Merge branch 'kbuild' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild
[deliverable/linux.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 <net/flow_keys.h>
38
39 /* A. Checksumming of received packets by device.
40 *
41 * CHECKSUM_NONE:
42 *
43 * Device failed to checksum this packet e.g. due to lack of capabilities.
44 * The packet contains full (though not verified) checksum in packet but
45 * not in skb->csum. Thus, skb->csum is undefined in this case.
46 *
47 * CHECKSUM_UNNECESSARY:
48 *
49 * The hardware you're dealing with doesn't calculate the full checksum
50 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
51 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
52 * if their checksums are okay. skb->csum is still undefined in this case
53 * though. It is a bad option, but, unfortunately, nowadays most vendors do
54 * this. Apparently with the secret goal to sell you new devices, when you
55 * will add new protocol to your host, f.e. IPv6 8)
56 *
57 * CHECKSUM_UNNECESSARY is applicable to following protocols:
58 * TCP: IPv6 and IPv4.
59 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
60 * zero UDP checksum for either IPv4 or IPv6, the networking stack
61 * may perform further validation in this case.
62 * GRE: only if the checksum is present in the header.
63 * SCTP: indicates the CRC in SCTP header has been validated.
64 *
65 * skb->csum_level indicates the number of consecutive checksums found in
66 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
67 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
68 * and a device is able to verify the checksums for UDP (possibly zero),
69 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
70 * two. If the device were only able to verify the UDP checksum and not
71 * GRE, either because it doesn't support GRE checksum of because GRE
72 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
73 * not considered in this case).
74 *
75 * CHECKSUM_COMPLETE:
76 *
77 * This is the most generic way. The device supplied checksum of the _whole_
78 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
79 * hardware doesn't need to parse L3/L4 headers to implement this.
80 *
81 * Note: Even if device supports only some protocols, but is able to produce
82 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
83 *
84 * CHECKSUM_PARTIAL:
85 *
86 * A checksum is set up to be offloaded to a device as described in the
87 * output description for CHECKSUM_PARTIAL. This may occur on a packet
88 * received directly from another Linux OS, e.g., a virtualized Linux kernel
89 * on the same host, or it may be set in the input path in GRO or remote
90 * checksum offload. For the purposes of checksum verification, the checksum
91 * referred to by skb->csum_start + skb->csum_offset and any preceding
92 * checksums in the packet are considered verified. Any checksums in the
93 * packet that are after the checksum being offloaded are not considered to
94 * be verified.
95 *
96 * B. Checksumming on output.
97 *
98 * CHECKSUM_NONE:
99 *
100 * The skb was already checksummed by the protocol, or a checksum is not
101 * required.
102 *
103 * CHECKSUM_PARTIAL:
104 *
105 * The device is required to checksum the packet as seen by hard_start_xmit()
106 * from skb->csum_start up to the end, and to record/write the checksum at
107 * offset skb->csum_start + skb->csum_offset.
108 *
109 * The device must show its capabilities in dev->features, set up at device
110 * setup time, e.g. netdev_features.h:
111 *
112 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
113 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
114 * IPv4. Sigh. Vendors like this way for an unknown reason.
115 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
116 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
117 * NETIF_F_... - Well, you get the picture.
118 *
119 * CHECKSUM_UNNECESSARY:
120 *
121 * Normally, the device will do per protocol specific checksumming. Protocol
122 * implementations that do not want the NIC to perform the checksum
123 * calculation should use this flag in their outgoing skbs.
124 *
125 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
126 * offload. Correspondingly, the FCoE protocol driver
127 * stack should use CHECKSUM_UNNECESSARY.
128 *
129 * Any questions? No questions, good. --ANK
130 */
131
132 /* Don't change this without changing skb_csum_unnecessary! */
133 #define CHECKSUM_NONE 0
134 #define CHECKSUM_UNNECESSARY 1
135 #define CHECKSUM_COMPLETE 2
136 #define CHECKSUM_PARTIAL 3
137
138 /* Maximum value in skb->csum_level */
139 #define SKB_MAX_CSUM_LEVEL 3
140
141 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
142 #define SKB_WITH_OVERHEAD(X) \
143 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
144 #define SKB_MAX_ORDER(X, ORDER) \
145 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
146 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
147 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
148
149 /* return minimum truesize of one skb containing X bytes of data */
150 #define SKB_TRUESIZE(X) ((X) + \
151 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
152 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
153
154 struct net_device;
155 struct scatterlist;
156 struct pipe_inode_info;
157 struct iov_iter;
158 struct napi_struct;
159
160 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
161 struct nf_conntrack {
162 atomic_t use;
163 };
164 #endif
165
166 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
167 struct nf_bridge_info {
168 atomic_t use;
169 unsigned int mask;
170 struct net_device *physindev;
171 struct net_device *physoutdev;
172 unsigned long data[32 / sizeof(unsigned long)];
173 };
174 #endif
175
176 struct sk_buff_head {
177 /* These two members must be first. */
178 struct sk_buff *next;
179 struct sk_buff *prev;
180
181 __u32 qlen;
182 spinlock_t lock;
183 };
184
185 struct sk_buff;
186
187 /* To allow 64K frame to be packed as single skb without frag_list we
188 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
189 * buffers which do not start on a page boundary.
190 *
191 * Since GRO uses frags we allocate at least 16 regardless of page
192 * size.
193 */
194 #if (65536/PAGE_SIZE + 1) < 16
195 #define MAX_SKB_FRAGS 16UL
196 #else
197 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
198 #endif
199
200 typedef struct skb_frag_struct skb_frag_t;
201
202 struct skb_frag_struct {
203 struct {
204 struct page *p;
205 } page;
206 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
207 __u32 page_offset;
208 __u32 size;
209 #else
210 __u16 page_offset;
211 __u16 size;
212 #endif
213 };
214
215 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
216 {
217 return frag->size;
218 }
219
220 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
221 {
222 frag->size = size;
223 }
224
225 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
226 {
227 frag->size += delta;
228 }
229
230 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
231 {
232 frag->size -= delta;
233 }
234
235 #define HAVE_HW_TIME_STAMP
236
237 /**
238 * struct skb_shared_hwtstamps - hardware time stamps
239 * @hwtstamp: hardware time stamp transformed into duration
240 * since arbitrary point in time
241 *
242 * Software time stamps generated by ktime_get_real() are stored in
243 * skb->tstamp.
244 *
245 * hwtstamps can only be compared against other hwtstamps from
246 * the same device.
247 *
248 * This structure is attached to packets as part of the
249 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
250 */
251 struct skb_shared_hwtstamps {
252 ktime_t hwtstamp;
253 };
254
255 /* Definitions for tx_flags in struct skb_shared_info */
256 enum {
257 /* generate hardware time stamp */
258 SKBTX_HW_TSTAMP = 1 << 0,
259
260 /* generate software time stamp when queueing packet to NIC */
261 SKBTX_SW_TSTAMP = 1 << 1,
262
263 /* device driver is going to provide hardware time stamp */
264 SKBTX_IN_PROGRESS = 1 << 2,
265
266 /* device driver supports TX zero-copy buffers */
267 SKBTX_DEV_ZEROCOPY = 1 << 3,
268
269 /* generate wifi status information (where possible) */
270 SKBTX_WIFI_STATUS = 1 << 4,
271
272 /* This indicates at least one fragment might be overwritten
273 * (as in vmsplice(), sendfile() ...)
274 * If we need to compute a TX checksum, we'll need to copy
275 * all frags to avoid possible bad checksum
276 */
277 SKBTX_SHARED_FRAG = 1 << 5,
278
279 /* generate software time stamp when entering packet scheduling */
280 SKBTX_SCHED_TSTAMP = 1 << 6,
281
282 /* generate software timestamp on peer data acknowledgment */
283 SKBTX_ACK_TSTAMP = 1 << 7,
284 };
285
286 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
287 SKBTX_SCHED_TSTAMP | \
288 SKBTX_ACK_TSTAMP)
289 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
290
291 /*
292 * The callback notifies userspace to release buffers when skb DMA is done in
293 * lower device, the skb last reference should be 0 when calling this.
294 * The zerocopy_success argument is true if zero copy transmit occurred,
295 * false on data copy or out of memory error caused by data copy attempt.
296 * The ctx field is used to track device context.
297 * The desc field is used to track userspace buffer index.
298 */
299 struct ubuf_info {
300 void (*callback)(struct ubuf_info *, bool zerocopy_success);
301 void *ctx;
302 unsigned long desc;
303 };
304
305 /* This data is invariant across clones and lives at
306 * the end of the header data, ie. at skb->end.
307 */
308 struct skb_shared_info {
309 unsigned char nr_frags;
310 __u8 tx_flags;
311 unsigned short gso_size;
312 /* Warning: this field is not always filled in (UFO)! */
313 unsigned short gso_segs;
314 unsigned short gso_type;
315 struct sk_buff *frag_list;
316 struct skb_shared_hwtstamps hwtstamps;
317 u32 tskey;
318 __be32 ip6_frag_id;
319
320 /*
321 * Warning : all fields before dataref are cleared in __alloc_skb()
322 */
323 atomic_t dataref;
324
325 /* Intermediate layers must ensure that destructor_arg
326 * remains valid until skb destructor */
327 void * destructor_arg;
328
329 /* must be last field, see pskb_expand_head() */
330 skb_frag_t frags[MAX_SKB_FRAGS];
331 };
332
333 /* We divide dataref into two halves. The higher 16 bits hold references
334 * to the payload part of skb->data. The lower 16 bits hold references to
335 * the entire skb->data. A clone of a headerless skb holds the length of
336 * the header in skb->hdr_len.
337 *
338 * All users must obey the rule that the skb->data reference count must be
339 * greater than or equal to the payload reference count.
340 *
341 * Holding a reference to the payload part means that the user does not
342 * care about modifications to the header part of skb->data.
343 */
344 #define SKB_DATAREF_SHIFT 16
345 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
346
347
348 enum {
349 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
350 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
351 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
352 };
353
354 enum {
355 SKB_GSO_TCPV4 = 1 << 0,
356 SKB_GSO_UDP = 1 << 1,
357
358 /* This indicates the skb is from an untrusted source. */
359 SKB_GSO_DODGY = 1 << 2,
360
361 /* This indicates the tcp segment has CWR set. */
362 SKB_GSO_TCP_ECN = 1 << 3,
363
364 SKB_GSO_TCPV6 = 1 << 4,
365
366 SKB_GSO_FCOE = 1 << 5,
367
368 SKB_GSO_GRE = 1 << 6,
369
370 SKB_GSO_GRE_CSUM = 1 << 7,
371
372 SKB_GSO_IPIP = 1 << 8,
373
374 SKB_GSO_SIT = 1 << 9,
375
376 SKB_GSO_UDP_TUNNEL = 1 << 10,
377
378 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
379
380 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
381 };
382
383 #if BITS_PER_LONG > 32
384 #define NET_SKBUFF_DATA_USES_OFFSET 1
385 #endif
386
387 #ifdef NET_SKBUFF_DATA_USES_OFFSET
388 typedef unsigned int sk_buff_data_t;
389 #else
390 typedef unsigned char *sk_buff_data_t;
391 #endif
392
393 /**
394 * struct skb_mstamp - multi resolution time stamps
395 * @stamp_us: timestamp in us resolution
396 * @stamp_jiffies: timestamp in jiffies
397 */
398 struct skb_mstamp {
399 union {
400 u64 v64;
401 struct {
402 u32 stamp_us;
403 u32 stamp_jiffies;
404 };
405 };
406 };
407
408 /**
409 * skb_mstamp_get - get current timestamp
410 * @cl: place to store timestamps
411 */
412 static inline void skb_mstamp_get(struct skb_mstamp *cl)
413 {
414 u64 val = local_clock();
415
416 do_div(val, NSEC_PER_USEC);
417 cl->stamp_us = (u32)val;
418 cl->stamp_jiffies = (u32)jiffies;
419 }
420
421 /**
422 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
423 * @t1: pointer to newest sample
424 * @t0: pointer to oldest sample
425 */
426 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
427 const struct skb_mstamp *t0)
428 {
429 s32 delta_us = t1->stamp_us - t0->stamp_us;
430 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
431
432 /* If delta_us is negative, this might be because interval is too big,
433 * or local_clock() drift is too big : fallback using jiffies.
434 */
435 if (delta_us <= 0 ||
436 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
437
438 delta_us = jiffies_to_usecs(delta_jiffies);
439
440 return delta_us;
441 }
442
443
444 /**
445 * struct sk_buff - socket buffer
446 * @next: Next buffer in list
447 * @prev: Previous buffer in list
448 * @tstamp: Time we arrived/left
449 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
450 * @sk: Socket we are owned by
451 * @dev: Device we arrived on/are leaving by
452 * @cb: Control buffer. Free for use by every layer. Put private vars here
453 * @_skb_refdst: destination entry (with norefcount bit)
454 * @sp: the security path, used for xfrm
455 * @len: Length of actual data
456 * @data_len: Data length
457 * @mac_len: Length of link layer header
458 * @hdr_len: writable header length of cloned skb
459 * @csum: Checksum (must include start/offset pair)
460 * @csum_start: Offset from skb->head where checksumming should start
461 * @csum_offset: Offset from csum_start where checksum should be stored
462 * @priority: Packet queueing priority
463 * @ignore_df: allow local fragmentation
464 * @cloned: Head may be cloned (check refcnt to be sure)
465 * @ip_summed: Driver fed us an IP checksum
466 * @nohdr: Payload reference only, must not modify header
467 * @nfctinfo: Relationship of this skb to the connection
468 * @pkt_type: Packet class
469 * @fclone: skbuff clone status
470 * @ipvs_property: skbuff is owned by ipvs
471 * @peeked: this packet has been seen already, so stats have been
472 * done for it, don't do them again
473 * @nf_trace: netfilter packet trace flag
474 * @protocol: Packet protocol from driver
475 * @destructor: Destruct function
476 * @nfct: Associated connection, if any
477 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
478 * @skb_iif: ifindex of device we arrived on
479 * @tc_index: Traffic control index
480 * @tc_verd: traffic control verdict
481 * @hash: the packet hash
482 * @queue_mapping: Queue mapping for multiqueue devices
483 * @xmit_more: More SKBs are pending for this queue
484 * @ndisc_nodetype: router type (from link layer)
485 * @ooo_okay: allow the mapping of a socket to a queue to be changed
486 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
487 * ports.
488 * @sw_hash: indicates hash was computed in software stack
489 * @wifi_acked_valid: wifi_acked was set
490 * @wifi_acked: whether frame was acked on wifi or not
491 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
492 * @napi_id: id of the NAPI struct this skb came from
493 * @secmark: security marking
494 * @mark: Generic packet mark
495 * @dropcount: total number of sk_receive_queue overflows
496 * @vlan_proto: vlan encapsulation protocol
497 * @vlan_tci: vlan tag control information
498 * @inner_protocol: Protocol (encapsulation)
499 * @inner_transport_header: Inner transport layer header (encapsulation)
500 * @inner_network_header: Network layer header (encapsulation)
501 * @inner_mac_header: Link layer header (encapsulation)
502 * @transport_header: Transport layer header
503 * @network_header: Network layer header
504 * @mac_header: Link layer header
505 * @tail: Tail pointer
506 * @end: End pointer
507 * @head: Head of buffer
508 * @data: Data head pointer
509 * @truesize: Buffer size
510 * @users: User count - see {datagram,tcp}.c
511 */
512
513 struct sk_buff {
514 union {
515 struct {
516 /* These two members must be first. */
517 struct sk_buff *next;
518 struct sk_buff *prev;
519
520 union {
521 ktime_t tstamp;
522 struct skb_mstamp skb_mstamp;
523 };
524 };
525 struct rb_node rbnode; /* used in netem & tcp stack */
526 };
527 struct sock *sk;
528 struct net_device *dev;
529
530 /*
531 * This is the control buffer. It is free to use for every
532 * layer. Please put your private variables there. If you
533 * want to keep them across layers you have to do a skb_clone()
534 * first. This is owned by whoever has the skb queued ATM.
535 */
536 char cb[48] __aligned(8);
537
538 unsigned long _skb_refdst;
539 void (*destructor)(struct sk_buff *skb);
540 #ifdef CONFIG_XFRM
541 struct sec_path *sp;
542 #endif
543 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
544 struct nf_conntrack *nfct;
545 #endif
546 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
547 struct nf_bridge_info *nf_bridge;
548 #endif
549 unsigned int len,
550 data_len;
551 __u16 mac_len,
552 hdr_len;
553
554 /* Following fields are _not_ copied in __copy_skb_header()
555 * Note that queue_mapping is here mostly to fill a hole.
556 */
557 kmemcheck_bitfield_begin(flags1);
558 __u16 queue_mapping;
559 __u8 cloned:1,
560 nohdr:1,
561 fclone:2,
562 peeked:1,
563 head_frag:1,
564 xmit_more:1;
565 /* one bit hole */
566 kmemcheck_bitfield_end(flags1);
567
568 /* fields enclosed in headers_start/headers_end are copied
569 * using a single memcpy() in __copy_skb_header()
570 */
571 /* private: */
572 __u32 headers_start[0];
573 /* public: */
574
575 /* if you move pkt_type around you also must adapt those constants */
576 #ifdef __BIG_ENDIAN_BITFIELD
577 #define PKT_TYPE_MAX (7 << 5)
578 #else
579 #define PKT_TYPE_MAX 7
580 #endif
581 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
582
583 __u8 __pkt_type_offset[0];
584 __u8 pkt_type:3;
585 __u8 pfmemalloc:1;
586 __u8 ignore_df:1;
587 __u8 nfctinfo:3;
588
589 __u8 nf_trace:1;
590 __u8 ip_summed:2;
591 __u8 ooo_okay:1;
592 __u8 l4_hash:1;
593 __u8 sw_hash:1;
594 __u8 wifi_acked_valid:1;
595 __u8 wifi_acked:1;
596
597 __u8 no_fcs:1;
598 /* Indicates the inner headers are valid in the skbuff. */
599 __u8 encapsulation:1;
600 __u8 encap_hdr_csum:1;
601 __u8 csum_valid:1;
602 __u8 csum_complete_sw:1;
603 __u8 csum_level:2;
604 __u8 csum_bad:1;
605
606 #ifdef CONFIG_IPV6_NDISC_NODETYPE
607 __u8 ndisc_nodetype:2;
608 #endif
609 __u8 ipvs_property:1;
610 __u8 inner_protocol_type:1;
611 __u8 remcsum_offload:1;
612 /* 3 or 5 bit hole */
613
614 #ifdef CONFIG_NET_SCHED
615 __u16 tc_index; /* traffic control index */
616 #ifdef CONFIG_NET_CLS_ACT
617 __u16 tc_verd; /* traffic control verdict */
618 #endif
619 #endif
620
621 union {
622 __wsum csum;
623 struct {
624 __u16 csum_start;
625 __u16 csum_offset;
626 };
627 };
628 __u32 priority;
629 int skb_iif;
630 __u32 hash;
631 __be16 vlan_proto;
632 __u16 vlan_tci;
633 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
634 union {
635 unsigned int napi_id;
636 unsigned int sender_cpu;
637 };
638 #endif
639 #ifdef CONFIG_NETWORK_SECMARK
640 __u32 secmark;
641 #endif
642 union {
643 __u32 mark;
644 __u32 dropcount;
645 __u32 reserved_tailroom;
646 };
647
648 union {
649 __be16 inner_protocol;
650 __u8 inner_ipproto;
651 };
652
653 __u16 inner_transport_header;
654 __u16 inner_network_header;
655 __u16 inner_mac_header;
656
657 __be16 protocol;
658 __u16 transport_header;
659 __u16 network_header;
660 __u16 mac_header;
661
662 /* private: */
663 __u32 headers_end[0];
664 /* public: */
665
666 /* These elements must be at the end, see alloc_skb() for details. */
667 sk_buff_data_t tail;
668 sk_buff_data_t end;
669 unsigned char *head,
670 *data;
671 unsigned int truesize;
672 atomic_t users;
673 };
674
675 #ifdef __KERNEL__
676 /*
677 * Handling routines are only of interest to the kernel
678 */
679 #include <linux/slab.h>
680
681
682 #define SKB_ALLOC_FCLONE 0x01
683 #define SKB_ALLOC_RX 0x02
684 #define SKB_ALLOC_NAPI 0x04
685
686 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
687 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
688 {
689 return unlikely(skb->pfmemalloc);
690 }
691
692 /*
693 * skb might have a dst pointer attached, refcounted or not.
694 * _skb_refdst low order bit is set if refcount was _not_ taken
695 */
696 #define SKB_DST_NOREF 1UL
697 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
698
699 /**
700 * skb_dst - returns skb dst_entry
701 * @skb: buffer
702 *
703 * Returns skb dst_entry, regardless of reference taken or not.
704 */
705 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
706 {
707 /* If refdst was not refcounted, check we still are in a
708 * rcu_read_lock section
709 */
710 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
711 !rcu_read_lock_held() &&
712 !rcu_read_lock_bh_held());
713 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
714 }
715
716 /**
717 * skb_dst_set - sets skb dst
718 * @skb: buffer
719 * @dst: dst entry
720 *
721 * Sets skb dst, assuming a reference was taken on dst and should
722 * be released by skb_dst_drop()
723 */
724 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
725 {
726 skb->_skb_refdst = (unsigned long)dst;
727 }
728
729 /**
730 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
731 * @skb: buffer
732 * @dst: dst entry
733 *
734 * Sets skb dst, assuming a reference was not taken on dst.
735 * If dst entry is cached, we do not take reference and dst_release
736 * will be avoided by refdst_drop. If dst entry is not cached, we take
737 * reference, so that last dst_release can destroy the dst immediately.
738 */
739 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
740 {
741 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
742 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
743 }
744
745 /**
746 * skb_dst_is_noref - Test if skb dst isn't refcounted
747 * @skb: buffer
748 */
749 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
750 {
751 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
752 }
753
754 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
755 {
756 return (struct rtable *)skb_dst(skb);
757 }
758
759 void kfree_skb(struct sk_buff *skb);
760 void kfree_skb_list(struct sk_buff *segs);
761 void skb_tx_error(struct sk_buff *skb);
762 void consume_skb(struct sk_buff *skb);
763 void __kfree_skb(struct sk_buff *skb);
764 extern struct kmem_cache *skbuff_head_cache;
765
766 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
767 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
768 bool *fragstolen, int *delta_truesize);
769
770 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
771 int node);
772 struct sk_buff *build_skb(void *data, unsigned int frag_size);
773 static inline struct sk_buff *alloc_skb(unsigned int size,
774 gfp_t priority)
775 {
776 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
777 }
778
779 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
780 unsigned long data_len,
781 int max_page_order,
782 int *errcode,
783 gfp_t gfp_mask);
784
785 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
786 struct sk_buff_fclones {
787 struct sk_buff skb1;
788
789 struct sk_buff skb2;
790
791 atomic_t fclone_ref;
792 };
793
794 /**
795 * skb_fclone_busy - check if fclone is busy
796 * @skb: buffer
797 *
798 * Returns true is skb is a fast clone, and its clone is not freed.
799 * Some drivers call skb_orphan() in their ndo_start_xmit(),
800 * so we also check that this didnt happen.
801 */
802 static inline bool skb_fclone_busy(const struct sock *sk,
803 const struct sk_buff *skb)
804 {
805 const struct sk_buff_fclones *fclones;
806
807 fclones = container_of(skb, struct sk_buff_fclones, skb1);
808
809 return skb->fclone == SKB_FCLONE_ORIG &&
810 atomic_read(&fclones->fclone_ref) > 1 &&
811 fclones->skb2.sk == sk;
812 }
813
814 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
815 gfp_t priority)
816 {
817 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
818 }
819
820 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
821 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
822 {
823 return __alloc_skb_head(priority, -1);
824 }
825
826 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
827 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
828 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
829 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
830 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
831 gfp_t gfp_mask, bool fclone);
832 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
833 gfp_t gfp_mask)
834 {
835 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
836 }
837
838 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
839 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
840 unsigned int headroom);
841 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
842 int newtailroom, gfp_t priority);
843 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
844 int offset, int len);
845 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
846 int len);
847 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
848 int skb_pad(struct sk_buff *skb, int pad);
849 #define dev_kfree_skb(a) consume_skb(a)
850
851 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
852 int getfrag(void *from, char *to, int offset,
853 int len, int odd, struct sk_buff *skb),
854 void *from, int length);
855
856 struct skb_seq_state {
857 __u32 lower_offset;
858 __u32 upper_offset;
859 __u32 frag_idx;
860 __u32 stepped_offset;
861 struct sk_buff *root_skb;
862 struct sk_buff *cur_skb;
863 __u8 *frag_data;
864 };
865
866 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
867 unsigned int to, struct skb_seq_state *st);
868 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
869 struct skb_seq_state *st);
870 void skb_abort_seq_read(struct skb_seq_state *st);
871
872 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
873 unsigned int to, struct ts_config *config,
874 struct ts_state *state);
875
876 /*
877 * Packet hash types specify the type of hash in skb_set_hash.
878 *
879 * Hash types refer to the protocol layer addresses which are used to
880 * construct a packet's hash. The hashes are used to differentiate or identify
881 * flows of the protocol layer for the hash type. Hash types are either
882 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
883 *
884 * Properties of hashes:
885 *
886 * 1) Two packets in different flows have different hash values
887 * 2) Two packets in the same flow should have the same hash value
888 *
889 * A hash at a higher layer is considered to be more specific. A driver should
890 * set the most specific hash possible.
891 *
892 * A driver cannot indicate a more specific hash than the layer at which a hash
893 * was computed. For instance an L3 hash cannot be set as an L4 hash.
894 *
895 * A driver may indicate a hash level which is less specific than the
896 * actual layer the hash was computed on. For instance, a hash computed
897 * at L4 may be considered an L3 hash. This should only be done if the
898 * driver can't unambiguously determine that the HW computed the hash at
899 * the higher layer. Note that the "should" in the second property above
900 * permits this.
901 */
902 enum pkt_hash_types {
903 PKT_HASH_TYPE_NONE, /* Undefined type */
904 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
905 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
906 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
907 };
908
909 static inline void
910 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
911 {
912 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
913 skb->sw_hash = 0;
914 skb->hash = hash;
915 }
916
917 void __skb_get_hash(struct sk_buff *skb);
918 static inline __u32 skb_get_hash(struct sk_buff *skb)
919 {
920 if (!skb->l4_hash && !skb->sw_hash)
921 __skb_get_hash(skb);
922
923 return skb->hash;
924 }
925
926 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
927 {
928 return skb->hash;
929 }
930
931 static inline void skb_clear_hash(struct sk_buff *skb)
932 {
933 skb->hash = 0;
934 skb->sw_hash = 0;
935 skb->l4_hash = 0;
936 }
937
938 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
939 {
940 if (!skb->l4_hash)
941 skb_clear_hash(skb);
942 }
943
944 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
945 {
946 to->hash = from->hash;
947 to->sw_hash = from->sw_hash;
948 to->l4_hash = from->l4_hash;
949 };
950
951 #ifdef NET_SKBUFF_DATA_USES_OFFSET
952 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
953 {
954 return skb->head + skb->end;
955 }
956
957 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
958 {
959 return skb->end;
960 }
961 #else
962 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
963 {
964 return skb->end;
965 }
966
967 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
968 {
969 return skb->end - skb->head;
970 }
971 #endif
972
973 /* Internal */
974 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
975
976 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
977 {
978 return &skb_shinfo(skb)->hwtstamps;
979 }
980
981 /**
982 * skb_queue_empty - check if a queue is empty
983 * @list: queue head
984 *
985 * Returns true if the queue is empty, false otherwise.
986 */
987 static inline int skb_queue_empty(const struct sk_buff_head *list)
988 {
989 return list->next == (const struct sk_buff *) list;
990 }
991
992 /**
993 * skb_queue_is_last - check if skb is the last entry in the queue
994 * @list: queue head
995 * @skb: buffer
996 *
997 * Returns true if @skb is the last buffer on the list.
998 */
999 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1000 const struct sk_buff *skb)
1001 {
1002 return skb->next == (const struct sk_buff *) list;
1003 }
1004
1005 /**
1006 * skb_queue_is_first - check if skb is the first entry in the queue
1007 * @list: queue head
1008 * @skb: buffer
1009 *
1010 * Returns true if @skb is the first buffer on the list.
1011 */
1012 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1013 const struct sk_buff *skb)
1014 {
1015 return skb->prev == (const struct sk_buff *) list;
1016 }
1017
1018 /**
1019 * skb_queue_next - return the next packet in the queue
1020 * @list: queue head
1021 * @skb: current buffer
1022 *
1023 * Return the next packet in @list after @skb. It is only valid to
1024 * call this if skb_queue_is_last() evaluates to false.
1025 */
1026 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1027 const struct sk_buff *skb)
1028 {
1029 /* This BUG_ON may seem severe, but if we just return then we
1030 * are going to dereference garbage.
1031 */
1032 BUG_ON(skb_queue_is_last(list, skb));
1033 return skb->next;
1034 }
1035
1036 /**
1037 * skb_queue_prev - return the prev packet in the queue
1038 * @list: queue head
1039 * @skb: current buffer
1040 *
1041 * Return the prev packet in @list before @skb. It is only valid to
1042 * call this if skb_queue_is_first() evaluates to false.
1043 */
1044 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1045 const struct sk_buff *skb)
1046 {
1047 /* This BUG_ON may seem severe, but if we just return then we
1048 * are going to dereference garbage.
1049 */
1050 BUG_ON(skb_queue_is_first(list, skb));
1051 return skb->prev;
1052 }
1053
1054 /**
1055 * skb_get - reference buffer
1056 * @skb: buffer to reference
1057 *
1058 * Makes another reference to a socket buffer and returns a pointer
1059 * to the buffer.
1060 */
1061 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1062 {
1063 atomic_inc(&skb->users);
1064 return skb;
1065 }
1066
1067 /*
1068 * If users == 1, we are the only owner and are can avoid redundant
1069 * atomic change.
1070 */
1071
1072 /**
1073 * skb_cloned - is the buffer a clone
1074 * @skb: buffer to check
1075 *
1076 * Returns true if the buffer was generated with skb_clone() and is
1077 * one of multiple shared copies of the buffer. Cloned buffers are
1078 * shared data so must not be written to under normal circumstances.
1079 */
1080 static inline int skb_cloned(const struct sk_buff *skb)
1081 {
1082 return skb->cloned &&
1083 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1084 }
1085
1086 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1087 {
1088 might_sleep_if(pri & __GFP_WAIT);
1089
1090 if (skb_cloned(skb))
1091 return pskb_expand_head(skb, 0, 0, pri);
1092
1093 return 0;
1094 }
1095
1096 /**
1097 * skb_header_cloned - is the header a clone
1098 * @skb: buffer to check
1099 *
1100 * Returns true if modifying the header part of the buffer requires
1101 * the data to be copied.
1102 */
1103 static inline int skb_header_cloned(const struct sk_buff *skb)
1104 {
1105 int dataref;
1106
1107 if (!skb->cloned)
1108 return 0;
1109
1110 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1111 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1112 return dataref != 1;
1113 }
1114
1115 /**
1116 * skb_header_release - release reference to header
1117 * @skb: buffer to operate on
1118 *
1119 * Drop a reference to the header part of the buffer. This is done
1120 * by acquiring a payload reference. You must not read from the header
1121 * part of skb->data after this.
1122 * Note : Check if you can use __skb_header_release() instead.
1123 */
1124 static inline void skb_header_release(struct sk_buff *skb)
1125 {
1126 BUG_ON(skb->nohdr);
1127 skb->nohdr = 1;
1128 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1129 }
1130
1131 /**
1132 * __skb_header_release - release reference to header
1133 * @skb: buffer to operate on
1134 *
1135 * Variant of skb_header_release() assuming skb is private to caller.
1136 * We can avoid one atomic operation.
1137 */
1138 static inline void __skb_header_release(struct sk_buff *skb)
1139 {
1140 skb->nohdr = 1;
1141 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1142 }
1143
1144
1145 /**
1146 * skb_shared - is the buffer shared
1147 * @skb: buffer to check
1148 *
1149 * Returns true if more than one person has a reference to this
1150 * buffer.
1151 */
1152 static inline int skb_shared(const struct sk_buff *skb)
1153 {
1154 return atomic_read(&skb->users) != 1;
1155 }
1156
1157 /**
1158 * skb_share_check - check if buffer is shared and if so clone it
1159 * @skb: buffer to check
1160 * @pri: priority for memory allocation
1161 *
1162 * If the buffer is shared the buffer is cloned and the old copy
1163 * drops a reference. A new clone with a single reference is returned.
1164 * If the buffer is not shared the original buffer is returned. When
1165 * being called from interrupt status or with spinlocks held pri must
1166 * be GFP_ATOMIC.
1167 *
1168 * NULL is returned on a memory allocation failure.
1169 */
1170 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1171 {
1172 might_sleep_if(pri & __GFP_WAIT);
1173 if (skb_shared(skb)) {
1174 struct sk_buff *nskb = skb_clone(skb, pri);
1175
1176 if (likely(nskb))
1177 consume_skb(skb);
1178 else
1179 kfree_skb(skb);
1180 skb = nskb;
1181 }
1182 return skb;
1183 }
1184
1185 /*
1186 * Copy shared buffers into a new sk_buff. We effectively do COW on
1187 * packets to handle cases where we have a local reader and forward
1188 * and a couple of other messy ones. The normal one is tcpdumping
1189 * a packet thats being forwarded.
1190 */
1191
1192 /**
1193 * skb_unshare - make a copy of a shared buffer
1194 * @skb: buffer to check
1195 * @pri: priority for memory allocation
1196 *
1197 * If the socket buffer is a clone then this function creates a new
1198 * copy of the data, drops a reference count on the old copy and returns
1199 * the new copy with the reference count at 1. If the buffer is not a clone
1200 * the original buffer is returned. When called with a spinlock held or
1201 * from interrupt state @pri must be %GFP_ATOMIC
1202 *
1203 * %NULL is returned on a memory allocation failure.
1204 */
1205 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1206 gfp_t pri)
1207 {
1208 might_sleep_if(pri & __GFP_WAIT);
1209 if (skb_cloned(skb)) {
1210 struct sk_buff *nskb = skb_copy(skb, pri);
1211
1212 /* Free our shared copy */
1213 if (likely(nskb))
1214 consume_skb(skb);
1215 else
1216 kfree_skb(skb);
1217 skb = nskb;
1218 }
1219 return skb;
1220 }
1221
1222 /**
1223 * skb_peek - peek at the head of an &sk_buff_head
1224 * @list_: list to peek at
1225 *
1226 * Peek an &sk_buff. Unlike most other operations you _MUST_
1227 * be careful with this one. A peek leaves the buffer on the
1228 * list and someone else may run off with it. You must hold
1229 * the appropriate locks or have a private queue to do this.
1230 *
1231 * Returns %NULL for an empty list or a pointer to the head element.
1232 * The reference count is not incremented and the reference is therefore
1233 * volatile. Use with caution.
1234 */
1235 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1236 {
1237 struct sk_buff *skb = list_->next;
1238
1239 if (skb == (struct sk_buff *)list_)
1240 skb = NULL;
1241 return skb;
1242 }
1243
1244 /**
1245 * skb_peek_next - peek skb following the given one from a queue
1246 * @skb: skb to start from
1247 * @list_: list to peek at
1248 *
1249 * Returns %NULL when the end of the list is met or a pointer to the
1250 * next element. The reference count is not incremented and the
1251 * reference is therefore volatile. Use with caution.
1252 */
1253 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1254 const struct sk_buff_head *list_)
1255 {
1256 struct sk_buff *next = skb->next;
1257
1258 if (next == (struct sk_buff *)list_)
1259 next = NULL;
1260 return next;
1261 }
1262
1263 /**
1264 * skb_peek_tail - peek at the tail of an &sk_buff_head
1265 * @list_: list to peek at
1266 *
1267 * Peek an &sk_buff. Unlike most other operations you _MUST_
1268 * be careful with this one. A peek leaves the buffer on the
1269 * list and someone else may run off with it. You must hold
1270 * the appropriate locks or have a private queue to do this.
1271 *
1272 * Returns %NULL for an empty list or a pointer to the tail element.
1273 * The reference count is not incremented and the reference is therefore
1274 * volatile. Use with caution.
1275 */
1276 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1277 {
1278 struct sk_buff *skb = list_->prev;
1279
1280 if (skb == (struct sk_buff *)list_)
1281 skb = NULL;
1282 return skb;
1283
1284 }
1285
1286 /**
1287 * skb_queue_len - get queue length
1288 * @list_: list to measure
1289 *
1290 * Return the length of an &sk_buff queue.
1291 */
1292 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1293 {
1294 return list_->qlen;
1295 }
1296
1297 /**
1298 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1299 * @list: queue to initialize
1300 *
1301 * This initializes only the list and queue length aspects of
1302 * an sk_buff_head object. This allows to initialize the list
1303 * aspects of an sk_buff_head without reinitializing things like
1304 * the spinlock. It can also be used for on-stack sk_buff_head
1305 * objects where the spinlock is known to not be used.
1306 */
1307 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1308 {
1309 list->prev = list->next = (struct sk_buff *)list;
1310 list->qlen = 0;
1311 }
1312
1313 /*
1314 * This function creates a split out lock class for each invocation;
1315 * this is needed for now since a whole lot of users of the skb-queue
1316 * infrastructure in drivers have different locking usage (in hardirq)
1317 * than the networking core (in softirq only). In the long run either the
1318 * network layer or drivers should need annotation to consolidate the
1319 * main types of usage into 3 classes.
1320 */
1321 static inline void skb_queue_head_init(struct sk_buff_head *list)
1322 {
1323 spin_lock_init(&list->lock);
1324 __skb_queue_head_init(list);
1325 }
1326
1327 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1328 struct lock_class_key *class)
1329 {
1330 skb_queue_head_init(list);
1331 lockdep_set_class(&list->lock, class);
1332 }
1333
1334 /*
1335 * Insert an sk_buff on a list.
1336 *
1337 * The "__skb_xxxx()" functions are the non-atomic ones that
1338 * can only be called with interrupts disabled.
1339 */
1340 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1341 struct sk_buff_head *list);
1342 static inline void __skb_insert(struct sk_buff *newsk,
1343 struct sk_buff *prev, struct sk_buff *next,
1344 struct sk_buff_head *list)
1345 {
1346 newsk->next = next;
1347 newsk->prev = prev;
1348 next->prev = prev->next = newsk;
1349 list->qlen++;
1350 }
1351
1352 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1353 struct sk_buff *prev,
1354 struct sk_buff *next)
1355 {
1356 struct sk_buff *first = list->next;
1357 struct sk_buff *last = list->prev;
1358
1359 first->prev = prev;
1360 prev->next = first;
1361
1362 last->next = next;
1363 next->prev = last;
1364 }
1365
1366 /**
1367 * skb_queue_splice - join two skb lists, this is designed for stacks
1368 * @list: the new list to add
1369 * @head: the place to add it in the first list
1370 */
1371 static inline void skb_queue_splice(const struct sk_buff_head *list,
1372 struct sk_buff_head *head)
1373 {
1374 if (!skb_queue_empty(list)) {
1375 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1376 head->qlen += list->qlen;
1377 }
1378 }
1379
1380 /**
1381 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1382 * @list: the new list to add
1383 * @head: the place to add it in the first list
1384 *
1385 * The list at @list is reinitialised
1386 */
1387 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1388 struct sk_buff_head *head)
1389 {
1390 if (!skb_queue_empty(list)) {
1391 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1392 head->qlen += list->qlen;
1393 __skb_queue_head_init(list);
1394 }
1395 }
1396
1397 /**
1398 * skb_queue_splice_tail - join two skb lists, each list being a queue
1399 * @list: the new list to add
1400 * @head: the place to add it in the first list
1401 */
1402 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1403 struct sk_buff_head *head)
1404 {
1405 if (!skb_queue_empty(list)) {
1406 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1407 head->qlen += list->qlen;
1408 }
1409 }
1410
1411 /**
1412 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1413 * @list: the new list to add
1414 * @head: the place to add it in the first list
1415 *
1416 * Each of the lists is a queue.
1417 * The list at @list is reinitialised
1418 */
1419 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1420 struct sk_buff_head *head)
1421 {
1422 if (!skb_queue_empty(list)) {
1423 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1424 head->qlen += list->qlen;
1425 __skb_queue_head_init(list);
1426 }
1427 }
1428
1429 /**
1430 * __skb_queue_after - queue a buffer at the list head
1431 * @list: list to use
1432 * @prev: place after this buffer
1433 * @newsk: buffer to queue
1434 *
1435 * Queue a buffer int the middle of a list. This function takes no locks
1436 * and you must therefore hold required locks before calling it.
1437 *
1438 * A buffer cannot be placed on two lists at the same time.
1439 */
1440 static inline void __skb_queue_after(struct sk_buff_head *list,
1441 struct sk_buff *prev,
1442 struct sk_buff *newsk)
1443 {
1444 __skb_insert(newsk, prev, prev->next, list);
1445 }
1446
1447 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1448 struct sk_buff_head *list);
1449
1450 static inline void __skb_queue_before(struct sk_buff_head *list,
1451 struct sk_buff *next,
1452 struct sk_buff *newsk)
1453 {
1454 __skb_insert(newsk, next->prev, next, list);
1455 }
1456
1457 /**
1458 * __skb_queue_head - queue a buffer at the list head
1459 * @list: list to use
1460 * @newsk: buffer to queue
1461 *
1462 * Queue a buffer at the start of a list. This function takes no locks
1463 * and you must therefore hold required locks before calling it.
1464 *
1465 * A buffer cannot be placed on two lists at the same time.
1466 */
1467 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1468 static inline void __skb_queue_head(struct sk_buff_head *list,
1469 struct sk_buff *newsk)
1470 {
1471 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1472 }
1473
1474 /**
1475 * __skb_queue_tail - queue a buffer at the list tail
1476 * @list: list to use
1477 * @newsk: buffer to queue
1478 *
1479 * Queue a buffer at the end of a list. This function takes no locks
1480 * and you must therefore hold required locks before calling it.
1481 *
1482 * A buffer cannot be placed on two lists at the same time.
1483 */
1484 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1485 static inline void __skb_queue_tail(struct sk_buff_head *list,
1486 struct sk_buff *newsk)
1487 {
1488 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1489 }
1490
1491 /*
1492 * remove sk_buff from list. _Must_ be called atomically, and with
1493 * the list known..
1494 */
1495 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1496 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1497 {
1498 struct sk_buff *next, *prev;
1499
1500 list->qlen--;
1501 next = skb->next;
1502 prev = skb->prev;
1503 skb->next = skb->prev = NULL;
1504 next->prev = prev;
1505 prev->next = next;
1506 }
1507
1508 /**
1509 * __skb_dequeue - remove from the head of the queue
1510 * @list: list to dequeue from
1511 *
1512 * Remove the head of the list. This function does not take any locks
1513 * so must be used with appropriate locks held only. The head item is
1514 * returned or %NULL if the list is empty.
1515 */
1516 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1517 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1518 {
1519 struct sk_buff *skb = skb_peek(list);
1520 if (skb)
1521 __skb_unlink(skb, list);
1522 return skb;
1523 }
1524
1525 /**
1526 * __skb_dequeue_tail - remove from the tail of the queue
1527 * @list: list to dequeue from
1528 *
1529 * Remove the tail of the list. This function does not take any locks
1530 * so must be used with appropriate locks held only. The tail item is
1531 * returned or %NULL if the list is empty.
1532 */
1533 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1534 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1535 {
1536 struct sk_buff *skb = skb_peek_tail(list);
1537 if (skb)
1538 __skb_unlink(skb, list);
1539 return skb;
1540 }
1541
1542
1543 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1544 {
1545 return skb->data_len;
1546 }
1547
1548 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1549 {
1550 return skb->len - skb->data_len;
1551 }
1552
1553 static inline int skb_pagelen(const struct sk_buff *skb)
1554 {
1555 int i, len = 0;
1556
1557 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1558 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1559 return len + skb_headlen(skb);
1560 }
1561
1562 /**
1563 * __skb_fill_page_desc - initialise a paged fragment in an skb
1564 * @skb: buffer containing fragment to be initialised
1565 * @i: paged fragment index to initialise
1566 * @page: the page to use for this fragment
1567 * @off: the offset to the data with @page
1568 * @size: the length of the data
1569 *
1570 * Initialises the @i'th fragment of @skb to point to &size bytes at
1571 * offset @off within @page.
1572 *
1573 * Does not take any additional reference on the fragment.
1574 */
1575 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1576 struct page *page, int off, int size)
1577 {
1578 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1579
1580 /*
1581 * Propagate page->pfmemalloc to the skb if we can. The problem is
1582 * that not all callers have unique ownership of the page. If
1583 * pfmemalloc is set, we check the mapping as a mapping implies
1584 * page->index is set (index and pfmemalloc share space).
1585 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1586 * do not lose pfmemalloc information as the pages would not be
1587 * allocated using __GFP_MEMALLOC.
1588 */
1589 frag->page.p = page;
1590 frag->page_offset = off;
1591 skb_frag_size_set(frag, size);
1592
1593 page = compound_head(page);
1594 if (page->pfmemalloc && !page->mapping)
1595 skb->pfmemalloc = true;
1596 }
1597
1598 /**
1599 * skb_fill_page_desc - initialise a paged fragment in an skb
1600 * @skb: buffer containing fragment to be initialised
1601 * @i: paged fragment index to initialise
1602 * @page: the page to use for this fragment
1603 * @off: the offset to the data with @page
1604 * @size: the length of the data
1605 *
1606 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1607 * @skb to point to @size bytes at offset @off within @page. In
1608 * addition updates @skb such that @i is the last fragment.
1609 *
1610 * Does not take any additional reference on the fragment.
1611 */
1612 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1613 struct page *page, int off, int size)
1614 {
1615 __skb_fill_page_desc(skb, i, page, off, size);
1616 skb_shinfo(skb)->nr_frags = i + 1;
1617 }
1618
1619 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1620 int size, unsigned int truesize);
1621
1622 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1623 unsigned int truesize);
1624
1625 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1626 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1627 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1628
1629 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1630 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1631 {
1632 return skb->head + skb->tail;
1633 }
1634
1635 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1636 {
1637 skb->tail = skb->data - skb->head;
1638 }
1639
1640 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1641 {
1642 skb_reset_tail_pointer(skb);
1643 skb->tail += offset;
1644 }
1645
1646 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1647 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1648 {
1649 return skb->tail;
1650 }
1651
1652 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1653 {
1654 skb->tail = skb->data;
1655 }
1656
1657 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1658 {
1659 skb->tail = skb->data + offset;
1660 }
1661
1662 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1663
1664 /*
1665 * Add data to an sk_buff
1666 */
1667 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1668 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1669 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1670 {
1671 unsigned char *tmp = skb_tail_pointer(skb);
1672 SKB_LINEAR_ASSERT(skb);
1673 skb->tail += len;
1674 skb->len += len;
1675 return tmp;
1676 }
1677
1678 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1679 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1680 {
1681 skb->data -= len;
1682 skb->len += len;
1683 return skb->data;
1684 }
1685
1686 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1687 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1688 {
1689 skb->len -= len;
1690 BUG_ON(skb->len < skb->data_len);
1691 return skb->data += len;
1692 }
1693
1694 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1695 {
1696 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1697 }
1698
1699 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1700
1701 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1702 {
1703 if (len > skb_headlen(skb) &&
1704 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1705 return NULL;
1706 skb->len -= len;
1707 return skb->data += len;
1708 }
1709
1710 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1711 {
1712 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1713 }
1714
1715 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1716 {
1717 if (likely(len <= skb_headlen(skb)))
1718 return 1;
1719 if (unlikely(len > skb->len))
1720 return 0;
1721 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1722 }
1723
1724 /**
1725 * skb_headroom - bytes at buffer head
1726 * @skb: buffer to check
1727 *
1728 * Return the number of bytes of free space at the head of an &sk_buff.
1729 */
1730 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1731 {
1732 return skb->data - skb->head;
1733 }
1734
1735 /**
1736 * skb_tailroom - bytes at buffer end
1737 * @skb: buffer to check
1738 *
1739 * Return the number of bytes of free space at the tail of an sk_buff
1740 */
1741 static inline int skb_tailroom(const struct sk_buff *skb)
1742 {
1743 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1744 }
1745
1746 /**
1747 * skb_availroom - bytes at buffer end
1748 * @skb: buffer to check
1749 *
1750 * Return the number of bytes of free space at the tail of an sk_buff
1751 * allocated by sk_stream_alloc()
1752 */
1753 static inline int skb_availroom(const struct sk_buff *skb)
1754 {
1755 if (skb_is_nonlinear(skb))
1756 return 0;
1757
1758 return skb->end - skb->tail - skb->reserved_tailroom;
1759 }
1760
1761 /**
1762 * skb_reserve - adjust headroom
1763 * @skb: buffer to alter
1764 * @len: bytes to move
1765 *
1766 * Increase the headroom of an empty &sk_buff by reducing the tail
1767 * room. This is only allowed for an empty buffer.
1768 */
1769 static inline void skb_reserve(struct sk_buff *skb, int len)
1770 {
1771 skb->data += len;
1772 skb->tail += len;
1773 }
1774
1775 #define ENCAP_TYPE_ETHER 0
1776 #define ENCAP_TYPE_IPPROTO 1
1777
1778 static inline void skb_set_inner_protocol(struct sk_buff *skb,
1779 __be16 protocol)
1780 {
1781 skb->inner_protocol = protocol;
1782 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1783 }
1784
1785 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1786 __u8 ipproto)
1787 {
1788 skb->inner_ipproto = ipproto;
1789 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1790 }
1791
1792 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1793 {
1794 skb->inner_mac_header = skb->mac_header;
1795 skb->inner_network_header = skb->network_header;
1796 skb->inner_transport_header = skb->transport_header;
1797 }
1798
1799 static inline void skb_reset_mac_len(struct sk_buff *skb)
1800 {
1801 skb->mac_len = skb->network_header - skb->mac_header;
1802 }
1803
1804 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1805 *skb)
1806 {
1807 return skb->head + skb->inner_transport_header;
1808 }
1809
1810 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1811 {
1812 skb->inner_transport_header = skb->data - skb->head;
1813 }
1814
1815 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1816 const int offset)
1817 {
1818 skb_reset_inner_transport_header(skb);
1819 skb->inner_transport_header += offset;
1820 }
1821
1822 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1823 {
1824 return skb->head + skb->inner_network_header;
1825 }
1826
1827 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1828 {
1829 skb->inner_network_header = skb->data - skb->head;
1830 }
1831
1832 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1833 const int offset)
1834 {
1835 skb_reset_inner_network_header(skb);
1836 skb->inner_network_header += offset;
1837 }
1838
1839 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1840 {
1841 return skb->head + skb->inner_mac_header;
1842 }
1843
1844 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1845 {
1846 skb->inner_mac_header = skb->data - skb->head;
1847 }
1848
1849 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1850 const int offset)
1851 {
1852 skb_reset_inner_mac_header(skb);
1853 skb->inner_mac_header += offset;
1854 }
1855 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1856 {
1857 return skb->transport_header != (typeof(skb->transport_header))~0U;
1858 }
1859
1860 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1861 {
1862 return skb->head + skb->transport_header;
1863 }
1864
1865 static inline void skb_reset_transport_header(struct sk_buff *skb)
1866 {
1867 skb->transport_header = skb->data - skb->head;
1868 }
1869
1870 static inline void skb_set_transport_header(struct sk_buff *skb,
1871 const int offset)
1872 {
1873 skb_reset_transport_header(skb);
1874 skb->transport_header += offset;
1875 }
1876
1877 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1878 {
1879 return skb->head + skb->network_header;
1880 }
1881
1882 static inline void skb_reset_network_header(struct sk_buff *skb)
1883 {
1884 skb->network_header = skb->data - skb->head;
1885 }
1886
1887 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1888 {
1889 skb_reset_network_header(skb);
1890 skb->network_header += offset;
1891 }
1892
1893 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1894 {
1895 return skb->head + skb->mac_header;
1896 }
1897
1898 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1899 {
1900 return skb->mac_header != (typeof(skb->mac_header))~0U;
1901 }
1902
1903 static inline void skb_reset_mac_header(struct sk_buff *skb)
1904 {
1905 skb->mac_header = skb->data - skb->head;
1906 }
1907
1908 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1909 {
1910 skb_reset_mac_header(skb);
1911 skb->mac_header += offset;
1912 }
1913
1914 static inline void skb_pop_mac_header(struct sk_buff *skb)
1915 {
1916 skb->mac_header = skb->network_header;
1917 }
1918
1919 static inline void skb_probe_transport_header(struct sk_buff *skb,
1920 const int offset_hint)
1921 {
1922 struct flow_keys keys;
1923
1924 if (skb_transport_header_was_set(skb))
1925 return;
1926 else if (skb_flow_dissect(skb, &keys))
1927 skb_set_transport_header(skb, keys.thoff);
1928 else
1929 skb_set_transport_header(skb, offset_hint);
1930 }
1931
1932 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1933 {
1934 if (skb_mac_header_was_set(skb)) {
1935 const unsigned char *old_mac = skb_mac_header(skb);
1936
1937 skb_set_mac_header(skb, -skb->mac_len);
1938 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1939 }
1940 }
1941
1942 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1943 {
1944 return skb->csum_start - skb_headroom(skb);
1945 }
1946
1947 static inline int skb_transport_offset(const struct sk_buff *skb)
1948 {
1949 return skb_transport_header(skb) - skb->data;
1950 }
1951
1952 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1953 {
1954 return skb->transport_header - skb->network_header;
1955 }
1956
1957 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1958 {
1959 return skb->inner_transport_header - skb->inner_network_header;
1960 }
1961
1962 static inline int skb_network_offset(const struct sk_buff *skb)
1963 {
1964 return skb_network_header(skb) - skb->data;
1965 }
1966
1967 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1968 {
1969 return skb_inner_network_header(skb) - skb->data;
1970 }
1971
1972 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1973 {
1974 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1975 }
1976
1977 /*
1978 * CPUs often take a performance hit when accessing unaligned memory
1979 * locations. The actual performance hit varies, it can be small if the
1980 * hardware handles it or large if we have to take an exception and fix it
1981 * in software.
1982 *
1983 * Since an ethernet header is 14 bytes network drivers often end up with
1984 * the IP header at an unaligned offset. The IP header can be aligned by
1985 * shifting the start of the packet by 2 bytes. Drivers should do this
1986 * with:
1987 *
1988 * skb_reserve(skb, NET_IP_ALIGN);
1989 *
1990 * The downside to this alignment of the IP header is that the DMA is now
1991 * unaligned. On some architectures the cost of an unaligned DMA is high
1992 * and this cost outweighs the gains made by aligning the IP header.
1993 *
1994 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1995 * to be overridden.
1996 */
1997 #ifndef NET_IP_ALIGN
1998 #define NET_IP_ALIGN 2
1999 #endif
2000
2001 /*
2002 * The networking layer reserves some headroom in skb data (via
2003 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2004 * the header has to grow. In the default case, if the header has to grow
2005 * 32 bytes or less we avoid the reallocation.
2006 *
2007 * Unfortunately this headroom changes the DMA alignment of the resulting
2008 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2009 * on some architectures. An architecture can override this value,
2010 * perhaps setting it to a cacheline in size (since that will maintain
2011 * cacheline alignment of the DMA). It must be a power of 2.
2012 *
2013 * Various parts of the networking layer expect at least 32 bytes of
2014 * headroom, you should not reduce this.
2015 *
2016 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2017 * to reduce average number of cache lines per packet.
2018 * get_rps_cpus() for example only access one 64 bytes aligned block :
2019 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2020 */
2021 #ifndef NET_SKB_PAD
2022 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2023 #endif
2024
2025 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2026
2027 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2028 {
2029 if (unlikely(skb_is_nonlinear(skb))) {
2030 WARN_ON(1);
2031 return;
2032 }
2033 skb->len = len;
2034 skb_set_tail_pointer(skb, len);
2035 }
2036
2037 void skb_trim(struct sk_buff *skb, unsigned int len);
2038
2039 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2040 {
2041 if (skb->data_len)
2042 return ___pskb_trim(skb, len);
2043 __skb_trim(skb, len);
2044 return 0;
2045 }
2046
2047 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2048 {
2049 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2050 }
2051
2052 /**
2053 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2054 * @skb: buffer to alter
2055 * @len: new length
2056 *
2057 * This is identical to pskb_trim except that the caller knows that
2058 * the skb is not cloned so we should never get an error due to out-
2059 * of-memory.
2060 */
2061 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2062 {
2063 int err = pskb_trim(skb, len);
2064 BUG_ON(err);
2065 }
2066
2067 /**
2068 * skb_orphan - orphan a buffer
2069 * @skb: buffer to orphan
2070 *
2071 * If a buffer currently has an owner then we call the owner's
2072 * destructor function and make the @skb unowned. The buffer continues
2073 * to exist but is no longer charged to its former owner.
2074 */
2075 static inline void skb_orphan(struct sk_buff *skb)
2076 {
2077 if (skb->destructor) {
2078 skb->destructor(skb);
2079 skb->destructor = NULL;
2080 skb->sk = NULL;
2081 } else {
2082 BUG_ON(skb->sk);
2083 }
2084 }
2085
2086 /**
2087 * skb_orphan_frags - orphan the frags contained in a buffer
2088 * @skb: buffer to orphan frags from
2089 * @gfp_mask: allocation mask for replacement pages
2090 *
2091 * For each frag in the SKB which needs a destructor (i.e. has an
2092 * owner) create a copy of that frag and release the original
2093 * page by calling the destructor.
2094 */
2095 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2096 {
2097 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2098 return 0;
2099 return skb_copy_ubufs(skb, gfp_mask);
2100 }
2101
2102 /**
2103 * __skb_queue_purge - empty a list
2104 * @list: list to empty
2105 *
2106 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2107 * the list and one reference dropped. This function does not take the
2108 * list lock and the caller must hold the relevant locks to use it.
2109 */
2110 void skb_queue_purge(struct sk_buff_head *list);
2111 static inline void __skb_queue_purge(struct sk_buff_head *list)
2112 {
2113 struct sk_buff *skb;
2114 while ((skb = __skb_dequeue(list)) != NULL)
2115 kfree_skb(skb);
2116 }
2117
2118 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2119 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2120 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2121
2122 void *netdev_alloc_frag(unsigned int fragsz);
2123
2124 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2125 gfp_t gfp_mask);
2126
2127 /**
2128 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2129 * @dev: network device to receive on
2130 * @length: length to allocate
2131 *
2132 * Allocate a new &sk_buff and assign it a usage count of one. The
2133 * buffer has unspecified headroom built in. Users should allocate
2134 * the headroom they think they need without accounting for the
2135 * built in space. The built in space is used for optimisations.
2136 *
2137 * %NULL is returned if there is no free memory. Although this function
2138 * allocates memory it can be called from an interrupt.
2139 */
2140 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2141 unsigned int length)
2142 {
2143 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2144 }
2145
2146 /* legacy helper around __netdev_alloc_skb() */
2147 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2148 gfp_t gfp_mask)
2149 {
2150 return __netdev_alloc_skb(NULL, length, gfp_mask);
2151 }
2152
2153 /* legacy helper around netdev_alloc_skb() */
2154 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2155 {
2156 return netdev_alloc_skb(NULL, length);
2157 }
2158
2159
2160 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2161 unsigned int length, gfp_t gfp)
2162 {
2163 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2164
2165 if (NET_IP_ALIGN && skb)
2166 skb_reserve(skb, NET_IP_ALIGN);
2167 return skb;
2168 }
2169
2170 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2171 unsigned int length)
2172 {
2173 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2174 }
2175
2176 void *napi_alloc_frag(unsigned int fragsz);
2177 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2178 unsigned int length, gfp_t gfp_mask);
2179 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2180 unsigned int length)
2181 {
2182 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2183 }
2184
2185 /**
2186 * __dev_alloc_pages - allocate page for network Rx
2187 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2188 * @order: size of the allocation
2189 *
2190 * Allocate a new page.
2191 *
2192 * %NULL is returned if there is no free memory.
2193 */
2194 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2195 unsigned int order)
2196 {
2197 /* This piece of code contains several assumptions.
2198 * 1. This is for device Rx, therefor a cold page is preferred.
2199 * 2. The expectation is the user wants a compound page.
2200 * 3. If requesting a order 0 page it will not be compound
2201 * due to the check to see if order has a value in prep_new_page
2202 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2203 * code in gfp_to_alloc_flags that should be enforcing this.
2204 */
2205 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2206
2207 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2208 }
2209
2210 static inline struct page *dev_alloc_pages(unsigned int order)
2211 {
2212 return __dev_alloc_pages(GFP_ATOMIC, order);
2213 }
2214
2215 /**
2216 * __dev_alloc_page - allocate a page for network Rx
2217 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2218 *
2219 * Allocate a new page.
2220 *
2221 * %NULL is returned if there is no free memory.
2222 */
2223 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2224 {
2225 return __dev_alloc_pages(gfp_mask, 0);
2226 }
2227
2228 static inline struct page *dev_alloc_page(void)
2229 {
2230 return __dev_alloc_page(GFP_ATOMIC);
2231 }
2232
2233 /**
2234 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2235 * @page: The page that was allocated from skb_alloc_page
2236 * @skb: The skb that may need pfmemalloc set
2237 */
2238 static inline void skb_propagate_pfmemalloc(struct page *page,
2239 struct sk_buff *skb)
2240 {
2241 if (page && page->pfmemalloc)
2242 skb->pfmemalloc = true;
2243 }
2244
2245 /**
2246 * skb_frag_page - retrieve the page referred to by a paged fragment
2247 * @frag: the paged fragment
2248 *
2249 * Returns the &struct page associated with @frag.
2250 */
2251 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2252 {
2253 return frag->page.p;
2254 }
2255
2256 /**
2257 * __skb_frag_ref - take an addition reference on a paged fragment.
2258 * @frag: the paged fragment
2259 *
2260 * Takes an additional reference on the paged fragment @frag.
2261 */
2262 static inline void __skb_frag_ref(skb_frag_t *frag)
2263 {
2264 get_page(skb_frag_page(frag));
2265 }
2266
2267 /**
2268 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2269 * @skb: the buffer
2270 * @f: the fragment offset.
2271 *
2272 * Takes an additional reference on the @f'th paged fragment of @skb.
2273 */
2274 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2275 {
2276 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2277 }
2278
2279 /**
2280 * __skb_frag_unref - release a reference on a paged fragment.
2281 * @frag: the paged fragment
2282 *
2283 * Releases a reference on the paged fragment @frag.
2284 */
2285 static inline void __skb_frag_unref(skb_frag_t *frag)
2286 {
2287 put_page(skb_frag_page(frag));
2288 }
2289
2290 /**
2291 * skb_frag_unref - release a reference on a paged fragment of an skb.
2292 * @skb: the buffer
2293 * @f: the fragment offset
2294 *
2295 * Releases a reference on the @f'th paged fragment of @skb.
2296 */
2297 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2298 {
2299 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2300 }
2301
2302 /**
2303 * skb_frag_address - gets the address of the data contained in a paged fragment
2304 * @frag: the paged fragment buffer
2305 *
2306 * Returns the address of the data within @frag. The page must already
2307 * be mapped.
2308 */
2309 static inline void *skb_frag_address(const skb_frag_t *frag)
2310 {
2311 return page_address(skb_frag_page(frag)) + frag->page_offset;
2312 }
2313
2314 /**
2315 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2316 * @frag: the paged fragment buffer
2317 *
2318 * Returns the address of the data within @frag. Checks that the page
2319 * is mapped and returns %NULL otherwise.
2320 */
2321 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2322 {
2323 void *ptr = page_address(skb_frag_page(frag));
2324 if (unlikely(!ptr))
2325 return NULL;
2326
2327 return ptr + frag->page_offset;
2328 }
2329
2330 /**
2331 * __skb_frag_set_page - sets the page contained in a paged fragment
2332 * @frag: the paged fragment
2333 * @page: the page to set
2334 *
2335 * Sets the fragment @frag to contain @page.
2336 */
2337 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2338 {
2339 frag->page.p = page;
2340 }
2341
2342 /**
2343 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2344 * @skb: the buffer
2345 * @f: the fragment offset
2346 * @page: the page to set
2347 *
2348 * Sets the @f'th fragment of @skb to contain @page.
2349 */
2350 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2351 struct page *page)
2352 {
2353 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2354 }
2355
2356 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2357
2358 /**
2359 * skb_frag_dma_map - maps a paged fragment via the DMA API
2360 * @dev: the device to map the fragment to
2361 * @frag: the paged fragment to map
2362 * @offset: the offset within the fragment (starting at the
2363 * fragment's own offset)
2364 * @size: the number of bytes to map
2365 * @dir: the direction of the mapping (%PCI_DMA_*)
2366 *
2367 * Maps the page associated with @frag to @device.
2368 */
2369 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2370 const skb_frag_t *frag,
2371 size_t offset, size_t size,
2372 enum dma_data_direction dir)
2373 {
2374 return dma_map_page(dev, skb_frag_page(frag),
2375 frag->page_offset + offset, size, dir);
2376 }
2377
2378 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2379 gfp_t gfp_mask)
2380 {
2381 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2382 }
2383
2384
2385 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2386 gfp_t gfp_mask)
2387 {
2388 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2389 }
2390
2391
2392 /**
2393 * skb_clone_writable - is the header of a clone writable
2394 * @skb: buffer to check
2395 * @len: length up to which to write
2396 *
2397 * Returns true if modifying the header part of the cloned buffer
2398 * does not requires the data to be copied.
2399 */
2400 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2401 {
2402 return !skb_header_cloned(skb) &&
2403 skb_headroom(skb) + len <= skb->hdr_len;
2404 }
2405
2406 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2407 int cloned)
2408 {
2409 int delta = 0;
2410
2411 if (headroom > skb_headroom(skb))
2412 delta = headroom - skb_headroom(skb);
2413
2414 if (delta || cloned)
2415 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2416 GFP_ATOMIC);
2417 return 0;
2418 }
2419
2420 /**
2421 * skb_cow - copy header of skb when it is required
2422 * @skb: buffer to cow
2423 * @headroom: needed headroom
2424 *
2425 * If the skb passed lacks sufficient headroom or its data part
2426 * is shared, data is reallocated. If reallocation fails, an error
2427 * is returned and original skb is not changed.
2428 *
2429 * The result is skb with writable area skb->head...skb->tail
2430 * and at least @headroom of space at head.
2431 */
2432 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2433 {
2434 return __skb_cow(skb, headroom, skb_cloned(skb));
2435 }
2436
2437 /**
2438 * skb_cow_head - skb_cow but only making the head writable
2439 * @skb: buffer to cow
2440 * @headroom: needed headroom
2441 *
2442 * This function is identical to skb_cow except that we replace the
2443 * skb_cloned check by skb_header_cloned. It should be used when
2444 * you only need to push on some header and do not need to modify
2445 * the data.
2446 */
2447 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2448 {
2449 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2450 }
2451
2452 /**
2453 * skb_padto - pad an skbuff up to a minimal size
2454 * @skb: buffer to pad
2455 * @len: minimal length
2456 *
2457 * Pads up a buffer to ensure the trailing bytes exist and are
2458 * blanked. If the buffer already contains sufficient data it
2459 * is untouched. Otherwise it is extended. Returns zero on
2460 * success. The skb is freed on error.
2461 */
2462 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2463 {
2464 unsigned int size = skb->len;
2465 if (likely(size >= len))
2466 return 0;
2467 return skb_pad(skb, len - size);
2468 }
2469
2470 /**
2471 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2472 * @skb: buffer to pad
2473 * @len: minimal length
2474 *
2475 * Pads up a buffer to ensure the trailing bytes exist and are
2476 * blanked. If the buffer already contains sufficient data it
2477 * is untouched. Otherwise it is extended. Returns zero on
2478 * success. The skb is freed on error.
2479 */
2480 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2481 {
2482 unsigned int size = skb->len;
2483
2484 if (unlikely(size < len)) {
2485 len -= size;
2486 if (skb_pad(skb, len))
2487 return -ENOMEM;
2488 __skb_put(skb, len);
2489 }
2490 return 0;
2491 }
2492
2493 static inline int skb_add_data(struct sk_buff *skb,
2494 struct iov_iter *from, int copy)
2495 {
2496 const int off = skb->len;
2497
2498 if (skb->ip_summed == CHECKSUM_NONE) {
2499 __wsum csum = 0;
2500 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2501 &csum, from) == copy) {
2502 skb->csum = csum_block_add(skb->csum, csum, off);
2503 return 0;
2504 }
2505 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2506 return 0;
2507
2508 __skb_trim(skb, off);
2509 return -EFAULT;
2510 }
2511
2512 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2513 const struct page *page, int off)
2514 {
2515 if (i) {
2516 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2517
2518 return page == skb_frag_page(frag) &&
2519 off == frag->page_offset + skb_frag_size(frag);
2520 }
2521 return false;
2522 }
2523
2524 static inline int __skb_linearize(struct sk_buff *skb)
2525 {
2526 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2527 }
2528
2529 /**
2530 * skb_linearize - convert paged skb to linear one
2531 * @skb: buffer to linarize
2532 *
2533 * If there is no free memory -ENOMEM is returned, otherwise zero
2534 * is returned and the old skb data released.
2535 */
2536 static inline int skb_linearize(struct sk_buff *skb)
2537 {
2538 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2539 }
2540
2541 /**
2542 * skb_has_shared_frag - can any frag be overwritten
2543 * @skb: buffer to test
2544 *
2545 * Return true if the skb has at least one frag that might be modified
2546 * by an external entity (as in vmsplice()/sendfile())
2547 */
2548 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2549 {
2550 return skb_is_nonlinear(skb) &&
2551 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2552 }
2553
2554 /**
2555 * skb_linearize_cow - make sure skb is linear and writable
2556 * @skb: buffer to process
2557 *
2558 * If there is no free memory -ENOMEM is returned, otherwise zero
2559 * is returned and the old skb data released.
2560 */
2561 static inline int skb_linearize_cow(struct sk_buff *skb)
2562 {
2563 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2564 __skb_linearize(skb) : 0;
2565 }
2566
2567 /**
2568 * skb_postpull_rcsum - update checksum for received skb after pull
2569 * @skb: buffer to update
2570 * @start: start of data before pull
2571 * @len: length of data pulled
2572 *
2573 * After doing a pull on a received packet, you need to call this to
2574 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2575 * CHECKSUM_NONE so that it can be recomputed from scratch.
2576 */
2577
2578 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2579 const void *start, unsigned int len)
2580 {
2581 if (skb->ip_summed == CHECKSUM_COMPLETE)
2582 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2583 }
2584
2585 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2586
2587 /**
2588 * pskb_trim_rcsum - trim received skb and update checksum
2589 * @skb: buffer to trim
2590 * @len: new length
2591 *
2592 * This is exactly the same as pskb_trim except that it ensures the
2593 * checksum of received packets are still valid after the operation.
2594 */
2595
2596 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2597 {
2598 if (likely(len >= skb->len))
2599 return 0;
2600 if (skb->ip_summed == CHECKSUM_COMPLETE)
2601 skb->ip_summed = CHECKSUM_NONE;
2602 return __pskb_trim(skb, len);
2603 }
2604
2605 #define skb_queue_walk(queue, skb) \
2606 for (skb = (queue)->next; \
2607 skb != (struct sk_buff *)(queue); \
2608 skb = skb->next)
2609
2610 #define skb_queue_walk_safe(queue, skb, tmp) \
2611 for (skb = (queue)->next, tmp = skb->next; \
2612 skb != (struct sk_buff *)(queue); \
2613 skb = tmp, tmp = skb->next)
2614
2615 #define skb_queue_walk_from(queue, skb) \
2616 for (; skb != (struct sk_buff *)(queue); \
2617 skb = skb->next)
2618
2619 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2620 for (tmp = skb->next; \
2621 skb != (struct sk_buff *)(queue); \
2622 skb = tmp, tmp = skb->next)
2623
2624 #define skb_queue_reverse_walk(queue, skb) \
2625 for (skb = (queue)->prev; \
2626 skb != (struct sk_buff *)(queue); \
2627 skb = skb->prev)
2628
2629 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2630 for (skb = (queue)->prev, tmp = skb->prev; \
2631 skb != (struct sk_buff *)(queue); \
2632 skb = tmp, tmp = skb->prev)
2633
2634 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2635 for (tmp = skb->prev; \
2636 skb != (struct sk_buff *)(queue); \
2637 skb = tmp, tmp = skb->prev)
2638
2639 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2640 {
2641 return skb_shinfo(skb)->frag_list != NULL;
2642 }
2643
2644 static inline void skb_frag_list_init(struct sk_buff *skb)
2645 {
2646 skb_shinfo(skb)->frag_list = NULL;
2647 }
2648
2649 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2650 {
2651 frag->next = skb_shinfo(skb)->frag_list;
2652 skb_shinfo(skb)->frag_list = frag;
2653 }
2654
2655 #define skb_walk_frags(skb, iter) \
2656 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2657
2658 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2659 int *peeked, int *off, int *err);
2660 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2661 int *err);
2662 unsigned int datagram_poll(struct file *file, struct socket *sock,
2663 struct poll_table_struct *wait);
2664 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2665 struct iov_iter *to, int size);
2666 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2667 struct msghdr *msg, int size)
2668 {
2669 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2670 }
2671 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2672 struct msghdr *msg);
2673 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2674 struct iov_iter *from, int len);
2675 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2676 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2677 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2678 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2679 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2680 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2681 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2682 int len, __wsum csum);
2683 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2684 struct pipe_inode_info *pipe, unsigned int len,
2685 unsigned int flags);
2686 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2687 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2688 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2689 int len, int hlen);
2690 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2691 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2692 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2693 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2694 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2695 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2696 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2697 int skb_vlan_pop(struct sk_buff *skb);
2698 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
2699
2700 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2701 {
2702 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2703 }
2704
2705 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2706 {
2707 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2708 }
2709
2710 struct skb_checksum_ops {
2711 __wsum (*update)(const void *mem, int len, __wsum wsum);
2712 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2713 };
2714
2715 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2716 __wsum csum, const struct skb_checksum_ops *ops);
2717 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2718 __wsum csum);
2719
2720 static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset,
2721 int len, void *data, int hlen, void *buffer)
2722 {
2723 if (hlen - offset >= len)
2724 return data + offset;
2725
2726 if (!skb ||
2727 skb_copy_bits(skb, offset, buffer, len) < 0)
2728 return NULL;
2729
2730 return buffer;
2731 }
2732
2733 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2734 int len, void *buffer)
2735 {
2736 return __skb_header_pointer(skb, offset, len, skb->data,
2737 skb_headlen(skb), buffer);
2738 }
2739
2740 /**
2741 * skb_needs_linearize - check if we need to linearize a given skb
2742 * depending on the given device features.
2743 * @skb: socket buffer to check
2744 * @features: net device features
2745 *
2746 * Returns true if either:
2747 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2748 * 2. skb is fragmented and the device does not support SG.
2749 */
2750 static inline bool skb_needs_linearize(struct sk_buff *skb,
2751 netdev_features_t features)
2752 {
2753 return skb_is_nonlinear(skb) &&
2754 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2755 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2756 }
2757
2758 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2759 void *to,
2760 const unsigned int len)
2761 {
2762 memcpy(to, skb->data, len);
2763 }
2764
2765 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2766 const int offset, void *to,
2767 const unsigned int len)
2768 {
2769 memcpy(to, skb->data + offset, len);
2770 }
2771
2772 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2773 const void *from,
2774 const unsigned int len)
2775 {
2776 memcpy(skb->data, from, len);
2777 }
2778
2779 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2780 const int offset,
2781 const void *from,
2782 const unsigned int len)
2783 {
2784 memcpy(skb->data + offset, from, len);
2785 }
2786
2787 void skb_init(void);
2788
2789 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2790 {
2791 return skb->tstamp;
2792 }
2793
2794 /**
2795 * skb_get_timestamp - get timestamp from a skb
2796 * @skb: skb to get stamp from
2797 * @stamp: pointer to struct timeval to store stamp in
2798 *
2799 * Timestamps are stored in the skb as offsets to a base timestamp.
2800 * This function converts the offset back to a struct timeval and stores
2801 * it in stamp.
2802 */
2803 static inline void skb_get_timestamp(const struct sk_buff *skb,
2804 struct timeval *stamp)
2805 {
2806 *stamp = ktime_to_timeval(skb->tstamp);
2807 }
2808
2809 static inline void skb_get_timestampns(const struct sk_buff *skb,
2810 struct timespec *stamp)
2811 {
2812 *stamp = ktime_to_timespec(skb->tstamp);
2813 }
2814
2815 static inline void __net_timestamp(struct sk_buff *skb)
2816 {
2817 skb->tstamp = ktime_get_real();
2818 }
2819
2820 static inline ktime_t net_timedelta(ktime_t t)
2821 {
2822 return ktime_sub(ktime_get_real(), t);
2823 }
2824
2825 static inline ktime_t net_invalid_timestamp(void)
2826 {
2827 return ktime_set(0, 0);
2828 }
2829
2830 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2831
2832 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2833
2834 void skb_clone_tx_timestamp(struct sk_buff *skb);
2835 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2836
2837 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2838
2839 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2840 {
2841 }
2842
2843 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2844 {
2845 return false;
2846 }
2847
2848 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2849
2850 /**
2851 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2852 *
2853 * PHY drivers may accept clones of transmitted packets for
2854 * timestamping via their phy_driver.txtstamp method. These drivers
2855 * must call this function to return the skb back to the stack, with
2856 * or without a timestamp.
2857 *
2858 * @skb: clone of the the original outgoing packet
2859 * @hwtstamps: hardware time stamps, may be NULL if not available
2860 *
2861 */
2862 void skb_complete_tx_timestamp(struct sk_buff *skb,
2863 struct skb_shared_hwtstamps *hwtstamps);
2864
2865 void __skb_tstamp_tx(struct sk_buff *orig_skb,
2866 struct skb_shared_hwtstamps *hwtstamps,
2867 struct sock *sk, int tstype);
2868
2869 /**
2870 * skb_tstamp_tx - queue clone of skb with send time stamps
2871 * @orig_skb: the original outgoing packet
2872 * @hwtstamps: hardware time stamps, may be NULL if not available
2873 *
2874 * If the skb has a socket associated, then this function clones the
2875 * skb (thus sharing the actual data and optional structures), stores
2876 * the optional hardware time stamping information (if non NULL) or
2877 * generates a software time stamp (otherwise), then queues the clone
2878 * to the error queue of the socket. Errors are silently ignored.
2879 */
2880 void skb_tstamp_tx(struct sk_buff *orig_skb,
2881 struct skb_shared_hwtstamps *hwtstamps);
2882
2883 static inline void sw_tx_timestamp(struct sk_buff *skb)
2884 {
2885 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2886 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2887 skb_tstamp_tx(skb, NULL);
2888 }
2889
2890 /**
2891 * skb_tx_timestamp() - Driver hook for transmit timestamping
2892 *
2893 * Ethernet MAC Drivers should call this function in their hard_xmit()
2894 * function immediately before giving the sk_buff to the MAC hardware.
2895 *
2896 * Specifically, one should make absolutely sure that this function is
2897 * called before TX completion of this packet can trigger. Otherwise
2898 * the packet could potentially already be freed.
2899 *
2900 * @skb: A socket buffer.
2901 */
2902 static inline void skb_tx_timestamp(struct sk_buff *skb)
2903 {
2904 skb_clone_tx_timestamp(skb);
2905 sw_tx_timestamp(skb);
2906 }
2907
2908 /**
2909 * skb_complete_wifi_ack - deliver skb with wifi status
2910 *
2911 * @skb: the original outgoing packet
2912 * @acked: ack status
2913 *
2914 */
2915 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2916
2917 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2918 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2919
2920 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2921 {
2922 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
2923 skb->csum_valid ||
2924 (skb->ip_summed == CHECKSUM_PARTIAL &&
2925 skb_checksum_start_offset(skb) >= 0));
2926 }
2927
2928 /**
2929 * skb_checksum_complete - Calculate checksum of an entire packet
2930 * @skb: packet to process
2931 *
2932 * This function calculates the checksum over the entire packet plus
2933 * the value of skb->csum. The latter can be used to supply the
2934 * checksum of a pseudo header as used by TCP/UDP. It returns the
2935 * checksum.
2936 *
2937 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2938 * this function can be used to verify that checksum on received
2939 * packets. In that case the function should return zero if the
2940 * checksum is correct. In particular, this function will return zero
2941 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2942 * hardware has already verified the correctness of the checksum.
2943 */
2944 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2945 {
2946 return skb_csum_unnecessary(skb) ?
2947 0 : __skb_checksum_complete(skb);
2948 }
2949
2950 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
2951 {
2952 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2953 if (skb->csum_level == 0)
2954 skb->ip_summed = CHECKSUM_NONE;
2955 else
2956 skb->csum_level--;
2957 }
2958 }
2959
2960 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
2961 {
2962 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2963 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
2964 skb->csum_level++;
2965 } else if (skb->ip_summed == CHECKSUM_NONE) {
2966 skb->ip_summed = CHECKSUM_UNNECESSARY;
2967 skb->csum_level = 0;
2968 }
2969 }
2970
2971 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
2972 {
2973 /* Mark current checksum as bad (typically called from GRO
2974 * path). In the case that ip_summed is CHECKSUM_NONE
2975 * this must be the first checksum encountered in the packet.
2976 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
2977 * checksum after the last one validated. For UDP, a zero
2978 * checksum can not be marked as bad.
2979 */
2980
2981 if (skb->ip_summed == CHECKSUM_NONE ||
2982 skb->ip_summed == CHECKSUM_UNNECESSARY)
2983 skb->csum_bad = 1;
2984 }
2985
2986 /* Check if we need to perform checksum complete validation.
2987 *
2988 * Returns true if checksum complete is needed, false otherwise
2989 * (either checksum is unnecessary or zero checksum is allowed).
2990 */
2991 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2992 bool zero_okay,
2993 __sum16 check)
2994 {
2995 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2996 skb->csum_valid = 1;
2997 __skb_decr_checksum_unnecessary(skb);
2998 return false;
2999 }
3000
3001 return true;
3002 }
3003
3004 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3005 * in checksum_init.
3006 */
3007 #define CHECKSUM_BREAK 76
3008
3009 /* Validate (init) checksum based on checksum complete.
3010 *
3011 * Return values:
3012 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3013 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3014 * checksum is stored in skb->csum for use in __skb_checksum_complete
3015 * non-zero: value of invalid checksum
3016 *
3017 */
3018 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3019 bool complete,
3020 __wsum psum)
3021 {
3022 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3023 if (!csum_fold(csum_add(psum, skb->csum))) {
3024 skb->csum_valid = 1;
3025 return 0;
3026 }
3027 } else if (skb->csum_bad) {
3028 /* ip_summed == CHECKSUM_NONE in this case */
3029 return 1;
3030 }
3031
3032 skb->csum = psum;
3033
3034 if (complete || skb->len <= CHECKSUM_BREAK) {
3035 __sum16 csum;
3036
3037 csum = __skb_checksum_complete(skb);
3038 skb->csum_valid = !csum;
3039 return csum;
3040 }
3041
3042 return 0;
3043 }
3044
3045 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3046 {
3047 return 0;
3048 }
3049
3050 /* Perform checksum validate (init). Note that this is a macro since we only
3051 * want to calculate the pseudo header which is an input function if necessary.
3052 * First we try to validate without any computation (checksum unnecessary) and
3053 * then calculate based on checksum complete calling the function to compute
3054 * pseudo header.
3055 *
3056 * Return values:
3057 * 0: checksum is validated or try to in skb_checksum_complete
3058 * non-zero: value of invalid checksum
3059 */
3060 #define __skb_checksum_validate(skb, proto, complete, \
3061 zero_okay, check, compute_pseudo) \
3062 ({ \
3063 __sum16 __ret = 0; \
3064 skb->csum_valid = 0; \
3065 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3066 __ret = __skb_checksum_validate_complete(skb, \
3067 complete, compute_pseudo(skb, proto)); \
3068 __ret; \
3069 })
3070
3071 #define skb_checksum_init(skb, proto, compute_pseudo) \
3072 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3073
3074 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3075 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3076
3077 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3078 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3079
3080 #define skb_checksum_validate_zero_check(skb, proto, check, \
3081 compute_pseudo) \
3082 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3083
3084 #define skb_checksum_simple_validate(skb) \
3085 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3086
3087 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3088 {
3089 return (skb->ip_summed == CHECKSUM_NONE &&
3090 skb->csum_valid && !skb->csum_bad);
3091 }
3092
3093 static inline void __skb_checksum_convert(struct sk_buff *skb,
3094 __sum16 check, __wsum pseudo)
3095 {
3096 skb->csum = ~pseudo;
3097 skb->ip_summed = CHECKSUM_COMPLETE;
3098 }
3099
3100 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3101 do { \
3102 if (__skb_checksum_convert_check(skb)) \
3103 __skb_checksum_convert(skb, check, \
3104 compute_pseudo(skb, proto)); \
3105 } while (0)
3106
3107 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3108 u16 start, u16 offset)
3109 {
3110 skb->ip_summed = CHECKSUM_PARTIAL;
3111 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3112 skb->csum_offset = offset - start;
3113 }
3114
3115 /* Update skbuf and packet to reflect the remote checksum offload operation.
3116 * When called, ptr indicates the starting point for skb->csum when
3117 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3118 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3119 */
3120 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3121 int start, int offset, bool nopartial)
3122 {
3123 __wsum delta;
3124
3125 if (!nopartial) {
3126 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3127 return;
3128 }
3129
3130 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3131 __skb_checksum_complete(skb);
3132 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3133 }
3134
3135 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3136
3137 /* Adjust skb->csum since we changed the packet */
3138 skb->csum = csum_add(skb->csum, delta);
3139 }
3140
3141 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3142 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3143 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3144 {
3145 if (nfct && atomic_dec_and_test(&nfct->use))
3146 nf_conntrack_destroy(nfct);
3147 }
3148 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3149 {
3150 if (nfct)
3151 atomic_inc(&nfct->use);
3152 }
3153 #endif
3154 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3155 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3156 {
3157 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3158 kfree(nf_bridge);
3159 }
3160 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3161 {
3162 if (nf_bridge)
3163 atomic_inc(&nf_bridge->use);
3164 }
3165 #endif /* CONFIG_BRIDGE_NETFILTER */
3166 static inline void nf_reset(struct sk_buff *skb)
3167 {
3168 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3169 nf_conntrack_put(skb->nfct);
3170 skb->nfct = NULL;
3171 #endif
3172 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3173 nf_bridge_put(skb->nf_bridge);
3174 skb->nf_bridge = NULL;
3175 #endif
3176 }
3177
3178 static inline void nf_reset_trace(struct sk_buff *skb)
3179 {
3180 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3181 skb->nf_trace = 0;
3182 #endif
3183 }
3184
3185 /* Note: This doesn't put any conntrack and bridge info in dst. */
3186 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3187 bool copy)
3188 {
3189 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3190 dst->nfct = src->nfct;
3191 nf_conntrack_get(src->nfct);
3192 if (copy)
3193 dst->nfctinfo = src->nfctinfo;
3194 #endif
3195 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3196 dst->nf_bridge = src->nf_bridge;
3197 nf_bridge_get(src->nf_bridge);
3198 #endif
3199 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3200 if (copy)
3201 dst->nf_trace = src->nf_trace;
3202 #endif
3203 }
3204
3205 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3206 {
3207 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3208 nf_conntrack_put(dst->nfct);
3209 #endif
3210 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3211 nf_bridge_put(dst->nf_bridge);
3212 #endif
3213 __nf_copy(dst, src, true);
3214 }
3215
3216 #ifdef CONFIG_NETWORK_SECMARK
3217 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3218 {
3219 to->secmark = from->secmark;
3220 }
3221
3222 static inline void skb_init_secmark(struct sk_buff *skb)
3223 {
3224 skb->secmark = 0;
3225 }
3226 #else
3227 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3228 { }
3229
3230 static inline void skb_init_secmark(struct sk_buff *skb)
3231 { }
3232 #endif
3233
3234 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3235 {
3236 return !skb->destructor &&
3237 #if IS_ENABLED(CONFIG_XFRM)
3238 !skb->sp &&
3239 #endif
3240 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3241 !skb->nfct &&
3242 #endif
3243 !skb->_skb_refdst &&
3244 !skb_has_frag_list(skb);
3245 }
3246
3247 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3248 {
3249 skb->queue_mapping = queue_mapping;
3250 }
3251
3252 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3253 {
3254 return skb->queue_mapping;
3255 }
3256
3257 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3258 {
3259 to->queue_mapping = from->queue_mapping;
3260 }
3261
3262 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3263 {
3264 skb->queue_mapping = rx_queue + 1;
3265 }
3266
3267 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3268 {
3269 return skb->queue_mapping - 1;
3270 }
3271
3272 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3273 {
3274 return skb->queue_mapping != 0;
3275 }
3276
3277 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
3278 unsigned int num_tx_queues);
3279
3280 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3281 {
3282 #ifdef CONFIG_XFRM
3283 return skb->sp;
3284 #else
3285 return NULL;
3286 #endif
3287 }
3288
3289 /* Keeps track of mac header offset relative to skb->head.
3290 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3291 * For non-tunnel skb it points to skb_mac_header() and for
3292 * tunnel skb it points to outer mac header.
3293 * Keeps track of level of encapsulation of network headers.
3294 */
3295 struct skb_gso_cb {
3296 int mac_offset;
3297 int encap_level;
3298 __u16 csum_start;
3299 };
3300 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3301
3302 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3303 {
3304 return (skb_mac_header(inner_skb) - inner_skb->head) -
3305 SKB_GSO_CB(inner_skb)->mac_offset;
3306 }
3307
3308 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3309 {
3310 int new_headroom, headroom;
3311 int ret;
3312
3313 headroom = skb_headroom(skb);
3314 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3315 if (ret)
3316 return ret;
3317
3318 new_headroom = skb_headroom(skb);
3319 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3320 return 0;
3321 }
3322
3323 /* Compute the checksum for a gso segment. First compute the checksum value
3324 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3325 * then add in skb->csum (checksum from csum_start to end of packet).
3326 * skb->csum and csum_start are then updated to reflect the checksum of the
3327 * resultant packet starting from the transport header-- the resultant checksum
3328 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3329 * header.
3330 */
3331 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3332 {
3333 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3334 skb_transport_offset(skb);
3335 __u16 csum;
3336
3337 csum = csum_fold(csum_partial(skb_transport_header(skb),
3338 plen, skb->csum));
3339 skb->csum = res;
3340 SKB_GSO_CB(skb)->csum_start -= plen;
3341
3342 return csum;
3343 }
3344
3345 static inline bool skb_is_gso(const struct sk_buff *skb)
3346 {
3347 return skb_shinfo(skb)->gso_size;
3348 }
3349
3350 /* Note: Should be called only if skb_is_gso(skb) is true */
3351 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3352 {
3353 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3354 }
3355
3356 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3357
3358 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3359 {
3360 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3361 * wanted then gso_type will be set. */
3362 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3363
3364 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3365 unlikely(shinfo->gso_type == 0)) {
3366 __skb_warn_lro_forwarding(skb);
3367 return true;
3368 }
3369 return false;
3370 }
3371
3372 static inline void skb_forward_csum(struct sk_buff *skb)
3373 {
3374 /* Unfortunately we don't support this one. Any brave souls? */
3375 if (skb->ip_summed == CHECKSUM_COMPLETE)
3376 skb->ip_summed = CHECKSUM_NONE;
3377 }
3378
3379 /**
3380 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3381 * @skb: skb to check
3382 *
3383 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3384 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3385 * use this helper, to document places where we make this assertion.
3386 */
3387 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3388 {
3389 #ifdef DEBUG
3390 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3391 #endif
3392 }
3393
3394 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3395
3396 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3397
3398 u32 skb_get_poff(const struct sk_buff *skb);
3399 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
3400 const struct flow_keys *keys, int hlen);
3401
3402 /**
3403 * skb_head_is_locked - Determine if the skb->head is locked down
3404 * @skb: skb to check
3405 *
3406 * The head on skbs build around a head frag can be removed if they are
3407 * not cloned. This function returns true if the skb head is locked down
3408 * due to either being allocated via kmalloc, or by being a clone with
3409 * multiple references to the head.
3410 */
3411 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3412 {
3413 return !skb->head_frag || skb_cloned(skb);
3414 }
3415
3416 /**
3417 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3418 *
3419 * @skb: GSO skb
3420 *
3421 * skb_gso_network_seglen is used to determine the real size of the
3422 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3423 *
3424 * The MAC/L2 header is not accounted for.
3425 */
3426 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3427 {
3428 unsigned int hdr_len = skb_transport_header(skb) -
3429 skb_network_header(skb);
3430 return hdr_len + skb_gso_transport_seglen(skb);
3431 }
3432 #endif /* __KERNEL__ */
3433 #endif /* _LINUX_SKBUFF_H */
This page took 0.10803 seconds and 6 git commands to generate.