1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * Definitions for the 'struct sk_buff' memory handlers.
4 *
5 * Authors:
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 */
9
10#ifndef _LINUX_SKBUFF_H
11#define _LINUX_SKBUFF_H
12
13#include <linux/kernel.h>
14#include <linux/compiler.h>
15#include <linux/time.h>
16#include <linux/bug.h>
17#include <linux/bvec.h>
18#include <linux/cache.h>
19#include <linux/rbtree.h>
20#include <linux/socket.h>
21#include <linux/refcount.h>
22
23#include <linux/atomic.h>
24#include <asm/types.h>
25#include <linux/spinlock.h>
26#include <linux/net.h>
27#include <linux/textsearch.h>
28#include <net/checksum.h>
29#include <linux/rcupdate.h>
30#include <linux/hrtimer.h>
31#include <linux/dma-mapping.h>
32#include <linux/netdev_features.h>
33#include <linux/sched.h>
34#include <linux/sched/clock.h>
35#include <net/flow_dissector.h>
36#include <linux/splice.h>
37#include <linux/in6.h>
38#include <linux/if_packet.h>
39#include <linux/llist.h>
40#include <net/flow.h>
41#include <net/page_pool.h>
42#if IS_ENABLED(CONFIG_NF_CONNTRACK)
43#include <linux/netfilter/nf_conntrack_common.h>
44#endif
45#include <net/net_debug.h>
46#include <net/dropreason.h>
47
48/**
49 * DOC: skb checksums
50 *
51 * The interface for checksum offload between the stack and networking drivers
52 * is as follows...
53 *
54 * IP checksum related features
55 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
56 *
57 * Drivers advertise checksum offload capabilities in the features of a device.
58 * From the stack's point of view these are capabilities offered by the driver.
59 * A driver typically only advertises features that it is capable of offloading
60 * to its device.
61 *
62 * .. flat-table:: Checksum related device features
63 * :widths: 1 10
64 *
65 * * - %NETIF_F_HW_CSUM
66 * - The driver (or its device) is able to compute one
67 * IP (one's complement) checksum for any combination
68 * of protocols or protocol layering. The checksum is
69 * computed and set in a packet per the CHECKSUM_PARTIAL
70 * interface (see below).
71 *
72 * * - %NETIF_F_IP_CSUM
73 * - Driver (device) is only able to checksum plain
74 * TCP or UDP packets over IPv4. These are specifically
75 * unencapsulated packets of the form IPv4|TCP or
76 * IPv4|UDP where the Protocol field in the IPv4 header
77 * is TCP or UDP. The IPv4 header may contain IP options.
78 * This feature cannot be set in features for a device
79 * with NETIF_F_HW_CSUM also set. This feature is being
80 * DEPRECATED (see below).
81 *
82 * * - %NETIF_F_IPV6_CSUM
83 * - Driver (device) is only able to checksum plain
84 * TCP or UDP packets over IPv6. These are specifically
85 * unencapsulated packets of the form IPv6|TCP or
86 * IPv6|UDP where the Next Header field in the IPv6
87 * header is either TCP or UDP. IPv6 extension headers
88 * are not supported with this feature. This feature
89 * cannot be set in features for a device with
90 * NETIF_F_HW_CSUM also set. This feature is being
91 * DEPRECATED (see below).
92 *
93 * * - %NETIF_F_RXCSUM
94 * - Driver (device) performs receive checksum offload.
95 * This flag is only used to disable the RX checksum
96 * feature for a device. The stack will accept receive
97 * checksum indication in packets received on a device
98 * regardless of whether NETIF_F_RXCSUM is set.
99 *
100 * Checksumming of received packets by device
101 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
102 *
103 * Indication of checksum verification is set in &sk_buff.ip_summed.
104 * Possible values are:
105 *
106 * - %CHECKSUM_NONE
107 *
108 * Device did not checksum this packet e.g. due to lack of capabilities.
109 * The packet contains full (though not verified) checksum in packet but
110 * not in skb->csum. Thus, skb->csum is undefined in this case.
111 *
112 * - %CHECKSUM_UNNECESSARY
113 *
114 * The hardware you're dealing with doesn't calculate the full checksum
115 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
116 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
117 * if their checksums are okay. &sk_buff.csum is still undefined in this case
118 * though. A driver or device must never modify the checksum field in the
119 * packet even if checksum is verified.
120 *
121 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
122 *
123 * - TCP: IPv6 and IPv4.
124 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
125 * zero UDP checksum for either IPv4 or IPv6, the networking stack
126 * may perform further validation in this case.
127 * - GRE: only if the checksum is present in the header.
128 * - SCTP: indicates the CRC in SCTP header has been validated.
129 * - FCOE: indicates the CRC in FC frame has been validated.
130 *
131 * &sk_buff.csum_level indicates the number of consecutive checksums found in
132 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
133 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
134 * and a device is able to verify the checksums for UDP (possibly zero),
135 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
136 * two. If the device were only able to verify the UDP checksum and not
137 * GRE, either because it doesn't support GRE checksum or because GRE
138 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
139 * not considered in this case).
140 *
141 * - %CHECKSUM_COMPLETE
142 *
143 * This is the most generic way. The device supplied checksum of the _whole_
144 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
145 * hardware doesn't need to parse L3/L4 headers to implement this.
146 *
147 * Notes:
148 *
149 * - Even if device supports only some protocols, but is able to produce
150 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
151 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
152 *
153 * - %CHECKSUM_PARTIAL
154 *
155 * A checksum is set up to be offloaded to a device as described in the
156 * output description for CHECKSUM_PARTIAL. This may occur on a packet
157 * received directly from another Linux OS, e.g., a virtualized Linux kernel
158 * on the same host, or it may be set in the input path in GRO or remote
159 * checksum offload. For the purposes of checksum verification, the checksum
160 * referred to by skb->csum_start + skb->csum_offset and any preceding
161 * checksums in the packet are considered verified. Any checksums in the
162 * packet that are after the checksum being offloaded are not considered to
163 * be verified.
164 *
165 * Checksumming on transmit for non-GSO
166 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
167 *
168 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
169 * Values are:
170 *
171 * - %CHECKSUM_PARTIAL
172 *
173 * The driver is required to checksum the packet as seen by hard_start_xmit()
174 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
175 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
176 * A driver may verify that the
177 * csum_start and csum_offset values are valid values given the length and
178 * offset of the packet, but it should not attempt to validate that the
179 * checksum refers to a legitimate transport layer checksum -- it is the
180 * purview of the stack to validate that csum_start and csum_offset are set
181 * correctly.
182 *
183 * When the stack requests checksum offload for a packet, the driver MUST
184 * ensure that the checksum is set correctly. A driver can either offload the
185 * checksum calculation to the device, or call skb_checksum_help (in the case
186 * that the device does not support offload for a particular checksum).
187 *
188 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
189 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
190 * checksum offload capability.
191 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
192 * on network device checksumming capabilities: if a packet does not match
193 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
194 * &sk_buff.csum_not_inet, see :ref:`crc`)
195 * is called to resolve the checksum.
196 *
197 * - %CHECKSUM_NONE
198 *
199 * The skb was already checksummed by the protocol, or a checksum is not
200 * required.
201 *
202 * - %CHECKSUM_UNNECESSARY
203 *
204 * This has the same meaning as CHECKSUM_NONE for checksum offload on
205 * output.
206 *
207 * - %CHECKSUM_COMPLETE
208 *
209 * Not used in checksum output. If a driver observes a packet with this value
210 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
211 *
212 * .. _crc:
213 *
214 * Non-IP checksum (CRC) offloads
215 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
216 *
217 * .. flat-table::
218 * :widths: 1 10
219 *
220 * * - %NETIF_F_SCTP_CRC
221 * - This feature indicates that a device is capable of
222 * offloading the SCTP CRC in a packet. To perform this offload the stack
223 * will set csum_start and csum_offset accordingly, set ip_summed to
224 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
225 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
226 * A driver that supports both IP checksum offload and SCTP CRC32c offload
227 * must verify which offload is configured for a packet by testing the
228 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
229 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
230 *
231 * * - %NETIF_F_FCOE_CRC
232 * - This feature indicates that a device is capable of offloading the FCOE
233 * CRC in a packet. To perform this offload the stack will set ip_summed
234 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
235 * accordingly. Note that there is no indication in the skbuff that the
236 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
237 * both IP checksum offload and FCOE CRC offload must verify which offload
238 * is configured for a packet, presumably by inspecting packet headers.
239 *
240 * Checksumming on output with GSO
241 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
242 *
243 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
244 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
245 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
246 * part of the GSO operation is implied. If a checksum is being offloaded
247 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
248 * csum_offset are set to refer to the outermost checksum being offloaded
249 * (two offloaded checksums are possible with UDP encapsulation).
250 */
251
252/* Don't change this without changing skb_csum_unnecessary! */
253#define CHECKSUM_NONE 0
254#define CHECKSUM_UNNECESSARY 1
255#define CHECKSUM_COMPLETE 2
256#define CHECKSUM_PARTIAL 3
257
258/* Maximum value in skb->csum_level */
259#define SKB_MAX_CSUM_LEVEL 3
260
261#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
262#define SKB_WITH_OVERHEAD(X) \
263 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
264#define SKB_MAX_ORDER(X, ORDER) \
265 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
266#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
267#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
268
269/* return minimum truesize of one skb containing X bytes of data */
270#define SKB_TRUESIZE(X) ((X) + \
271 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
272 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
273
274struct ahash_request;
275struct net_device;
276struct scatterlist;
277struct pipe_inode_info;
278struct iov_iter;
279struct napi_struct;
280struct bpf_prog;
281union bpf_attr;
282struct skb_ext;
283
284#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
285struct nf_bridge_info {
286 enum {
287 BRNF_PROTO_UNCHANGED,
288 BRNF_PROTO_8021Q,
289 BRNF_PROTO_PPPOE
290 } orig_proto:8;
291 u8 pkt_otherhost:1;
292 u8 in_prerouting:1;
293 u8 bridged_dnat:1;
294 __u16 frag_max_size;
295 struct net_device *physindev;
296
297 /* always valid & non-NULL from FORWARD on, for physdev match */
298 struct net_device *physoutdev;
299 union {
300 /* prerouting: detect dnat in orig/reply direction */
301 __be32 ipv4_daddr;
302 struct in6_addr ipv6_daddr;
303
304 /* after prerouting + nat detected: store original source
305 * mac since neigh resolution overwrites it, only used while
306 * skb is out in neigh layer.
307 */
308 char neigh_header[8];
309 };
310};
311#endif
312
313#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
314/* Chain in tc_skb_ext will be used to share the tc chain with
315 * ovs recirc_id. It will be set to the current chain by tc
316 * and read by ovs to recirc_id.
317 */
318struct tc_skb_ext {
319 __u32 chain;
320 __u16 mru;
321 __u16 zone;
322 u8 post_ct:1;
323 u8 post_ct_snat:1;
324 u8 post_ct_dnat:1;
325};
326#endif
327
328struct sk_buff_head {
329 /* These two members must be first to match sk_buff. */
330 struct_group_tagged(sk_buff_list, list,
331 struct sk_buff *next;
332 struct sk_buff *prev;
333 );
334
335 __u32 qlen;
336 spinlock_t lock;
337};
338
339struct sk_buff;
340
341/* To allow 64K frame to be packed as single skb without frag_list we
342 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
343 * buffers which do not start on a page boundary.
344 *
345 * Since GRO uses frags we allocate at least 16 regardless of page
346 * size.
347 */
348#if (65536/PAGE_SIZE + 1) < 16
349#define MAX_SKB_FRAGS 16UL
350#else
351#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
352#endif
353extern int sysctl_max_skb_frags;
354
355/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
356 * segment using its current segmentation instead.
357 */
358#define GSO_BY_FRAGS 0xFFFF
359
360typedef struct bio_vec skb_frag_t;
361
362/**
363 * skb_frag_size() - Returns the size of a skb fragment
364 * @frag: skb fragment
365 */
366static inline unsigned int skb_frag_size(const skb_frag_t *frag)
367{
368 return frag->bv_len;
369}
370
371/**
372 * skb_frag_size_set() - Sets the size of a skb fragment
373 * @frag: skb fragment
374 * @size: size of fragment
375 */
376static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
377{
378 frag->bv_len = size;
379}
380
381/**
382 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
383 * @frag: skb fragment
384 * @delta: value to add
385 */
386static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
387{
388 frag->bv_len += delta;
389}
390
391/**
392 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
393 * @frag: skb fragment
394 * @delta: value to subtract
395 */
396static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
397{
398 frag->bv_len -= delta;
399}
400
401/**
402 * skb_frag_must_loop - Test if %p is a high memory page
403 * @p: fragment's page
404 */
405static inline bool skb_frag_must_loop(struct page *p)
406{
407#if defined(CONFIG_HIGHMEM)
408 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
409 return true;
410#endif
411 return false;
412}
413
414/**
415 * skb_frag_foreach_page - loop over pages in a fragment
416 *
417 * @f: skb frag to operate on
418 * @f_off: offset from start of f->bv_page
419 * @f_len: length from f_off to loop over
420 * @p: (temp var) current page
421 * @p_off: (temp var) offset from start of current page,
422 * non-zero only on first page.
423 * @p_len: (temp var) length in current page,
424 * < PAGE_SIZE only on first and last page.
425 * @copied: (temp var) length so far, excluding current p_len.
426 *
427 * A fragment can hold a compound page, in which case per-page
428 * operations, notably kmap_atomic, must be called for each
429 * regular page.
430 */
431#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
432 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
433 p_off = (f_off) & (PAGE_SIZE - 1), \
434 p_len = skb_frag_must_loop(p) ? \
435 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
436 copied = 0; \
437 copied < f_len; \
438 copied += p_len, p++, p_off = 0, \
439 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
440
441#define HAVE_HW_TIME_STAMP
442
443/**
444 * struct skb_shared_hwtstamps - hardware time stamps
445 * @hwtstamp: hardware time stamp transformed into duration
446 * since arbitrary point in time
447 * @netdev_data: address/cookie of network device driver used as
448 * reference to actual hardware time stamp
449 *
450 * Software time stamps generated by ktime_get_real() are stored in
451 * skb->tstamp.
452 *
453 * hwtstamps can only be compared against other hwtstamps from
454 * the same device.
455 *
456 * This structure is attached to packets as part of the
457 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
458 */
459struct skb_shared_hwtstamps {
460 union {
461 ktime_t hwtstamp;
462 void *netdev_data;
463 };
464};
465
466/* Definitions for tx_flags in struct skb_shared_info */
467enum {
468 /* generate hardware time stamp */
469 SKBTX_HW_TSTAMP = 1 << 0,
470
471 /* generate software time stamp when queueing packet to NIC */
472 SKBTX_SW_TSTAMP = 1 << 1,
473
474 /* device driver is going to provide hardware time stamp */
475 SKBTX_IN_PROGRESS = 1 << 2,
476
477 /* generate hardware time stamp based on cycles if supported */
478 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
479
480 /* generate wifi status information (where possible) */
481 SKBTX_WIFI_STATUS = 1 << 4,
482
483 /* determine hardware time stamp based on time or cycles */
484 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
485
486 /* generate software time stamp when entering packet scheduling */
487 SKBTX_SCHED_TSTAMP = 1 << 6,
488};
489
490#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
491 SKBTX_SCHED_TSTAMP)
492#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
493 SKBTX_HW_TSTAMP_USE_CYCLES | \
494 SKBTX_ANY_SW_TSTAMP)
495
496/* Definitions for flags in struct skb_shared_info */
497enum {
498 /* use zcopy routines */
499 SKBFL_ZEROCOPY_ENABLE = BIT(0),
500
501 /* This indicates at least one fragment might be overwritten
502 * (as in vmsplice(), sendfile() ...)
503 * If we need to compute a TX checksum, we'll need to copy
504 * all frags to avoid possible bad checksum
505 */
506 SKBFL_SHARED_FRAG = BIT(1),
507
508 /* segment contains only zerocopy data and should not be
509 * charged to the kernel memory.
510 */
511 SKBFL_PURE_ZEROCOPY = BIT(2),
512
513 SKBFL_DONT_ORPHAN = BIT(3),
514
515 /* page references are managed by the ubuf_info, so it's safe to
516 * use frags only up until ubuf_info is released
517 */
518 SKBFL_MANAGED_FRAG_REFS = BIT(4),
519};
520
521#define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
522#define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
523 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
524
525/*
526 * The callback notifies userspace to release buffers when skb DMA is done in
527 * lower device, the skb last reference should be 0 when calling this.
528 * The zerocopy_success argument is true if zero copy transmit occurred,
529 * false on data copy or out of memory error caused by data copy attempt.
530 * The ctx field is used to track device context.
531 * The desc field is used to track userspace buffer index.
532 */
533struct ubuf_info {
534 void (*callback)(struct sk_buff *, struct ubuf_info *,
535 bool zerocopy_success);
536 union {
537 struct {
538 unsigned long desc;
539 void *ctx;
540 };
541 struct {
542 u32 id;
543 u16 len;
544 u16 zerocopy:1;
545 u32 bytelen;
546 };
547 };
548 refcount_t refcnt;
549 u8 flags;
550
551 struct mmpin {
552 struct user_struct *user;
553 unsigned int num_pg;
554 } mmp;
555};
556
557#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
558
559int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
560void mm_unaccount_pinned_pages(struct mmpin *mmp);
561
562/* This data is invariant across clones and lives at
563 * the end of the header data, ie. at skb->end.
564 */
565struct skb_shared_info {
566 __u8 flags;
567 __u8 meta_len;
568 __u8 nr_frags;
569 __u8 tx_flags;
570 unsigned short gso_size;
571 /* Warning: this field is not always filled in (UFO)! */
572 unsigned short gso_segs;
573 struct sk_buff *frag_list;
574 struct skb_shared_hwtstamps hwtstamps;
575 unsigned int gso_type;
576 u32 tskey;
577
578 /*
579 * Warning : all fields before dataref are cleared in __alloc_skb()
580 */
581 atomic_t dataref;
582 unsigned int xdp_frags_size;
583
584 /* Intermediate layers must ensure that destructor_arg
585 * remains valid until skb destructor */
586 void * destructor_arg;
587
588 /* must be last field, see pskb_expand_head() */
589 skb_frag_t frags[MAX_SKB_FRAGS];
590};
591
592/**
593 * DOC: dataref and headerless skbs
594 *
595 * Transport layers send out clones of payload skbs they hold for
596 * retransmissions. To allow lower layers of the stack to prepend their headers
597 * we split &skb_shared_info.dataref into two halves.
598 * The lower 16 bits count the overall number of references.
599 * The higher 16 bits indicate how many of the references are payload-only.
600 * skb_header_cloned() checks if skb is allowed to add / write the headers.
601 *
602 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
603 * (via __skb_header_release()). Any clone created from marked skb will get
604 * &sk_buff.hdr_len populated with the available headroom.
605 * If there's the only clone in existence it's able to modify the headroom
606 * at will. The sequence of calls inside the transport layer is::
607 *
608 * <alloc skb>
609 * skb_reserve()
610 * __skb_header_release()
611 * skb_clone()
612 * // send the clone down the stack
613 *
614 * This is not a very generic construct and it depends on the transport layers
615 * doing the right thing. In practice there's usually only one payload-only skb.
616 * Having multiple payload-only skbs with different lengths of hdr_len is not
617 * possible. The payload-only skbs should never leave their owner.
618 */
619#define SKB_DATAREF_SHIFT 16
620#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
621
622
623enum {
624 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
625 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
626 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
627};
628
629enum {
630 SKB_GSO_TCPV4 = 1 << 0,
631
632 /* This indicates the skb is from an untrusted source. */
633 SKB_GSO_DODGY = 1 << 1,
634
635 /* This indicates the tcp segment has CWR set. */
636 SKB_GSO_TCP_ECN = 1 << 2,
637
638 SKB_GSO_TCP_FIXEDID = 1 << 3,
639
640 SKB_GSO_TCPV6 = 1 << 4,
641
642 SKB_GSO_FCOE = 1 << 5,
643
644 SKB_GSO_GRE = 1 << 6,
645
646 SKB_GSO_GRE_CSUM = 1 << 7,
647
648 SKB_GSO_IPXIP4 = 1 << 8,
649
650 SKB_GSO_IPXIP6 = 1 << 9,
651
652 SKB_GSO_UDP_TUNNEL = 1 << 10,
653
654 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
655
656 SKB_GSO_PARTIAL = 1 << 12,
657
658 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
659
660 SKB_GSO_SCTP = 1 << 14,
661
662 SKB_GSO_ESP = 1 << 15,
663
664 SKB_GSO_UDP = 1 << 16,
665
666 SKB_GSO_UDP_L4 = 1 << 17,
667
668 SKB_GSO_FRAGLIST = 1 << 18,
669};
670
671#if BITS_PER_LONG > 32
672#define NET_SKBUFF_DATA_USES_OFFSET 1
673#endif
674
675#ifdef NET_SKBUFF_DATA_USES_OFFSET
676typedef unsigned int sk_buff_data_t;
677#else
678typedef unsigned char *sk_buff_data_t;
679#endif
680
681/**
682 * DOC: Basic sk_buff geometry
683 *
684 * struct sk_buff itself is a metadata structure and does not hold any packet
685 * data. All the data is held in associated buffers.
686 *
687 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
688 * into two parts:
689 *
690 * - data buffer, containing headers and sometimes payload;
691 * this is the part of the skb operated on by the common helpers
692 * such as skb_put() or skb_pull();
693 * - shared info (struct skb_shared_info) which holds an array of pointers
694 * to read-only data in the (page, offset, length) format.
695 *
696 * Optionally &skb_shared_info.frag_list may point to another skb.
697 *
698 * Basic diagram may look like this::
699 *
700 * ---------------
701 * | sk_buff |
702 * ---------------
703 * ,--------------------------- + head
704 * / ,----------------- + data
705 * / / ,----------- + tail
706 * | | | , + end
707 * | | | |
708 * v v v v
709 * -----------------------------------------------
710 * | headroom | data | tailroom | skb_shared_info |
711 * -----------------------------------------------
712 * + [page frag]
713 * + [page frag]
714 * + [page frag]
715 * + [page frag] ---------
716 * + frag_list --> | sk_buff |
717 * ---------
718 *
719 */
720
721/**
722 * struct sk_buff - socket buffer
723 * @next: Next buffer in list
724 * @prev: Previous buffer in list
725 * @tstamp: Time we arrived/left
726 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
727 * for retransmit timer
728 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
729 * @list: queue head
730 * @ll_node: anchor in an llist (eg socket defer_list)
731 * @sk: Socket we are owned by
732 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
733 * fragmentation management
734 * @dev: Device we arrived on/are leaving by
735 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
736 * @cb: Control buffer. Free for use by every layer. Put private vars here
737 * @_skb_refdst: destination entry (with norefcount bit)
738 * @sp: the security path, used for xfrm
739 * @len: Length of actual data
740 * @data_len: Data length
741 * @mac_len: Length of link layer header
742 * @hdr_len: writable header length of cloned skb
743 * @csum: Checksum (must include start/offset pair)
744 * @csum_start: Offset from skb->head where checksumming should start
745 * @csum_offset: Offset from csum_start where checksum should be stored
746 * @priority: Packet queueing priority
747 * @ignore_df: allow local fragmentation
748 * @cloned: Head may be cloned (check refcnt to be sure)
749 * @ip_summed: Driver fed us an IP checksum
750 * @nohdr: Payload reference only, must not modify header
751 * @pkt_type: Packet class
752 * @fclone: skbuff clone status
753 * @ipvs_property: skbuff is owned by ipvs
754 * @inner_protocol_type: whether the inner protocol is
755 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
756 * @remcsum_offload: remote checksum offload is enabled
757 * @offload_fwd_mark: Packet was L2-forwarded in hardware
758 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
759 * @tc_skip_classify: do not classify packet. set by IFB device
760 * @tc_at_ingress: used within tc_classify to distinguish in/egress
761 * @redirected: packet was redirected by packet classifier
762 * @from_ingress: packet was redirected from the ingress path
763 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
764 * @peeked: this packet has been seen already, so stats have been
765 * done for it, don't do them again
766 * @nf_trace: netfilter packet trace flag
767 * @protocol: Packet protocol from driver
768 * @destructor: Destruct function
769 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
770 * @_sk_redir: socket redirection information for skmsg
771 * @_nfct: Associated connection, if any (with nfctinfo bits)
772 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
773 * @skb_iif: ifindex of device we arrived on
774 * @tc_index: Traffic control index
775 * @hash: the packet hash
776 * @queue_mapping: Queue mapping for multiqueue devices
777 * @head_frag: skb was allocated from page fragments,
778 * not allocated by kmalloc() or vmalloc().
779 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
780 * @pp_recycle: mark the packet for recycling instead of freeing (implies
781 * page_pool support on driver)
782 * @active_extensions: active extensions (skb_ext_id types)
783 * @ndisc_nodetype: router type (from link layer)
784 * @ooo_okay: allow the mapping of a socket to a queue to be changed
785 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
786 * ports.
787 * @sw_hash: indicates hash was computed in software stack
788 * @wifi_acked_valid: wifi_acked was set
789 * @wifi_acked: whether frame was acked on wifi or not
790 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
791 * @encapsulation: indicates the inner headers in the skbuff are valid
792 * @encap_hdr_csum: software checksum is needed
793 * @csum_valid: checksum is already valid
794 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
795 * @csum_complete_sw: checksum was completed by software
796 * @csum_level: indicates the number of consecutive checksums found in
797 * the packet minus one that have been verified as
798 * CHECKSUM_UNNECESSARY (max 3)
799 * @dst_pending_confirm: need to confirm neighbour
800 * @decrypted: Decrypted SKB
801 * @slow_gro: state present at GRO time, slower prepare step required
802 * @mono_delivery_time: When set, skb->tstamp has the
803 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
804 * skb->tstamp has the (rcv) timestamp at ingress and
805 * delivery_time at egress.
806 * @napi_id: id of the NAPI struct this skb came from
807 * @sender_cpu: (aka @napi_id) source CPU in XPS
808 * @alloc_cpu: CPU which did the skb allocation.
809 * @secmark: security marking
810 * @mark: Generic packet mark
811 * @reserved_tailroom: (aka @mark) number of bytes of free space available
812 * at the tail of an sk_buff
813 * @vlan_present: VLAN tag is present
814 * @vlan_proto: vlan encapsulation protocol
815 * @vlan_tci: vlan tag control information
816 * @inner_protocol: Protocol (encapsulation)
817 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
818 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
819 * @inner_transport_header: Inner transport layer header (encapsulation)
820 * @inner_network_header: Network layer header (encapsulation)
821 * @inner_mac_header: Link layer header (encapsulation)
822 * @transport_header: Transport layer header
823 * @network_header: Network layer header
824 * @mac_header: Link layer header
825 * @kcov_handle: KCOV remote handle for remote coverage collection
826 * @tail: Tail pointer
827 * @end: End pointer
828 * @head: Head of buffer
829 * @data: Data head pointer
830 * @truesize: Buffer size
831 * @users: User count - see {datagram,tcp}.c
832 * @extensions: allocated extensions, valid if active_extensions is nonzero
833 */
834
835struct sk_buff {
836 union {
837 struct {
838 /* These two members must be first to match sk_buff_head. */
839 struct sk_buff *next;
840 struct sk_buff *prev;
841
842 union {
843 struct net_device *dev;
844 /* Some protocols might use this space to store information,
845 * while device pointer would be NULL.
846 * UDP receive path is one user.
847 */
848 unsigned long dev_scratch;
849 };
850 };
851 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
852 struct list_head list;
853 struct llist_node ll_node;
854 };
855
856 union {
857 struct sock *sk;
858 int ip_defrag_offset;
859 };
860
861 union {
862 ktime_t tstamp;
863 u64 skb_mstamp_ns; /* earliest departure time */
864 };
865 /*
866 * This is the control buffer. It is free to use for every
867 * layer. Please put your private variables there. If you
868 * want to keep them across layers you have to do a skb_clone()
869 * first. This is owned by whoever has the skb queued ATM.
870 */
871 char cb[48] __aligned(8);
872
873 union {
874 struct {
875 unsigned long _skb_refdst;
876 void (*destructor)(struct sk_buff *skb);
877 };
878 struct list_head tcp_tsorted_anchor;
879#ifdef CONFIG_NET_SOCK_MSG
880 unsigned long _sk_redir;
881#endif
882 };
883
884#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
885 unsigned long _nfct;
886#endif
887 unsigned int len,
888 data_len;
889 __u16 mac_len,
890 hdr_len;
891
892 /* Following fields are _not_ copied in __copy_skb_header()
893 * Note that queue_mapping is here mostly to fill a hole.
894 */
895 __u16 queue_mapping;
896
897/* if you move cloned around you also must adapt those constants */
898#ifdef __BIG_ENDIAN_BITFIELD
899#define CLONED_MASK (1 << 7)
900#else
901#define CLONED_MASK 1
902#endif
903#define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
904
905 /* private: */
906 __u8 __cloned_offset[0];
907 /* public: */
908 __u8 cloned:1,
909 nohdr:1,
910 fclone:2,
911 peeked:1,
912 head_frag:1,
913 pfmemalloc:1,
914 pp_recycle:1; /* page_pool recycle indicator */
915#ifdef CONFIG_SKB_EXTENSIONS
916 __u8 active_extensions;
917#endif
918
919 /* Fields enclosed in headers group are copied
920 * using a single memcpy() in __copy_skb_header()
921 */
922 struct_group(headers,
923
924 /* private: */
925 __u8 __pkt_type_offset[0];
926 /* public: */
927 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
928 __u8 ignore_df:1;
929 __u8 nf_trace:1;
930 __u8 ip_summed:2;
931 __u8 ooo_okay:1;
932
933 __u8 l4_hash:1;
934 __u8 sw_hash:1;
935 __u8 wifi_acked_valid:1;
936 __u8 wifi_acked:1;
937 __u8 no_fcs:1;
938 /* Indicates the inner headers are valid in the skbuff. */
939 __u8 encapsulation:1;
940 __u8 encap_hdr_csum:1;
941 __u8 csum_valid:1;
942
943 /* private: */
944 __u8 __pkt_vlan_present_offset[0];
945 /* public: */
946 __u8 vlan_present:1; /* See PKT_VLAN_PRESENT_BIT */
947 __u8 csum_complete_sw:1;
948 __u8 csum_level:2;
949 __u8 dst_pending_confirm:1;
950 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
951#ifdef CONFIG_NET_CLS_ACT
952 __u8 tc_skip_classify:1;
953 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
954#endif
955#ifdef CONFIG_IPV6_NDISC_NODETYPE
956 __u8 ndisc_nodetype:2;
957#endif
958
959 __u8 ipvs_property:1;
960 __u8 inner_protocol_type:1;
961 __u8 remcsum_offload:1;
962#ifdef CONFIG_NET_SWITCHDEV
963 __u8 offload_fwd_mark:1;
964 __u8 offload_l3_fwd_mark:1;
965#endif
966 __u8 redirected:1;
967#ifdef CONFIG_NET_REDIRECT
968 __u8 from_ingress:1;
969#endif
970#ifdef CONFIG_NETFILTER_SKIP_EGRESS
971 __u8 nf_skip_egress:1;
972#endif
973#ifdef CONFIG_TLS_DEVICE
974 __u8 decrypted:1;
975#endif
976 __u8 slow_gro:1;
977 __u8 csum_not_inet:1;
978
979#ifdef CONFIG_NET_SCHED
980 __u16 tc_index; /* traffic control index */
981#endif
982
983 union {
984 __wsum csum;
985 struct {
986 __u16 csum_start;
987 __u16 csum_offset;
988 };
989 };
990 __u32 priority;
991 int skb_iif;
992 __u32 hash;
993 __be16 vlan_proto;
994 __u16 vlan_tci;
995#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
996 union {
997 unsigned int napi_id;
998 unsigned int sender_cpu;
999 };
1000#endif
1001 u16 alloc_cpu;
1002#ifdef CONFIG_NETWORK_SECMARK
1003 __u32 secmark;
1004#endif
1005
1006 union {
1007 __u32 mark;
1008 __u32 reserved_tailroom;
1009 };
1010
1011 union {
1012 __be16 inner_protocol;
1013 __u8 inner_ipproto;
1014 };
1015
1016 __u16 inner_transport_header;
1017 __u16 inner_network_header;
1018 __u16 inner_mac_header;
1019
1020 __be16 protocol;
1021 __u16 transport_header;
1022 __u16 network_header;
1023 __u16 mac_header;
1024
1025#ifdef CONFIG_KCOV
1026 u64 kcov_handle;
1027#endif
1028
1029 ); /* end headers group */
1030
1031 /* These elements must be at the end, see alloc_skb() for details. */
1032 sk_buff_data_t tail;
1033 sk_buff_data_t end;
1034 unsigned char *head,
1035 *data;
1036 unsigned int truesize;
1037 refcount_t users;
1038
1039#ifdef CONFIG_SKB_EXTENSIONS
1040 /* only useable after checking ->active_extensions != 0 */
1041 struct skb_ext *extensions;
1042#endif
1043};
1044
1045/* if you move pkt_type around you also must adapt those constants */
1046#ifdef __BIG_ENDIAN_BITFIELD
1047#define PKT_TYPE_MAX (7 << 5)
1048#else
1049#define PKT_TYPE_MAX 7
1050#endif
1051#define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1052
1053/* if you move pkt_vlan_present, tc_at_ingress, or mono_delivery_time
1054 * around, you also must adapt these constants.
1055 */
1056#ifdef __BIG_ENDIAN_BITFIELD
1057#define PKT_VLAN_PRESENT_BIT 7
1058#define TC_AT_INGRESS_MASK (1 << 0)
1059#define SKB_MONO_DELIVERY_TIME_MASK (1 << 2)
1060#else
1061#define PKT_VLAN_PRESENT_BIT 0
1062#define TC_AT_INGRESS_MASK (1 << 7)
1063#define SKB_MONO_DELIVERY_TIME_MASK (1 << 5)
1064#endif
1065#define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset)
1066
1067#ifdef __KERNEL__
1068/*
1069 * Handling routines are only of interest to the kernel
1070 */
1071
1072#define SKB_ALLOC_FCLONE 0x01
1073#define SKB_ALLOC_RX 0x02
1074#define SKB_ALLOC_NAPI 0x04
1075
1076/**
1077 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1078 * @skb: buffer
1079 */
1080static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1081{
1082 return unlikely(skb->pfmemalloc);
1083}
1084
1085/*
1086 * skb might have a dst pointer attached, refcounted or not.
1087 * _skb_refdst low order bit is set if refcount was _not_ taken
1088 */
1089#define SKB_DST_NOREF 1UL
1090#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1091
1092/**
1093 * skb_dst - returns skb dst_entry
1094 * @skb: buffer
1095 *
1096 * Returns skb dst_entry, regardless of reference taken or not.
1097 */
1098static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1099{
1100 /* If refdst was not refcounted, check we still are in a
1101 * rcu_read_lock section
1102 */
1103 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1104 !rcu_read_lock_held() &&
1105 !rcu_read_lock_bh_held());
1106 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1107}
1108
1109/**
1110 * skb_dst_set - sets skb dst
1111 * @skb: buffer
1112 * @dst: dst entry
1113 *
1114 * Sets skb dst, assuming a reference was taken on dst and should
1115 * be released by skb_dst_drop()
1116 */
1117static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1118{
1119 skb->slow_gro |= !!dst;
1120 skb->_skb_refdst = (unsigned long)dst;
1121}
1122
1123/**
1124 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1125 * @skb: buffer
1126 * @dst: dst entry
1127 *
1128 * Sets skb dst, assuming a reference was not taken on dst.
1129 * If dst entry is cached, we do not take reference and dst_release
1130 * will be avoided by refdst_drop. If dst entry is not cached, we take
1131 * reference, so that last dst_release can destroy the dst immediately.
1132 */
1133static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1134{
1135 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1136 skb->slow_gro |= !!dst;
1137 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1138}
1139
1140/**
1141 * skb_dst_is_noref - Test if skb dst isn't refcounted
1142 * @skb: buffer
1143 */
1144static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1145{
1146 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1147}
1148
1149/**
1150 * skb_rtable - Returns the skb &rtable
1151 * @skb: buffer
1152 */
1153static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1154{
1155 return (struct rtable *)skb_dst(skb);
1156}
1157
1158/* For mangling skb->pkt_type from user space side from applications
1159 * such as nft, tc, etc, we only allow a conservative subset of
1160 * possible pkt_types to be set.
1161*/
1162static inline bool skb_pkt_type_ok(u32 ptype)
1163{
1164 return ptype <= PACKET_OTHERHOST;
1165}
1166
1167/**
1168 * skb_napi_id - Returns the skb's NAPI id
1169 * @skb: buffer
1170 */
1171static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1172{
1173#ifdef CONFIG_NET_RX_BUSY_POLL
1174 return skb->napi_id;
1175#else
1176 return 0;
1177#endif
1178}
1179
1180/**
1181 * skb_unref - decrement the skb's reference count
1182 * @skb: buffer
1183 *
1184 * Returns true if we can free the skb.
1185 */
1186static inline bool skb_unref(struct sk_buff *skb)
1187{
1188 if (unlikely(!skb))
1189 return false;
1190 if (likely(refcount_read(&skb->users) == 1))
1191 smp_rmb();
1192 else if (likely(!refcount_dec_and_test(&skb->users)))
1193 return false;
1194
1195 return true;
1196}
1197
1198void kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1199
1200/**
1201 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1202 * @skb: buffer to free
1203 */
1204static inline void kfree_skb(struct sk_buff *skb)
1205{
1206 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1207}
1208
1209void skb_release_head_state(struct sk_buff *skb);
1210void kfree_skb_list_reason(struct sk_buff *segs,
1211 enum skb_drop_reason reason);
1212void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1213void skb_tx_error(struct sk_buff *skb);
1214
1215static inline void kfree_skb_list(struct sk_buff *segs)
1216{
1217 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1218}
1219
1220#ifdef CONFIG_TRACEPOINTS
1221void consume_skb(struct sk_buff *skb);
1222#else
1223static inline void consume_skb(struct sk_buff *skb)
1224{
1225 return kfree_skb(skb);
1226}
1227#endif
1228
1229void __consume_stateless_skb(struct sk_buff *skb);
1230void __kfree_skb(struct sk_buff *skb);
1231extern struct kmem_cache *skbuff_head_cache;
1232
1233void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1234bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1235 bool *fragstolen, int *delta_truesize);
1236
1237struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1238 int node);
1239struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1240struct sk_buff *build_skb(void *data, unsigned int frag_size);
1241struct sk_buff *build_skb_around(struct sk_buff *skb,
1242 void *data, unsigned int frag_size);
1243void skb_attempt_defer_free(struct sk_buff *skb);
1244
1245struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1246
1247/**
1248 * alloc_skb - allocate a network buffer
1249 * @size: size to allocate
1250 * @priority: allocation mask
1251 *
1252 * This function is a convenient wrapper around __alloc_skb().
1253 */
1254static inline struct sk_buff *alloc_skb(unsigned int size,
1255 gfp_t priority)
1256{
1257 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1258}
1259
1260struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1261 unsigned long data_len,
1262 int max_page_order,
1263 int *errcode,
1264 gfp_t gfp_mask);
1265struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1266
1267/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1268struct sk_buff_fclones {
1269 struct sk_buff skb1;
1270
1271 struct sk_buff skb2;
1272
1273 refcount_t fclone_ref;
1274};
1275
1276/**
1277 * skb_fclone_busy - check if fclone is busy
1278 * @sk: socket
1279 * @skb: buffer
1280 *
1281 * Returns true if skb is a fast clone, and its clone is not freed.
1282 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1283 * so we also check that this didnt happen.
1284 */
1285static inline bool skb_fclone_busy(const struct sock *sk,
1286 const struct sk_buff *skb)
1287{
1288 const struct sk_buff_fclones *fclones;
1289
1290 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1291
1292 return skb->fclone == SKB_FCLONE_ORIG &&
1293 refcount_read(&fclones->fclone_ref) > 1 &&
1294 READ_ONCE(fclones->skb2.sk) == sk;
1295}
1296
1297/**
1298 * alloc_skb_fclone - allocate a network buffer from fclone cache
1299 * @size: size to allocate
1300 * @priority: allocation mask
1301 *
1302 * This function is a convenient wrapper around __alloc_skb().
1303 */
1304static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1305 gfp_t priority)
1306{
1307 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1308}
1309
1310struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1311void skb_headers_offset_update(struct sk_buff *skb, int off);
1312int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1313struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1314void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1315struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1316struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1317 gfp_t gfp_mask, bool fclone);
1318static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1319 gfp_t gfp_mask)
1320{
1321 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1322}
1323
1324int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1325struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1326 unsigned int headroom);
1327struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1328struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1329 int newtailroom, gfp_t priority);
1330int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1331 int offset, int len);
1332int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1333 int offset, int len);
1334int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1335int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1336
1337/**
1338 * skb_pad - zero pad the tail of an skb
1339 * @skb: buffer to pad
1340 * @pad: space to pad
1341 *
1342 * Ensure that a buffer is followed by a padding area that is zero
1343 * filled. Used by network drivers which may DMA or transfer data
1344 * beyond the buffer end onto the wire.
1345 *
1346 * May return error in out of memory cases. The skb is freed on error.
1347 */
1348static inline int skb_pad(struct sk_buff *skb, int pad)
1349{
1350 return __skb_pad(skb, pad, true);
1351}
1352#define dev_kfree_skb(a) consume_skb(a)
1353
1354int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1355 int offset, size_t size);
1356
1357struct skb_seq_state {
1358 __u32 lower_offset;
1359 __u32 upper_offset;
1360 __u32 frag_idx;
1361 __u32 stepped_offset;
1362 struct sk_buff *root_skb;
1363 struct sk_buff *cur_skb;
1364 __u8 *frag_data;
1365 __u32 frag_off;
1366};
1367
1368void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1369 unsigned int to, struct skb_seq_state *st);
1370unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1371 struct skb_seq_state *st);
1372void skb_abort_seq_read(struct skb_seq_state *st);
1373
1374unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1375 unsigned int to, struct ts_config *config);
1376
1377/*
1378 * Packet hash types specify the type of hash in skb_set_hash.
1379 *
1380 * Hash types refer to the protocol layer addresses which are used to
1381 * construct a packet's hash. The hashes are used to differentiate or identify
1382 * flows of the protocol layer for the hash type. Hash types are either
1383 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1384 *
1385 * Properties of hashes:
1386 *
1387 * 1) Two packets in different flows have different hash values
1388 * 2) Two packets in the same flow should have the same hash value
1389 *
1390 * A hash at a higher layer is considered to be more specific. A driver should
1391 * set the most specific hash possible.
1392 *
1393 * A driver cannot indicate a more specific hash than the layer at which a hash
1394 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1395 *
1396 * A driver may indicate a hash level which is less specific than the
1397 * actual layer the hash was computed on. For instance, a hash computed
1398 * at L4 may be considered an L3 hash. This should only be done if the
1399 * driver can't unambiguously determine that the HW computed the hash at
1400 * the higher layer. Note that the "should" in the second property above
1401 * permits this.
1402 */
1403enum pkt_hash_types {
1404 PKT_HASH_TYPE_NONE, /* Undefined type */
1405 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1406 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1407 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1408};
1409
1410static inline void skb_clear_hash(struct sk_buff *skb)
1411{
1412 skb->hash = 0;
1413 skb->sw_hash = 0;
1414 skb->l4_hash = 0;
1415}
1416
1417static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1418{
1419 if (!skb->l4_hash)
1420 skb_clear_hash(skb);
1421}
1422
1423static inline void
1424__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1425{
1426 skb->l4_hash = is_l4;
1427 skb->sw_hash = is_sw;
1428 skb->hash = hash;
1429}
1430
1431static inline void
1432skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1433{
1434 /* Used by drivers to set hash from HW */
1435 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1436}
1437
1438static inline void
1439__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1440{
1441 __skb_set_hash(skb, hash, true, is_l4);
1442}
1443
1444void __skb_get_hash(struct sk_buff *skb);
1445u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1446u32 skb_get_poff(const struct sk_buff *skb);
1447u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1448 const struct flow_keys_basic *keys, int hlen);
1449__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1450 const void *data, int hlen_proto);
1451
1452static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1453 int thoff, u8 ip_proto)
1454{
1455 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1456}
1457
1458void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1459 const struct flow_dissector_key *key,
1460 unsigned int key_count);
1461
1462struct bpf_flow_dissector;
1463bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1464 __be16 proto, int nhoff, int hlen, unsigned int flags);
1465
1466bool __skb_flow_dissect(const struct net *net,
1467 const struct sk_buff *skb,
1468 struct flow_dissector *flow_dissector,
1469 void *target_container, const void *data,
1470 __be16 proto, int nhoff, int hlen, unsigned int flags);
1471
1472static inline bool skb_flow_dissect(const struct sk_buff *skb,
1473 struct flow_dissector *flow_dissector,
1474 void *target_container, unsigned int flags)
1475{
1476 return __skb_flow_dissect(NULL, skb, flow_dissector,
1477 target_container, NULL, 0, 0, 0, flags);
1478}
1479
1480static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1481 struct flow_keys *flow,
1482 unsigned int flags)
1483{
1484 memset(flow, 0, sizeof(*flow));
1485 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1486 flow, NULL, 0, 0, 0, flags);
1487}
1488
1489static inline bool
1490skb_flow_dissect_flow_keys_basic(const struct net *net,
1491 const struct sk_buff *skb,
1492 struct flow_keys_basic *flow,
1493 const void *data, __be16 proto,
1494 int nhoff, int hlen, unsigned int flags)
1495{
1496 memset(flow, 0, sizeof(*flow));
1497 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1498 data, proto, nhoff, hlen, flags);
1499}
1500
1501void skb_flow_dissect_meta(const struct sk_buff *skb,
1502 struct flow_dissector *flow_dissector,
1503 void *target_container);
1504
1505/* Gets a skb connection tracking info, ctinfo map should be a
1506 * map of mapsize to translate enum ip_conntrack_info states
1507 * to user states.
1508 */
1509void
1510skb_flow_dissect_ct(const struct sk_buff *skb,
1511 struct flow_dissector *flow_dissector,
1512 void *target_container,
1513 u16 *ctinfo_map, size_t mapsize,
1514 bool post_ct, u16 zone);
1515void
1516skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1517 struct flow_dissector *flow_dissector,
1518 void *target_container);
1519
1520void skb_flow_dissect_hash(const struct sk_buff *skb,
1521 struct flow_dissector *flow_dissector,
1522 void *target_container);
1523
1524static inline __u32 skb_get_hash(struct sk_buff *skb)
1525{
1526 if (!skb->l4_hash && !skb->sw_hash)
1527 __skb_get_hash(skb);
1528
1529 return skb->hash;
1530}
1531
1532static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1533{
1534 if (!skb->l4_hash && !skb->sw_hash) {
1535 struct flow_keys keys;
1536 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1537
1538 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1539 }
1540
1541 return skb->hash;
1542}
1543
1544__u32 skb_get_hash_perturb(const struct sk_buff *skb,
1545 const siphash_key_t *perturb);
1546
1547static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1548{
1549 return skb->hash;
1550}
1551
1552static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1553{
1554 to->hash = from->hash;
1555 to->sw_hash = from->sw_hash;
1556 to->l4_hash = from->l4_hash;
1557};
1558
1559static inline void skb_copy_decrypted(struct sk_buff *to,
1560 const struct sk_buff *from)
1561{
1562#ifdef CONFIG_TLS_DEVICE
1563 to->decrypted = from->decrypted;
1564#endif
1565}
1566
1567#ifdef NET_SKBUFF_DATA_USES_OFFSET
1568static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1569{
1570 return skb->head + skb->end;
1571}
1572
1573static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1574{
1575 return skb->end;
1576}
1577
1578static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1579{
1580 skb->end = offset;
1581}
1582#else
1583static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1584{
1585 return skb->end;
1586}
1587
1588static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1589{
1590 return skb->end - skb->head;
1591}
1592
1593static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1594{
1595 skb->end = skb->head + offset;
1596}
1597#endif
1598
1599struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1600 struct ubuf_info *uarg);
1601
1602void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1603
1604void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1605 bool success);
1606
1607int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1608 struct sk_buff *skb, struct iov_iter *from,
1609 size_t length);
1610
1611static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1612 struct msghdr *msg, int len)
1613{
1614 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1615}
1616
1617int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1618 struct msghdr *msg, int len,
1619 struct ubuf_info *uarg);
1620
1621/* Internal */
1622#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1623
1624static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1625{
1626 return &skb_shinfo(skb)->hwtstamps;
1627}
1628
1629static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1630{
1631 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1632
1633 return is_zcopy ? skb_uarg(skb) : NULL;
1634}
1635
1636static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1637{
1638 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1639}
1640
1641static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1642{
1643 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1644}
1645
1646static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1647 const struct sk_buff *skb2)
1648{
1649 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1650}
1651
1652static inline void net_zcopy_get(struct ubuf_info *uarg)
1653{
1654 refcount_inc(&uarg->refcnt);
1655}
1656
1657static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1658{
1659 skb_shinfo(skb)->destructor_arg = uarg;
1660 skb_shinfo(skb)->flags |= uarg->flags;
1661}
1662
1663static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1664 bool *have_ref)
1665{
1666 if (skb && uarg && !skb_zcopy(skb)) {
1667 if (unlikely(have_ref && *have_ref))
1668 *have_ref = false;
1669 else
1670 net_zcopy_get(uarg);
1671 skb_zcopy_init(skb, uarg);
1672 }
1673}
1674
1675static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1676{
1677 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1678 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1679}
1680
1681static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1682{
1683 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1684}
1685
1686static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1687{
1688 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1689}
1690
1691static inline void net_zcopy_put(struct ubuf_info *uarg)
1692{
1693 if (uarg)
1694 uarg->callback(NULL, uarg, true);
1695}
1696
1697static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1698{
1699 if (uarg) {
1700 if (uarg->callback == msg_zerocopy_callback)
1701 msg_zerocopy_put_abort(uarg, have_uref);
1702 else if (have_uref)
1703 net_zcopy_put(uarg);
1704 }
1705}
1706
1707/* Release a reference on a zerocopy structure */
1708static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1709{
1710 struct ubuf_info *uarg = skb_zcopy(skb);
1711
1712 if (uarg) {
1713 if (!skb_zcopy_is_nouarg(skb))
1714 uarg->callback(skb, uarg, zerocopy_success);
1715
1716 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1717 }
1718}
1719
1720void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1721
1722static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1723{
1724 if (unlikely(skb_zcopy_managed(skb)))
1725 __skb_zcopy_downgrade_managed(skb);
1726}
1727
1728static inline void skb_mark_not_on_list(struct sk_buff *skb)
1729{
1730 skb->next = NULL;
1731}
1732
1733/* Iterate through singly-linked GSO fragments of an skb. */
1734#define skb_list_walk_safe(first, skb, next_skb) \
1735 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1736 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1737
1738static inline void skb_list_del_init(struct sk_buff *skb)
1739{
1740 __list_del_entry(&skb->list);
1741 skb_mark_not_on_list(skb);
1742}
1743
1744/**
1745 * skb_queue_empty - check if a queue is empty
1746 * @list: queue head
1747 *
1748 * Returns true if the queue is empty, false otherwise.
1749 */
1750static inline int skb_queue_empty(const struct sk_buff_head *list)
1751{
1752 return list->next == (const struct sk_buff *) list;
1753}
1754
1755/**
1756 * skb_queue_empty_lockless - check if a queue is empty
1757 * @list: queue head
1758 *
1759 * Returns true if the queue is empty, false otherwise.
1760 * This variant can be used in lockless contexts.
1761 */
1762static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1763{
1764 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1765}
1766
1767
1768/**
1769 * skb_queue_is_last - check if skb is the last entry in the queue
1770 * @list: queue head
1771 * @skb: buffer
1772 *
1773 * Returns true if @skb is the last buffer on the list.
1774 */
1775static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1776 const struct sk_buff *skb)
1777{
1778 return skb->next == (const struct sk_buff *) list;
1779}
1780
1781/**
1782 * skb_queue_is_first - check if skb is the first entry in the queue
1783 * @list: queue head
1784 * @skb: buffer
1785 *
1786 * Returns true if @skb is the first buffer on the list.
1787 */
1788static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1789 const struct sk_buff *skb)
1790{
1791 return skb->prev == (const struct sk_buff *) list;
1792}
1793
1794/**
1795 * skb_queue_next - return the next packet in the queue
1796 * @list: queue head
1797 * @skb: current buffer
1798 *
1799 * Return the next packet in @list after @skb. It is only valid to
1800 * call this if skb_queue_is_last() evaluates to false.
1801 */
1802static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1803 const struct sk_buff *skb)
1804{
1805 /* This BUG_ON may seem severe, but if we just return then we
1806 * are going to dereference garbage.
1807 */
1808 BUG_ON(skb_queue_is_last(list, skb));
1809 return skb->next;
1810}
1811
1812/**
1813 * skb_queue_prev - return the prev packet in the queue
1814 * @list: queue head
1815 * @skb: current buffer
1816 *
1817 * Return the prev packet in @list before @skb. It is only valid to
1818 * call this if skb_queue_is_first() evaluates to false.
1819 */
1820static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1821 const struct sk_buff *skb)
1822{
1823 /* This BUG_ON may seem severe, but if we just return then we
1824 * are going to dereference garbage.
1825 */
1826 BUG_ON(skb_queue_is_first(list, skb));
1827 return skb->prev;
1828}
1829
1830/**
1831 * skb_get - reference buffer
1832 * @skb: buffer to reference
1833 *
1834 * Makes another reference to a socket buffer and returns a pointer
1835 * to the buffer.
1836 */
1837static inline struct sk_buff *skb_get(struct sk_buff *skb)
1838{
1839 refcount_inc(&skb->users);
1840 return skb;
1841}
1842
1843/*
1844 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1845 */
1846
1847/**
1848 * skb_cloned - is the buffer a clone
1849 * @skb: buffer to check
1850 *
1851 * Returns true if the buffer was generated with skb_clone() and is
1852 * one of multiple shared copies of the buffer. Cloned buffers are
1853 * shared data so must not be written to under normal circumstances.
1854 */
1855static inline int skb_cloned(const struct sk_buff *skb)
1856{
1857 return skb->cloned &&
1858 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1859}
1860
1861static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1862{
1863 might_sleep_if(gfpflags_allow_blocking(pri));
1864
1865 if (skb_cloned(skb))
1866 return pskb_expand_head(skb, 0, 0, pri);
1867
1868 return 0;
1869}
1870
1871/* This variant of skb_unclone() makes sure skb->truesize
1872 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1873 *
1874 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1875 * when various debugging features are in place.
1876 */
1877int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1878static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1879{
1880 might_sleep_if(gfpflags_allow_blocking(pri));
1881
1882 if (skb_cloned(skb))
1883 return __skb_unclone_keeptruesize(skb, pri);
1884 return 0;
1885}
1886
1887/**
1888 * skb_header_cloned - is the header a clone
1889 * @skb: buffer to check
1890 *
1891 * Returns true if modifying the header part of the buffer requires
1892 * the data to be copied.
1893 */
1894static inline int skb_header_cloned(const struct sk_buff *skb)
1895{
1896 int dataref;
1897
1898 if (!skb->cloned)
1899 return 0;
1900
1901 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1902 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1903 return dataref != 1;
1904}
1905
1906static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1907{
1908 might_sleep_if(gfpflags_allow_blocking(pri));
1909
1910 if (skb_header_cloned(skb))
1911 return pskb_expand_head(skb, 0, 0, pri);
1912
1913 return 0;
1914}
1915
1916/**
1917 * __skb_header_release() - allow clones to use the headroom
1918 * @skb: buffer to operate on
1919 *
1920 * See "DOC: dataref and headerless skbs".
1921 */
1922static inline void __skb_header_release(struct sk_buff *skb)
1923{
1924 skb->nohdr = 1;
1925 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1926}
1927
1928
1929/**
1930 * skb_shared - is the buffer shared
1931 * @skb: buffer to check
1932 *
1933 * Returns true if more than one person has a reference to this
1934 * buffer.
1935 */
1936static inline int skb_shared(const struct sk_buff *skb)
1937{
1938 return refcount_read(&skb->users) != 1;
1939}
1940
1941/**
1942 * skb_share_check - check if buffer is shared and if so clone it
1943 * @skb: buffer to check
1944 * @pri: priority for memory allocation
1945 *
1946 * If the buffer is shared the buffer is cloned and the old copy
1947 * drops a reference. A new clone with a single reference is returned.
1948 * If the buffer is not shared the original buffer is returned. When
1949 * being called from interrupt status or with spinlocks held pri must
1950 * be GFP_ATOMIC.
1951 *
1952 * NULL is returned on a memory allocation failure.
1953 */
1954static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1955{
1956 might_sleep_if(gfpflags_allow_blocking(pri));
1957 if (skb_shared(skb)) {
1958 struct sk_buff *nskb = skb_clone(skb, pri);
1959
1960 if (likely(nskb))
1961 consume_skb(skb);
1962 else
1963 kfree_skb(skb);
1964 skb = nskb;
1965 }
1966 return skb;
1967}
1968
1969/*
1970 * Copy shared buffers into a new sk_buff. We effectively do COW on
1971 * packets to handle cases where we have a local reader and forward
1972 * and a couple of other messy ones. The normal one is tcpdumping
1973 * a packet thats being forwarded.
1974 */
1975
1976/**
1977 * skb_unshare - make a copy of a shared buffer
1978 * @skb: buffer to check
1979 * @pri: priority for memory allocation
1980 *
1981 * If the socket buffer is a clone then this function creates a new
1982 * copy of the data, drops a reference count on the old copy and returns
1983 * the new copy with the reference count at 1. If the buffer is not a clone
1984 * the original buffer is returned. When called with a spinlock held or
1985 * from interrupt state @pri must be %GFP_ATOMIC
1986 *
1987 * %NULL is returned on a memory allocation failure.
1988 */
1989static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1990 gfp_t pri)
1991{
1992 might_sleep_if(gfpflags_allow_blocking(pri));
1993 if (skb_cloned(skb)) {
1994 struct sk_buff *nskb = skb_copy(skb, pri);
1995
1996 /* Free our shared copy */
1997 if (likely(nskb))
1998 consume_skb(skb);
1999 else
2000 kfree_skb(skb);
2001 skb = nskb;
2002 }
2003 return skb;
2004}
2005
2006/**
2007 * skb_peek - peek at the head of an &sk_buff_head
2008 * @list_: list to peek at
2009 *
2010 * Peek an &sk_buff. Unlike most other operations you _MUST_
2011 * be careful with this one. A peek leaves the buffer on the
2012 * list and someone else may run off with it. You must hold
2013 * the appropriate locks or have a private queue to do this.
2014 *
2015 * Returns %NULL for an empty list or a pointer to the head element.
2016 * The reference count is not incremented and the reference is therefore
2017 * volatile. Use with caution.
2018 */
2019static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2020{
2021 struct sk_buff *skb = list_->next;
2022
2023 if (skb == (struct sk_buff *)list_)
2024 skb = NULL;
2025 return skb;
2026}
2027
2028/**
2029 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2030 * @list_: list to peek at
2031 *
2032 * Like skb_peek(), but the caller knows that the list is not empty.
2033 */
2034static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2035{
2036 return list_->next;
2037}
2038
2039/**
2040 * skb_peek_next - peek skb following the given one from a queue
2041 * @skb: skb to start from
2042 * @list_: list to peek at
2043 *
2044 * Returns %NULL when the end of the list is met or a pointer to the
2045 * next element. The reference count is not incremented and the
2046 * reference is therefore volatile. Use with caution.
2047 */
2048static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2049 const struct sk_buff_head *list_)
2050{
2051 struct sk_buff *next = skb->next;
2052
2053 if (next == (struct sk_buff *)list_)
2054 next = NULL;
2055 return next;
2056}
2057
2058/**
2059 * skb_peek_tail - peek at the tail of an &sk_buff_head
2060 * @list_: list to peek at
2061 *
2062 * Peek an &sk_buff. Unlike most other operations you _MUST_
2063 * be careful with this one. A peek leaves the buffer on the
2064 * list and someone else may run off with it. You must hold
2065 * the appropriate locks or have a private queue to do this.
2066 *
2067 * Returns %NULL for an empty list or a pointer to the tail element.
2068 * The reference count is not incremented and the reference is therefore
2069 * volatile. Use with caution.
2070 */
2071static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2072{
2073 struct sk_buff *skb = READ_ONCE(list_->prev);
2074
2075 if (skb == (struct sk_buff *)list_)
2076 skb = NULL;
2077 return skb;
2078
2079}
2080
2081/**
2082 * skb_queue_len - get queue length
2083 * @list_: list to measure
2084 *
2085 * Return the length of an &sk_buff queue.
2086 */
2087static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2088{
2089 return list_->qlen;
2090}
2091
2092/**
2093 * skb_queue_len_lockless - get queue length
2094 * @list_: list to measure
2095 *
2096 * Return the length of an &sk_buff queue.
2097 * This variant can be used in lockless contexts.
2098 */
2099static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2100{
2101 return READ_ONCE(list_->qlen);
2102}
2103
2104/**
2105 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2106 * @list: queue to initialize
2107 *
2108 * This initializes only the list and queue length aspects of
2109 * an sk_buff_head object. This allows to initialize the list
2110 * aspects of an sk_buff_head without reinitializing things like
2111 * the spinlock. It can also be used for on-stack sk_buff_head
2112 * objects where the spinlock is known to not be used.
2113 */
2114static inline void __skb_queue_head_init(struct sk_buff_head *list)
2115{
2116 list->prev = list->next = (struct sk_buff *)list;
2117 list->qlen = 0;
2118}
2119
2120/*
2121 * This function creates a split out lock class for each invocation;
2122 * this is needed for now since a whole lot of users of the skb-queue
2123 * infrastructure in drivers have different locking usage (in hardirq)
2124 * than the networking core (in softirq only). In the long run either the
2125 * network layer or drivers should need annotation to consolidate the
2126 * main types of usage into 3 classes.
2127 */
2128static inline void skb_queue_head_init(struct sk_buff_head *list)
2129{
2130 spin_lock_init(&list->lock);
2131 __skb_queue_head_init(list);
2132}
2133
2134static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2135 struct lock_class_key *class)
2136{
2137 skb_queue_head_init(list);
2138 lockdep_set_class(&list->lock, class);
2139}
2140
2141/*
2142 * Insert an sk_buff on a list.
2143 *
2144 * The "__skb_xxxx()" functions are the non-atomic ones that
2145 * can only be called with interrupts disabled.
2146 */
2147static inline void __skb_insert(struct sk_buff *newsk,
2148 struct sk_buff *prev, struct sk_buff *next,
2149 struct sk_buff_head *list)
2150{
2151 /* See skb_queue_empty_lockless() and skb_peek_tail()
2152 * for the opposite READ_ONCE()
2153 */
2154 WRITE_ONCE(newsk->next, next);
2155 WRITE_ONCE(newsk->prev, prev);
2156 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2157 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2158 WRITE_ONCE(list->qlen, list->qlen + 1);
2159}
2160
2161static inline void __skb_queue_splice(const struct sk_buff_head *list,
2162 struct sk_buff *prev,
2163 struct sk_buff *next)
2164{
2165 struct sk_buff *first = list->next;
2166 struct sk_buff *last = list->prev;
2167
2168 WRITE_ONCE(first->prev, prev);
2169 WRITE_ONCE(prev->next, first);
2170
2171 WRITE_ONCE(last->next, next);
2172 WRITE_ONCE(next->prev, last);
2173}
2174
2175/**
2176 * skb_queue_splice - join two skb lists, this is designed for stacks
2177 * @list: the new list to add
2178 * @head: the place to add it in the first list
2179 */
2180static inline void skb_queue_splice(const struct sk_buff_head *list,
2181 struct sk_buff_head *head)
2182{
2183 if (!skb_queue_empty(list)) {
2184 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2185 head->qlen += list->qlen;
2186 }
2187}
2188
2189/**
2190 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2191 * @list: the new list to add
2192 * @head: the place to add it in the first list
2193 *
2194 * The list at @list is reinitialised
2195 */
2196static inline void skb_queue_splice_init(struct sk_buff_head *list,
2197 struct sk_buff_head *head)
2198{
2199 if (!skb_queue_empty(list)) {
2200 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2201 head->qlen += list->qlen;
2202 __skb_queue_head_init(list);
2203 }
2204}
2205
2206/**
2207 * skb_queue_splice_tail - join two skb lists, each list being a queue
2208 * @list: the new list to add
2209 * @head: the place to add it in the first list
2210 */
2211static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2212 struct sk_buff_head *head)
2213{
2214 if (!skb_queue_empty(list)) {
2215 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2216 head->qlen += list->qlen;
2217 }
2218}
2219
2220/**
2221 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2222 * @list: the new list to add
2223 * @head: the place to add it in the first list
2224 *
2225 * Each of the lists is a queue.
2226 * The list at @list is reinitialised
2227 */
2228static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2229 struct sk_buff_head *head)
2230{
2231 if (!skb_queue_empty(list)) {
2232 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2233 head->qlen += list->qlen;
2234 __skb_queue_head_init(list);
2235 }
2236}
2237
2238/**
2239 * __skb_queue_after - queue a buffer at the list head
2240 * @list: list to use
2241 * @prev: place after this buffer
2242 * @newsk: buffer to queue
2243 *
2244 * Queue a buffer int the middle of a list. This function takes no locks
2245 * and you must therefore hold required locks before calling it.
2246 *
2247 * A buffer cannot be placed on two lists at the same time.
2248 */
2249static inline void __skb_queue_after(struct sk_buff_head *list,
2250 struct sk_buff *prev,
2251 struct sk_buff *newsk)
2252{
2253 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2254}
2255
2256void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2257 struct sk_buff_head *list);
2258
2259static inline void __skb_queue_before(struct sk_buff_head *list,
2260 struct sk_buff *next,
2261 struct sk_buff *newsk)
2262{
2263 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2264}
2265
2266/**
2267 * __skb_queue_head - queue a buffer at the list head
2268 * @list: list to use
2269 * @newsk: buffer to queue
2270 *
2271 * Queue a buffer at the start of a list. This function takes no locks
2272 * and you must therefore hold required locks before calling it.
2273 *
2274 * A buffer cannot be placed on two lists at the same time.
2275 */
2276static inline void __skb_queue_head(struct sk_buff_head *list,
2277 struct sk_buff *newsk)
2278{
2279 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2280}
2281void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2282
2283/**
2284 * __skb_queue_tail - queue a buffer at the list tail
2285 * @list: list to use
2286 * @newsk: buffer to queue
2287 *
2288 * Queue a buffer at the end of a list. This function takes no locks
2289 * and you must therefore hold required locks before calling it.
2290 *
2291 * A buffer cannot be placed on two lists at the same time.
2292 */
2293static inline void __skb_queue_tail(struct sk_buff_head *list,
2294 struct sk_buff *newsk)
2295{
2296 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2297}
2298void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2299
2300/*
2301 * remove sk_buff from list. _Must_ be called atomically, and with
2302 * the list known..
2303 */
2304void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2305static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2306{
2307 struct sk_buff *next, *prev;
2308
2309 WRITE_ONCE(list->qlen, list->qlen - 1);
2310 next = skb->next;
2311 prev = skb->prev;
2312 skb->next = skb->prev = NULL;
2313 WRITE_ONCE(next->prev, prev);
2314 WRITE_ONCE(prev->next, next);
2315}
2316
2317/**
2318 * __skb_dequeue - remove from the head of the queue
2319 * @list: list to dequeue from
2320 *
2321 * Remove the head of the list. This function does not take any locks
2322 * so must be used with appropriate locks held only. The head item is
2323 * returned or %NULL if the list is empty.
2324 */
2325static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2326{
2327 struct sk_buff *skb = skb_peek(list);
2328 if (skb)
2329 __skb_unlink(skb, list);
2330 return skb;
2331}
2332struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2333
2334/**
2335 * __skb_dequeue_tail - remove from the tail of the queue
2336 * @list: list to dequeue from
2337 *
2338 * Remove the tail of the list. This function does not take any locks
2339 * so must be used with appropriate locks held only. The tail item is
2340 * returned or %NULL if the list is empty.
2341 */
2342static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2343{
2344 struct sk_buff *skb = skb_peek_tail(list);
2345 if (skb)
2346 __skb_unlink(skb, list);
2347 return skb;
2348}
2349struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2350
2351
2352static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2353{
2354 return skb->data_len;
2355}
2356
2357static inline unsigned int skb_headlen(const struct sk_buff *skb)
2358{
2359 return skb->len - skb->data_len;
2360}
2361
2362static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2363{
2364 unsigned int i, len = 0;
2365
2366 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2367 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2368 return len;
2369}
2370
2371static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2372{
2373 return skb_headlen(skb) + __skb_pagelen(skb);
2374}
2375
2376static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2377 int i, struct page *page,
2378 int off, int size)
2379{
2380 skb_frag_t *frag = &shinfo->frags[i];
2381
2382 /*
2383 * Propagate page pfmemalloc to the skb if we can. The problem is
2384 * that not all callers have unique ownership of the page but rely
2385 * on page_is_pfmemalloc doing the right thing(tm).
2386 */
2387 frag->bv_page = page;
2388 frag->bv_offset = off;
2389 skb_frag_size_set(frag, size);
2390}
2391
2392/**
2393 * skb_len_add - adds a number to len fields of skb
2394 * @skb: buffer to add len to
2395 * @delta: number of bytes to add
2396 */
2397static inline void skb_len_add(struct sk_buff *skb, int delta)
2398{
2399 skb->len += delta;
2400 skb->data_len += delta;
2401 skb->truesize += delta;
2402}
2403
2404/**
2405 * __skb_fill_page_desc - initialise a paged fragment in an skb
2406 * @skb: buffer containing fragment to be initialised
2407 * @i: paged fragment index to initialise
2408 * @page: the page to use for this fragment
2409 * @off: the offset to the data with @page
2410 * @size: the length of the data
2411 *
2412 * Initialises the @i'th fragment of @skb to point to &size bytes at
2413 * offset @off within @page.
2414 *
2415 * Does not take any additional reference on the fragment.
2416 */
2417static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2418 struct page *page, int off, int size)
2419{
2420 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size);
2421 page = compound_head(page);
2422 if (page_is_pfmemalloc(page))
2423 skb->pfmemalloc = true;
2424}
2425
2426/**
2427 * skb_fill_page_desc - initialise a paged fragment in an skb
2428 * @skb: buffer containing fragment to be initialised
2429 * @i: paged fragment index to initialise
2430 * @page: the page to use for this fragment
2431 * @off: the offset to the data with @page
2432 * @size: the length of the data
2433 *
2434 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2435 * @skb to point to @size bytes at offset @off within @page. In
2436 * addition updates @skb such that @i is the last fragment.
2437 *
2438 * Does not take any additional reference on the fragment.
2439 */
2440static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2441 struct page *page, int off, int size)
2442{
2443 __skb_fill_page_desc(skb, i, page, off, size);
2444 skb_shinfo(skb)->nr_frags = i + 1;
2445}
2446
2447void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2448 int size, unsigned int truesize);
2449
2450void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2451 unsigned int truesize);
2452
2453#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2454
2455#ifdef NET_SKBUFF_DATA_USES_OFFSET
2456static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2457{
2458 return skb->head + skb->tail;
2459}
2460
2461static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2462{
2463 skb->tail = skb->data - skb->head;
2464}
2465
2466static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2467{
2468 skb_reset_tail_pointer(skb);
2469 skb->tail += offset;
2470}
2471
2472#else /* NET_SKBUFF_DATA_USES_OFFSET */
2473static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2474{
2475 return skb->tail;
2476}
2477
2478static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2479{
2480 skb->tail = skb->data;
2481}
2482
2483static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2484{
2485 skb->tail = skb->data + offset;
2486}
2487
2488#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2489
2490static inline void skb_assert_len(struct sk_buff *skb)
2491{
2492#ifdef CONFIG_DEBUG_NET
2493 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2494 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2495#endif /* CONFIG_DEBUG_NET */
2496}
2497
2498/*
2499 * Add data to an sk_buff
2500 */
2501void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2502void *skb_put(struct sk_buff *skb, unsigned int len);
2503static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2504{
2505 void *tmp = skb_tail_pointer(skb);
2506 SKB_LINEAR_ASSERT(skb);
2507 skb->tail += len;
2508 skb->len += len;
2509 return tmp;
2510}
2511
2512static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2513{
2514 void *tmp = __skb_put(skb, len);
2515
2516 memset(tmp, 0, len);
2517 return tmp;
2518}
2519
2520static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2521 unsigned int len)
2522{
2523 void *tmp = __skb_put(skb, len);
2524
2525 memcpy(tmp, data, len);
2526 return tmp;
2527}
2528
2529static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2530{
2531 *(u8 *)__skb_put(skb, 1) = val;
2532}
2533
2534static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2535{
2536 void *tmp = skb_put(skb, len);
2537
2538 memset(tmp, 0, len);
2539
2540 return tmp;
2541}
2542
2543static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2544 unsigned int len)
2545{
2546 void *tmp = skb_put(skb, len);
2547
2548 memcpy(tmp, data, len);
2549
2550 return tmp;
2551}
2552
2553static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2554{
2555 *(u8 *)skb_put(skb, 1) = val;
2556}
2557
2558void *skb_push(struct sk_buff *skb, unsigned int len);
2559static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2560{
2561 skb->data -= len;
2562 skb->len += len;
2563 return skb->data;
2564}
2565
2566void *skb_pull(struct sk_buff *skb, unsigned int len);
2567static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2568{
2569 skb->len -= len;
2570 if (unlikely(skb->len < skb->data_len)) {
2571#if defined(CONFIG_DEBUG_NET)
2572 skb->len += len;
2573 pr_err("__skb_pull(len=%u)\n", len);
2574 skb_dump(KERN_ERR, skb, false);
2575#endif
2576 BUG();
2577 }
2578 return skb->data += len;
2579}
2580
2581static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2582{
2583 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2584}
2585
2586void *skb_pull_data(struct sk_buff *skb, size_t len);
2587
2588void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2589
2590static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2591{
2592 if (len > skb_headlen(skb) &&
2593 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2594 return NULL;
2595 skb->len -= len;
2596 return skb->data += len;
2597}
2598
2599static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2600{
2601 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2602}
2603
2604static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2605{
2606 if (likely(len <= skb_headlen(skb)))
2607 return true;
2608 if (unlikely(len > skb->len))
2609 return false;
2610 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2611}
2612
2613void skb_condense(struct sk_buff *skb);
2614
2615/**
2616 * skb_headroom - bytes at buffer head
2617 * @skb: buffer to check
2618 *
2619 * Return the number of bytes of free space at the head of an &sk_buff.
2620 */
2621static inline unsigned int skb_headroom(const struct sk_buff *skb)
2622{
2623 return skb->data - skb->head;
2624}
2625
2626/**
2627 * skb_tailroom - bytes at buffer end
2628 * @skb: buffer to check
2629 *
2630 * Return the number of bytes of free space at the tail of an sk_buff
2631 */
2632static inline int skb_tailroom(const struct sk_buff *skb)
2633{
2634 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2635}
2636
2637/**
2638 * skb_availroom - bytes at buffer end
2639 * @skb: buffer to check
2640 *
2641 * Return the number of bytes of free space at the tail of an sk_buff
2642 * allocated by sk_stream_alloc()
2643 */
2644static inline int skb_availroom(const struct sk_buff *skb)
2645{
2646 if (skb_is_nonlinear(skb))
2647 return 0;
2648
2649 return skb->end - skb->tail - skb->reserved_tailroom;
2650}
2651
2652/**
2653 * skb_reserve - adjust headroom
2654 * @skb: buffer to alter
2655 * @len: bytes to move
2656 *
2657 * Increase the headroom of an empty &sk_buff by reducing the tail
2658 * room. This is only allowed for an empty buffer.
2659 */
2660static inline void skb_reserve(struct sk_buff *skb, int len)
2661{
2662 skb->data += len;
2663 skb->tail += len;
2664}
2665
2666/**
2667 * skb_tailroom_reserve - adjust reserved_tailroom
2668 * @skb: buffer to alter
2669 * @mtu: maximum amount of headlen permitted
2670 * @needed_tailroom: minimum amount of reserved_tailroom
2671 *
2672 * Set reserved_tailroom so that headlen can be as large as possible but
2673 * not larger than mtu and tailroom cannot be smaller than
2674 * needed_tailroom.
2675 * The required headroom should already have been reserved before using
2676 * this function.
2677 */
2678static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2679 unsigned int needed_tailroom)
2680{
2681 SKB_LINEAR_ASSERT(skb);
2682 if (mtu < skb_tailroom(skb) - needed_tailroom)
2683 /* use at most mtu */
2684 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2685 else
2686 /* use up to all available space */
2687 skb->reserved_tailroom = needed_tailroom;
2688}
2689
2690#define ENCAP_TYPE_ETHER 0
2691#define ENCAP_TYPE_IPPROTO 1
2692
2693static inline void skb_set_inner_protocol(struct sk_buff *skb,
2694 __be16 protocol)
2695{
2696 skb->inner_protocol = protocol;
2697 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2698}
2699
2700static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2701 __u8 ipproto)
2702{
2703 skb->inner_ipproto = ipproto;
2704 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2705}
2706
2707static inline void skb_reset_inner_headers(struct sk_buff *skb)
2708{
2709 skb->inner_mac_header = skb->mac_header;
2710 skb->inner_network_header = skb->network_header;
2711 skb->inner_transport_header = skb->transport_header;
2712}
2713
2714static inline void skb_reset_mac_len(struct sk_buff *skb)
2715{
2716 skb->mac_len = skb->network_header - skb->mac_header;
2717}
2718
2719static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2720 *skb)
2721{
2722 return skb->head + skb->inner_transport_header;
2723}
2724
2725static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2726{
2727 return skb_inner_transport_header(skb) - skb->data;
2728}
2729
2730static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2731{
2732 skb->inner_transport_header = skb->data - skb->head;
2733}
2734
2735static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2736 const int offset)
2737{
2738 skb_reset_inner_transport_header(skb);
2739 skb->inner_transport_header += offset;
2740}
2741
2742static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2743{
2744 return skb->head + skb->inner_network_header;
2745}
2746
2747static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2748{
2749 skb->inner_network_header = skb->data - skb->head;
2750}
2751
2752static inline void skb_set_inner_network_header(struct sk_buff *skb,
2753 const int offset)
2754{
2755 skb_reset_inner_network_header(skb);
2756 skb->inner_network_header += offset;
2757}
2758
2759static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2760{
2761 return skb->head + skb->inner_mac_header;
2762}
2763
2764static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2765{
2766 skb->inner_mac_header = skb->data - skb->head;
2767}
2768
2769static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2770 const int offset)
2771{
2772 skb_reset_inner_mac_header(skb);
2773 skb->inner_mac_header += offset;
2774}
2775static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2776{
2777 return skb->transport_header != (typeof(skb->transport_header))~0U;
2778}
2779
2780static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2781{
2782 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2783 return skb->head + skb->transport_header;
2784}
2785
2786static inline void skb_reset_transport_header(struct sk_buff *skb)
2787{
2788 skb->transport_header = skb->data - skb->head;
2789}
2790
2791static inline void skb_set_transport_header(struct sk_buff *skb,
2792 const int offset)
2793{
2794 skb_reset_transport_header(skb);
2795 skb->transport_header += offset;
2796}
2797
2798static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2799{
2800 return skb->head + skb->network_header;
2801}
2802
2803static inline void skb_reset_network_header(struct sk_buff *skb)
2804{
2805 skb->network_header = skb->data - skb->head;
2806}
2807
2808static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2809{
2810 skb_reset_network_header(skb);
2811 skb->network_header += offset;
2812}
2813
2814static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2815{
2816 return skb->mac_header != (typeof(skb->mac_header))~0U;
2817}
2818
2819static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2820{
2821 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2822 return skb->head + skb->mac_header;
2823}
2824
2825static inline int skb_mac_offset(const struct sk_buff *skb)
2826{
2827 return skb_mac_header(skb) - skb->data;
2828}
2829
2830static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2831{
2832 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2833 return skb->network_header - skb->mac_header;
2834}
2835
2836static inline void skb_unset_mac_header(struct sk_buff *skb)
2837{
2838 skb->mac_header = (typeof(skb->mac_header))~0U;
2839}
2840
2841static inline void skb_reset_mac_header(struct sk_buff *skb)
2842{
2843 skb->mac_header = skb->data - skb->head;
2844}
2845
2846static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2847{
2848 skb_reset_mac_header(skb);
2849 skb->mac_header += offset;
2850}
2851
2852static inline void skb_pop_mac_header(struct sk_buff *skb)
2853{
2854 skb->mac_header = skb->network_header;
2855}
2856
2857static inline void skb_probe_transport_header(struct sk_buff *skb)
2858{
2859 struct flow_keys_basic keys;
2860
2861 if (skb_transport_header_was_set(skb))
2862 return;
2863
2864 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2865 NULL, 0, 0, 0, 0))
2866 skb_set_transport_header(skb, keys.control.thoff);
2867}
2868
2869static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2870{
2871 if (skb_mac_header_was_set(skb)) {
2872 const unsigned char *old_mac = skb_mac_header(skb);
2873
2874 skb_set_mac_header(skb, -skb->mac_len);
2875 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2876 }
2877}
2878
2879static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2880{
2881 return skb->csum_start - skb_headroom(skb);
2882}
2883
2884static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2885{
2886 return skb->head + skb->csum_start;
2887}
2888
2889static inline int skb_transport_offset(const struct sk_buff *skb)
2890{
2891 return skb_transport_header(skb) - skb->data;
2892}
2893
2894static inline u32 skb_network_header_len(const struct sk_buff *skb)
2895{
2896 return skb->transport_header - skb->network_header;
2897}
2898
2899static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2900{
2901 return skb->inner_transport_header - skb->inner_network_header;
2902}
2903
2904static inline int skb_network_offset(const struct sk_buff *skb)
2905{
2906 return skb_network_header(skb) - skb->data;
2907}
2908
2909static inline int skb_inner_network_offset(const struct sk_buff *skb)
2910{
2911 return skb_inner_network_header(skb) - skb->data;
2912}
2913
2914static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2915{
2916 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2917}
2918
2919/*
2920 * CPUs often take a performance hit when accessing unaligned memory
2921 * locations. The actual performance hit varies, it can be small if the
2922 * hardware handles it or large if we have to take an exception and fix it
2923 * in software.
2924 *
2925 * Since an ethernet header is 14 bytes network drivers often end up with
2926 * the IP header at an unaligned offset. The IP header can be aligned by
2927 * shifting the start of the packet by 2 bytes. Drivers should do this
2928 * with:
2929 *
2930 * skb_reserve(skb, NET_IP_ALIGN);
2931 *
2932 * The downside to this alignment of the IP header is that the DMA is now
2933 * unaligned. On some architectures the cost of an unaligned DMA is high
2934 * and this cost outweighs the gains made by aligning the IP header.
2935 *
2936 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2937 * to be overridden.
2938 */
2939#ifndef NET_IP_ALIGN
2940#define NET_IP_ALIGN 2
2941#endif
2942
2943/*
2944 * The networking layer reserves some headroom in skb data (via
2945 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2946 * the header has to grow. In the default case, if the header has to grow
2947 * 32 bytes or less we avoid the reallocation.
2948 *
2949 * Unfortunately this headroom changes the DMA alignment of the resulting
2950 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2951 * on some architectures. An architecture can override this value,
2952 * perhaps setting it to a cacheline in size (since that will maintain
2953 * cacheline alignment of the DMA). It must be a power of 2.
2954 *
2955 * Various parts of the networking layer expect at least 32 bytes of
2956 * headroom, you should not reduce this.
2957 *
2958 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2959 * to reduce average number of cache lines per packet.
2960 * get_rps_cpu() for example only access one 64 bytes aligned block :
2961 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2962 */
2963#ifndef NET_SKB_PAD
2964#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2965#endif
2966
2967int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2968
2969static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2970{
2971 if (WARN_ON(skb_is_nonlinear(skb)))
2972 return;
2973 skb->len = len;
2974 skb_set_tail_pointer(skb, len);
2975}
2976
2977static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2978{
2979 __skb_set_length(skb, len);
2980}
2981
2982void skb_trim(struct sk_buff *skb, unsigned int len);
2983
2984static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2985{
2986 if (skb->data_len)
2987 return ___pskb_trim(skb, len);
2988 __skb_trim(skb, len);
2989 return 0;
2990}
2991
2992static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2993{
2994 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2995}
2996
2997/**
2998 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2999 * @skb: buffer to alter
3000 * @len: new length
3001 *
3002 * This is identical to pskb_trim except that the caller knows that
3003 * the skb is not cloned so we should never get an error due to out-
3004 * of-memory.
3005 */
3006static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3007{
3008 int err = pskb_trim(skb, len);
3009 BUG_ON(err);
3010}
3011
3012static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3013{
3014 unsigned int diff = len - skb->len;
3015
3016 if (skb_tailroom(skb) < diff) {
3017 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3018 GFP_ATOMIC);
3019 if (ret)
3020 return ret;
3021 }
3022 __skb_set_length(skb, len);
3023 return 0;
3024}
3025
3026/**
3027 * skb_orphan - orphan a buffer
3028 * @skb: buffer to orphan
3029 *
3030 * If a buffer currently has an owner then we call the owner's
3031 * destructor function and make the @skb unowned. The buffer continues
3032 * to exist but is no longer charged to its former owner.
3033 */
3034static inline void skb_orphan(struct sk_buff *skb)
3035{
3036 if (skb->destructor) {
3037 skb->destructor(skb);
3038 skb->destructor = NULL;
3039 skb->sk = NULL;
3040 } else {
3041 BUG_ON(skb->sk);
3042 }
3043}
3044
3045/**
3046 * skb_orphan_frags - orphan the frags contained in a buffer
3047 * @skb: buffer to orphan frags from
3048 * @gfp_mask: allocation mask for replacement pages
3049 *
3050 * For each frag in the SKB which needs a destructor (i.e. has an
3051 * owner) create a copy of that frag and release the original
3052 * page by calling the destructor.
3053 */
3054static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3055{
3056 if (likely(!skb_zcopy(skb)))
3057 return 0;
3058 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3059 return 0;
3060 return skb_copy_ubufs(skb, gfp_mask);
3061}
3062
3063/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3064static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3065{
3066 if (likely(!skb_zcopy(skb)))
3067 return 0;
3068 return skb_copy_ubufs(skb, gfp_mask);
3069}
3070
3071/**
3072 * __skb_queue_purge - empty a list
3073 * @list: list to empty
3074 *
3075 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3076 * the list and one reference dropped. This function does not take the
3077 * list lock and the caller must hold the relevant locks to use it.
3078 */
3079static inline void __skb_queue_purge(struct sk_buff_head *list)
3080{
3081 struct sk_buff *skb;
3082 while ((skb = __skb_dequeue(list)) != NULL)
3083 kfree_skb(skb);
3084}
3085void skb_queue_purge(struct sk_buff_head *list);
3086
3087unsigned int skb_rbtree_purge(struct rb_root *root);
3088
3089void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3090
3091/**
3092 * netdev_alloc_frag - allocate a page fragment
3093 * @fragsz: fragment size
3094 *
3095 * Allocates a frag from a page for receive buffer.
3096 * Uses GFP_ATOMIC allocations.
3097 */
3098static inline void *netdev_alloc_frag(unsigned int fragsz)
3099{
3100 return __netdev_alloc_frag_align(fragsz, ~0u);
3101}
3102
3103static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3104 unsigned int align)
3105{
3106 WARN_ON_ONCE(!is_power_of_2(align));
3107 return __netdev_alloc_frag_align(fragsz, -align);
3108}
3109
3110struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3111 gfp_t gfp_mask);
3112
3113/**
3114 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3115 * @dev: network device to receive on
3116 * @length: length to allocate
3117 *
3118 * Allocate a new &sk_buff and assign it a usage count of one. The
3119 * buffer has unspecified headroom built in. Users should allocate
3120 * the headroom they think they need without accounting for the
3121 * built in space. The built in space is used for optimisations.
3122 *
3123 * %NULL is returned if there is no free memory. Although this function
3124 * allocates memory it can be called from an interrupt.
3125 */
3126static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3127 unsigned int length)
3128{
3129 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3130}
3131
3132/* legacy helper around __netdev_alloc_skb() */
3133static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3134 gfp_t gfp_mask)
3135{
3136 return __netdev_alloc_skb(NULL, length, gfp_mask);
3137}
3138
3139/* legacy helper around netdev_alloc_skb() */
3140static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3141{
3142 return netdev_alloc_skb(NULL, length);
3143}
3144
3145
3146static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3147 unsigned int length, gfp_t gfp)
3148{
3149 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3150
3151 if (NET_IP_ALIGN && skb)
3152 skb_reserve(skb, NET_IP_ALIGN);
3153 return skb;
3154}
3155
3156static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3157 unsigned int length)
3158{
3159 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3160}
3161
3162static inline void skb_free_frag(void *addr)
3163{
3164 page_frag_free(addr);
3165}
3166
3167void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3168
3169static inline void *napi_alloc_frag(unsigned int fragsz)
3170{
3171 return __napi_alloc_frag_align(fragsz, ~0u);
3172}
3173
3174static inline void *napi_alloc_frag_align(unsigned int fragsz,
3175 unsigned int align)
3176{
3177 WARN_ON_ONCE(!is_power_of_2(align));
3178 return __napi_alloc_frag_align(fragsz, -align);
3179}
3180
3181struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3182 unsigned int length, gfp_t gfp_mask);
3183static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3184 unsigned int length)
3185{
3186 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3187}
3188void napi_consume_skb(struct sk_buff *skb, int budget);
3189
3190void napi_skb_free_stolen_head(struct sk_buff *skb);
3191void __kfree_skb_defer(struct sk_buff *skb);
3192
3193/**
3194 * __dev_alloc_pages - allocate page for network Rx
3195 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3196 * @order: size of the allocation
3197 *
3198 * Allocate a new page.
3199 *
3200 * %NULL is returned if there is no free memory.
3201*/
3202static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3203 unsigned int order)
3204{
3205 /* This piece of code contains several assumptions.
3206 * 1. This is for device Rx, therefor a cold page is preferred.
3207 * 2. The expectation is the user wants a compound page.
3208 * 3. If requesting a order 0 page it will not be compound
3209 * due to the check to see if order has a value in prep_new_page
3210 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3211 * code in gfp_to_alloc_flags that should be enforcing this.
3212 */
3213 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3214
3215 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3216}
3217
3218static inline struct page *dev_alloc_pages(unsigned int order)
3219{
3220 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3221}
3222
3223/**
3224 * __dev_alloc_page - allocate a page for network Rx
3225 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3226 *
3227 * Allocate a new page.
3228 *
3229 * %NULL is returned if there is no free memory.
3230 */
3231static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3232{
3233 return __dev_alloc_pages(gfp_mask, 0);
3234}
3235
3236static inline struct page *dev_alloc_page(void)
3237{
3238 return dev_alloc_pages(0);
3239}
3240
3241/**
3242 * dev_page_is_reusable - check whether a page can be reused for network Rx
3243 * @page: the page to test
3244 *
3245 * A page shouldn't be considered for reusing/recycling if it was allocated
3246 * under memory pressure or at a distant memory node.
3247 *
3248 * Returns false if this page should be returned to page allocator, true
3249 * otherwise.
3250 */
3251static inline bool dev_page_is_reusable(const struct page *page)
3252{
3253 return likely(page_to_nid(page) == numa_mem_id() &&
3254 !page_is_pfmemalloc(page));
3255}
3256
3257/**
3258 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3259 * @page: The page that was allocated from skb_alloc_page
3260 * @skb: The skb that may need pfmemalloc set
3261 */
3262static inline void skb_propagate_pfmemalloc(const struct page *page,
3263 struct sk_buff *skb)
3264{
3265 if (page_is_pfmemalloc(page))
3266 skb->pfmemalloc = true;
3267}
3268
3269/**
3270 * skb_frag_off() - Returns the offset of a skb fragment
3271 * @frag: the paged fragment
3272 */
3273static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3274{
3275 return frag->bv_offset;
3276}
3277
3278/**
3279 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3280 * @frag: skb fragment
3281 * @delta: value to add
3282 */
3283static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3284{
3285 frag->bv_offset += delta;
3286}
3287
3288/**
3289 * skb_frag_off_set() - Sets the offset of a skb fragment
3290 * @frag: skb fragment
3291 * @offset: offset of fragment
3292 */
3293static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3294{
3295 frag->bv_offset = offset;
3296}
3297
3298/**
3299 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3300 * @fragto: skb fragment where offset is set
3301 * @fragfrom: skb fragment offset is copied from
3302 */
3303static inline void skb_frag_off_copy(skb_frag_t *fragto,
3304 const skb_frag_t *fragfrom)
3305{
3306 fragto->bv_offset = fragfrom->bv_offset;
3307}
3308
3309/**
3310 * skb_frag_page - retrieve the page referred to by a paged fragment
3311 * @frag: the paged fragment
3312 *
3313 * Returns the &struct page associated with @frag.
3314 */
3315static inline struct page *skb_frag_page(const skb_frag_t *frag)
3316{
3317 return frag->bv_page;
3318}
3319
3320/**
3321 * __skb_frag_ref - take an addition reference on a paged fragment.
3322 * @frag: the paged fragment
3323 *
3324 * Takes an additional reference on the paged fragment @frag.
3325 */
3326static inline void __skb_frag_ref(skb_frag_t *frag)
3327{
3328 get_page(skb_frag_page(frag));
3329}
3330
3331/**
3332 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3333 * @skb: the buffer
3334 * @f: the fragment offset.
3335 *
3336 * Takes an additional reference on the @f'th paged fragment of @skb.
3337 */
3338static inline void skb_frag_ref(struct sk_buff *skb, int f)
3339{
3340 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3341}
3342
3343/**
3344 * __skb_frag_unref - release a reference on a paged fragment.
3345 * @frag: the paged fragment
3346 * @recycle: recycle the page if allocated via page_pool
3347 *
3348 * Releases a reference on the paged fragment @frag
3349 * or recycles the page via the page_pool API.
3350 */
3351static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3352{
3353 struct page *page = skb_frag_page(frag);
3354
3355#ifdef CONFIG_PAGE_POOL
3356 if (recycle && page_pool_return_skb_page(page))
3357 return;
3358#endif
3359 put_page(page);
3360}
3361
3362/**
3363 * skb_frag_unref - release a reference on a paged fragment of an skb.
3364 * @skb: the buffer
3365 * @f: the fragment offset
3366 *
3367 * Releases a reference on the @f'th paged fragment of @skb.
3368 */
3369static inline void skb_frag_unref(struct sk_buff *skb, int f)
3370{
3371 struct skb_shared_info *shinfo = skb_shinfo(skb);
3372
3373 if (!skb_zcopy_managed(skb))
3374 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3375}
3376
3377/**
3378 * skb_frag_address - gets the address of the data contained in a paged fragment
3379 * @frag: the paged fragment buffer
3380 *
3381 * Returns the address of the data within @frag. The page must already
3382 * be mapped.
3383 */
3384static inline void *skb_frag_address(const skb_frag_t *frag)
3385{
3386 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3387}
3388
3389/**
3390 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3391 * @frag: the paged fragment buffer
3392 *
3393 * Returns the address of the data within @frag. Checks that the page
3394 * is mapped and returns %NULL otherwise.
3395 */
3396static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3397{
3398 void *ptr = page_address(skb_frag_page(frag));
3399 if (unlikely(!ptr))
3400 return NULL;
3401
3402 return ptr + skb_frag_off(frag);
3403}
3404
3405/**
3406 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3407 * @fragto: skb fragment where page is set
3408 * @fragfrom: skb fragment page is copied from
3409 */
3410static inline void skb_frag_page_copy(skb_frag_t *fragto,
3411 const skb_frag_t *fragfrom)
3412{
3413 fragto->bv_page = fragfrom->bv_page;
3414}
3415
3416/**
3417 * __skb_frag_set_page - sets the page contained in a paged fragment
3418 * @frag: the paged fragment
3419 * @page: the page to set
3420 *
3421 * Sets the fragment @frag to contain @page.
3422 */
3423static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3424{
3425 frag->bv_page = page;
3426}
3427
3428/**
3429 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3430 * @skb: the buffer
3431 * @f: the fragment offset
3432 * @page: the page to set
3433 *
3434 * Sets the @f'th fragment of @skb to contain @page.
3435 */
3436static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3437 struct page *page)
3438{
3439 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3440}
3441
3442bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3443
3444/**
3445 * skb_frag_dma_map - maps a paged fragment via the DMA API
3446 * @dev: the device to map the fragment to
3447 * @frag: the paged fragment to map
3448 * @offset: the offset within the fragment (starting at the
3449 * fragment's own offset)
3450 * @size: the number of bytes to map
3451 * @dir: the direction of the mapping (``PCI_DMA_*``)
3452 *
3453 * Maps the page associated with @frag to @device.
3454 */
3455static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3456 const skb_frag_t *frag,
3457 size_t offset, size_t size,
3458 enum dma_data_direction dir)
3459{
3460 return dma_map_page(dev, skb_frag_page(frag),
3461 skb_frag_off(frag) + offset, size, dir);
3462}
3463
3464static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3465 gfp_t gfp_mask)
3466{
3467 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3468}
3469
3470
3471static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3472 gfp_t gfp_mask)
3473{
3474 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3475}
3476
3477
3478/**
3479 * skb_clone_writable - is the header of a clone writable
3480 * @skb: buffer to check
3481 * @len: length up to which to write
3482 *
3483 * Returns true if modifying the header part of the cloned buffer
3484 * does not requires the data to be copied.
3485 */
3486static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3487{
3488 return !skb_header_cloned(skb) &&
3489 skb_headroom(skb) + len <= skb->hdr_len;
3490}
3491
3492static inline int skb_try_make_writable(struct sk_buff *skb,
3493 unsigned int write_len)
3494{
3495 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3496 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3497}
3498
3499static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3500 int cloned)
3501{
3502 int delta = 0;
3503
3504 if (headroom > skb_headroom(skb))
3505 delta = headroom - skb_headroom(skb);
3506
3507 if (delta || cloned)
3508 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3509 GFP_ATOMIC);
3510 return 0;
3511}
3512
3513/**
3514 * skb_cow - copy header of skb when it is required
3515 * @skb: buffer to cow
3516 * @headroom: needed headroom
3517 *
3518 * If the skb passed lacks sufficient headroom or its data part
3519 * is shared, data is reallocated. If reallocation fails, an error
3520 * is returned and original skb is not changed.
3521 *
3522 * The result is skb with writable area skb->head...skb->tail
3523 * and at least @headroom of space at head.
3524 */
3525static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3526{
3527 return __skb_cow(skb, headroom, skb_cloned(skb));
3528}
3529
3530/**
3531 * skb_cow_head - skb_cow but only making the head writable
3532 * @skb: buffer to cow
3533 * @headroom: needed headroom
3534 *
3535 * This function is identical to skb_cow except that we replace the
3536 * skb_cloned check by skb_header_cloned. It should be used when
3537 * you only need to push on some header and do not need to modify
3538 * the data.
3539 */
3540static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3541{
3542 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3543}
3544
3545/**
3546 * skb_padto - pad an skbuff up to a minimal size
3547 * @skb: buffer to pad
3548 * @len: minimal length
3549 *
3550 * Pads up a buffer to ensure the trailing bytes exist and are
3551 * blanked. If the buffer already contains sufficient data it
3552 * is untouched. Otherwise it is extended. Returns zero on
3553 * success. The skb is freed on error.
3554 */
3555static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3556{
3557 unsigned int size = skb->len;
3558 if (likely(size >= len))
3559 return 0;
3560 return skb_pad(skb, len - size);
3561}
3562
3563/**
3564 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3565 * @skb: buffer to pad
3566 * @len: minimal length
3567 * @free_on_error: free buffer on error
3568 *
3569 * Pads up a buffer to ensure the trailing bytes exist and are
3570 * blanked. If the buffer already contains sufficient data it
3571 * is untouched. Otherwise it is extended. Returns zero on
3572 * success. The skb is freed on error if @free_on_error is true.
3573 */
3574static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3575 unsigned int len,
3576 bool free_on_error)
3577{
3578 unsigned int size = skb->len;
3579
3580 if (unlikely(size < len)) {
3581 len -= size;
3582 if (__skb_pad(skb, len, free_on_error))
3583 return -ENOMEM;
3584 __skb_put(skb, len);
3585 }
3586 return 0;
3587}
3588
3589/**
3590 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3591 * @skb: buffer to pad
3592 * @len: minimal length
3593 *
3594 * Pads up a buffer to ensure the trailing bytes exist and are
3595 * blanked. If the buffer already contains sufficient data it
3596 * is untouched. Otherwise it is extended. Returns zero on
3597 * success. The skb is freed on error.
3598 */
3599static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3600{
3601 return __skb_put_padto(skb, len, true);
3602}
3603
3604static inline int skb_add_data(struct sk_buff *skb,
3605 struct iov_iter *from, int copy)
3606{
3607 const int off = skb->len;
3608
3609 if (skb->ip_summed == CHECKSUM_NONE) {
3610 __wsum csum = 0;
3611 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3612 &csum, from)) {
3613 skb->csum = csum_block_add(skb->csum, csum, off);
3614 return 0;
3615 }
3616 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3617 return 0;
3618
3619 __skb_trim(skb, off);
3620 return -EFAULT;
3621}
3622
3623static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3624 const struct page *page, int off)
3625{
3626 if (skb_zcopy(skb))
3627 return false;
3628 if (i) {
3629 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3630
3631 return page == skb_frag_page(frag) &&
3632 off == skb_frag_off(frag) + skb_frag_size(frag);
3633 }
3634 return false;
3635}
3636
3637static inline int __skb_linearize(struct sk_buff *skb)
3638{
3639 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3640}
3641
3642/**
3643 * skb_linearize - convert paged skb to linear one
3644 * @skb: buffer to linarize
3645 *
3646 * If there is no free memory -ENOMEM is returned, otherwise zero
3647 * is returned and the old skb data released.
3648 */
3649static inline int skb_linearize(struct sk_buff *skb)
3650{
3651 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3652}
3653
3654/**
3655 * skb_has_shared_frag - can any frag be overwritten
3656 * @skb: buffer to test
3657 *
3658 * Return true if the skb has at least one frag that might be modified
3659 * by an external entity (as in vmsplice()/sendfile())
3660 */
3661static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3662{
3663 return skb_is_nonlinear(skb) &&
3664 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3665}
3666
3667/**
3668 * skb_linearize_cow - make sure skb is linear and writable
3669 * @skb: buffer to process
3670 *
3671 * If there is no free memory -ENOMEM is returned, otherwise zero
3672 * is returned and the old skb data released.
3673 */
3674static inline int skb_linearize_cow(struct sk_buff *skb)
3675{
3676 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3677 __skb_linearize(skb) : 0;
3678}
3679
3680static __always_inline void
3681__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3682 unsigned int off)
3683{
3684 if (skb->ip_summed == CHECKSUM_COMPLETE)
3685 skb->csum = csum_block_sub(skb->csum,
3686 csum_partial(start, len, 0), off);
3687 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3688 skb_checksum_start_offset(skb) < 0)
3689 skb->ip_summed = CHECKSUM_NONE;
3690}
3691
3692/**
3693 * skb_postpull_rcsum - update checksum for received skb after pull
3694 * @skb: buffer to update
3695 * @start: start of data before pull
3696 * @len: length of data pulled
3697 *
3698 * After doing a pull on a received packet, you need to call this to
3699 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3700 * CHECKSUM_NONE so that it can be recomputed from scratch.
3701 */
3702static inline void skb_postpull_rcsum(struct sk_buff *skb,
3703 const void *start, unsigned int len)
3704{
3705 if (skb->ip_summed == CHECKSUM_COMPLETE)
3706 skb->csum = wsum_negate(csum_partial(start, len,
3707 wsum_negate(skb->csum)));
3708 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3709 skb_checksum_start_offset(skb) < 0)
3710 skb->ip_summed = CHECKSUM_NONE;
3711}
3712
3713static __always_inline void
3714__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3715 unsigned int off)
3716{
3717 if (skb->ip_summed == CHECKSUM_COMPLETE)
3718 skb->csum = csum_block_add(skb->csum,
3719 csum_partial(start, len, 0), off);
3720}
3721
3722/**
3723 * skb_postpush_rcsum - update checksum for received skb after push
3724 * @skb: buffer to update
3725 * @start: start of data after push
3726 * @len: length of data pushed
3727 *
3728 * After doing a push on a received packet, you need to call this to
3729 * update the CHECKSUM_COMPLETE checksum.
3730 */
3731static inline void skb_postpush_rcsum(struct sk_buff *skb,
3732 const void *start, unsigned int len)
3733{
3734 __skb_postpush_rcsum(skb, start, len, 0);
3735}
3736
3737void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3738
3739/**
3740 * skb_push_rcsum - push skb and update receive checksum
3741 * @skb: buffer to update
3742 * @len: length of data pulled
3743 *
3744 * This function performs an skb_push on the packet and updates
3745 * the CHECKSUM_COMPLETE checksum. It should be used on
3746 * receive path processing instead of skb_push unless you know
3747 * that the checksum difference is zero (e.g., a valid IP header)
3748 * or you are setting ip_summed to CHECKSUM_NONE.
3749 */
3750static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3751{
3752 skb_push(skb, len);
3753 skb_postpush_rcsum(skb, skb->data, len);
3754 return skb->data;
3755}
3756
3757int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3758/**
3759 * pskb_trim_rcsum - trim received skb and update checksum
3760 * @skb: buffer to trim
3761 * @len: new length
3762 *
3763 * This is exactly the same as pskb_trim except that it ensures the
3764 * checksum of received packets are still valid after the operation.
3765 * It can change skb pointers.
3766 */
3767
3768static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3769{
3770 if (likely(len >= skb->len))
3771 return 0;
3772 return pskb_trim_rcsum_slow(skb, len);
3773}
3774
3775static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3776{
3777 if (skb->ip_summed == CHECKSUM_COMPLETE)
3778 skb->ip_summed = CHECKSUM_NONE;
3779 __skb_trim(skb, len);
3780 return 0;
3781}
3782
3783static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3784{
3785 if (skb->ip_summed == CHECKSUM_COMPLETE)
3786 skb->ip_summed = CHECKSUM_NONE;
3787 return __skb_grow(skb, len);
3788}
3789
3790#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3791#define skb_rb_first(root) rb_to_skb(rb_first(root))
3792#define skb_rb_last(root) rb_to_skb(rb_last(root))
3793#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3794#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3795
3796#define skb_queue_walk(queue, skb) \
3797 for (skb = (queue)->next; \
3798 skb != (struct sk_buff *)(queue); \
3799 skb = skb->next)
3800
3801#define skb_queue_walk_safe(queue, skb, tmp) \
3802 for (skb = (queue)->next, tmp = skb->next; \
3803 skb != (struct sk_buff *)(queue); \
3804 skb = tmp, tmp = skb->next)
3805
3806#define skb_queue_walk_from(queue, skb) \
3807 for (; skb != (struct sk_buff *)(queue); \
3808 skb = skb->next)
3809
3810#define skb_rbtree_walk(skb, root) \
3811 for (skb = skb_rb_first(root); skb != NULL; \
3812 skb = skb_rb_next(skb))
3813
3814#define skb_rbtree_walk_from(skb) \
3815 for (; skb != NULL; \
3816 skb = skb_rb_next(skb))
3817
3818#define skb_rbtree_walk_from_safe(skb, tmp) \
3819 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3820 skb = tmp)
3821
3822#define skb_queue_walk_from_safe(queue, skb, tmp) \
3823 for (tmp = skb->next; \
3824 skb != (struct sk_buff *)(queue); \
3825 skb = tmp, tmp = skb->next)
3826
3827#define skb_queue_reverse_walk(queue, skb) \
3828 for (skb = (queue)->prev; \
3829 skb != (struct sk_buff *)(queue); \
3830 skb = skb->prev)
3831
3832#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3833 for (skb = (queue)->prev, tmp = skb->prev; \
3834 skb != (struct sk_buff *)(queue); \
3835 skb = tmp, tmp = skb->prev)
3836
3837#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3838 for (tmp = skb->prev; \
3839 skb != (struct sk_buff *)(queue); \
3840 skb = tmp, tmp = skb->prev)
3841
3842static inline bool skb_has_frag_list(const struct sk_buff *skb)
3843{
3844 return skb_shinfo(skb)->frag_list != NULL;
3845}
3846
3847static inline void skb_frag_list_init(struct sk_buff *skb)
3848{
3849 skb_shinfo(skb)->frag_list = NULL;
3850}
3851
3852#define skb_walk_frags(skb, iter) \
3853 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3854
3855
3856int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3857 int *err, long *timeo_p,
3858 const struct sk_buff *skb);
3859struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3860 struct sk_buff_head *queue,
3861 unsigned int flags,
3862 int *off, int *err,
3863 struct sk_buff **last);
3864struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3865 struct sk_buff_head *queue,
3866 unsigned int flags, int *off, int *err,
3867 struct sk_buff **last);
3868struct sk_buff *__skb_recv_datagram(struct sock *sk,
3869 struct sk_buff_head *sk_queue,
3870 unsigned int flags, int *off, int *err);
3871struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3872__poll_t datagram_poll(struct file *file, struct socket *sock,
3873 struct poll_table_struct *wait);
3874int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3875 struct iov_iter *to, int size);
3876static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3877 struct msghdr *msg, int size)
3878{
3879 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3880}
3881int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3882 struct msghdr *msg);
3883int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3884 struct iov_iter *to, int len,
3885 struct ahash_request *hash);
3886int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3887 struct iov_iter *from, int len);
3888int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3889void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3890void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3891static inline void skb_free_datagram_locked(struct sock *sk,
3892 struct sk_buff *skb)
3893{
3894 __skb_free_datagram_locked(sk, skb, 0);
3895}
3896int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3897int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3898int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3899__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3900 int len);
3901int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3902 struct pipe_inode_info *pipe, unsigned int len,
3903 unsigned int flags);
3904int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3905 int len);
3906int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3907void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3908unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3909int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3910 int len, int hlen);
3911void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3912int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3913void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3914bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3915bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3916struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3917struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3918 unsigned int offset);
3919struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3920int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
3921int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3922int skb_vlan_pop(struct sk_buff *skb);
3923int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3924int skb_eth_pop(struct sk_buff *skb);
3925int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3926 const unsigned char *src);
3927int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3928 int mac_len, bool ethernet);
3929int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3930 bool ethernet);
3931int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3932int skb_mpls_dec_ttl(struct sk_buff *skb);
3933struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3934 gfp_t gfp);
3935
3936static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3937{
3938 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3939}
3940
3941static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3942{
3943 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3944}
3945
3946struct skb_checksum_ops {
3947 __wsum (*update)(const void *mem, int len, __wsum wsum);
3948 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3949};
3950
3951extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3952
3953__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3954 __wsum csum, const struct skb_checksum_ops *ops);
3955__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3956 __wsum csum);
3957
3958static inline void * __must_check
3959__skb_header_pointer(const struct sk_buff *skb, int offset, int len,
3960 const void *data, int hlen, void *buffer)
3961{
3962 if (likely(hlen - offset >= len))
3963 return (void *)data + offset;
3964
3965 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
3966 return NULL;
3967
3968 return buffer;
3969}
3970
3971static inline void * __must_check
3972skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3973{
3974 return __skb_header_pointer(skb, offset, len, skb->data,
3975 skb_headlen(skb), buffer);
3976}
3977
3978/**
3979 * skb_needs_linearize - check if we need to linearize a given skb
3980 * depending on the given device features.
3981 * @skb: socket buffer to check
3982 * @features: net device features
3983 *
3984 * Returns true if either:
3985 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3986 * 2. skb is fragmented and the device does not support SG.
3987 */
3988static inline bool skb_needs_linearize(struct sk_buff *skb,
3989 netdev_features_t features)
3990{
3991 return skb_is_nonlinear(skb) &&
3992 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3993 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3994}
3995
3996static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3997 void *to,
3998 const unsigned int len)
3999{
4000 memcpy(to, skb->data, len);
4001}
4002
4003static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4004 const int offset, void *to,
4005 const unsigned int len)
4006{
4007 memcpy(to, skb->data + offset, len);
4008}
4009
4010static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4011 const void *from,
4012 const unsigned int len)
4013{
4014 memcpy(skb->data, from, len);
4015}
4016
4017static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4018 const int offset,
4019 const void *from,
4020 const unsigned int len)
4021{
4022 memcpy(skb->data + offset, from, len);
4023}
4024
4025void skb_init(void);
4026
4027static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4028{
4029 return skb->tstamp;
4030}
4031
4032/**
4033 * skb_get_timestamp - get timestamp from a skb
4034 * @skb: skb to get stamp from
4035 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4036 *
4037 * Timestamps are stored in the skb as offsets to a base timestamp.
4038 * This function converts the offset back to a struct timeval and stores
4039 * it in stamp.
4040 */
4041static inline void skb_get_timestamp(const struct sk_buff *skb,
4042 struct __kernel_old_timeval *stamp)
4043{
4044 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4045}
4046
4047static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4048 struct __kernel_sock_timeval *stamp)
4049{
4050 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4051
4052 stamp->tv_sec = ts.tv_sec;
4053 stamp->tv_usec = ts.tv_nsec / 1000;
4054}
4055
4056static inline void skb_get_timestampns(const struct sk_buff *skb,
4057 struct __kernel_old_timespec *stamp)
4058{
4059 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4060
4061 stamp->tv_sec = ts.tv_sec;
4062 stamp->tv_nsec = ts.tv_nsec;
4063}
4064
4065static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4066 struct __kernel_timespec *stamp)
4067{
4068 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4069
4070 stamp->tv_sec = ts.tv_sec;
4071 stamp->tv_nsec = ts.tv_nsec;
4072}
4073
4074static inline void __net_timestamp(struct sk_buff *skb)
4075{
4076 skb->tstamp = ktime_get_real();
4077 skb->mono_delivery_time = 0;
4078}
4079
4080static inline ktime_t net_timedelta(ktime_t t)
4081{
4082 return ktime_sub(ktime_get_real(), t);
4083}
4084
4085static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4086 bool mono)
4087{
4088 skb->tstamp = kt;
4089 skb->mono_delivery_time = kt && mono;
4090}
4091
4092DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4093
4094/* It is used in the ingress path to clear the delivery_time.
4095 * If needed, set the skb->tstamp to the (rcv) timestamp.
4096 */
4097static inline void skb_clear_delivery_time(struct sk_buff *skb)
4098{
4099 if (skb->mono_delivery_time) {
4100 skb->mono_delivery_time = 0;
4101 if (static_branch_unlikely(&netstamp_needed_key))
4102 skb->tstamp = ktime_get_real();
4103 else
4104 skb->tstamp = 0;
4105 }
4106}
4107
4108static inline void skb_clear_tstamp(struct sk_buff *skb)
4109{
4110 if (skb->mono_delivery_time)
4111 return;
4112
4113 skb->tstamp = 0;
4114}
4115
4116static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4117{
4118 if (skb->mono_delivery_time)
4119 return 0;
4120
4121 return skb->tstamp;
4122}
4123
4124static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4125{
4126 if (!skb->mono_delivery_time && skb->tstamp)
4127 return skb->tstamp;
4128
4129 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4130 return ktime_get_real();
4131
4132 return 0;
4133}
4134
4135static inline u8 skb_metadata_len(const struct sk_buff *skb)
4136{
4137 return skb_shinfo(skb)->meta_len;
4138}
4139
4140static inline void *skb_metadata_end(const struct sk_buff *skb)
4141{
4142 return skb_mac_header(skb);
4143}
4144
4145static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4146 const struct sk_buff *skb_b,
4147 u8 meta_len)
4148{
4149 const void *a = skb_metadata_end(skb_a);
4150 const void *b = skb_metadata_end(skb_b);
4151 /* Using more efficient varaiant than plain call to memcmp(). */
4152#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4153 u64 diffs = 0;
4154
4155 switch (meta_len) {
4156#define __it(x, op) (x -= sizeof(u##op))
4157#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4158 case 32: diffs |= __it_diff(a, b, 64);
4159 fallthrough;
4160 case 24: diffs |= __it_diff(a, b, 64);
4161 fallthrough;
4162 case 16: diffs |= __it_diff(a, b, 64);
4163 fallthrough;
4164 case 8: diffs |= __it_diff(a, b, 64);
4165 break;
4166 case 28: diffs |= __it_diff(a, b, 64);
4167 fallthrough;
4168 case 20: diffs |= __it_diff(a, b, 64);
4169 fallthrough;
4170 case 12: diffs |= __it_diff(a, b, 64);
4171 fallthrough;
4172 case 4: diffs |= __it_diff(a, b, 32);
4173 break;
4174 }
4175 return diffs;
4176#else
4177 return memcmp(a - meta_len, b - meta_len, meta_len);
4178#endif
4179}
4180
4181static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4182 const struct sk_buff *skb_b)
4183{
4184 u8 len_a = skb_metadata_len(skb_a);
4185 u8 len_b = skb_metadata_len(skb_b);
4186
4187 if (!(len_a | len_b))
4188 return false;
4189
4190 return len_a != len_b ?
4191 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4192}
4193
4194static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4195{
4196 skb_shinfo(skb)->meta_len = meta_len;
4197}
4198
4199static inline void skb_metadata_clear(struct sk_buff *skb)
4200{
4201 skb_metadata_set(skb, 0);
4202}
4203
4204struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4205
4206#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4207
4208void skb_clone_tx_timestamp(struct sk_buff *skb);
4209bool skb_defer_rx_timestamp(struct sk_buff *skb);
4210
4211#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4212
4213static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4214{
4215}
4216
4217static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4218{
4219 return false;
4220}
4221
4222#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4223
4224/**
4225 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4226 *
4227 * PHY drivers may accept clones of transmitted packets for
4228 * timestamping via their phy_driver.txtstamp method. These drivers
4229 * must call this function to return the skb back to the stack with a
4230 * timestamp.
4231 *
4232 * @skb: clone of the original outgoing packet
4233 * @hwtstamps: hardware time stamps
4234 *
4235 */
4236void skb_complete_tx_timestamp(struct sk_buff *skb,
4237 struct skb_shared_hwtstamps *hwtstamps);
4238
4239void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4240 struct skb_shared_hwtstamps *hwtstamps,
4241 struct sock *sk, int tstype);
4242
4243/**
4244 * skb_tstamp_tx - queue clone of skb with send time stamps
4245 * @orig_skb: the original outgoing packet
4246 * @hwtstamps: hardware time stamps, may be NULL if not available
4247 *
4248 * If the skb has a socket associated, then this function clones the
4249 * skb (thus sharing the actual data and optional structures), stores
4250 * the optional hardware time stamping information (if non NULL) or
4251 * generates a software time stamp (otherwise), then queues the clone
4252 * to the error queue of the socket. Errors are silently ignored.
4253 */
4254void skb_tstamp_tx(struct sk_buff *orig_skb,
4255 struct skb_shared_hwtstamps *hwtstamps);
4256
4257/**
4258 * skb_tx_timestamp() - Driver hook for transmit timestamping
4259 *
4260 * Ethernet MAC Drivers should call this function in their hard_xmit()
4261 * function immediately before giving the sk_buff to the MAC hardware.
4262 *
4263 * Specifically, one should make absolutely sure that this function is
4264 * called before TX completion of this packet can trigger. Otherwise
4265 * the packet could potentially already be freed.
4266 *
4267 * @skb: A socket buffer.
4268 */
4269static inline void skb_tx_timestamp(struct sk_buff *skb)
4270{
4271 skb_clone_tx_timestamp(skb);
4272 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4273 skb_tstamp_tx(skb, NULL);
4274}
4275
4276/**
4277 * skb_complete_wifi_ack - deliver skb with wifi status
4278 *
4279 * @skb: the original outgoing packet
4280 * @acked: ack status
4281 *
4282 */
4283void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4284
4285__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4286__sum16 __skb_checksum_complete(struct sk_buff *skb);
4287
4288static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4289{
4290 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4291 skb->csum_valid ||
4292 (skb->ip_summed == CHECKSUM_PARTIAL &&
4293 skb_checksum_start_offset(skb) >= 0));
4294}
4295
4296/**
4297 * skb_checksum_complete - Calculate checksum of an entire packet
4298 * @skb: packet to process
4299 *
4300 * This function calculates the checksum over the entire packet plus
4301 * the value of skb->csum. The latter can be used to supply the
4302 * checksum of a pseudo header as used by TCP/UDP. It returns the
4303 * checksum.
4304 *
4305 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4306 * this function can be used to verify that checksum on received
4307 * packets. In that case the function should return zero if the
4308 * checksum is correct. In particular, this function will return zero
4309 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4310 * hardware has already verified the correctness of the checksum.
4311 */
4312static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4313{
4314 return skb_csum_unnecessary(skb) ?
4315 0 : __skb_checksum_complete(skb);
4316}
4317
4318static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4319{
4320 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4321 if (skb->csum_level == 0)
4322 skb->ip_summed = CHECKSUM_NONE;
4323 else
4324 skb->csum_level--;
4325 }
4326}
4327
4328static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4329{
4330 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4331 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4332 skb->csum_level++;
4333 } else if (skb->ip_summed == CHECKSUM_NONE) {
4334 skb->ip_summed = CHECKSUM_UNNECESSARY;
4335 skb->csum_level = 0;
4336 }
4337}
4338
4339static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4340{
4341 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4342 skb->ip_summed = CHECKSUM_NONE;
4343 skb->csum_level = 0;
4344 }
4345}
4346
4347/* Check if we need to perform checksum complete validation.
4348 *
4349 * Returns true if checksum complete is needed, false otherwise
4350 * (either checksum is unnecessary or zero checksum is allowed).
4351 */
4352static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4353 bool zero_okay,
4354 __sum16 check)
4355{
4356 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4357 skb->csum_valid = 1;
4358 __skb_decr_checksum_unnecessary(skb);
4359 return false;
4360 }
4361
4362 return true;
4363}
4364
4365/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4366 * in checksum_init.
4367 */
4368#define CHECKSUM_BREAK 76
4369
4370/* Unset checksum-complete
4371 *
4372 * Unset checksum complete can be done when packet is being modified
4373 * (uncompressed for instance) and checksum-complete value is
4374 * invalidated.
4375 */
4376static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4377{
4378 if (skb->ip_summed == CHECKSUM_COMPLETE)
4379 skb->ip_summed = CHECKSUM_NONE;
4380}
4381
4382/* Validate (init) checksum based on checksum complete.
4383 *
4384 * Return values:
4385 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4386 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4387 * checksum is stored in skb->csum for use in __skb_checksum_complete
4388 * non-zero: value of invalid checksum
4389 *
4390 */
4391static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4392 bool complete,
4393 __wsum psum)
4394{
4395 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4396 if (!csum_fold(csum_add(psum, skb->csum))) {
4397 skb->csum_valid = 1;
4398 return 0;
4399 }
4400 }
4401
4402 skb->csum = psum;
4403
4404 if (complete || skb->len <= CHECKSUM_BREAK) {
4405 __sum16 csum;
4406
4407 csum = __skb_checksum_complete(skb);
4408 skb->csum_valid = !csum;
4409 return csum;
4410 }
4411
4412 return 0;
4413}
4414
4415static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4416{
4417 return 0;
4418}
4419
4420/* Perform checksum validate (init). Note that this is a macro since we only
4421 * want to calculate the pseudo header which is an input function if necessary.
4422 * First we try to validate without any computation (checksum unnecessary) and
4423 * then calculate based on checksum complete calling the function to compute
4424 * pseudo header.
4425 *
4426 * Return values:
4427 * 0: checksum is validated or try to in skb_checksum_complete
4428 * non-zero: value of invalid checksum
4429 */
4430#define __skb_checksum_validate(skb, proto, complete, \
4431 zero_okay, check, compute_pseudo) \
4432({ \
4433 __sum16 __ret = 0; \
4434 skb->csum_valid = 0; \
4435 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4436 __ret = __skb_checksum_validate_complete(skb, \
4437 complete, compute_pseudo(skb, proto)); \
4438 __ret; \
4439})
4440
4441#define skb_checksum_init(skb, proto, compute_pseudo) \
4442 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4443
4444#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4445 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4446
4447#define skb_checksum_validate(skb, proto, compute_pseudo) \
4448 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4449
4450#define skb_checksum_validate_zero_check(skb, proto, check, \
4451 compute_pseudo) \
4452 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4453
4454#define skb_checksum_simple_validate(skb) \
4455 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4456
4457static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4458{
4459 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4460}
4461
4462static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4463{
4464 skb->csum = ~pseudo;
4465 skb->ip_summed = CHECKSUM_COMPLETE;
4466}
4467
4468#define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4469do { \
4470 if (__skb_checksum_convert_check(skb)) \
4471 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4472} while (0)
4473
4474static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4475 u16 start, u16 offset)
4476{
4477 skb->ip_summed = CHECKSUM_PARTIAL;
4478 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4479 skb->csum_offset = offset - start;
4480}
4481
4482/* Update skbuf and packet to reflect the remote checksum offload operation.
4483 * When called, ptr indicates the starting point for skb->csum when
4484 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4485 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4486 */
4487static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4488 int start, int offset, bool nopartial)
4489{
4490 __wsum delta;
4491
4492 if (!nopartial) {
4493 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4494 return;
4495 }
4496
4497 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4498 __skb_checksum_complete(skb);
4499 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4500 }
4501
4502 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4503
4504 /* Adjust skb->csum since we changed the packet */
4505 skb->csum = csum_add(skb->csum, delta);
4506}
4507
4508static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4509{
4510#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4511 return (void *)(skb->_nfct & NFCT_PTRMASK);
4512#else
4513 return NULL;
4514#endif
4515}
4516
4517static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4518{
4519#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4520 return skb->_nfct;
4521#else
4522 return 0UL;
4523#endif
4524}
4525
4526static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4527{
4528#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4529 skb->slow_gro |= !!nfct;
4530 skb->_nfct = nfct;
4531#endif
4532}
4533
4534#ifdef CONFIG_SKB_EXTENSIONS
4535enum skb_ext_id {
4536#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4537 SKB_EXT_BRIDGE_NF,
4538#endif
4539#ifdef CONFIG_XFRM
4540 SKB_EXT_SEC_PATH,
4541#endif
4542#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4543 TC_SKB_EXT,
4544#endif
4545#if IS_ENABLED(CONFIG_MPTCP)
4546 SKB_EXT_MPTCP,
4547#endif
4548#if IS_ENABLED(CONFIG_MCTP_FLOWS)
4549 SKB_EXT_MCTP,
4550#endif
4551 SKB_EXT_NUM, /* must be last */
4552};
4553
4554/**
4555 * struct skb_ext - sk_buff extensions
4556 * @refcnt: 1 on allocation, deallocated on 0
4557 * @offset: offset to add to @data to obtain extension address
4558 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4559 * @data: start of extension data, variable sized
4560 *
4561 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4562 * to use 'u8' types while allowing up to 2kb worth of extension data.
4563 */
4564struct skb_ext {
4565 refcount_t refcnt;
4566 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4567 u8 chunks; /* same */
4568 char data[] __aligned(8);
4569};
4570
4571struct skb_ext *__skb_ext_alloc(gfp_t flags);
4572void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4573 struct skb_ext *ext);
4574void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4575void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4576void __skb_ext_put(struct skb_ext *ext);
4577
4578static inline void skb_ext_put(struct sk_buff *skb)
4579{
4580 if (skb->active_extensions)
4581 __skb_ext_put(skb->extensions);
4582}
4583
4584static inline void __skb_ext_copy(struct sk_buff *dst,
4585 const struct sk_buff *src)
4586{
4587 dst->active_extensions = src->active_extensions;
4588
4589 if (src->active_extensions) {
4590 struct skb_ext *ext = src->extensions;
4591
4592 refcount_inc(&ext->refcnt);
4593 dst->extensions = ext;
4594 }
4595}
4596
4597static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4598{
4599 skb_ext_put(dst);
4600 __skb_ext_copy(dst, src);
4601}
4602
4603static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4604{
4605 return !!ext->offset[i];
4606}
4607
4608static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4609{
4610 return skb->active_extensions & (1 << id);
4611}
4612
4613static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4614{
4615 if (skb_ext_exist(skb, id))
4616 __skb_ext_del(skb, id);
4617}
4618
4619static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4620{
4621 if (skb_ext_exist(skb, id)) {
4622 struct skb_ext *ext = skb->extensions;
4623
4624 return (void *)ext + (ext->offset[id] << 3);
4625 }
4626
4627 return NULL;
4628}
4629
4630static inline void skb_ext_reset(struct sk_buff *skb)
4631{
4632 if (unlikely(skb->active_extensions)) {
4633 __skb_ext_put(skb->extensions);
4634 skb->active_extensions = 0;
4635 }
4636}
4637
4638static inline bool skb_has_extensions(struct sk_buff *skb)
4639{
4640 return unlikely(skb->active_extensions);
4641}
4642#else
4643static inline void skb_ext_put(struct sk_buff *skb) {}
4644static inline void skb_ext_reset(struct sk_buff *skb) {}
4645static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4646static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4647static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4648static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4649#endif /* CONFIG_SKB_EXTENSIONS */
4650
4651static inline void nf_reset_ct(struct sk_buff *skb)
4652{
4653#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4654 nf_conntrack_put(skb_nfct(skb));
4655 skb->_nfct = 0;
4656#endif
4657}
4658
4659static inline void nf_reset_trace(struct sk_buff *skb)
4660{
4661#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4662 skb->nf_trace = 0;
4663#endif
4664}
4665
4666static inline void ipvs_reset(struct sk_buff *skb)
4667{
4668#if IS_ENABLED(CONFIG_IP_VS)
4669 skb->ipvs_property = 0;
4670#endif
4671}
4672
4673/* Note: This doesn't put any conntrack info in dst. */
4674static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4675 bool copy)
4676{
4677#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4678 dst->_nfct = src->_nfct;
4679 nf_conntrack_get(skb_nfct(src));
4680#endif
4681#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4682 if (copy)
4683 dst->nf_trace = src->nf_trace;
4684#endif
4685}
4686
4687static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4688{
4689#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4690 nf_conntrack_put(skb_nfct(dst));
4691#endif
4692 dst->slow_gro = src->slow_gro;
4693 __nf_copy(dst, src, true);
4694}
4695
4696#ifdef CONFIG_NETWORK_SECMARK
4697static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4698{
4699 to->secmark = from->secmark;
4700}
4701
4702static inline void skb_init_secmark(struct sk_buff *skb)
4703{
4704 skb->secmark = 0;
4705}
4706#else
4707static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4708{ }
4709
4710static inline void skb_init_secmark(struct sk_buff *skb)
4711{ }
4712#endif
4713
4714static inline int secpath_exists(const struct sk_buff *skb)
4715{
4716#ifdef CONFIG_XFRM
4717 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4718#else
4719 return 0;
4720#endif
4721}
4722
4723static inline bool skb_irq_freeable(const struct sk_buff *skb)
4724{
4725 return !skb->destructor &&
4726 !secpath_exists(skb) &&
4727 !skb_nfct(skb) &&
4728 !skb->_skb_refdst &&
4729 !skb_has_frag_list(skb);
4730}
4731
4732static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4733{
4734 skb->queue_mapping = queue_mapping;
4735}
4736
4737static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4738{
4739 return skb->queue_mapping;
4740}
4741
4742static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4743{
4744 to->queue_mapping = from->queue_mapping;
4745}
4746
4747static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4748{
4749 skb->queue_mapping = rx_queue + 1;
4750}
4751
4752static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4753{
4754 return skb->queue_mapping - 1;
4755}
4756
4757static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4758{
4759 return skb->queue_mapping != 0;
4760}
4761
4762static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4763{
4764 skb->dst_pending_confirm = val;
4765}
4766
4767static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4768{
4769 return skb->dst_pending_confirm != 0;
4770}
4771
4772static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4773{
4774#ifdef CONFIG_XFRM
4775 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4776#else
4777 return NULL;
4778#endif
4779}
4780
4781/* Keeps track of mac header offset relative to skb->head.
4782 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4783 * For non-tunnel skb it points to skb_mac_header() and for
4784 * tunnel skb it points to outer mac header.
4785 * Keeps track of level of encapsulation of network headers.
4786 */
4787struct skb_gso_cb {
4788 union {
4789 int mac_offset;
4790 int data_offset;
4791 };
4792 int encap_level;
4793 __wsum csum;
4794 __u16 csum_start;
4795};
4796#define SKB_GSO_CB_OFFSET 32
4797#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4798
4799static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4800{
4801 return (skb_mac_header(inner_skb) - inner_skb->head) -
4802 SKB_GSO_CB(inner_skb)->mac_offset;
4803}
4804
4805static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4806{
4807 int new_headroom, headroom;
4808 int ret;
4809
4810 headroom = skb_headroom(skb);
4811 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4812 if (ret)
4813 return ret;
4814
4815 new_headroom = skb_headroom(skb);
4816 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4817 return 0;
4818}
4819
4820static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4821{
4822 /* Do not update partial checksums if remote checksum is enabled. */
4823 if (skb->remcsum_offload)
4824 return;
4825
4826