Merge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
[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 * @offload_fwd_mark: fwding offload mark
616 * @mark: Generic packet mark
617 * @vlan_proto: vlan encapsulation protocol
618 * @vlan_tci: vlan tag control information
619 * @inner_protocol: Protocol (encapsulation)
620 * @inner_transport_header: Inner transport layer header (encapsulation)
621 * @inner_network_header: Network layer header (encapsulation)
622 * @inner_mac_header: Link layer header (encapsulation)
623 * @transport_header: Transport layer header
624 * @network_header: Network layer header
625 * @mac_header: Link layer header
626 * @tail: Tail pointer
627 * @end: End pointer
628 * @head: Head of buffer
629 * @data: Data head pointer
630 * @truesize: Buffer size
631 * @users: User count - see {datagram,tcp}.c
632 */
633
634 struct sk_buff {
635 union {
636 struct {
637 /* These two members must be first. */
638 struct sk_buff *next;
639 struct sk_buff *prev;
640
641 union {
642 ktime_t tstamp;
643 struct skb_mstamp skb_mstamp;
644 };
645 };
646 struct rb_node rbnode; /* used in netem & tcp stack */
647 };
648 struct sock *sk;
649 struct net_device *dev;
650
651 /*
652 * This is the control buffer. It is free to use for every
653 * layer. Please put your private variables there. If you
654 * want to keep them across layers you have to do a skb_clone()
655 * first. This is owned by whoever has the skb queued ATM.
656 */
657 char cb[48] __aligned(8);
658
659 unsigned long _skb_refdst;
660 void (*destructor)(struct sk_buff *skb);
661 #ifdef CONFIG_XFRM
662 struct sec_path *sp;
663 #endif
664 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
665 struct nf_conntrack *nfct;
666 #endif
667 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
668 struct nf_bridge_info *nf_bridge;
669 #endif
670 unsigned int len,
671 data_len;
672 __u16 mac_len,
673 hdr_len;
674
675 /* Following fields are _not_ copied in __copy_skb_header()
676 * Note that queue_mapping is here mostly to fill a hole.
677 */
678 kmemcheck_bitfield_begin(flags1);
679 __u16 queue_mapping;
680 __u8 cloned:1,
681 nohdr:1,
682 fclone:2,
683 peeked:1,
684 head_frag:1,
685 xmit_more:1;
686 /* one bit hole */
687 kmemcheck_bitfield_end(flags1);
688
689 /* fields enclosed in headers_start/headers_end are copied
690 * using a single memcpy() in __copy_skb_header()
691 */
692 /* private: */
693 __u32 headers_start[0];
694 /* public: */
695
696 /* if you move pkt_type around you also must adapt those constants */
697 #ifdef __BIG_ENDIAN_BITFIELD
698 #define PKT_TYPE_MAX (7 << 5)
699 #else
700 #define PKT_TYPE_MAX 7
701 #endif
702 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
703
704 __u8 __pkt_type_offset[0];
705 __u8 pkt_type:3;
706 __u8 pfmemalloc:1;
707 __u8 ignore_df:1;
708 __u8 nfctinfo:3;
709
710 __u8 nf_trace:1;
711 __u8 ip_summed:2;
712 __u8 ooo_okay:1;
713 __u8 l4_hash:1;
714 __u8 sw_hash:1;
715 __u8 wifi_acked_valid:1;
716 __u8 wifi_acked:1;
717
718 __u8 no_fcs:1;
719 /* Indicates the inner headers are valid in the skbuff. */
720 __u8 encapsulation:1;
721 __u8 encap_hdr_csum:1;
722 __u8 csum_valid:1;
723 __u8 csum_complete_sw:1;
724 __u8 csum_level:2;
725 __u8 csum_bad:1;
726
727 #ifdef CONFIG_IPV6_NDISC_NODETYPE
728 __u8 ndisc_nodetype:2;
729 #endif
730 __u8 ipvs_property:1;
731 __u8 inner_protocol_type:1;
732 __u8 remcsum_offload:1;
733 /* 3 or 5 bit hole */
734
735 #ifdef CONFIG_NET_SCHED
736 __u16 tc_index; /* traffic control index */
737 #ifdef CONFIG_NET_CLS_ACT
738 __u16 tc_verd; /* traffic control verdict */
739 #endif
740 #endif
741
742 union {
743 __wsum csum;
744 struct {
745 __u16 csum_start;
746 __u16 csum_offset;
747 };
748 };
749 __u32 priority;
750 int skb_iif;
751 __u32 hash;
752 __be16 vlan_proto;
753 __u16 vlan_tci;
754 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
755 union {
756 unsigned int napi_id;
757 unsigned int sender_cpu;
758 };
759 #endif
760 union {
761 #ifdef CONFIG_NETWORK_SECMARK
762 __u32 secmark;
763 #endif
764 #ifdef CONFIG_NET_SWITCHDEV
765 __u32 offload_fwd_mark;
766 #endif
767 };
768
769 union {
770 __u32 mark;
771 __u32 reserved_tailroom;
772 };
773
774 union {
775 __be16 inner_protocol;
776 __u8 inner_ipproto;
777 };
778
779 __u16 inner_transport_header;
780 __u16 inner_network_header;
781 __u16 inner_mac_header;
782
783 __be16 protocol;
784 __u16 transport_header;
785 __u16 network_header;
786 __u16 mac_header;
787
788 /* private: */
789 __u32 headers_end[0];
790 /* public: */
791
792 /* These elements must be at the end, see alloc_skb() for details. */
793 sk_buff_data_t tail;
794 sk_buff_data_t end;
795 unsigned char *head,
796 *data;
797 unsigned int truesize;
798 atomic_t users;
799 };
800
801 #ifdef __KERNEL__
802 /*
803 * Handling routines are only of interest to the kernel
804 */
805 #include <linux/slab.h>
806
807
808 #define SKB_ALLOC_FCLONE 0x01
809 #define SKB_ALLOC_RX 0x02
810 #define SKB_ALLOC_NAPI 0x04
811
812 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
813 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
814 {
815 return unlikely(skb->pfmemalloc);
816 }
817
818 /*
819 * skb might have a dst pointer attached, refcounted or not.
820 * _skb_refdst low order bit is set if refcount was _not_ taken
821 */
822 #define SKB_DST_NOREF 1UL
823 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
824
825 /**
826 * skb_dst - returns skb dst_entry
827 * @skb: buffer
828 *
829 * Returns skb dst_entry, regardless of reference taken or not.
830 */
831 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
832 {
833 /* If refdst was not refcounted, check we still are in a
834 * rcu_read_lock section
835 */
836 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
837 !rcu_read_lock_held() &&
838 !rcu_read_lock_bh_held());
839 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
840 }
841
842 /**
843 * skb_dst_set - sets skb dst
844 * @skb: buffer
845 * @dst: dst entry
846 *
847 * Sets skb dst, assuming a reference was taken on dst and should
848 * be released by skb_dst_drop()
849 */
850 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
851 {
852 skb->_skb_refdst = (unsigned long)dst;
853 }
854
855 /**
856 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
857 * @skb: buffer
858 * @dst: dst entry
859 *
860 * Sets skb dst, assuming a reference was not taken on dst.
861 * If dst entry is cached, we do not take reference and dst_release
862 * will be avoided by refdst_drop. If dst entry is not cached, we take
863 * reference, so that last dst_release can destroy the dst immediately.
864 */
865 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
866 {
867 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
868 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
869 }
870
871 /**
872 * skb_dst_is_noref - Test if skb dst isn't refcounted
873 * @skb: buffer
874 */
875 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
876 {
877 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
878 }
879
880 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
881 {
882 return (struct rtable *)skb_dst(skb);
883 }
884
885 /* For mangling skb->pkt_type from user space side from applications
886 * such as nft, tc, etc, we only allow a conservative subset of
887 * possible pkt_types to be set.
888 */
889 static inline bool skb_pkt_type_ok(u32 ptype)
890 {
891 return ptype <= PACKET_OTHERHOST;
892 }
893
894 void kfree_skb(struct sk_buff *skb);
895 void kfree_skb_list(struct sk_buff *segs);
896 void skb_tx_error(struct sk_buff *skb);
897 void consume_skb(struct sk_buff *skb);
898 void __kfree_skb(struct sk_buff *skb);
899 extern struct kmem_cache *skbuff_head_cache;
900
901 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
902 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
903 bool *fragstolen, int *delta_truesize);
904
905 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
906 int node);
907 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
908 struct sk_buff *build_skb(void *data, unsigned int frag_size);
909 static inline struct sk_buff *alloc_skb(unsigned int size,
910 gfp_t priority)
911 {
912 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
913 }
914
915 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
916 unsigned long data_len,
917 int max_page_order,
918 int *errcode,
919 gfp_t gfp_mask);
920
921 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
922 struct sk_buff_fclones {
923 struct sk_buff skb1;
924
925 struct sk_buff skb2;
926
927 atomic_t fclone_ref;
928 };
929
930 /**
931 * skb_fclone_busy - check if fclone is busy
932 * @skb: buffer
933 *
934 * Returns true if skb is a fast clone, and its clone is not freed.
935 * Some drivers call skb_orphan() in their ndo_start_xmit(),
936 * so we also check that this didnt happen.
937 */
938 static inline bool skb_fclone_busy(const struct sock *sk,
939 const struct sk_buff *skb)
940 {
941 const struct sk_buff_fclones *fclones;
942
943 fclones = container_of(skb, struct sk_buff_fclones, skb1);
944
945 return skb->fclone == SKB_FCLONE_ORIG &&
946 atomic_read(&fclones->fclone_ref) > 1 &&
947 fclones->skb2.sk == sk;
948 }
949
950 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
951 gfp_t priority)
952 {
953 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
954 }
955
956 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
957 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
958 {
959 return __alloc_skb_head(priority, -1);
960 }
961
962 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
963 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
964 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
965 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
966 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
967 gfp_t gfp_mask, bool fclone);
968 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
969 gfp_t gfp_mask)
970 {
971 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
972 }
973
974 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
975 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
976 unsigned int headroom);
977 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
978 int newtailroom, gfp_t priority);
979 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
980 int offset, int len);
981 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
982 int len);
983 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
984 int skb_pad(struct sk_buff *skb, int pad);
985 #define dev_kfree_skb(a) consume_skb(a)
986
987 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
988 int getfrag(void *from, char *to, int offset,
989 int len, int odd, struct sk_buff *skb),
990 void *from, int length);
991
992 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
993 int offset, size_t size);
994
995 struct skb_seq_state {
996 __u32 lower_offset;
997 __u32 upper_offset;
998 __u32 frag_idx;
999 __u32 stepped_offset;
1000 struct sk_buff *root_skb;
1001 struct sk_buff *cur_skb;
1002 __u8 *frag_data;
1003 };
1004
1005 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1006 unsigned int to, struct skb_seq_state *st);
1007 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1008 struct skb_seq_state *st);
1009 void skb_abort_seq_read(struct skb_seq_state *st);
1010
1011 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1012 unsigned int to, struct ts_config *config);
1013
1014 /*
1015 * Packet hash types specify the type of hash in skb_set_hash.
1016 *
1017 * Hash types refer to the protocol layer addresses which are used to
1018 * construct a packet's hash. The hashes are used to differentiate or identify
1019 * flows of the protocol layer for the hash type. Hash types are either
1020 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1021 *
1022 * Properties of hashes:
1023 *
1024 * 1) Two packets in different flows have different hash values
1025 * 2) Two packets in the same flow should have the same hash value
1026 *
1027 * A hash at a higher layer is considered to be more specific. A driver should
1028 * set the most specific hash possible.
1029 *
1030 * A driver cannot indicate a more specific hash than the layer at which a hash
1031 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1032 *
1033 * A driver may indicate a hash level which is less specific than the
1034 * actual layer the hash was computed on. For instance, a hash computed
1035 * at L4 may be considered an L3 hash. This should only be done if the
1036 * driver can't unambiguously determine that the HW computed the hash at
1037 * the higher layer. Note that the "should" in the second property above
1038 * permits this.
1039 */
1040 enum pkt_hash_types {
1041 PKT_HASH_TYPE_NONE, /* Undefined type */
1042 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1043 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1044 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1045 };
1046
1047 static inline void skb_clear_hash(struct sk_buff *skb)
1048 {
1049 skb->hash = 0;
1050 skb->sw_hash = 0;
1051 skb->l4_hash = 0;
1052 }
1053
1054 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1055 {
1056 if (!skb->l4_hash)
1057 skb_clear_hash(skb);
1058 }
1059
1060 static inline void
1061 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1062 {
1063 skb->l4_hash = is_l4;
1064 skb->sw_hash = is_sw;
1065 skb->hash = hash;
1066 }
1067
1068 static inline void
1069 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1070 {
1071 /* Used by drivers to set hash from HW */
1072 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1073 }
1074
1075 static inline void
1076 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1077 {
1078 __skb_set_hash(skb, hash, true, is_l4);
1079 }
1080
1081 void __skb_get_hash(struct sk_buff *skb);
1082 u32 __skb_get_hash_symmetric(struct sk_buff *skb);
1083 u32 skb_get_poff(const struct sk_buff *skb);
1084 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1085 const struct flow_keys *keys, int hlen);
1086 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1087 void *data, int hlen_proto);
1088
1089 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1090 int thoff, u8 ip_proto)
1091 {
1092 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1093 }
1094
1095 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1096 const struct flow_dissector_key *key,
1097 unsigned int key_count);
1098
1099 bool __skb_flow_dissect(const struct sk_buff *skb,
1100 struct flow_dissector *flow_dissector,
1101 void *target_container,
1102 void *data, __be16 proto, int nhoff, int hlen,
1103 unsigned int flags);
1104
1105 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1106 struct flow_dissector *flow_dissector,
1107 void *target_container, unsigned int flags)
1108 {
1109 return __skb_flow_dissect(skb, flow_dissector, target_container,
1110 NULL, 0, 0, 0, flags);
1111 }
1112
1113 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1114 struct flow_keys *flow,
1115 unsigned int flags)
1116 {
1117 memset(flow, 0, sizeof(*flow));
1118 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1119 NULL, 0, 0, 0, flags);
1120 }
1121
1122 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1123 void *data, __be16 proto,
1124 int nhoff, int hlen,
1125 unsigned int flags)
1126 {
1127 memset(flow, 0, sizeof(*flow));
1128 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1129 data, proto, nhoff, hlen, flags);
1130 }
1131
1132 static inline __u32 skb_get_hash(struct sk_buff *skb)
1133 {
1134 if (!skb->l4_hash && !skb->sw_hash)
1135 __skb_get_hash(skb);
1136
1137 return skb->hash;
1138 }
1139
1140 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1141
1142 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1143 {
1144 if (!skb->l4_hash && !skb->sw_hash) {
1145 struct flow_keys keys;
1146 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1147
1148 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1149 }
1150
1151 return skb->hash;
1152 }
1153
1154 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1155
1156 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1157 {
1158 if (!skb->l4_hash && !skb->sw_hash) {
1159 struct flow_keys keys;
1160 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1161
1162 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1163 }
1164
1165 return skb->hash;
1166 }
1167
1168 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1169
1170 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1171 {
1172 return skb->hash;
1173 }
1174
1175 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1176 {
1177 to->hash = from->hash;
1178 to->sw_hash = from->sw_hash;
1179 to->l4_hash = from->l4_hash;
1180 };
1181
1182 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1183 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1184 {
1185 return skb->head + skb->end;
1186 }
1187
1188 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1189 {
1190 return skb->end;
1191 }
1192 #else
1193 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1194 {
1195 return skb->end;
1196 }
1197
1198 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1199 {
1200 return skb->end - skb->head;
1201 }
1202 #endif
1203
1204 /* Internal */
1205 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1206
1207 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1208 {
1209 return &skb_shinfo(skb)->hwtstamps;
1210 }
1211
1212 /**
1213 * skb_queue_empty - check if a queue is empty
1214 * @list: queue head
1215 *
1216 * Returns true if the queue is empty, false otherwise.
1217 */
1218 static inline int skb_queue_empty(const struct sk_buff_head *list)
1219 {
1220 return list->next == (const struct sk_buff *) list;
1221 }
1222
1223 /**
1224 * skb_queue_is_last - check if skb is the last entry in the queue
1225 * @list: queue head
1226 * @skb: buffer
1227 *
1228 * Returns true if @skb is the last buffer on the list.
1229 */
1230 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1231 const struct sk_buff *skb)
1232 {
1233 return skb->next == (const struct sk_buff *) list;
1234 }
1235
1236 /**
1237 * skb_queue_is_first - check if skb is the first entry in the queue
1238 * @list: queue head
1239 * @skb: buffer
1240 *
1241 * Returns true if @skb is the first buffer on the list.
1242 */
1243 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1244 const struct sk_buff *skb)
1245 {
1246 return skb->prev == (const struct sk_buff *) list;
1247 }
1248
1249 /**
1250 * skb_queue_next - return the next packet in the queue
1251 * @list: queue head
1252 * @skb: current buffer
1253 *
1254 * Return the next packet in @list after @skb. It is only valid to
1255 * call this if skb_queue_is_last() evaluates to false.
1256 */
1257 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1258 const struct sk_buff *skb)
1259 {
1260 /* This BUG_ON may seem severe, but if we just return then we
1261 * are going to dereference garbage.
1262 */
1263 BUG_ON(skb_queue_is_last(list, skb));
1264 return skb->next;
1265 }
1266
1267 /**
1268 * skb_queue_prev - return the prev packet in the queue
1269 * @list: queue head
1270 * @skb: current buffer
1271 *
1272 * Return the prev packet in @list before @skb. It is only valid to
1273 * call this if skb_queue_is_first() evaluates to false.
1274 */
1275 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1276 const struct sk_buff *skb)
1277 {
1278 /* This BUG_ON may seem severe, but if we just return then we
1279 * are going to dereference garbage.
1280 */
1281 BUG_ON(skb_queue_is_first(list, skb));
1282 return skb->prev;
1283 }
1284
1285 /**
1286 * skb_get - reference buffer
1287 * @skb: buffer to reference
1288 *
1289 * Makes another reference to a socket buffer and returns a pointer
1290 * to the buffer.
1291 */
1292 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1293 {
1294 atomic_inc(&skb->users);
1295 return skb;
1296 }
1297
1298 /*
1299 * If users == 1, we are the only owner and are can avoid redundant
1300 * atomic change.
1301 */
1302
1303 /**
1304 * skb_cloned - is the buffer a clone
1305 * @skb: buffer to check
1306 *
1307 * Returns true if the buffer was generated with skb_clone() and is
1308 * one of multiple shared copies of the buffer. Cloned buffers are
1309 * shared data so must not be written to under normal circumstances.
1310 */
1311 static inline int skb_cloned(const struct sk_buff *skb)
1312 {
1313 return skb->cloned &&
1314 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1315 }
1316
1317 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1318 {
1319 might_sleep_if(gfpflags_allow_blocking(pri));
1320
1321 if (skb_cloned(skb))
1322 return pskb_expand_head(skb, 0, 0, pri);
1323
1324 return 0;
1325 }
1326
1327 /**
1328 * skb_header_cloned - is the header a clone
1329 * @skb: buffer to check
1330 *
1331 * Returns true if modifying the header part of the buffer requires
1332 * the data to be copied.
1333 */
1334 static inline int skb_header_cloned(const struct sk_buff *skb)
1335 {
1336 int dataref;
1337
1338 if (!skb->cloned)
1339 return 0;
1340
1341 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1342 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1343 return dataref != 1;
1344 }
1345
1346 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1347 {
1348 might_sleep_if(gfpflags_allow_blocking(pri));
1349
1350 if (skb_header_cloned(skb))
1351 return pskb_expand_head(skb, 0, 0, pri);
1352
1353 return 0;
1354 }
1355
1356 /**
1357 * skb_header_release - release reference to header
1358 * @skb: buffer to operate on
1359 *
1360 * Drop a reference to the header part of the buffer. This is done
1361 * by acquiring a payload reference. You must not read from the header
1362 * part of skb->data after this.
1363 * Note : Check if you can use __skb_header_release() instead.
1364 */
1365 static inline void skb_header_release(struct sk_buff *skb)
1366 {
1367 BUG_ON(skb->nohdr);
1368 skb->nohdr = 1;
1369 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1370 }
1371
1372 /**
1373 * __skb_header_release - release reference to header
1374 * @skb: buffer to operate on
1375 *
1376 * Variant of skb_header_release() assuming skb is private to caller.
1377 * We can avoid one atomic operation.
1378 */
1379 static inline void __skb_header_release(struct sk_buff *skb)
1380 {
1381 skb->nohdr = 1;
1382 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1383 }
1384
1385
1386 /**
1387 * skb_shared - is the buffer shared
1388 * @skb: buffer to check
1389 *
1390 * Returns true if more than one person has a reference to this
1391 * buffer.
1392 */
1393 static inline int skb_shared(const struct sk_buff *skb)
1394 {
1395 return atomic_read(&skb->users) != 1;
1396 }
1397
1398 /**
1399 * skb_share_check - check if buffer is shared and if so clone it
1400 * @skb: buffer to check
1401 * @pri: priority for memory allocation
1402 *
1403 * If the buffer is shared the buffer is cloned and the old copy
1404 * drops a reference. A new clone with a single reference is returned.
1405 * If the buffer is not shared the original buffer is returned. When
1406 * being called from interrupt status or with spinlocks held pri must
1407 * be GFP_ATOMIC.
1408 *
1409 * NULL is returned on a memory allocation failure.
1410 */
1411 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1412 {
1413 might_sleep_if(gfpflags_allow_blocking(pri));
1414 if (skb_shared(skb)) {
1415 struct sk_buff *nskb = skb_clone(skb, pri);
1416
1417 if (likely(nskb))
1418 consume_skb(skb);
1419 else
1420 kfree_skb(skb);
1421 skb = nskb;
1422 }
1423 return skb;
1424 }
1425
1426 /*
1427 * Copy shared buffers into a new sk_buff. We effectively do COW on
1428 * packets to handle cases where we have a local reader and forward
1429 * and a couple of other messy ones. The normal one is tcpdumping
1430 * a packet thats being forwarded.
1431 */
1432
1433 /**
1434 * skb_unshare - make a copy of a shared buffer
1435 * @skb: buffer to check
1436 * @pri: priority for memory allocation
1437 *
1438 * If the socket buffer is a clone then this function creates a new
1439 * copy of the data, drops a reference count on the old copy and returns
1440 * the new copy with the reference count at 1. If the buffer is not a clone
1441 * the original buffer is returned. When called with a spinlock held or
1442 * from interrupt state @pri must be %GFP_ATOMIC
1443 *
1444 * %NULL is returned on a memory allocation failure.
1445 */
1446 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1447 gfp_t pri)
1448 {
1449 might_sleep_if(gfpflags_allow_blocking(pri));
1450 if (skb_cloned(skb)) {
1451 struct sk_buff *nskb = skb_copy(skb, pri);
1452
1453 /* Free our shared copy */
1454 if (likely(nskb))
1455 consume_skb(skb);
1456 else
1457 kfree_skb(skb);
1458 skb = nskb;
1459 }
1460 return skb;
1461 }
1462
1463 /**
1464 * skb_peek - peek at the head of an &sk_buff_head
1465 * @list_: list to peek at
1466 *
1467 * Peek an &sk_buff. Unlike most other operations you _MUST_
1468 * be careful with this one. A peek leaves the buffer on the
1469 * list and someone else may run off with it. You must hold
1470 * the appropriate locks or have a private queue to do this.
1471 *
1472 * Returns %NULL for an empty list or a pointer to the head element.
1473 * The reference count is not incremented and the reference is therefore
1474 * volatile. Use with caution.
1475 */
1476 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1477 {
1478 struct sk_buff *skb = list_->next;
1479
1480 if (skb == (struct sk_buff *)list_)
1481 skb = NULL;
1482 return skb;
1483 }
1484
1485 /**
1486 * skb_peek_next - peek skb following the given one from a queue
1487 * @skb: skb to start from
1488 * @list_: list to peek at
1489 *
1490 * Returns %NULL when the end of the list is met or a pointer to the
1491 * next element. The reference count is not incremented and the
1492 * reference is therefore volatile. Use with caution.
1493 */
1494 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1495 const struct sk_buff_head *list_)
1496 {
1497 struct sk_buff *next = skb->next;
1498
1499 if (next == (struct sk_buff *)list_)
1500 next = NULL;
1501 return next;
1502 }
1503
1504 /**
1505 * skb_peek_tail - peek at the tail of an &sk_buff_head
1506 * @list_: list to peek at
1507 *
1508 * Peek an &sk_buff. Unlike most other operations you _MUST_
1509 * be careful with this one. A peek leaves the buffer on the
1510 * list and someone else may run off with it. You must hold
1511 * the appropriate locks or have a private queue to do this.
1512 *
1513 * Returns %NULL for an empty list or a pointer to the tail element.
1514 * The reference count is not incremented and the reference is therefore
1515 * volatile. Use with caution.
1516 */
1517 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1518 {
1519 struct sk_buff *skb = list_->prev;
1520
1521 if (skb == (struct sk_buff *)list_)
1522 skb = NULL;
1523 return skb;
1524
1525 }
1526
1527 /**
1528 * skb_queue_len - get queue length
1529 * @list_: list to measure
1530 *
1531 * Return the length of an &sk_buff queue.
1532 */
1533 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1534 {
1535 return list_->qlen;
1536 }
1537
1538 /**
1539 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1540 * @list: queue to initialize
1541 *
1542 * This initializes only the list and queue length aspects of
1543 * an sk_buff_head object. This allows to initialize the list
1544 * aspects of an sk_buff_head without reinitializing things like
1545 * the spinlock. It can also be used for on-stack sk_buff_head
1546 * objects where the spinlock is known to not be used.
1547 */
1548 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1549 {
1550 list->prev = list->next = (struct sk_buff *)list;
1551 list->qlen = 0;
1552 }
1553
1554 /*
1555 * This function creates a split out lock class for each invocation;
1556 * this is needed for now since a whole lot of users of the skb-queue
1557 * infrastructure in drivers have different locking usage (in hardirq)
1558 * than the networking core (in softirq only). In the long run either the
1559 * network layer or drivers should need annotation to consolidate the
1560 * main types of usage into 3 classes.
1561 */
1562 static inline void skb_queue_head_init(struct sk_buff_head *list)
1563 {
1564 spin_lock_init(&list->lock);
1565 __skb_queue_head_init(list);
1566 }
1567
1568 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1569 struct lock_class_key *class)
1570 {
1571 skb_queue_head_init(list);
1572 lockdep_set_class(&list->lock, class);
1573 }
1574
1575 /*
1576 * Insert an sk_buff on a list.
1577 *
1578 * The "__skb_xxxx()" functions are the non-atomic ones that
1579 * can only be called with interrupts disabled.
1580 */
1581 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1582 struct sk_buff_head *list);
1583 static inline void __skb_insert(struct sk_buff *newsk,
1584 struct sk_buff *prev, struct sk_buff *next,
1585 struct sk_buff_head *list)
1586 {
1587 newsk->next = next;
1588 newsk->prev = prev;
1589 next->prev = prev->next = newsk;
1590 list->qlen++;
1591 }
1592
1593 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1594 struct sk_buff *prev,
1595 struct sk_buff *next)
1596 {
1597 struct sk_buff *first = list->next;
1598 struct sk_buff *last = list->prev;
1599
1600 first->prev = prev;
1601 prev->next = first;
1602
1603 last->next = next;
1604 next->prev = last;
1605 }
1606
1607 /**
1608 * skb_queue_splice - join two skb lists, this is designed for stacks
1609 * @list: the new list to add
1610 * @head: the place to add it in the first list
1611 */
1612 static inline void skb_queue_splice(const struct sk_buff_head *list,
1613 struct sk_buff_head *head)
1614 {
1615 if (!skb_queue_empty(list)) {
1616 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1617 head->qlen += list->qlen;
1618 }
1619 }
1620
1621 /**
1622 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1623 * @list: the new list to add
1624 * @head: the place to add it in the first list
1625 *
1626 * The list at @list is reinitialised
1627 */
1628 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1629 struct sk_buff_head *head)
1630 {
1631 if (!skb_queue_empty(list)) {
1632 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1633 head->qlen += list->qlen;
1634 __skb_queue_head_init(list);
1635 }
1636 }
1637
1638 /**
1639 * skb_queue_splice_tail - join two skb lists, each list being a queue
1640 * @list: the new list to add
1641 * @head: the place to add it in the first list
1642 */
1643 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1644 struct sk_buff_head *head)
1645 {
1646 if (!skb_queue_empty(list)) {
1647 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1648 head->qlen += list->qlen;
1649 }
1650 }
1651
1652 /**
1653 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1654 * @list: the new list to add
1655 * @head: the place to add it in the first list
1656 *
1657 * Each of the lists is a queue.
1658 * The list at @list is reinitialised
1659 */
1660 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1661 struct sk_buff_head *head)
1662 {
1663 if (!skb_queue_empty(list)) {
1664 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1665 head->qlen += list->qlen;
1666 __skb_queue_head_init(list);
1667 }
1668 }
1669
1670 /**
1671 * __skb_queue_after - queue a buffer at the list head
1672 * @list: list to use
1673 * @prev: place after this buffer
1674 * @newsk: buffer to queue
1675 *
1676 * Queue a buffer int the middle of a list. This function takes no locks
1677 * and you must therefore hold required locks before calling it.
1678 *
1679 * A buffer cannot be placed on two lists at the same time.
1680 */
1681 static inline void __skb_queue_after(struct sk_buff_head *list,
1682 struct sk_buff *prev,
1683 struct sk_buff *newsk)
1684 {
1685 __skb_insert(newsk, prev, prev->next, list);
1686 }
1687
1688 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1689 struct sk_buff_head *list);
1690
1691 static inline void __skb_queue_before(struct sk_buff_head *list,
1692 struct sk_buff *next,
1693 struct sk_buff *newsk)
1694 {
1695 __skb_insert(newsk, next->prev, next, list);
1696 }
1697
1698 /**
1699 * __skb_queue_head - queue a buffer at the list head
1700 * @list: list to use
1701 * @newsk: buffer to queue
1702 *
1703 * Queue a buffer at the start of a list. This function takes no locks
1704 * and you must therefore hold required locks before calling it.
1705 *
1706 * A buffer cannot be placed on two lists at the same time.
1707 */
1708 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1709 static inline void __skb_queue_head(struct sk_buff_head *list,
1710 struct sk_buff *newsk)
1711 {
1712 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1713 }
1714
1715 /**
1716 * __skb_queue_tail - queue a buffer at the list tail
1717 * @list: list to use
1718 * @newsk: buffer to queue
1719 *
1720 * Queue a buffer at the end of a list. This function takes no locks
1721 * and you must therefore hold required locks before calling it.
1722 *
1723 * A buffer cannot be placed on two lists at the same time.
1724 */
1725 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1726 static inline void __skb_queue_tail(struct sk_buff_head *list,
1727 struct sk_buff *newsk)
1728 {
1729 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1730 }
1731
1732 /*
1733 * remove sk_buff from list. _Must_ be called atomically, and with
1734 * the list known..
1735 */
1736 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1737 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1738 {
1739 struct sk_buff *next, *prev;
1740
1741 list->qlen--;
1742 next = skb->next;
1743 prev = skb->prev;
1744 skb->next = skb->prev = NULL;
1745 next->prev = prev;
1746 prev->next = next;
1747 }
1748
1749 /**
1750 * __skb_dequeue - remove from the head of the queue
1751 * @list: list to dequeue from
1752 *
1753 * Remove the head of the list. This function does not take any locks
1754 * so must be used with appropriate locks held only. The head item is
1755 * returned or %NULL if the list is empty.
1756 */
1757 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1758 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1759 {
1760 struct sk_buff *skb = skb_peek(list);
1761 if (skb)
1762 __skb_unlink(skb, list);
1763 return skb;
1764 }
1765
1766 /**
1767 * __skb_dequeue_tail - remove from the tail of the queue
1768 * @list: list to dequeue from
1769 *
1770 * Remove the tail of the list. This function does not take any locks
1771 * so must be used with appropriate locks held only. The tail item is
1772 * returned or %NULL if the list is empty.
1773 */
1774 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1775 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1776 {
1777 struct sk_buff *skb = skb_peek_tail(list);
1778 if (skb)
1779 __skb_unlink(skb, list);
1780 return skb;
1781 }
1782
1783
1784 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1785 {
1786 return skb->data_len;
1787 }
1788
1789 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1790 {
1791 return skb->len - skb->data_len;
1792 }
1793
1794 static inline int skb_pagelen(const struct sk_buff *skb)
1795 {
1796 int i, len = 0;
1797
1798 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1799 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1800 return len + skb_headlen(skb);
1801 }
1802
1803 /**
1804 * __skb_fill_page_desc - initialise a paged fragment in an skb
1805 * @skb: buffer containing fragment to be initialised
1806 * @i: paged fragment index to initialise
1807 * @page: the page to use for this fragment
1808 * @off: the offset to the data with @page
1809 * @size: the length of the data
1810 *
1811 * Initialises the @i'th fragment of @skb to point to &size bytes at
1812 * offset @off within @page.
1813 *
1814 * Does not take any additional reference on the fragment.
1815 */
1816 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1817 struct page *page, int off, int size)
1818 {
1819 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1820
1821 /*
1822 * Propagate page pfmemalloc to the skb if we can. The problem is
1823 * that not all callers have unique ownership of the page but rely
1824 * on page_is_pfmemalloc doing the right thing(tm).
1825 */
1826 frag->page.p = page;
1827 frag->page_offset = off;
1828 skb_frag_size_set(frag, size);
1829
1830 page = compound_head(page);
1831 if (page_is_pfmemalloc(page))
1832 skb->pfmemalloc = true;
1833 }
1834
1835 /**
1836 * skb_fill_page_desc - initialise a paged fragment in an skb
1837 * @skb: buffer containing fragment to be initialised
1838 * @i: paged fragment index to initialise
1839 * @page: the page to use for this fragment
1840 * @off: the offset to the data with @page
1841 * @size: the length of the data
1842 *
1843 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1844 * @skb to point to @size bytes at offset @off within @page. In
1845 * addition updates @skb such that @i is the last fragment.
1846 *
1847 * Does not take any additional reference on the fragment.
1848 */
1849 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1850 struct page *page, int off, int size)
1851 {
1852 __skb_fill_page_desc(skb, i, page, off, size);
1853 skb_shinfo(skb)->nr_frags = i + 1;
1854 }
1855
1856 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1857 int size, unsigned int truesize);
1858
1859 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1860 unsigned int truesize);
1861
1862 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1863 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1864 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1865
1866 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1867 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1868 {
1869 return skb->head + skb->tail;
1870 }
1871
1872 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1873 {
1874 skb->tail = skb->data - skb->head;
1875 }
1876
1877 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1878 {
1879 skb_reset_tail_pointer(skb);
1880 skb->tail += offset;
1881 }
1882
1883 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1884 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1885 {
1886 return skb->tail;
1887 }
1888
1889 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1890 {
1891 skb->tail = skb->data;
1892 }
1893
1894 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1895 {
1896 skb->tail = skb->data + offset;
1897 }
1898
1899 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1900
1901 /*
1902 * Add data to an sk_buff
1903 */
1904 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1905 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1906 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1907 {
1908 unsigned char *tmp = skb_tail_pointer(skb);
1909 SKB_LINEAR_ASSERT(skb);
1910 skb->tail += len;
1911 skb->len += len;
1912 return tmp;
1913 }
1914
1915 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1916 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1917 {
1918 skb->data -= len;
1919 skb->len += len;
1920 return skb->data;
1921 }
1922
1923 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1924 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1925 {
1926 skb->len -= len;
1927 BUG_ON(skb->len < skb->data_len);
1928 return skb->data += len;
1929 }
1930
1931 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1932 {
1933 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1934 }
1935
1936 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1937
1938 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1939 {
1940 if (len > skb_headlen(skb) &&
1941 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1942 return NULL;
1943 skb->len -= len;
1944 return skb->data += len;
1945 }
1946
1947 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1948 {
1949 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1950 }
1951
1952 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1953 {
1954 if (likely(len <= skb_headlen(skb)))
1955 return 1;
1956 if (unlikely(len > skb->len))
1957 return 0;
1958 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1959 }
1960
1961 /**
1962 * skb_headroom - bytes at buffer head
1963 * @skb: buffer to check
1964 *
1965 * Return the number of bytes of free space at the head of an &sk_buff.
1966 */
1967 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1968 {
1969 return skb->data - skb->head;
1970 }
1971
1972 /**
1973 * skb_tailroom - bytes at buffer end
1974 * @skb: buffer to check
1975 *
1976 * Return the number of bytes of free space at the tail of an sk_buff
1977 */
1978 static inline int skb_tailroom(const struct sk_buff *skb)
1979 {
1980 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1981 }
1982
1983 /**
1984 * skb_availroom - bytes at buffer end
1985 * @skb: buffer to check
1986 *
1987 * Return the number of bytes of free space at the tail of an sk_buff
1988 * allocated by sk_stream_alloc()
1989 */
1990 static inline int skb_availroom(const struct sk_buff *skb)
1991 {
1992 if (skb_is_nonlinear(skb))
1993 return 0;
1994
1995 return skb->end - skb->tail - skb->reserved_tailroom;
1996 }
1997
1998 /**
1999 * skb_reserve - adjust headroom
2000 * @skb: buffer to alter
2001 * @len: bytes to move
2002 *
2003 * Increase the headroom of an empty &sk_buff by reducing the tail
2004 * room. This is only allowed for an empty buffer.
2005 */
2006 static inline void skb_reserve(struct sk_buff *skb, int len)
2007 {
2008 skb->data += len;
2009 skb->tail += len;
2010 }
2011
2012 /**
2013 * skb_tailroom_reserve - adjust reserved_tailroom
2014 * @skb: buffer to alter
2015 * @mtu: maximum amount of headlen permitted
2016 * @needed_tailroom: minimum amount of reserved_tailroom
2017 *
2018 * Set reserved_tailroom so that headlen can be as large as possible but
2019 * not larger than mtu and tailroom cannot be smaller than
2020 * needed_tailroom.
2021 * The required headroom should already have been reserved before using
2022 * this function.
2023 */
2024 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2025 unsigned int needed_tailroom)
2026 {
2027 SKB_LINEAR_ASSERT(skb);
2028 if (mtu < skb_tailroom(skb) - needed_tailroom)
2029 /* use at most mtu */
2030 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2031 else
2032 /* use up to all available space */
2033 skb->reserved_tailroom = needed_tailroom;
2034 }
2035
2036 #define ENCAP_TYPE_ETHER 0
2037 #define ENCAP_TYPE_IPPROTO 1
2038
2039 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2040 __be16 protocol)
2041 {
2042 skb->inner_protocol = protocol;
2043 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2044 }
2045
2046 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2047 __u8 ipproto)
2048 {
2049 skb->inner_ipproto = ipproto;
2050 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2051 }
2052
2053 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2054 {
2055 skb->inner_mac_header = skb->mac_header;
2056 skb->inner_network_header = skb->network_header;
2057 skb->inner_transport_header = skb->transport_header;
2058 }
2059
2060 static inline void skb_reset_mac_len(struct sk_buff *skb)
2061 {
2062 skb->mac_len = skb->network_header - skb->mac_header;
2063 }
2064
2065 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2066 *skb)
2067 {
2068 return skb->head + skb->inner_transport_header;
2069 }
2070
2071 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2072 {
2073 return skb_inner_transport_header(skb) - skb->data;
2074 }
2075
2076 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2077 {
2078 skb->inner_transport_header = skb->data - skb->head;
2079 }
2080
2081 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2082 const int offset)
2083 {
2084 skb_reset_inner_transport_header(skb);
2085 skb->inner_transport_header += offset;
2086 }
2087
2088 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2089 {
2090 return skb->head + skb->inner_network_header;
2091 }
2092
2093 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2094 {
2095 skb->inner_network_header = skb->data - skb->head;
2096 }
2097
2098 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2099 const int offset)
2100 {
2101 skb_reset_inner_network_header(skb);
2102 skb->inner_network_header += offset;
2103 }
2104
2105 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2106 {
2107 return skb->head + skb->inner_mac_header;
2108 }
2109
2110 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2111 {
2112 skb->inner_mac_header = skb->data - skb->head;
2113 }
2114
2115 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2116 const int offset)
2117 {
2118 skb_reset_inner_mac_header(skb);
2119 skb->inner_mac_header += offset;
2120 }
2121 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2122 {
2123 return skb->transport_header != (typeof(skb->transport_header))~0U;
2124 }
2125
2126 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2127 {
2128 return skb->head + skb->transport_header;
2129 }
2130
2131 static inline void skb_reset_transport_header(struct sk_buff *skb)
2132 {
2133 skb->transport_header = skb->data - skb->head;
2134 }
2135
2136 static inline void skb_set_transport_header(struct sk_buff *skb,
2137 const int offset)
2138 {
2139 skb_reset_transport_header(skb);
2140 skb->transport_header += offset;
2141 }
2142
2143 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2144 {
2145 return skb->head + skb->network_header;
2146 }
2147
2148 static inline void skb_reset_network_header(struct sk_buff *skb)
2149 {
2150 skb->network_header = skb->data - skb->head;
2151 }
2152
2153 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2154 {
2155 skb_reset_network_header(skb);
2156 skb->network_header += offset;
2157 }
2158
2159 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2160 {
2161 return skb->head + skb->mac_header;
2162 }
2163
2164 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2165 {
2166 return skb->mac_header != (typeof(skb->mac_header))~0U;
2167 }
2168
2169 static inline void skb_reset_mac_header(struct sk_buff *skb)
2170 {
2171 skb->mac_header = skb->data - skb->head;
2172 }
2173
2174 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2175 {
2176 skb_reset_mac_header(skb);
2177 skb->mac_header += offset;
2178 }
2179
2180 static inline void skb_pop_mac_header(struct sk_buff *skb)
2181 {
2182 skb->mac_header = skb->network_header;
2183 }
2184
2185 static inline void skb_probe_transport_header(struct sk_buff *skb,
2186 const int offset_hint)
2187 {
2188 struct flow_keys keys;
2189
2190 if (skb_transport_header_was_set(skb))
2191 return;
2192 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2193 skb_set_transport_header(skb, keys.control.thoff);
2194 else
2195 skb_set_transport_header(skb, offset_hint);
2196 }
2197
2198 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2199 {
2200 if (skb_mac_header_was_set(skb)) {
2201 const unsigned char *old_mac = skb_mac_header(skb);
2202
2203 skb_set_mac_header(skb, -skb->mac_len);
2204 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2205 }
2206 }
2207
2208 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2209 {
2210 return skb->csum_start - skb_headroom(skb);
2211 }
2212
2213 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2214 {
2215 return skb->head + skb->csum_start;
2216 }
2217
2218 static inline int skb_transport_offset(const struct sk_buff *skb)
2219 {
2220 return skb_transport_header(skb) - skb->data;
2221 }
2222
2223 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2224 {
2225 return skb->transport_header - skb->network_header;
2226 }
2227
2228 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2229 {
2230 return skb->inner_transport_header - skb->inner_network_header;
2231 }
2232
2233 static inline int skb_network_offset(const struct sk_buff *skb)
2234 {
2235 return skb_network_header(skb) - skb->data;
2236 }
2237
2238 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2239 {
2240 return skb_inner_network_header(skb) - skb->data;
2241 }
2242
2243 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2244 {
2245 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2246 }
2247
2248 /*
2249 * CPUs often take a performance hit when accessing unaligned memory
2250 * locations. The actual performance hit varies, it can be small if the
2251 * hardware handles it or large if we have to take an exception and fix it
2252 * in software.
2253 *
2254 * Since an ethernet header is 14 bytes network drivers often end up with
2255 * the IP header at an unaligned offset. The IP header can be aligned by
2256 * shifting the start of the packet by 2 bytes. Drivers should do this
2257 * with:
2258 *
2259 * skb_reserve(skb, NET_IP_ALIGN);
2260 *
2261 * The downside to this alignment of the IP header is that the DMA is now
2262 * unaligned. On some architectures the cost of an unaligned DMA is high
2263 * and this cost outweighs the gains made by aligning the IP header.
2264 *
2265 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2266 * to be overridden.
2267 */
2268 #ifndef NET_IP_ALIGN
2269 #define NET_IP_ALIGN 2
2270 #endif
2271
2272 /*
2273 * The networking layer reserves some headroom in skb data (via
2274 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2275 * the header has to grow. In the default case, if the header has to grow
2276 * 32 bytes or less we avoid the reallocation.
2277 *
2278 * Unfortunately this headroom changes the DMA alignment of the resulting
2279 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2280 * on some architectures. An architecture can override this value,
2281 * perhaps setting it to a cacheline in size (since that will maintain
2282 * cacheline alignment of the DMA). It must be a power of 2.
2283 *
2284 * Various parts of the networking layer expect at least 32 bytes of
2285 * headroom, you should not reduce this.
2286 *
2287 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2288 * to reduce average number of cache lines per packet.
2289 * get_rps_cpus() for example only access one 64 bytes aligned block :
2290 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2291 */
2292 #ifndef NET_SKB_PAD
2293 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2294 #endif
2295
2296 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2297
2298 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2299 {
2300 if (unlikely(skb_is_nonlinear(skb))) {
2301 WARN_ON(1);
2302 return;
2303 }
2304 skb->len = len;
2305 skb_set_tail_pointer(skb, len);
2306 }
2307
2308 void skb_trim(struct sk_buff *skb, unsigned int len);
2309
2310 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2311 {
2312 if (skb->data_len)
2313 return ___pskb_trim(skb, len);
2314 __skb_trim(skb, len);
2315 return 0;
2316 }
2317
2318 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2319 {
2320 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2321 }
2322
2323 /**
2324 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2325 * @skb: buffer to alter
2326 * @len: new length
2327 *
2328 * This is identical to pskb_trim except that the caller knows that
2329 * the skb is not cloned so we should never get an error due to out-
2330 * of-memory.
2331 */
2332 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2333 {
2334 int err = pskb_trim(skb, len);
2335 BUG_ON(err);
2336 }
2337
2338 /**
2339 * skb_orphan - orphan a buffer
2340 * @skb: buffer to orphan
2341 *
2342 * If a buffer currently has an owner then we call the owner's
2343 * destructor function and make the @skb unowned. The buffer continues
2344 * to exist but is no longer charged to its former owner.
2345 */
2346 static inline void skb_orphan(struct sk_buff *skb)
2347 {
2348 if (skb->destructor) {
2349 skb->destructor(skb);
2350 skb->destructor = NULL;
2351 skb->sk = NULL;
2352 } else {
2353 BUG_ON(skb->sk);
2354 }
2355 }
2356
2357 /**
2358 * skb_orphan_frags - orphan the frags contained in a buffer
2359 * @skb: buffer to orphan frags from
2360 * @gfp_mask: allocation mask for replacement pages
2361 *
2362 * For each frag in the SKB which needs a destructor (i.e. has an
2363 * owner) create a copy of that frag and release the original
2364 * page by calling the destructor.
2365 */
2366 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2367 {
2368 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2369 return 0;
2370 return skb_copy_ubufs(skb, gfp_mask);
2371 }
2372
2373 /**
2374 * __skb_queue_purge - empty a list
2375 * @list: list to empty
2376 *
2377 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2378 * the list and one reference dropped. This function does not take the
2379 * list lock and the caller must hold the relevant locks to use it.
2380 */
2381 void skb_queue_purge(struct sk_buff_head *list);
2382 static inline void __skb_queue_purge(struct sk_buff_head *list)
2383 {
2384 struct sk_buff *skb;
2385 while ((skb = __skb_dequeue(list)) != NULL)
2386 kfree_skb(skb);
2387 }
2388
2389 void *netdev_alloc_frag(unsigned int fragsz);
2390
2391 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2392 gfp_t gfp_mask);
2393
2394 /**
2395 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2396 * @dev: network device to receive on
2397 * @length: length to allocate
2398 *
2399 * Allocate a new &sk_buff and assign it a usage count of one. The
2400 * buffer has unspecified headroom built in. Users should allocate
2401 * the headroom they think they need without accounting for the
2402 * built in space. The built in space is used for optimisations.
2403 *
2404 * %NULL is returned if there is no free memory. Although this function
2405 * allocates memory it can be called from an interrupt.
2406 */
2407 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2408 unsigned int length)
2409 {
2410 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2411 }
2412
2413 /* legacy helper around __netdev_alloc_skb() */
2414 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2415 gfp_t gfp_mask)
2416 {
2417 return __netdev_alloc_skb(NULL, length, gfp_mask);
2418 }
2419
2420 /* legacy helper around netdev_alloc_skb() */
2421 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2422 {
2423 return netdev_alloc_skb(NULL, length);
2424 }
2425
2426
2427 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2428 unsigned int length, gfp_t gfp)
2429 {
2430 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2431
2432 if (NET_IP_ALIGN && skb)
2433 skb_reserve(skb, NET_IP_ALIGN);
2434 return skb;
2435 }
2436
2437 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2438 unsigned int length)
2439 {
2440 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2441 }
2442
2443 static inline void skb_free_frag(void *addr)
2444 {
2445 __free_page_frag(addr);
2446 }
2447
2448 void *napi_alloc_frag(unsigned int fragsz);
2449 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2450 unsigned int length, gfp_t gfp_mask);
2451 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2452 unsigned int length)
2453 {
2454 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2455 }
2456 void napi_consume_skb(struct sk_buff *skb, int budget);
2457
2458 void __kfree_skb_flush(void);
2459 void __kfree_skb_defer(struct sk_buff *skb);
2460
2461 /**
2462 * __dev_alloc_pages - allocate page for network Rx
2463 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2464 * @order: size of the allocation
2465 *
2466 * Allocate a new page.
2467 *
2468 * %NULL is returned if there is no free memory.
2469 */
2470 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2471 unsigned int order)
2472 {
2473 /* This piece of code contains several assumptions.
2474 * 1. This is for device Rx, therefor a cold page is preferred.
2475 * 2. The expectation is the user wants a compound page.
2476 * 3. If requesting a order 0 page it will not be compound
2477 * due to the check to see if order has a value in prep_new_page
2478 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2479 * code in gfp_to_alloc_flags that should be enforcing this.
2480 */
2481 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2482
2483 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2484 }
2485
2486 static inline struct page *dev_alloc_pages(unsigned int order)
2487 {
2488 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2489 }
2490
2491 /**
2492 * __dev_alloc_page - allocate a page for network Rx
2493 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2494 *
2495 * Allocate a new page.
2496 *
2497 * %NULL is returned if there is no free memory.
2498 */
2499 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2500 {
2501 return __dev_alloc_pages(gfp_mask, 0);
2502 }
2503
2504 static inline struct page *dev_alloc_page(void)
2505 {
2506 return dev_alloc_pages(0);
2507 }
2508
2509 /**
2510 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2511 * @page: The page that was allocated from skb_alloc_page
2512 * @skb: The skb that may need pfmemalloc set
2513 */
2514 static inline void skb_propagate_pfmemalloc(struct page *page,
2515 struct sk_buff *skb)
2516 {
2517 if (page_is_pfmemalloc(page))
2518 skb->pfmemalloc = true;
2519 }
2520
2521 /**
2522 * skb_frag_page - retrieve the page referred to by a paged fragment
2523 * @frag: the paged fragment
2524 *
2525 * Returns the &struct page associated with @frag.
2526 */
2527 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2528 {
2529 return frag->page.p;
2530 }
2531
2532 /**
2533 * __skb_frag_ref - take an addition reference on a paged fragment.
2534 * @frag: the paged fragment
2535 *
2536 * Takes an additional reference on the paged fragment @frag.
2537 */
2538 static inline void __skb_frag_ref(skb_frag_t *frag)
2539 {
2540 get_page(skb_frag_page(frag));
2541 }
2542
2543 /**
2544 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2545 * @skb: the buffer
2546 * @f: the fragment offset.
2547 *
2548 * Takes an additional reference on the @f'th paged fragment of @skb.
2549 */
2550 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2551 {
2552 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2553 }
2554
2555 /**
2556 * __skb_frag_unref - release a reference on a paged fragment.
2557 * @frag: the paged fragment
2558 *
2559 * Releases a reference on the paged fragment @frag.
2560 */
2561 static inline void __skb_frag_unref(skb_frag_t *frag)
2562 {
2563 put_page(skb_frag_page(frag));
2564 }
2565
2566 /**
2567 * skb_frag_unref - release a reference on a paged fragment of an skb.
2568 * @skb: the buffer
2569 * @f: the fragment offset
2570 *
2571 * Releases a reference on the @f'th paged fragment of @skb.
2572 */
2573 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2574 {
2575 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2576 }
2577
2578 /**
2579 * skb_frag_address - gets the address of the data contained in a paged fragment
2580 * @frag: the paged fragment buffer
2581 *
2582 * Returns the address of the data within @frag. The page must already
2583 * be mapped.
2584 */
2585 static inline void *skb_frag_address(const skb_frag_t *frag)
2586 {
2587 return page_address(skb_frag_page(frag)) + frag->page_offset;
2588 }
2589
2590 /**
2591 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2592 * @frag: the paged fragment buffer
2593 *
2594 * Returns the address of the data within @frag. Checks that the page
2595 * is mapped and returns %NULL otherwise.
2596 */
2597 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2598 {
2599 void *ptr = page_address(skb_frag_page(frag));
2600 if (unlikely(!ptr))
2601 return NULL;
2602
2603 return ptr + frag->page_offset;
2604 }
2605
2606 /**
2607 * __skb_frag_set_page - sets the page contained in a paged fragment
2608 * @frag: the paged fragment
2609 * @page: the page to set
2610 *
2611 * Sets the fragment @frag to contain @page.
2612 */
2613 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2614 {
2615 frag->page.p = page;
2616 }
2617
2618 /**
2619 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2620 * @skb: the buffer
2621 * @f: the fragment offset
2622 * @page: the page to set
2623 *
2624 * Sets the @f'th fragment of @skb to contain @page.
2625 */
2626 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2627 struct page *page)
2628 {
2629 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2630 }
2631
2632 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2633
2634 /**
2635 * skb_frag_dma_map - maps a paged fragment via the DMA API
2636 * @dev: the device to map the fragment to
2637 * @frag: the paged fragment to map
2638 * @offset: the offset within the fragment (starting at the
2639 * fragment's own offset)
2640 * @size: the number of bytes to map
2641 * @dir: the direction of the mapping (%PCI_DMA_*)
2642 *
2643 * Maps the page associated with @frag to @device.
2644 */
2645 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2646 const skb_frag_t *frag,
2647 size_t offset, size_t size,
2648 enum dma_data_direction dir)
2649 {
2650 return dma_map_page(dev, skb_frag_page(frag),
2651 frag->page_offset + offset, size, dir);
2652 }
2653
2654 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2655 gfp_t gfp_mask)
2656 {
2657 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2658 }
2659
2660
2661 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2662 gfp_t gfp_mask)
2663 {
2664 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2665 }
2666
2667
2668 /**
2669 * skb_clone_writable - is the header of a clone writable
2670 * @skb: buffer to check
2671 * @len: length up to which to write
2672 *
2673 * Returns true if modifying the header part of the cloned buffer
2674 * does not requires the data to be copied.
2675 */
2676 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2677 {
2678 return !skb_header_cloned(skb) &&
2679 skb_headroom(skb) + len <= skb->hdr_len;
2680 }
2681
2682 static inline int skb_try_make_writable(struct sk_buff *skb,
2683 unsigned int write_len)
2684 {
2685 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2686 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2687 }
2688
2689 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2690 int cloned)
2691 {
2692 int delta = 0;
2693
2694 if (headroom > skb_headroom(skb))
2695 delta = headroom - skb_headroom(skb);
2696
2697 if (delta || cloned)
2698 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2699 GFP_ATOMIC);
2700 return 0;
2701 }
2702
2703 /**
2704 * skb_cow - copy header of skb when it is required
2705 * @skb: buffer to cow
2706 * @headroom: needed headroom
2707 *
2708 * If the skb passed lacks sufficient headroom or its data part
2709 * is shared, data is reallocated. If reallocation fails, an error
2710 * is returned and original skb is not changed.
2711 *
2712 * The result is skb with writable area skb->head...skb->tail
2713 * and at least @headroom of space at head.
2714 */
2715 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2716 {
2717 return __skb_cow(skb, headroom, skb_cloned(skb));
2718 }
2719
2720 /**
2721 * skb_cow_head - skb_cow but only making the head writable
2722 * @skb: buffer to cow
2723 * @headroom: needed headroom
2724 *
2725 * This function is identical to skb_cow except that we replace the
2726 * skb_cloned check by skb_header_cloned. It should be used when
2727 * you only need to push on some header and do not need to modify
2728 * the data.
2729 */
2730 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2731 {
2732 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2733 }
2734
2735 /**
2736 * skb_padto - pad an skbuff up to a minimal size
2737 * @skb: buffer to pad
2738 * @len: minimal length
2739 *
2740 * Pads up a buffer to ensure the trailing bytes exist and are
2741 * blanked. If the buffer already contains sufficient data it
2742 * is untouched. Otherwise it is extended. Returns zero on
2743 * success. The skb is freed on error.
2744 */
2745 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2746 {
2747 unsigned int size = skb->len;
2748 if (likely(size >= len))
2749 return 0;
2750 return skb_pad(skb, len - size);
2751 }
2752
2753 /**
2754 * skb_put_padto - increase size and 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_put_padto(struct sk_buff *skb, unsigned int len)
2764 {
2765 unsigned int size = skb->len;
2766
2767 if (unlikely(size < len)) {
2768 len -= size;
2769 if (skb_pad(skb, len))
2770 return -ENOMEM;
2771 __skb_put(skb, len);
2772 }
2773 return 0;
2774 }
2775
2776 static inline int skb_add_data(struct sk_buff *skb,
2777 struct iov_iter *from, int copy)
2778 {
2779 const int off = skb->len;
2780
2781 if (skb->ip_summed == CHECKSUM_NONE) {
2782 __wsum csum = 0;
2783 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2784 &csum, from) == copy) {
2785 skb->csum = csum_block_add(skb->csum, csum, off);
2786 return 0;
2787 }
2788 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2789 return 0;
2790
2791 __skb_trim(skb, off);
2792 return -EFAULT;
2793 }
2794
2795 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2796 const struct page *page, int off)
2797 {
2798 if (i) {
2799 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2800
2801 return page == skb_frag_page(frag) &&
2802 off == frag->page_offset + skb_frag_size(frag);
2803 }
2804 return false;
2805 }
2806
2807 static inline int __skb_linearize(struct sk_buff *skb)
2808 {
2809 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2810 }
2811
2812 /**
2813 * skb_linearize - convert paged skb to linear one
2814 * @skb: buffer to linarize
2815 *
2816 * If there is no free memory -ENOMEM is returned, otherwise zero
2817 * is returned and the old skb data released.
2818 */
2819 static inline int skb_linearize(struct sk_buff *skb)
2820 {
2821 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2822 }
2823
2824 /**
2825 * skb_has_shared_frag - can any frag be overwritten
2826 * @skb: buffer to test
2827 *
2828 * Return true if the skb has at least one frag that might be modified
2829 * by an external entity (as in vmsplice()/sendfile())
2830 */
2831 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2832 {
2833 return skb_is_nonlinear(skb) &&
2834 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2835 }
2836
2837 /**
2838 * skb_linearize_cow - make sure skb is linear and writable
2839 * @skb: buffer to process
2840 *
2841 * If there is no free memory -ENOMEM is returned, otherwise zero
2842 * is returned and the old skb data released.
2843 */
2844 static inline int skb_linearize_cow(struct sk_buff *skb)
2845 {
2846 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2847 __skb_linearize(skb) : 0;
2848 }
2849
2850 /**
2851 * skb_postpull_rcsum - update checksum for received skb after pull
2852 * @skb: buffer to update
2853 * @start: start of data before pull
2854 * @len: length of data pulled
2855 *
2856 * After doing a pull on a received packet, you need to call this to
2857 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2858 * CHECKSUM_NONE so that it can be recomputed from scratch.
2859 */
2860
2861 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2862 const void *start, unsigned int len)
2863 {
2864 if (skb->ip_summed == CHECKSUM_COMPLETE)
2865 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2866 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2867 skb_checksum_start_offset(skb) < 0)
2868 skb->ip_summed = CHECKSUM_NONE;
2869 }
2870
2871 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2872
2873 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2874 const void *start, unsigned int len)
2875 {
2876 /* For performing the reverse operation to skb_postpull_rcsum(),
2877 * we can instead of ...
2878 *
2879 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2880 *
2881 * ... just use this equivalent version here to save a few
2882 * instructions. Feeding csum of 0 in csum_partial() and later
2883 * on adding skb->csum is equivalent to feed skb->csum in the
2884 * first place.
2885 */
2886 if (skb->ip_summed == CHECKSUM_COMPLETE)
2887 skb->csum = csum_partial(start, len, skb->csum);
2888 }
2889
2890 /**
2891 * skb_push_rcsum - push skb and update receive checksum
2892 * @skb: buffer to update
2893 * @len: length of data pulled
2894 *
2895 * This function performs an skb_push on the packet and updates
2896 * the CHECKSUM_COMPLETE checksum. It should be used on
2897 * receive path processing instead of skb_push unless you know
2898 * that the checksum difference is zero (e.g., a valid IP header)
2899 * or you are setting ip_summed to CHECKSUM_NONE.
2900 */
2901 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2902 unsigned int len)
2903 {
2904 skb_push(skb, len);
2905 skb_postpush_rcsum(skb, skb->data, len);
2906 return skb->data;
2907 }
2908
2909 /**
2910 * pskb_trim_rcsum - trim received skb and update checksum
2911 * @skb: buffer to trim
2912 * @len: new length
2913 *
2914 * This is exactly the same as pskb_trim except that it ensures the
2915 * checksum of received packets are still valid after the operation.
2916 */
2917
2918 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2919 {
2920 if (likely(len >= skb->len))
2921 return 0;
2922 if (skb->ip_summed == CHECKSUM_COMPLETE)
2923 skb->ip_summed = CHECKSUM_NONE;
2924 return __pskb_trim(skb, len);
2925 }
2926
2927 #define skb_queue_walk(queue, skb) \
2928 for (skb = (queue)->next; \
2929 skb != (struct sk_buff *)(queue); \
2930 skb = skb->next)
2931
2932 #define skb_queue_walk_safe(queue, skb, tmp) \
2933 for (skb = (queue)->next, tmp = skb->next; \
2934 skb != (struct sk_buff *)(queue); \
2935 skb = tmp, tmp = skb->next)
2936
2937 #define skb_queue_walk_from(queue, skb) \
2938 for (; skb != (struct sk_buff *)(queue); \
2939 skb = skb->next)
2940
2941 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2942 for (tmp = skb->next; \
2943 skb != (struct sk_buff *)(queue); \
2944 skb = tmp, tmp = skb->next)
2945
2946 #define skb_queue_reverse_walk(queue, skb) \
2947 for (skb = (queue)->prev; \
2948 skb != (struct sk_buff *)(queue); \
2949 skb = skb->prev)
2950
2951 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2952 for (skb = (queue)->prev, tmp = skb->prev; \
2953 skb != (struct sk_buff *)(queue); \
2954 skb = tmp, tmp = skb->prev)
2955
2956 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2957 for (tmp = skb->prev; \
2958 skb != (struct sk_buff *)(queue); \
2959 skb = tmp, tmp = skb->prev)
2960
2961 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2962 {
2963 return skb_shinfo(skb)->frag_list != NULL;
2964 }
2965
2966 static inline void skb_frag_list_init(struct sk_buff *skb)
2967 {
2968 skb_shinfo(skb)->frag_list = NULL;
2969 }
2970
2971 #define skb_walk_frags(skb, iter) \
2972 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2973
2974
2975 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2976 const struct sk_buff *skb);
2977 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2978 int *peeked, int *off, int *err,
2979 struct sk_buff **last);
2980 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2981 int *peeked, int *off, int *err);
2982 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2983 int *err);
2984 unsigned int datagram_poll(struct file *file, struct socket *sock,
2985 struct poll_table_struct *wait);
2986 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2987 struct iov_iter *to, int size);
2988 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2989 struct msghdr *msg, int size)
2990 {
2991 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2992 }
2993 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2994 struct msghdr *msg);
2995 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2996 struct iov_iter *from, int len);
2997 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2998 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2999 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3000 static inline void skb_free_datagram_locked(struct sock *sk,
3001 struct sk_buff *skb)
3002 {
3003 __skb_free_datagram_locked(sk, skb, 0);
3004 }
3005 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3006 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3007 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3008 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3009 int len, __wsum csum);
3010 ssize_t skb_socket_splice(struct sock *sk,
3011 struct pipe_inode_info *pipe,
3012 struct splice_pipe_desc *spd);
3013 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3014 struct pipe_inode_info *pipe, unsigned int len,
3015 unsigned int flags,
3016 ssize_t (*splice_cb)(struct sock *,
3017 struct pipe_inode_info *,
3018 struct splice_pipe_desc *));
3019 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3020 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3021 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3022 int len, int hlen);
3023 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3024 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3025 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3026 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3027 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3028 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3029 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3030 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3031 int skb_vlan_pop(struct sk_buff *skb);
3032 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3033 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3034 gfp_t gfp);
3035
3036 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3037 {
3038 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3039 }
3040
3041 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3042 {
3043 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3044 }
3045
3046 struct skb_checksum_ops {
3047 __wsum (*update)(const void *mem, int len, __wsum wsum);
3048 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3049 };
3050
3051 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3052 __wsum csum, const struct skb_checksum_ops *ops);
3053 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3054 __wsum csum);
3055
3056 static inline void * __must_check
3057 __skb_header_pointer(const struct sk_buff *skb, int offset,
3058 int len, void *data, int hlen, void *buffer)
3059 {
3060 if (hlen - offset >= len)
3061 return data + offset;
3062
3063 if (!skb ||
3064 skb_copy_bits(skb, offset, buffer, len) < 0)
3065 return NULL;
3066
3067 return buffer;
3068 }
3069
3070 static inline void * __must_check
3071 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3072 {
3073 return __skb_header_pointer(skb, offset, len, skb->data,
3074 skb_headlen(skb), buffer);
3075 }
3076
3077 /**
3078 * skb_needs_linearize - check if we need to linearize a given skb
3079 * depending on the given device features.
3080 * @skb: socket buffer to check
3081 * @features: net device features
3082 *
3083 * Returns true if either:
3084 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3085 * 2. skb is fragmented and the device does not support SG.
3086 */
3087 static inline bool skb_needs_linearize(struct sk_buff *skb,
3088 netdev_features_t features)
3089 {
3090 return skb_is_nonlinear(skb) &&
3091 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3092 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3093 }
3094
3095 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3096 void *to,
3097 const unsigned int len)
3098 {
3099 memcpy(to, skb->data, len);
3100 }
3101
3102 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3103 const int offset, void *to,
3104 const unsigned int len)
3105 {
3106 memcpy(to, skb->data + offset, len);
3107 }
3108
3109 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3110 const void *from,
3111 const unsigned int len)
3112 {
3113 memcpy(skb->data, from, len);
3114 }
3115
3116 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3117 const int offset,
3118 const void *from,
3119 const unsigned int len)
3120 {
3121 memcpy(skb->data + offset, from, len);
3122 }
3123
3124 void skb_init(void);
3125
3126 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3127 {
3128 return skb->tstamp;
3129 }
3130
3131 /**
3132 * skb_get_timestamp - get timestamp from a skb
3133 * @skb: skb to get stamp from
3134 * @stamp: pointer to struct timeval to store stamp in
3135 *
3136 * Timestamps are stored in the skb as offsets to a base timestamp.
3137 * This function converts the offset back to a struct timeval and stores
3138 * it in stamp.
3139 */
3140 static inline void skb_get_timestamp(const struct sk_buff *skb,
3141 struct timeval *stamp)
3142 {
3143 *stamp = ktime_to_timeval(skb->tstamp);
3144 }
3145
3146 static inline void skb_get_timestampns(const struct sk_buff *skb,
3147 struct timespec *stamp)
3148 {
3149 *stamp = ktime_to_timespec(skb->tstamp);
3150 }
3151
3152 static inline void __net_timestamp(struct sk_buff *skb)
3153 {
3154 skb->tstamp = ktime_get_real();
3155 }
3156
3157 static inline ktime_t net_timedelta(ktime_t t)
3158 {
3159 return ktime_sub(ktime_get_real(), t);
3160 }
3161
3162 static inline ktime_t net_invalid_timestamp(void)
3163 {
3164 return ktime_set(0, 0);
3165 }
3166
3167 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3168
3169 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3170
3171 void skb_clone_tx_timestamp(struct sk_buff *skb);
3172 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3173
3174 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3175
3176 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3177 {
3178 }
3179
3180 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3181 {
3182 return false;
3183 }
3184
3185 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3186
3187 /**
3188 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3189 *
3190 * PHY drivers may accept clones of transmitted packets for
3191 * timestamping via their phy_driver.txtstamp method. These drivers
3192 * must call this function to return the skb back to the stack with a
3193 * timestamp.
3194 *
3195 * @skb: clone of the the original outgoing packet
3196 * @hwtstamps: hardware time stamps
3197 *
3198 */
3199 void skb_complete_tx_timestamp(struct sk_buff *skb,
3200 struct skb_shared_hwtstamps *hwtstamps);
3201
3202 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3203 struct skb_shared_hwtstamps *hwtstamps,
3204 struct sock *sk, int tstype);
3205
3206 /**
3207 * skb_tstamp_tx - queue clone of skb with send time stamps
3208 * @orig_skb: the original outgoing packet
3209 * @hwtstamps: hardware time stamps, may be NULL if not available
3210 *
3211 * If the skb has a socket associated, then this function clones the
3212 * skb (thus sharing the actual data and optional structures), stores
3213 * the optional hardware time stamping information (if non NULL) or
3214 * generates a software time stamp (otherwise), then queues the clone
3215 * to the error queue of the socket. Errors are silently ignored.
3216 */
3217 void skb_tstamp_tx(struct sk_buff *orig_skb,
3218 struct skb_shared_hwtstamps *hwtstamps);
3219
3220 static inline void sw_tx_timestamp(struct sk_buff *skb)
3221 {
3222 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3223 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3224 skb_tstamp_tx(skb, NULL);
3225 }
3226
3227 /**
3228 * skb_tx_timestamp() - Driver hook for transmit timestamping
3229 *
3230 * Ethernet MAC Drivers should call this function in their hard_xmit()
3231 * function immediately before giving the sk_buff to the MAC hardware.
3232 *
3233 * Specifically, one should make absolutely sure that this function is
3234 * called before TX completion of this packet can trigger. Otherwise
3235 * the packet could potentially already be freed.
3236 *
3237 * @skb: A socket buffer.
3238 */
3239 static inline void skb_tx_timestamp(struct sk_buff *skb)
3240 {
3241 skb_clone_tx_timestamp(skb);
3242 sw_tx_timestamp(skb);
3243 }
3244
3245 /**
3246 * skb_complete_wifi_ack - deliver skb with wifi status
3247 *
3248 * @skb: the original outgoing packet
3249 * @acked: ack status
3250 *
3251 */
3252 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3253
3254 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3255 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3256
3257 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3258 {
3259 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3260 skb->csum_valid ||
3261 (skb->ip_summed == CHECKSUM_PARTIAL &&
3262 skb_checksum_start_offset(skb) >= 0));
3263 }
3264
3265 /**
3266 * skb_checksum_complete - Calculate checksum of an entire packet
3267 * @skb: packet to process
3268 *
3269 * This function calculates the checksum over the entire packet plus
3270 * the value of skb->csum. The latter can be used to supply the
3271 * checksum of a pseudo header as used by TCP/UDP. It returns the
3272 * checksum.
3273 *
3274 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3275 * this function can be used to verify that checksum on received
3276 * packets. In that case the function should return zero if the
3277 * checksum is correct. In particular, this function will return zero
3278 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3279 * hardware has already verified the correctness of the checksum.
3280 */
3281 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3282 {
3283 return skb_csum_unnecessary(skb) ?
3284 0 : __skb_checksum_complete(skb);
3285 }
3286
3287 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3288 {
3289 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3290 if (skb->csum_level == 0)
3291 skb->ip_summed = CHECKSUM_NONE;
3292 else
3293 skb->csum_level--;
3294 }
3295 }
3296
3297 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3298 {
3299 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3300 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3301 skb->csum_level++;
3302 } else if (skb->ip_summed == CHECKSUM_NONE) {
3303 skb->ip_summed = CHECKSUM_UNNECESSARY;
3304 skb->csum_level = 0;
3305 }
3306 }
3307
3308 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3309 {
3310 /* Mark current checksum as bad (typically called from GRO
3311 * path). In the case that ip_summed is CHECKSUM_NONE
3312 * this must be the first checksum encountered in the packet.
3313 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3314 * checksum after the last one validated. For UDP, a zero
3315 * checksum can not be marked as bad.
3316 */
3317
3318 if (skb->ip_summed == CHECKSUM_NONE ||
3319 skb->ip_summed == CHECKSUM_UNNECESSARY)
3320 skb->csum_bad = 1;
3321 }
3322
3323 /* Check if we need to perform checksum complete validation.
3324 *
3325 * Returns true if checksum complete is needed, false otherwise
3326 * (either checksum is unnecessary or zero checksum is allowed).
3327 */
3328 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3329 bool zero_okay,
3330 __sum16 check)
3331 {
3332 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3333 skb->csum_valid = 1;
3334 __skb_decr_checksum_unnecessary(skb);
3335 return false;
3336 }
3337
3338 return true;
3339 }
3340
3341 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3342 * in checksum_init.
3343 */
3344 #define CHECKSUM_BREAK 76
3345
3346 /* Unset checksum-complete
3347 *
3348 * Unset checksum complete can be done when packet is being modified
3349 * (uncompressed for instance) and checksum-complete value is
3350 * invalidated.
3351 */
3352 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3353 {
3354 if (skb->ip_summed == CHECKSUM_COMPLETE)
3355 skb->ip_summed = CHECKSUM_NONE;
3356 }
3357
3358 /* Validate (init) checksum based on checksum complete.
3359 *
3360 * Return values:
3361 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3362 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3363 * checksum is stored in skb->csum for use in __skb_checksum_complete
3364 * non-zero: value of invalid checksum
3365 *
3366 */
3367 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3368 bool complete,
3369 __wsum psum)
3370 {
3371 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3372 if (!csum_fold(csum_add(psum, skb->csum))) {
3373 skb->csum_valid = 1;
3374 return 0;
3375 }
3376 } else if (skb->csum_bad) {
3377 /* ip_summed == CHECKSUM_NONE in this case */
3378 return (__force __sum16)1;
3379 }
3380
3381 skb->csum = psum;
3382
3383 if (complete || skb->len <= CHECKSUM_BREAK) {
3384 __sum16 csum;
3385
3386 csum = __skb_checksum_complete(skb);
3387 skb->csum_valid = !csum;
3388 return csum;
3389 }
3390
3391 return 0;
3392 }
3393
3394 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3395 {
3396 return 0;
3397 }
3398
3399 /* Perform checksum validate (init). Note that this is a macro since we only
3400 * want to calculate the pseudo header which is an input function if necessary.
3401 * First we try to validate without any computation (checksum unnecessary) and
3402 * then calculate based on checksum complete calling the function to compute
3403 * pseudo header.
3404 *
3405 * Return values:
3406 * 0: checksum is validated or try to in skb_checksum_complete
3407 * non-zero: value of invalid checksum
3408 */
3409 #define __skb_checksum_validate(skb, proto, complete, \
3410 zero_okay, check, compute_pseudo) \
3411 ({ \
3412 __sum16 __ret = 0; \
3413 skb->csum_valid = 0; \
3414 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3415 __ret = __skb_checksum_validate_complete(skb, \
3416 complete, compute_pseudo(skb, proto)); \
3417 __ret; \
3418 })
3419
3420 #define skb_checksum_init(skb, proto, compute_pseudo) \
3421 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3422
3423 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3424 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3425
3426 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3427 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3428
3429 #define skb_checksum_validate_zero_check(skb, proto, check, \
3430 compute_pseudo) \
3431 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3432
3433 #define skb_checksum_simple_validate(skb) \
3434 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3435
3436 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3437 {
3438 return (skb->ip_summed == CHECKSUM_NONE &&
3439 skb->csum_valid && !skb->csum_bad);
3440 }
3441
3442 static inline void __skb_checksum_convert(struct sk_buff *skb,
3443 __sum16 check, __wsum pseudo)
3444 {
3445 skb->csum = ~pseudo;
3446 skb->ip_summed = CHECKSUM_COMPLETE;
3447 }
3448
3449 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3450 do { \
3451 if (__skb_checksum_convert_check(skb)) \
3452 __skb_checksum_convert(skb, check, \
3453 compute_pseudo(skb, proto)); \
3454 } while (0)
3455
3456 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3457 u16 start, u16 offset)
3458 {
3459 skb->ip_summed = CHECKSUM_PARTIAL;
3460 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3461 skb->csum_offset = offset - start;
3462 }
3463
3464 /* Update skbuf and packet to reflect the remote checksum offload operation.
3465 * When called, ptr indicates the starting point for skb->csum when
3466 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3467 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3468 */
3469 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3470 int start, int offset, bool nopartial)
3471 {
3472 __wsum delta;
3473
3474 if (!nopartial) {
3475 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3476 return;
3477 }
3478
3479 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3480 __skb_checksum_complete(skb);
3481 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3482 }
3483
3484 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3485
3486 /* Adjust skb->csum since we changed the packet */
3487 skb->csum = csum_add(skb->csum, delta);
3488 }
3489
3490 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3491 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3492 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3493 {
3494 if (nfct && atomic_dec_and_test(&nfct->use))
3495 nf_conntrack_destroy(nfct);
3496 }
3497 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3498 {
3499 if (nfct)
3500 atomic_inc(&nfct->use);
3501 }
3502 #endif
3503 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3504 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3505 {
3506 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3507 kfree(nf_bridge);
3508 }
3509 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3510 {
3511 if (nf_bridge)
3512 atomic_inc(&nf_bridge->use);
3513 }
3514 #endif /* CONFIG_BRIDGE_NETFILTER */
3515 static inline void nf_reset(struct sk_buff *skb)
3516 {
3517 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3518 nf_conntrack_put(skb->nfct);
3519 skb->nfct = NULL;
3520 #endif
3521 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3522 nf_bridge_put(skb->nf_bridge);
3523 skb->nf_bridge = NULL;
3524 #endif
3525 }
3526
3527 static inline void nf_reset_trace(struct sk_buff *skb)
3528 {
3529 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3530 skb->nf_trace = 0;
3531 #endif
3532 }
3533
3534 /* Note: This doesn't put any conntrack and bridge info in dst. */
3535 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3536 bool copy)
3537 {
3538 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3539 dst->nfct = src->nfct;
3540 nf_conntrack_get(src->nfct);
3541 if (copy)
3542 dst->nfctinfo = src->nfctinfo;
3543 #endif
3544 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3545 dst->nf_bridge = src->nf_bridge;
3546 nf_bridge_get(src->nf_bridge);
3547 #endif
3548 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3549 if (copy)
3550 dst->nf_trace = src->nf_trace;
3551 #endif
3552 }
3553
3554 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3555 {
3556 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3557 nf_conntrack_put(dst->nfct);
3558 #endif
3559 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3560 nf_bridge_put(dst->nf_bridge);
3561 #endif
3562 __nf_copy(dst, src, true);
3563 }
3564
3565 #ifdef CONFIG_NETWORK_SECMARK
3566 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3567 {
3568 to->secmark = from->secmark;
3569 }
3570
3571 static inline void skb_init_secmark(struct sk_buff *skb)
3572 {
3573 skb->secmark = 0;
3574 }
3575 #else
3576 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3577 { }
3578
3579 static inline void skb_init_secmark(struct sk_buff *skb)
3580 { }
3581 #endif
3582
3583 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3584 {
3585 return !skb->destructor &&
3586 #if IS_ENABLED(CONFIG_XFRM)
3587 !skb->sp &&
3588 #endif
3589 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3590 !skb->nfct &&
3591 #endif
3592 !skb->_skb_refdst &&
3593 !skb_has_frag_list(skb);
3594 }
3595
3596 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3597 {
3598 skb->queue_mapping = queue_mapping;
3599 }
3600
3601 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3602 {
3603 return skb->queue_mapping;
3604 }
3605
3606 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3607 {
3608 to->queue_mapping = from->queue_mapping;
3609 }
3610
3611 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3612 {
3613 skb->queue_mapping = rx_queue + 1;
3614 }
3615
3616 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3617 {
3618 return skb->queue_mapping - 1;
3619 }
3620
3621 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3622 {
3623 return skb->queue_mapping != 0;
3624 }
3625
3626 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3627 {
3628 #ifdef CONFIG_XFRM
3629 return skb->sp;
3630 #else
3631 return NULL;
3632 #endif
3633 }
3634
3635 /* Keeps track of mac header offset relative to skb->head.
3636 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3637 * For non-tunnel skb it points to skb_mac_header() and for
3638 * tunnel skb it points to outer mac header.
3639 * Keeps track of level of encapsulation of network headers.
3640 */
3641 struct skb_gso_cb {
3642 union {
3643 int mac_offset;
3644 int data_offset;
3645 };
3646 int encap_level;
3647 __wsum csum;
3648 __u16 csum_start;
3649 };
3650 #define SKB_SGO_CB_OFFSET 32
3651 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3652
3653 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3654 {
3655 return (skb_mac_header(inner_skb) - inner_skb->head) -
3656 SKB_GSO_CB(inner_skb)->mac_offset;
3657 }
3658
3659 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3660 {
3661 int new_headroom, headroom;
3662 int ret;
3663
3664 headroom = skb_headroom(skb);
3665 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3666 if (ret)
3667 return ret;
3668
3669 new_headroom = skb_headroom(skb);
3670 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3671 return 0;
3672 }
3673
3674 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3675 {
3676 /* Do not update partial checksums if remote checksum is enabled. */
3677 if (skb->remcsum_offload)
3678 return;
3679
3680 SKB_GSO_CB(skb)->csum = res;
3681 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3682 }
3683
3684 /* Compute the checksum for a gso segment. First compute the checksum value
3685 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3686 * then add in skb->csum (checksum from csum_start to end of packet).
3687 * skb->csum and csum_start are then updated to reflect the checksum of the
3688 * resultant packet starting from the transport header-- the resultant checksum
3689 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3690 * header.
3691 */
3692 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3693 {
3694 unsigned char *csum_start = skb_transport_header(skb);
3695 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3696 __wsum partial = SKB_GSO_CB(skb)->csum;
3697
3698 SKB_GSO_CB(skb)->csum = res;
3699 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3700
3701 return csum_fold(csum_partial(csum_start, plen, partial));
3702 }
3703
3704 static inline bool skb_is_gso(const struct sk_buff *skb)
3705 {
3706 return skb_shinfo(skb)->gso_size;
3707 }
3708
3709 /* Note: Should be called only if skb_is_gso(skb) is true */
3710 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3711 {
3712 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3713 }
3714
3715 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3716
3717 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3718 {
3719 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3720 * wanted then gso_type will be set. */
3721 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3722
3723 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3724 unlikely(shinfo->gso_type == 0)) {
3725 __skb_warn_lro_forwarding(skb);
3726 return true;
3727 }
3728 return false;
3729 }
3730
3731 static inline void skb_forward_csum(struct sk_buff *skb)
3732 {
3733 /* Unfortunately we don't support this one. Any brave souls? */
3734 if (skb->ip_summed == CHECKSUM_COMPLETE)
3735 skb->ip_summed = CHECKSUM_NONE;
3736 }
3737
3738 /**
3739 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3740 * @skb: skb to check
3741 *
3742 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3743 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3744 * use this helper, to document places where we make this assertion.
3745 */
3746 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3747 {
3748 #ifdef DEBUG
3749 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3750 #endif
3751 }
3752
3753 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3754
3755 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3756 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3757 unsigned int transport_len,
3758 __sum16(*skb_chkf)(struct sk_buff *skb));
3759
3760 /**
3761 * skb_head_is_locked - Determine if the skb->head is locked down
3762 * @skb: skb to check
3763 *
3764 * The head on skbs build around a head frag can be removed if they are
3765 * not cloned. This function returns true if the skb head is locked down
3766 * due to either being allocated via kmalloc, or by being a clone with
3767 * multiple references to the head.
3768 */
3769 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3770 {
3771 return !skb->head_frag || skb_cloned(skb);
3772 }
3773
3774 /**
3775 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3776 *
3777 * @skb: GSO skb
3778 *
3779 * skb_gso_network_seglen is used to determine the real size of the
3780 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3781 *
3782 * The MAC/L2 header is not accounted for.
3783 */
3784 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3785 {
3786 unsigned int hdr_len = skb_transport_header(skb) -
3787 skb_network_header(skb);
3788 return hdr_len + skb_gso_transport_seglen(skb);
3789 }
3790
3791 /* Local Checksum Offload.
3792 * Compute outer checksum based on the assumption that the
3793 * inner checksum will be offloaded later.
3794 * See Documentation/networking/checksum-offloads.txt for
3795 * explanation of how this works.
3796 * Fill in outer checksum adjustment (e.g. with sum of outer
3797 * pseudo-header) before calling.
3798 * Also ensure that inner checksum is in linear data area.
3799 */
3800 static inline __wsum lco_csum(struct sk_buff *skb)
3801 {
3802 unsigned char *csum_start = skb_checksum_start(skb);
3803 unsigned char *l4_hdr = skb_transport_header(skb);
3804 __wsum partial;
3805
3806 /* Start with complement of inner checksum adjustment */
3807 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3808 skb->csum_offset));
3809
3810 /* Add in checksum of our headers (incl. outer checksum
3811 * adjustment filled in by caller) and return result.
3812 */
3813 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3814 }
3815
3816 #endif /* __KERNEL__ */
3817 #endif /* _LINUX_SKBUFF_H */
This page took 0.186839 seconds and 5 git commands to generate.