]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/linux/skbuff.h
Merge tag 'trace-3.16-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[mirror_ubuntu-artful-kernel.git] / include / linux / skbuff.h
1 /*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
16
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23
24 #include <linux/atomic.h>
25 #include <asm/types.h>
26 #include <linux/spinlock.h>
27 #include <linux/net.h>
28 #include <linux/textsearch.h>
29 #include <net/checksum.h>
30 #include <linux/rcupdate.h>
31 #include <linux/dmaengine.h>
32 #include <linux/hrtimer.h>
33 #include <linux/dma-mapping.h>
34 #include <linux/netdev_features.h>
35 #include <linux/sched.h>
36 #include <net/flow_keys.h>
37
38 /* A. Checksumming of received packets by device.
39 *
40 * CHECKSUM_NONE:
41 *
42 * Device failed to checksum this packet e.g. due to lack of capabilities.
43 * The packet contains full (though not verified) checksum in packet but
44 * not in skb->csum. Thus, skb->csum is undefined in this case.
45 *
46 * CHECKSUM_UNNECESSARY:
47 *
48 * The hardware you're dealing with doesn't calculate the full checksum
49 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
50 * for specific protocols e.g. TCP/UDP/SCTP, then, for such packets it will
51 * set CHECKSUM_UNNECESSARY if their checksums are okay. skb->csum is still
52 * undefined in this case though. It is a bad option, but, unfortunately,
53 * nowadays most vendors do this. Apparently with the secret goal to sell
54 * you new devices, when you will add new protocol to your host, f.e. IPv6 8)
55 *
56 * CHECKSUM_COMPLETE:
57 *
58 * This is the most generic way. The device supplied checksum of the _whole_
59 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
60 * hardware doesn't need to parse L3/L4 headers to implement this.
61 *
62 * Note: Even if device supports only some protocols, but is able to produce
63 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
64 *
65 * CHECKSUM_PARTIAL:
66 *
67 * This is identical to the case for output below. This may occur on a packet
68 * received directly from another Linux OS, e.g., a virtualized Linux kernel
69 * on the same host. The packet can be treated in the same way as
70 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the
71 * checksum must be filled in by the OS or the hardware.
72 *
73 * B. Checksumming on output.
74 *
75 * CHECKSUM_NONE:
76 *
77 * The skb was already checksummed by the protocol, or a checksum is not
78 * required.
79 *
80 * CHECKSUM_PARTIAL:
81 *
82 * The device is required to checksum the packet as seen by hard_start_xmit()
83 * from skb->csum_start up to the end, and to record/write the checksum at
84 * offset skb->csum_start + skb->csum_offset.
85 *
86 * The device must show its capabilities in dev->features, set up at device
87 * setup time, e.g. netdev_features.h:
88 *
89 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
90 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
91 * IPv4. Sigh. Vendors like this way for an unknown reason.
92 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
93 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
94 * NETIF_F_... - Well, you get the picture.
95 *
96 * CHECKSUM_UNNECESSARY:
97 *
98 * Normally, the device will do per protocol specific checksumming. Protocol
99 * implementations that do not want the NIC to perform the checksum
100 * calculation should use this flag in their outgoing skbs.
101 *
102 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
103 * offload. Correspondingly, the FCoE protocol driver
104 * stack should use CHECKSUM_UNNECESSARY.
105 *
106 * Any questions? No questions, good. --ANK
107 */
108
109 /* Don't change this without changing skb_csum_unnecessary! */
110 #define CHECKSUM_NONE 0
111 #define CHECKSUM_UNNECESSARY 1
112 #define CHECKSUM_COMPLETE 2
113 #define CHECKSUM_PARTIAL 3
114
115 #define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \
116 ~(SMP_CACHE_BYTES - 1))
117 #define SKB_WITH_OVERHEAD(X) \
118 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
119 #define SKB_MAX_ORDER(X, ORDER) \
120 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
121 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
122 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
123
124 /* return minimum truesize of one skb containing X bytes of data */
125 #define SKB_TRUESIZE(X) ((X) + \
126 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
127 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
128
129 struct net_device;
130 struct scatterlist;
131 struct pipe_inode_info;
132
133 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
134 struct nf_conntrack {
135 atomic_t use;
136 };
137 #endif
138
139 #ifdef CONFIG_BRIDGE_NETFILTER
140 struct nf_bridge_info {
141 atomic_t use;
142 unsigned int mask;
143 struct net_device *physindev;
144 struct net_device *physoutdev;
145 unsigned long data[32 / sizeof(unsigned long)];
146 };
147 #endif
148
149 struct sk_buff_head {
150 /* These two members must be first. */
151 struct sk_buff *next;
152 struct sk_buff *prev;
153
154 __u32 qlen;
155 spinlock_t lock;
156 };
157
158 struct sk_buff;
159
160 /* To allow 64K frame to be packed as single skb without frag_list we
161 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
162 * buffers which do not start on a page boundary.
163 *
164 * Since GRO uses frags we allocate at least 16 regardless of page
165 * size.
166 */
167 #if (65536/PAGE_SIZE + 1) < 16
168 #define MAX_SKB_FRAGS 16UL
169 #else
170 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
171 #endif
172
173 typedef struct skb_frag_struct skb_frag_t;
174
175 struct skb_frag_struct {
176 struct {
177 struct page *p;
178 } page;
179 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
180 __u32 page_offset;
181 __u32 size;
182 #else
183 __u16 page_offset;
184 __u16 size;
185 #endif
186 };
187
188 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
189 {
190 return frag->size;
191 }
192
193 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
194 {
195 frag->size = size;
196 }
197
198 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
199 {
200 frag->size += delta;
201 }
202
203 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
204 {
205 frag->size -= delta;
206 }
207
208 #define HAVE_HW_TIME_STAMP
209
210 /**
211 * struct skb_shared_hwtstamps - hardware time stamps
212 * @hwtstamp: hardware time stamp transformed into duration
213 * since arbitrary point in time
214 * @syststamp: hwtstamp transformed to system time base
215 *
216 * Software time stamps generated by ktime_get_real() are stored in
217 * skb->tstamp. The relation between the different kinds of time
218 * stamps is as follows:
219 *
220 * syststamp and tstamp can be compared against each other in
221 * arbitrary combinations. The accuracy of a
222 * syststamp/tstamp/"syststamp from other device" comparison is
223 * limited by the accuracy of the transformation into system time
224 * base. This depends on the device driver and its underlying
225 * hardware.
226 *
227 * hwtstamps can only be compared against other hwtstamps from
228 * the same device.
229 *
230 * This structure is attached to packets as part of the
231 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
232 */
233 struct skb_shared_hwtstamps {
234 ktime_t hwtstamp;
235 ktime_t syststamp;
236 };
237
238 /* Definitions for tx_flags in struct skb_shared_info */
239 enum {
240 /* generate hardware time stamp */
241 SKBTX_HW_TSTAMP = 1 << 0,
242
243 /* generate software time stamp */
244 SKBTX_SW_TSTAMP = 1 << 1,
245
246 /* device driver is going to provide hardware time stamp */
247 SKBTX_IN_PROGRESS = 1 << 2,
248
249 /* device driver supports TX zero-copy buffers */
250 SKBTX_DEV_ZEROCOPY = 1 << 3,
251
252 /* generate wifi status information (where possible) */
253 SKBTX_WIFI_STATUS = 1 << 4,
254
255 /* This indicates at least one fragment might be overwritten
256 * (as in vmsplice(), sendfile() ...)
257 * If we need to compute a TX checksum, we'll need to copy
258 * all frags to avoid possible bad checksum
259 */
260 SKBTX_SHARED_FRAG = 1 << 5,
261 };
262
263 /*
264 * The callback notifies userspace to release buffers when skb DMA is done in
265 * lower device, the skb last reference should be 0 when calling this.
266 * The zerocopy_success argument is true if zero copy transmit occurred,
267 * false on data copy or out of memory error caused by data copy attempt.
268 * The ctx field is used to track device context.
269 * The desc field is used to track userspace buffer index.
270 */
271 struct ubuf_info {
272 void (*callback)(struct ubuf_info *, bool zerocopy_success);
273 void *ctx;
274 unsigned long desc;
275 };
276
277 /* This data is invariant across clones and lives at
278 * the end of the header data, ie. at skb->end.
279 */
280 struct skb_shared_info {
281 unsigned char nr_frags;
282 __u8 tx_flags;
283 unsigned short gso_size;
284 /* Warning: this field is not always filled in (UFO)! */
285 unsigned short gso_segs;
286 unsigned short gso_type;
287 struct sk_buff *frag_list;
288 struct skb_shared_hwtstamps hwtstamps;
289 __be32 ip6_frag_id;
290
291 /*
292 * Warning : all fields before dataref are cleared in __alloc_skb()
293 */
294 atomic_t dataref;
295
296 /* Intermediate layers must ensure that destructor_arg
297 * remains valid until skb destructor */
298 void * destructor_arg;
299
300 /* must be last field, see pskb_expand_head() */
301 skb_frag_t frags[MAX_SKB_FRAGS];
302 };
303
304 /* We divide dataref into two halves. The higher 16 bits hold references
305 * to the payload part of skb->data. The lower 16 bits hold references to
306 * the entire skb->data. A clone of a headerless skb holds the length of
307 * the header in skb->hdr_len.
308 *
309 * All users must obey the rule that the skb->data reference count must be
310 * greater than or equal to the payload reference count.
311 *
312 * Holding a reference to the payload part means that the user does not
313 * care about modifications to the header part of skb->data.
314 */
315 #define SKB_DATAREF_SHIFT 16
316 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
317
318
319 enum {
320 SKB_FCLONE_UNAVAILABLE,
321 SKB_FCLONE_ORIG,
322 SKB_FCLONE_CLONE,
323 };
324
325 enum {
326 SKB_GSO_TCPV4 = 1 << 0,
327 SKB_GSO_UDP = 1 << 1,
328
329 /* This indicates the skb is from an untrusted source. */
330 SKB_GSO_DODGY = 1 << 2,
331
332 /* This indicates the tcp segment has CWR set. */
333 SKB_GSO_TCP_ECN = 1 << 3,
334
335 SKB_GSO_TCPV6 = 1 << 4,
336
337 SKB_GSO_FCOE = 1 << 5,
338
339 SKB_GSO_GRE = 1 << 6,
340
341 SKB_GSO_IPIP = 1 << 7,
342
343 SKB_GSO_SIT = 1 << 8,
344
345 SKB_GSO_UDP_TUNNEL = 1 << 9,
346
347 SKB_GSO_MPLS = 1 << 10,
348
349 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
350
351 SKB_GSO_GRE_CSUM = 1 << 12,
352 };
353
354 #if BITS_PER_LONG > 32
355 #define NET_SKBUFF_DATA_USES_OFFSET 1
356 #endif
357
358 #ifdef NET_SKBUFF_DATA_USES_OFFSET
359 typedef unsigned int sk_buff_data_t;
360 #else
361 typedef unsigned char *sk_buff_data_t;
362 #endif
363
364 /**
365 * struct skb_mstamp - multi resolution time stamps
366 * @stamp_us: timestamp in us resolution
367 * @stamp_jiffies: timestamp in jiffies
368 */
369 struct skb_mstamp {
370 union {
371 u64 v64;
372 struct {
373 u32 stamp_us;
374 u32 stamp_jiffies;
375 };
376 };
377 };
378
379 /**
380 * skb_mstamp_get - get current timestamp
381 * @cl: place to store timestamps
382 */
383 static inline void skb_mstamp_get(struct skb_mstamp *cl)
384 {
385 u64 val = local_clock();
386
387 do_div(val, NSEC_PER_USEC);
388 cl->stamp_us = (u32)val;
389 cl->stamp_jiffies = (u32)jiffies;
390 }
391
392 /**
393 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
394 * @t1: pointer to newest sample
395 * @t0: pointer to oldest sample
396 */
397 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
398 const struct skb_mstamp *t0)
399 {
400 s32 delta_us = t1->stamp_us - t0->stamp_us;
401 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
402
403 /* If delta_us is negative, this might be because interval is too big,
404 * or local_clock() drift is too big : fallback using jiffies.
405 */
406 if (delta_us <= 0 ||
407 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
408
409 delta_us = jiffies_to_usecs(delta_jiffies);
410
411 return delta_us;
412 }
413
414
415 /**
416 * struct sk_buff - socket buffer
417 * @next: Next buffer in list
418 * @prev: Previous buffer in list
419 * @tstamp: Time we arrived/left
420 * @sk: Socket we are owned by
421 * @dev: Device we arrived on/are leaving by
422 * @cb: Control buffer. Free for use by every layer. Put private vars here
423 * @_skb_refdst: destination entry (with norefcount bit)
424 * @sp: the security path, used for xfrm
425 * @len: Length of actual data
426 * @data_len: Data length
427 * @mac_len: Length of link layer header
428 * @hdr_len: writable header length of cloned skb
429 * @csum: Checksum (must include start/offset pair)
430 * @csum_start: Offset from skb->head where checksumming should start
431 * @csum_offset: Offset from csum_start where checksum should be stored
432 * @priority: Packet queueing priority
433 * @ignore_df: allow local fragmentation
434 * @cloned: Head may be cloned (check refcnt to be sure)
435 * @ip_summed: Driver fed us an IP checksum
436 * @nohdr: Payload reference only, must not modify header
437 * @nfctinfo: Relationship of this skb to the connection
438 * @pkt_type: Packet class
439 * @fclone: skbuff clone status
440 * @ipvs_property: skbuff is owned by ipvs
441 * @peeked: this packet has been seen already, so stats have been
442 * done for it, don't do them again
443 * @nf_trace: netfilter packet trace flag
444 * @protocol: Packet protocol from driver
445 * @destructor: Destruct function
446 * @nfct: Associated connection, if any
447 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
448 * @skb_iif: ifindex of device we arrived on
449 * @tc_index: Traffic control index
450 * @tc_verd: traffic control verdict
451 * @hash: the packet hash
452 * @queue_mapping: Queue mapping for multiqueue devices
453 * @ndisc_nodetype: router type (from link layer)
454 * @ooo_okay: allow the mapping of a socket to a queue to be changed
455 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
456 * ports.
457 * @wifi_acked_valid: wifi_acked was set
458 * @wifi_acked: whether frame was acked on wifi or not
459 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
460 * @dma_cookie: a cookie to one of several possible DMA operations
461 * done by skb DMA functions
462 * @napi_id: id of the NAPI struct this skb came from
463 * @secmark: security marking
464 * @mark: Generic packet mark
465 * @dropcount: total number of sk_receive_queue overflows
466 * @vlan_proto: vlan encapsulation protocol
467 * @vlan_tci: vlan tag control information
468 * @inner_protocol: Protocol (encapsulation)
469 * @inner_transport_header: Inner transport layer header (encapsulation)
470 * @inner_network_header: Network layer header (encapsulation)
471 * @inner_mac_header: Link layer header (encapsulation)
472 * @transport_header: Transport layer header
473 * @network_header: Network layer header
474 * @mac_header: Link layer header
475 * @tail: Tail pointer
476 * @end: End pointer
477 * @head: Head of buffer
478 * @data: Data head pointer
479 * @truesize: Buffer size
480 * @users: User count - see {datagram,tcp}.c
481 */
482
483 struct sk_buff {
484 /* These two members must be first. */
485 struct sk_buff *next;
486 struct sk_buff *prev;
487
488 union {
489 ktime_t tstamp;
490 struct skb_mstamp skb_mstamp;
491 };
492
493 struct sock *sk;
494 struct net_device *dev;
495
496 /*
497 * This is the control buffer. It is free to use for every
498 * layer. Please put your private variables there. If you
499 * want to keep them across layers you have to do a skb_clone()
500 * first. This is owned by whoever has the skb queued ATM.
501 */
502 char cb[48] __aligned(8);
503
504 unsigned long _skb_refdst;
505 #ifdef CONFIG_XFRM
506 struct sec_path *sp;
507 #endif
508 unsigned int len,
509 data_len;
510 __u16 mac_len,
511 hdr_len;
512 union {
513 __wsum csum;
514 struct {
515 __u16 csum_start;
516 __u16 csum_offset;
517 };
518 };
519 __u32 priority;
520 kmemcheck_bitfield_begin(flags1);
521 __u8 ignore_df:1,
522 cloned:1,
523 ip_summed:2,
524 nohdr:1,
525 nfctinfo:3;
526 __u8 pkt_type:3,
527 fclone:2,
528 ipvs_property:1,
529 peeked:1,
530 nf_trace:1;
531 kmemcheck_bitfield_end(flags1);
532 __be16 protocol;
533
534 void (*destructor)(struct sk_buff *skb);
535 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
536 struct nf_conntrack *nfct;
537 #endif
538 #ifdef CONFIG_BRIDGE_NETFILTER
539 struct nf_bridge_info *nf_bridge;
540 #endif
541
542 int skb_iif;
543
544 __u32 hash;
545
546 __be16 vlan_proto;
547 __u16 vlan_tci;
548
549 #ifdef CONFIG_NET_SCHED
550 __u16 tc_index; /* traffic control index */
551 #ifdef CONFIG_NET_CLS_ACT
552 __u16 tc_verd; /* traffic control verdict */
553 #endif
554 #endif
555
556 __u16 queue_mapping;
557 kmemcheck_bitfield_begin(flags2);
558 #ifdef CONFIG_IPV6_NDISC_NODETYPE
559 __u8 ndisc_nodetype:2;
560 #endif
561 __u8 pfmemalloc:1;
562 __u8 ooo_okay:1;
563 __u8 l4_hash:1;
564 __u8 wifi_acked_valid:1;
565 __u8 wifi_acked:1;
566 __u8 no_fcs:1;
567 __u8 head_frag:1;
568 /* Encapsulation protocol and NIC drivers should use
569 * this flag to indicate to each other if the skb contains
570 * encapsulated packet or not and maybe use the inner packet
571 * headers if needed
572 */
573 __u8 encapsulation:1;
574 __u8 encap_hdr_csum:1;
575 __u8 csum_valid:1;
576 __u8 csum_complete_sw:1;
577 /* 3/5 bit hole (depending on ndisc_nodetype presence) */
578 kmemcheck_bitfield_end(flags2);
579
580 #if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
581 union {
582 unsigned int napi_id;
583 dma_cookie_t dma_cookie;
584 };
585 #endif
586 #ifdef CONFIG_NETWORK_SECMARK
587 __u32 secmark;
588 #endif
589 union {
590 __u32 mark;
591 __u32 dropcount;
592 __u32 reserved_tailroom;
593 };
594
595 __be16 inner_protocol;
596 __u16 inner_transport_header;
597 __u16 inner_network_header;
598 __u16 inner_mac_header;
599 __u16 transport_header;
600 __u16 network_header;
601 __u16 mac_header;
602 /* These elements must be at the end, see alloc_skb() for details. */
603 sk_buff_data_t tail;
604 sk_buff_data_t end;
605 unsigned char *head,
606 *data;
607 unsigned int truesize;
608 atomic_t users;
609 };
610
611 #ifdef __KERNEL__
612 /*
613 * Handling routines are only of interest to the kernel
614 */
615 #include <linux/slab.h>
616
617
618 #define SKB_ALLOC_FCLONE 0x01
619 #define SKB_ALLOC_RX 0x02
620
621 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
622 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
623 {
624 return unlikely(skb->pfmemalloc);
625 }
626
627 /*
628 * skb might have a dst pointer attached, refcounted or not.
629 * _skb_refdst low order bit is set if refcount was _not_ taken
630 */
631 #define SKB_DST_NOREF 1UL
632 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
633
634 /**
635 * skb_dst - returns skb dst_entry
636 * @skb: buffer
637 *
638 * Returns skb dst_entry, regardless of reference taken or not.
639 */
640 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
641 {
642 /* If refdst was not refcounted, check we still are in a
643 * rcu_read_lock section
644 */
645 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
646 !rcu_read_lock_held() &&
647 !rcu_read_lock_bh_held());
648 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
649 }
650
651 /**
652 * skb_dst_set - sets skb dst
653 * @skb: buffer
654 * @dst: dst entry
655 *
656 * Sets skb dst, assuming a reference was taken on dst and should
657 * be released by skb_dst_drop()
658 */
659 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
660 {
661 skb->_skb_refdst = (unsigned long)dst;
662 }
663
664 void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
665 bool force);
666
667 /**
668 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
669 * @skb: buffer
670 * @dst: dst entry
671 *
672 * Sets skb dst, assuming a reference was not taken on dst.
673 * If dst entry is cached, we do not take reference and dst_release
674 * will be avoided by refdst_drop. If dst entry is not cached, we take
675 * reference, so that last dst_release can destroy the dst immediately.
676 */
677 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
678 {
679 __skb_dst_set_noref(skb, dst, false);
680 }
681
682 /**
683 * skb_dst_set_noref_force - sets skb dst, without taking reference
684 * @skb: buffer
685 * @dst: dst entry
686 *
687 * Sets skb dst, assuming a reference was not taken on dst.
688 * No reference is taken and no dst_release will be called. While for
689 * cached dsts deferred reclaim is a basic feature, for entries that are
690 * not cached it is caller's job to guarantee that last dst_release for
691 * provided dst happens when nobody uses it, eg. after a RCU grace period.
692 */
693 static inline void skb_dst_set_noref_force(struct sk_buff *skb,
694 struct dst_entry *dst)
695 {
696 __skb_dst_set_noref(skb, dst, true);
697 }
698
699 /**
700 * skb_dst_is_noref - Test if skb dst isn't refcounted
701 * @skb: buffer
702 */
703 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
704 {
705 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
706 }
707
708 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
709 {
710 return (struct rtable *)skb_dst(skb);
711 }
712
713 void kfree_skb(struct sk_buff *skb);
714 void kfree_skb_list(struct sk_buff *segs);
715 void skb_tx_error(struct sk_buff *skb);
716 void consume_skb(struct sk_buff *skb);
717 void __kfree_skb(struct sk_buff *skb);
718 extern struct kmem_cache *skbuff_head_cache;
719
720 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
721 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
722 bool *fragstolen, int *delta_truesize);
723
724 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
725 int node);
726 struct sk_buff *build_skb(void *data, unsigned int frag_size);
727 static inline struct sk_buff *alloc_skb(unsigned int size,
728 gfp_t priority)
729 {
730 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
731 }
732
733 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
734 gfp_t priority)
735 {
736 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
737 }
738
739 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
740 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
741 {
742 return __alloc_skb_head(priority, -1);
743 }
744
745 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
746 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
747 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
748 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
749 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
750 gfp_t gfp_mask, bool fclone);
751 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
752 gfp_t gfp_mask)
753 {
754 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
755 }
756
757 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
758 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
759 unsigned int headroom);
760 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
761 int newtailroom, gfp_t priority);
762 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
763 int offset, int len);
764 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
765 int len);
766 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
767 int skb_pad(struct sk_buff *skb, int pad);
768 #define dev_kfree_skb(a) consume_skb(a)
769
770 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
771 int getfrag(void *from, char *to, int offset,
772 int len, int odd, struct sk_buff *skb),
773 void *from, int length);
774
775 struct skb_seq_state {
776 __u32 lower_offset;
777 __u32 upper_offset;
778 __u32 frag_idx;
779 __u32 stepped_offset;
780 struct sk_buff *root_skb;
781 struct sk_buff *cur_skb;
782 __u8 *frag_data;
783 };
784
785 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
786 unsigned int to, struct skb_seq_state *st);
787 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
788 struct skb_seq_state *st);
789 void skb_abort_seq_read(struct skb_seq_state *st);
790
791 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
792 unsigned int to, struct ts_config *config,
793 struct ts_state *state);
794
795 /*
796 * Packet hash types specify the type of hash in skb_set_hash.
797 *
798 * Hash types refer to the protocol layer addresses which are used to
799 * construct a packet's hash. The hashes are used to differentiate or identify
800 * flows of the protocol layer for the hash type. Hash types are either
801 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
802 *
803 * Properties of hashes:
804 *
805 * 1) Two packets in different flows have different hash values
806 * 2) Two packets in the same flow should have the same hash value
807 *
808 * A hash at a higher layer is considered to be more specific. A driver should
809 * set the most specific hash possible.
810 *
811 * A driver cannot indicate a more specific hash than the layer at which a hash
812 * was computed. For instance an L3 hash cannot be set as an L4 hash.
813 *
814 * A driver may indicate a hash level which is less specific than the
815 * actual layer the hash was computed on. For instance, a hash computed
816 * at L4 may be considered an L3 hash. This should only be done if the
817 * driver can't unambiguously determine that the HW computed the hash at
818 * the higher layer. Note that the "should" in the second property above
819 * permits this.
820 */
821 enum pkt_hash_types {
822 PKT_HASH_TYPE_NONE, /* Undefined type */
823 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
824 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
825 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
826 };
827
828 static inline void
829 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
830 {
831 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
832 skb->hash = hash;
833 }
834
835 void __skb_get_hash(struct sk_buff *skb);
836 static inline __u32 skb_get_hash(struct sk_buff *skb)
837 {
838 if (!skb->l4_hash)
839 __skb_get_hash(skb);
840
841 return skb->hash;
842 }
843
844 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
845 {
846 return skb->hash;
847 }
848
849 static inline void skb_clear_hash(struct sk_buff *skb)
850 {
851 skb->hash = 0;
852 skb->l4_hash = 0;
853 }
854
855 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
856 {
857 if (!skb->l4_hash)
858 skb_clear_hash(skb);
859 }
860
861 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
862 {
863 to->hash = from->hash;
864 to->l4_hash = from->l4_hash;
865 };
866
867 #ifdef NET_SKBUFF_DATA_USES_OFFSET
868 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
869 {
870 return skb->head + skb->end;
871 }
872
873 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
874 {
875 return skb->end;
876 }
877 #else
878 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
879 {
880 return skb->end;
881 }
882
883 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
884 {
885 return skb->end - skb->head;
886 }
887 #endif
888
889 /* Internal */
890 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
891
892 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
893 {
894 return &skb_shinfo(skb)->hwtstamps;
895 }
896
897 /**
898 * skb_queue_empty - check if a queue is empty
899 * @list: queue head
900 *
901 * Returns true if the queue is empty, false otherwise.
902 */
903 static inline int skb_queue_empty(const struct sk_buff_head *list)
904 {
905 return list->next == (const struct sk_buff *) list;
906 }
907
908 /**
909 * skb_queue_is_last - check if skb is the last entry in the queue
910 * @list: queue head
911 * @skb: buffer
912 *
913 * Returns true if @skb is the last buffer on the list.
914 */
915 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
916 const struct sk_buff *skb)
917 {
918 return skb->next == (const struct sk_buff *) list;
919 }
920
921 /**
922 * skb_queue_is_first - check if skb is the first entry in the queue
923 * @list: queue head
924 * @skb: buffer
925 *
926 * Returns true if @skb is the first buffer on the list.
927 */
928 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
929 const struct sk_buff *skb)
930 {
931 return skb->prev == (const struct sk_buff *) list;
932 }
933
934 /**
935 * skb_queue_next - return the next packet in the queue
936 * @list: queue head
937 * @skb: current buffer
938 *
939 * Return the next packet in @list after @skb. It is only valid to
940 * call this if skb_queue_is_last() evaluates to false.
941 */
942 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
943 const struct sk_buff *skb)
944 {
945 /* This BUG_ON may seem severe, but if we just return then we
946 * are going to dereference garbage.
947 */
948 BUG_ON(skb_queue_is_last(list, skb));
949 return skb->next;
950 }
951
952 /**
953 * skb_queue_prev - return the prev packet in the queue
954 * @list: queue head
955 * @skb: current buffer
956 *
957 * Return the prev packet in @list before @skb. It is only valid to
958 * call this if skb_queue_is_first() evaluates to false.
959 */
960 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
961 const struct sk_buff *skb)
962 {
963 /* This BUG_ON may seem severe, but if we just return then we
964 * are going to dereference garbage.
965 */
966 BUG_ON(skb_queue_is_first(list, skb));
967 return skb->prev;
968 }
969
970 /**
971 * skb_get - reference buffer
972 * @skb: buffer to reference
973 *
974 * Makes another reference to a socket buffer and returns a pointer
975 * to the buffer.
976 */
977 static inline struct sk_buff *skb_get(struct sk_buff *skb)
978 {
979 atomic_inc(&skb->users);
980 return skb;
981 }
982
983 /*
984 * If users == 1, we are the only owner and are can avoid redundant
985 * atomic change.
986 */
987
988 /**
989 * skb_cloned - is the buffer a clone
990 * @skb: buffer to check
991 *
992 * Returns true if the buffer was generated with skb_clone() and is
993 * one of multiple shared copies of the buffer. Cloned buffers are
994 * shared data so must not be written to under normal circumstances.
995 */
996 static inline int skb_cloned(const struct sk_buff *skb)
997 {
998 return skb->cloned &&
999 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1000 }
1001
1002 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1003 {
1004 might_sleep_if(pri & __GFP_WAIT);
1005
1006 if (skb_cloned(skb))
1007 return pskb_expand_head(skb, 0, 0, pri);
1008
1009 return 0;
1010 }
1011
1012 /**
1013 * skb_header_cloned - is the header a clone
1014 * @skb: buffer to check
1015 *
1016 * Returns true if modifying the header part of the buffer requires
1017 * the data to be copied.
1018 */
1019 static inline int skb_header_cloned(const struct sk_buff *skb)
1020 {
1021 int dataref;
1022
1023 if (!skb->cloned)
1024 return 0;
1025
1026 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1027 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1028 return dataref != 1;
1029 }
1030
1031 /**
1032 * skb_header_release - release reference to header
1033 * @skb: buffer to operate on
1034 *
1035 * Drop a reference to the header part of the buffer. This is done
1036 * by acquiring a payload reference. You must not read from the header
1037 * part of skb->data after this.
1038 */
1039 static inline void skb_header_release(struct sk_buff *skb)
1040 {
1041 BUG_ON(skb->nohdr);
1042 skb->nohdr = 1;
1043 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1044 }
1045
1046 /**
1047 * skb_shared - is the buffer shared
1048 * @skb: buffer to check
1049 *
1050 * Returns true if more than one person has a reference to this
1051 * buffer.
1052 */
1053 static inline int skb_shared(const struct sk_buff *skb)
1054 {
1055 return atomic_read(&skb->users) != 1;
1056 }
1057
1058 /**
1059 * skb_share_check - check if buffer is shared and if so clone it
1060 * @skb: buffer to check
1061 * @pri: priority for memory allocation
1062 *
1063 * If the buffer is shared the buffer is cloned and the old copy
1064 * drops a reference. A new clone with a single reference is returned.
1065 * If the buffer is not shared the original buffer is returned. When
1066 * being called from interrupt status or with spinlocks held pri must
1067 * be GFP_ATOMIC.
1068 *
1069 * NULL is returned on a memory allocation failure.
1070 */
1071 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1072 {
1073 might_sleep_if(pri & __GFP_WAIT);
1074 if (skb_shared(skb)) {
1075 struct sk_buff *nskb = skb_clone(skb, pri);
1076
1077 if (likely(nskb))
1078 consume_skb(skb);
1079 else
1080 kfree_skb(skb);
1081 skb = nskb;
1082 }
1083 return skb;
1084 }
1085
1086 /*
1087 * Copy shared buffers into a new sk_buff. We effectively do COW on
1088 * packets to handle cases where we have a local reader and forward
1089 * and a couple of other messy ones. The normal one is tcpdumping
1090 * a packet thats being forwarded.
1091 */
1092
1093 /**
1094 * skb_unshare - make a copy of a shared buffer
1095 * @skb: buffer to check
1096 * @pri: priority for memory allocation
1097 *
1098 * If the socket buffer is a clone then this function creates a new
1099 * copy of the data, drops a reference count on the old copy and returns
1100 * the new copy with the reference count at 1. If the buffer is not a clone
1101 * the original buffer is returned. When called with a spinlock held or
1102 * from interrupt state @pri must be %GFP_ATOMIC
1103 *
1104 * %NULL is returned on a memory allocation failure.
1105 */
1106 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1107 gfp_t pri)
1108 {
1109 might_sleep_if(pri & __GFP_WAIT);
1110 if (skb_cloned(skb)) {
1111 struct sk_buff *nskb = skb_copy(skb, pri);
1112 kfree_skb(skb); /* Free our shared copy */
1113 skb = nskb;
1114 }
1115 return skb;
1116 }
1117
1118 /**
1119 * skb_peek - peek at the head of an &sk_buff_head
1120 * @list_: list to peek at
1121 *
1122 * Peek an &sk_buff. Unlike most other operations you _MUST_
1123 * be careful with this one. A peek leaves the buffer on the
1124 * list and someone else may run off with it. You must hold
1125 * the appropriate locks or have a private queue to do this.
1126 *
1127 * Returns %NULL for an empty list or a pointer to the head element.
1128 * The reference count is not incremented and the reference is therefore
1129 * volatile. Use with caution.
1130 */
1131 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1132 {
1133 struct sk_buff *skb = list_->next;
1134
1135 if (skb == (struct sk_buff *)list_)
1136 skb = NULL;
1137 return skb;
1138 }
1139
1140 /**
1141 * skb_peek_next - peek skb following the given one from a queue
1142 * @skb: skb to start from
1143 * @list_: list to peek at
1144 *
1145 * Returns %NULL when the end of the list is met or a pointer to the
1146 * next element. The reference count is not incremented and the
1147 * reference is therefore volatile. Use with caution.
1148 */
1149 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1150 const struct sk_buff_head *list_)
1151 {
1152 struct sk_buff *next = skb->next;
1153
1154 if (next == (struct sk_buff *)list_)
1155 next = NULL;
1156 return next;
1157 }
1158
1159 /**
1160 * skb_peek_tail - peek at the tail of an &sk_buff_head
1161 * @list_: list to peek at
1162 *
1163 * Peek an &sk_buff. Unlike most other operations you _MUST_
1164 * be careful with this one. A peek leaves the buffer on the
1165 * list and someone else may run off with it. You must hold
1166 * the appropriate locks or have a private queue to do this.
1167 *
1168 * Returns %NULL for an empty list or a pointer to the tail element.
1169 * The reference count is not incremented and the reference is therefore
1170 * volatile. Use with caution.
1171 */
1172 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1173 {
1174 struct sk_buff *skb = list_->prev;
1175
1176 if (skb == (struct sk_buff *)list_)
1177 skb = NULL;
1178 return skb;
1179
1180 }
1181
1182 /**
1183 * skb_queue_len - get queue length
1184 * @list_: list to measure
1185 *
1186 * Return the length of an &sk_buff queue.
1187 */
1188 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1189 {
1190 return list_->qlen;
1191 }
1192
1193 /**
1194 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1195 * @list: queue to initialize
1196 *
1197 * This initializes only the list and queue length aspects of
1198 * an sk_buff_head object. This allows to initialize the list
1199 * aspects of an sk_buff_head without reinitializing things like
1200 * the spinlock. It can also be used for on-stack sk_buff_head
1201 * objects where the spinlock is known to not be used.
1202 */
1203 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1204 {
1205 list->prev = list->next = (struct sk_buff *)list;
1206 list->qlen = 0;
1207 }
1208
1209 /*
1210 * This function creates a split out lock class for each invocation;
1211 * this is needed for now since a whole lot of users of the skb-queue
1212 * infrastructure in drivers have different locking usage (in hardirq)
1213 * than the networking core (in softirq only). In the long run either the
1214 * network layer or drivers should need annotation to consolidate the
1215 * main types of usage into 3 classes.
1216 */
1217 static inline void skb_queue_head_init(struct sk_buff_head *list)
1218 {
1219 spin_lock_init(&list->lock);
1220 __skb_queue_head_init(list);
1221 }
1222
1223 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1224 struct lock_class_key *class)
1225 {
1226 skb_queue_head_init(list);
1227 lockdep_set_class(&list->lock, class);
1228 }
1229
1230 /*
1231 * Insert an sk_buff on a list.
1232 *
1233 * The "__skb_xxxx()" functions are the non-atomic ones that
1234 * can only be called with interrupts disabled.
1235 */
1236 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1237 struct sk_buff_head *list);
1238 static inline void __skb_insert(struct sk_buff *newsk,
1239 struct sk_buff *prev, struct sk_buff *next,
1240 struct sk_buff_head *list)
1241 {
1242 newsk->next = next;
1243 newsk->prev = prev;
1244 next->prev = prev->next = newsk;
1245 list->qlen++;
1246 }
1247
1248 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1249 struct sk_buff *prev,
1250 struct sk_buff *next)
1251 {
1252 struct sk_buff *first = list->next;
1253 struct sk_buff *last = list->prev;
1254
1255 first->prev = prev;
1256 prev->next = first;
1257
1258 last->next = next;
1259 next->prev = last;
1260 }
1261
1262 /**
1263 * skb_queue_splice - join two skb lists, this is designed for stacks
1264 * @list: the new list to add
1265 * @head: the place to add it in the first list
1266 */
1267 static inline void skb_queue_splice(const struct sk_buff_head *list,
1268 struct sk_buff_head *head)
1269 {
1270 if (!skb_queue_empty(list)) {
1271 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1272 head->qlen += list->qlen;
1273 }
1274 }
1275
1276 /**
1277 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1278 * @list: the new list to add
1279 * @head: the place to add it in the first list
1280 *
1281 * The list at @list is reinitialised
1282 */
1283 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1284 struct sk_buff_head *head)
1285 {
1286 if (!skb_queue_empty(list)) {
1287 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1288 head->qlen += list->qlen;
1289 __skb_queue_head_init(list);
1290 }
1291 }
1292
1293 /**
1294 * skb_queue_splice_tail - join two skb lists, each list being a queue
1295 * @list: the new list to add
1296 * @head: the place to add it in the first list
1297 */
1298 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1299 struct sk_buff_head *head)
1300 {
1301 if (!skb_queue_empty(list)) {
1302 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1303 head->qlen += list->qlen;
1304 }
1305 }
1306
1307 /**
1308 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1309 * @list: the new list to add
1310 * @head: the place to add it in the first list
1311 *
1312 * Each of the lists is a queue.
1313 * The list at @list is reinitialised
1314 */
1315 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1316 struct sk_buff_head *head)
1317 {
1318 if (!skb_queue_empty(list)) {
1319 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1320 head->qlen += list->qlen;
1321 __skb_queue_head_init(list);
1322 }
1323 }
1324
1325 /**
1326 * __skb_queue_after - queue a buffer at the list head
1327 * @list: list to use
1328 * @prev: place after this buffer
1329 * @newsk: buffer to queue
1330 *
1331 * Queue a buffer int the middle of a list. This function takes no locks
1332 * and you must therefore hold required locks before calling it.
1333 *
1334 * A buffer cannot be placed on two lists at the same time.
1335 */
1336 static inline void __skb_queue_after(struct sk_buff_head *list,
1337 struct sk_buff *prev,
1338 struct sk_buff *newsk)
1339 {
1340 __skb_insert(newsk, prev, prev->next, list);
1341 }
1342
1343 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1344 struct sk_buff_head *list);
1345
1346 static inline void __skb_queue_before(struct sk_buff_head *list,
1347 struct sk_buff *next,
1348 struct sk_buff *newsk)
1349 {
1350 __skb_insert(newsk, next->prev, next, list);
1351 }
1352
1353 /**
1354 * __skb_queue_head - queue a buffer at the list head
1355 * @list: list to use
1356 * @newsk: buffer to queue
1357 *
1358 * Queue a buffer at the start of a list. This function takes no locks
1359 * and you must therefore hold required locks before calling it.
1360 *
1361 * A buffer cannot be placed on two lists at the same time.
1362 */
1363 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1364 static inline void __skb_queue_head(struct sk_buff_head *list,
1365 struct sk_buff *newsk)
1366 {
1367 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1368 }
1369
1370 /**
1371 * __skb_queue_tail - queue a buffer at the list tail
1372 * @list: list to use
1373 * @newsk: buffer to queue
1374 *
1375 * Queue a buffer at the end of a list. This function takes no locks
1376 * and you must therefore hold required locks before calling it.
1377 *
1378 * A buffer cannot be placed on two lists at the same time.
1379 */
1380 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1381 static inline void __skb_queue_tail(struct sk_buff_head *list,
1382 struct sk_buff *newsk)
1383 {
1384 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1385 }
1386
1387 /*
1388 * remove sk_buff from list. _Must_ be called atomically, and with
1389 * the list known..
1390 */
1391 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1392 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1393 {
1394 struct sk_buff *next, *prev;
1395
1396 list->qlen--;
1397 next = skb->next;
1398 prev = skb->prev;
1399 skb->next = skb->prev = NULL;
1400 next->prev = prev;
1401 prev->next = next;
1402 }
1403
1404 /**
1405 * __skb_dequeue - remove from the head of the queue
1406 * @list: list to dequeue from
1407 *
1408 * Remove the head of the list. This function does not take any locks
1409 * so must be used with appropriate locks held only. The head item is
1410 * returned or %NULL if the list is empty.
1411 */
1412 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1413 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1414 {
1415 struct sk_buff *skb = skb_peek(list);
1416 if (skb)
1417 __skb_unlink(skb, list);
1418 return skb;
1419 }
1420
1421 /**
1422 * __skb_dequeue_tail - remove from the tail of the queue
1423 * @list: list to dequeue from
1424 *
1425 * Remove the tail of the list. This function does not take any locks
1426 * so must be used with appropriate locks held only. The tail item is
1427 * returned or %NULL if the list is empty.
1428 */
1429 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1430 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1431 {
1432 struct sk_buff *skb = skb_peek_tail(list);
1433 if (skb)
1434 __skb_unlink(skb, list);
1435 return skb;
1436 }
1437
1438
1439 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1440 {
1441 return skb->data_len;
1442 }
1443
1444 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1445 {
1446 return skb->len - skb->data_len;
1447 }
1448
1449 static inline int skb_pagelen(const struct sk_buff *skb)
1450 {
1451 int i, len = 0;
1452
1453 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1454 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1455 return len + skb_headlen(skb);
1456 }
1457
1458 /**
1459 * __skb_fill_page_desc - initialise a paged fragment in an skb
1460 * @skb: buffer containing fragment to be initialised
1461 * @i: paged fragment index to initialise
1462 * @page: the page to use for this fragment
1463 * @off: the offset to the data with @page
1464 * @size: the length of the data
1465 *
1466 * Initialises the @i'th fragment of @skb to point to &size bytes at
1467 * offset @off within @page.
1468 *
1469 * Does not take any additional reference on the fragment.
1470 */
1471 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1472 struct page *page, int off, int size)
1473 {
1474 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1475
1476 /*
1477 * Propagate page->pfmemalloc to the skb if we can. The problem is
1478 * that not all callers have unique ownership of the page. If
1479 * pfmemalloc is set, we check the mapping as a mapping implies
1480 * page->index is set (index and pfmemalloc share space).
1481 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1482 * do not lose pfmemalloc information as the pages would not be
1483 * allocated using __GFP_MEMALLOC.
1484 */
1485 frag->page.p = page;
1486 frag->page_offset = off;
1487 skb_frag_size_set(frag, size);
1488
1489 page = compound_head(page);
1490 if (page->pfmemalloc && !page->mapping)
1491 skb->pfmemalloc = true;
1492 }
1493
1494 /**
1495 * skb_fill_page_desc - initialise a paged fragment in an skb
1496 * @skb: buffer containing fragment to be initialised
1497 * @i: paged fragment index to initialise
1498 * @page: the page to use for this fragment
1499 * @off: the offset to the data with @page
1500 * @size: the length of the data
1501 *
1502 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1503 * @skb to point to @size bytes at offset @off within @page. In
1504 * addition updates @skb such that @i is the last fragment.
1505 *
1506 * Does not take any additional reference on the fragment.
1507 */
1508 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1509 struct page *page, int off, int size)
1510 {
1511 __skb_fill_page_desc(skb, i, page, off, size);
1512 skb_shinfo(skb)->nr_frags = i + 1;
1513 }
1514
1515 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1516 int size, unsigned int truesize);
1517
1518 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1519 unsigned int truesize);
1520
1521 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1522 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1523 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1524
1525 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1526 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1527 {
1528 return skb->head + skb->tail;
1529 }
1530
1531 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1532 {
1533 skb->tail = skb->data - skb->head;
1534 }
1535
1536 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1537 {
1538 skb_reset_tail_pointer(skb);
1539 skb->tail += offset;
1540 }
1541
1542 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1543 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1544 {
1545 return skb->tail;
1546 }
1547
1548 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1549 {
1550 skb->tail = skb->data;
1551 }
1552
1553 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1554 {
1555 skb->tail = skb->data + offset;
1556 }
1557
1558 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1559
1560 /*
1561 * Add data to an sk_buff
1562 */
1563 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1564 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1565 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1566 {
1567 unsigned char *tmp = skb_tail_pointer(skb);
1568 SKB_LINEAR_ASSERT(skb);
1569 skb->tail += len;
1570 skb->len += len;
1571 return tmp;
1572 }
1573
1574 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1575 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1576 {
1577 skb->data -= len;
1578 skb->len += len;
1579 return skb->data;
1580 }
1581
1582 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1583 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1584 {
1585 skb->len -= len;
1586 BUG_ON(skb->len < skb->data_len);
1587 return skb->data += len;
1588 }
1589
1590 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1591 {
1592 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1593 }
1594
1595 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1596
1597 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1598 {
1599 if (len > skb_headlen(skb) &&
1600 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1601 return NULL;
1602 skb->len -= len;
1603 return skb->data += len;
1604 }
1605
1606 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1607 {
1608 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1609 }
1610
1611 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1612 {
1613 if (likely(len <= skb_headlen(skb)))
1614 return 1;
1615 if (unlikely(len > skb->len))
1616 return 0;
1617 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1618 }
1619
1620 /**
1621 * skb_headroom - bytes at buffer head
1622 * @skb: buffer to check
1623 *
1624 * Return the number of bytes of free space at the head of an &sk_buff.
1625 */
1626 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1627 {
1628 return skb->data - skb->head;
1629 }
1630
1631 /**
1632 * skb_tailroom - bytes at buffer end
1633 * @skb: buffer to check
1634 *
1635 * Return the number of bytes of free space at the tail of an sk_buff
1636 */
1637 static inline int skb_tailroom(const struct sk_buff *skb)
1638 {
1639 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1640 }
1641
1642 /**
1643 * skb_availroom - bytes at buffer end
1644 * @skb: buffer to check
1645 *
1646 * Return the number of bytes of free space at the tail of an sk_buff
1647 * allocated by sk_stream_alloc()
1648 */
1649 static inline int skb_availroom(const struct sk_buff *skb)
1650 {
1651 if (skb_is_nonlinear(skb))
1652 return 0;
1653
1654 return skb->end - skb->tail - skb->reserved_tailroom;
1655 }
1656
1657 /**
1658 * skb_reserve - adjust headroom
1659 * @skb: buffer to alter
1660 * @len: bytes to move
1661 *
1662 * Increase the headroom of an empty &sk_buff by reducing the tail
1663 * room. This is only allowed for an empty buffer.
1664 */
1665 static inline void skb_reserve(struct sk_buff *skb, int len)
1666 {
1667 skb->data += len;
1668 skb->tail += len;
1669 }
1670
1671 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1672 {
1673 skb->inner_mac_header = skb->mac_header;
1674 skb->inner_network_header = skb->network_header;
1675 skb->inner_transport_header = skb->transport_header;
1676 }
1677
1678 static inline void skb_reset_mac_len(struct sk_buff *skb)
1679 {
1680 skb->mac_len = skb->network_header - skb->mac_header;
1681 }
1682
1683 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1684 *skb)
1685 {
1686 return skb->head + skb->inner_transport_header;
1687 }
1688
1689 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1690 {
1691 skb->inner_transport_header = skb->data - skb->head;
1692 }
1693
1694 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1695 const int offset)
1696 {
1697 skb_reset_inner_transport_header(skb);
1698 skb->inner_transport_header += offset;
1699 }
1700
1701 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1702 {
1703 return skb->head + skb->inner_network_header;
1704 }
1705
1706 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1707 {
1708 skb->inner_network_header = skb->data - skb->head;
1709 }
1710
1711 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1712 const int offset)
1713 {
1714 skb_reset_inner_network_header(skb);
1715 skb->inner_network_header += offset;
1716 }
1717
1718 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1719 {
1720 return skb->head + skb->inner_mac_header;
1721 }
1722
1723 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1724 {
1725 skb->inner_mac_header = skb->data - skb->head;
1726 }
1727
1728 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1729 const int offset)
1730 {
1731 skb_reset_inner_mac_header(skb);
1732 skb->inner_mac_header += offset;
1733 }
1734 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1735 {
1736 return skb->transport_header != (typeof(skb->transport_header))~0U;
1737 }
1738
1739 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1740 {
1741 return skb->head + skb->transport_header;
1742 }
1743
1744 static inline void skb_reset_transport_header(struct sk_buff *skb)
1745 {
1746 skb->transport_header = skb->data - skb->head;
1747 }
1748
1749 static inline void skb_set_transport_header(struct sk_buff *skb,
1750 const int offset)
1751 {
1752 skb_reset_transport_header(skb);
1753 skb->transport_header += offset;
1754 }
1755
1756 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1757 {
1758 return skb->head + skb->network_header;
1759 }
1760
1761 static inline void skb_reset_network_header(struct sk_buff *skb)
1762 {
1763 skb->network_header = skb->data - skb->head;
1764 }
1765
1766 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1767 {
1768 skb_reset_network_header(skb);
1769 skb->network_header += offset;
1770 }
1771
1772 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1773 {
1774 return skb->head + skb->mac_header;
1775 }
1776
1777 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1778 {
1779 return skb->mac_header != (typeof(skb->mac_header))~0U;
1780 }
1781
1782 static inline void skb_reset_mac_header(struct sk_buff *skb)
1783 {
1784 skb->mac_header = skb->data - skb->head;
1785 }
1786
1787 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1788 {
1789 skb_reset_mac_header(skb);
1790 skb->mac_header += offset;
1791 }
1792
1793 static inline void skb_pop_mac_header(struct sk_buff *skb)
1794 {
1795 skb->mac_header = skb->network_header;
1796 }
1797
1798 static inline void skb_probe_transport_header(struct sk_buff *skb,
1799 const int offset_hint)
1800 {
1801 struct flow_keys keys;
1802
1803 if (skb_transport_header_was_set(skb))
1804 return;
1805 else if (skb_flow_dissect(skb, &keys))
1806 skb_set_transport_header(skb, keys.thoff);
1807 else
1808 skb_set_transport_header(skb, offset_hint);
1809 }
1810
1811 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1812 {
1813 if (skb_mac_header_was_set(skb)) {
1814 const unsigned char *old_mac = skb_mac_header(skb);
1815
1816 skb_set_mac_header(skb, -skb->mac_len);
1817 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1818 }
1819 }
1820
1821 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1822 {
1823 return skb->csum_start - skb_headroom(skb);
1824 }
1825
1826 static inline int skb_transport_offset(const struct sk_buff *skb)
1827 {
1828 return skb_transport_header(skb) - skb->data;
1829 }
1830
1831 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1832 {
1833 return skb->transport_header - skb->network_header;
1834 }
1835
1836 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1837 {
1838 return skb->inner_transport_header - skb->inner_network_header;
1839 }
1840
1841 static inline int skb_network_offset(const struct sk_buff *skb)
1842 {
1843 return skb_network_header(skb) - skb->data;
1844 }
1845
1846 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1847 {
1848 return skb_inner_network_header(skb) - skb->data;
1849 }
1850
1851 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1852 {
1853 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1854 }
1855
1856 /*
1857 * CPUs often take a performance hit when accessing unaligned memory
1858 * locations. The actual performance hit varies, it can be small if the
1859 * hardware handles it or large if we have to take an exception and fix it
1860 * in software.
1861 *
1862 * Since an ethernet header is 14 bytes network drivers often end up with
1863 * the IP header at an unaligned offset. The IP header can be aligned by
1864 * shifting the start of the packet by 2 bytes. Drivers should do this
1865 * with:
1866 *
1867 * skb_reserve(skb, NET_IP_ALIGN);
1868 *
1869 * The downside to this alignment of the IP header is that the DMA is now
1870 * unaligned. On some architectures the cost of an unaligned DMA is high
1871 * and this cost outweighs the gains made by aligning the IP header.
1872 *
1873 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1874 * to be overridden.
1875 */
1876 #ifndef NET_IP_ALIGN
1877 #define NET_IP_ALIGN 2
1878 #endif
1879
1880 /*
1881 * The networking layer reserves some headroom in skb data (via
1882 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1883 * the header has to grow. In the default case, if the header has to grow
1884 * 32 bytes or less we avoid the reallocation.
1885 *
1886 * Unfortunately this headroom changes the DMA alignment of the resulting
1887 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1888 * on some architectures. An architecture can override this value,
1889 * perhaps setting it to a cacheline in size (since that will maintain
1890 * cacheline alignment of the DMA). It must be a power of 2.
1891 *
1892 * Various parts of the networking layer expect at least 32 bytes of
1893 * headroom, you should not reduce this.
1894 *
1895 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1896 * to reduce average number of cache lines per packet.
1897 * get_rps_cpus() for example only access one 64 bytes aligned block :
1898 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
1899 */
1900 #ifndef NET_SKB_PAD
1901 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
1902 #endif
1903
1904 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1905
1906 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1907 {
1908 if (unlikely(skb_is_nonlinear(skb))) {
1909 WARN_ON(1);
1910 return;
1911 }
1912 skb->len = len;
1913 skb_set_tail_pointer(skb, len);
1914 }
1915
1916 void skb_trim(struct sk_buff *skb, unsigned int len);
1917
1918 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1919 {
1920 if (skb->data_len)
1921 return ___pskb_trim(skb, len);
1922 __skb_trim(skb, len);
1923 return 0;
1924 }
1925
1926 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1927 {
1928 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1929 }
1930
1931 /**
1932 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1933 * @skb: buffer to alter
1934 * @len: new length
1935 *
1936 * This is identical to pskb_trim except that the caller knows that
1937 * the skb is not cloned so we should never get an error due to out-
1938 * of-memory.
1939 */
1940 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1941 {
1942 int err = pskb_trim(skb, len);
1943 BUG_ON(err);
1944 }
1945
1946 /**
1947 * skb_orphan - orphan a buffer
1948 * @skb: buffer to orphan
1949 *
1950 * If a buffer currently has an owner then we call the owner's
1951 * destructor function and make the @skb unowned. The buffer continues
1952 * to exist but is no longer charged to its former owner.
1953 */
1954 static inline void skb_orphan(struct sk_buff *skb)
1955 {
1956 if (skb->destructor) {
1957 skb->destructor(skb);
1958 skb->destructor = NULL;
1959 skb->sk = NULL;
1960 } else {
1961 BUG_ON(skb->sk);
1962 }
1963 }
1964
1965 /**
1966 * skb_orphan_frags - orphan the frags contained in a buffer
1967 * @skb: buffer to orphan frags from
1968 * @gfp_mask: allocation mask for replacement pages
1969 *
1970 * For each frag in the SKB which needs a destructor (i.e. has an
1971 * owner) create a copy of that frag and release the original
1972 * page by calling the destructor.
1973 */
1974 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1975 {
1976 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1977 return 0;
1978 return skb_copy_ubufs(skb, gfp_mask);
1979 }
1980
1981 /**
1982 * __skb_queue_purge - empty a list
1983 * @list: list to empty
1984 *
1985 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1986 * the list and one reference dropped. This function does not take the
1987 * list lock and the caller must hold the relevant locks to use it.
1988 */
1989 void skb_queue_purge(struct sk_buff_head *list);
1990 static inline void __skb_queue_purge(struct sk_buff_head *list)
1991 {
1992 struct sk_buff *skb;
1993 while ((skb = __skb_dequeue(list)) != NULL)
1994 kfree_skb(skb);
1995 }
1996
1997 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
1998 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
1999 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2000
2001 void *netdev_alloc_frag(unsigned int fragsz);
2002
2003 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2004 gfp_t gfp_mask);
2005
2006 /**
2007 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2008 * @dev: network device to receive on
2009 * @length: length to allocate
2010 *
2011 * Allocate a new &sk_buff and assign it a usage count of one. The
2012 * buffer has unspecified headroom built in. Users should allocate
2013 * the headroom they think they need without accounting for the
2014 * built in space. The built in space is used for optimisations.
2015 *
2016 * %NULL is returned if there is no free memory. Although this function
2017 * allocates memory it can be called from an interrupt.
2018 */
2019 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2020 unsigned int length)
2021 {
2022 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2023 }
2024
2025 /* legacy helper around __netdev_alloc_skb() */
2026 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2027 gfp_t gfp_mask)
2028 {
2029 return __netdev_alloc_skb(NULL, length, gfp_mask);
2030 }
2031
2032 /* legacy helper around netdev_alloc_skb() */
2033 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2034 {
2035 return netdev_alloc_skb(NULL, length);
2036 }
2037
2038
2039 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2040 unsigned int length, gfp_t gfp)
2041 {
2042 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2043
2044 if (NET_IP_ALIGN && skb)
2045 skb_reserve(skb, NET_IP_ALIGN);
2046 return skb;
2047 }
2048
2049 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2050 unsigned int length)
2051 {
2052 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2053 }
2054
2055 /**
2056 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
2057 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2058 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2059 * @order: size of the allocation
2060 *
2061 * Allocate a new page.
2062 *
2063 * %NULL is returned if there is no free memory.
2064 */
2065 static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2066 struct sk_buff *skb,
2067 unsigned int order)
2068 {
2069 struct page *page;
2070
2071 gfp_mask |= __GFP_COLD;
2072
2073 if (!(gfp_mask & __GFP_NOMEMALLOC))
2074 gfp_mask |= __GFP_MEMALLOC;
2075
2076 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2077 if (skb && page && page->pfmemalloc)
2078 skb->pfmemalloc = true;
2079
2080 return page;
2081 }
2082
2083 /**
2084 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2085 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2086 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2087 *
2088 * Allocate a new page.
2089 *
2090 * %NULL is returned if there is no free memory.
2091 */
2092 static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2093 struct sk_buff *skb)
2094 {
2095 return __skb_alloc_pages(gfp_mask, skb, 0);
2096 }
2097
2098 /**
2099 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2100 * @page: The page that was allocated from skb_alloc_page
2101 * @skb: The skb that may need pfmemalloc set
2102 */
2103 static inline void skb_propagate_pfmemalloc(struct page *page,
2104 struct sk_buff *skb)
2105 {
2106 if (page && page->pfmemalloc)
2107 skb->pfmemalloc = true;
2108 }
2109
2110 /**
2111 * skb_frag_page - retrieve the page referred to by a paged fragment
2112 * @frag: the paged fragment
2113 *
2114 * Returns the &struct page associated with @frag.
2115 */
2116 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2117 {
2118 return frag->page.p;
2119 }
2120
2121 /**
2122 * __skb_frag_ref - take an addition reference on a paged fragment.
2123 * @frag: the paged fragment
2124 *
2125 * Takes an additional reference on the paged fragment @frag.
2126 */
2127 static inline void __skb_frag_ref(skb_frag_t *frag)
2128 {
2129 get_page(skb_frag_page(frag));
2130 }
2131
2132 /**
2133 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2134 * @skb: the buffer
2135 * @f: the fragment offset.
2136 *
2137 * Takes an additional reference on the @f'th paged fragment of @skb.
2138 */
2139 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2140 {
2141 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2142 }
2143
2144 /**
2145 * __skb_frag_unref - release a reference on a paged fragment.
2146 * @frag: the paged fragment
2147 *
2148 * Releases a reference on the paged fragment @frag.
2149 */
2150 static inline void __skb_frag_unref(skb_frag_t *frag)
2151 {
2152 put_page(skb_frag_page(frag));
2153 }
2154
2155 /**
2156 * skb_frag_unref - release a reference on a paged fragment of an skb.
2157 * @skb: the buffer
2158 * @f: the fragment offset
2159 *
2160 * Releases a reference on the @f'th paged fragment of @skb.
2161 */
2162 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2163 {
2164 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2165 }
2166
2167 /**
2168 * skb_frag_address - gets the address of the data contained in a paged fragment
2169 * @frag: the paged fragment buffer
2170 *
2171 * Returns the address of the data within @frag. The page must already
2172 * be mapped.
2173 */
2174 static inline void *skb_frag_address(const skb_frag_t *frag)
2175 {
2176 return page_address(skb_frag_page(frag)) + frag->page_offset;
2177 }
2178
2179 /**
2180 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2181 * @frag: the paged fragment buffer
2182 *
2183 * Returns the address of the data within @frag. Checks that the page
2184 * is mapped and returns %NULL otherwise.
2185 */
2186 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2187 {
2188 void *ptr = page_address(skb_frag_page(frag));
2189 if (unlikely(!ptr))
2190 return NULL;
2191
2192 return ptr + frag->page_offset;
2193 }
2194
2195 /**
2196 * __skb_frag_set_page - sets the page contained in a paged fragment
2197 * @frag: the paged fragment
2198 * @page: the page to set
2199 *
2200 * Sets the fragment @frag to contain @page.
2201 */
2202 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2203 {
2204 frag->page.p = page;
2205 }
2206
2207 /**
2208 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2209 * @skb: the buffer
2210 * @f: the fragment offset
2211 * @page: the page to set
2212 *
2213 * Sets the @f'th fragment of @skb to contain @page.
2214 */
2215 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2216 struct page *page)
2217 {
2218 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2219 }
2220
2221 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2222
2223 /**
2224 * skb_frag_dma_map - maps a paged fragment via the DMA API
2225 * @dev: the device to map the fragment to
2226 * @frag: the paged fragment to map
2227 * @offset: the offset within the fragment (starting at the
2228 * fragment's own offset)
2229 * @size: the number of bytes to map
2230 * @dir: the direction of the mapping (%PCI_DMA_*)
2231 *
2232 * Maps the page associated with @frag to @device.
2233 */
2234 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2235 const skb_frag_t *frag,
2236 size_t offset, size_t size,
2237 enum dma_data_direction dir)
2238 {
2239 return dma_map_page(dev, skb_frag_page(frag),
2240 frag->page_offset + offset, size, dir);
2241 }
2242
2243 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2244 gfp_t gfp_mask)
2245 {
2246 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2247 }
2248
2249
2250 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2251 gfp_t gfp_mask)
2252 {
2253 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2254 }
2255
2256
2257 /**
2258 * skb_clone_writable - is the header of a clone writable
2259 * @skb: buffer to check
2260 * @len: length up to which to write
2261 *
2262 * Returns true if modifying the header part of the cloned buffer
2263 * does not requires the data to be copied.
2264 */
2265 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2266 {
2267 return !skb_header_cloned(skb) &&
2268 skb_headroom(skb) + len <= skb->hdr_len;
2269 }
2270
2271 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2272 int cloned)
2273 {
2274 int delta = 0;
2275
2276 if (headroom > skb_headroom(skb))
2277 delta = headroom - skb_headroom(skb);
2278
2279 if (delta || cloned)
2280 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2281 GFP_ATOMIC);
2282 return 0;
2283 }
2284
2285 /**
2286 * skb_cow - copy header of skb when it is required
2287 * @skb: buffer to cow
2288 * @headroom: needed headroom
2289 *
2290 * If the skb passed lacks sufficient headroom or its data part
2291 * is shared, data is reallocated. If reallocation fails, an error
2292 * is returned and original skb is not changed.
2293 *
2294 * The result is skb with writable area skb->head...skb->tail
2295 * and at least @headroom of space at head.
2296 */
2297 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2298 {
2299 return __skb_cow(skb, headroom, skb_cloned(skb));
2300 }
2301
2302 /**
2303 * skb_cow_head - skb_cow but only making the head writable
2304 * @skb: buffer to cow
2305 * @headroom: needed headroom
2306 *
2307 * This function is identical to skb_cow except that we replace the
2308 * skb_cloned check by skb_header_cloned. It should be used when
2309 * you only need to push on some header and do not need to modify
2310 * the data.
2311 */
2312 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2313 {
2314 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2315 }
2316
2317 /**
2318 * skb_padto - pad an skbuff up to a minimal size
2319 * @skb: buffer to pad
2320 * @len: minimal length
2321 *
2322 * Pads up a buffer to ensure the trailing bytes exist and are
2323 * blanked. If the buffer already contains sufficient data it
2324 * is untouched. Otherwise it is extended. Returns zero on
2325 * success. The skb is freed on error.
2326 */
2327
2328 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2329 {
2330 unsigned int size = skb->len;
2331 if (likely(size >= len))
2332 return 0;
2333 return skb_pad(skb, len - size);
2334 }
2335
2336 static inline int skb_add_data(struct sk_buff *skb,
2337 char __user *from, int copy)
2338 {
2339 const int off = skb->len;
2340
2341 if (skb->ip_summed == CHECKSUM_NONE) {
2342 int err = 0;
2343 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
2344 copy, 0, &err);
2345 if (!err) {
2346 skb->csum = csum_block_add(skb->csum, csum, off);
2347 return 0;
2348 }
2349 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2350 return 0;
2351
2352 __skb_trim(skb, off);
2353 return -EFAULT;
2354 }
2355
2356 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2357 const struct page *page, int off)
2358 {
2359 if (i) {
2360 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2361
2362 return page == skb_frag_page(frag) &&
2363 off == frag->page_offset + skb_frag_size(frag);
2364 }
2365 return false;
2366 }
2367
2368 static inline int __skb_linearize(struct sk_buff *skb)
2369 {
2370 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2371 }
2372
2373 /**
2374 * skb_linearize - convert paged skb to linear one
2375 * @skb: buffer to linarize
2376 *
2377 * If there is no free memory -ENOMEM is returned, otherwise zero
2378 * is returned and the old skb data released.
2379 */
2380 static inline int skb_linearize(struct sk_buff *skb)
2381 {
2382 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2383 }
2384
2385 /**
2386 * skb_has_shared_frag - can any frag be overwritten
2387 * @skb: buffer to test
2388 *
2389 * Return true if the skb has at least one frag that might be modified
2390 * by an external entity (as in vmsplice()/sendfile())
2391 */
2392 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2393 {
2394 return skb_is_nonlinear(skb) &&
2395 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2396 }
2397
2398 /**
2399 * skb_linearize_cow - make sure skb is linear and writable
2400 * @skb: buffer to process
2401 *
2402 * If there is no free memory -ENOMEM is returned, otherwise zero
2403 * is returned and the old skb data released.
2404 */
2405 static inline int skb_linearize_cow(struct sk_buff *skb)
2406 {
2407 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2408 __skb_linearize(skb) : 0;
2409 }
2410
2411 /**
2412 * skb_postpull_rcsum - update checksum for received skb after pull
2413 * @skb: buffer to update
2414 * @start: start of data before pull
2415 * @len: length of data pulled
2416 *
2417 * After doing a pull on a received packet, you need to call this to
2418 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2419 * CHECKSUM_NONE so that it can be recomputed from scratch.
2420 */
2421
2422 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2423 const void *start, unsigned int len)
2424 {
2425 if (skb->ip_summed == CHECKSUM_COMPLETE)
2426 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2427 }
2428
2429 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2430
2431 /**
2432 * pskb_trim_rcsum - trim received skb and update checksum
2433 * @skb: buffer to trim
2434 * @len: new length
2435 *
2436 * This is exactly the same as pskb_trim except that it ensures the
2437 * checksum of received packets are still valid after the operation.
2438 */
2439
2440 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2441 {
2442 if (likely(len >= skb->len))
2443 return 0;
2444 if (skb->ip_summed == CHECKSUM_COMPLETE)
2445 skb->ip_summed = CHECKSUM_NONE;
2446 return __pskb_trim(skb, len);
2447 }
2448
2449 #define skb_queue_walk(queue, skb) \
2450 for (skb = (queue)->next; \
2451 skb != (struct sk_buff *)(queue); \
2452 skb = skb->next)
2453
2454 #define skb_queue_walk_safe(queue, skb, tmp) \
2455 for (skb = (queue)->next, tmp = skb->next; \
2456 skb != (struct sk_buff *)(queue); \
2457 skb = tmp, tmp = skb->next)
2458
2459 #define skb_queue_walk_from(queue, skb) \
2460 for (; skb != (struct sk_buff *)(queue); \
2461 skb = skb->next)
2462
2463 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2464 for (tmp = skb->next; \
2465 skb != (struct sk_buff *)(queue); \
2466 skb = tmp, tmp = skb->next)
2467
2468 #define skb_queue_reverse_walk(queue, skb) \
2469 for (skb = (queue)->prev; \
2470 skb != (struct sk_buff *)(queue); \
2471 skb = skb->prev)
2472
2473 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2474 for (skb = (queue)->prev, tmp = skb->prev; \
2475 skb != (struct sk_buff *)(queue); \
2476 skb = tmp, tmp = skb->prev)
2477
2478 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2479 for (tmp = skb->prev; \
2480 skb != (struct sk_buff *)(queue); \
2481 skb = tmp, tmp = skb->prev)
2482
2483 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2484 {
2485 return skb_shinfo(skb)->frag_list != NULL;
2486 }
2487
2488 static inline void skb_frag_list_init(struct sk_buff *skb)
2489 {
2490 skb_shinfo(skb)->frag_list = NULL;
2491 }
2492
2493 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2494 {
2495 frag->next = skb_shinfo(skb)->frag_list;
2496 skb_shinfo(skb)->frag_list = frag;
2497 }
2498
2499 #define skb_walk_frags(skb, iter) \
2500 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2501
2502 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2503 int *peeked, int *off, int *err);
2504 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2505 int *err);
2506 unsigned int datagram_poll(struct file *file, struct socket *sock,
2507 struct poll_table_struct *wait);
2508 int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2509 struct iovec *to, int size);
2510 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2511 struct iovec *iov);
2512 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2513 const struct iovec *from, int from_offset,
2514 int len);
2515 int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2516 int offset, size_t count);
2517 int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2518 const struct iovec *to, int to_offset,
2519 int size);
2520 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2521 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2522 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2523 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2524 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2525 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2526 int len, __wsum csum);
2527 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2528 struct pipe_inode_info *pipe, unsigned int len,
2529 unsigned int flags);
2530 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2531 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2532 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2533 int len, int hlen);
2534 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2535 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2536 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2537 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2538 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2539
2540 struct skb_checksum_ops {
2541 __wsum (*update)(const void *mem, int len, __wsum wsum);
2542 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2543 };
2544
2545 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2546 __wsum csum, const struct skb_checksum_ops *ops);
2547 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2548 __wsum csum);
2549
2550 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2551 int len, void *buffer)
2552 {
2553 int hlen = skb_headlen(skb);
2554
2555 if (hlen - offset >= len)
2556 return skb->data + offset;
2557
2558 if (skb_copy_bits(skb, offset, buffer, len) < 0)
2559 return NULL;
2560
2561 return buffer;
2562 }
2563
2564 /**
2565 * skb_needs_linearize - check if we need to linearize a given skb
2566 * depending on the given device features.
2567 * @skb: socket buffer to check
2568 * @features: net device features
2569 *
2570 * Returns true if either:
2571 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2572 * 2. skb is fragmented and the device does not support SG.
2573 */
2574 static inline bool skb_needs_linearize(struct sk_buff *skb,
2575 netdev_features_t features)
2576 {
2577 return skb_is_nonlinear(skb) &&
2578 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2579 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2580 }
2581
2582 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2583 void *to,
2584 const unsigned int len)
2585 {
2586 memcpy(to, skb->data, len);
2587 }
2588
2589 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2590 const int offset, void *to,
2591 const unsigned int len)
2592 {
2593 memcpy(to, skb->data + offset, len);
2594 }
2595
2596 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2597 const void *from,
2598 const unsigned int len)
2599 {
2600 memcpy(skb->data, from, len);
2601 }
2602
2603 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2604 const int offset,
2605 const void *from,
2606 const unsigned int len)
2607 {
2608 memcpy(skb->data + offset, from, len);
2609 }
2610
2611 void skb_init(void);
2612
2613 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2614 {
2615 return skb->tstamp;
2616 }
2617
2618 /**
2619 * skb_get_timestamp - get timestamp from a skb
2620 * @skb: skb to get stamp from
2621 * @stamp: pointer to struct timeval to store stamp in
2622 *
2623 * Timestamps are stored in the skb as offsets to a base timestamp.
2624 * This function converts the offset back to a struct timeval and stores
2625 * it in stamp.
2626 */
2627 static inline void skb_get_timestamp(const struct sk_buff *skb,
2628 struct timeval *stamp)
2629 {
2630 *stamp = ktime_to_timeval(skb->tstamp);
2631 }
2632
2633 static inline void skb_get_timestampns(const struct sk_buff *skb,
2634 struct timespec *stamp)
2635 {
2636 *stamp = ktime_to_timespec(skb->tstamp);
2637 }
2638
2639 static inline void __net_timestamp(struct sk_buff *skb)
2640 {
2641 skb->tstamp = ktime_get_real();
2642 }
2643
2644 static inline ktime_t net_timedelta(ktime_t t)
2645 {
2646 return ktime_sub(ktime_get_real(), t);
2647 }
2648
2649 static inline ktime_t net_invalid_timestamp(void)
2650 {
2651 return ktime_set(0, 0);
2652 }
2653
2654 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2655
2656 void skb_clone_tx_timestamp(struct sk_buff *skb);
2657 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2658
2659 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2660
2661 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2662 {
2663 }
2664
2665 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2666 {
2667 return false;
2668 }
2669
2670 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2671
2672 /**
2673 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2674 *
2675 * PHY drivers may accept clones of transmitted packets for
2676 * timestamping via their phy_driver.txtstamp method. These drivers
2677 * must call this function to return the skb back to the stack, with
2678 * or without a timestamp.
2679 *
2680 * @skb: clone of the the original outgoing packet
2681 * @hwtstamps: hardware time stamps, may be NULL if not available
2682 *
2683 */
2684 void skb_complete_tx_timestamp(struct sk_buff *skb,
2685 struct skb_shared_hwtstamps *hwtstamps);
2686
2687 /**
2688 * skb_tstamp_tx - queue clone of skb with send time stamps
2689 * @orig_skb: the original outgoing packet
2690 * @hwtstamps: hardware time stamps, may be NULL if not available
2691 *
2692 * If the skb has a socket associated, then this function clones the
2693 * skb (thus sharing the actual data and optional structures), stores
2694 * the optional hardware time stamping information (if non NULL) or
2695 * generates a software time stamp (otherwise), then queues the clone
2696 * to the error queue of the socket. Errors are silently ignored.
2697 */
2698 void skb_tstamp_tx(struct sk_buff *orig_skb,
2699 struct skb_shared_hwtstamps *hwtstamps);
2700
2701 static inline void sw_tx_timestamp(struct sk_buff *skb)
2702 {
2703 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2704 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2705 skb_tstamp_tx(skb, NULL);
2706 }
2707
2708 /**
2709 * skb_tx_timestamp() - Driver hook for transmit timestamping
2710 *
2711 * Ethernet MAC Drivers should call this function in their hard_xmit()
2712 * function immediately before giving the sk_buff to the MAC hardware.
2713 *
2714 * Specifically, one should make absolutely sure that this function is
2715 * called before TX completion of this packet can trigger. Otherwise
2716 * the packet could potentially already be freed.
2717 *
2718 * @skb: A socket buffer.
2719 */
2720 static inline void skb_tx_timestamp(struct sk_buff *skb)
2721 {
2722 skb_clone_tx_timestamp(skb);
2723 sw_tx_timestamp(skb);
2724 }
2725
2726 /**
2727 * skb_complete_wifi_ack - deliver skb with wifi status
2728 *
2729 * @skb: the original outgoing packet
2730 * @acked: ack status
2731 *
2732 */
2733 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2734
2735 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2736 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2737
2738 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2739 {
2740 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
2741 }
2742
2743 /**
2744 * skb_checksum_complete - Calculate checksum of an entire packet
2745 * @skb: packet to process
2746 *
2747 * This function calculates the checksum over the entire packet plus
2748 * the value of skb->csum. The latter can be used to supply the
2749 * checksum of a pseudo header as used by TCP/UDP. It returns the
2750 * checksum.
2751 *
2752 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2753 * this function can be used to verify that checksum on received
2754 * packets. In that case the function should return zero if the
2755 * checksum is correct. In particular, this function will return zero
2756 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2757 * hardware has already verified the correctness of the checksum.
2758 */
2759 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2760 {
2761 return skb_csum_unnecessary(skb) ?
2762 0 : __skb_checksum_complete(skb);
2763 }
2764
2765 /* Check if we need to perform checksum complete validation.
2766 *
2767 * Returns true if checksum complete is needed, false otherwise
2768 * (either checksum is unnecessary or zero checksum is allowed).
2769 */
2770 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2771 bool zero_okay,
2772 __sum16 check)
2773 {
2774 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2775 skb->csum_valid = 1;
2776 return false;
2777 }
2778
2779 return true;
2780 }
2781
2782 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2783 * in checksum_init.
2784 */
2785 #define CHECKSUM_BREAK 76
2786
2787 /* Validate (init) checksum based on checksum complete.
2788 *
2789 * Return values:
2790 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2791 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2792 * checksum is stored in skb->csum for use in __skb_checksum_complete
2793 * non-zero: value of invalid checksum
2794 *
2795 */
2796 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2797 bool complete,
2798 __wsum psum)
2799 {
2800 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2801 if (!csum_fold(csum_add(psum, skb->csum))) {
2802 skb->csum_valid = 1;
2803 return 0;
2804 }
2805 }
2806
2807 skb->csum = psum;
2808
2809 if (complete || skb->len <= CHECKSUM_BREAK) {
2810 __sum16 csum;
2811
2812 csum = __skb_checksum_complete(skb);
2813 skb->csum_valid = !csum;
2814 return csum;
2815 }
2816
2817 return 0;
2818 }
2819
2820 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
2821 {
2822 return 0;
2823 }
2824
2825 /* Perform checksum validate (init). Note that this is a macro since we only
2826 * want to calculate the pseudo header which is an input function if necessary.
2827 * First we try to validate without any computation (checksum unnecessary) and
2828 * then calculate based on checksum complete calling the function to compute
2829 * pseudo header.
2830 *
2831 * Return values:
2832 * 0: checksum is validated or try to in skb_checksum_complete
2833 * non-zero: value of invalid checksum
2834 */
2835 #define __skb_checksum_validate(skb, proto, complete, \
2836 zero_okay, check, compute_pseudo) \
2837 ({ \
2838 __sum16 __ret = 0; \
2839 skb->csum_valid = 0; \
2840 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
2841 __ret = __skb_checksum_validate_complete(skb, \
2842 complete, compute_pseudo(skb, proto)); \
2843 __ret; \
2844 })
2845
2846 #define skb_checksum_init(skb, proto, compute_pseudo) \
2847 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
2848
2849 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
2850 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
2851
2852 #define skb_checksum_validate(skb, proto, compute_pseudo) \
2853 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
2854
2855 #define skb_checksum_validate_zero_check(skb, proto, check, \
2856 compute_pseudo) \
2857 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
2858
2859 #define skb_checksum_simple_validate(skb) \
2860 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
2861
2862 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2863 void nf_conntrack_destroy(struct nf_conntrack *nfct);
2864 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2865 {
2866 if (nfct && atomic_dec_and_test(&nfct->use))
2867 nf_conntrack_destroy(nfct);
2868 }
2869 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2870 {
2871 if (nfct)
2872 atomic_inc(&nfct->use);
2873 }
2874 #endif
2875 #ifdef CONFIG_BRIDGE_NETFILTER
2876 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2877 {
2878 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2879 kfree(nf_bridge);
2880 }
2881 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2882 {
2883 if (nf_bridge)
2884 atomic_inc(&nf_bridge->use);
2885 }
2886 #endif /* CONFIG_BRIDGE_NETFILTER */
2887 static inline void nf_reset(struct sk_buff *skb)
2888 {
2889 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2890 nf_conntrack_put(skb->nfct);
2891 skb->nfct = NULL;
2892 #endif
2893 #ifdef CONFIG_BRIDGE_NETFILTER
2894 nf_bridge_put(skb->nf_bridge);
2895 skb->nf_bridge = NULL;
2896 #endif
2897 }
2898
2899 static inline void nf_reset_trace(struct sk_buff *skb)
2900 {
2901 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2902 skb->nf_trace = 0;
2903 #endif
2904 }
2905
2906 /* Note: This doesn't put any conntrack and bridge info in dst. */
2907 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2908 {
2909 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2910 dst->nfct = src->nfct;
2911 nf_conntrack_get(src->nfct);
2912 dst->nfctinfo = src->nfctinfo;
2913 #endif
2914 #ifdef CONFIG_BRIDGE_NETFILTER
2915 dst->nf_bridge = src->nf_bridge;
2916 nf_bridge_get(src->nf_bridge);
2917 #endif
2918 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2919 dst->nf_trace = src->nf_trace;
2920 #endif
2921 }
2922
2923 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2924 {
2925 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2926 nf_conntrack_put(dst->nfct);
2927 #endif
2928 #ifdef CONFIG_BRIDGE_NETFILTER
2929 nf_bridge_put(dst->nf_bridge);
2930 #endif
2931 __nf_copy(dst, src);
2932 }
2933
2934 #ifdef CONFIG_NETWORK_SECMARK
2935 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2936 {
2937 to->secmark = from->secmark;
2938 }
2939
2940 static inline void skb_init_secmark(struct sk_buff *skb)
2941 {
2942 skb->secmark = 0;
2943 }
2944 #else
2945 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2946 { }
2947
2948 static inline void skb_init_secmark(struct sk_buff *skb)
2949 { }
2950 #endif
2951
2952 static inline bool skb_irq_freeable(const struct sk_buff *skb)
2953 {
2954 return !skb->destructor &&
2955 #if IS_ENABLED(CONFIG_XFRM)
2956 !skb->sp &&
2957 #endif
2958 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
2959 !skb->nfct &&
2960 #endif
2961 !skb->_skb_refdst &&
2962 !skb_has_frag_list(skb);
2963 }
2964
2965 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2966 {
2967 skb->queue_mapping = queue_mapping;
2968 }
2969
2970 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
2971 {
2972 return skb->queue_mapping;
2973 }
2974
2975 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2976 {
2977 to->queue_mapping = from->queue_mapping;
2978 }
2979
2980 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
2981 {
2982 skb->queue_mapping = rx_queue + 1;
2983 }
2984
2985 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
2986 {
2987 return skb->queue_mapping - 1;
2988 }
2989
2990 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
2991 {
2992 return skb->queue_mapping != 0;
2993 }
2994
2995 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2996 unsigned int num_tx_queues);
2997
2998 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
2999 {
3000 #ifdef CONFIG_XFRM
3001 return skb->sp;
3002 #else
3003 return NULL;
3004 #endif
3005 }
3006
3007 /* Keeps track of mac header offset relative to skb->head.
3008 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3009 * For non-tunnel skb it points to skb_mac_header() and for
3010 * tunnel skb it points to outer mac header.
3011 * Keeps track of level of encapsulation of network headers.
3012 */
3013 struct skb_gso_cb {
3014 int mac_offset;
3015 int encap_level;
3016 __u16 csum_start;
3017 };
3018 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3019
3020 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3021 {
3022 return (skb_mac_header(inner_skb) - inner_skb->head) -
3023 SKB_GSO_CB(inner_skb)->mac_offset;
3024 }
3025
3026 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3027 {
3028 int new_headroom, headroom;
3029 int ret;
3030
3031 headroom = skb_headroom(skb);
3032 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3033 if (ret)
3034 return ret;
3035
3036 new_headroom = skb_headroom(skb);
3037 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3038 return 0;
3039 }
3040
3041 /* Compute the checksum for a gso segment. First compute the checksum value
3042 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3043 * then add in skb->csum (checksum from csum_start to end of packet).
3044 * skb->csum and csum_start are then updated to reflect the checksum of the
3045 * resultant packet starting from the transport header-- the resultant checksum
3046 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3047 * header.
3048 */
3049 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3050 {
3051 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3052 skb_transport_offset(skb);
3053 __u16 csum;
3054
3055 csum = csum_fold(csum_partial(skb_transport_header(skb),
3056 plen, skb->csum));
3057 skb->csum = res;
3058 SKB_GSO_CB(skb)->csum_start -= plen;
3059
3060 return csum;
3061 }
3062
3063 static inline bool skb_is_gso(const struct sk_buff *skb)
3064 {
3065 return skb_shinfo(skb)->gso_size;
3066 }
3067
3068 /* Note: Should be called only if skb_is_gso(skb) is true */
3069 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3070 {
3071 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3072 }
3073
3074 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3075
3076 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3077 {
3078 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3079 * wanted then gso_type will be set. */
3080 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3081
3082 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3083 unlikely(shinfo->gso_type == 0)) {
3084 __skb_warn_lro_forwarding(skb);
3085 return true;
3086 }
3087 return false;
3088 }
3089
3090 static inline void skb_forward_csum(struct sk_buff *skb)
3091 {
3092 /* Unfortunately we don't support this one. Any brave souls? */
3093 if (skb->ip_summed == CHECKSUM_COMPLETE)
3094 skb->ip_summed = CHECKSUM_NONE;
3095 }
3096
3097 /**
3098 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3099 * @skb: skb to check
3100 *
3101 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3102 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3103 * use this helper, to document places where we make this assertion.
3104 */
3105 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3106 {
3107 #ifdef DEBUG
3108 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3109 #endif
3110 }
3111
3112 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3113
3114 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3115
3116 u32 __skb_get_poff(const struct sk_buff *skb);
3117
3118 /**
3119 * skb_head_is_locked - Determine if the skb->head is locked down
3120 * @skb: skb to check
3121 *
3122 * The head on skbs build around a head frag can be removed if they are
3123 * not cloned. This function returns true if the skb head is locked down
3124 * due to either being allocated via kmalloc, or by being a clone with
3125 * multiple references to the head.
3126 */
3127 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3128 {
3129 return !skb->head_frag || skb_cloned(skb);
3130 }
3131
3132 /**
3133 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3134 *
3135 * @skb: GSO skb
3136 *
3137 * skb_gso_network_seglen is used to determine the real size of the
3138 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3139 *
3140 * The MAC/L2 header is not accounted for.
3141 */
3142 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3143 {
3144 unsigned int hdr_len = skb_transport_header(skb) -
3145 skb_network_header(skb);
3146 return hdr_len + skb_gso_transport_seglen(skb);
3147 }
3148 #endif /* __KERNEL__ */
3149 #endif /* _LINUX_SKBUFF_H */