Merge remote-tracking branch 'regulator/for-next'
[deliverable/linux.git] / include / linux / skbuff.h
1 /*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
16
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
25
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <net/flow_dissector.h>
38 #include <linux/splice.h>
39 #include <linux/in6.h>
40 #include <linux/if_packet.h>
41 #include <net/flow.h>
42
43 /* The interface for checksum offload between the stack and networking drivers
44 * is as follows...
45 *
46 * A. IP checksum related features
47 *
48 * Drivers advertise checksum offload capabilities in the features of a device.
49 * From the stack's point of view these are capabilities offered by the driver,
50 * a driver typically only advertises features that it is capable of offloading
51 * to its device.
52 *
53 * The checksum related features are:
54 *
55 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
56 * IP (one's complement) checksum for any combination
57 * of protocols or protocol layering. The checksum is
58 * computed and set in a packet per the CHECKSUM_PARTIAL
59 * interface (see below).
60 *
61 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
62 * TCP or UDP packets over IPv4. These are specifically
63 * unencapsulated packets of the form IPv4|TCP or
64 * IPv4|UDP where the Protocol field in the IPv4 header
65 * is TCP or UDP. The IPv4 header may contain IP options
66 * This feature cannot be set in features for a device
67 * with NETIF_F_HW_CSUM also set. This feature is being
68 * DEPRECATED (see below).
69 *
70 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
71 * TCP or UDP packets over IPv6. These are specifically
72 * unencapsulated packets of the form IPv6|TCP or
73 * IPv4|UDP where the Next Header field in the IPv6
74 * header is either TCP or UDP. IPv6 extension headers
75 * are not supported with this feature. This feature
76 * cannot be set in features for a device with
77 * NETIF_F_HW_CSUM also set. This feature is being
78 * DEPRECATED (see below).
79 *
80 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
81 * This flag is used only used to disable the RX checksum
82 * feature for a device. The stack will accept receive
83 * checksum indication in packets received on a device
84 * regardless of whether NETIF_F_RXCSUM is set.
85 *
86 * B. Checksumming of received packets by device. Indication of checksum
87 * verification is in set skb->ip_summed. Possible values are:
88 *
89 * CHECKSUM_NONE:
90 *
91 * Device did not checksum this packet e.g. due to lack of capabilities.
92 * The packet contains full (though not verified) checksum in packet but
93 * not in skb->csum. Thus, skb->csum is undefined in this case.
94 *
95 * CHECKSUM_UNNECESSARY:
96 *
97 * The hardware you're dealing with doesn't calculate the full checksum
98 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
99 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
100 * if their checksums are okay. skb->csum is still undefined in this case
101 * though. A driver or device must never modify the checksum field in the
102 * packet even if checksum is verified.
103 *
104 * CHECKSUM_UNNECESSARY is applicable to following protocols:
105 * TCP: IPv6 and IPv4.
106 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
107 * zero UDP checksum for either IPv4 or IPv6, the networking stack
108 * may perform further validation in this case.
109 * GRE: only if the checksum is present in the header.
110 * SCTP: indicates the CRC in SCTP header has been validated.
111 *
112 * skb->csum_level indicates the number of consecutive checksums found in
113 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
114 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
115 * and a device is able to verify the checksums for UDP (possibly zero),
116 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
117 * two. If the device were only able to verify the UDP checksum and not
118 * GRE, either because it doesn't support GRE checksum of because GRE
119 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
120 * not considered in this case).
121 *
122 * CHECKSUM_COMPLETE:
123 *
124 * This is the most generic way. The device supplied checksum of the _whole_
125 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
126 * hardware doesn't need to parse L3/L4 headers to implement this.
127 *
128 * Note: Even if device supports only some protocols, but is able to produce
129 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
130 *
131 * CHECKSUM_PARTIAL:
132 *
133 * A checksum is set up to be offloaded to a device as described in the
134 * output description for CHECKSUM_PARTIAL. This may occur on a packet
135 * received directly from another Linux OS, e.g., a virtualized Linux kernel
136 * on the same host, or it may be set in the input path in GRO or remote
137 * checksum offload. For the purposes of checksum verification, the checksum
138 * referred to by skb->csum_start + skb->csum_offset and any preceding
139 * checksums in the packet are considered verified. Any checksums in the
140 * packet that are after the checksum being offloaded are not considered to
141 * be verified.
142 *
143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
144 * in the skb->ip_summed for a packet. Values are:
145 *
146 * CHECKSUM_PARTIAL:
147 *
148 * The driver is required to checksum the packet as seen by hard_start_xmit()
149 * from skb->csum_start up to the end, and to record/write the checksum at
150 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
151 * csum_start and csum_offset values are valid values given the length and
152 * offset of the packet, however they should not attempt to validate that the
153 * checksum refers to a legitimate transport layer checksum-- it is the
154 * purview of the stack to validate that csum_start and csum_offset are set
155 * correctly.
156 *
157 * When the stack requests checksum offload for a packet, the driver MUST
158 * ensure that the checksum is set correctly. A driver can either offload the
159 * checksum calculation to the device, or call skb_checksum_help (in the case
160 * that the device does not support offload for a particular checksum).
161 *
162 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
163 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
164 * checksum offload capability. If a device has limited checksum capabilities
165 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
166 * described above) a helper function can be called to resolve
167 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
168 * function takes a spec argument that describes the protocol layer that is
169 * supported for checksum offload and can be called for each packet. If a
170 * packet does not match the specification for offload, skb_checksum_help
171 * is called to resolve the checksum.
172 *
173 * CHECKSUM_NONE:
174 *
175 * The skb was already checksummed by the protocol, or a checksum is not
176 * required.
177 *
178 * CHECKSUM_UNNECESSARY:
179 *
180 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
181 * output.
182 *
183 * CHECKSUM_COMPLETE:
184 * Not used in checksum output. If a driver observes a packet with this value
185 * set in skbuff, if should treat as CHECKSUM_NONE being set.
186 *
187 * D. Non-IP checksum (CRC) offloads
188 *
189 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
190 * offloading the SCTP CRC in a packet. To perform this offload the stack
191 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
192 * accordingly. Note the there is no indication in the skbuff that the
193 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
194 * both IP checksum offload and SCTP CRC offload must verify which offload
195 * is configured for a packet presumably by inspecting packet headers.
196 *
197 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
198 * offloading the FCOE CRC in a packet. To perform this offload the stack
199 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
200 * accordingly. Note the there is no indication in the skbuff that the
201 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
202 * both IP checksum offload and FCOE CRC offload must verify which offload
203 * is configured for a packet presumably by inspecting packet headers.
204 *
205 * E. Checksumming on output with GSO.
206 *
207 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
208 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
209 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
210 * part of the GSO operation is implied. If a checksum is being offloaded
211 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
212 * are set to refer to the outermost checksum being offload (two offloaded
213 * checksums are possible with UDP encapsulation).
214 */
215
216 /* Don't change this without changing skb_csum_unnecessary! */
217 #define CHECKSUM_NONE 0
218 #define CHECKSUM_UNNECESSARY 1
219 #define CHECKSUM_COMPLETE 2
220 #define CHECKSUM_PARTIAL 3
221
222 /* Maximum value in skb->csum_level */
223 #define SKB_MAX_CSUM_LEVEL 3
224
225 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
226 #define SKB_WITH_OVERHEAD(X) \
227 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
228 #define SKB_MAX_ORDER(X, ORDER) \
229 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
230 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
231 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
232
233 /* return minimum truesize of one skb containing X bytes of data */
234 #define SKB_TRUESIZE(X) ((X) + \
235 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
236 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
237
238 struct net_device;
239 struct scatterlist;
240 struct pipe_inode_info;
241 struct iov_iter;
242 struct napi_struct;
243
244 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
245 struct nf_conntrack {
246 atomic_t use;
247 };
248 #endif
249
250 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
251 struct nf_bridge_info {
252 atomic_t use;
253 enum {
254 BRNF_PROTO_UNCHANGED,
255 BRNF_PROTO_8021Q,
256 BRNF_PROTO_PPPOE
257 } orig_proto:8;
258 u8 pkt_otherhost:1;
259 u8 in_prerouting:1;
260 u8 bridged_dnat:1;
261 __u16 frag_max_size;
262 struct net_device *physindev;
263
264 /* always valid & non-NULL from FORWARD on, for physdev match */
265 struct net_device *physoutdev;
266 union {
267 /* prerouting: detect dnat in orig/reply direction */
268 __be32 ipv4_daddr;
269 struct in6_addr ipv6_daddr;
270
271 /* after prerouting + nat detected: store original source
272 * mac since neigh resolution overwrites it, only used while
273 * skb is out in neigh layer.
274 */
275 char neigh_header[8];
276 };
277 };
278 #endif
279
280 struct sk_buff_head {
281 /* These two members must be first. */
282 struct sk_buff *next;
283 struct sk_buff *prev;
284
285 __u32 qlen;
286 spinlock_t lock;
287 };
288
289 struct sk_buff;
290
291 /* To allow 64K frame to be packed as single skb without frag_list we
292 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
293 * buffers which do not start on a page boundary.
294 *
295 * Since GRO uses frags we allocate at least 16 regardless of page
296 * size.
297 */
298 #if (65536/PAGE_SIZE + 1) < 16
299 #define MAX_SKB_FRAGS 16UL
300 #else
301 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
302 #endif
303 extern int sysctl_max_skb_frags;
304
305 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
306 * segment using its current segmentation instead.
307 */
308 #define GSO_BY_FRAGS 0xFFFF
309
310 typedef struct skb_frag_struct skb_frag_t;
311
312 struct skb_frag_struct {
313 struct {
314 struct page *p;
315 } page;
316 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
317 __u32 page_offset;
318 __u32 size;
319 #else
320 __u16 page_offset;
321 __u16 size;
322 #endif
323 };
324
325 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
326 {
327 return frag->size;
328 }
329
330 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
331 {
332 frag->size = size;
333 }
334
335 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
336 {
337 frag->size += delta;
338 }
339
340 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
341 {
342 frag->size -= delta;
343 }
344
345 #define HAVE_HW_TIME_STAMP
346
347 /**
348 * struct skb_shared_hwtstamps - hardware time stamps
349 * @hwtstamp: hardware time stamp transformed into duration
350 * since arbitrary point in time
351 *
352 * Software time stamps generated by ktime_get_real() are stored in
353 * skb->tstamp.
354 *
355 * hwtstamps can only be compared against other hwtstamps from
356 * the same device.
357 *
358 * This structure is attached to packets as part of the
359 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
360 */
361 struct skb_shared_hwtstamps {
362 ktime_t hwtstamp;
363 };
364
365 /* Definitions for tx_flags in struct skb_shared_info */
366 enum {
367 /* generate hardware time stamp */
368 SKBTX_HW_TSTAMP = 1 << 0,
369
370 /* generate software time stamp when queueing packet to NIC */
371 SKBTX_SW_TSTAMP = 1 << 1,
372
373 /* device driver is going to provide hardware time stamp */
374 SKBTX_IN_PROGRESS = 1 << 2,
375
376 /* device driver supports TX zero-copy buffers */
377 SKBTX_DEV_ZEROCOPY = 1 << 3,
378
379 /* generate wifi status information (where possible) */
380 SKBTX_WIFI_STATUS = 1 << 4,
381
382 /* This indicates at least one fragment might be overwritten
383 * (as in vmsplice(), sendfile() ...)
384 * If we need to compute a TX checksum, we'll need to copy
385 * all frags to avoid possible bad checksum
386 */
387 SKBTX_SHARED_FRAG = 1 << 5,
388
389 /* generate software time stamp when entering packet scheduling */
390 SKBTX_SCHED_TSTAMP = 1 << 6,
391 };
392
393 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
394 SKBTX_SCHED_TSTAMP)
395 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
396
397 /*
398 * The callback notifies userspace to release buffers when skb DMA is done in
399 * lower device, the skb last reference should be 0 when calling this.
400 * The zerocopy_success argument is true if zero copy transmit occurred,
401 * false on data copy or out of memory error caused by data copy attempt.
402 * The ctx field is used to track device context.
403 * The desc field is used to track userspace buffer index.
404 */
405 struct ubuf_info {
406 void (*callback)(struct ubuf_info *, bool zerocopy_success);
407 void *ctx;
408 unsigned long desc;
409 };
410
411 /* This data is invariant across clones and lives at
412 * the end of the header data, ie. at skb->end.
413 */
414 struct skb_shared_info {
415 unsigned char nr_frags;
416 __u8 tx_flags;
417 unsigned short gso_size;
418 /* Warning: this field is not always filled in (UFO)! */
419 unsigned short gso_segs;
420 unsigned short gso_type;
421 struct sk_buff *frag_list;
422 struct skb_shared_hwtstamps hwtstamps;
423 u32 tskey;
424 __be32 ip6_frag_id;
425
426 /*
427 * Warning : all fields before dataref are cleared in __alloc_skb()
428 */
429 atomic_t dataref;
430
431 /* Intermediate layers must ensure that destructor_arg
432 * remains valid until skb destructor */
433 void * destructor_arg;
434
435 /* must be last field, see pskb_expand_head() */
436 skb_frag_t frags[MAX_SKB_FRAGS];
437 };
438
439 /* We divide dataref into two halves. The higher 16 bits hold references
440 * to the payload part of skb->data. The lower 16 bits hold references to
441 * the entire skb->data. A clone of a headerless skb holds the length of
442 * the header in skb->hdr_len.
443 *
444 * All users must obey the rule that the skb->data reference count must be
445 * greater than or equal to the payload reference count.
446 *
447 * Holding a reference to the payload part means that the user does not
448 * care about modifications to the header part of skb->data.
449 */
450 #define SKB_DATAREF_SHIFT 16
451 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
452
453
454 enum {
455 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
456 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
457 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
458 };
459
460 enum {
461 SKB_GSO_TCPV4 = 1 << 0,
462 SKB_GSO_UDP = 1 << 1,
463
464 /* This indicates the skb is from an untrusted source. */
465 SKB_GSO_DODGY = 1 << 2,
466
467 /* This indicates the tcp segment has CWR set. */
468 SKB_GSO_TCP_ECN = 1 << 3,
469
470 SKB_GSO_TCP_FIXEDID = 1 << 4,
471
472 SKB_GSO_TCPV6 = 1 << 5,
473
474 SKB_GSO_FCOE = 1 << 6,
475
476 SKB_GSO_GRE = 1 << 7,
477
478 SKB_GSO_GRE_CSUM = 1 << 8,
479
480 SKB_GSO_IPXIP4 = 1 << 9,
481
482 SKB_GSO_IPXIP6 = 1 << 10,
483
484 SKB_GSO_UDP_TUNNEL = 1 << 11,
485
486 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
487
488 SKB_GSO_PARTIAL = 1 << 13,
489
490 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
491
492 SKB_GSO_SCTP = 1 << 15,
493 };
494
495 #if BITS_PER_LONG > 32
496 #define NET_SKBUFF_DATA_USES_OFFSET 1
497 #endif
498
499 #ifdef NET_SKBUFF_DATA_USES_OFFSET
500 typedef unsigned int sk_buff_data_t;
501 #else
502 typedef unsigned char *sk_buff_data_t;
503 #endif
504
505 /**
506 * struct skb_mstamp - multi resolution time stamps
507 * @stamp_us: timestamp in us resolution
508 * @stamp_jiffies: timestamp in jiffies
509 */
510 struct skb_mstamp {
511 union {
512 u64 v64;
513 struct {
514 u32 stamp_us;
515 u32 stamp_jiffies;
516 };
517 };
518 };
519
520 /**
521 * skb_mstamp_get - get current timestamp
522 * @cl: place to store timestamps
523 */
524 static inline void skb_mstamp_get(struct skb_mstamp *cl)
525 {
526 u64 val = local_clock();
527
528 do_div(val, NSEC_PER_USEC);
529 cl->stamp_us = (u32)val;
530 cl->stamp_jiffies = (u32)jiffies;
531 }
532
533 /**
534 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
535 * @t1: pointer to newest sample
536 * @t0: pointer to oldest sample
537 */
538 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
539 const struct skb_mstamp *t0)
540 {
541 s32 delta_us = t1->stamp_us - t0->stamp_us;
542 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
543
544 /* If delta_us is negative, this might be because interval is too big,
545 * or local_clock() drift is too big : fallback using jiffies.
546 */
547 if (delta_us <= 0 ||
548 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
549
550 delta_us = jiffies_to_usecs(delta_jiffies);
551
552 return delta_us;
553 }
554
555 static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
556 const struct skb_mstamp *t0)
557 {
558 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
559
560 if (!diff)
561 diff = t1->stamp_us - t0->stamp_us;
562 return diff > 0;
563 }
564
565 /**
566 * struct sk_buff - socket buffer
567 * @next: Next buffer in list
568 * @prev: Previous buffer in list
569 * @tstamp: Time we arrived/left
570 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
571 * @sk: Socket we are owned by
572 * @dev: Device we arrived on/are leaving by
573 * @cb: Control buffer. Free for use by every layer. Put private vars here
574 * @_skb_refdst: destination entry (with norefcount bit)
575 * @sp: the security path, used for xfrm
576 * @len: Length of actual data
577 * @data_len: Data length
578 * @mac_len: Length of link layer header
579 * @hdr_len: writable header length of cloned skb
580 * @csum: Checksum (must include start/offset pair)
581 * @csum_start: Offset from skb->head where checksumming should start
582 * @csum_offset: Offset from csum_start where checksum should be stored
583 * @priority: Packet queueing priority
584 * @ignore_df: allow local fragmentation
585 * @cloned: Head may be cloned (check refcnt to be sure)
586 * @ip_summed: Driver fed us an IP checksum
587 * @nohdr: Payload reference only, must not modify header
588 * @nfctinfo: Relationship of this skb to the connection
589 * @pkt_type: Packet class
590 * @fclone: skbuff clone status
591 * @ipvs_property: skbuff is owned by ipvs
592 * @peeked: this packet has been seen already, so stats have been
593 * done for it, don't do them again
594 * @nf_trace: netfilter packet trace flag
595 * @protocol: Packet protocol from driver
596 * @destructor: Destruct function
597 * @nfct: Associated connection, if any
598 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
599 * @skb_iif: ifindex of device we arrived on
600 * @tc_index: Traffic control index
601 * @tc_verd: traffic control verdict
602 * @hash: the packet hash
603 * @queue_mapping: Queue mapping for multiqueue devices
604 * @xmit_more: More SKBs are pending for this queue
605 * @ndisc_nodetype: router type (from link layer)
606 * @ooo_okay: allow the mapping of a socket to a queue to be changed
607 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
608 * ports.
609 * @sw_hash: indicates hash was computed in software stack
610 * @wifi_acked_valid: wifi_acked was set
611 * @wifi_acked: whether frame was acked on wifi or not
612 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
613 * @napi_id: id of the NAPI struct this skb came from
614 * @secmark: security marking
615 * @mark: Generic packet mark
616 * @vlan_proto: vlan encapsulation protocol
617 * @vlan_tci: vlan tag control information
618 * @inner_protocol: Protocol (encapsulation)
619 * @inner_transport_header: Inner transport layer header (encapsulation)
620 * @inner_network_header: Network layer header (encapsulation)
621 * @inner_mac_header: Link layer header (encapsulation)
622 * @transport_header: Transport layer header
623 * @network_header: Network layer header
624 * @mac_header: Link layer header
625 * @tail: Tail pointer
626 * @end: End pointer
627 * @head: Head of buffer
628 * @data: Data head pointer
629 * @truesize: Buffer size
630 * @users: User count - see {datagram,tcp}.c
631 */
632
633 struct sk_buff {
634 union {
635 struct {
636 /* These two members must be first. */
637 struct sk_buff *next;
638 struct sk_buff *prev;
639
640 union {
641 ktime_t tstamp;
642 struct skb_mstamp skb_mstamp;
643 };
644 };
645 struct rb_node rbnode; /* used in netem & tcp stack */
646 };
647 struct sock *sk;
648 struct net_device *dev;
649
650 /*
651 * This is the control buffer. It is free to use for every
652 * layer. Please put your private variables there. If you
653 * want to keep them across layers you have to do a skb_clone()
654 * first. This is owned by whoever has the skb queued ATM.
655 */
656 char cb[48] __aligned(8);
657
658 unsigned long _skb_refdst;
659 void (*destructor)(struct sk_buff *skb);
660 #ifdef CONFIG_XFRM
661 struct sec_path *sp;
662 #endif
663 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
664 struct nf_conntrack *nfct;
665 #endif
666 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
667 struct nf_bridge_info *nf_bridge;
668 #endif
669 unsigned int len,
670 data_len;
671 __u16 mac_len,
672 hdr_len;
673
674 /* Following fields are _not_ copied in __copy_skb_header()
675 * Note that queue_mapping is here mostly to fill a hole.
676 */
677 kmemcheck_bitfield_begin(flags1);
678 __u16 queue_mapping;
679 __u8 cloned:1,
680 nohdr:1,
681 fclone:2,
682 peeked:1,
683 head_frag:1,
684 xmit_more:1;
685 /* one bit hole */
686 kmemcheck_bitfield_end(flags1);
687
688 /* fields enclosed in headers_start/headers_end are copied
689 * using a single memcpy() in __copy_skb_header()
690 */
691 /* private: */
692 __u32 headers_start[0];
693 /* public: */
694
695 /* if you move pkt_type around you also must adapt those constants */
696 #ifdef __BIG_ENDIAN_BITFIELD
697 #define PKT_TYPE_MAX (7 << 5)
698 #else
699 #define PKT_TYPE_MAX 7
700 #endif
701 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
702
703 __u8 __pkt_type_offset[0];
704 __u8 pkt_type:3;
705 __u8 pfmemalloc:1;
706 __u8 ignore_df:1;
707 __u8 nfctinfo:3;
708
709 __u8 nf_trace:1;
710 __u8 ip_summed:2;
711 __u8 ooo_okay:1;
712 __u8 l4_hash:1;
713 __u8 sw_hash:1;
714 __u8 wifi_acked_valid:1;
715 __u8 wifi_acked:1;
716
717 __u8 no_fcs:1;
718 /* Indicates the inner headers are valid in the skbuff. */
719 __u8 encapsulation:1;
720 __u8 encap_hdr_csum:1;
721 __u8 csum_valid:1;
722 __u8 csum_complete_sw:1;
723 __u8 csum_level:2;
724 __u8 csum_bad:1;
725
726 #ifdef CONFIG_IPV6_NDISC_NODETYPE
727 __u8 ndisc_nodetype:2;
728 #endif
729 __u8 ipvs_property:1;
730 __u8 inner_protocol_type:1;
731 __u8 remcsum_offload:1;
732 #ifdef CONFIG_NET_SWITCHDEV
733 __u8 offload_fwd_mark:1;
734 #endif
735 /* 2, 4 or 5 bit hole */
736
737 #ifdef CONFIG_NET_SCHED
738 __u16 tc_index; /* traffic control index */
739 #ifdef CONFIG_NET_CLS_ACT
740 __u16 tc_verd; /* traffic control verdict */
741 #endif
742 #endif
743
744 union {
745 __wsum csum;
746 struct {
747 __u16 csum_start;
748 __u16 csum_offset;
749 };
750 };
751 __u32 priority;
752 int skb_iif;
753 __u32 hash;
754 __be16 vlan_proto;
755 __u16 vlan_tci;
756 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
757 union {
758 unsigned int napi_id;
759 unsigned int sender_cpu;
760 };
761 #endif
762 #ifdef CONFIG_NETWORK_SECMARK
763 __u32 secmark;
764 #endif
765
766 union {
767 __u32 mark;
768 __u32 reserved_tailroom;
769 };
770
771 union {
772 __be16 inner_protocol;
773 __u8 inner_ipproto;
774 };
775
776 __u16 inner_transport_header;
777 __u16 inner_network_header;
778 __u16 inner_mac_header;
779
780 __be16 protocol;
781 __u16 transport_header;
782 __u16 network_header;
783 __u16 mac_header;
784
785 /* private: */
786 __u32 headers_end[0];
787 /* public: */
788
789 /* These elements must be at the end, see alloc_skb() for details. */
790 sk_buff_data_t tail;
791 sk_buff_data_t end;
792 unsigned char *head,
793 *data;
794 unsigned int truesize;
795 atomic_t users;
796 };
797
798 #ifdef __KERNEL__
799 /*
800 * Handling routines are only of interest to the kernel
801 */
802 #include <linux/slab.h>
803
804
805 #define SKB_ALLOC_FCLONE 0x01
806 #define SKB_ALLOC_RX 0x02
807 #define SKB_ALLOC_NAPI 0x04
808
809 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
810 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
811 {
812 return unlikely(skb->pfmemalloc);
813 }
814
815 /*
816 * skb might have a dst pointer attached, refcounted or not.
817 * _skb_refdst low order bit is set if refcount was _not_ taken
818 */
819 #define SKB_DST_NOREF 1UL
820 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
821
822 /**
823 * skb_dst - returns skb dst_entry
824 * @skb: buffer
825 *
826 * Returns skb dst_entry, regardless of reference taken or not.
827 */
828 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
829 {
830 /* If refdst was not refcounted, check we still are in a
831 * rcu_read_lock section
832 */
833 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
834 !rcu_read_lock_held() &&
835 !rcu_read_lock_bh_held());
836 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
837 }
838
839 /**
840 * skb_dst_set - sets skb dst
841 * @skb: buffer
842 * @dst: dst entry
843 *
844 * Sets skb dst, assuming a reference was taken on dst and should
845 * be released by skb_dst_drop()
846 */
847 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
848 {
849 skb->_skb_refdst = (unsigned long)dst;
850 }
851
852 /**
853 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
854 * @skb: buffer
855 * @dst: dst entry
856 *
857 * Sets skb dst, assuming a reference was not taken on dst.
858 * If dst entry is cached, we do not take reference and dst_release
859 * will be avoided by refdst_drop. If dst entry is not cached, we take
860 * reference, so that last dst_release can destroy the dst immediately.
861 */
862 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
863 {
864 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
865 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
866 }
867
868 /**
869 * skb_dst_is_noref - Test if skb dst isn't refcounted
870 * @skb: buffer
871 */
872 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
873 {
874 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
875 }
876
877 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
878 {
879 return (struct rtable *)skb_dst(skb);
880 }
881
882 /* For mangling skb->pkt_type from user space side from applications
883 * such as nft, tc, etc, we only allow a conservative subset of
884 * possible pkt_types to be set.
885 */
886 static inline bool skb_pkt_type_ok(u32 ptype)
887 {
888 return ptype <= PACKET_OTHERHOST;
889 }
890
891 void kfree_skb(struct sk_buff *skb);
892 void kfree_skb_list(struct sk_buff *segs);
893 void skb_tx_error(struct sk_buff *skb);
894 void consume_skb(struct sk_buff *skb);
895 void __kfree_skb(struct sk_buff *skb);
896 extern struct kmem_cache *skbuff_head_cache;
897
898 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
899 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
900 bool *fragstolen, int *delta_truesize);
901
902 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
903 int node);
904 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
905 struct sk_buff *build_skb(void *data, unsigned int frag_size);
906 static inline struct sk_buff *alloc_skb(unsigned int size,
907 gfp_t priority)
908 {
909 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
910 }
911
912 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
913 unsigned long data_len,
914 int max_page_order,
915 int *errcode,
916 gfp_t gfp_mask);
917
918 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
919 struct sk_buff_fclones {
920 struct sk_buff skb1;
921
922 struct sk_buff skb2;
923
924 atomic_t fclone_ref;
925 };
926
927 /**
928 * skb_fclone_busy - check if fclone is busy
929 * @skb: buffer
930 *
931 * Returns true if skb is a fast clone, and its clone is not freed.
932 * Some drivers call skb_orphan() in their ndo_start_xmit(),
933 * so we also check that this didnt happen.
934 */
935 static inline bool skb_fclone_busy(const struct sock *sk,
936 const struct sk_buff *skb)
937 {
938 const struct sk_buff_fclones *fclones;
939
940 fclones = container_of(skb, struct sk_buff_fclones, skb1);
941
942 return skb->fclone == SKB_FCLONE_ORIG &&
943 atomic_read(&fclones->fclone_ref) > 1 &&
944 fclones->skb2.sk == sk;
945 }
946
947 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
948 gfp_t priority)
949 {
950 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
951 }
952
953 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
954 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
955 {
956 return __alloc_skb_head(priority, -1);
957 }
958
959 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
960 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
961 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
962 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
963 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
964 gfp_t gfp_mask, bool fclone);
965 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
966 gfp_t gfp_mask)
967 {
968 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
969 }
970
971 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
972 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
973 unsigned int headroom);
974 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
975 int newtailroom, gfp_t priority);
976 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
977 int offset, int len);
978 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
979 int len);
980 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
981 int skb_pad(struct sk_buff *skb, int pad);
982 #define dev_kfree_skb(a) consume_skb(a)
983
984 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
985 int getfrag(void *from, char *to, int offset,
986 int len, int odd, struct sk_buff *skb),
987 void *from, int length);
988
989 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
990 int offset, size_t size);
991
992 struct skb_seq_state {
993 __u32 lower_offset;
994 __u32 upper_offset;
995 __u32 frag_idx;
996 __u32 stepped_offset;
997 struct sk_buff *root_skb;
998 struct sk_buff *cur_skb;
999 __u8 *frag_data;
1000 };
1001
1002 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1003 unsigned int to, struct skb_seq_state *st);
1004 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1005 struct skb_seq_state *st);
1006 void skb_abort_seq_read(struct skb_seq_state *st);
1007
1008 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1009 unsigned int to, struct ts_config *config);
1010
1011 /*
1012 * Packet hash types specify the type of hash in skb_set_hash.
1013 *
1014 * Hash types refer to the protocol layer addresses which are used to
1015 * construct a packet's hash. The hashes are used to differentiate or identify
1016 * flows of the protocol layer for the hash type. Hash types are either
1017 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1018 *
1019 * Properties of hashes:
1020 *
1021 * 1) Two packets in different flows have different hash values
1022 * 2) Two packets in the same flow should have the same hash value
1023 *
1024 * A hash at a higher layer is considered to be more specific. A driver should
1025 * set the most specific hash possible.
1026 *
1027 * A driver cannot indicate a more specific hash than the layer at which a hash
1028 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1029 *
1030 * A driver may indicate a hash level which is less specific than the
1031 * actual layer the hash was computed on. For instance, a hash computed
1032 * at L4 may be considered an L3 hash. This should only be done if the
1033 * driver can't unambiguously determine that the HW computed the hash at
1034 * the higher layer. Note that the "should" in the second property above
1035 * permits this.
1036 */
1037 enum pkt_hash_types {
1038 PKT_HASH_TYPE_NONE, /* Undefined type */
1039 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1040 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1041 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1042 };
1043
1044 static inline void skb_clear_hash(struct sk_buff *skb)
1045 {
1046 skb->hash = 0;
1047 skb->sw_hash = 0;
1048 skb->l4_hash = 0;
1049 }
1050
1051 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1052 {
1053 if (!skb->l4_hash)
1054 skb_clear_hash(skb);
1055 }
1056
1057 static inline void
1058 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1059 {
1060 skb->l4_hash = is_l4;
1061 skb->sw_hash = is_sw;
1062 skb->hash = hash;
1063 }
1064
1065 static inline void
1066 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1067 {
1068 /* Used by drivers to set hash from HW */
1069 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1070 }
1071
1072 static inline void
1073 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1074 {
1075 __skb_set_hash(skb, hash, true, is_l4);
1076 }
1077
1078 void __skb_get_hash(struct sk_buff *skb);
1079 u32 __skb_get_hash_symmetric(struct sk_buff *skb);
1080 u32 skb_get_poff(const struct sk_buff *skb);
1081 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1082 const struct flow_keys *keys, int hlen);
1083 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1084 void *data, int hlen_proto);
1085
1086 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1087 int thoff, u8 ip_proto)
1088 {
1089 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1090 }
1091
1092 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1093 const struct flow_dissector_key *key,
1094 unsigned int key_count);
1095
1096 bool __skb_flow_dissect(const struct sk_buff *skb,
1097 struct flow_dissector *flow_dissector,
1098 void *target_container,
1099 void *data, __be16 proto, int nhoff, int hlen,
1100 unsigned int flags);
1101
1102 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1103 struct flow_dissector *flow_dissector,
1104 void *target_container, unsigned int flags)
1105 {
1106 return __skb_flow_dissect(skb, flow_dissector, target_container,
1107 NULL, 0, 0, 0, flags);
1108 }
1109
1110 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1111 struct flow_keys *flow,
1112 unsigned int flags)
1113 {
1114 memset(flow, 0, sizeof(*flow));
1115 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1116 NULL, 0, 0, 0, flags);
1117 }
1118
1119 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1120 void *data, __be16 proto,
1121 int nhoff, int hlen,
1122 unsigned int flags)
1123 {
1124 memset(flow, 0, sizeof(*flow));
1125 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1126 data, proto, nhoff, hlen, flags);
1127 }
1128
1129 static inline __u32 skb_get_hash(struct sk_buff *skb)
1130 {
1131 if (!skb->l4_hash && !skb->sw_hash)
1132 __skb_get_hash(skb);
1133
1134 return skb->hash;
1135 }
1136
1137 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1138
1139 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1140 {
1141 if (!skb->l4_hash && !skb->sw_hash) {
1142 struct flow_keys keys;
1143 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1144
1145 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1146 }
1147
1148 return skb->hash;
1149 }
1150
1151 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1152
1153 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1154 {
1155 if (!skb->l4_hash && !skb->sw_hash) {
1156 struct flow_keys keys;
1157 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1158
1159 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1160 }
1161
1162 return skb->hash;
1163 }
1164
1165 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1166
1167 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1168 {
1169 return skb->hash;
1170 }
1171
1172 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1173 {
1174 to->hash = from->hash;
1175 to->sw_hash = from->sw_hash;
1176 to->l4_hash = from->l4_hash;
1177 };
1178
1179 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1180 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1181 {
1182 return skb->head + skb->end;
1183 }
1184
1185 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1186 {
1187 return skb->end;
1188 }
1189 #else
1190 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1191 {
1192 return skb->end;
1193 }
1194
1195 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1196 {
1197 return skb->end - skb->head;
1198 }
1199 #endif
1200
1201 /* Internal */
1202 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1203
1204 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1205 {
1206 return &skb_shinfo(skb)->hwtstamps;
1207 }
1208
1209 /**
1210 * skb_queue_empty - check if a queue is empty
1211 * @list: queue head
1212 *
1213 * Returns true if the queue is empty, false otherwise.
1214 */
1215 static inline int skb_queue_empty(const struct sk_buff_head *list)
1216 {
1217 return list->next == (const struct sk_buff *) list;
1218 }
1219
1220 /**
1221 * skb_queue_is_last - check if skb is the last entry in the queue
1222 * @list: queue head
1223 * @skb: buffer
1224 *
1225 * Returns true if @skb is the last buffer on the list.
1226 */
1227 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1228 const struct sk_buff *skb)
1229 {
1230 return skb->next == (const struct sk_buff *) list;
1231 }
1232
1233 /**
1234 * skb_queue_is_first - check if skb is the first entry in the queue
1235 * @list: queue head
1236 * @skb: buffer
1237 *
1238 * Returns true if @skb is the first buffer on the list.
1239 */
1240 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1241 const struct sk_buff *skb)
1242 {
1243 return skb->prev == (const struct sk_buff *) list;
1244 }
1245
1246 /**
1247 * skb_queue_next - return the next packet in the queue
1248 * @list: queue head
1249 * @skb: current buffer
1250 *
1251 * Return the next packet in @list after @skb. It is only valid to
1252 * call this if skb_queue_is_last() evaluates to false.
1253 */
1254 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1255 const struct sk_buff *skb)
1256 {
1257 /* This BUG_ON may seem severe, but if we just return then we
1258 * are going to dereference garbage.
1259 */
1260 BUG_ON(skb_queue_is_last(list, skb));
1261 return skb->next;
1262 }
1263
1264 /**
1265 * skb_queue_prev - return the prev packet in the queue
1266 * @list: queue head
1267 * @skb: current buffer
1268 *
1269 * Return the prev packet in @list before @skb. It is only valid to
1270 * call this if skb_queue_is_first() evaluates to false.
1271 */
1272 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1273 const struct sk_buff *skb)
1274 {
1275 /* This BUG_ON may seem severe, but if we just return then we
1276 * are going to dereference garbage.
1277 */
1278 BUG_ON(skb_queue_is_first(list, skb));
1279 return skb->prev;
1280 }
1281
1282 /**
1283 * skb_get - reference buffer
1284 * @skb: buffer to reference
1285 *
1286 * Makes another reference to a socket buffer and returns a pointer
1287 * to the buffer.
1288 */
1289 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1290 {
1291 atomic_inc(&skb->users);
1292 return skb;
1293 }
1294
1295 /*
1296 * If users == 1, we are the only owner and are can avoid redundant
1297 * atomic change.
1298 */
1299
1300 /**
1301 * skb_cloned - is the buffer a clone
1302 * @skb: buffer to check
1303 *
1304 * Returns true if the buffer was generated with skb_clone() and is
1305 * one of multiple shared copies of the buffer. Cloned buffers are
1306 * shared data so must not be written to under normal circumstances.
1307 */
1308 static inline int skb_cloned(const struct sk_buff *skb)
1309 {
1310 return skb->cloned &&
1311 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1312 }
1313
1314 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1315 {
1316 might_sleep_if(gfpflags_allow_blocking(pri));
1317
1318 if (skb_cloned(skb))
1319 return pskb_expand_head(skb, 0, 0, pri);
1320
1321 return 0;
1322 }
1323
1324 /**
1325 * skb_header_cloned - is the header a clone
1326 * @skb: buffer to check
1327 *
1328 * Returns true if modifying the header part of the buffer requires
1329 * the data to be copied.
1330 */
1331 static inline int skb_header_cloned(const struct sk_buff *skb)
1332 {
1333 int dataref;
1334
1335 if (!skb->cloned)
1336 return 0;
1337
1338 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1339 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1340 return dataref != 1;
1341 }
1342
1343 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1344 {
1345 might_sleep_if(gfpflags_allow_blocking(pri));
1346
1347 if (skb_header_cloned(skb))
1348 return pskb_expand_head(skb, 0, 0, pri);
1349
1350 return 0;
1351 }
1352
1353 /**
1354 * skb_header_release - release reference to header
1355 * @skb: buffer to operate on
1356 *
1357 * Drop a reference to the header part of the buffer. This is done
1358 * by acquiring a payload reference. You must not read from the header
1359 * part of skb->data after this.
1360 * Note : Check if you can use __skb_header_release() instead.
1361 */
1362 static inline void skb_header_release(struct sk_buff *skb)
1363 {
1364 BUG_ON(skb->nohdr);
1365 skb->nohdr = 1;
1366 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1367 }
1368
1369 /**
1370 * __skb_header_release - release reference to header
1371 * @skb: buffer to operate on
1372 *
1373 * Variant of skb_header_release() assuming skb is private to caller.
1374 * We can avoid one atomic operation.
1375 */
1376 static inline void __skb_header_release(struct sk_buff *skb)
1377 {
1378 skb->nohdr = 1;
1379 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1380 }
1381
1382
1383 /**
1384 * skb_shared - is the buffer shared
1385 * @skb: buffer to check
1386 *
1387 * Returns true if more than one person has a reference to this
1388 * buffer.
1389 */
1390 static inline int skb_shared(const struct sk_buff *skb)
1391 {
1392 return atomic_read(&skb->users) != 1;
1393 }
1394
1395 /**
1396 * skb_share_check - check if buffer is shared and if so clone it
1397 * @skb: buffer to check
1398 * @pri: priority for memory allocation
1399 *
1400 * If the buffer is shared the buffer is cloned and the old copy
1401 * drops a reference. A new clone with a single reference is returned.
1402 * If the buffer is not shared the original buffer is returned. When
1403 * being called from interrupt status or with spinlocks held pri must
1404 * be GFP_ATOMIC.
1405 *
1406 * NULL is returned on a memory allocation failure.
1407 */
1408 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1409 {
1410 might_sleep_if(gfpflags_allow_blocking(pri));
1411 if (skb_shared(skb)) {
1412 struct sk_buff *nskb = skb_clone(skb, pri);
1413
1414 if (likely(nskb))
1415 consume_skb(skb);
1416 else
1417 kfree_skb(skb);
1418 skb = nskb;
1419 }
1420 return skb;
1421 }
1422
1423 /*
1424 * Copy shared buffers into a new sk_buff. We effectively do COW on
1425 * packets to handle cases where we have a local reader and forward
1426 * and a couple of other messy ones. The normal one is tcpdumping
1427 * a packet thats being forwarded.
1428 */
1429
1430 /**
1431 * skb_unshare - make a copy of a shared buffer
1432 * @skb: buffer to check
1433 * @pri: priority for memory allocation
1434 *
1435 * If the socket buffer is a clone then this function creates a new
1436 * copy of the data, drops a reference count on the old copy and returns
1437 * the new copy with the reference count at 1. If the buffer is not a clone
1438 * the original buffer is returned. When called with a spinlock held or
1439 * from interrupt state @pri must be %GFP_ATOMIC
1440 *
1441 * %NULL is returned on a memory allocation failure.
1442 */
1443 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1444 gfp_t pri)
1445 {
1446 might_sleep_if(gfpflags_allow_blocking(pri));
1447 if (skb_cloned(skb)) {
1448 struct sk_buff *nskb = skb_copy(skb, pri);
1449
1450 /* Free our shared copy */
1451 if (likely(nskb))
1452 consume_skb(skb);
1453 else
1454 kfree_skb(skb);
1455 skb = nskb;
1456 }
1457 return skb;
1458 }
1459
1460 /**
1461 * skb_peek - peek at the head of an &sk_buff_head
1462 * @list_: list to peek at
1463 *
1464 * Peek an &sk_buff. Unlike most other operations you _MUST_
1465 * be careful with this one. A peek leaves the buffer on the
1466 * list and someone else may run off with it. You must hold
1467 * the appropriate locks or have a private queue to do this.
1468 *
1469 * Returns %NULL for an empty list or a pointer to the head element.
1470 * The reference count is not incremented and the reference is therefore
1471 * volatile. Use with caution.
1472 */
1473 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1474 {
1475 struct sk_buff *skb = list_->next;
1476
1477 if (skb == (struct sk_buff *)list_)
1478 skb = NULL;
1479 return skb;
1480 }
1481
1482 /**
1483 * skb_peek_next - peek skb following the given one from a queue
1484 * @skb: skb to start from
1485 * @list_: list to peek at
1486 *
1487 * Returns %NULL when the end of the list is met or a pointer to the
1488 * next element. The reference count is not incremented and the
1489 * reference is therefore volatile. Use with caution.
1490 */
1491 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1492 const struct sk_buff_head *list_)
1493 {
1494 struct sk_buff *next = skb->next;
1495
1496 if (next == (struct sk_buff *)list_)
1497 next = NULL;
1498 return next;
1499 }
1500
1501 /**
1502 * skb_peek_tail - peek at the tail of an &sk_buff_head
1503 * @list_: list to peek at
1504 *
1505 * Peek an &sk_buff. Unlike most other operations you _MUST_
1506 * be careful with this one. A peek leaves the buffer on the
1507 * list and someone else may run off with it. You must hold
1508 * the appropriate locks or have a private queue to do this.
1509 *
1510 * Returns %NULL for an empty list or a pointer to the tail element.
1511 * The reference count is not incremented and the reference is therefore
1512 * volatile. Use with caution.
1513 */
1514 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1515 {
1516 struct sk_buff *skb = list_->prev;
1517
1518 if (skb == (struct sk_buff *)list_)
1519 skb = NULL;
1520 return skb;
1521
1522 }
1523
1524 /**
1525 * skb_queue_len - get queue length
1526 * @list_: list to measure
1527 *
1528 * Return the length of an &sk_buff queue.
1529 */
1530 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1531 {
1532 return list_->qlen;
1533 }
1534
1535 /**
1536 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1537 * @list: queue to initialize
1538 *
1539 * This initializes only the list and queue length aspects of
1540 * an sk_buff_head object. This allows to initialize the list
1541 * aspects of an sk_buff_head without reinitializing things like
1542 * the spinlock. It can also be used for on-stack sk_buff_head
1543 * objects where the spinlock is known to not be used.
1544 */
1545 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1546 {
1547 list->prev = list->next = (struct sk_buff *)list;
1548 list->qlen = 0;
1549 }
1550
1551 /*
1552 * This function creates a split out lock class for each invocation;
1553 * this is needed for now since a whole lot of users of the skb-queue
1554 * infrastructure in drivers have different locking usage (in hardirq)
1555 * than the networking core (in softirq only). In the long run either the
1556 * network layer or drivers should need annotation to consolidate the
1557 * main types of usage into 3 classes.
1558 */
1559 static inline void skb_queue_head_init(struct sk_buff_head *list)
1560 {
1561 spin_lock_init(&list->lock);
1562 __skb_queue_head_init(list);
1563 }
1564
1565 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1566 struct lock_class_key *class)
1567 {
1568 skb_queue_head_init(list);
1569 lockdep_set_class(&list->lock, class);
1570 }
1571
1572 /*
1573 * Insert an sk_buff on a list.
1574 *
1575 * The "__skb_xxxx()" functions are the non-atomic ones that
1576 * can only be called with interrupts disabled.
1577 */
1578 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1579 struct sk_buff_head *list);
1580 static inline void __skb_insert(struct sk_buff *newsk,
1581 struct sk_buff *prev, struct sk_buff *next,
1582 struct sk_buff_head *list)
1583 {
1584 newsk->next = next;
1585 newsk->prev = prev;
1586 next->prev = prev->next = newsk;
1587 list->qlen++;
1588 }
1589
1590 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1591 struct sk_buff *prev,
1592 struct sk_buff *next)
1593 {
1594 struct sk_buff *first = list->next;
1595 struct sk_buff *last = list->prev;
1596
1597 first->prev = prev;
1598 prev->next = first;
1599
1600 last->next = next;
1601 next->prev = last;
1602 }
1603
1604 /**
1605 * skb_queue_splice - join two skb lists, this is designed for stacks
1606 * @list: the new list to add
1607 * @head: the place to add it in the first list
1608 */
1609 static inline void skb_queue_splice(const struct sk_buff_head *list,
1610 struct sk_buff_head *head)
1611 {
1612 if (!skb_queue_empty(list)) {
1613 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1614 head->qlen += list->qlen;
1615 }
1616 }
1617
1618 /**
1619 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1620 * @list: the new list to add
1621 * @head: the place to add it in the first list
1622 *
1623 * The list at @list is reinitialised
1624 */
1625 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1626 struct sk_buff_head *head)
1627 {
1628 if (!skb_queue_empty(list)) {
1629 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1630 head->qlen += list->qlen;
1631 __skb_queue_head_init(list);
1632 }
1633 }
1634
1635 /**
1636 * skb_queue_splice_tail - join two skb lists, each list being a queue
1637 * @list: the new list to add
1638 * @head: the place to add it in the first list
1639 */
1640 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1641 struct sk_buff_head *head)
1642 {
1643 if (!skb_queue_empty(list)) {
1644 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1645 head->qlen += list->qlen;
1646 }
1647 }
1648
1649 /**
1650 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1651 * @list: the new list to add
1652 * @head: the place to add it in the first list
1653 *
1654 * Each of the lists is a queue.
1655 * The list at @list is reinitialised
1656 */
1657 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1658 struct sk_buff_head *head)
1659 {
1660 if (!skb_queue_empty(list)) {
1661 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1662 head->qlen += list->qlen;
1663 __skb_queue_head_init(list);
1664 }
1665 }
1666
1667 /**
1668 * __skb_queue_after - queue a buffer at the list head
1669 * @list: list to use
1670 * @prev: place after this buffer
1671 * @newsk: buffer to queue
1672 *
1673 * Queue a buffer int the middle of a list. This function takes no locks
1674 * and you must therefore hold required locks before calling it.
1675 *
1676 * A buffer cannot be placed on two lists at the same time.
1677 */
1678 static inline void __skb_queue_after(struct sk_buff_head *list,
1679 struct sk_buff *prev,
1680 struct sk_buff *newsk)
1681 {
1682 __skb_insert(newsk, prev, prev->next, list);
1683 }
1684
1685 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1686 struct sk_buff_head *list);
1687
1688 static inline void __skb_queue_before(struct sk_buff_head *list,
1689 struct sk_buff *next,
1690 struct sk_buff *newsk)
1691 {
1692 __skb_insert(newsk, next->prev, next, list);
1693 }
1694
1695 /**
1696 * __skb_queue_head - queue a buffer at the list head
1697 * @list: list to use
1698 * @newsk: buffer to queue
1699 *
1700 * Queue a buffer at the start of a list. This function takes no locks
1701 * and you must therefore hold required locks before calling it.
1702 *
1703 * A buffer cannot be placed on two lists at the same time.
1704 */
1705 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1706 static inline void __skb_queue_head(struct sk_buff_head *list,
1707 struct sk_buff *newsk)
1708 {
1709 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1710 }
1711
1712 /**
1713 * __skb_queue_tail - queue a buffer at the list tail
1714 * @list: list to use
1715 * @newsk: buffer to queue
1716 *
1717 * Queue a buffer at the end of a list. This function takes no locks
1718 * and you must therefore hold required locks before calling it.
1719 *
1720 * A buffer cannot be placed on two lists at the same time.
1721 */
1722 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1723 static inline void __skb_queue_tail(struct sk_buff_head *list,
1724 struct sk_buff *newsk)
1725 {
1726 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1727 }
1728
1729 /*
1730 * remove sk_buff from list. _Must_ be called atomically, and with
1731 * the list known..
1732 */
1733 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1734 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1735 {
1736 struct sk_buff *next, *prev;
1737
1738 list->qlen--;
1739 next = skb->next;
1740 prev = skb->prev;
1741 skb->next = skb->prev = NULL;
1742 next->prev = prev;
1743 prev->next = next;
1744 }
1745
1746 /**
1747 * __skb_dequeue - remove from the head of the queue
1748 * @list: list to dequeue from
1749 *
1750 * Remove the head of the list. This function does not take any locks
1751 * so must be used with appropriate locks held only. The head item is
1752 * returned or %NULL if the list is empty.
1753 */
1754 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1755 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1756 {
1757 struct sk_buff *skb = skb_peek(list);
1758 if (skb)
1759 __skb_unlink(skb, list);
1760 return skb;
1761 }
1762
1763 /**
1764 * __skb_dequeue_tail - remove from the tail of the queue
1765 * @list: list to dequeue from
1766 *
1767 * Remove the tail of the list. This function does not take any locks
1768 * so must be used with appropriate locks held only. The tail item is
1769 * returned or %NULL if the list is empty.
1770 */
1771 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1772 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1773 {
1774 struct sk_buff *skb = skb_peek_tail(list);
1775 if (skb)
1776 __skb_unlink(skb, list);
1777 return skb;
1778 }
1779
1780
1781 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1782 {
1783 return skb->data_len;
1784 }
1785
1786 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1787 {
1788 return skb->len - skb->data_len;
1789 }
1790
1791 static inline int skb_pagelen(const struct sk_buff *skb)
1792 {
1793 int i, len = 0;
1794
1795 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1796 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1797 return len + skb_headlen(skb);
1798 }
1799
1800 /**
1801 * __skb_fill_page_desc - initialise a paged fragment in an skb
1802 * @skb: buffer containing fragment to be initialised
1803 * @i: paged fragment index to initialise
1804 * @page: the page to use for this fragment
1805 * @off: the offset to the data with @page
1806 * @size: the length of the data
1807 *
1808 * Initialises the @i'th fragment of @skb to point to &size bytes at
1809 * offset @off within @page.
1810 *
1811 * Does not take any additional reference on the fragment.
1812 */
1813 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1814 struct page *page, int off, int size)
1815 {
1816 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1817
1818 /*
1819 * Propagate page pfmemalloc to the skb if we can. The problem is
1820 * that not all callers have unique ownership of the page but rely
1821 * on page_is_pfmemalloc doing the right thing(tm).
1822 */
1823 frag->page.p = page;
1824 frag->page_offset = off;
1825 skb_frag_size_set(frag, size);
1826
1827 page = compound_head(page);
1828 if (page_is_pfmemalloc(page))
1829 skb->pfmemalloc = true;
1830 }
1831
1832 /**
1833 * skb_fill_page_desc - initialise a paged fragment in an skb
1834 * @skb: buffer containing fragment to be initialised
1835 * @i: paged fragment index to initialise
1836 * @page: the page to use for this fragment
1837 * @off: the offset to the data with @page
1838 * @size: the length of the data
1839 *
1840 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1841 * @skb to point to @size bytes at offset @off within @page. In
1842 * addition updates @skb such that @i is the last fragment.
1843 *
1844 * Does not take any additional reference on the fragment.
1845 */
1846 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1847 struct page *page, int off, int size)
1848 {
1849 __skb_fill_page_desc(skb, i, page, off, size);
1850 skb_shinfo(skb)->nr_frags = i + 1;
1851 }
1852
1853 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1854 int size, unsigned int truesize);
1855
1856 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1857 unsigned int truesize);
1858
1859 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1860 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1861 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1862
1863 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1864 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1865 {
1866 return skb->head + skb->tail;
1867 }
1868
1869 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1870 {
1871 skb->tail = skb->data - skb->head;
1872 }
1873
1874 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1875 {
1876 skb_reset_tail_pointer(skb);
1877 skb->tail += offset;
1878 }
1879
1880 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1881 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1882 {
1883 return skb->tail;
1884 }
1885
1886 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1887 {
1888 skb->tail = skb->data;
1889 }
1890
1891 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1892 {
1893 skb->tail = skb->data + offset;
1894 }
1895
1896 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1897
1898 /*
1899 * Add data to an sk_buff
1900 */
1901 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1902 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1903 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1904 {
1905 unsigned char *tmp = skb_tail_pointer(skb);
1906 SKB_LINEAR_ASSERT(skb);
1907 skb->tail += len;
1908 skb->len += len;
1909 return tmp;
1910 }
1911
1912 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1913 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1914 {
1915 skb->data -= len;
1916 skb->len += len;
1917 return skb->data;
1918 }
1919
1920 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1921 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1922 {
1923 skb->len -= len;
1924 BUG_ON(skb->len < skb->data_len);
1925 return skb->data += len;
1926 }
1927
1928 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1929 {
1930 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1931 }
1932
1933 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1934
1935 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1936 {
1937 if (len > skb_headlen(skb) &&
1938 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1939 return NULL;
1940 skb->len -= len;
1941 return skb->data += len;
1942 }
1943
1944 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1945 {
1946 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1947 }
1948
1949 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1950 {
1951 if (likely(len <= skb_headlen(skb)))
1952 return 1;
1953 if (unlikely(len > skb->len))
1954 return 0;
1955 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1956 }
1957
1958 /**
1959 * skb_headroom - bytes at buffer head
1960 * @skb: buffer to check
1961 *
1962 * Return the number of bytes of free space at the head of an &sk_buff.
1963 */
1964 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1965 {
1966 return skb->data - skb->head;
1967 }
1968
1969 /**
1970 * skb_tailroom - bytes at buffer end
1971 * @skb: buffer to check
1972 *
1973 * Return the number of bytes of free space at the tail of an sk_buff
1974 */
1975 static inline int skb_tailroom(const struct sk_buff *skb)
1976 {
1977 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1978 }
1979
1980 /**
1981 * skb_availroom - bytes at buffer end
1982 * @skb: buffer to check
1983 *
1984 * Return the number of bytes of free space at the tail of an sk_buff
1985 * allocated by sk_stream_alloc()
1986 */
1987 static inline int skb_availroom(const struct sk_buff *skb)
1988 {
1989 if (skb_is_nonlinear(skb))
1990 return 0;
1991
1992 return skb->end - skb->tail - skb->reserved_tailroom;
1993 }
1994
1995 /**
1996 * skb_reserve - adjust headroom
1997 * @skb: buffer to alter
1998 * @len: bytes to move
1999 *
2000 * Increase the headroom of an empty &sk_buff by reducing the tail
2001 * room. This is only allowed for an empty buffer.
2002 */
2003 static inline void skb_reserve(struct sk_buff *skb, int len)
2004 {
2005 skb->data += len;
2006 skb->tail += len;
2007 }
2008
2009 /**
2010 * skb_tailroom_reserve - adjust reserved_tailroom
2011 * @skb: buffer to alter
2012 * @mtu: maximum amount of headlen permitted
2013 * @needed_tailroom: minimum amount of reserved_tailroom
2014 *
2015 * Set reserved_tailroom so that headlen can be as large as possible but
2016 * not larger than mtu and tailroom cannot be smaller than
2017 * needed_tailroom.
2018 * The required headroom should already have been reserved before using
2019 * this function.
2020 */
2021 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2022 unsigned int needed_tailroom)
2023 {
2024 SKB_LINEAR_ASSERT(skb);
2025 if (mtu < skb_tailroom(skb) - needed_tailroom)
2026 /* use at most mtu */
2027 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2028 else
2029 /* use up to all available space */
2030 skb->reserved_tailroom = needed_tailroom;
2031 }
2032
2033 #define ENCAP_TYPE_ETHER 0
2034 #define ENCAP_TYPE_IPPROTO 1
2035
2036 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2037 __be16 protocol)
2038 {
2039 skb->inner_protocol = protocol;
2040 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2041 }
2042
2043 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2044 __u8 ipproto)
2045 {
2046 skb->inner_ipproto = ipproto;
2047 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2048 }
2049
2050 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2051 {
2052 skb->inner_mac_header = skb->mac_header;
2053 skb->inner_network_header = skb->network_header;
2054 skb->inner_transport_header = skb->transport_header;
2055 }
2056
2057 static inline void skb_reset_mac_len(struct sk_buff *skb)
2058 {
2059 skb->mac_len = skb->network_header - skb->mac_header;
2060 }
2061
2062 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2063 *skb)
2064 {
2065 return skb->head + skb->inner_transport_header;
2066 }
2067
2068 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2069 {
2070 return skb_inner_transport_header(skb) - skb->data;
2071 }
2072
2073 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2074 {
2075 skb->inner_transport_header = skb->data - skb->head;
2076 }
2077
2078 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2079 const int offset)
2080 {
2081 skb_reset_inner_transport_header(skb);
2082 skb->inner_transport_header += offset;
2083 }
2084
2085 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2086 {
2087 return skb->head + skb->inner_network_header;
2088 }
2089
2090 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2091 {
2092 skb->inner_network_header = skb->data - skb->head;
2093 }
2094
2095 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2096 const int offset)
2097 {
2098 skb_reset_inner_network_header(skb);
2099 skb->inner_network_header += offset;
2100 }
2101
2102 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2103 {
2104 return skb->head + skb->inner_mac_header;
2105 }
2106
2107 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2108 {
2109 skb->inner_mac_header = skb->data - skb->head;
2110 }
2111
2112 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2113 const int offset)
2114 {
2115 skb_reset_inner_mac_header(skb);
2116 skb->inner_mac_header += offset;
2117 }
2118 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2119 {
2120 return skb->transport_header != (typeof(skb->transport_header))~0U;
2121 }
2122
2123 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2124 {
2125 return skb->head + skb->transport_header;
2126 }
2127
2128 static inline void skb_reset_transport_header(struct sk_buff *skb)
2129 {
2130 skb->transport_header = skb->data - skb->head;
2131 }
2132
2133 static inline void skb_set_transport_header(struct sk_buff *skb,
2134 const int offset)
2135 {
2136 skb_reset_transport_header(skb);
2137 skb->transport_header += offset;
2138 }
2139
2140 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2141 {
2142 return skb->head + skb->network_header;
2143 }
2144
2145 static inline void skb_reset_network_header(struct sk_buff *skb)
2146 {
2147 skb->network_header = skb->data - skb->head;
2148 }
2149
2150 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2151 {
2152 skb_reset_network_header(skb);
2153 skb->network_header += offset;
2154 }
2155
2156 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2157 {
2158 return skb->head + skb->mac_header;
2159 }
2160
2161 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2162 {
2163 return skb->mac_header != (typeof(skb->mac_header))~0U;
2164 }
2165
2166 static inline void skb_reset_mac_header(struct sk_buff *skb)
2167 {
2168 skb->mac_header = skb->data - skb->head;
2169 }
2170
2171 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2172 {
2173 skb_reset_mac_header(skb);
2174 skb->mac_header += offset;
2175 }
2176
2177 static inline void skb_pop_mac_header(struct sk_buff *skb)
2178 {
2179 skb->mac_header = skb->network_header;
2180 }
2181
2182 static inline void skb_probe_transport_header(struct sk_buff *skb,
2183 const int offset_hint)
2184 {
2185 struct flow_keys keys;
2186
2187 if (skb_transport_header_was_set(skb))
2188 return;
2189 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2190 skb_set_transport_header(skb, keys.control.thoff);
2191 else
2192 skb_set_transport_header(skb, offset_hint);
2193 }
2194
2195 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2196 {
2197 if (skb_mac_header_was_set(skb)) {
2198 const unsigned char *old_mac = skb_mac_header(skb);
2199
2200 skb_set_mac_header(skb, -skb->mac_len);
2201 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2202 }
2203 }
2204
2205 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2206 {
2207 return skb->csum_start - skb_headroom(skb);
2208 }
2209
2210 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2211 {
2212 return skb->head + skb->csum_start;
2213 }
2214
2215 static inline int skb_transport_offset(const struct sk_buff *skb)
2216 {
2217 return skb_transport_header(skb) - skb->data;
2218 }
2219
2220 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2221 {
2222 return skb->transport_header - skb->network_header;
2223 }
2224
2225 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2226 {
2227 return skb->inner_transport_header - skb->inner_network_header;
2228 }
2229
2230 static inline int skb_network_offset(const struct sk_buff *skb)
2231 {
2232 return skb_network_header(skb) - skb->data;
2233 }
2234
2235 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2236 {
2237 return skb_inner_network_header(skb) - skb->data;
2238 }
2239
2240 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2241 {
2242 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2243 }
2244
2245 /*
2246 * CPUs often take a performance hit when accessing unaligned memory
2247 * locations. The actual performance hit varies, it can be small if the
2248 * hardware handles it or large if we have to take an exception and fix it
2249 * in software.
2250 *
2251 * Since an ethernet header is 14 bytes network drivers often end up with
2252 * the IP header at an unaligned offset. The IP header can be aligned by
2253 * shifting the start of the packet by 2 bytes. Drivers should do this
2254 * with:
2255 *
2256 * skb_reserve(skb, NET_IP_ALIGN);
2257 *
2258 * The downside to this alignment of the IP header is that the DMA is now
2259 * unaligned. On some architectures the cost of an unaligned DMA is high
2260 * and this cost outweighs the gains made by aligning the IP header.
2261 *
2262 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2263 * to be overridden.
2264 */
2265 #ifndef NET_IP_ALIGN
2266 #define NET_IP_ALIGN 2
2267 #endif
2268
2269 /*
2270 * The networking layer reserves some headroom in skb data (via
2271 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2272 * the header has to grow. In the default case, if the header has to grow
2273 * 32 bytes or less we avoid the reallocation.
2274 *
2275 * Unfortunately this headroom changes the DMA alignment of the resulting
2276 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2277 * on some architectures. An architecture can override this value,
2278 * perhaps setting it to a cacheline in size (since that will maintain
2279 * cacheline alignment of the DMA). It must be a power of 2.
2280 *
2281 * Various parts of the networking layer expect at least 32 bytes of
2282 * headroom, you should not reduce this.
2283 *
2284 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2285 * to reduce average number of cache lines per packet.
2286 * get_rps_cpus() for example only access one 64 bytes aligned block :
2287 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2288 */
2289 #ifndef NET_SKB_PAD
2290 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2291 #endif
2292
2293 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2294
2295 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2296 {
2297 if (unlikely(skb_is_nonlinear(skb))) {
2298 WARN_ON(1);
2299 return;
2300 }
2301 skb->len = len;
2302 skb_set_tail_pointer(skb, len);
2303 }
2304
2305 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2306 {
2307 __skb_set_length(skb, len);
2308 }
2309
2310 void skb_trim(struct sk_buff *skb, unsigned int len);
2311
2312 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2313 {
2314 if (skb->data_len)
2315 return ___pskb_trim(skb, len);
2316 __skb_trim(skb, len);
2317 return 0;
2318 }
2319
2320 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2321 {
2322 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2323 }
2324
2325 /**
2326 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2327 * @skb: buffer to alter
2328 * @len: new length
2329 *
2330 * This is identical to pskb_trim except that the caller knows that
2331 * the skb is not cloned so we should never get an error due to out-
2332 * of-memory.
2333 */
2334 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2335 {
2336 int err = pskb_trim(skb, len);
2337 BUG_ON(err);
2338 }
2339
2340 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2341 {
2342 unsigned int diff = len - skb->len;
2343
2344 if (skb_tailroom(skb) < diff) {
2345 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2346 GFP_ATOMIC);
2347 if (ret)
2348 return ret;
2349 }
2350 __skb_set_length(skb, len);
2351 return 0;
2352 }
2353
2354 /**
2355 * skb_orphan - orphan a buffer
2356 * @skb: buffer to orphan
2357 *
2358 * If a buffer currently has an owner then we call the owner's
2359 * destructor function and make the @skb unowned. The buffer continues
2360 * to exist but is no longer charged to its former owner.
2361 */
2362 static inline void skb_orphan(struct sk_buff *skb)
2363 {
2364 if (skb->destructor) {
2365 skb->destructor(skb);
2366 skb->destructor = NULL;
2367 skb->sk = NULL;
2368 } else {
2369 BUG_ON(skb->sk);
2370 }
2371 }
2372
2373 /**
2374 * skb_orphan_frags - orphan the frags contained in a buffer
2375 * @skb: buffer to orphan frags from
2376 * @gfp_mask: allocation mask for replacement pages
2377 *
2378 * For each frag in the SKB which needs a destructor (i.e. has an
2379 * owner) create a copy of that frag and release the original
2380 * page by calling the destructor.
2381 */
2382 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2383 {
2384 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2385 return 0;
2386 return skb_copy_ubufs(skb, gfp_mask);
2387 }
2388
2389 /**
2390 * __skb_queue_purge - empty a list
2391 * @list: list to empty
2392 *
2393 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2394 * the list and one reference dropped. This function does not take the
2395 * list lock and the caller must hold the relevant locks to use it.
2396 */
2397 void skb_queue_purge(struct sk_buff_head *list);
2398 static inline void __skb_queue_purge(struct sk_buff_head *list)
2399 {
2400 struct sk_buff *skb;
2401 while ((skb = __skb_dequeue(list)) != NULL)
2402 kfree_skb(skb);
2403 }
2404
2405 void skb_rbtree_purge(struct rb_root *root);
2406
2407 void *netdev_alloc_frag(unsigned int fragsz);
2408
2409 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2410 gfp_t gfp_mask);
2411
2412 /**
2413 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2414 * @dev: network device to receive on
2415 * @length: length to allocate
2416 *
2417 * Allocate a new &sk_buff and assign it a usage count of one. The
2418 * buffer has unspecified headroom built in. Users should allocate
2419 * the headroom they think they need without accounting for the
2420 * built in space. The built in space is used for optimisations.
2421 *
2422 * %NULL is returned if there is no free memory. Although this function
2423 * allocates memory it can be called from an interrupt.
2424 */
2425 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2426 unsigned int length)
2427 {
2428 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2429 }
2430
2431 /* legacy helper around __netdev_alloc_skb() */
2432 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2433 gfp_t gfp_mask)
2434 {
2435 return __netdev_alloc_skb(NULL, length, gfp_mask);
2436 }
2437
2438 /* legacy helper around netdev_alloc_skb() */
2439 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2440 {
2441 return netdev_alloc_skb(NULL, length);
2442 }
2443
2444
2445 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2446 unsigned int length, gfp_t gfp)
2447 {
2448 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2449
2450 if (NET_IP_ALIGN && skb)
2451 skb_reserve(skb, NET_IP_ALIGN);
2452 return skb;
2453 }
2454
2455 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2456 unsigned int length)
2457 {
2458 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2459 }
2460
2461 static inline void skb_free_frag(void *addr)
2462 {
2463 __free_page_frag(addr);
2464 }
2465
2466 void *napi_alloc_frag(unsigned int fragsz);
2467 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2468 unsigned int length, gfp_t gfp_mask);
2469 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2470 unsigned int length)
2471 {
2472 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2473 }
2474 void napi_consume_skb(struct sk_buff *skb, int budget);
2475
2476 void __kfree_skb_flush(void);
2477 void __kfree_skb_defer(struct sk_buff *skb);
2478
2479 /**
2480 * __dev_alloc_pages - allocate page for network Rx
2481 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2482 * @order: size of the allocation
2483 *
2484 * Allocate a new page.
2485 *
2486 * %NULL is returned if there is no free memory.
2487 */
2488 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2489 unsigned int order)
2490 {
2491 /* This piece of code contains several assumptions.
2492 * 1. This is for device Rx, therefor a cold page is preferred.
2493 * 2. The expectation is the user wants a compound page.
2494 * 3. If requesting a order 0 page it will not be compound
2495 * due to the check to see if order has a value in prep_new_page
2496 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2497 * code in gfp_to_alloc_flags that should be enforcing this.
2498 */
2499 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2500
2501 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2502 }
2503
2504 static inline struct page *dev_alloc_pages(unsigned int order)
2505 {
2506 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2507 }
2508
2509 /**
2510 * __dev_alloc_page - allocate a page for network Rx
2511 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2512 *
2513 * Allocate a new page.
2514 *
2515 * %NULL is returned if there is no free memory.
2516 */
2517 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2518 {
2519 return __dev_alloc_pages(gfp_mask, 0);
2520 }
2521
2522 static inline struct page *dev_alloc_page(void)
2523 {
2524 return dev_alloc_pages(0);
2525 }
2526
2527 /**
2528 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2529 * @page: The page that was allocated from skb_alloc_page
2530 * @skb: The skb that may need pfmemalloc set
2531 */
2532 static inline void skb_propagate_pfmemalloc(struct page *page,
2533 struct sk_buff *skb)
2534 {
2535 if (page_is_pfmemalloc(page))
2536 skb->pfmemalloc = true;
2537 }
2538
2539 /**
2540 * skb_frag_page - retrieve the page referred to by a paged fragment
2541 * @frag: the paged fragment
2542 *
2543 * Returns the &struct page associated with @frag.
2544 */
2545 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2546 {
2547 return frag->page.p;
2548 }
2549
2550 /**
2551 * __skb_frag_ref - take an addition reference on a paged fragment.
2552 * @frag: the paged fragment
2553 *
2554 * Takes an additional reference on the paged fragment @frag.
2555 */
2556 static inline void __skb_frag_ref(skb_frag_t *frag)
2557 {
2558 get_page(skb_frag_page(frag));
2559 }
2560
2561 /**
2562 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2563 * @skb: the buffer
2564 * @f: the fragment offset.
2565 *
2566 * Takes an additional reference on the @f'th paged fragment of @skb.
2567 */
2568 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2569 {
2570 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2571 }
2572
2573 /**
2574 * __skb_frag_unref - release a reference on a paged fragment.
2575 * @frag: the paged fragment
2576 *
2577 * Releases a reference on the paged fragment @frag.
2578 */
2579 static inline void __skb_frag_unref(skb_frag_t *frag)
2580 {
2581 put_page(skb_frag_page(frag));
2582 }
2583
2584 /**
2585 * skb_frag_unref - release a reference on a paged fragment of an skb.
2586 * @skb: the buffer
2587 * @f: the fragment offset
2588 *
2589 * Releases a reference on the @f'th paged fragment of @skb.
2590 */
2591 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2592 {
2593 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2594 }
2595
2596 /**
2597 * skb_frag_address - gets the address of the data contained in a paged fragment
2598 * @frag: the paged fragment buffer
2599 *
2600 * Returns the address of the data within @frag. The page must already
2601 * be mapped.
2602 */
2603 static inline void *skb_frag_address(const skb_frag_t *frag)
2604 {
2605 return page_address(skb_frag_page(frag)) + frag->page_offset;
2606 }
2607
2608 /**
2609 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2610 * @frag: the paged fragment buffer
2611 *
2612 * Returns the address of the data within @frag. Checks that the page
2613 * is mapped and returns %NULL otherwise.
2614 */
2615 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2616 {
2617 void *ptr = page_address(skb_frag_page(frag));
2618 if (unlikely(!ptr))
2619 return NULL;
2620
2621 return ptr + frag->page_offset;
2622 }
2623
2624 /**
2625 * __skb_frag_set_page - sets the page contained in a paged fragment
2626 * @frag: the paged fragment
2627 * @page: the page to set
2628 *
2629 * Sets the fragment @frag to contain @page.
2630 */
2631 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2632 {
2633 frag->page.p = page;
2634 }
2635
2636 /**
2637 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2638 * @skb: the buffer
2639 * @f: the fragment offset
2640 * @page: the page to set
2641 *
2642 * Sets the @f'th fragment of @skb to contain @page.
2643 */
2644 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2645 struct page *page)
2646 {
2647 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2648 }
2649
2650 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2651
2652 /**
2653 * skb_frag_dma_map - maps a paged fragment via the DMA API
2654 * @dev: the device to map the fragment to
2655 * @frag: the paged fragment to map
2656 * @offset: the offset within the fragment (starting at the
2657 * fragment's own offset)
2658 * @size: the number of bytes to map
2659 * @dir: the direction of the mapping (%PCI_DMA_*)
2660 *
2661 * Maps the page associated with @frag to @device.
2662 */
2663 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2664 const skb_frag_t *frag,
2665 size_t offset, size_t size,
2666 enum dma_data_direction dir)
2667 {
2668 return dma_map_page(dev, skb_frag_page(frag),
2669 frag->page_offset + offset, size, dir);
2670 }
2671
2672 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2673 gfp_t gfp_mask)
2674 {
2675 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2676 }
2677
2678
2679 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2680 gfp_t gfp_mask)
2681 {
2682 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2683 }
2684
2685
2686 /**
2687 * skb_clone_writable - is the header of a clone writable
2688 * @skb: buffer to check
2689 * @len: length up to which to write
2690 *
2691 * Returns true if modifying the header part of the cloned buffer
2692 * does not requires the data to be copied.
2693 */
2694 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2695 {
2696 return !skb_header_cloned(skb) &&
2697 skb_headroom(skb) + len <= skb->hdr_len;
2698 }
2699
2700 static inline int skb_try_make_writable(struct sk_buff *skb,
2701 unsigned int write_len)
2702 {
2703 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2704 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2705 }
2706
2707 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2708 int cloned)
2709 {
2710 int delta = 0;
2711
2712 if (headroom > skb_headroom(skb))
2713 delta = headroom - skb_headroom(skb);
2714
2715 if (delta || cloned)
2716 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2717 GFP_ATOMIC);
2718 return 0;
2719 }
2720
2721 /**
2722 * skb_cow - copy header of skb when it is required
2723 * @skb: buffer to cow
2724 * @headroom: needed headroom
2725 *
2726 * If the skb passed lacks sufficient headroom or its data part
2727 * is shared, data is reallocated. If reallocation fails, an error
2728 * is returned and original skb is not changed.
2729 *
2730 * The result is skb with writable area skb->head...skb->tail
2731 * and at least @headroom of space at head.
2732 */
2733 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2734 {
2735 return __skb_cow(skb, headroom, skb_cloned(skb));
2736 }
2737
2738 /**
2739 * skb_cow_head - skb_cow but only making the head writable
2740 * @skb: buffer to cow
2741 * @headroom: needed headroom
2742 *
2743 * This function is identical to skb_cow except that we replace the
2744 * skb_cloned check by skb_header_cloned. It should be used when
2745 * you only need to push on some header and do not need to modify
2746 * the data.
2747 */
2748 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2749 {
2750 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2751 }
2752
2753 /**
2754 * skb_padto - pad an skbuff up to a minimal size
2755 * @skb: buffer to pad
2756 * @len: minimal length
2757 *
2758 * Pads up a buffer to ensure the trailing bytes exist and are
2759 * blanked. If the buffer already contains sufficient data it
2760 * is untouched. Otherwise it is extended. Returns zero on
2761 * success. The skb is freed on error.
2762 */
2763 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2764 {
2765 unsigned int size = skb->len;
2766 if (likely(size >= len))
2767 return 0;
2768 return skb_pad(skb, len - size);
2769 }
2770
2771 /**
2772 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2773 * @skb: buffer to pad
2774 * @len: minimal length
2775 *
2776 * Pads up a buffer to ensure the trailing bytes exist and are
2777 * blanked. If the buffer already contains sufficient data it
2778 * is untouched. Otherwise it is extended. Returns zero on
2779 * success. The skb is freed on error.
2780 */
2781 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2782 {
2783 unsigned int size = skb->len;
2784
2785 if (unlikely(size < len)) {
2786 len -= size;
2787 if (skb_pad(skb, len))
2788 return -ENOMEM;
2789 __skb_put(skb, len);
2790 }
2791 return 0;
2792 }
2793
2794 static inline int skb_add_data(struct sk_buff *skb,
2795 struct iov_iter *from, int copy)
2796 {
2797 const int off = skb->len;
2798
2799 if (skb->ip_summed == CHECKSUM_NONE) {
2800 __wsum csum = 0;
2801 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2802 &csum, from) == copy) {
2803 skb->csum = csum_block_add(skb->csum, csum, off);
2804 return 0;
2805 }
2806 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2807 return 0;
2808
2809 __skb_trim(skb, off);
2810 return -EFAULT;
2811 }
2812
2813 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2814 const struct page *page, int off)
2815 {
2816 if (i) {
2817 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2818
2819 return page == skb_frag_page(frag) &&
2820 off == frag->page_offset + skb_frag_size(frag);
2821 }
2822 return false;
2823 }
2824
2825 static inline int __skb_linearize(struct sk_buff *skb)
2826 {
2827 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2828 }
2829
2830 /**
2831 * skb_linearize - convert paged skb to linear one
2832 * @skb: buffer to linarize
2833 *
2834 * If there is no free memory -ENOMEM is returned, otherwise zero
2835 * is returned and the old skb data released.
2836 */
2837 static inline int skb_linearize(struct sk_buff *skb)
2838 {
2839 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2840 }
2841
2842 /**
2843 * skb_has_shared_frag - can any frag be overwritten
2844 * @skb: buffer to test
2845 *
2846 * Return true if the skb has at least one frag that might be modified
2847 * by an external entity (as in vmsplice()/sendfile())
2848 */
2849 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2850 {
2851 return skb_is_nonlinear(skb) &&
2852 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2853 }
2854
2855 /**
2856 * skb_linearize_cow - make sure skb is linear and writable
2857 * @skb: buffer to process
2858 *
2859 * If there is no free memory -ENOMEM is returned, otherwise zero
2860 * is returned and the old skb data released.
2861 */
2862 static inline int skb_linearize_cow(struct sk_buff *skb)
2863 {
2864 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2865 __skb_linearize(skb) : 0;
2866 }
2867
2868 static __always_inline void
2869 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2870 unsigned int off)
2871 {
2872 if (skb->ip_summed == CHECKSUM_COMPLETE)
2873 skb->csum = csum_block_sub(skb->csum,
2874 csum_partial(start, len, 0), off);
2875 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2876 skb_checksum_start_offset(skb) < 0)
2877 skb->ip_summed = CHECKSUM_NONE;
2878 }
2879
2880 /**
2881 * skb_postpull_rcsum - update checksum for received skb after pull
2882 * @skb: buffer to update
2883 * @start: start of data before pull
2884 * @len: length of data pulled
2885 *
2886 * After doing a pull on a received packet, you need to call this to
2887 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2888 * CHECKSUM_NONE so that it can be recomputed from scratch.
2889 */
2890 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2891 const void *start, unsigned int len)
2892 {
2893 __skb_postpull_rcsum(skb, start, len, 0);
2894 }
2895
2896 static __always_inline void
2897 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2898 unsigned int off)
2899 {
2900 if (skb->ip_summed == CHECKSUM_COMPLETE)
2901 skb->csum = csum_block_add(skb->csum,
2902 csum_partial(start, len, 0), off);
2903 }
2904
2905 /**
2906 * skb_postpush_rcsum - update checksum for received skb after push
2907 * @skb: buffer to update
2908 * @start: start of data after push
2909 * @len: length of data pushed
2910 *
2911 * After doing a push on a received packet, you need to call this to
2912 * update the CHECKSUM_COMPLETE checksum.
2913 */
2914 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2915 const void *start, unsigned int len)
2916 {
2917 __skb_postpush_rcsum(skb, start, len, 0);
2918 }
2919
2920 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2921
2922 /**
2923 * skb_push_rcsum - push skb and update receive checksum
2924 * @skb: buffer to update
2925 * @len: length of data pulled
2926 *
2927 * This function performs an skb_push on the packet and updates
2928 * the CHECKSUM_COMPLETE checksum. It should be used on
2929 * receive path processing instead of skb_push unless you know
2930 * that the checksum difference is zero (e.g., a valid IP header)
2931 * or you are setting ip_summed to CHECKSUM_NONE.
2932 */
2933 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2934 unsigned int len)
2935 {
2936 skb_push(skb, len);
2937 skb_postpush_rcsum(skb, skb->data, len);
2938 return skb->data;
2939 }
2940
2941 /**
2942 * pskb_trim_rcsum - trim received skb and update checksum
2943 * @skb: buffer to trim
2944 * @len: new length
2945 *
2946 * This is exactly the same as pskb_trim except that it ensures the
2947 * checksum of received packets are still valid after the operation.
2948 */
2949
2950 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2951 {
2952 if (likely(len >= skb->len))
2953 return 0;
2954 if (skb->ip_summed == CHECKSUM_COMPLETE)
2955 skb->ip_summed = CHECKSUM_NONE;
2956 return __pskb_trim(skb, len);
2957 }
2958
2959 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2960 {
2961 if (skb->ip_summed == CHECKSUM_COMPLETE)
2962 skb->ip_summed = CHECKSUM_NONE;
2963 __skb_trim(skb, len);
2964 return 0;
2965 }
2966
2967 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
2968 {
2969 if (skb->ip_summed == CHECKSUM_COMPLETE)
2970 skb->ip_summed = CHECKSUM_NONE;
2971 return __skb_grow(skb, len);
2972 }
2973
2974 #define skb_queue_walk(queue, skb) \
2975 for (skb = (queue)->next; \
2976 skb != (struct sk_buff *)(queue); \
2977 skb = skb->next)
2978
2979 #define skb_queue_walk_safe(queue, skb, tmp) \
2980 for (skb = (queue)->next, tmp = skb->next; \
2981 skb != (struct sk_buff *)(queue); \
2982 skb = tmp, tmp = skb->next)
2983
2984 #define skb_queue_walk_from(queue, skb) \
2985 for (; skb != (struct sk_buff *)(queue); \
2986 skb = skb->next)
2987
2988 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2989 for (tmp = skb->next; \
2990 skb != (struct sk_buff *)(queue); \
2991 skb = tmp, tmp = skb->next)
2992
2993 #define skb_queue_reverse_walk(queue, skb) \
2994 for (skb = (queue)->prev; \
2995 skb != (struct sk_buff *)(queue); \
2996 skb = skb->prev)
2997
2998 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2999 for (skb = (queue)->prev, tmp = skb->prev; \
3000 skb != (struct sk_buff *)(queue); \
3001 skb = tmp, tmp = skb->prev)
3002
3003 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3004 for (tmp = skb->prev; \
3005 skb != (struct sk_buff *)(queue); \
3006 skb = tmp, tmp = skb->prev)
3007
3008 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3009 {
3010 return skb_shinfo(skb)->frag_list != NULL;
3011 }
3012
3013 static inline void skb_frag_list_init(struct sk_buff *skb)
3014 {
3015 skb_shinfo(skb)->frag_list = NULL;
3016 }
3017
3018 #define skb_walk_frags(skb, iter) \
3019 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3020
3021
3022 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3023 const struct sk_buff *skb);
3024 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3025 int *peeked, int *off, int *err,
3026 struct sk_buff **last);
3027 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3028 int *peeked, int *off, int *err);
3029 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3030 int *err);
3031 unsigned int datagram_poll(struct file *file, struct socket *sock,
3032 struct poll_table_struct *wait);
3033 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3034 struct iov_iter *to, int size);
3035 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3036 struct msghdr *msg, int size)
3037 {
3038 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3039 }
3040 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3041 struct msghdr *msg);
3042 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3043 struct iov_iter *from, int len);
3044 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3045 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3046 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3047 static inline void skb_free_datagram_locked(struct sock *sk,
3048 struct sk_buff *skb)
3049 {
3050 __skb_free_datagram_locked(sk, skb, 0);
3051 }
3052 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3053 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3054 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3055 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3056 int len, __wsum csum);
3057 ssize_t skb_socket_splice(struct sock *sk,
3058 struct pipe_inode_info *pipe,
3059 struct splice_pipe_desc *spd);
3060 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3061 struct pipe_inode_info *pipe, unsigned int len,
3062 unsigned int flags,
3063 ssize_t (*splice_cb)(struct sock *,
3064 struct pipe_inode_info *,
3065 struct splice_pipe_desc *));
3066 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3067 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3068 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3069 int len, int hlen);
3070 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3071 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3072 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3073 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3074 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3075 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3076 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3077 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3078 int skb_vlan_pop(struct sk_buff *skb);
3079 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3080 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3081 gfp_t gfp);
3082
3083 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3084 {
3085 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3086 }
3087
3088 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3089 {
3090 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3091 }
3092
3093 struct skb_checksum_ops {
3094 __wsum (*update)(const void *mem, int len, __wsum wsum);
3095 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3096 };
3097
3098 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3099 __wsum csum, const struct skb_checksum_ops *ops);
3100 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3101 __wsum csum);
3102
3103 static inline void * __must_check
3104 __skb_header_pointer(const struct sk_buff *skb, int offset,
3105 int len, void *data, int hlen, void *buffer)
3106 {
3107 if (hlen - offset >= len)
3108 return data + offset;
3109
3110 if (!skb ||
3111 skb_copy_bits(skb, offset, buffer, len) < 0)
3112 return NULL;
3113
3114 return buffer;
3115 }
3116
3117 static inline void * __must_check
3118 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3119 {
3120 return __skb_header_pointer(skb, offset, len, skb->data,
3121 skb_headlen(skb), buffer);
3122 }
3123
3124 /**
3125 * skb_needs_linearize - check if we need to linearize a given skb
3126 * depending on the given device features.
3127 * @skb: socket buffer to check
3128 * @features: net device features
3129 *
3130 * Returns true if either:
3131 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3132 * 2. skb is fragmented and the device does not support SG.
3133 */
3134 static inline bool skb_needs_linearize(struct sk_buff *skb,
3135 netdev_features_t features)
3136 {
3137 return skb_is_nonlinear(skb) &&
3138 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3139 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3140 }
3141
3142 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3143 void *to,
3144 const unsigned int len)
3145 {
3146 memcpy(to, skb->data, len);
3147 }
3148
3149 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3150 const int offset, void *to,
3151 const unsigned int len)
3152 {
3153 memcpy(to, skb->data + offset, len);
3154 }
3155
3156 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3157 const void *from,
3158 const unsigned int len)
3159 {
3160 memcpy(skb->data, from, len);
3161 }
3162
3163 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3164 const int offset,
3165 const void *from,
3166 const unsigned int len)
3167 {
3168 memcpy(skb->data + offset, from, len);
3169 }
3170
3171 void skb_init(void);
3172
3173 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3174 {
3175 return skb->tstamp;
3176 }
3177
3178 /**
3179 * skb_get_timestamp - get timestamp from a skb
3180 * @skb: skb to get stamp from
3181 * @stamp: pointer to struct timeval to store stamp in
3182 *
3183 * Timestamps are stored in the skb as offsets to a base timestamp.
3184 * This function converts the offset back to a struct timeval and stores
3185 * it in stamp.
3186 */
3187 static inline void skb_get_timestamp(const struct sk_buff *skb,
3188 struct timeval *stamp)
3189 {
3190 *stamp = ktime_to_timeval(skb->tstamp);
3191 }
3192
3193 static inline void skb_get_timestampns(const struct sk_buff *skb,
3194 struct timespec *stamp)
3195 {
3196 *stamp = ktime_to_timespec(skb->tstamp);
3197 }
3198
3199 static inline void __net_timestamp(struct sk_buff *skb)
3200 {
3201 skb->tstamp = ktime_get_real();
3202 }
3203
3204 static inline ktime_t net_timedelta(ktime_t t)
3205 {
3206 return ktime_sub(ktime_get_real(), t);
3207 }
3208
3209 static inline ktime_t net_invalid_timestamp(void)
3210 {
3211 return ktime_set(0, 0);
3212 }
3213
3214 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3215
3216 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3217
3218 void skb_clone_tx_timestamp(struct sk_buff *skb);
3219 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3220
3221 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3222
3223 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3224 {
3225 }
3226
3227 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3228 {
3229 return false;
3230 }
3231
3232 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3233
3234 /**
3235 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3236 *
3237 * PHY drivers may accept clones of transmitted packets for
3238 * timestamping via their phy_driver.txtstamp method. These drivers
3239 * must call this function to return the skb back to the stack with a
3240 * timestamp.
3241 *
3242 * @skb: clone of the the original outgoing packet
3243 * @hwtstamps: hardware time stamps
3244 *
3245 */
3246 void skb_complete_tx_timestamp(struct sk_buff *skb,
3247 struct skb_shared_hwtstamps *hwtstamps);
3248
3249 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3250 struct skb_shared_hwtstamps *hwtstamps,
3251 struct sock *sk, int tstype);
3252
3253 /**
3254 * skb_tstamp_tx - queue clone of skb with send time stamps
3255 * @orig_skb: the original outgoing packet
3256 * @hwtstamps: hardware time stamps, may be NULL if not available
3257 *
3258 * If the skb has a socket associated, then this function clones the
3259 * skb (thus sharing the actual data and optional structures), stores
3260 * the optional hardware time stamping information (if non NULL) or
3261 * generates a software time stamp (otherwise), then queues the clone
3262 * to the error queue of the socket. Errors are silently ignored.
3263 */
3264 void skb_tstamp_tx(struct sk_buff *orig_skb,
3265 struct skb_shared_hwtstamps *hwtstamps);
3266
3267 static inline void sw_tx_timestamp(struct sk_buff *skb)
3268 {
3269 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3270 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3271 skb_tstamp_tx(skb, NULL);
3272 }
3273
3274 /**
3275 * skb_tx_timestamp() - Driver hook for transmit timestamping
3276 *
3277 * Ethernet MAC Drivers should call this function in their hard_xmit()
3278 * function immediately before giving the sk_buff to the MAC hardware.
3279 *
3280 * Specifically, one should make absolutely sure that this function is
3281 * called before TX completion of this packet can trigger. Otherwise
3282 * the packet could potentially already be freed.
3283 *
3284 * @skb: A socket buffer.
3285 */
3286 static inline void skb_tx_timestamp(struct sk_buff *skb)
3287 {
3288 skb_clone_tx_timestamp(skb);
3289 sw_tx_timestamp(skb);
3290 }
3291
3292 /**
3293 * skb_complete_wifi_ack - deliver skb with wifi status
3294 *
3295 * @skb: the original outgoing packet
3296 * @acked: ack status
3297 *
3298 */
3299 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3300
3301 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3302 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3303
3304 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3305 {
3306 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3307 skb->csum_valid ||
3308 (skb->ip_summed == CHECKSUM_PARTIAL &&
3309 skb_checksum_start_offset(skb) >= 0));
3310 }
3311
3312 /**
3313 * skb_checksum_complete - Calculate checksum of an entire packet
3314 * @skb: packet to process
3315 *
3316 * This function calculates the checksum over the entire packet plus
3317 * the value of skb->csum. The latter can be used to supply the
3318 * checksum of a pseudo header as used by TCP/UDP. It returns the
3319 * checksum.
3320 *
3321 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3322 * this function can be used to verify that checksum on received
3323 * packets. In that case the function should return zero if the
3324 * checksum is correct. In particular, this function will return zero
3325 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3326 * hardware has already verified the correctness of the checksum.
3327 */
3328 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3329 {
3330 return skb_csum_unnecessary(skb) ?
3331 0 : __skb_checksum_complete(skb);
3332 }
3333
3334 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3335 {
3336 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3337 if (skb->csum_level == 0)
3338 skb->ip_summed = CHECKSUM_NONE;
3339 else
3340 skb->csum_level--;
3341 }
3342 }
3343
3344 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3345 {
3346 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3347 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3348 skb->csum_level++;
3349 } else if (skb->ip_summed == CHECKSUM_NONE) {
3350 skb->ip_summed = CHECKSUM_UNNECESSARY;
3351 skb->csum_level = 0;
3352 }
3353 }
3354
3355 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3356 {
3357 /* Mark current checksum as bad (typically called from GRO
3358 * path). In the case that ip_summed is CHECKSUM_NONE
3359 * this must be the first checksum encountered in the packet.
3360 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3361 * checksum after the last one validated. For UDP, a zero
3362 * checksum can not be marked as bad.
3363 */
3364
3365 if (skb->ip_summed == CHECKSUM_NONE ||
3366 skb->ip_summed == CHECKSUM_UNNECESSARY)
3367 skb->csum_bad = 1;
3368 }
3369
3370 /* Check if we need to perform checksum complete validation.
3371 *
3372 * Returns true if checksum complete is needed, false otherwise
3373 * (either checksum is unnecessary or zero checksum is allowed).
3374 */
3375 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3376 bool zero_okay,
3377 __sum16 check)
3378 {
3379 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3380 skb->csum_valid = 1;
3381 __skb_decr_checksum_unnecessary(skb);
3382 return false;
3383 }
3384
3385 return true;
3386 }
3387
3388 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3389 * in checksum_init.
3390 */
3391 #define CHECKSUM_BREAK 76
3392
3393 /* Unset checksum-complete
3394 *
3395 * Unset checksum complete can be done when packet is being modified
3396 * (uncompressed for instance) and checksum-complete value is
3397 * invalidated.
3398 */
3399 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3400 {
3401 if (skb->ip_summed == CHECKSUM_COMPLETE)
3402 skb->ip_summed = CHECKSUM_NONE;
3403 }
3404
3405 /* Validate (init) checksum based on checksum complete.
3406 *
3407 * Return values:
3408 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3409 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3410 * checksum is stored in skb->csum for use in __skb_checksum_complete
3411 * non-zero: value of invalid checksum
3412 *
3413 */
3414 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3415 bool complete,
3416 __wsum psum)
3417 {
3418 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3419 if (!csum_fold(csum_add(psum, skb->csum))) {
3420 skb->csum_valid = 1;
3421 return 0;
3422 }
3423 } else if (skb->csum_bad) {
3424 /* ip_summed == CHECKSUM_NONE in this case */
3425 return (__force __sum16)1;
3426 }
3427
3428 skb->csum = psum;
3429
3430 if (complete || skb->len <= CHECKSUM_BREAK) {
3431 __sum16 csum;
3432
3433 csum = __skb_checksum_complete(skb);
3434 skb->csum_valid = !csum;
3435 return csum;
3436 }
3437
3438 return 0;
3439 }
3440
3441 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3442 {
3443 return 0;
3444 }
3445
3446 /* Perform checksum validate (init). Note that this is a macro since we only
3447 * want to calculate the pseudo header which is an input function if necessary.
3448 * First we try to validate without any computation (checksum unnecessary) and
3449 * then calculate based on checksum complete calling the function to compute
3450 * pseudo header.
3451 *
3452 * Return values:
3453 * 0: checksum is validated or try to in skb_checksum_complete
3454 * non-zero: value of invalid checksum
3455 */
3456 #define __skb_checksum_validate(skb, proto, complete, \
3457 zero_okay, check, compute_pseudo) \
3458 ({ \
3459 __sum16 __ret = 0; \
3460 skb->csum_valid = 0; \
3461 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3462 __ret = __skb_checksum_validate_complete(skb, \
3463 complete, compute_pseudo(skb, proto)); \
3464 __ret; \
3465 })
3466
3467 #define skb_checksum_init(skb, proto, compute_pseudo) \
3468 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3469
3470 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3471 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3472
3473 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3474 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3475
3476 #define skb_checksum_validate_zero_check(skb, proto, check, \
3477 compute_pseudo) \
3478 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3479
3480 #define skb_checksum_simple_validate(skb) \
3481 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3482
3483 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3484 {
3485 return (skb->ip_summed == CHECKSUM_NONE &&
3486 skb->csum_valid && !skb->csum_bad);
3487 }
3488
3489 static inline void __skb_checksum_convert(struct sk_buff *skb,
3490 __sum16 check, __wsum pseudo)
3491 {
3492 skb->csum = ~pseudo;
3493 skb->ip_summed = CHECKSUM_COMPLETE;
3494 }
3495
3496 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3497 do { \
3498 if (__skb_checksum_convert_check(skb)) \
3499 __skb_checksum_convert(skb, check, \
3500 compute_pseudo(skb, proto)); \
3501 } while (0)
3502
3503 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3504 u16 start, u16 offset)
3505 {
3506 skb->ip_summed = CHECKSUM_PARTIAL;
3507 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3508 skb->csum_offset = offset - start;
3509 }
3510
3511 /* Update skbuf and packet to reflect the remote checksum offload operation.
3512 * When called, ptr indicates the starting point for skb->csum when
3513 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3514 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3515 */
3516 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3517 int start, int offset, bool nopartial)
3518 {
3519 __wsum delta;
3520
3521 if (!nopartial) {
3522 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3523 return;
3524 }
3525
3526 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3527 __skb_checksum_complete(skb);
3528 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3529 }
3530
3531 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3532
3533 /* Adjust skb->csum since we changed the packet */
3534 skb->csum = csum_add(skb->csum, delta);
3535 }
3536
3537 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3538 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3539 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3540 {
3541 if (nfct && atomic_dec_and_test(&nfct->use))
3542 nf_conntrack_destroy(nfct);
3543 }
3544 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3545 {
3546 if (nfct)
3547 atomic_inc(&nfct->use);
3548 }
3549 #endif
3550 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3551 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3552 {
3553 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3554 kfree(nf_bridge);
3555 }
3556 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3557 {
3558 if (nf_bridge)
3559 atomic_inc(&nf_bridge->use);
3560 }
3561 #endif /* CONFIG_BRIDGE_NETFILTER */
3562 static inline void nf_reset(struct sk_buff *skb)
3563 {
3564 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3565 nf_conntrack_put(skb->nfct);
3566 skb->nfct = NULL;
3567 #endif
3568 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3569 nf_bridge_put(skb->nf_bridge);
3570 skb->nf_bridge = NULL;
3571 #endif
3572 }
3573
3574 static inline void nf_reset_trace(struct sk_buff *skb)
3575 {
3576 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3577 skb->nf_trace = 0;
3578 #endif
3579 }
3580
3581 /* Note: This doesn't put any conntrack and bridge info in dst. */
3582 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3583 bool copy)
3584 {
3585 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3586 dst->nfct = src->nfct;
3587 nf_conntrack_get(src->nfct);
3588 if (copy)
3589 dst->nfctinfo = src->nfctinfo;
3590 #endif
3591 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3592 dst->nf_bridge = src->nf_bridge;
3593 nf_bridge_get(src->nf_bridge);
3594 #endif
3595 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3596 if (copy)
3597 dst->nf_trace = src->nf_trace;
3598 #endif
3599 }
3600
3601 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3602 {
3603 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3604 nf_conntrack_put(dst->nfct);
3605 #endif
3606 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3607 nf_bridge_put(dst->nf_bridge);
3608 #endif
3609 __nf_copy(dst, src, true);
3610 }
3611
3612 #ifdef CONFIG_NETWORK_SECMARK
3613 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3614 {
3615 to->secmark = from->secmark;
3616 }
3617
3618 static inline void skb_init_secmark(struct sk_buff *skb)
3619 {
3620 skb->secmark = 0;
3621 }
3622 #else
3623 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3624 { }
3625
3626 static inline void skb_init_secmark(struct sk_buff *skb)
3627 { }
3628 #endif
3629
3630 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3631 {
3632 return !skb->destructor &&
3633 #if IS_ENABLED(CONFIG_XFRM)
3634 !skb->sp &&
3635 #endif
3636 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3637 !skb->nfct &&
3638 #endif
3639 !skb->_skb_refdst &&
3640 !skb_has_frag_list(skb);
3641 }
3642
3643 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3644 {
3645 skb->queue_mapping = queue_mapping;
3646 }
3647
3648 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3649 {
3650 return skb->queue_mapping;
3651 }
3652
3653 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3654 {
3655 to->queue_mapping = from->queue_mapping;
3656 }
3657
3658 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3659 {
3660 skb->queue_mapping = rx_queue + 1;
3661 }
3662
3663 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3664 {
3665 return skb->queue_mapping - 1;
3666 }
3667
3668 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3669 {
3670 return skb->queue_mapping != 0;
3671 }
3672
3673 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3674 {
3675 #ifdef CONFIG_XFRM
3676 return skb->sp;
3677 #else
3678 return NULL;
3679 #endif
3680 }
3681
3682 /* Keeps track of mac header offset relative to skb->head.
3683 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3684 * For non-tunnel skb it points to skb_mac_header() and for
3685 * tunnel skb it points to outer mac header.
3686 * Keeps track of level of encapsulation of network headers.
3687 */
3688 struct skb_gso_cb {
3689 union {
3690 int mac_offset;
3691 int data_offset;
3692 };
3693 int encap_level;
3694 __wsum csum;
3695 __u16 csum_start;
3696 };
3697 #define SKB_SGO_CB_OFFSET 32
3698 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3699
3700 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3701 {
3702 return (skb_mac_header(inner_skb) - inner_skb->head) -
3703 SKB_GSO_CB(inner_skb)->mac_offset;
3704 }
3705
3706 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3707 {
3708 int new_headroom, headroom;
3709 int ret;
3710
3711 headroom = skb_headroom(skb);
3712 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3713 if (ret)
3714 return ret;
3715
3716 new_headroom = skb_headroom(skb);
3717 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3718 return 0;
3719 }
3720
3721 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3722 {
3723 /* Do not update partial checksums if remote checksum is enabled. */
3724 if (skb->remcsum_offload)
3725 return;
3726
3727 SKB_GSO_CB(skb)->csum = res;
3728 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3729 }
3730
3731 /* Compute the checksum for a gso segment. First compute the checksum value
3732 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3733 * then add in skb->csum (checksum from csum_start to end of packet).
3734 * skb->csum and csum_start are then updated to reflect the checksum of the
3735 * resultant packet starting from the transport header-- the resultant checksum
3736 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3737 * header.
3738 */
3739 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3740 {
3741 unsigned char *csum_start = skb_transport_header(skb);
3742 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3743 __wsum partial = SKB_GSO_CB(skb)->csum;
3744
3745 SKB_GSO_CB(skb)->csum = res;
3746 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3747
3748 return csum_fold(csum_partial(csum_start, plen, partial));
3749 }
3750
3751 static inline bool skb_is_gso(const struct sk_buff *skb)
3752 {
3753 return skb_shinfo(skb)->gso_size;
3754 }
3755
3756 /* Note: Should be called only if skb_is_gso(skb) is true */
3757 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3758 {
3759 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3760 }
3761
3762 static inline void skb_gso_reset(struct sk_buff *skb)
3763 {
3764 skb_shinfo(skb)->gso_size = 0;
3765 skb_shinfo(skb)->gso_segs = 0;
3766 skb_shinfo(skb)->gso_type = 0;
3767 }
3768
3769 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3770
3771 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3772 {
3773 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3774 * wanted then gso_type will be set. */
3775 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3776
3777 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3778 unlikely(shinfo->gso_type == 0)) {
3779 __skb_warn_lro_forwarding(skb);
3780 return true;
3781 }
3782 return false;
3783 }
3784
3785 static inline void skb_forward_csum(struct sk_buff *skb)
3786 {
3787 /* Unfortunately we don't support this one. Any brave souls? */
3788 if (skb->ip_summed == CHECKSUM_COMPLETE)
3789 skb->ip_summed = CHECKSUM_NONE;
3790 }
3791
3792 /**
3793 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3794 * @skb: skb to check
3795 *
3796 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3797 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3798 * use this helper, to document places where we make this assertion.
3799 */
3800 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3801 {
3802 #ifdef DEBUG
3803 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3804 #endif
3805 }
3806
3807 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3808
3809 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3810 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3811 unsigned int transport_len,
3812 __sum16(*skb_chkf)(struct sk_buff *skb));
3813
3814 /**
3815 * skb_head_is_locked - Determine if the skb->head is locked down
3816 * @skb: skb to check
3817 *
3818 * The head on skbs build around a head frag can be removed if they are
3819 * not cloned. This function returns true if the skb head is locked down
3820 * due to either being allocated via kmalloc, or by being a clone with
3821 * multiple references to the head.
3822 */
3823 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3824 {
3825 return !skb->head_frag || skb_cloned(skb);
3826 }
3827
3828 /**
3829 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3830 *
3831 * @skb: GSO skb
3832 *
3833 * skb_gso_network_seglen is used to determine the real size of the
3834 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3835 *
3836 * The MAC/L2 header is not accounted for.
3837 */
3838 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3839 {
3840 unsigned int hdr_len = skb_transport_header(skb) -
3841 skb_network_header(skb);
3842 return hdr_len + skb_gso_transport_seglen(skb);
3843 }
3844
3845 /* Local Checksum Offload.
3846 * Compute outer checksum based on the assumption that the
3847 * inner checksum will be offloaded later.
3848 * See Documentation/networking/checksum-offloads.txt for
3849 * explanation of how this works.
3850 * Fill in outer checksum adjustment (e.g. with sum of outer
3851 * pseudo-header) before calling.
3852 * Also ensure that inner checksum is in linear data area.
3853 */
3854 static inline __wsum lco_csum(struct sk_buff *skb)
3855 {
3856 unsigned char *csum_start = skb_checksum_start(skb);
3857 unsigned char *l4_hdr = skb_transport_header(skb);
3858 __wsum partial;
3859
3860 /* Start with complement of inner checksum adjustment */
3861 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3862 skb->csum_offset));
3863
3864 /* Add in checksum of our headers (incl. outer checksum
3865 * adjustment filled in by caller) and return result.
3866 */
3867 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3868 }
3869
3870 #endif /* __KERNEL__ */
3871 #endif /* _LINUX_SKBUFF_H */
This page took 0.110909 seconds and 5 git commands to generate.