942 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
943 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
944 unsigned int headroom);
945 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
946 int newtailroom, gfp_t priority);
947 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
948 int offset, int len);
949 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
950 int len);
951 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
952 int skb_pad(struct sk_buff *skb, int pad);
953 #define dev_kfree_skb(a) consume_skb(a)
954
955 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
956 int getfrag(void *from, char *to, int offset,
957 int len, int odd, struct sk_buff *skb),
958 void *from, int length);
959
960 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
961 int offset, size_t size);
962
963 struct skb_seq_state {
964 __u32 lower_offset;
965 __u32 upper_offset;
966 __u32 frag_idx;
967 __u32 stepped_offset;
968 struct sk_buff *root_skb;
969 struct sk_buff *cur_skb;
970 __u8 *frag_data;
971 };
972
973 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
974 unsigned int to, struct skb_seq_state *st);
975 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
976 struct skb_seq_state *st);
977 void skb_abort_seq_read(struct skb_seq_state *st);
978
979 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
980 unsigned int to, struct ts_config *config);
981
982 /*
983 * Packet hash types specify the type of hash in skb_set_hash.
984 *
985 * Hash types refer to the protocol layer addresses which are used to
986 * construct a packet's hash. The hashes are used to differentiate or identify
987 * flows of the protocol layer for the hash type. Hash types are either
988 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
989 *
990 * Properties of hashes:
991 *
992 * 1) Two packets in different flows have different hash values
993 * 2) Two packets in the same flow should have the same hash value
994 *
995 * A hash at a higher layer is considered to be more specific. A driver should
996 * set the most specific hash possible.
997 *
998 * A driver cannot indicate a more specific hash than the layer at which a hash
999 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1000 *
1001 * A driver may indicate a hash level which is less specific than the
1002 * actual layer the hash was computed on. For instance, a hash computed
1003 * at L4 may be considered an L3 hash. This should only be done if the
1004 * driver can't unambiguously determine that the HW computed the hash at
1005 * the higher layer. Note that the "should" in the second property above
1006 * permits this.
1007 */
1008 enum pkt_hash_types {
1009 PKT_HASH_TYPE_NONE, /* Undefined type */
1010 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1011 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1012 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1013 };
1014
1015 static inline void skb_clear_hash(struct sk_buff *skb)
1016 {
1017 skb->hash = 0;
1018 skb->sw_hash = 0;
1019 skb->l4_hash = 0;
1020 }
1021
1022 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1023 {
1024 if (!skb->l4_hash)
1025 skb_clear_hash(skb);
1026 }
1027
1028 static inline void
1029 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1030 {
1031 skb->l4_hash = is_l4;
1032 skb->sw_hash = is_sw;
1033 skb->hash = hash;
1034 }
1035
1036 static inline void
1037 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1038 {
1039 /* Used by drivers to set hash from HW */
1040 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1041 }
1042
1043 static inline void
1044 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1045 {
1046 __skb_set_hash(skb, hash, true, is_l4);
1047 }
1048
1049 void __skb_get_hash(struct sk_buff *skb);
1050 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1051 u32 skb_get_poff(const struct sk_buff *skb);
1052 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1053 const struct flow_keys *keys, int hlen);
1054 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1055 void *data, int hlen_proto);
1056
1057 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1058 int thoff, u8 ip_proto)
1059 {
1060 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1061 }
1062
1063 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1064 const struct flow_dissector_key *key,
1065 unsigned int key_count);
1066
1067 bool __skb_flow_dissect(const struct sk_buff *skb,
1068 struct flow_dissector *flow_dissector,
1069 void *target_container,
1070 void *data, __be16 proto, int nhoff, int hlen,
1071 unsigned int flags);
1072
1073 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1074 struct flow_dissector *flow_dissector,
1075 void *target_container, unsigned int flags)
1076 {
1077 return __skb_flow_dissect(skb, flow_dissector, target_container,
1078 NULL, 0, 0, 0, flags);
1079 }
1080
1081 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1082 struct flow_keys *flow,
1083 unsigned int flags)
1084 {
1085 memset(flow, 0, sizeof(*flow));
1086 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1087 NULL, 0, 0, 0, flags);
1088 }
1089
1090 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1091 void *data, __be16 proto,
1092 int nhoff, int hlen,
1093 unsigned int flags)
1094 {
1095 memset(flow, 0, sizeof(*flow));
1096 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1097 data, proto, nhoff, hlen, flags);
1098 }
1099
1100 static inline __u32 skb_get_hash(struct sk_buff *skb)
1101 {
1102 if (!skb->l4_hash && !skb->sw_hash)
1103 __skb_get_hash(skb);
1104
1105 return skb->hash;
1106 }
1107
1108 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1109
1110 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1111 {
1112 if (!skb->l4_hash && !skb->sw_hash) {
1113 struct flow_keys keys;
1114 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1115
1116 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1117 }
1118
1119 return skb->hash;
1120 }
1121
1122 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1123
1124 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1125 {
1126 if (!skb->l4_hash && !skb->sw_hash) {
1127 struct flow_keys keys;
1128 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1129
1130 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1131 }
1132
1133 return skb->hash;
1134 }
1135
1136 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1137
1138 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1139 {
1140 return skb->hash;
1141 }
1142
1143 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1144 {
1145 to->hash = from->hash;
1146 to->sw_hash = from->sw_hash;
1147 to->l4_hash = from->l4_hash;
1148 };
1149
1150 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1151 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1152 {
1153 return skb->head + skb->end;
1154 }
1155
1156 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1157 {
1158 return skb->end;
1159 }
1160 #else
1161 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1162 {
1163 return skb->end;
1164 }
1165
1166 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1167 {
1168 return skb->end - skb->head;
1169 }
1170 #endif
1171
1172 /* Internal */
1173 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1174
1175 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1176 {
1177 return &skb_shinfo(skb)->hwtstamps;
1178 }
1179
1180 /**
1181 * skb_queue_empty - check if a queue is empty
1182 * @list: queue head
1183 *
1184 * Returns true if the queue is empty, false otherwise.
1185 */
1186 static inline int skb_queue_empty(const struct sk_buff_head *list)
1187 {
1188 return list->next == (const struct sk_buff *) list;
1189 }
1190
1191 /**
1192 * skb_queue_is_last - check if skb is the last entry in the queue
1193 * @list: queue head
1194 * @skb: buffer
1195 *
1196 * Returns true if @skb is the last buffer on the list.
1197 */
1198 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1199 const struct sk_buff *skb)
1200 {
1201 return skb->next == (const struct sk_buff *) list;
1202 }
1203
1204 /**
1205 * skb_queue_is_first - check if skb is the first entry in the queue
1206 * @list: queue head
1207 * @skb: buffer
1208 *
1209 * Returns true if @skb is the first buffer on the list.
1210 */
1211 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1212 const struct sk_buff *skb)
1213 {
1214 return skb->prev == (const struct sk_buff *) list;
1215 }
1216
1217 /**
1218 * skb_queue_next - return the next packet in the queue
1219 * @list: queue head
1220 * @skb: current buffer
1221 *
1222 * Return the next packet in @list after @skb. It is only valid to
1223 * call this if skb_queue_is_last() evaluates to false.
1224 */
1225 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1226 const struct sk_buff *skb)
1227 {
1228 /* This BUG_ON may seem severe, but if we just return then we
1229 * are going to dereference garbage.
1230 */
1231 BUG_ON(skb_queue_is_last(list, skb));
1232 return skb->next;
1233 }
1234
1235 /**
1236 * skb_queue_prev - return the prev packet in the queue
1237 * @list: queue head
1238 * @skb: current buffer
1239 *
1240 * Return the prev packet in @list before @skb. It is only valid to
1241 * call this if skb_queue_is_first() evaluates to false.
1242 */
1243 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1244 const struct sk_buff *skb)
1245 {
1246 /* This BUG_ON may seem severe, but if we just return then we
1247 * are going to dereference garbage.
1248 */
1249 BUG_ON(skb_queue_is_first(list, skb));
1250 return skb->prev;
1251 }
1252
1253 /**
1254 * skb_get - reference buffer
1255 * @skb: buffer to reference
1256 *
1257 * Makes another reference to a socket buffer and returns a pointer
1258 * to the buffer.
1259 */
1260 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1261 {
1262 atomic_inc(&skb->users);
1263 return skb;
1264 }
1265
1266 /*
1267 * If users == 1, we are the only owner and are can avoid redundant
1268 * atomic change.
1269 */
1270
1271 /**
1272 * skb_cloned - is the buffer a clone
1273 * @skb: buffer to check
1274 *
1275 * Returns true if the buffer was generated with skb_clone() and is
1276 * one of multiple shared copies of the buffer. Cloned buffers are
1277 * shared data so must not be written to under normal circumstances.
1278 */
1279 static inline int skb_cloned(const struct sk_buff *skb)
1280 {
1281 return skb->cloned &&
1282 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1283 }
1284
1285 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1286 {
1287 might_sleep_if(gfpflags_allow_blocking(pri));
1288
1289 if (skb_cloned(skb))
1290 return pskb_expand_head(skb, 0, 0, pri);
1291
1292 return 0;
1293 }
1294
1295 /**
1296 * skb_header_cloned - is the header a clone
1297 * @skb: buffer to check
1298 *
1299 * Returns true if modifying the header part of the buffer requires
1300 * the data to be copied.
1301 */
1302 static inline int skb_header_cloned(const struct sk_buff *skb)
1303 {
1304 int dataref;
1305
1306 if (!skb->cloned)
1307 return 0;
1308
1309 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1310 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1311 return dataref != 1;
1312 }
1313
1314 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1315 {
1316 might_sleep_if(gfpflags_allow_blocking(pri));
1317
1318 if (skb_header_cloned(skb))
1319 return pskb_expand_head(skb, 0, 0, pri);
1320
1321 return 0;
1322 }
1323
1324 /**
1325 * skb_header_release - release reference to header
1326 * @skb: buffer to operate on
1327 *
1328 * Drop a reference to the header part of the buffer. This is done
1329 * by acquiring a payload reference. You must not read from the header
1330 * part of skb->data after this.
1331 * Note : Check if you can use __skb_header_release() instead.
1332 */
1333 static inline void skb_header_release(struct sk_buff *skb)
1334 {
1335 BUG_ON(skb->nohdr);
1336 skb->nohdr = 1;
1337 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1338 }
1339
1340 /**
1341 * __skb_header_release - release reference to header
1342 * @skb: buffer to operate on
1343 *
1344 * Variant of skb_header_release() assuming skb is private to caller.
1345 * We can avoid one atomic operation.
1346 */
1347 static inline void __skb_header_release(struct sk_buff *skb)
1348 {
1349 skb->nohdr = 1;
1350 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1351 }
1352
1353
1354 /**
1355 * skb_shared - is the buffer shared
1356 * @skb: buffer to check
1357 *
1358 * Returns true if more than one person has a reference to this
1359 * buffer.
1360 */
1361 static inline int skb_shared(const struct sk_buff *skb)
1362 {
1363 return atomic_read(&skb->users) != 1;
1364 }
1365
1366 /**
1367 * skb_share_check - check if buffer is shared and if so clone it
1368 * @skb: buffer to check
1369 * @pri: priority for memory allocation
1370 *
1371 * If the buffer is shared the buffer is cloned and the old copy
1372 * drops a reference. A new clone with a single reference is returned.
1373 * If the buffer is not shared the original buffer is returned. When
1374 * being called from interrupt status or with spinlocks held pri must
1375 * be GFP_ATOMIC.
1376 *
1377 * NULL is returned on a memory allocation failure.
1378 */
1379 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1380 {
1381 might_sleep_if(gfpflags_allow_blocking(pri));
1382 if (skb_shared(skb)) {
1383 struct sk_buff *nskb = skb_clone(skb, pri);
1384
1385 if (likely(nskb))
1386 consume_skb(skb);
1387 else
1388 kfree_skb(skb);
1389 skb = nskb;
1390 }
1391 return skb;
1392 }
1393
1394 /*
1395 * Copy shared buffers into a new sk_buff. We effectively do COW on
1396 * packets to handle cases where we have a local reader and forward
1397 * and a couple of other messy ones. The normal one is tcpdumping
1398 * a packet thats being forwarded.
1399 */
1400
1401 /**
1402 * skb_unshare - make a copy of a shared buffer
1403 * @skb: buffer to check
1404 * @pri: priority for memory allocation
1405 *
1406 * If the socket buffer is a clone then this function creates a new
1407 * copy of the data, drops a reference count on the old copy and returns
1408 * the new copy with the reference count at 1. If the buffer is not a clone
1409 * the original buffer is returned. When called with a spinlock held or
1410 * from interrupt state @pri must be %GFP_ATOMIC
1411 *
1412 * %NULL is returned on a memory allocation failure.
1413 */
1414 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1415 gfp_t pri)
1416 {
1417 might_sleep_if(gfpflags_allow_blocking(pri));
1418 if (skb_cloned(skb)) {
1419 struct sk_buff *nskb = skb_copy(skb, pri);
1420
1421 /* Free our shared copy */
1422 if (likely(nskb))
1423 consume_skb(skb);
1424 else
1425 kfree_skb(skb);
1426 skb = nskb;
1427 }
1428 return skb;
1429 }
1430
1431 /**
1432 * skb_peek - peek at the head of an &sk_buff_head
1433 * @list_: list to peek at
1434 *
1435 * Peek an &sk_buff. Unlike most other operations you _MUST_
1436 * be careful with this one. A peek leaves the buffer on the
1437 * list and someone else may run off with it. You must hold
1438 * the appropriate locks or have a private queue to do this.
1439 *
1440 * Returns %NULL for an empty list or a pointer to the head element.
1441 * The reference count is not incremented and the reference is therefore
1442 * volatile. Use with caution.
1443 */
1444 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1445 {
1446 struct sk_buff *skb = list_->next;
1447
1448 if (skb == (struct sk_buff *)list_)
1449 skb = NULL;
1450 return skb;
1451 }
1452
1453 /**
1454 * skb_peek_next - peek skb following the given one from a queue
1455 * @skb: skb to start from
1456 * @list_: list to peek at
1457 *
1458 * Returns %NULL when the end of the list is met or a pointer to the
1459 * next element. The reference count is not incremented and the
1460 * reference is therefore volatile. Use with caution.
1461 */
1462 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1463 const struct sk_buff_head *list_)
1464 {
1465 struct sk_buff *next = skb->next;
1466
1467 if (next == (struct sk_buff *)list_)
1468 next = NULL;
1469 return next;
1470 }
1471
1472 /**
1473 * skb_peek_tail - peek at the tail of an &sk_buff_head
1474 * @list_: list to peek at
1475 *
1476 * Peek an &sk_buff. Unlike most other operations you _MUST_
1477 * be careful with this one. A peek leaves the buffer on the
1478 * list and someone else may run off with it. You must hold
1479 * the appropriate locks or have a private queue to do this.
1480 *
1481 * Returns %NULL for an empty list or a pointer to the tail element.
1482 * The reference count is not incremented and the reference is therefore
1483 * volatile. Use with caution.
1484 */
1485 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1486 {
1487 struct sk_buff *skb = list_->prev;
1488
1489 if (skb == (struct sk_buff *)list_)
1490 skb = NULL;
1491 return skb;
1492
1493 }
1494
1495 /**
1496 * skb_queue_len - get queue length
1497 * @list_: list to measure
1498 *
1499 * Return the length of an &sk_buff queue.
1500 */
1501 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1502 {
1503 return list_->qlen;
1504 }
1505
1506 /**
1507 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1508 * @list: queue to initialize
1509 *
1510 * This initializes only the list and queue length aspects of
1511 * an sk_buff_head object. This allows to initialize the list
1512 * aspects of an sk_buff_head without reinitializing things like
1513 * the spinlock. It can also be used for on-stack sk_buff_head
1514 * objects where the spinlock is known to not be used.
1515 */
1516 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1517 {
1518 list->prev = list->next = (struct sk_buff *)list;
1519 list->qlen = 0;
1520 }
1521
1522 /*
1523 * This function creates a split out lock class for each invocation;
1524 * this is needed for now since a whole lot of users of the skb-queue
1525 * infrastructure in drivers have different locking usage (in hardirq)
1526 * than the networking core (in softirq only). In the long run either the
1527 * network layer or drivers should need annotation to consolidate the
1528 * main types of usage into 3 classes.
1529 */
1530 static inline void skb_queue_head_init(struct sk_buff_head *list)
1531 {
1532 spin_lock_init(&list->lock);
1533 __skb_queue_head_init(list);
1534 }
1535
1536 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1537 struct lock_class_key *class)
1538 {
1539 skb_queue_head_init(list);
1540 lockdep_set_class(&list->lock, class);
1541 }
1542
1543 /*
1544 * Insert an sk_buff on a list.
1545 *
1546 * The "__skb_xxxx()" functions are the non-atomic ones that
1547 * can only be called with interrupts disabled.
1548 */
1549 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1550 struct sk_buff_head *list);
1551 static inline void __skb_insert(struct sk_buff *newsk,
1552 struct sk_buff *prev, struct sk_buff *next,
1553 struct sk_buff_head *list)
1554 {
1555 newsk->next = next;
1556 newsk->prev = prev;
1557 next->prev = prev->next = newsk;
1558 list->qlen++;
1559 }
1560
1561 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1562 struct sk_buff *prev,
1563 struct sk_buff *next)
1564 {
1565 struct sk_buff *first = list->next;
1566 struct sk_buff *last = list->prev;
1567
1568 first->prev = prev;
1569 prev->next = first;
1570
1571 last->next = next;
1572 next->prev = last;
1573 }
1574
1575 /**
1576 * skb_queue_splice - join two skb lists, this is designed for stacks
1577 * @list: the new list to add
1578 * @head: the place to add it in the first list
1579 */
1580 static inline void skb_queue_splice(const struct sk_buff_head *list,
1581 struct sk_buff_head *head)
1582 {
1583 if (!skb_queue_empty(list)) {
1584 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1585 head->qlen += list->qlen;
1586 }
1587 }
1588
1589 /**
1590 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1591 * @list: the new list to add
1592 * @head: the place to add it in the first list
1593 *
1594 * The list at @list is reinitialised
1595 */
1596 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1597 struct sk_buff_head *head)
1598 {
1599 if (!skb_queue_empty(list)) {
1600 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1601 head->qlen += list->qlen;
1602 __skb_queue_head_init(list);
1603 }
1604 }
1605
1606 /**
1607 * skb_queue_splice_tail - join two skb lists, each list being a queue
1608 * @list: the new list to add
1609 * @head: the place to add it in the first list
1610 */
1611 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1612 struct sk_buff_head *head)
1613 {
1614 if (!skb_queue_empty(list)) {
1615 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1616 head->qlen += list->qlen;
1617 }
1618 }
1619
1620 /**
1621 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1622 * @list: the new list to add
1623 * @head: the place to add it in the first list
1624 *
1625 * Each of the lists is a queue.
1626 * The list at @list is reinitialised
1627 */
1628 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1629 struct sk_buff_head *head)
1630 {
1631 if (!skb_queue_empty(list)) {
1632 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1633 head->qlen += list->qlen;
1634 __skb_queue_head_init(list);
1635 }
1636 }
1637
1638 /**
1639 * __skb_queue_after - queue a buffer at the list head
1640 * @list: list to use
1641 * @prev: place after this buffer
1642 * @newsk: buffer to queue
1643 *
1644 * Queue a buffer int the middle of a list. This function takes no locks
1645 * and you must therefore hold required locks before calling it.
1646 *
1647 * A buffer cannot be placed on two lists at the same time.
1648 */
1649 static inline void __skb_queue_after(struct sk_buff_head *list,
1650 struct sk_buff *prev,
1651 struct sk_buff *newsk)
1652 {
1653 __skb_insert(newsk, prev, prev->next, list);
1654 }
1655
1656 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1657 struct sk_buff_head *list);
1658
1659 static inline void __skb_queue_before(struct sk_buff_head *list,
1660 struct sk_buff *next,
1661 struct sk_buff *newsk)
1662 {
1663 __skb_insert(newsk, next->prev, next, list);
1664 }
1665
1666 /**
1667 * __skb_queue_head - queue a buffer at the list head
1668 * @list: list to use
1669 * @newsk: buffer to queue
1670 *
1671 * Queue a buffer at the start of a list. This function takes no locks
1672 * and you must therefore hold required locks before calling it.
1673 *
1674 * A buffer cannot be placed on two lists at the same time.
1675 */
1676 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1677 static inline void __skb_queue_head(struct sk_buff_head *list,
1678 struct sk_buff *newsk)
1679 {
1680 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1681 }
1682
1683 /**
1684 * __skb_queue_tail - queue a buffer at the list tail
1685 * @list: list to use
1686 * @newsk: buffer to queue
1687 *
1688 * Queue a buffer at the end of a list. This function takes no locks
1689 * and you must therefore hold required locks before calling it.
1690 *
1691 * A buffer cannot be placed on two lists at the same time.
1692 */
1693 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1694 static inline void __skb_queue_tail(struct sk_buff_head *list,
1695 struct sk_buff *newsk)
1696 {
1697 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1698 }
1699
1700 /*
1701 * remove sk_buff from list. _Must_ be called atomically, and with
1702 * the list known..
1703 */
1704 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1705 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1706 {
1707 struct sk_buff *next, *prev;
1708
1709 list->qlen--;
1710 next = skb->next;
1711 prev = skb->prev;
1712 skb->next = skb->prev = NULL;
1713 next->prev = prev;
1714 prev->next = next;
1715 }
1716
1717 /**
1718 * __skb_dequeue - remove from the head of the queue
1719 * @list: list to dequeue from
1720 *
1721 * Remove the head of the list. This function does not take any locks
1722 * so must be used with appropriate locks held only. The head item is
1723 * returned or %NULL if the list is empty.
1724 */
1725 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1726 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1727 {
1728 struct sk_buff *skb = skb_peek(list);
1729 if (skb)
1730 __skb_unlink(skb, list);
1731 return skb;
1732 }
1733
1734 /**
1735 * __skb_dequeue_tail - remove from the tail of the queue
1736 * @list: list to dequeue from
1737 *
1738 * Remove the tail of the list. This function does not take any locks
1739 * so must be used with appropriate locks held only. The tail item is
1740 * returned or %NULL if the list is empty.
1741 */
1742 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1743 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1744 {
1745 struct sk_buff *skb = skb_peek_tail(list);
1746 if (skb)
1747 __skb_unlink(skb, list);
1748 return skb;
1749 }
1750
1751
1752 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1753 {
1754 return skb->data_len;
1755 }
1756
1757 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1758 {
1759 return skb->len - skb->data_len;
1760 }
1761
1762 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1763 {
1764 unsigned int i, len = 0;
1765
1766 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1767 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1768 return len + skb_headlen(skb);
1769 }
1770
1771 /**
1772 * __skb_fill_page_desc - initialise a paged fragment in an skb
1773 * @skb: buffer containing fragment to be initialised
1774 * @i: paged fragment index to initialise
1775 * @page: the page to use for this fragment
1776 * @off: the offset to the data with @page
1777 * @size: the length of the data
1778 *
1779 * Initialises the @i'th fragment of @skb to point to &size bytes at
1780 * offset @off within @page.
1781 *
1782 * Does not take any additional reference on the fragment.
1783 */
1784 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1785 struct page *page, int off, int size)
1786 {
1787 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1788
1789 /*
1790 * Propagate page pfmemalloc to the skb if we can. The problem is
1791 * that not all callers have unique ownership of the page but rely
1792 * on page_is_pfmemalloc doing the right thing(tm).
1793 */
1794 frag->page.p = page;
1795 frag->page_offset = off;
1796 skb_frag_size_set(frag, size);
1797
1798 page = compound_head(page);
1799 if (page_is_pfmemalloc(page))
1800 skb->pfmemalloc = true;
1801 }
1802
1803 /**
1804 * skb_fill_page_desc - initialise a paged fragment in an skb
1805 * @skb: buffer containing fragment to be initialised
1806 * @i: paged fragment index to initialise
1807 * @page: the page to use for this fragment
1808 * @off: the offset to the data with @page
1809 * @size: the length of the data
1810 *
1811 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1812 * @skb to point to @size bytes at offset @off within @page. In
1813 * addition updates @skb such that @i is the last fragment.
1814 *
1815 * Does not take any additional reference on the fragment.
1816 */
1817 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1818 struct page *page, int off, int size)
1819 {
1820 __skb_fill_page_desc(skb, i, page, off, size);
1821 skb_shinfo(skb)->nr_frags = i + 1;
1822 }
1823
1824 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1825 int size, unsigned int truesize);
1826
1827 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1828 unsigned int truesize);
1829
1830 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1831 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1832 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1833
1834 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1835 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1836 {
1837 return skb->head + skb->tail;
1838 }
1839
1840 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1841 {
1842 skb->tail = skb->data - skb->head;
1843 }
1844
1845 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1846 {
1847 skb_reset_tail_pointer(skb);
1848 skb->tail += offset;
1849 }
1850
1851 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1852 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1853 {
1854 return skb->tail;
1855 }
1856
1857 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1858 {
1859 skb->tail = skb->data;
1860 }
1861
1862 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1863 {
1864 skb->tail = skb->data + offset;
1865 }
1866
1867 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1868
1869 /*
1870 * Add data to an sk_buff
1871 */
1872 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1873 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1874 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1875 {
1876 unsigned char *tmp = skb_tail_pointer(skb);
1877 SKB_LINEAR_ASSERT(skb);
1878 skb->tail += len;
1879 skb->len += len;
1880 return tmp;
1881 }
1882
1883 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1884 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1885 {
1886 skb->data -= len;
1887 skb->len += len;
1888 return skb->data;
1889 }
1890
1891 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1892 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1893 {
1894 skb->len -= len;
1895 BUG_ON(skb->len < skb->data_len);
1896 return skb->data += len;
1897 }
1898
1899 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1900 {
1901 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1902 }
1903
1904 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1905
1906 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1907 {
1908 if (len > skb_headlen(skb) &&
1909 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1910 return NULL;
1911 skb->len -= len;
1912 return skb->data += len;
1913 }
1914
1915 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1916 {
1917 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1918 }
1919
1920 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1921 {
1922 if (likely(len <= skb_headlen(skb)))
1923 return 1;
1924 if (unlikely(len > skb->len))
1925 return 0;
1926 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1927 }
1928
1929 void skb_condense(struct sk_buff *skb);
1930
1931 /**
1932 * skb_headroom - bytes at buffer head
1933 * @skb: buffer to check
1934 *
1935 * Return the number of bytes of free space at the head of an &sk_buff.
1936 */
1937 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1938 {
1939 return skb->data - skb->head;
1940 }
1941
1942 /**
1943 * skb_tailroom - bytes at buffer end
1944 * @skb: buffer to check
1945 *
1946 * Return the number of bytes of free space at the tail of an sk_buff
1947 */
1948 static inline int skb_tailroom(const struct sk_buff *skb)
1949 {
1950 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1951 }
1952
1953 /**
1954 * skb_availroom - bytes at buffer end
1955 * @skb: buffer to check
1956 *
1957 * Return the number of bytes of free space at the tail of an sk_buff
1958 * allocated by sk_stream_alloc()
1959 */
1960 static inline int skb_availroom(const struct sk_buff *skb)
1961 {
1962 if (skb_is_nonlinear(skb))
1963 return 0;
1964
1965 return skb->end - skb->tail - skb->reserved_tailroom;
1966 }
1967
1968 /**
1969 * skb_reserve - adjust headroom
1970 * @skb: buffer to alter
1971 * @len: bytes to move
1972 *
1973 * Increase the headroom of an empty &sk_buff by reducing the tail
1974 * room. This is only allowed for an empty buffer.
1975 */
1976 static inline void skb_reserve(struct sk_buff *skb, int len)
1977 {
1978 skb->data += len;
1979 skb->tail += len;
1980 }
1981
1982 /**
1983 * skb_tailroom_reserve - adjust reserved_tailroom
1984 * @skb: buffer to alter
1985 * @mtu: maximum amount of headlen permitted
1986 * @needed_tailroom: minimum amount of reserved_tailroom
1987 *
1988 * Set reserved_tailroom so that headlen can be as large as possible but
1989 * not larger than mtu and tailroom cannot be smaller than
1990 * needed_tailroom.
1991 * The required headroom should already have been reserved before using
1992 * this function.
1993 */
1994 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
1995 unsigned int needed_tailroom)
1996 {
1997 SKB_LINEAR_ASSERT(skb);
1998 if (mtu < skb_tailroom(skb) - needed_tailroom)
1999 /* use at most mtu */
2000 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2001 else
2002 /* use up to all available space */
2003 skb->reserved_tailroom = needed_tailroom;
2004 }
2005
2006 #define ENCAP_TYPE_ETHER 0
2007 #define ENCAP_TYPE_IPPROTO 1
2008
2009 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2010 __be16 protocol)
2011 {
2012 skb->inner_protocol = protocol;
2013 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2014 }
2015
2016 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2017 __u8 ipproto)
2018 {
2019 skb->inner_ipproto = ipproto;
2020 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2021 }
2022
2023 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2024 {
2025 skb->inner_mac_header = skb->mac_header;
2026 skb->inner_network_header = skb->network_header;
2027 skb->inner_transport_header = skb->transport_header;
2028 }
2029
2030 static inline void skb_reset_mac_len(struct sk_buff *skb)
2031 {
2032 skb->mac_len = skb->network_header - skb->mac_header;
2033 }
2034
2035 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2036 *skb)
2037 {
2038 return skb->head + skb->inner_transport_header;
2039 }
2040
2041 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2042 {
2043 return skb_inner_transport_header(skb) - skb->data;
2044 }
2045
2046 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2047 {
2048 skb->inner_transport_header = skb->data - skb->head;
2049 }
2050
2051 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2052 const int offset)
2053 {
2054 skb_reset_inner_transport_header(skb);
2055 skb->inner_transport_header += offset;
2056 }
2057
2058 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2059 {
2060 return skb->head + skb->inner_network_header;
2061 }
2062
2063 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2064 {
2065 skb->inner_network_header = skb->data - skb->head;
2066 }
2067
2068 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2069 const int offset)
2070 {
2071 skb_reset_inner_network_header(skb);
2072 skb->inner_network_header += offset;
2073 }
2074
2075 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2076 {
2077 return skb->head + skb->inner_mac_header;
2078 }
2079
2080 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2081 {
2082 skb->inner_mac_header = skb->data - skb->head;
2083 }
2084
2085 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2086 const int offset)
2087 {
2088 skb_reset_inner_mac_header(skb);
2089 skb->inner_mac_header += offset;
2090 }
2091 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2092 {
2093 return skb->transport_header != (typeof(skb->transport_header))~0U;
2094 }
2095
2096 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2097 {
2098 return skb->head + skb->transport_header;
2099 }
2100
2101 static inline void skb_reset_transport_header(struct sk_buff *skb)
2102 {
2103 skb->transport_header = skb->data - skb->head;
2104 }
2105
2106 static inline void skb_set_transport_header(struct sk_buff *skb,
2107 const int offset)
2108 {
2109 skb_reset_transport_header(skb);
2110 skb->transport_header += offset;
2111 }
2112
2113 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2114 {
2115 return skb->head + skb->network_header;
2116 }
2117
2118 static inline void skb_reset_network_header(struct sk_buff *skb)
2119 {
2120 skb->network_header = skb->data - skb->head;
2121 }
2122
2123 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2124 {
2125 skb_reset_network_header(skb);
2126 skb->network_header += offset;
2127 }
2128
2129 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2130 {
2131 return skb->head + skb->mac_header;
2132 }
2133
2134 static inline int skb_mac_offset(const struct sk_buff *skb)
2135 {
2136 return skb_mac_header(skb) - skb->data;
2137 }
2138
2139 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2140 {
2141 return skb->mac_header != (typeof(skb->mac_header))~0U;
2142 }
2143
2144 static inline void skb_reset_mac_header(struct sk_buff *skb)
2145 {
2146 skb->mac_header = skb->data - skb->head;
2147 }
2148
2149 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2150 {
2151 skb_reset_mac_header(skb);
2152 skb->mac_header += offset;
2153 }
2154
2155 static inline void skb_pop_mac_header(struct sk_buff *skb)
2156 {
2157 skb->mac_header = skb->network_header;
2158 }
2159
2160 static inline void skb_probe_transport_header(struct sk_buff *skb,
2161 const int offset_hint)
2162 {
2163 struct flow_keys keys;
2164
2165 if (skb_transport_header_was_set(skb))
2166 return;
2167 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2168 skb_set_transport_header(skb, keys.control.thoff);
2169 else
2170 skb_set_transport_header(skb, offset_hint);
2171 }
2172
2173 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2174 {
2175 if (skb_mac_header_was_set(skb)) {
2176 const unsigned char *old_mac = skb_mac_header(skb);
2177
2178 skb_set_mac_header(skb, -skb->mac_len);
2179 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2180 }
2181 }
2182
2183 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2184 {
2185 return skb->csum_start - skb_headroom(skb);
2186 }
2187
2188 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2189 {
2190 return skb->head + skb->csum_start;
2191 }
2192
2193 static inline int skb_transport_offset(const struct sk_buff *skb)
2194 {
2195 return skb_transport_header(skb) - skb->data;
2196 }
2197
2198 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2199 {
2200 return skb->transport_header - skb->network_header;
2201 }
2202
2203 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2204 {
2205 return skb->inner_transport_header - skb->inner_network_header;
2206 }
2207
2208 static inline int skb_network_offset(const struct sk_buff *skb)
2209 {
2210 return skb_network_header(skb) - skb->data;
2211 }
2212
2213 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2214 {
2215 return skb_inner_network_header(skb) - skb->data;
2216 }
2217
2218 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2219 {
2220 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2221 }
2222
2223 /*
2224 * CPUs often take a performance hit when accessing unaligned memory
2225 * locations. The actual performance hit varies, it can be small if the
2226 * hardware handles it or large if we have to take an exception and fix it
2227 * in software.
2228 *
2229 * Since an ethernet header is 14 bytes network drivers often end up with
2230 * the IP header at an unaligned offset. The IP header can be aligned by
2231 * shifting the start of the packet by 2 bytes. Drivers should do this
2232 * with:
2233 *
2234 * skb_reserve(skb, NET_IP_ALIGN);
2235 *
2236 * The downside to this alignment of the IP header is that the DMA is now
2237 * unaligned. On some architectures the cost of an unaligned DMA is high
2238 * and this cost outweighs the gains made by aligning the IP header.
2239 *
2240 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2241 * to be overridden.
2242 */
2243 #ifndef NET_IP_ALIGN
2244 #define NET_IP_ALIGN 2
2245 #endif
2246
2247 /*
2248 * The networking layer reserves some headroom in skb data (via
2249 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2250 * the header has to grow. In the default case, if the header has to grow
2251 * 32 bytes or less we avoid the reallocation.
2252 *
2253 * Unfortunately this headroom changes the DMA alignment of the resulting
2254 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2255 * on some architectures. An architecture can override this value,
2256 * perhaps setting it to a cacheline in size (since that will maintain
2257 * cacheline alignment of the DMA). It must be a power of 2.
2258 *
2259 * Various parts of the networking layer expect at least 32 bytes of
2260 * headroom, you should not reduce this.
2261 *
2262 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2263 * to reduce average number of cache lines per packet.
2264 * get_rps_cpus() for example only access one 64 bytes aligned block :
2265 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2266 */
2267 #ifndef NET_SKB_PAD
2268 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2269 #endif
2270
2271 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2272
2273 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2274 {
2275 if (unlikely(skb_is_nonlinear(skb))) {
2276 WARN_ON(1);
2277 return;
2278 }
2279 skb->len = len;
2280 skb_set_tail_pointer(skb, len);
2281 }
2282
2283 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2284 {
2285 __skb_set_length(skb, len);
2286 }
2287
2288 void skb_trim(struct sk_buff *skb, unsigned int len);
2289
2290 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2291 {
2292 if (skb->data_len)
2293 return ___pskb_trim(skb, len);
2294 __skb_trim(skb, len);
2295 return 0;
2296 }
2297
2298 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2299 {
2300 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2301 }
2302
2303 /**
2304 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2305 * @skb: buffer to alter
2306 * @len: new length
2307 *
2308 * This is identical to pskb_trim except that the caller knows that
2309 * the skb is not cloned so we should never get an error due to out-
2310 * of-memory.
2311 */
2312 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2313 {
2314 int err = pskb_trim(skb, len);
2315 BUG_ON(err);
2316 }
2317
2318 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2319 {
2320 unsigned int diff = len - skb->len;
2321
2322 if (skb_tailroom(skb) < diff) {
2323 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2324 GFP_ATOMIC);
2325 if (ret)
2326 return ret;
2327 }
2328 __skb_set_length(skb, len);
2329 return 0;
2330 }
2331
2332 /**
2333 * skb_orphan - orphan a buffer
2334 * @skb: buffer to orphan
2335 *
2336 * If a buffer currently has an owner then we call the owner's
2337 * destructor function and make the @skb unowned. The buffer continues
2338 * to exist but is no longer charged to its former owner.
2339 */
2340 static inline void skb_orphan(struct sk_buff *skb)
2341 {
2342 if (skb->destructor) {
2343 skb->destructor(skb);
2344 skb->destructor = NULL;
2345 skb->sk = NULL;
2346 } else {
2347 BUG_ON(skb->sk);
2348 }
2349 }
2350
2351 /**
2352 * skb_orphan_frags - orphan the frags contained in a buffer
2353 * @skb: buffer to orphan frags from
2354 * @gfp_mask: allocation mask for replacement pages
2355 *
2356 * For each frag in the SKB which needs a destructor (i.e. has an
2357 * owner) create a copy of that frag and release the original
2358 * page by calling the destructor.
2359 */
2360 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2361 {
2362 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2363 return 0;
2364 return skb_copy_ubufs(skb, gfp_mask);
2365 }
2366
2367 /**
2368 * __skb_queue_purge - empty a list
2369 * @list: list to empty
2370 *
2371 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2372 * the list and one reference dropped. This function does not take the
2373 * list lock and the caller must hold the relevant locks to use it.
2374 */
2375 void skb_queue_purge(struct sk_buff_head *list);
2376 static inline void __skb_queue_purge(struct sk_buff_head *list)
2377 {
2378 struct sk_buff *skb;
2379 while ((skb = __skb_dequeue(list)) != NULL)
2380 kfree_skb(skb);
2381 }
2382
2383 void skb_rbtree_purge(struct rb_root *root);
2384
2385 void *netdev_alloc_frag(unsigned int fragsz);
2386
2387 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2388 gfp_t gfp_mask);
2389
2390 /**
2391 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2392 * @dev: network device to receive on
2393 * @length: length to allocate
2394 *
2395 * Allocate a new &sk_buff and assign it a usage count of one. The
2396 * buffer has unspecified headroom built in. Users should allocate
2397 * the headroom they think they need without accounting for the
2398 * built in space. The built in space is used for optimisations.
2399 *
2400 * %NULL is returned if there is no free memory. Although this function
2401 * allocates memory it can be called from an interrupt.
2402 */
2403 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2404 unsigned int length)
2405 {
2406 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2407 }
2408
2409 /* legacy helper around __netdev_alloc_skb() */
2410 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2411 gfp_t gfp_mask)
2412 {
2413 return __netdev_alloc_skb(NULL, length, gfp_mask);
2414 }
2415
2416 /* legacy helper around netdev_alloc_skb() */
2417 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2418 {
2419 return netdev_alloc_skb(NULL, length);
2420 }
2421
2422
2423 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2424 unsigned int length, gfp_t gfp)
2425 {
2426 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2427
2428 if (NET_IP_ALIGN && skb)
2429 skb_reserve(skb, NET_IP_ALIGN);
2430 return skb;
2431 }
2432
2433 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2434 unsigned int length)
2435 {
2436 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2437 }
2438
2439 static inline void skb_free_frag(void *addr)
2440 {
2441 page_frag_free(addr);
2442 }
2443
2444 void *napi_alloc_frag(unsigned int fragsz);
2445 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2446 unsigned int length, gfp_t gfp_mask);
2447 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2448 unsigned int length)
2449 {
2450 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2451 }
2452 void napi_consume_skb(struct sk_buff *skb, int budget);
2453
2454 void __kfree_skb_flush(void);
2455 void __kfree_skb_defer(struct sk_buff *skb);
2456
2457 /**
2458 * __dev_alloc_pages - allocate page for network Rx
2459 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2460 * @order: size of the allocation
2461 *
2462 * Allocate a new page.
2463 *
2464 * %NULL is returned if there is no free memory.
2465 */
2466 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2467 unsigned int order)
2468 {
2469 /* This piece of code contains several assumptions.
2470 * 1. This is for device Rx, therefor a cold page is preferred.
2471 * 2. The expectation is the user wants a compound page.
2472 * 3. If requesting a order 0 page it will not be compound
2473 * due to the check to see if order has a value in prep_new_page
2474 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2475 * code in gfp_to_alloc_flags that should be enforcing this.
2476 */
2477 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2478
2479 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2480 }
2481
2482 static inline struct page *dev_alloc_pages(unsigned int order)
2483 {
2484 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2485 }
2486
2487 /**
2488 * __dev_alloc_page - allocate a page for network Rx
2489 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2490 *
2491 * Allocate a new page.
2492 *
2493 * %NULL is returned if there is no free memory.
2494 */
2495 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2496 {
2497 return __dev_alloc_pages(gfp_mask, 0);
2498 }
2499
2500 static inline struct page *dev_alloc_page(void)
2501 {
2502 return dev_alloc_pages(0);
2503 }
2504
2505 /**
2506 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2507 * @page: The page that was allocated from skb_alloc_page
2508 * @skb: The skb that may need pfmemalloc set
2509 */
2510 static inline void skb_propagate_pfmemalloc(struct page *page,
2511 struct sk_buff *skb)
2512 {
2513 if (page_is_pfmemalloc(page))
2514 skb->pfmemalloc = true;
2515 }
2516
2517 /**
2518 * skb_frag_page - retrieve the page referred to by a paged fragment
2519 * @frag: the paged fragment
2520 *
2521 * Returns the &struct page associated with @frag.
2522 */
2523 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2524 {
2525 return frag->page.p;
2526 }
2527
2528 /**
2529 * __skb_frag_ref - take an addition reference on a paged fragment.
2530 * @frag: the paged fragment
2531 *
2532 * Takes an additional reference on the paged fragment @frag.
2533 */
2534 static inline void __skb_frag_ref(skb_frag_t *frag)
2535 {
2536 get_page(skb_frag_page(frag));
2537 }
2538
2539 /**
2540 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2541 * @skb: the buffer
2542 * @f: the fragment offset.
2543 *
2544 * Takes an additional reference on the @f'th paged fragment of @skb.
2545 */
2546 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2547 {
2548 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2549 }
2550
2551 /**
2552 * __skb_frag_unref - release a reference on a paged fragment.
2553 * @frag: the paged fragment
2554 *
2555 * Releases a reference on the paged fragment @frag.
2556 */
2557 static inline void __skb_frag_unref(skb_frag_t *frag)
2558 {
2559 put_page(skb_frag_page(frag));
2560 }
2561
2562 /**
2563 * skb_frag_unref - release a reference on a paged fragment of an skb.
2564 * @skb: the buffer
2565 * @f: the fragment offset
2566 *
2567 * Releases a reference on the @f'th paged fragment of @skb.
2568 */
2569 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2570 {
2571 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2572 }
2573
2574 /**
2575 * skb_frag_address - gets the address of the data contained in a paged fragment
2576 * @frag: the paged fragment buffer
2577 *
2578 * Returns the address of the data within @frag. The page must already
2579 * be mapped.
2580 */
2581 static inline void *skb_frag_address(const skb_frag_t *frag)
2582 {
2583 return page_address(skb_frag_page(frag)) + frag->page_offset;
2584 }
2585
2586 /**
2587 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2588 * @frag: the paged fragment buffer
2589 *
2590 * Returns the address of the data within @frag. Checks that the page
2591 * is mapped and returns %NULL otherwise.
2592 */
2593 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2594 {
2595 void *ptr = page_address(skb_frag_page(frag));
2596 if (unlikely(!ptr))
2597 return NULL;
2598
2599 return ptr + frag->page_offset;
2600 }
2601
2602 /**
2603 * __skb_frag_set_page - sets the page contained in a paged fragment
2604 * @frag: the paged fragment
2605 * @page: the page to set
2606 *
2607 * Sets the fragment @frag to contain @page.
2608 */
2609 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2610 {
2611 frag->page.p = page;
2612 }
2613
2614 /**
2615 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2616 * @skb: the buffer
2617 * @f: the fragment offset
2618 * @page: the page to set
2619 *
2620 * Sets the @f'th fragment of @skb to contain @page.
2621 */
2622 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2623 struct page *page)
2624 {
2625 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2626 }
2627
2628 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2629
2630 /**
2631 * skb_frag_dma_map - maps a paged fragment via the DMA API
2632 * @dev: the device to map the fragment to
2633 * @frag: the paged fragment to map
2634 * @offset: the offset within the fragment (starting at the
2635 * fragment's own offset)
2636 * @size: the number of bytes to map
2637 * @dir: the direction of the mapping (%PCI_DMA_*)
2638 *
2639 * Maps the page associated with @frag to @device.
2640 */
2641 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2642 const skb_frag_t *frag,
2643 size_t offset, size_t size,
2644 enum dma_data_direction dir)
2645 {
2646 return dma_map_page(dev, skb_frag_page(frag),
2647 frag->page_offset + offset, size, dir);
2648 }
2649
2650 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2651 gfp_t gfp_mask)
2652 {
2653 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2654 }
2655
2656
2657 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2658 gfp_t gfp_mask)
2659 {
2660 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2661 }
2662
2663
2664 /**
2665 * skb_clone_writable - is the header of a clone writable
2666 * @skb: buffer to check
2667 * @len: length up to which to write
2668 *
2669 * Returns true if modifying the header part of the cloned buffer
2670 * does not requires the data to be copied.
2671 */
2672 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2673 {
2674 return !skb_header_cloned(skb) &&
2675 skb_headroom(skb) + len <= skb->hdr_len;
2676 }
2677
2678 static inline int skb_try_make_writable(struct sk_buff *skb,
2679 unsigned int write_len)
2680 {
2681 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2682 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2683 }
2684
2685 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2686 int cloned)
2687 {
2688 int delta = 0;
2689
2690 if (headroom > skb_headroom(skb))
2691 delta = headroom - skb_headroom(skb);
2692
2693 if (delta || cloned)
2694 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2695 GFP_ATOMIC);
2696 return 0;
2697 }
2698
2699 /**
2700 * skb_cow - copy header of skb when it is required
2701 * @skb: buffer to cow
2702 * @headroom: needed headroom
2703 *
2704 * If the skb passed lacks sufficient headroom or its data part
2705 * is shared, data is reallocated. If reallocation fails, an error
2706 * is returned and original skb is not changed.
2707 *
2708 * The result is skb with writable area skb->head...skb->tail
2709 * and at least @headroom of space at head.
2710 */
2711 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2712 {
2713 return __skb_cow(skb, headroom, skb_cloned(skb));
2714 }
2715
2716 /**
2717 * skb_cow_head - skb_cow but only making the head writable
2718 * @skb: buffer to cow
2719 * @headroom: needed headroom
2720 *
2721 * This function is identical to skb_cow except that we replace the
2722 * skb_cloned check by skb_header_cloned. It should be used when
2723 * you only need to push on some header and do not need to modify
2724 * the data.
2725 */
2726 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2727 {
2728 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2729 }
2730
2731 /**
2732 * skb_padto - pad an skbuff up to a minimal size
2733 * @skb: buffer to pad
2734 * @len: minimal length
2735 *
2736 * Pads up a buffer to ensure the trailing bytes exist and are
2737 * blanked. If the buffer already contains sufficient data it
2738 * is untouched. Otherwise it is extended. Returns zero on
2739 * success. The skb is freed on error.
2740 */
2741 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2742 {
2743 unsigned int size = skb->len;
2744 if (likely(size >= len))
2745 return 0;
2746 return skb_pad(skb, len - size);
2747 }
2748
2749 /**
2750 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2751 * @skb: buffer to pad
2752 * @len: minimal length
2753 *
2754 * Pads up a buffer to ensure the trailing bytes exist and are
2755 * blanked. If the buffer already contains sufficient data it
2756 * is untouched. Otherwise it is extended. Returns zero on
2757 * success. The skb is freed on error.
2758 */
2759 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2760 {
2761 unsigned int size = skb->len;
2762
2763 if (unlikely(size < len)) {
2764 len -= size;
2765 if (skb_pad(skb, len))
2766 return -ENOMEM;
2767 __skb_put(skb, len);
2768 }
2769 return 0;
2770 }
2771
2772 static inline int skb_add_data(struct sk_buff *skb,
2773 struct iov_iter *from, int copy)
2774 {
2775 const int off = skb->len;
2776
2777 if (skb->ip_summed == CHECKSUM_NONE) {
2778 __wsum csum = 0;
2779 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2780 &csum, from)) {
2781 skb->csum = csum_block_add(skb->csum, csum, off);
2782 return 0;
2783 }
2784 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2785 return 0;
2786
2787 __skb_trim(skb, off);
2788 return -EFAULT;
2789 }
2790
2791 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2792 const struct page *page, int off)
2793 {
2794 if (i) {
2795 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2796
2797 return page == skb_frag_page(frag) &&
2798 off == frag->page_offset + skb_frag_size(frag);
2799 }
2800 return false;
2801 }
2802
2803 static inline int __skb_linearize(struct sk_buff *skb)
2804 {
2805 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2806 }
2807
2808 /**
2809 * skb_linearize - convert paged skb to linear one
2810 * @skb: buffer to linarize
2811 *
2812 * If there is no free memory -ENOMEM is returned, otherwise zero
2813 * is returned and the old skb data released.
2814 */
2815 static inline int skb_linearize(struct sk_buff *skb)
2816 {
2817 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2818 }
2819
2820 /**
2821 * skb_has_shared_frag - can any frag be overwritten
2822 * @skb: buffer to test
2823 *
2824 * Return true if the skb has at least one frag that might be modified
2825 * by an external entity (as in vmsplice()/sendfile())
2826 */
2827 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2828 {
2829 return skb_is_nonlinear(skb) &&
2830 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2831 }
2832
2833 /**
2834 * skb_linearize_cow - make sure skb is linear and writable
2835 * @skb: buffer to process
2836 *
2837 * If there is no free memory -ENOMEM is returned, otherwise zero
2838 * is returned and the old skb data released.
2839 */
2840 static inline int skb_linearize_cow(struct sk_buff *skb)
2841 {
2842 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2843 __skb_linearize(skb) : 0;
2844 }
2845
2846 static __always_inline void
2847 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2848 unsigned int off)
2849 {
2850 if (skb->ip_summed == CHECKSUM_COMPLETE)
2851 skb->csum = csum_block_sub(skb->csum,
2852 csum_partial(start, len, 0), off);
2853 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2854 skb_checksum_start_offset(skb) < 0)
2855 skb->ip_summed = CHECKSUM_NONE;
2856 }
2857
2858 /**
2859 * skb_postpull_rcsum - update checksum for received skb after pull
2860 * @skb: buffer to update
2861 * @start: start of data before pull
2862 * @len: length of data pulled
2863 *
2864 * After doing a pull on a received packet, you need to call this to
2865 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2866 * CHECKSUM_NONE so that it can be recomputed from scratch.
2867 */
2868 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2869 const void *start, unsigned int len)
2870 {
2871 __skb_postpull_rcsum(skb, start, len, 0);
2872 }
2873
2874 static __always_inline void
2875 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2876 unsigned int off)
2877 {
2878 if (skb->ip_summed == CHECKSUM_COMPLETE)
2879 skb->csum = csum_block_add(skb->csum,
2880 csum_partial(start, len, 0), off);
2881 }
2882
2883 /**
2884 * skb_postpush_rcsum - update checksum for received skb after push
2885 * @skb: buffer to update
2886 * @start: start of data after push
2887 * @len: length of data pushed
2888 *
2889 * After doing a push on a received packet, you need to call this to
2890 * update the CHECKSUM_COMPLETE checksum.
2891 */
2892 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2893 const void *start, unsigned int len)
2894 {
2895 __skb_postpush_rcsum(skb, start, len, 0);
2896 }
2897
2898 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2899
2900 /**
2901 * skb_push_rcsum - push skb and update receive checksum
2902 * @skb: buffer to update
2903 * @len: length of data pulled
2904 *
2905 * This function performs an skb_push on the packet and updates
2906 * the CHECKSUM_COMPLETE checksum. It should be used on
2907 * receive path processing instead of skb_push unless you know
2908 * that the checksum difference is zero (e.g., a valid IP header)
2909 * or you are setting ip_summed to CHECKSUM_NONE.
2910 */
2911 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2912 unsigned int len)
2913 {
2914 skb_push(skb, len);
2915 skb_postpush_rcsum(skb, skb->data, len);
2916 return skb->data;
2917 }
2918
2919 /**
2920 * pskb_trim_rcsum - trim received skb and update checksum
2921 * @skb: buffer to trim
2922 * @len: new length
2923 *
2924 * This is exactly the same as pskb_trim except that it ensures the
2925 * checksum of received packets are still valid after the operation.
2926 */
2927
2928 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2929 {
2930 if (likely(len >= skb->len))
2931 return 0;
2932 if (skb->ip_summed == CHECKSUM_COMPLETE)
2933 skb->ip_summed = CHECKSUM_NONE;
2934 return __pskb_trim(skb, len);
2935 }
2936
2937 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2938 {
2939 if (skb->ip_summed == CHECKSUM_COMPLETE)
2940 skb->ip_summed = CHECKSUM_NONE;
2941 __skb_trim(skb, len);
2942 return 0;
2943 }
2944
2945 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
2946 {
2947 if (skb->ip_summed == CHECKSUM_COMPLETE)
2948 skb->ip_summed = CHECKSUM_NONE;
2949 return __skb_grow(skb, len);
2950 }
2951
2952 #define skb_queue_walk(queue, skb) \
2953 for (skb = (queue)->next; \
2954 skb != (struct sk_buff *)(queue); \
2955 skb = skb->next)
2956
2957 #define skb_queue_walk_safe(queue, skb, tmp) \
2958 for (skb = (queue)->next, tmp = skb->next; \
2959 skb != (struct sk_buff *)(queue); \
2960 skb = tmp, tmp = skb->next)
2961
2962 #define skb_queue_walk_from(queue, skb) \
2963 for (; skb != (struct sk_buff *)(queue); \
2964 skb = skb->next)
2965
2966 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2967 for (tmp = skb->next; \
2968 skb != (struct sk_buff *)(queue); \
2969 skb = tmp, tmp = skb->next)
2970
2971 #define skb_queue_reverse_walk(queue, skb) \
2972 for (skb = (queue)->prev; \
2973 skb != (struct sk_buff *)(queue); \
2974 skb = skb->prev)
2975
2976 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2977 for (skb = (queue)->prev, tmp = skb->prev; \
2978 skb != (struct sk_buff *)(queue); \
2979 skb = tmp, tmp = skb->prev)
2980
2981 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2982 for (tmp = skb->prev; \
2983 skb != (struct sk_buff *)(queue); \
2984 skb = tmp, tmp = skb->prev)
2985
2986 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2987 {
2988 return skb_shinfo(skb)->frag_list != NULL;
2989 }
2990
2991 static inline void skb_frag_list_init(struct sk_buff *skb)
2992 {
2993 skb_shinfo(skb)->frag_list = NULL;
2994 }
2995
2996 #define skb_walk_frags(skb, iter) \
2997 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2998
2999
3000 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3001 const struct sk_buff *skb);
3002 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3003 struct sk_buff_head *queue,
3004 unsigned int flags,
3005 void (*destructor)(struct sock *sk,
3006 struct sk_buff *skb),
3007 int *peeked, int *off, int *err,
3008 struct sk_buff **last);
3009 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3010 void (*destructor)(struct sock *sk,
3011 struct sk_buff *skb),
3012 int *peeked, int *off, int *err,
3013 struct sk_buff **last);
3014 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3015 void (*destructor)(struct sock *sk,
3016 struct sk_buff *skb),
3017 int *peeked, int *off, int *err);
3018 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3019 int *err);
3020 unsigned int datagram_poll(struct file *file, struct socket *sock,
3021 struct poll_table_struct *wait);
3022 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3023 struct iov_iter *to, int size);
3024 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3025 struct msghdr *msg, int size)
3026 {
3027 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3028 }
3029 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3030 struct msghdr *msg);
3031 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3032 struct iov_iter *from, int len);
3033 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3034 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3035 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3036 static inline void skb_free_datagram_locked(struct sock *sk,
3037 struct sk_buff *skb)
3038 {
3039 __skb_free_datagram_locked(sk, skb, 0);
3040 }
3041 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3042 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3043 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3044 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3045 int len, __wsum csum);
3046 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3047 struct pipe_inode_info *pipe, unsigned int len,
3048 unsigned int flags);
3049 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3050 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3051 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3052 int len, int hlen);
3053 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3054 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3055 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3056 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3057 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3058 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3059 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3060 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3061 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3062 int skb_vlan_pop(struct sk_buff *skb);
3063 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3064 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3065 gfp_t gfp);
3066
3067 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3068 {
3069 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3070 }
3071
3072 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3073 {
3074 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3075 }
3076
3077 struct skb_checksum_ops {
3078 __wsum (*update)(const void *mem, int len, __wsum wsum);
3079 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3080 };
3081
3082 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3083
3084 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3085 __wsum csum, const struct skb_checksum_ops *ops);
3086 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3087 __wsum csum);
3088
3089 static inline void * __must_check
3090 __skb_header_pointer(const struct sk_buff *skb, int offset,
3091 int len, void *data, int hlen, void *buffer)
3092 {
3093 if (hlen - offset >= len)
3094 return data + offset;
3095
3096 if (!skb ||
3097 skb_copy_bits(skb, offset, buffer, len) < 0)
3098 return NULL;
3099
3100 return buffer;
3101 }
3102
3103 static inline void * __must_check
3104 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3105 {
3106 return __skb_header_pointer(skb, offset, len, skb->data,
3107 skb_headlen(skb), buffer);
3108 }
3109
3110 /**
3111 * skb_needs_linearize - check if we need to linearize a given skb
3112 * depending on the given device features.
3113 * @skb: socket buffer to check
3114 * @features: net device features
3115 *
3116 * Returns true if either:
3117 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3118 * 2. skb is fragmented and the device does not support SG.
3119 */
3120 static inline bool skb_needs_linearize(struct sk_buff *skb,
3121 netdev_features_t features)
3122 {
3123 return skb_is_nonlinear(skb) &&
3124 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3125 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3126 }
3127
3128 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3129 void *to,
3130 const unsigned int len)
3131 {
3132 memcpy(to, skb->data, len);
3133 }
3134
3135 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3136 const int offset, void *to,
3137 const unsigned int len)
3138 {
3139 memcpy(to, skb->data + offset, len);
3140 }
3141
3142 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3143 const void *from,
3144 const unsigned int len)
3145 {
3146 memcpy(skb->data, from, len);
3147 }
3148
3149 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3150 const int offset,
3151 const void *from,
3152 const unsigned int len)
3153 {
3154 memcpy(skb->data + offset, from, len);
3155 }
3156
3157 void skb_init(void);
3158
3159 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3160 {
3161 return skb->tstamp;
3162 }
3163
3164 /**
3165 * skb_get_timestamp - get timestamp from a skb
3166 * @skb: skb to get stamp from
3167 * @stamp: pointer to struct timeval to store stamp in
3168 *
3169 * Timestamps are stored in the skb as offsets to a base timestamp.
3170 * This function converts the offset back to a struct timeval and stores
3171 * it in stamp.
3172 */
3173 static inline void skb_get_timestamp(const struct sk_buff *skb,
3174 struct timeval *stamp)
3175 {
3176 *stamp = ktime_to_timeval(skb->tstamp);
3177 }
3178
3179 static inline void skb_get_timestampns(const struct sk_buff *skb,
3180 struct timespec *stamp)
3181 {
3182 *stamp = ktime_to_timespec(skb->tstamp);
3183 }
3184
3185 static inline void __net_timestamp(struct sk_buff *skb)
3186 {
3187 skb->tstamp = ktime_get_real();
3188 }
3189
3190 static inline ktime_t net_timedelta(ktime_t t)
3191 {
3192 return ktime_sub(ktime_get_real(), t);
3193 }
3194
3195 static inline ktime_t net_invalid_timestamp(void)
3196 {
3197 return 0;
3198 }
3199
3200 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3201
3202 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3203
3204 void skb_clone_tx_timestamp(struct sk_buff *skb);
3205 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3206
3207 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3208
3209 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3210 {
3211 }
3212
3213 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3214 {
3215 return false;
3216 }
3217
3218 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3219
3220 /**
3221 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3222 *
3223 * PHY drivers may accept clones of transmitted packets for
3224 * timestamping via their phy_driver.txtstamp method. These drivers
3225 * must call this function to return the skb back to the stack with a
3226 * timestamp.
3227 *
3228 * @skb: clone of the the original outgoing packet
3229 * @hwtstamps: hardware time stamps
3230 *
3231 */
3232 void skb_complete_tx_timestamp(struct sk_buff *skb,
3233 struct skb_shared_hwtstamps *hwtstamps);
3234
3235 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3236 struct skb_shared_hwtstamps *hwtstamps,
3237 struct sock *sk, int tstype);
3238
3239 /**
3240 * skb_tstamp_tx - queue clone of skb with send time stamps
3241 * @orig_skb: the original outgoing packet
3242 * @hwtstamps: hardware time stamps, may be NULL if not available
3243 *
3244 * If the skb has a socket associated, then this function clones the
3245 * skb (thus sharing the actual data and optional structures), stores
3246 * the optional hardware time stamping information (if non NULL) or
3247 * generates a software time stamp (otherwise), then queues the clone
3248 * to the error queue of the socket. Errors are silently ignored.
3249 */
3250 void skb_tstamp_tx(struct sk_buff *orig_skb,
3251 struct skb_shared_hwtstamps *hwtstamps);
3252
3253 static inline void sw_tx_timestamp(struct sk_buff *skb)
3254 {
3255 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3256 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3257 skb_tstamp_tx(skb, NULL);
3258 }
3259
3260 /**
3261 * skb_tx_timestamp() - Driver hook for transmit timestamping
3262 *
3263 * Ethernet MAC Drivers should call this function in their hard_xmit()
3264 * function immediately before giving the sk_buff to the MAC hardware.
3265 *
3266 * Specifically, one should make absolutely sure that this function is
3267 * called before TX completion of this packet can trigger. Otherwise
3268 * the packet could potentially already be freed.
3269 *
3270 * @skb: A socket buffer.
3271 */
3272 static inline void skb_tx_timestamp(struct sk_buff *skb)
3273 {
3274 skb_clone_tx_timestamp(skb);
3275 sw_tx_timestamp(skb);
3276 }
3277
3278 /**
3279 * skb_complete_wifi_ack - deliver skb with wifi status
3280 *
3281 * @skb: the original outgoing packet
3282 * @acked: ack status
3283 *
3284 */
3285 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3286
3287 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3288 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3289
3290 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3291 {
3292 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3293 skb->csum_valid ||
3294 (skb->ip_summed == CHECKSUM_PARTIAL &&
3295 skb_checksum_start_offset(skb) >= 0));
3296 }
3297
3298 /**
3299 * skb_checksum_complete - Calculate checksum of an entire packet
3300 * @skb: packet to process
3301 *
3302 * This function calculates the checksum over the entire packet plus
3303 * the value of skb->csum. The latter can be used to supply the
3304 * checksum of a pseudo header as used by TCP/UDP. It returns the
3305 * checksum.
3306 *
3307 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3308 * this function can be used to verify that checksum on received
3309 * packets. In that case the function should return zero if the
3310 * checksum is correct. In particular, this function will return zero
3311 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3312 * hardware has already verified the correctness of the checksum.
3313 */
3314 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3315 {
3316 return skb_csum_unnecessary(skb) ?
3317 0 : __skb_checksum_complete(skb);
3318 }
3319
3320 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3321 {
3322 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3323 if (skb->csum_level == 0)
3324 skb->ip_summed = CHECKSUM_NONE;
3325 else
3326 skb->csum_level--;
3327 }
3328 }
3329
3330 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3331 {
3332 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3333 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3334 skb->csum_level++;
3335 } else if (skb->ip_summed == CHECKSUM_NONE) {
3336 skb->ip_summed = CHECKSUM_UNNECESSARY;
3337 skb->csum_level = 0;
3338 }
3339 }
3340
3341 /* Check if we need to perform checksum complete validation.
3342 *
3343 * Returns true if checksum complete is needed, false otherwise
3344 * (either checksum is unnecessary or zero checksum is allowed).
3345 */
3346 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3347 bool zero_okay,
3348 __sum16 check)
3349 {
3350 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3351 skb->csum_valid = 1;
3352 __skb_decr_checksum_unnecessary(skb);
3353 return false;
3354 }
3355
3356 return true;
3357 }
3358
3359 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3360 * in checksum_init.
3361 */
3362 #define CHECKSUM_BREAK 76
3363
3364 /* Unset checksum-complete
3365 *
3366 * Unset checksum complete can be done when packet is being modified
3367 * (uncompressed for instance) and checksum-complete value is
3368 * invalidated.
3369 */
3370 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3371 {
3372 if (skb->ip_summed == CHECKSUM_COMPLETE)
3373 skb->ip_summed = CHECKSUM_NONE;
3374 }
3375
3376 /* Validate (init) checksum based on checksum complete.
3377 *
3378 * Return values:
3379 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3380 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3381 * checksum is stored in skb->csum for use in __skb_checksum_complete
3382 * non-zero: value of invalid checksum
3383 *
3384 */
3385 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3386 bool complete,
3387 __wsum psum)
3388 {
3389 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3390 if (!csum_fold(csum_add(psum, skb->csum))) {
3391 skb->csum_valid = 1;
3392 return 0;
3393 }
3394 }
3395
3396 skb->csum = psum;
3397
3398 if (complete || skb->len <= CHECKSUM_BREAK) {
3399 __sum16 csum;
3400
3401 csum = __skb_checksum_complete(skb);
3402 skb->csum_valid = !csum;
3403 return csum;
3404 }
3405
3406 return 0;
3407 }
3408
3409 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3410 {
3411 return 0;
3412 }
3413
3414 /* Perform checksum validate (init). Note that this is a macro since we only
3415 * want to calculate the pseudo header which is an input function if necessary.
3416 * First we try to validate without any computation (checksum unnecessary) and
3417 * then calculate based on checksum complete calling the function to compute
3418 * pseudo header.
3419 *
3420 * Return values:
3421 * 0: checksum is validated or try to in skb_checksum_complete
3422 * non-zero: value of invalid checksum
3423 */
3424 #define __skb_checksum_validate(skb, proto, complete, \
3425 zero_okay, check, compute_pseudo) \
3426 ({ \
3427 __sum16 __ret = 0; \
3428 skb->csum_valid = 0; \
3429 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3430 __ret = __skb_checksum_validate_complete(skb, \
3431 complete, compute_pseudo(skb, proto)); \
3432 __ret; \
3433 })
3434
3435 #define skb_checksum_init(skb, proto, compute_pseudo) \
3436 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3437
3438 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3439 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3440
3441 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3442 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3443
3444 #define skb_checksum_validate_zero_check(skb, proto, check, \
3445 compute_pseudo) \
3446 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3447
3448 #define skb_checksum_simple_validate(skb) \
3449 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3450
3451 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3452 {
3453 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3454 }
3455
3456 static inline void __skb_checksum_convert(struct sk_buff *skb,
3457 __sum16 check, __wsum pseudo)
3458 {
3459 skb->csum = ~pseudo;
3460 skb->ip_summed = CHECKSUM_COMPLETE;
3461 }
3462
3463 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3464 do { \
3465 if (__skb_checksum_convert_check(skb)) \
3466 __skb_checksum_convert(skb, check, \
3467 compute_pseudo(skb, proto)); \
3468 } while (0)
3469
3470 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3471 u16 start, u16 offset)
3472 {
3473 skb->ip_summed = CHECKSUM_PARTIAL;
3474 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3475 skb->csum_offset = offset - start;
3476 }
3477
3478 /* Update skbuf and packet to reflect the remote checksum offload operation.
3479 * When called, ptr indicates the starting point for skb->csum when
3480 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3481 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3482 */
3483 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3484 int start, int offset, bool nopartial)
3485 {
3486 __wsum delta;
3487
3488 if (!nopartial) {
3489 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3490 return;
3491 }
3492
3493 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3494 __skb_checksum_complete(skb);
3495 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3496 }
3497
3498 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3499
3500 /* Adjust skb->csum since we changed the packet */
3501 skb->csum = csum_add(skb->csum, delta);
3502 }
3503
3504 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3505 {
3506 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3507 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3508 #else
3509 return NULL;
3510 #endif
3511 }
3512
3513 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3514 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3515 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3516 {
3517 if (nfct && atomic_dec_and_test(&nfct->use))
3518 nf_conntrack_destroy(nfct);
3519 }
3520 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3521 {
3522 if (nfct)
3523 atomic_inc(&nfct->use);
3524 }
3525 #endif
3526 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3527 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3528 {
3529 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3530 kfree(nf_bridge);
3531 }
3532 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3533 {
3534 if (nf_bridge)
3535 atomic_inc(&nf_bridge->use);
3536 }
3537 #endif /* CONFIG_BRIDGE_NETFILTER */
3538 static inline void nf_reset(struct sk_buff *skb)
3539 {
3540 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3541 nf_conntrack_put(skb_nfct(skb));
3542 skb->_nfct = 0;
3543 #endif
3544 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3545 nf_bridge_put(skb->nf_bridge);
3546 skb->nf_bridge = NULL;
3547 #endif
3548 }
3549
3550 static inline void nf_reset_trace(struct sk_buff *skb)
3551 {
3552 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3553 skb->nf_trace = 0;
3554 #endif
3555 }
3556
3557 /* Note: This doesn't put any conntrack and bridge info in dst. */
3558 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3559 bool copy)
3560 {
3561 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3562 dst->_nfct = src->_nfct;
3563 nf_conntrack_get(skb_nfct(src));
3564 #endif
3565 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3566 dst->nf_bridge = src->nf_bridge;
3567 nf_bridge_get(src->nf_bridge);
3568 #endif
3569 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3570 if (copy)
3571 dst->nf_trace = src->nf_trace;
3572 #endif
3573 }
3574
3575 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3576 {
3577 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3578 nf_conntrack_put(skb_nfct(dst));
3579 #endif
3580 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3581 nf_bridge_put(dst->nf_bridge);
3582 #endif
3583 __nf_copy(dst, src, true);
3584 }
3585
3586 #ifdef CONFIG_NETWORK_SECMARK
3587 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3588 {
3589 to->secmark = from->secmark;
3590 }
3591
3592 static inline void skb_init_secmark(struct sk_buff *skb)
3593 {
3594 skb->secmark = 0;
3595 }
3596 #else
3597 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3598 { }
3599
3600 static inline void skb_init_secmark(struct sk_buff *skb)
3601 { }
3602 #endif
3603
3604 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3605 {
3606 return !skb->destructor &&
3607 #if IS_ENABLED(CONFIG_XFRM)
3608 !skb->sp &&
3609 #endif
3610 !skb_nfct(skb) &&
3611 !skb->_skb_refdst &&
3612 !skb_has_frag_list(skb);
3613 }
3614
3615 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3616 {
3617 skb->queue_mapping = queue_mapping;
3618 }
3619
3620 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3621 {
3622 return skb->queue_mapping;
3623 }
3624
3625 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3626 {
3627 to->queue_mapping = from->queue_mapping;
3628 }
3629
3630 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3631 {
3632 skb->queue_mapping = rx_queue + 1;
3633 }
3634
3635 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3636 {
3637 return skb->queue_mapping - 1;
3638 }
3639
3640 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3641 {
3642 return skb->queue_mapping != 0;
3643 }
3644
3645 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3646 {
3647 skb->dst_pending_confirm = val;
3648 }
3649
3650 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3651 {
3652 return skb->dst_pending_confirm != 0;
3653 }
3654
3655 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3656 {
3657 #ifdef CONFIG_XFRM
3658 return skb->sp;
3659 #else
3660 return NULL;
3661 #endif
3662 }
3663
3664 /* Keeps track of mac header offset relative to skb->head.
3665 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3666 * For non-tunnel skb it points to skb_mac_header() and for
3667 * tunnel skb it points to outer mac header.
3668 * Keeps track of level of encapsulation of network headers.
3669 */
3670 struct skb_gso_cb {
3671 union {
3672 int mac_offset;
3673 int data_offset;
3674 };
3675 int encap_level;
3676 __wsum csum;
3677 __u16 csum_start;
3678 };
3679 #define SKB_SGO_CB_OFFSET 32
3680 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3681
3682 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3683 {
3684 return (skb_mac_header(inner_skb) - inner_skb->head) -
3685 SKB_GSO_CB(inner_skb)->mac_offset;
3686 }
3687
3688 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3689 {
3690 int new_headroom, headroom;
3691 int ret;
3692
3693 headroom = skb_headroom(skb);
3694 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3695 if (ret)
3696 return ret;
3697
3698 new_headroom = skb_headroom(skb);
3699 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3700 return 0;
3701 }
3702
3703 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3704 {
3705 /* Do not update partial checksums if remote checksum is enabled. */
3706 if (skb->remcsum_offload)
3707 return;
3708
3709 SKB_GSO_CB(skb)->csum = res;
3710 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3711 }
3712
3713 /* Compute the checksum for a gso segment. First compute the checksum value
3714 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3715 * then add in skb->csum (checksum from csum_start to end of packet).
3716 * skb->csum and csum_start are then updated to reflect the checksum of the
3717 * resultant packet starting from the transport header-- the resultant checksum
3718 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3719 * header.
3720 */
3721 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3722 {
3723 unsigned char *csum_start = skb_transport_header(skb);
3724 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3725 __wsum partial = SKB_GSO_CB(skb)->csum;
3726
3727 SKB_GSO_CB(skb)->csum = res;
3728 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3729
3730 return csum_fold(csum_partial(csum_start, plen, partial));
3731 }
3732
3733 static inline bool skb_is_gso(const struct sk_buff *skb)
3734 {
3735 return skb_shinfo(skb)->gso_size;
3736 }
3737
3738 /* Note: Should be called only if skb_is_gso(skb) is true */
3739 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3740 {
3741 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3742 }
3743
3744 static inline void skb_gso_reset(struct sk_buff *skb)
3745 {
3746 skb_shinfo(skb)->gso_size = 0;
3747 skb_shinfo(skb)->gso_segs = 0;
3748 skb_shinfo(skb)->gso_type = 0;
3749 }
3750
3751 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3752
3753 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3754 {
3755 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3756 * wanted then gso_type will be set. */
3757 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3758
3759 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3760 unlikely(shinfo->gso_type == 0)) {
3761 __skb_warn_lro_forwarding(skb);
3762 return true;
3763 }
3764 return false;
3765 }
3766
3767 static inline void skb_forward_csum(struct sk_buff *skb)
3768 {
3769 /* Unfortunately we don't support this one. Any brave souls? */
3770 if (skb->ip_summed == CHECKSUM_COMPLETE)
3771 skb->ip_summed = CHECKSUM_NONE;
3772 }
3773
3774 /**
3775 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3776 * @skb: skb to check
3777 *
3778 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3779 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3780 * use this helper, to document places where we make this assertion.
3781 */
3782 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3783 {
3784 #ifdef DEBUG
3785 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3786 #endif
3787 }
3788
3789 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3790
3791 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3792 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3793 unsigned int transport_len,
3794 __sum16(*skb_chkf)(struct sk_buff *skb));
3795
3796 /**
3797 * skb_head_is_locked - Determine if the skb->head is locked down
3798 * @skb: skb to check
3799 *
3800 * The head on skbs build around a head frag can be removed if they are
3801 * not cloned. This function returns true if the skb head is locked down
3802 * due to either being allocated via kmalloc, or by being a clone with
3803 * multiple references to the head.
3804 */
3805 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3806 {
3807 return !skb->head_frag || skb_cloned(skb);
3808 }
3809
3810 /**
3811 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3812 *
3813 * @skb: GSO skb
3814 *
3815 * skb_gso_network_seglen is used to determine the real size of the
3816 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3817 *
3818 * The MAC/L2 header is not accounted for.
3819 */
3820 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3821 {
3822 unsigned int hdr_len = skb_transport_header(skb) -
3823 skb_network_header(skb);
3824 return hdr_len + skb_gso_transport_seglen(skb);
3825 }
3826
3827 /* Local Checksum Offload.
3828 * Compute outer checksum based on the assumption that the
3829 * inner checksum will be offloaded later.
3830 * See Documentation/networking/checksum-offloads.txt for
3831 * explanation of how this works.
3832 * Fill in outer checksum adjustment (e.g. with sum of outer
3833 * pseudo-header) before calling.
3834 * Also ensure that inner checksum is in linear data area.
3835 */
3836 static inline __wsum lco_csum(struct sk_buff *skb)
3837 {
3838 unsigned char *csum_start = skb_checksum_start(skb);
3839 unsigned char *l4_hdr = skb_transport_header(skb);
3840 __wsum partial;
3841
3842 /* Start with complement of inner checksum adjustment */
3843 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3844 skb->csum_offset));
3845
3846 /* Add in checksum of our headers (incl. outer checksum
3847 * adjustment filled in by caller) and return result.
3848 */
3849 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3850 }
3851
3852 #endif /* __KERNEL__ */
3853 #endif /* _LINUX_SKBUFF_H */