]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - include/linux/skbuff.h
net: move __skb_tx_hash to dev.c
[mirror_ubuntu-bionic-kernel.git] / include / linux / skbuff.h
1 /*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
16
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
25
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <net/flow_dissector.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 enum {
170 BRNF_PROTO_UNCHANGED,
171 BRNF_PROTO_8021Q,
172 BRNF_PROTO_PPPOE
173 } orig_proto;
174 bool pkt_otherhost;
175 unsigned int mask;
176 struct net_device *physindev;
177 struct net_device *physoutdev;
178 char neigh_header[8];
179 };
180 #endif
181
182 struct sk_buff_head {
183 /* These two members must be first. */
184 struct sk_buff *next;
185 struct sk_buff *prev;
186
187 __u32 qlen;
188 spinlock_t lock;
189 };
190
191 struct sk_buff;
192
193 /* To allow 64K frame to be packed as single skb without frag_list we
194 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
195 * buffers which do not start on a page boundary.
196 *
197 * Since GRO uses frags we allocate at least 16 regardless of page
198 * size.
199 */
200 #if (65536/PAGE_SIZE + 1) < 16
201 #define MAX_SKB_FRAGS 16UL
202 #else
203 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
204 #endif
205
206 typedef struct skb_frag_struct skb_frag_t;
207
208 struct skb_frag_struct {
209 struct {
210 struct page *p;
211 } page;
212 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
213 __u32 page_offset;
214 __u32 size;
215 #else
216 __u16 page_offset;
217 __u16 size;
218 #endif
219 };
220
221 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
222 {
223 return frag->size;
224 }
225
226 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
227 {
228 frag->size = size;
229 }
230
231 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
232 {
233 frag->size += delta;
234 }
235
236 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
237 {
238 frag->size -= delta;
239 }
240
241 #define HAVE_HW_TIME_STAMP
242
243 /**
244 * struct skb_shared_hwtstamps - hardware time stamps
245 * @hwtstamp: hardware time stamp transformed into duration
246 * since arbitrary point in time
247 *
248 * Software time stamps generated by ktime_get_real() are stored in
249 * skb->tstamp.
250 *
251 * hwtstamps can only be compared against other hwtstamps from
252 * the same device.
253 *
254 * This structure is attached to packets as part of the
255 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
256 */
257 struct skb_shared_hwtstamps {
258 ktime_t hwtstamp;
259 };
260
261 /* Definitions for tx_flags in struct skb_shared_info */
262 enum {
263 /* generate hardware time stamp */
264 SKBTX_HW_TSTAMP = 1 << 0,
265
266 /* generate software time stamp when queueing packet to NIC */
267 SKBTX_SW_TSTAMP = 1 << 1,
268
269 /* device driver is going to provide hardware time stamp */
270 SKBTX_IN_PROGRESS = 1 << 2,
271
272 /* device driver supports TX zero-copy buffers */
273 SKBTX_DEV_ZEROCOPY = 1 << 3,
274
275 /* generate wifi status information (where possible) */
276 SKBTX_WIFI_STATUS = 1 << 4,
277
278 /* This indicates at least one fragment might be overwritten
279 * (as in vmsplice(), sendfile() ...)
280 * If we need to compute a TX checksum, we'll need to copy
281 * all frags to avoid possible bad checksum
282 */
283 SKBTX_SHARED_FRAG = 1 << 5,
284
285 /* generate software time stamp when entering packet scheduling */
286 SKBTX_SCHED_TSTAMP = 1 << 6,
287
288 /* generate software timestamp on peer data acknowledgment */
289 SKBTX_ACK_TSTAMP = 1 << 7,
290 };
291
292 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
293 SKBTX_SCHED_TSTAMP | \
294 SKBTX_ACK_TSTAMP)
295 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
296
297 /*
298 * The callback notifies userspace to release buffers when skb DMA is done in
299 * lower device, the skb last reference should be 0 when calling this.
300 * The zerocopy_success argument is true if zero copy transmit occurred,
301 * false on data copy or out of memory error caused by data copy attempt.
302 * The ctx field is used to track device context.
303 * The desc field is used to track userspace buffer index.
304 */
305 struct ubuf_info {
306 void (*callback)(struct ubuf_info *, bool zerocopy_success);
307 void *ctx;
308 unsigned long desc;
309 };
310
311 /* This data is invariant across clones and lives at
312 * the end of the header data, ie. at skb->end.
313 */
314 struct skb_shared_info {
315 unsigned char nr_frags;
316 __u8 tx_flags;
317 unsigned short gso_size;
318 /* Warning: this field is not always filled in (UFO)! */
319 unsigned short gso_segs;
320 unsigned short gso_type;
321 struct sk_buff *frag_list;
322 struct skb_shared_hwtstamps hwtstamps;
323 u32 tskey;
324 __be32 ip6_frag_id;
325
326 /*
327 * Warning : all fields before dataref are cleared in __alloc_skb()
328 */
329 atomic_t dataref;
330
331 /* Intermediate layers must ensure that destructor_arg
332 * remains valid until skb destructor */
333 void * destructor_arg;
334
335 /* must be last field, see pskb_expand_head() */
336 skb_frag_t frags[MAX_SKB_FRAGS];
337 };
338
339 /* We divide dataref into two halves. The higher 16 bits hold references
340 * to the payload part of skb->data. The lower 16 bits hold references to
341 * the entire skb->data. A clone of a headerless skb holds the length of
342 * the header in skb->hdr_len.
343 *
344 * All users must obey the rule that the skb->data reference count must be
345 * greater than or equal to the payload reference count.
346 *
347 * Holding a reference to the payload part means that the user does not
348 * care about modifications to the header part of skb->data.
349 */
350 #define SKB_DATAREF_SHIFT 16
351 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
352
353
354 enum {
355 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
356 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
357 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
358 };
359
360 enum {
361 SKB_GSO_TCPV4 = 1 << 0,
362 SKB_GSO_UDP = 1 << 1,
363
364 /* This indicates the skb is from an untrusted source. */
365 SKB_GSO_DODGY = 1 << 2,
366
367 /* This indicates the tcp segment has CWR set. */
368 SKB_GSO_TCP_ECN = 1 << 3,
369
370 SKB_GSO_TCPV6 = 1 << 4,
371
372 SKB_GSO_FCOE = 1 << 5,
373
374 SKB_GSO_GRE = 1 << 6,
375
376 SKB_GSO_GRE_CSUM = 1 << 7,
377
378 SKB_GSO_IPIP = 1 << 8,
379
380 SKB_GSO_SIT = 1 << 9,
381
382 SKB_GSO_UDP_TUNNEL = 1 << 10,
383
384 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
385
386 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
387 };
388
389 #if BITS_PER_LONG > 32
390 #define NET_SKBUFF_DATA_USES_OFFSET 1
391 #endif
392
393 #ifdef NET_SKBUFF_DATA_USES_OFFSET
394 typedef unsigned int sk_buff_data_t;
395 #else
396 typedef unsigned char *sk_buff_data_t;
397 #endif
398
399 /**
400 * struct skb_mstamp - multi resolution time stamps
401 * @stamp_us: timestamp in us resolution
402 * @stamp_jiffies: timestamp in jiffies
403 */
404 struct skb_mstamp {
405 union {
406 u64 v64;
407 struct {
408 u32 stamp_us;
409 u32 stamp_jiffies;
410 };
411 };
412 };
413
414 /**
415 * skb_mstamp_get - get current timestamp
416 * @cl: place to store timestamps
417 */
418 static inline void skb_mstamp_get(struct skb_mstamp *cl)
419 {
420 u64 val = local_clock();
421
422 do_div(val, NSEC_PER_USEC);
423 cl->stamp_us = (u32)val;
424 cl->stamp_jiffies = (u32)jiffies;
425 }
426
427 /**
428 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
429 * @t1: pointer to newest sample
430 * @t0: pointer to oldest sample
431 */
432 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
433 const struct skb_mstamp *t0)
434 {
435 s32 delta_us = t1->stamp_us - t0->stamp_us;
436 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
437
438 /* If delta_us is negative, this might be because interval is too big,
439 * or local_clock() drift is too big : fallback using jiffies.
440 */
441 if (delta_us <= 0 ||
442 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
443
444 delta_us = jiffies_to_usecs(delta_jiffies);
445
446 return delta_us;
447 }
448
449
450 /**
451 * struct sk_buff - socket buffer
452 * @next: Next buffer in list
453 * @prev: Previous buffer in list
454 * @tstamp: Time we arrived/left
455 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
456 * @sk: Socket we are owned by
457 * @dev: Device we arrived on/are leaving by
458 * @cb: Control buffer. Free for use by every layer. Put private vars here
459 * @_skb_refdst: destination entry (with norefcount bit)
460 * @sp: the security path, used for xfrm
461 * @len: Length of actual data
462 * @data_len: Data length
463 * @mac_len: Length of link layer header
464 * @hdr_len: writable header length of cloned skb
465 * @csum: Checksum (must include start/offset pair)
466 * @csum_start: Offset from skb->head where checksumming should start
467 * @csum_offset: Offset from csum_start where checksum should be stored
468 * @priority: Packet queueing priority
469 * @ignore_df: allow local fragmentation
470 * @cloned: Head may be cloned (check refcnt to be sure)
471 * @ip_summed: Driver fed us an IP checksum
472 * @nohdr: Payload reference only, must not modify header
473 * @nfctinfo: Relationship of this skb to the connection
474 * @pkt_type: Packet class
475 * @fclone: skbuff clone status
476 * @ipvs_property: skbuff is owned by ipvs
477 * @peeked: this packet has been seen already, so stats have been
478 * done for it, don't do them again
479 * @nf_trace: netfilter packet trace flag
480 * @protocol: Packet protocol from driver
481 * @destructor: Destruct function
482 * @nfct: Associated connection, if any
483 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
484 * @skb_iif: ifindex of device we arrived on
485 * @tc_index: Traffic control index
486 * @tc_verd: traffic control verdict
487 * @hash: the packet hash
488 * @queue_mapping: Queue mapping for multiqueue devices
489 * @xmit_more: More SKBs are pending for this queue
490 * @ndisc_nodetype: router type (from link layer)
491 * @ooo_okay: allow the mapping of a socket to a queue to be changed
492 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
493 * ports.
494 * @sw_hash: indicates hash was computed in software stack
495 * @wifi_acked_valid: wifi_acked was set
496 * @wifi_acked: whether frame was acked on wifi or not
497 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
498 * @napi_id: id of the NAPI struct this skb came from
499 * @secmark: security marking
500 * @mark: Generic packet mark
501 * @vlan_proto: vlan encapsulation protocol
502 * @vlan_tci: vlan tag control information
503 * @inner_protocol: Protocol (encapsulation)
504 * @inner_transport_header: Inner transport layer header (encapsulation)
505 * @inner_network_header: Network layer header (encapsulation)
506 * @inner_mac_header: Link layer header (encapsulation)
507 * @transport_header: Transport layer header
508 * @network_header: Network layer header
509 * @mac_header: Link layer header
510 * @tail: Tail pointer
511 * @end: End pointer
512 * @head: Head of buffer
513 * @data: Data head pointer
514 * @truesize: Buffer size
515 * @users: User count - see {datagram,tcp}.c
516 */
517
518 struct sk_buff {
519 union {
520 struct {
521 /* These two members must be first. */
522 struct sk_buff *next;
523 struct sk_buff *prev;
524
525 union {
526 ktime_t tstamp;
527 struct skb_mstamp skb_mstamp;
528 };
529 };
530 struct rb_node rbnode; /* used in netem & tcp stack */
531 };
532 struct sock *sk;
533 struct net_device *dev;
534
535 /*
536 * This is the control buffer. It is free to use for every
537 * layer. Please put your private variables there. If you
538 * want to keep them across layers you have to do a skb_clone()
539 * first. This is owned by whoever has the skb queued ATM.
540 */
541 char cb[48] __aligned(8);
542
543 unsigned long _skb_refdst;
544 void (*destructor)(struct sk_buff *skb);
545 #ifdef CONFIG_XFRM
546 struct sec_path *sp;
547 #endif
548 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
549 struct nf_conntrack *nfct;
550 #endif
551 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
552 struct nf_bridge_info *nf_bridge;
553 #endif
554 unsigned int len,
555 data_len;
556 __u16 mac_len,
557 hdr_len;
558
559 /* Following fields are _not_ copied in __copy_skb_header()
560 * Note that queue_mapping is here mostly to fill a hole.
561 */
562 kmemcheck_bitfield_begin(flags1);
563 __u16 queue_mapping;
564 __u8 cloned:1,
565 nohdr:1,
566 fclone:2,
567 peeked:1,
568 head_frag:1,
569 xmit_more:1;
570 /* one bit hole */
571 kmemcheck_bitfield_end(flags1);
572
573 /* fields enclosed in headers_start/headers_end are copied
574 * using a single memcpy() in __copy_skb_header()
575 */
576 /* private: */
577 __u32 headers_start[0];
578 /* public: */
579
580 /* if you move pkt_type around you also must adapt those constants */
581 #ifdef __BIG_ENDIAN_BITFIELD
582 #define PKT_TYPE_MAX (7 << 5)
583 #else
584 #define PKT_TYPE_MAX 7
585 #endif
586 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
587
588 __u8 __pkt_type_offset[0];
589 __u8 pkt_type:3;
590 __u8 pfmemalloc:1;
591 __u8 ignore_df:1;
592 __u8 nfctinfo:3;
593
594 __u8 nf_trace:1;
595 __u8 ip_summed:2;
596 __u8 ooo_okay:1;
597 __u8 l4_hash:1;
598 __u8 sw_hash:1;
599 __u8 wifi_acked_valid:1;
600 __u8 wifi_acked:1;
601
602 __u8 no_fcs:1;
603 /* Indicates the inner headers are valid in the skbuff. */
604 __u8 encapsulation:1;
605 __u8 encap_hdr_csum:1;
606 __u8 csum_valid:1;
607 __u8 csum_complete_sw:1;
608 __u8 csum_level:2;
609 __u8 csum_bad:1;
610
611 #ifdef CONFIG_IPV6_NDISC_NODETYPE
612 __u8 ndisc_nodetype:2;
613 #endif
614 __u8 ipvs_property:1;
615 __u8 inner_protocol_type:1;
616 __u8 remcsum_offload:1;
617 /* 3 or 5 bit hole */
618
619 #ifdef CONFIG_NET_SCHED
620 __u16 tc_index; /* traffic control index */
621 #ifdef CONFIG_NET_CLS_ACT
622 __u16 tc_verd; /* traffic control verdict */
623 #endif
624 #endif
625
626 union {
627 __wsum csum;
628 struct {
629 __u16 csum_start;
630 __u16 csum_offset;
631 };
632 };
633 __u32 priority;
634 int skb_iif;
635 __u32 hash;
636 __be16 vlan_proto;
637 __u16 vlan_tci;
638 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
639 union {
640 unsigned int napi_id;
641 unsigned int sender_cpu;
642 };
643 #endif
644 #ifdef CONFIG_NETWORK_SECMARK
645 __u32 secmark;
646 #endif
647 union {
648 __u32 mark;
649 __u32 reserved_tailroom;
650 };
651
652 union {
653 __be16 inner_protocol;
654 __u8 inner_ipproto;
655 };
656
657 __u16 inner_transport_header;
658 __u16 inner_network_header;
659 __u16 inner_mac_header;
660
661 __be16 protocol;
662 __u16 transport_header;
663 __u16 network_header;
664 __u16 mac_header;
665
666 /* private: */
667 __u32 headers_end[0];
668 /* public: */
669
670 /* These elements must be at the end, see alloc_skb() for details. */
671 sk_buff_data_t tail;
672 sk_buff_data_t end;
673 unsigned char *head,
674 *data;
675 unsigned int truesize;
676 atomic_t users;
677 };
678
679 #ifdef __KERNEL__
680 /*
681 * Handling routines are only of interest to the kernel
682 */
683 #include <linux/slab.h>
684
685
686 #define SKB_ALLOC_FCLONE 0x01
687 #define SKB_ALLOC_RX 0x02
688 #define SKB_ALLOC_NAPI 0x04
689
690 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
691 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
692 {
693 return unlikely(skb->pfmemalloc);
694 }
695
696 /*
697 * skb might have a dst pointer attached, refcounted or not.
698 * _skb_refdst low order bit is set if refcount was _not_ taken
699 */
700 #define SKB_DST_NOREF 1UL
701 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
702
703 /**
704 * skb_dst - returns skb dst_entry
705 * @skb: buffer
706 *
707 * Returns skb dst_entry, regardless of reference taken or not.
708 */
709 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
710 {
711 /* If refdst was not refcounted, check we still are in a
712 * rcu_read_lock section
713 */
714 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
715 !rcu_read_lock_held() &&
716 !rcu_read_lock_bh_held());
717 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
718 }
719
720 /**
721 * skb_dst_set - sets skb dst
722 * @skb: buffer
723 * @dst: dst entry
724 *
725 * Sets skb dst, assuming a reference was taken on dst and should
726 * be released by skb_dst_drop()
727 */
728 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
729 {
730 skb->_skb_refdst = (unsigned long)dst;
731 }
732
733 /**
734 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
735 * @skb: buffer
736 * @dst: dst entry
737 *
738 * Sets skb dst, assuming a reference was not taken on dst.
739 * If dst entry is cached, we do not take reference and dst_release
740 * will be avoided by refdst_drop. If dst entry is not cached, we take
741 * reference, so that last dst_release can destroy the dst immediately.
742 */
743 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
744 {
745 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
746 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
747 }
748
749 /**
750 * skb_dst_is_noref - Test if skb dst isn't refcounted
751 * @skb: buffer
752 */
753 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
754 {
755 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
756 }
757
758 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
759 {
760 return (struct rtable *)skb_dst(skb);
761 }
762
763 void kfree_skb(struct sk_buff *skb);
764 void kfree_skb_list(struct sk_buff *segs);
765 void skb_tx_error(struct sk_buff *skb);
766 void consume_skb(struct sk_buff *skb);
767 void __kfree_skb(struct sk_buff *skb);
768 extern struct kmem_cache *skbuff_head_cache;
769
770 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
771 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
772 bool *fragstolen, int *delta_truesize);
773
774 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
775 int node);
776 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
777 struct sk_buff *build_skb(void *data, unsigned int frag_size);
778 static inline struct sk_buff *alloc_skb(unsigned int size,
779 gfp_t priority)
780 {
781 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
782 }
783
784 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
785 unsigned long data_len,
786 int max_page_order,
787 int *errcode,
788 gfp_t gfp_mask);
789
790 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
791 struct sk_buff_fclones {
792 struct sk_buff skb1;
793
794 struct sk_buff skb2;
795
796 atomic_t fclone_ref;
797 };
798
799 /**
800 * skb_fclone_busy - check if fclone is busy
801 * @skb: buffer
802 *
803 * Returns true is skb is a fast clone, and its clone is not freed.
804 * Some drivers call skb_orphan() in their ndo_start_xmit(),
805 * so we also check that this didnt happen.
806 */
807 static inline bool skb_fclone_busy(const struct sock *sk,
808 const struct sk_buff *skb)
809 {
810 const struct sk_buff_fclones *fclones;
811
812 fclones = container_of(skb, struct sk_buff_fclones, skb1);
813
814 return skb->fclone == SKB_FCLONE_ORIG &&
815 atomic_read(&fclones->fclone_ref) > 1 &&
816 fclones->skb2.sk == sk;
817 }
818
819 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
820 gfp_t priority)
821 {
822 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
823 }
824
825 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
826 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
827 {
828 return __alloc_skb_head(priority, -1);
829 }
830
831 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
832 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
833 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
834 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
835 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
836 gfp_t gfp_mask, bool fclone);
837 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
838 gfp_t gfp_mask)
839 {
840 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
841 }
842
843 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
844 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
845 unsigned int headroom);
846 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
847 int newtailroom, gfp_t priority);
848 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
849 int offset, int len);
850 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
851 int len);
852 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
853 int skb_pad(struct sk_buff *skb, int pad);
854 #define dev_kfree_skb(a) consume_skb(a)
855
856 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
857 int getfrag(void *from, char *to, int offset,
858 int len, int odd, struct sk_buff *skb),
859 void *from, int length);
860
861 struct skb_seq_state {
862 __u32 lower_offset;
863 __u32 upper_offset;
864 __u32 frag_idx;
865 __u32 stepped_offset;
866 struct sk_buff *root_skb;
867 struct sk_buff *cur_skb;
868 __u8 *frag_data;
869 };
870
871 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
872 unsigned int to, struct skb_seq_state *st);
873 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
874 struct skb_seq_state *st);
875 void skb_abort_seq_read(struct skb_seq_state *st);
876
877 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
878 unsigned int to, struct ts_config *config);
879
880 /*
881 * Packet hash types specify the type of hash in skb_set_hash.
882 *
883 * Hash types refer to the protocol layer addresses which are used to
884 * construct a packet's hash. The hashes are used to differentiate or identify
885 * flows of the protocol layer for the hash type. Hash types are either
886 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
887 *
888 * Properties of hashes:
889 *
890 * 1) Two packets in different flows have different hash values
891 * 2) Two packets in the same flow should have the same hash value
892 *
893 * A hash at a higher layer is considered to be more specific. A driver should
894 * set the most specific hash possible.
895 *
896 * A driver cannot indicate a more specific hash than the layer at which a hash
897 * was computed. For instance an L3 hash cannot be set as an L4 hash.
898 *
899 * A driver may indicate a hash level which is less specific than the
900 * actual layer the hash was computed on. For instance, a hash computed
901 * at L4 may be considered an L3 hash. This should only be done if the
902 * driver can't unambiguously determine that the HW computed the hash at
903 * the higher layer. Note that the "should" in the second property above
904 * permits this.
905 */
906 enum pkt_hash_types {
907 PKT_HASH_TYPE_NONE, /* Undefined type */
908 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
909 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
910 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
911 };
912
913 static inline void
914 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
915 {
916 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
917 skb->sw_hash = 0;
918 skb->hash = hash;
919 }
920
921 static inline __u32 skb_get_hash(struct sk_buff *skb)
922 {
923 if (!skb->l4_hash && !skb->sw_hash)
924 __skb_get_hash(skb);
925
926 return skb->hash;
927 }
928
929 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
930
931 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
932 {
933 return skb->hash;
934 }
935
936 static inline void skb_clear_hash(struct sk_buff *skb)
937 {
938 skb->hash = 0;
939 skb->sw_hash = 0;
940 skb->l4_hash = 0;
941 }
942
943 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
944 {
945 if (!skb->l4_hash)
946 skb_clear_hash(skb);
947 }
948
949 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
950 {
951 to->hash = from->hash;
952 to->sw_hash = from->sw_hash;
953 to->l4_hash = from->l4_hash;
954 };
955
956 static inline void skb_sender_cpu_clear(struct sk_buff *skb)
957 {
958 #ifdef CONFIG_XPS
959 skb->sender_cpu = 0;
960 #endif
961 }
962
963 #ifdef NET_SKBUFF_DATA_USES_OFFSET
964 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
965 {
966 return skb->head + skb->end;
967 }
968
969 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
970 {
971 return skb->end;
972 }
973 #else
974 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
975 {
976 return skb->end;
977 }
978
979 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
980 {
981 return skb->end - skb->head;
982 }
983 #endif
984
985 /* Internal */
986 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
987
988 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
989 {
990 return &skb_shinfo(skb)->hwtstamps;
991 }
992
993 /**
994 * skb_queue_empty - check if a queue is empty
995 * @list: queue head
996 *
997 * Returns true if the queue is empty, false otherwise.
998 */
999 static inline int skb_queue_empty(const struct sk_buff_head *list)
1000 {
1001 return list->next == (const struct sk_buff *) list;
1002 }
1003
1004 /**
1005 * skb_queue_is_last - check if skb is the last entry in the queue
1006 * @list: queue head
1007 * @skb: buffer
1008 *
1009 * Returns true if @skb is the last buffer on the list.
1010 */
1011 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1012 const struct sk_buff *skb)
1013 {
1014 return skb->next == (const struct sk_buff *) list;
1015 }
1016
1017 /**
1018 * skb_queue_is_first - check if skb is the first entry in the queue
1019 * @list: queue head
1020 * @skb: buffer
1021 *
1022 * Returns true if @skb is the first buffer on the list.
1023 */
1024 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1025 const struct sk_buff *skb)
1026 {
1027 return skb->prev == (const struct sk_buff *) list;
1028 }
1029
1030 /**
1031 * skb_queue_next - return the next packet in the queue
1032 * @list: queue head
1033 * @skb: current buffer
1034 *
1035 * Return the next packet in @list after @skb. It is only valid to
1036 * call this if skb_queue_is_last() evaluates to false.
1037 */
1038 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1039 const struct sk_buff *skb)
1040 {
1041 /* This BUG_ON may seem severe, but if we just return then we
1042 * are going to dereference garbage.
1043 */
1044 BUG_ON(skb_queue_is_last(list, skb));
1045 return skb->next;
1046 }
1047
1048 /**
1049 * skb_queue_prev - return the prev packet in the queue
1050 * @list: queue head
1051 * @skb: current buffer
1052 *
1053 * Return the prev packet in @list before @skb. It is only valid to
1054 * call this if skb_queue_is_first() evaluates to false.
1055 */
1056 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1057 const struct sk_buff *skb)
1058 {
1059 /* This BUG_ON may seem severe, but if we just return then we
1060 * are going to dereference garbage.
1061 */
1062 BUG_ON(skb_queue_is_first(list, skb));
1063 return skb->prev;
1064 }
1065
1066 /**
1067 * skb_get - reference buffer
1068 * @skb: buffer to reference
1069 *
1070 * Makes another reference to a socket buffer and returns a pointer
1071 * to the buffer.
1072 */
1073 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1074 {
1075 atomic_inc(&skb->users);
1076 return skb;
1077 }
1078
1079 /*
1080 * If users == 1, we are the only owner and are can avoid redundant
1081 * atomic change.
1082 */
1083
1084 /**
1085 * skb_cloned - is the buffer a clone
1086 * @skb: buffer to check
1087 *
1088 * Returns true if the buffer was generated with skb_clone() and is
1089 * one of multiple shared copies of the buffer. Cloned buffers are
1090 * shared data so must not be written to under normal circumstances.
1091 */
1092 static inline int skb_cloned(const struct sk_buff *skb)
1093 {
1094 return skb->cloned &&
1095 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1096 }
1097
1098 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1099 {
1100 might_sleep_if(pri & __GFP_WAIT);
1101
1102 if (skb_cloned(skb))
1103 return pskb_expand_head(skb, 0, 0, pri);
1104
1105 return 0;
1106 }
1107
1108 /**
1109 * skb_header_cloned - is the header a clone
1110 * @skb: buffer to check
1111 *
1112 * Returns true if modifying the header part of the buffer requires
1113 * the data to be copied.
1114 */
1115 static inline int skb_header_cloned(const struct sk_buff *skb)
1116 {
1117 int dataref;
1118
1119 if (!skb->cloned)
1120 return 0;
1121
1122 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1123 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1124 return dataref != 1;
1125 }
1126
1127 /**
1128 * skb_header_release - release reference to header
1129 * @skb: buffer to operate on
1130 *
1131 * Drop a reference to the header part of the buffer. This is done
1132 * by acquiring a payload reference. You must not read from the header
1133 * part of skb->data after this.
1134 * Note : Check if you can use __skb_header_release() instead.
1135 */
1136 static inline void skb_header_release(struct sk_buff *skb)
1137 {
1138 BUG_ON(skb->nohdr);
1139 skb->nohdr = 1;
1140 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1141 }
1142
1143 /**
1144 * __skb_header_release - release reference to header
1145 * @skb: buffer to operate on
1146 *
1147 * Variant of skb_header_release() assuming skb is private to caller.
1148 * We can avoid one atomic operation.
1149 */
1150 static inline void __skb_header_release(struct sk_buff *skb)
1151 {
1152 skb->nohdr = 1;
1153 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1154 }
1155
1156
1157 /**
1158 * skb_shared - is the buffer shared
1159 * @skb: buffer to check
1160 *
1161 * Returns true if more than one person has a reference to this
1162 * buffer.
1163 */
1164 static inline int skb_shared(const struct sk_buff *skb)
1165 {
1166 return atomic_read(&skb->users) != 1;
1167 }
1168
1169 /**
1170 * skb_share_check - check if buffer is shared and if so clone it
1171 * @skb: buffer to check
1172 * @pri: priority for memory allocation
1173 *
1174 * If the buffer is shared the buffer is cloned and the old copy
1175 * drops a reference. A new clone with a single reference is returned.
1176 * If the buffer is not shared the original buffer is returned. When
1177 * being called from interrupt status or with spinlocks held pri must
1178 * be GFP_ATOMIC.
1179 *
1180 * NULL is returned on a memory allocation failure.
1181 */
1182 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1183 {
1184 might_sleep_if(pri & __GFP_WAIT);
1185 if (skb_shared(skb)) {
1186 struct sk_buff *nskb = skb_clone(skb, pri);
1187
1188 if (likely(nskb))
1189 consume_skb(skb);
1190 else
1191 kfree_skb(skb);
1192 skb = nskb;
1193 }
1194 return skb;
1195 }
1196
1197 /*
1198 * Copy shared buffers into a new sk_buff. We effectively do COW on
1199 * packets to handle cases where we have a local reader and forward
1200 * and a couple of other messy ones. The normal one is tcpdumping
1201 * a packet thats being forwarded.
1202 */
1203
1204 /**
1205 * skb_unshare - make a copy of a shared buffer
1206 * @skb: buffer to check
1207 * @pri: priority for memory allocation
1208 *
1209 * If the socket buffer is a clone then this function creates a new
1210 * copy of the data, drops a reference count on the old copy and returns
1211 * the new copy with the reference count at 1. If the buffer is not a clone
1212 * the original buffer is returned. When called with a spinlock held or
1213 * from interrupt state @pri must be %GFP_ATOMIC
1214 *
1215 * %NULL is returned on a memory allocation failure.
1216 */
1217 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1218 gfp_t pri)
1219 {
1220 might_sleep_if(pri & __GFP_WAIT);
1221 if (skb_cloned(skb)) {
1222 struct sk_buff *nskb = skb_copy(skb, pri);
1223
1224 /* Free our shared copy */
1225 if (likely(nskb))
1226 consume_skb(skb);
1227 else
1228 kfree_skb(skb);
1229 skb = nskb;
1230 }
1231 return skb;
1232 }
1233
1234 /**
1235 * skb_peek - peek at the head of an &sk_buff_head
1236 * @list_: list to peek at
1237 *
1238 * Peek an &sk_buff. Unlike most other operations you _MUST_
1239 * be careful with this one. A peek leaves the buffer on the
1240 * list and someone else may run off with it. You must hold
1241 * the appropriate locks or have a private queue to do this.
1242 *
1243 * Returns %NULL for an empty list or a pointer to the head element.
1244 * The reference count is not incremented and the reference is therefore
1245 * volatile. Use with caution.
1246 */
1247 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1248 {
1249 struct sk_buff *skb = list_->next;
1250
1251 if (skb == (struct sk_buff *)list_)
1252 skb = NULL;
1253 return skb;
1254 }
1255
1256 /**
1257 * skb_peek_next - peek skb following the given one from a queue
1258 * @skb: skb to start from
1259 * @list_: list to peek at
1260 *
1261 * Returns %NULL when the end of the list is met or a pointer to the
1262 * next element. The reference count is not incremented and the
1263 * reference is therefore volatile. Use with caution.
1264 */
1265 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1266 const struct sk_buff_head *list_)
1267 {
1268 struct sk_buff *next = skb->next;
1269
1270 if (next == (struct sk_buff *)list_)
1271 next = NULL;
1272 return next;
1273 }
1274
1275 /**
1276 * skb_peek_tail - peek at the tail of an &sk_buff_head
1277 * @list_: list to peek at
1278 *
1279 * Peek an &sk_buff. Unlike most other operations you _MUST_
1280 * be careful with this one. A peek leaves the buffer on the
1281 * list and someone else may run off with it. You must hold
1282 * the appropriate locks or have a private queue to do this.
1283 *
1284 * Returns %NULL for an empty list or a pointer to the tail element.
1285 * The reference count is not incremented and the reference is therefore
1286 * volatile. Use with caution.
1287 */
1288 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1289 {
1290 struct sk_buff *skb = list_->prev;
1291
1292 if (skb == (struct sk_buff *)list_)
1293 skb = NULL;
1294 return skb;
1295
1296 }
1297
1298 /**
1299 * skb_queue_len - get queue length
1300 * @list_: list to measure
1301 *
1302 * Return the length of an &sk_buff queue.
1303 */
1304 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1305 {
1306 return list_->qlen;
1307 }
1308
1309 /**
1310 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1311 * @list: queue to initialize
1312 *
1313 * This initializes only the list and queue length aspects of
1314 * an sk_buff_head object. This allows to initialize the list
1315 * aspects of an sk_buff_head without reinitializing things like
1316 * the spinlock. It can also be used for on-stack sk_buff_head
1317 * objects where the spinlock is known to not be used.
1318 */
1319 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1320 {
1321 list->prev = list->next = (struct sk_buff *)list;
1322 list->qlen = 0;
1323 }
1324
1325 /*
1326 * This function creates a split out lock class for each invocation;
1327 * this is needed for now since a whole lot of users of the skb-queue
1328 * infrastructure in drivers have different locking usage (in hardirq)
1329 * than the networking core (in softirq only). In the long run either the
1330 * network layer or drivers should need annotation to consolidate the
1331 * main types of usage into 3 classes.
1332 */
1333 static inline void skb_queue_head_init(struct sk_buff_head *list)
1334 {
1335 spin_lock_init(&list->lock);
1336 __skb_queue_head_init(list);
1337 }
1338
1339 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1340 struct lock_class_key *class)
1341 {
1342 skb_queue_head_init(list);
1343 lockdep_set_class(&list->lock, class);
1344 }
1345
1346 /*
1347 * Insert an sk_buff on a list.
1348 *
1349 * The "__skb_xxxx()" functions are the non-atomic ones that
1350 * can only be called with interrupts disabled.
1351 */
1352 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1353 struct sk_buff_head *list);
1354 static inline void __skb_insert(struct sk_buff *newsk,
1355 struct sk_buff *prev, struct sk_buff *next,
1356 struct sk_buff_head *list)
1357 {
1358 newsk->next = next;
1359 newsk->prev = prev;
1360 next->prev = prev->next = newsk;
1361 list->qlen++;
1362 }
1363
1364 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1365 struct sk_buff *prev,
1366 struct sk_buff *next)
1367 {
1368 struct sk_buff *first = list->next;
1369 struct sk_buff *last = list->prev;
1370
1371 first->prev = prev;
1372 prev->next = first;
1373
1374 last->next = next;
1375 next->prev = last;
1376 }
1377
1378 /**
1379 * skb_queue_splice - join two skb lists, this is designed for stacks
1380 * @list: the new list to add
1381 * @head: the place to add it in the first list
1382 */
1383 static inline void skb_queue_splice(const struct sk_buff_head *list,
1384 struct sk_buff_head *head)
1385 {
1386 if (!skb_queue_empty(list)) {
1387 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1388 head->qlen += list->qlen;
1389 }
1390 }
1391
1392 /**
1393 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1394 * @list: the new list to add
1395 * @head: the place to add it in the first list
1396 *
1397 * The list at @list is reinitialised
1398 */
1399 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1400 struct sk_buff_head *head)
1401 {
1402 if (!skb_queue_empty(list)) {
1403 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1404 head->qlen += list->qlen;
1405 __skb_queue_head_init(list);
1406 }
1407 }
1408
1409 /**
1410 * skb_queue_splice_tail - join two skb lists, each list being a queue
1411 * @list: the new list to add
1412 * @head: the place to add it in the first list
1413 */
1414 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1415 struct sk_buff_head *head)
1416 {
1417 if (!skb_queue_empty(list)) {
1418 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1419 head->qlen += list->qlen;
1420 }
1421 }
1422
1423 /**
1424 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1425 * @list: the new list to add
1426 * @head: the place to add it in the first list
1427 *
1428 * Each of the lists is a queue.
1429 * The list at @list is reinitialised
1430 */
1431 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1432 struct sk_buff_head *head)
1433 {
1434 if (!skb_queue_empty(list)) {
1435 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1436 head->qlen += list->qlen;
1437 __skb_queue_head_init(list);
1438 }
1439 }
1440
1441 /**
1442 * __skb_queue_after - queue a buffer at the list head
1443 * @list: list to use
1444 * @prev: place after this buffer
1445 * @newsk: buffer to queue
1446 *
1447 * Queue a buffer int the middle of a list. This function takes no locks
1448 * and you must therefore hold required locks before calling it.
1449 *
1450 * A buffer cannot be placed on two lists at the same time.
1451 */
1452 static inline void __skb_queue_after(struct sk_buff_head *list,
1453 struct sk_buff *prev,
1454 struct sk_buff *newsk)
1455 {
1456 __skb_insert(newsk, prev, prev->next, list);
1457 }
1458
1459 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1460 struct sk_buff_head *list);
1461
1462 static inline void __skb_queue_before(struct sk_buff_head *list,
1463 struct sk_buff *next,
1464 struct sk_buff *newsk)
1465 {
1466 __skb_insert(newsk, next->prev, next, list);
1467 }
1468
1469 /**
1470 * __skb_queue_head - queue a buffer at the list head
1471 * @list: list to use
1472 * @newsk: buffer to queue
1473 *
1474 * Queue a buffer at the start of a list. This function takes no locks
1475 * and you must therefore hold required locks before calling it.
1476 *
1477 * A buffer cannot be placed on two lists at the same time.
1478 */
1479 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1480 static inline void __skb_queue_head(struct sk_buff_head *list,
1481 struct sk_buff *newsk)
1482 {
1483 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1484 }
1485
1486 /**
1487 * __skb_queue_tail - queue a buffer at the list tail
1488 * @list: list to use
1489 * @newsk: buffer to queue
1490 *
1491 * Queue a buffer at the end of a list. This function takes no locks
1492 * and you must therefore hold required locks before calling it.
1493 *
1494 * A buffer cannot be placed on two lists at the same time.
1495 */
1496 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1497 static inline void __skb_queue_tail(struct sk_buff_head *list,
1498 struct sk_buff *newsk)
1499 {
1500 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1501 }
1502
1503 /*
1504 * remove sk_buff from list. _Must_ be called atomically, and with
1505 * the list known..
1506 */
1507 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1508 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1509 {
1510 struct sk_buff *next, *prev;
1511
1512 list->qlen--;
1513 next = skb->next;
1514 prev = skb->prev;
1515 skb->next = skb->prev = NULL;
1516 next->prev = prev;
1517 prev->next = next;
1518 }
1519
1520 /**
1521 * __skb_dequeue - remove from the head of the queue
1522 * @list: list to dequeue from
1523 *
1524 * Remove the head of the list. This function does not take any locks
1525 * so must be used with appropriate locks held only. The head item is
1526 * returned or %NULL if the list is empty.
1527 */
1528 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1529 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1530 {
1531 struct sk_buff *skb = skb_peek(list);
1532 if (skb)
1533 __skb_unlink(skb, list);
1534 return skb;
1535 }
1536
1537 /**
1538 * __skb_dequeue_tail - remove from the tail of the queue
1539 * @list: list to dequeue from
1540 *
1541 * Remove the tail of the list. This function does not take any locks
1542 * so must be used with appropriate locks held only. The tail item is
1543 * returned or %NULL if the list is empty.
1544 */
1545 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1546 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1547 {
1548 struct sk_buff *skb = skb_peek_tail(list);
1549 if (skb)
1550 __skb_unlink(skb, list);
1551 return skb;
1552 }
1553
1554
1555 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1556 {
1557 return skb->data_len;
1558 }
1559
1560 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1561 {
1562 return skb->len - skb->data_len;
1563 }
1564
1565 static inline int skb_pagelen(const struct sk_buff *skb)
1566 {
1567 int i, len = 0;
1568
1569 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1570 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1571 return len + skb_headlen(skb);
1572 }
1573
1574 /**
1575 * __skb_fill_page_desc - initialise a paged fragment in an skb
1576 * @skb: buffer containing fragment to be initialised
1577 * @i: paged fragment index to initialise
1578 * @page: the page to use for this fragment
1579 * @off: the offset to the data with @page
1580 * @size: the length of the data
1581 *
1582 * Initialises the @i'th fragment of @skb to point to &size bytes at
1583 * offset @off within @page.
1584 *
1585 * Does not take any additional reference on the fragment.
1586 */
1587 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1588 struct page *page, int off, int size)
1589 {
1590 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1591
1592 /*
1593 * Propagate page->pfmemalloc to the skb if we can. The problem is
1594 * that not all callers have unique ownership of the page. If
1595 * pfmemalloc is set, we check the mapping as a mapping implies
1596 * page->index is set (index and pfmemalloc share space).
1597 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1598 * do not lose pfmemalloc information as the pages would not be
1599 * allocated using __GFP_MEMALLOC.
1600 */
1601 frag->page.p = page;
1602 frag->page_offset = off;
1603 skb_frag_size_set(frag, size);
1604
1605 page = compound_head(page);
1606 if (page->pfmemalloc && !page->mapping)
1607 skb->pfmemalloc = true;
1608 }
1609
1610 /**
1611 * skb_fill_page_desc - initialise a paged fragment in an skb
1612 * @skb: buffer containing fragment to be initialised
1613 * @i: paged fragment index to initialise
1614 * @page: the page to use for this fragment
1615 * @off: the offset to the data with @page
1616 * @size: the length of the data
1617 *
1618 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1619 * @skb to point to @size bytes at offset @off within @page. In
1620 * addition updates @skb such that @i is the last fragment.
1621 *
1622 * Does not take any additional reference on the fragment.
1623 */
1624 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1625 struct page *page, int off, int size)
1626 {
1627 __skb_fill_page_desc(skb, i, page, off, size);
1628 skb_shinfo(skb)->nr_frags = i + 1;
1629 }
1630
1631 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1632 int size, unsigned int truesize);
1633
1634 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1635 unsigned int truesize);
1636
1637 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1638 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1639 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1640
1641 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1642 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1643 {
1644 return skb->head + skb->tail;
1645 }
1646
1647 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1648 {
1649 skb->tail = skb->data - skb->head;
1650 }
1651
1652 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1653 {
1654 skb_reset_tail_pointer(skb);
1655 skb->tail += offset;
1656 }
1657
1658 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1659 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1660 {
1661 return skb->tail;
1662 }
1663
1664 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1665 {
1666 skb->tail = skb->data;
1667 }
1668
1669 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1670 {
1671 skb->tail = skb->data + offset;
1672 }
1673
1674 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1675
1676 /*
1677 * Add data to an sk_buff
1678 */
1679 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1680 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1681 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1682 {
1683 unsigned char *tmp = skb_tail_pointer(skb);
1684 SKB_LINEAR_ASSERT(skb);
1685 skb->tail += len;
1686 skb->len += len;
1687 return tmp;
1688 }
1689
1690 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1691 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1692 {
1693 skb->data -= len;
1694 skb->len += len;
1695 return skb->data;
1696 }
1697
1698 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1699 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1700 {
1701 skb->len -= len;
1702 BUG_ON(skb->len < skb->data_len);
1703 return skb->data += len;
1704 }
1705
1706 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1707 {
1708 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1709 }
1710
1711 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1712
1713 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1714 {
1715 if (len > skb_headlen(skb) &&
1716 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1717 return NULL;
1718 skb->len -= len;
1719 return skb->data += len;
1720 }
1721
1722 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1723 {
1724 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1725 }
1726
1727 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1728 {
1729 if (likely(len <= skb_headlen(skb)))
1730 return 1;
1731 if (unlikely(len > skb->len))
1732 return 0;
1733 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1734 }
1735
1736 /**
1737 * skb_headroom - bytes at buffer head
1738 * @skb: buffer to check
1739 *
1740 * Return the number of bytes of free space at the head of an &sk_buff.
1741 */
1742 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1743 {
1744 return skb->data - skb->head;
1745 }
1746
1747 /**
1748 * skb_tailroom - bytes at buffer end
1749 * @skb: buffer to check
1750 *
1751 * Return the number of bytes of free space at the tail of an sk_buff
1752 */
1753 static inline int skb_tailroom(const struct sk_buff *skb)
1754 {
1755 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1756 }
1757
1758 /**
1759 * skb_availroom - bytes at buffer end
1760 * @skb: buffer to check
1761 *
1762 * Return the number of bytes of free space at the tail of an sk_buff
1763 * allocated by sk_stream_alloc()
1764 */
1765 static inline int skb_availroom(const struct sk_buff *skb)
1766 {
1767 if (skb_is_nonlinear(skb))
1768 return 0;
1769
1770 return skb->end - skb->tail - skb->reserved_tailroom;
1771 }
1772
1773 /**
1774 * skb_reserve - adjust headroom
1775 * @skb: buffer to alter
1776 * @len: bytes to move
1777 *
1778 * Increase the headroom of an empty &sk_buff by reducing the tail
1779 * room. This is only allowed for an empty buffer.
1780 */
1781 static inline void skb_reserve(struct sk_buff *skb, int len)
1782 {
1783 skb->data += len;
1784 skb->tail += len;
1785 }
1786
1787 #define ENCAP_TYPE_ETHER 0
1788 #define ENCAP_TYPE_IPPROTO 1
1789
1790 static inline void skb_set_inner_protocol(struct sk_buff *skb,
1791 __be16 protocol)
1792 {
1793 skb->inner_protocol = protocol;
1794 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1795 }
1796
1797 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1798 __u8 ipproto)
1799 {
1800 skb->inner_ipproto = ipproto;
1801 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1802 }
1803
1804 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1805 {
1806 skb->inner_mac_header = skb->mac_header;
1807 skb->inner_network_header = skb->network_header;
1808 skb->inner_transport_header = skb->transport_header;
1809 }
1810
1811 static inline void skb_reset_mac_len(struct sk_buff *skb)
1812 {
1813 skb->mac_len = skb->network_header - skb->mac_header;
1814 }
1815
1816 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1817 *skb)
1818 {
1819 return skb->head + skb->inner_transport_header;
1820 }
1821
1822 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1823 {
1824 skb->inner_transport_header = skb->data - skb->head;
1825 }
1826
1827 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1828 const int offset)
1829 {
1830 skb_reset_inner_transport_header(skb);
1831 skb->inner_transport_header += offset;
1832 }
1833
1834 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1835 {
1836 return skb->head + skb->inner_network_header;
1837 }
1838
1839 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1840 {
1841 skb->inner_network_header = skb->data - skb->head;
1842 }
1843
1844 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1845 const int offset)
1846 {
1847 skb_reset_inner_network_header(skb);
1848 skb->inner_network_header += offset;
1849 }
1850
1851 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1852 {
1853 return skb->head + skb->inner_mac_header;
1854 }
1855
1856 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1857 {
1858 skb->inner_mac_header = skb->data - skb->head;
1859 }
1860
1861 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1862 const int offset)
1863 {
1864 skb_reset_inner_mac_header(skb);
1865 skb->inner_mac_header += offset;
1866 }
1867 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1868 {
1869 return skb->transport_header != (typeof(skb->transport_header))~0U;
1870 }
1871
1872 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1873 {
1874 return skb->head + skb->transport_header;
1875 }
1876
1877 static inline void skb_reset_transport_header(struct sk_buff *skb)
1878 {
1879 skb->transport_header = skb->data - skb->head;
1880 }
1881
1882 static inline void skb_set_transport_header(struct sk_buff *skb,
1883 const int offset)
1884 {
1885 skb_reset_transport_header(skb);
1886 skb->transport_header += offset;
1887 }
1888
1889 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1890 {
1891 return skb->head + skb->network_header;
1892 }
1893
1894 static inline void skb_reset_network_header(struct sk_buff *skb)
1895 {
1896 skb->network_header = skb->data - skb->head;
1897 }
1898
1899 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1900 {
1901 skb_reset_network_header(skb);
1902 skb->network_header += offset;
1903 }
1904
1905 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1906 {
1907 return skb->head + skb->mac_header;
1908 }
1909
1910 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1911 {
1912 return skb->mac_header != (typeof(skb->mac_header))~0U;
1913 }
1914
1915 static inline void skb_reset_mac_header(struct sk_buff *skb)
1916 {
1917 skb->mac_header = skb->data - skb->head;
1918 }
1919
1920 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1921 {
1922 skb_reset_mac_header(skb);
1923 skb->mac_header += offset;
1924 }
1925
1926 static inline void skb_pop_mac_header(struct sk_buff *skb)
1927 {
1928 skb->mac_header = skb->network_header;
1929 }
1930
1931 static inline void skb_probe_transport_header(struct sk_buff *skb,
1932 const int offset_hint)
1933 {
1934 struct flow_keys keys;
1935
1936 if (skb_transport_header_was_set(skb))
1937 return;
1938 else if (skb_flow_dissect(skb, &keys))
1939 skb_set_transport_header(skb, keys.thoff);
1940 else
1941 skb_set_transport_header(skb, offset_hint);
1942 }
1943
1944 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1945 {
1946 if (skb_mac_header_was_set(skb)) {
1947 const unsigned char *old_mac = skb_mac_header(skb);
1948
1949 skb_set_mac_header(skb, -skb->mac_len);
1950 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1951 }
1952 }
1953
1954 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1955 {
1956 return skb->csum_start - skb_headroom(skb);
1957 }
1958
1959 static inline int skb_transport_offset(const struct sk_buff *skb)
1960 {
1961 return skb_transport_header(skb) - skb->data;
1962 }
1963
1964 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1965 {
1966 return skb->transport_header - skb->network_header;
1967 }
1968
1969 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1970 {
1971 return skb->inner_transport_header - skb->inner_network_header;
1972 }
1973
1974 static inline int skb_network_offset(const struct sk_buff *skb)
1975 {
1976 return skb_network_header(skb) - skb->data;
1977 }
1978
1979 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1980 {
1981 return skb_inner_network_header(skb) - skb->data;
1982 }
1983
1984 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1985 {
1986 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1987 }
1988
1989 /*
1990 * CPUs often take a performance hit when accessing unaligned memory
1991 * locations. The actual performance hit varies, it can be small if the
1992 * hardware handles it or large if we have to take an exception and fix it
1993 * in software.
1994 *
1995 * Since an ethernet header is 14 bytes network drivers often end up with
1996 * the IP header at an unaligned offset. The IP header can be aligned by
1997 * shifting the start of the packet by 2 bytes. Drivers should do this
1998 * with:
1999 *
2000 * skb_reserve(skb, NET_IP_ALIGN);
2001 *
2002 * The downside to this alignment of the IP header is that the DMA is now
2003 * unaligned. On some architectures the cost of an unaligned DMA is high
2004 * and this cost outweighs the gains made by aligning the IP header.
2005 *
2006 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2007 * to be overridden.
2008 */
2009 #ifndef NET_IP_ALIGN
2010 #define NET_IP_ALIGN 2
2011 #endif
2012
2013 /*
2014 * The networking layer reserves some headroom in skb data (via
2015 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2016 * the header has to grow. In the default case, if the header has to grow
2017 * 32 bytes or less we avoid the reallocation.
2018 *
2019 * Unfortunately this headroom changes the DMA alignment of the resulting
2020 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2021 * on some architectures. An architecture can override this value,
2022 * perhaps setting it to a cacheline in size (since that will maintain
2023 * cacheline alignment of the DMA). It must be a power of 2.
2024 *
2025 * Various parts of the networking layer expect at least 32 bytes of
2026 * headroom, you should not reduce this.
2027 *
2028 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2029 * to reduce average number of cache lines per packet.
2030 * get_rps_cpus() for example only access one 64 bytes aligned block :
2031 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2032 */
2033 #ifndef NET_SKB_PAD
2034 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2035 #endif
2036
2037 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2038
2039 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2040 {
2041 if (unlikely(skb_is_nonlinear(skb))) {
2042 WARN_ON(1);
2043 return;
2044 }
2045 skb->len = len;
2046 skb_set_tail_pointer(skb, len);
2047 }
2048
2049 void skb_trim(struct sk_buff *skb, unsigned int len);
2050
2051 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2052 {
2053 if (skb->data_len)
2054 return ___pskb_trim(skb, len);
2055 __skb_trim(skb, len);
2056 return 0;
2057 }
2058
2059 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2060 {
2061 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2062 }
2063
2064 /**
2065 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2066 * @skb: buffer to alter
2067 * @len: new length
2068 *
2069 * This is identical to pskb_trim except that the caller knows that
2070 * the skb is not cloned so we should never get an error due to out-
2071 * of-memory.
2072 */
2073 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2074 {
2075 int err = pskb_trim(skb, len);
2076 BUG_ON(err);
2077 }
2078
2079 /**
2080 * skb_orphan - orphan a buffer
2081 * @skb: buffer to orphan
2082 *
2083 * If a buffer currently has an owner then we call the owner's
2084 * destructor function and make the @skb unowned. The buffer continues
2085 * to exist but is no longer charged to its former owner.
2086 */
2087 static inline void skb_orphan(struct sk_buff *skb)
2088 {
2089 if (skb->destructor) {
2090 skb->destructor(skb);
2091 skb->destructor = NULL;
2092 skb->sk = NULL;
2093 } else {
2094 BUG_ON(skb->sk);
2095 }
2096 }
2097
2098 /**
2099 * skb_orphan_frags - orphan the frags contained in a buffer
2100 * @skb: buffer to orphan frags from
2101 * @gfp_mask: allocation mask for replacement pages
2102 *
2103 * For each frag in the SKB which needs a destructor (i.e. has an
2104 * owner) create a copy of that frag and release the original
2105 * page by calling the destructor.
2106 */
2107 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2108 {
2109 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2110 return 0;
2111 return skb_copy_ubufs(skb, gfp_mask);
2112 }
2113
2114 /**
2115 * __skb_queue_purge - empty a list
2116 * @list: list to empty
2117 *
2118 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2119 * the list and one reference dropped. This function does not take the
2120 * list lock and the caller must hold the relevant locks to use it.
2121 */
2122 void skb_queue_purge(struct sk_buff_head *list);
2123 static inline void __skb_queue_purge(struct sk_buff_head *list)
2124 {
2125 struct sk_buff *skb;
2126 while ((skb = __skb_dequeue(list)) != NULL)
2127 kfree_skb(skb);
2128 }
2129
2130 void *netdev_alloc_frag(unsigned int fragsz);
2131
2132 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2133 gfp_t gfp_mask);
2134
2135 /**
2136 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2137 * @dev: network device to receive on
2138 * @length: length to allocate
2139 *
2140 * Allocate a new &sk_buff and assign it a usage count of one. The
2141 * buffer has unspecified headroom built in. Users should allocate
2142 * the headroom they think they need without accounting for the
2143 * built in space. The built in space is used for optimisations.
2144 *
2145 * %NULL is returned if there is no free memory. Although this function
2146 * allocates memory it can be called from an interrupt.
2147 */
2148 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2149 unsigned int length)
2150 {
2151 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2152 }
2153
2154 /* legacy helper around __netdev_alloc_skb() */
2155 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2156 gfp_t gfp_mask)
2157 {
2158 return __netdev_alloc_skb(NULL, length, gfp_mask);
2159 }
2160
2161 /* legacy helper around netdev_alloc_skb() */
2162 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2163 {
2164 return netdev_alloc_skb(NULL, length);
2165 }
2166
2167
2168 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2169 unsigned int length, gfp_t gfp)
2170 {
2171 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2172
2173 if (NET_IP_ALIGN && skb)
2174 skb_reserve(skb, NET_IP_ALIGN);
2175 return skb;
2176 }
2177
2178 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2179 unsigned int length)
2180 {
2181 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2182 }
2183
2184 static inline void skb_free_frag(void *addr)
2185 {
2186 __free_page_frag(addr);
2187 }
2188
2189 void *napi_alloc_frag(unsigned int fragsz);
2190 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2191 unsigned int length, gfp_t gfp_mask);
2192 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2193 unsigned int length)
2194 {
2195 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2196 }
2197
2198 /**
2199 * __dev_alloc_pages - allocate page for network Rx
2200 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2201 * @order: size of the allocation
2202 *
2203 * Allocate a new page.
2204 *
2205 * %NULL is returned if there is no free memory.
2206 */
2207 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2208 unsigned int order)
2209 {
2210 /* This piece of code contains several assumptions.
2211 * 1. This is for device Rx, therefor a cold page is preferred.
2212 * 2. The expectation is the user wants a compound page.
2213 * 3. If requesting a order 0 page it will not be compound
2214 * due to the check to see if order has a value in prep_new_page
2215 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2216 * code in gfp_to_alloc_flags that should be enforcing this.
2217 */
2218 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2219
2220 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2221 }
2222
2223 static inline struct page *dev_alloc_pages(unsigned int order)
2224 {
2225 return __dev_alloc_pages(GFP_ATOMIC, order);
2226 }
2227
2228 /**
2229 * __dev_alloc_page - allocate a page for network Rx
2230 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2231 *
2232 * Allocate a new page.
2233 *
2234 * %NULL is returned if there is no free memory.
2235 */
2236 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2237 {
2238 return __dev_alloc_pages(gfp_mask, 0);
2239 }
2240
2241 static inline struct page *dev_alloc_page(void)
2242 {
2243 return __dev_alloc_page(GFP_ATOMIC);
2244 }
2245
2246 /**
2247 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2248 * @page: The page that was allocated from skb_alloc_page
2249 * @skb: The skb that may need pfmemalloc set
2250 */
2251 static inline void skb_propagate_pfmemalloc(struct page *page,
2252 struct sk_buff *skb)
2253 {
2254 if (page && page->pfmemalloc)
2255 skb->pfmemalloc = true;
2256 }
2257
2258 /**
2259 * skb_frag_page - retrieve the page referred to by a paged fragment
2260 * @frag: the paged fragment
2261 *
2262 * Returns the &struct page associated with @frag.
2263 */
2264 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2265 {
2266 return frag->page.p;
2267 }
2268
2269 /**
2270 * __skb_frag_ref - take an addition reference on a paged fragment.
2271 * @frag: the paged fragment
2272 *
2273 * Takes an additional reference on the paged fragment @frag.
2274 */
2275 static inline void __skb_frag_ref(skb_frag_t *frag)
2276 {
2277 get_page(skb_frag_page(frag));
2278 }
2279
2280 /**
2281 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2282 * @skb: the buffer
2283 * @f: the fragment offset.
2284 *
2285 * Takes an additional reference on the @f'th paged fragment of @skb.
2286 */
2287 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2288 {
2289 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2290 }
2291
2292 /**
2293 * __skb_frag_unref - release a reference on a paged fragment.
2294 * @frag: the paged fragment
2295 *
2296 * Releases a reference on the paged fragment @frag.
2297 */
2298 static inline void __skb_frag_unref(skb_frag_t *frag)
2299 {
2300 put_page(skb_frag_page(frag));
2301 }
2302
2303 /**
2304 * skb_frag_unref - release a reference on a paged fragment of an skb.
2305 * @skb: the buffer
2306 * @f: the fragment offset
2307 *
2308 * Releases a reference on the @f'th paged fragment of @skb.
2309 */
2310 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2311 {
2312 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2313 }
2314
2315 /**
2316 * skb_frag_address - gets the address of the data contained in a paged fragment
2317 * @frag: the paged fragment buffer
2318 *
2319 * Returns the address of the data within @frag. The page must already
2320 * be mapped.
2321 */
2322 static inline void *skb_frag_address(const skb_frag_t *frag)
2323 {
2324 return page_address(skb_frag_page(frag)) + frag->page_offset;
2325 }
2326
2327 /**
2328 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2329 * @frag: the paged fragment buffer
2330 *
2331 * Returns the address of the data within @frag. Checks that the page
2332 * is mapped and returns %NULL otherwise.
2333 */
2334 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2335 {
2336 void *ptr = page_address(skb_frag_page(frag));
2337 if (unlikely(!ptr))
2338 return NULL;
2339
2340 return ptr + frag->page_offset;
2341 }
2342
2343 /**
2344 * __skb_frag_set_page - sets the page contained in a paged fragment
2345 * @frag: the paged fragment
2346 * @page: the page to set
2347 *
2348 * Sets the fragment @frag to contain @page.
2349 */
2350 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2351 {
2352 frag->page.p = page;
2353 }
2354
2355 /**
2356 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2357 * @skb: the buffer
2358 * @f: the fragment offset
2359 * @page: the page to set
2360 *
2361 * Sets the @f'th fragment of @skb to contain @page.
2362 */
2363 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2364 struct page *page)
2365 {
2366 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2367 }
2368
2369 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2370
2371 /**
2372 * skb_frag_dma_map - maps a paged fragment via the DMA API
2373 * @dev: the device to map the fragment to
2374 * @frag: the paged fragment to map
2375 * @offset: the offset within the fragment (starting at the
2376 * fragment's own offset)
2377 * @size: the number of bytes to map
2378 * @dir: the direction of the mapping (%PCI_DMA_*)
2379 *
2380 * Maps the page associated with @frag to @device.
2381 */
2382 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2383 const skb_frag_t *frag,
2384 size_t offset, size_t size,
2385 enum dma_data_direction dir)
2386 {
2387 return dma_map_page(dev, skb_frag_page(frag),
2388 frag->page_offset + offset, size, dir);
2389 }
2390
2391 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2392 gfp_t gfp_mask)
2393 {
2394 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2395 }
2396
2397
2398 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2399 gfp_t gfp_mask)
2400 {
2401 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2402 }
2403
2404
2405 /**
2406 * skb_clone_writable - is the header of a clone writable
2407 * @skb: buffer to check
2408 * @len: length up to which to write
2409 *
2410 * Returns true if modifying the header part of the cloned buffer
2411 * does not requires the data to be copied.
2412 */
2413 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2414 {
2415 return !skb_header_cloned(skb) &&
2416 skb_headroom(skb) + len <= skb->hdr_len;
2417 }
2418
2419 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2420 int cloned)
2421 {
2422 int delta = 0;
2423
2424 if (headroom > skb_headroom(skb))
2425 delta = headroom - skb_headroom(skb);
2426
2427 if (delta || cloned)
2428 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2429 GFP_ATOMIC);
2430 return 0;
2431 }
2432
2433 /**
2434 * skb_cow - copy header of skb when it is required
2435 * @skb: buffer to cow
2436 * @headroom: needed headroom
2437 *
2438 * If the skb passed lacks sufficient headroom or its data part
2439 * is shared, data is reallocated. If reallocation fails, an error
2440 * is returned and original skb is not changed.
2441 *
2442 * The result is skb with writable area skb->head...skb->tail
2443 * and at least @headroom of space at head.
2444 */
2445 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2446 {
2447 return __skb_cow(skb, headroom, skb_cloned(skb));
2448 }
2449
2450 /**
2451 * skb_cow_head - skb_cow but only making the head writable
2452 * @skb: buffer to cow
2453 * @headroom: needed headroom
2454 *
2455 * This function is identical to skb_cow except that we replace the
2456 * skb_cloned check by skb_header_cloned. It should be used when
2457 * you only need to push on some header and do not need to modify
2458 * the data.
2459 */
2460 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2461 {
2462 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2463 }
2464
2465 /**
2466 * skb_padto - pad an skbuff up to a minimal size
2467 * @skb: buffer to pad
2468 * @len: minimal length
2469 *
2470 * Pads up a buffer to ensure the trailing bytes exist and are
2471 * blanked. If the buffer already contains sufficient data it
2472 * is untouched. Otherwise it is extended. Returns zero on
2473 * success. The skb is freed on error.
2474 */
2475 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2476 {
2477 unsigned int size = skb->len;
2478 if (likely(size >= len))
2479 return 0;
2480 return skb_pad(skb, len - size);
2481 }
2482
2483 /**
2484 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2485 * @skb: buffer to pad
2486 * @len: minimal length
2487 *
2488 * Pads up a buffer to ensure the trailing bytes exist and are
2489 * blanked. If the buffer already contains sufficient data it
2490 * is untouched. Otherwise it is extended. Returns zero on
2491 * success. The skb is freed on error.
2492 */
2493 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2494 {
2495 unsigned int size = skb->len;
2496
2497 if (unlikely(size < len)) {
2498 len -= size;
2499 if (skb_pad(skb, len))
2500 return -ENOMEM;
2501 __skb_put(skb, len);
2502 }
2503 return 0;
2504 }
2505
2506 static inline int skb_add_data(struct sk_buff *skb,
2507 struct iov_iter *from, int copy)
2508 {
2509 const int off = skb->len;
2510
2511 if (skb->ip_summed == CHECKSUM_NONE) {
2512 __wsum csum = 0;
2513 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2514 &csum, from) == copy) {
2515 skb->csum = csum_block_add(skb->csum, csum, off);
2516 return 0;
2517 }
2518 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2519 return 0;
2520
2521 __skb_trim(skb, off);
2522 return -EFAULT;
2523 }
2524
2525 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2526 const struct page *page, int off)
2527 {
2528 if (i) {
2529 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2530
2531 return page == skb_frag_page(frag) &&
2532 off == frag->page_offset + skb_frag_size(frag);
2533 }
2534 return false;
2535 }
2536
2537 static inline int __skb_linearize(struct sk_buff *skb)
2538 {
2539 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2540 }
2541
2542 /**
2543 * skb_linearize - convert paged skb to linear one
2544 * @skb: buffer to linarize
2545 *
2546 * If there is no free memory -ENOMEM is returned, otherwise zero
2547 * is returned and the old skb data released.
2548 */
2549 static inline int skb_linearize(struct sk_buff *skb)
2550 {
2551 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2552 }
2553
2554 /**
2555 * skb_has_shared_frag - can any frag be overwritten
2556 * @skb: buffer to test
2557 *
2558 * Return true if the skb has at least one frag that might be modified
2559 * by an external entity (as in vmsplice()/sendfile())
2560 */
2561 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2562 {
2563 return skb_is_nonlinear(skb) &&
2564 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2565 }
2566
2567 /**
2568 * skb_linearize_cow - make sure skb is linear and writable
2569 * @skb: buffer to process
2570 *
2571 * If there is no free memory -ENOMEM is returned, otherwise zero
2572 * is returned and the old skb data released.
2573 */
2574 static inline int skb_linearize_cow(struct sk_buff *skb)
2575 {
2576 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2577 __skb_linearize(skb) : 0;
2578 }
2579
2580 /**
2581 * skb_postpull_rcsum - update checksum for received skb after pull
2582 * @skb: buffer to update
2583 * @start: start of data before pull
2584 * @len: length of data pulled
2585 *
2586 * After doing a pull on a received packet, you need to call this to
2587 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2588 * CHECKSUM_NONE so that it can be recomputed from scratch.
2589 */
2590
2591 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2592 const void *start, unsigned int len)
2593 {
2594 if (skb->ip_summed == CHECKSUM_COMPLETE)
2595 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2596 }
2597
2598 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2599
2600 /**
2601 * pskb_trim_rcsum - trim received skb and update checksum
2602 * @skb: buffer to trim
2603 * @len: new length
2604 *
2605 * This is exactly the same as pskb_trim except that it ensures the
2606 * checksum of received packets are still valid after the operation.
2607 */
2608
2609 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2610 {
2611 if (likely(len >= skb->len))
2612 return 0;
2613 if (skb->ip_summed == CHECKSUM_COMPLETE)
2614 skb->ip_summed = CHECKSUM_NONE;
2615 return __pskb_trim(skb, len);
2616 }
2617
2618 #define skb_queue_walk(queue, skb) \
2619 for (skb = (queue)->next; \
2620 skb != (struct sk_buff *)(queue); \
2621 skb = skb->next)
2622
2623 #define skb_queue_walk_safe(queue, skb, tmp) \
2624 for (skb = (queue)->next, tmp = skb->next; \
2625 skb != (struct sk_buff *)(queue); \
2626 skb = tmp, tmp = skb->next)
2627
2628 #define skb_queue_walk_from(queue, skb) \
2629 for (; skb != (struct sk_buff *)(queue); \
2630 skb = skb->next)
2631
2632 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2633 for (tmp = skb->next; \
2634 skb != (struct sk_buff *)(queue); \
2635 skb = tmp, tmp = skb->next)
2636
2637 #define skb_queue_reverse_walk(queue, skb) \
2638 for (skb = (queue)->prev; \
2639 skb != (struct sk_buff *)(queue); \
2640 skb = skb->prev)
2641
2642 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2643 for (skb = (queue)->prev, tmp = skb->prev; \
2644 skb != (struct sk_buff *)(queue); \
2645 skb = tmp, tmp = skb->prev)
2646
2647 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2648 for (tmp = skb->prev; \
2649 skb != (struct sk_buff *)(queue); \
2650 skb = tmp, tmp = skb->prev)
2651
2652 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2653 {
2654 return skb_shinfo(skb)->frag_list != NULL;
2655 }
2656
2657 static inline void skb_frag_list_init(struct sk_buff *skb)
2658 {
2659 skb_shinfo(skb)->frag_list = NULL;
2660 }
2661
2662 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2663 {
2664 frag->next = skb_shinfo(skb)->frag_list;
2665 skb_shinfo(skb)->frag_list = frag;
2666 }
2667
2668 #define skb_walk_frags(skb, iter) \
2669 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2670
2671 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2672 int *peeked, int *off, int *err);
2673 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2674 int *err);
2675 unsigned int datagram_poll(struct file *file, struct socket *sock,
2676 struct poll_table_struct *wait);
2677 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2678 struct iov_iter *to, int size);
2679 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2680 struct msghdr *msg, int size)
2681 {
2682 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2683 }
2684 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2685 struct msghdr *msg);
2686 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2687 struct iov_iter *from, int len);
2688 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2689 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2690 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2691 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2692 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2693 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2694 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2695 int len, __wsum csum);
2696 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2697 struct pipe_inode_info *pipe, unsigned int len,
2698 unsigned int flags);
2699 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2700 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2701 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2702 int len, int hlen);
2703 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2704 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2705 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2706 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2707 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2708 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2709 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2710 int skb_vlan_pop(struct sk_buff *skb);
2711 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
2712
2713 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2714 {
2715 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2716 }
2717
2718 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2719 {
2720 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2721 }
2722
2723 struct skb_checksum_ops {
2724 __wsum (*update)(const void *mem, int len, __wsum wsum);
2725 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2726 };
2727
2728 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2729 __wsum csum, const struct skb_checksum_ops *ops);
2730 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2731 __wsum csum);
2732
2733 static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset,
2734 int len, void *data, int hlen, void *buffer)
2735 {
2736 if (hlen - offset >= len)
2737 return data + offset;
2738
2739 if (!skb ||
2740 skb_copy_bits(skb, offset, buffer, len) < 0)
2741 return NULL;
2742
2743 return buffer;
2744 }
2745
2746 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2747 int len, void *buffer)
2748 {
2749 return __skb_header_pointer(skb, offset, len, skb->data,
2750 skb_headlen(skb), buffer);
2751 }
2752
2753 /**
2754 * skb_needs_linearize - check if we need to linearize a given skb
2755 * depending on the given device features.
2756 * @skb: socket buffer to check
2757 * @features: net device features
2758 *
2759 * Returns true if either:
2760 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2761 * 2. skb is fragmented and the device does not support SG.
2762 */
2763 static inline bool skb_needs_linearize(struct sk_buff *skb,
2764 netdev_features_t features)
2765 {
2766 return skb_is_nonlinear(skb) &&
2767 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2768 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2769 }
2770
2771 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2772 void *to,
2773 const unsigned int len)
2774 {
2775 memcpy(to, skb->data, len);
2776 }
2777
2778 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2779 const int offset, void *to,
2780 const unsigned int len)
2781 {
2782 memcpy(to, skb->data + offset, len);
2783 }
2784
2785 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2786 const void *from,
2787 const unsigned int len)
2788 {
2789 memcpy(skb->data, from, len);
2790 }
2791
2792 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2793 const int offset,
2794 const void *from,
2795 const unsigned int len)
2796 {
2797 memcpy(skb->data + offset, from, len);
2798 }
2799
2800 void skb_init(void);
2801
2802 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2803 {
2804 return skb->tstamp;
2805 }
2806
2807 /**
2808 * skb_get_timestamp - get timestamp from a skb
2809 * @skb: skb to get stamp from
2810 * @stamp: pointer to struct timeval to store stamp in
2811 *
2812 * Timestamps are stored in the skb as offsets to a base timestamp.
2813 * This function converts the offset back to a struct timeval and stores
2814 * it in stamp.
2815 */
2816 static inline void skb_get_timestamp(const struct sk_buff *skb,
2817 struct timeval *stamp)
2818 {
2819 *stamp = ktime_to_timeval(skb->tstamp);
2820 }
2821
2822 static inline void skb_get_timestampns(const struct sk_buff *skb,
2823 struct timespec *stamp)
2824 {
2825 *stamp = ktime_to_timespec(skb->tstamp);
2826 }
2827
2828 static inline void __net_timestamp(struct sk_buff *skb)
2829 {
2830 skb->tstamp = ktime_get_real();
2831 }
2832
2833 static inline ktime_t net_timedelta(ktime_t t)
2834 {
2835 return ktime_sub(ktime_get_real(), t);
2836 }
2837
2838 static inline ktime_t net_invalid_timestamp(void)
2839 {
2840 return ktime_set(0, 0);
2841 }
2842
2843 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2844
2845 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2846
2847 void skb_clone_tx_timestamp(struct sk_buff *skb);
2848 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2849
2850 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2851
2852 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2853 {
2854 }
2855
2856 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2857 {
2858 return false;
2859 }
2860
2861 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2862
2863 /**
2864 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2865 *
2866 * PHY drivers may accept clones of transmitted packets for
2867 * timestamping via their phy_driver.txtstamp method. These drivers
2868 * must call this function to return the skb back to the stack, with
2869 * or without a timestamp.
2870 *
2871 * @skb: clone of the the original outgoing packet
2872 * @hwtstamps: hardware time stamps, may be NULL if not available
2873 *
2874 */
2875 void skb_complete_tx_timestamp(struct sk_buff *skb,
2876 struct skb_shared_hwtstamps *hwtstamps);
2877
2878 void __skb_tstamp_tx(struct sk_buff *orig_skb,
2879 struct skb_shared_hwtstamps *hwtstamps,
2880 struct sock *sk, int tstype);
2881
2882 /**
2883 * skb_tstamp_tx - queue clone of skb with send time stamps
2884 * @orig_skb: the original outgoing packet
2885 * @hwtstamps: hardware time stamps, may be NULL if not available
2886 *
2887 * If the skb has a socket associated, then this function clones the
2888 * skb (thus sharing the actual data and optional structures), stores
2889 * the optional hardware time stamping information (if non NULL) or
2890 * generates a software time stamp (otherwise), then queues the clone
2891 * to the error queue of the socket. Errors are silently ignored.
2892 */
2893 void skb_tstamp_tx(struct sk_buff *orig_skb,
2894 struct skb_shared_hwtstamps *hwtstamps);
2895
2896 static inline void sw_tx_timestamp(struct sk_buff *skb)
2897 {
2898 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2899 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2900 skb_tstamp_tx(skb, NULL);
2901 }
2902
2903 /**
2904 * skb_tx_timestamp() - Driver hook for transmit timestamping
2905 *
2906 * Ethernet MAC Drivers should call this function in their hard_xmit()
2907 * function immediately before giving the sk_buff to the MAC hardware.
2908 *
2909 * Specifically, one should make absolutely sure that this function is
2910 * called before TX completion of this packet can trigger. Otherwise
2911 * the packet could potentially already be freed.
2912 *
2913 * @skb: A socket buffer.
2914 */
2915 static inline void skb_tx_timestamp(struct sk_buff *skb)
2916 {
2917 skb_clone_tx_timestamp(skb);
2918 sw_tx_timestamp(skb);
2919 }
2920
2921 /**
2922 * skb_complete_wifi_ack - deliver skb with wifi status
2923 *
2924 * @skb: the original outgoing packet
2925 * @acked: ack status
2926 *
2927 */
2928 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2929
2930 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2931 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2932
2933 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2934 {
2935 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
2936 skb->csum_valid ||
2937 (skb->ip_summed == CHECKSUM_PARTIAL &&
2938 skb_checksum_start_offset(skb) >= 0));
2939 }
2940
2941 /**
2942 * skb_checksum_complete - Calculate checksum of an entire packet
2943 * @skb: packet to process
2944 *
2945 * This function calculates the checksum over the entire packet plus
2946 * the value of skb->csum. The latter can be used to supply the
2947 * checksum of a pseudo header as used by TCP/UDP. It returns the
2948 * checksum.
2949 *
2950 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2951 * this function can be used to verify that checksum on received
2952 * packets. In that case the function should return zero if the
2953 * checksum is correct. In particular, this function will return zero
2954 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2955 * hardware has already verified the correctness of the checksum.
2956 */
2957 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2958 {
2959 return skb_csum_unnecessary(skb) ?
2960 0 : __skb_checksum_complete(skb);
2961 }
2962
2963 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
2964 {
2965 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2966 if (skb->csum_level == 0)
2967 skb->ip_summed = CHECKSUM_NONE;
2968 else
2969 skb->csum_level--;
2970 }
2971 }
2972
2973 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
2974 {
2975 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2976 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
2977 skb->csum_level++;
2978 } else if (skb->ip_summed == CHECKSUM_NONE) {
2979 skb->ip_summed = CHECKSUM_UNNECESSARY;
2980 skb->csum_level = 0;
2981 }
2982 }
2983
2984 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
2985 {
2986 /* Mark current checksum as bad (typically called from GRO
2987 * path). In the case that ip_summed is CHECKSUM_NONE
2988 * this must be the first checksum encountered in the packet.
2989 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
2990 * checksum after the last one validated. For UDP, a zero
2991 * checksum can not be marked as bad.
2992 */
2993
2994 if (skb->ip_summed == CHECKSUM_NONE ||
2995 skb->ip_summed == CHECKSUM_UNNECESSARY)
2996 skb->csum_bad = 1;
2997 }
2998
2999 /* Check if we need to perform checksum complete validation.
3000 *
3001 * Returns true if checksum complete is needed, false otherwise
3002 * (either checksum is unnecessary or zero checksum is allowed).
3003 */
3004 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3005 bool zero_okay,
3006 __sum16 check)
3007 {
3008 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3009 skb->csum_valid = 1;
3010 __skb_decr_checksum_unnecessary(skb);
3011 return false;
3012 }
3013
3014 return true;
3015 }
3016
3017 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3018 * in checksum_init.
3019 */
3020 #define CHECKSUM_BREAK 76
3021
3022 /* Unset checksum-complete
3023 *
3024 * Unset checksum complete can be done when packet is being modified
3025 * (uncompressed for instance) and checksum-complete value is
3026 * invalidated.
3027 */
3028 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3029 {
3030 if (skb->ip_summed == CHECKSUM_COMPLETE)
3031 skb->ip_summed = CHECKSUM_NONE;
3032 }
3033
3034 /* Validate (init) checksum based on checksum complete.
3035 *
3036 * Return values:
3037 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3038 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3039 * checksum is stored in skb->csum for use in __skb_checksum_complete
3040 * non-zero: value of invalid checksum
3041 *
3042 */
3043 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3044 bool complete,
3045 __wsum psum)
3046 {
3047 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3048 if (!csum_fold(csum_add(psum, skb->csum))) {
3049 skb->csum_valid = 1;
3050 return 0;
3051 }
3052 } else if (skb->csum_bad) {
3053 /* ip_summed == CHECKSUM_NONE in this case */
3054 return 1;
3055 }
3056
3057 skb->csum = psum;
3058
3059 if (complete || skb->len <= CHECKSUM_BREAK) {
3060 __sum16 csum;
3061
3062 csum = __skb_checksum_complete(skb);
3063 skb->csum_valid = !csum;
3064 return csum;
3065 }
3066
3067 return 0;
3068 }
3069
3070 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3071 {
3072 return 0;
3073 }
3074
3075 /* Perform checksum validate (init). Note that this is a macro since we only
3076 * want to calculate the pseudo header which is an input function if necessary.
3077 * First we try to validate without any computation (checksum unnecessary) and
3078 * then calculate based on checksum complete calling the function to compute
3079 * pseudo header.
3080 *
3081 * Return values:
3082 * 0: checksum is validated or try to in skb_checksum_complete
3083 * non-zero: value of invalid checksum
3084 */
3085 #define __skb_checksum_validate(skb, proto, complete, \
3086 zero_okay, check, compute_pseudo) \
3087 ({ \
3088 __sum16 __ret = 0; \
3089 skb->csum_valid = 0; \
3090 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3091 __ret = __skb_checksum_validate_complete(skb, \
3092 complete, compute_pseudo(skb, proto)); \
3093 __ret; \
3094 })
3095
3096 #define skb_checksum_init(skb, proto, compute_pseudo) \
3097 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3098
3099 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3100 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3101
3102 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3103 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3104
3105 #define skb_checksum_validate_zero_check(skb, proto, check, \
3106 compute_pseudo) \
3107 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3108
3109 #define skb_checksum_simple_validate(skb) \
3110 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3111
3112 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3113 {
3114 return (skb->ip_summed == CHECKSUM_NONE &&
3115 skb->csum_valid && !skb->csum_bad);
3116 }
3117
3118 static inline void __skb_checksum_convert(struct sk_buff *skb,
3119 __sum16 check, __wsum pseudo)
3120 {
3121 skb->csum = ~pseudo;
3122 skb->ip_summed = CHECKSUM_COMPLETE;
3123 }
3124
3125 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3126 do { \
3127 if (__skb_checksum_convert_check(skb)) \
3128 __skb_checksum_convert(skb, check, \
3129 compute_pseudo(skb, proto)); \
3130 } while (0)
3131
3132 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3133 u16 start, u16 offset)
3134 {
3135 skb->ip_summed = CHECKSUM_PARTIAL;
3136 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3137 skb->csum_offset = offset - start;
3138 }
3139
3140 /* Update skbuf and packet to reflect the remote checksum offload operation.
3141 * When called, ptr indicates the starting point for skb->csum when
3142 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3143 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3144 */
3145 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3146 int start, int offset, bool nopartial)
3147 {
3148 __wsum delta;
3149
3150 if (!nopartial) {
3151 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3152 return;
3153 }
3154
3155 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3156 __skb_checksum_complete(skb);
3157 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3158 }
3159
3160 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3161
3162 /* Adjust skb->csum since we changed the packet */
3163 skb->csum = csum_add(skb->csum, delta);
3164 }
3165
3166 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3167 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3168 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3169 {
3170 if (nfct && atomic_dec_and_test(&nfct->use))
3171 nf_conntrack_destroy(nfct);
3172 }
3173 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3174 {
3175 if (nfct)
3176 atomic_inc(&nfct->use);
3177 }
3178 #endif
3179 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3180 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3181 {
3182 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3183 kfree(nf_bridge);
3184 }
3185 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3186 {
3187 if (nf_bridge)
3188 atomic_inc(&nf_bridge->use);
3189 }
3190 #endif /* CONFIG_BRIDGE_NETFILTER */
3191 static inline void nf_reset(struct sk_buff *skb)
3192 {
3193 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3194 nf_conntrack_put(skb->nfct);
3195 skb->nfct = NULL;
3196 #endif
3197 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3198 nf_bridge_put(skb->nf_bridge);
3199 skb->nf_bridge = NULL;
3200 #endif
3201 }
3202
3203 static inline void nf_reset_trace(struct sk_buff *skb)
3204 {
3205 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3206 skb->nf_trace = 0;
3207 #endif
3208 }
3209
3210 /* Note: This doesn't put any conntrack and bridge info in dst. */
3211 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3212 bool copy)
3213 {
3214 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3215 dst->nfct = src->nfct;
3216 nf_conntrack_get(src->nfct);
3217 if (copy)
3218 dst->nfctinfo = src->nfctinfo;
3219 #endif
3220 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3221 dst->nf_bridge = src->nf_bridge;
3222 nf_bridge_get(src->nf_bridge);
3223 #endif
3224 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3225 if (copy)
3226 dst->nf_trace = src->nf_trace;
3227 #endif
3228 }
3229
3230 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3231 {
3232 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3233 nf_conntrack_put(dst->nfct);
3234 #endif
3235 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3236 nf_bridge_put(dst->nf_bridge);
3237 #endif
3238 __nf_copy(dst, src, true);
3239 }
3240
3241 #ifdef CONFIG_NETWORK_SECMARK
3242 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3243 {
3244 to->secmark = from->secmark;
3245 }
3246
3247 static inline void skb_init_secmark(struct sk_buff *skb)
3248 {
3249 skb->secmark = 0;
3250 }
3251 #else
3252 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3253 { }
3254
3255 static inline void skb_init_secmark(struct sk_buff *skb)
3256 { }
3257 #endif
3258
3259 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3260 {
3261 return !skb->destructor &&
3262 #if IS_ENABLED(CONFIG_XFRM)
3263 !skb->sp &&
3264 #endif
3265 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3266 !skb->nfct &&
3267 #endif
3268 !skb->_skb_refdst &&
3269 !skb_has_frag_list(skb);
3270 }
3271
3272 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3273 {
3274 skb->queue_mapping = queue_mapping;
3275 }
3276
3277 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3278 {
3279 return skb->queue_mapping;
3280 }
3281
3282 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3283 {
3284 to->queue_mapping = from->queue_mapping;
3285 }
3286
3287 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3288 {
3289 skb->queue_mapping = rx_queue + 1;
3290 }
3291
3292 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3293 {
3294 return skb->queue_mapping - 1;
3295 }
3296
3297 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3298 {
3299 return skb->queue_mapping != 0;
3300 }
3301
3302 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3303 {
3304 #ifdef CONFIG_XFRM
3305 return skb->sp;
3306 #else
3307 return NULL;
3308 #endif
3309 }
3310
3311 /* Keeps track of mac header offset relative to skb->head.
3312 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3313 * For non-tunnel skb it points to skb_mac_header() and for
3314 * tunnel skb it points to outer mac header.
3315 * Keeps track of level of encapsulation of network headers.
3316 */
3317 struct skb_gso_cb {
3318 int mac_offset;
3319 int encap_level;
3320 __u16 csum_start;
3321 };
3322 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3323
3324 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3325 {
3326 return (skb_mac_header(inner_skb) - inner_skb->head) -
3327 SKB_GSO_CB(inner_skb)->mac_offset;
3328 }
3329
3330 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3331 {
3332 int new_headroom, headroom;
3333 int ret;
3334
3335 headroom = skb_headroom(skb);
3336 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3337 if (ret)
3338 return ret;
3339
3340 new_headroom = skb_headroom(skb);
3341 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3342 return 0;
3343 }
3344
3345 /* Compute the checksum for a gso segment. First compute the checksum value
3346 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3347 * then add in skb->csum (checksum from csum_start to end of packet).
3348 * skb->csum and csum_start are then updated to reflect the checksum of the
3349 * resultant packet starting from the transport header-- the resultant checksum
3350 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3351 * header.
3352 */
3353 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3354 {
3355 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3356 skb_transport_offset(skb);
3357 __u16 csum;
3358
3359 csum = csum_fold(csum_partial(skb_transport_header(skb),
3360 plen, skb->csum));
3361 skb->csum = res;
3362 SKB_GSO_CB(skb)->csum_start -= plen;
3363
3364 return csum;
3365 }
3366
3367 static inline bool skb_is_gso(const struct sk_buff *skb)
3368 {
3369 return skb_shinfo(skb)->gso_size;
3370 }
3371
3372 /* Note: Should be called only if skb_is_gso(skb) is true */
3373 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3374 {
3375 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3376 }
3377
3378 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3379
3380 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3381 {
3382 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3383 * wanted then gso_type will be set. */
3384 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3385
3386 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3387 unlikely(shinfo->gso_type == 0)) {
3388 __skb_warn_lro_forwarding(skb);
3389 return true;
3390 }
3391 return false;
3392 }
3393
3394 static inline void skb_forward_csum(struct sk_buff *skb)
3395 {
3396 /* Unfortunately we don't support this one. Any brave souls? */
3397 if (skb->ip_summed == CHECKSUM_COMPLETE)
3398 skb->ip_summed = CHECKSUM_NONE;
3399 }
3400
3401 /**
3402 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3403 * @skb: skb to check
3404 *
3405 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3406 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3407 * use this helper, to document places where we make this assertion.
3408 */
3409 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3410 {
3411 #ifdef DEBUG
3412 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3413 #endif
3414 }
3415
3416 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3417
3418 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3419 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3420 unsigned int transport_len,
3421 __sum16(*skb_chkf)(struct sk_buff *skb));
3422
3423 /**
3424 * skb_head_is_locked - Determine if the skb->head is locked down
3425 * @skb: skb to check
3426 *
3427 * The head on skbs build around a head frag can be removed if they are
3428 * not cloned. This function returns true if the skb head is locked down
3429 * due to either being allocated via kmalloc, or by being a clone with
3430 * multiple references to the head.
3431 */
3432 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3433 {
3434 return !skb->head_frag || skb_cloned(skb);
3435 }
3436
3437 /**
3438 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3439 *
3440 * @skb: GSO skb
3441 *
3442 * skb_gso_network_seglen is used to determine the real size of the
3443 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3444 *
3445 * The MAC/L2 header is not accounted for.
3446 */
3447 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3448 {
3449 unsigned int hdr_len = skb_transport_header(skb) -
3450 skb_network_header(skb);
3451 return hdr_len + skb_gso_transport_seglen(skb);
3452 }
3453 #endif /* __KERNEL__ */
3454 #endif /* _LINUX_SKBUFF_H */