Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 <net/checksum.h>
27#include <linux/rcupdate.h>
28#include <linux/dma-mapping.h>
29#include <linux/netdev_features.h>
30#include <net/flow_dissector.h>
31#include <linux/in6.h>
32#include <linux/if_packet.h>
33#include <linux/llist.h>
34#include <net/flow.h>
35#if IS_ENABLED(CONFIG_NF_CONNTRACK)
36#include <linux/netfilter/nf_conntrack_common.h>
37#endif
38#include <net/net_debug.h>
39#include <net/dropreason-core.h>
40#include <net/netmem.h>
41
42/**
43 * DOC: skb checksums
44 *
45 * The interface for checksum offload between the stack and networking drivers
46 * is as follows...
47 *
48 * IP checksum related features
49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
50 *
51 * Drivers advertise checksum offload capabilities in the features of a device.
52 * From the stack's point of view these are capabilities offered by the driver.
53 * A driver typically only advertises features that it is capable of offloading
54 * to its device.
55 *
56 * .. flat-table:: Checksum related device features
57 * :widths: 1 10
58 *
59 * * - %NETIF_F_HW_CSUM
60 * - The driver (or its device) is able to compute one
61 * IP (one's complement) checksum for any combination
62 * of protocols or protocol layering. The checksum is
63 * computed and set in a packet per the CHECKSUM_PARTIAL
64 * interface (see below).
65 *
66 * * - %NETIF_F_IP_CSUM
67 * - Driver (device) is only able to checksum plain
68 * TCP or UDP packets over IPv4. These are specifically
69 * unencapsulated packets of the form IPv4|TCP or
70 * IPv4|UDP where the Protocol field in the IPv4 header
71 * is TCP or UDP. The IPv4 header may contain IP options.
72 * This feature cannot be set in features for a device
73 * with NETIF_F_HW_CSUM also set. This feature is being
74 * DEPRECATED (see below).
75 *
76 * * - %NETIF_F_IPV6_CSUM
77 * - Driver (device) is only able to checksum plain
78 * TCP or UDP packets over IPv6. These are specifically
79 * unencapsulated packets of the form IPv6|TCP or
80 * IPv6|UDP where the Next Header field in the IPv6
81 * header is either TCP or UDP. IPv6 extension headers
82 * are not supported with this feature. This feature
83 * cannot be set in features for a device with
84 * NETIF_F_HW_CSUM also set. This feature is being
85 * DEPRECATED (see below).
86 *
87 * * - %NETIF_F_RXCSUM
88 * - Driver (device) performs receive checksum offload.
89 * This flag is only used to disable the RX checksum
90 * feature for a device. The stack will accept receive
91 * checksum indication in packets received on a device
92 * regardless of whether NETIF_F_RXCSUM is set.
93 *
94 * Checksumming of received packets by device
95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
96 *
97 * Indication of checksum verification is set in &sk_buff.ip_summed.
98 * Possible values are:
99 *
100 * - %CHECKSUM_NONE
101 *
102 * Device did not checksum this packet e.g. due to lack of capabilities.
103 * The packet contains full (though not verified) checksum in packet but
104 * not in skb->csum. Thus, skb->csum is undefined in this case.
105 *
106 * - %CHECKSUM_UNNECESSARY
107 *
108 * The hardware you're dealing with doesn't calculate the full checksum
109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
111 * if their checksums are okay. &sk_buff.csum is still undefined in this case
112 * though. A driver or device must never modify the checksum field in the
113 * packet even if checksum is verified.
114 *
115 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
116 *
117 * - TCP: IPv6 and IPv4.
118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
119 * zero UDP checksum for either IPv4 or IPv6, the networking stack
120 * may perform further validation in this case.
121 * - GRE: only if the checksum is present in the header.
122 * - SCTP: indicates the CRC in SCTP header has been validated.
123 * - FCOE: indicates the CRC in FC frame has been validated.
124 *
125 * &sk_buff.csum_level indicates the number of consecutive checksums found in
126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
128 * and a device is able to verify the checksums for UDP (possibly zero),
129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
130 * two. If the device were only able to verify the UDP checksum and not
131 * GRE, either because it doesn't support GRE checksum or because GRE
132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
133 * not considered in this case).
134 *
135 * - %CHECKSUM_COMPLETE
136 *
137 * This is the most generic way. The device supplied checksum of the _whole_
138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
139 * hardware doesn't need to parse L3/L4 headers to implement this.
140 *
141 * Notes:
142 *
143 * - Even if device supports only some protocols, but is able to produce
144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
146 *
147 * - %CHECKSUM_PARTIAL
148 *
149 * A checksum is set up to be offloaded to a device as described in the
150 * output description for CHECKSUM_PARTIAL. This may occur on a packet
151 * received directly from another Linux OS, e.g., a virtualized Linux kernel
152 * on the same host, or it may be set in the input path in GRO or remote
153 * checksum offload. For the purposes of checksum verification, the checksum
154 * referred to by skb->csum_start + skb->csum_offset and any preceding
155 * checksums in the packet are considered verified. Any checksums in the
156 * packet that are after the checksum being offloaded are not considered to
157 * be verified.
158 *
159 * Checksumming on transmit for non-GSO
160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
161 *
162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
163 * Values are:
164 *
165 * - %CHECKSUM_PARTIAL
166 *
167 * The driver is required to checksum the packet as seen by hard_start_xmit()
168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
169 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
170 * A driver may verify that the
171 * csum_start and csum_offset values are valid values given the length and
172 * offset of the packet, but it should not attempt to validate that the
173 * checksum refers to a legitimate transport layer checksum -- it is the
174 * purview of the stack to validate that csum_start and csum_offset are set
175 * correctly.
176 *
177 * When the stack requests checksum offload for a packet, the driver MUST
178 * ensure that the checksum is set correctly. A driver can either offload the
179 * checksum calculation to the device, or call skb_checksum_help (in the case
180 * that the device does not support offload for a particular checksum).
181 *
182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
184 * checksum offload capability.
185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
186 * on network device checksumming capabilities: if a packet does not match
187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
188 * &sk_buff.csum_not_inet, see :ref:`crc`)
189 * is called to resolve the checksum.
190 *
191 * - %CHECKSUM_NONE
192 *
193 * The skb was already checksummed by the protocol, or a checksum is not
194 * required.
195 *
196 * - %CHECKSUM_UNNECESSARY
197 *
198 * This has the same meaning as CHECKSUM_NONE for checksum offload on
199 * output.
200 *
201 * - %CHECKSUM_COMPLETE
202 *
203 * Not used in checksum output. If a driver observes a packet with this value
204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
205 *
206 * .. _crc:
207 *
208 * Non-IP checksum (CRC) offloads
209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
210 *
211 * .. flat-table::
212 * :widths: 1 10
213 *
214 * * - %NETIF_F_SCTP_CRC
215 * - This feature indicates that a device is capable of
216 * offloading the SCTP CRC in a packet. To perform this offload the stack
217 * will set csum_start and csum_offset accordingly, set ip_summed to
218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
220 * A driver that supports both IP checksum offload and SCTP CRC32c offload
221 * must verify which offload is configured for a packet by testing the
222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
224 *
225 * * - %NETIF_F_FCOE_CRC
226 * - This feature indicates that a device is capable of offloading the FCOE
227 * CRC in a packet. To perform this offload the stack will set ip_summed
228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
229 * accordingly. Note that there is no indication in the skbuff that the
230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
231 * both IP checksum offload and FCOE CRC offload must verify which offload
232 * is configured for a packet, presumably by inspecting packet headers.
233 *
234 * Checksumming on output with GSO
235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
236 *
237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
240 * part of the GSO operation is implied. If a checksum is being offloaded
241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
242 * csum_offset are set to refer to the outermost checksum being offloaded
243 * (two offloaded checksums are possible with UDP encapsulation).
244 */
245
246/* Don't change this without changing skb_csum_unnecessary! */
247#define CHECKSUM_NONE 0
248#define CHECKSUM_UNNECESSARY 1
249#define CHECKSUM_COMPLETE 2
250#define CHECKSUM_PARTIAL 3
251
252/* Maximum value in skb->csum_level */
253#define SKB_MAX_CSUM_LEVEL 3
254
255#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
256#define SKB_WITH_OVERHEAD(X) \
257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
258
259/* For X bytes available in skb->head, what is the minimal
260 * allocation needed, knowing struct skb_shared_info needs
261 * to be aligned.
262 */
263#define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \
264 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
265
266#define SKB_MAX_ORDER(X, ORDER) \
267 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
268#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
269#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
270
271/* return minimum truesize of one skb containing X bytes of data */
272#define SKB_TRUESIZE(X) ((X) + \
273 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
274 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
275
276struct ahash_request;
277struct net_device;
278struct scatterlist;
279struct pipe_inode_info;
280struct iov_iter;
281struct napi_struct;
282struct bpf_prog;
283union bpf_attr;
284struct skb_ext;
285struct ts_config;
286
287#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
288struct nf_bridge_info {
289 enum {
290 BRNF_PROTO_UNCHANGED,
291 BRNF_PROTO_8021Q,
292 BRNF_PROTO_PPPOE
293 } orig_proto:8;
294 u8 pkt_otherhost:1;
295 u8 in_prerouting:1;
296 u8 bridged_dnat:1;
297 u8 sabotage_in_done:1;
298 __u16 frag_max_size;
299 int physinif;
300
301 /* always valid & non-NULL from FORWARD on, for physdev match */
302 struct net_device *physoutdev;
303 union {
304 /* prerouting: detect dnat in orig/reply direction */
305 __be32 ipv4_daddr;
306 struct in6_addr ipv6_daddr;
307
308 /* after prerouting + nat detected: store original source
309 * mac since neigh resolution overwrites it, only used while
310 * skb is out in neigh layer.
311 */
312 char neigh_header[8];
313 };
314};
315#endif
316
317#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
318/* Chain in tc_skb_ext will be used to share the tc chain with
319 * ovs recirc_id. It will be set to the current chain by tc
320 * and read by ovs to recirc_id.
321 */
322struct tc_skb_ext {
323 union {
324 u64 act_miss_cookie;
325 __u32 chain;
326 };
327 __u16 mru;
328 __u16 zone;
329 u8 post_ct:1;
330 u8 post_ct_snat:1;
331 u8 post_ct_dnat:1;
332 u8 act_miss:1; /* Set if act_miss_cookie is used */
333 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */
334};
335#endif
336
337struct sk_buff_head {
338 /* These two members must be first to match sk_buff. */
339 struct_group_tagged(sk_buff_list, list,
340 struct sk_buff *next;
341 struct sk_buff *prev;
342 );
343
344 __u32 qlen;
345 spinlock_t lock;
346};
347
348struct sk_buff;
349
350#ifndef CONFIG_MAX_SKB_FRAGS
351# define CONFIG_MAX_SKB_FRAGS 17
352#endif
353
354#define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS
355
356extern int sysctl_max_skb_frags;
357
358/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
359 * segment using its current segmentation instead.
360 */
361#define GSO_BY_FRAGS 0xFFFF
362
363typedef struct skb_frag {
364 netmem_ref netmem;
365 unsigned int len;
366 unsigned int offset;
367} skb_frag_t;
368
369/**
370 * skb_frag_size() - Returns the size of a skb fragment
371 * @frag: skb fragment
372 */
373static inline unsigned int skb_frag_size(const skb_frag_t *frag)
374{
375 return frag->len;
376}
377
378/**
379 * skb_frag_size_set() - Sets the size of a skb fragment
380 * @frag: skb fragment
381 * @size: size of fragment
382 */
383static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
384{
385 frag->len = size;
386}
387
388/**
389 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
390 * @frag: skb fragment
391 * @delta: value to add
392 */
393static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
394{
395 frag->len += delta;
396}
397
398/**
399 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
400 * @frag: skb fragment
401 * @delta: value to subtract
402 */
403static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
404{
405 frag->len -= delta;
406}
407
408/**
409 * skb_frag_must_loop - Test if %p is a high memory page
410 * @p: fragment's page
411 */
412static inline bool skb_frag_must_loop(struct page *p)
413{
414#if defined(CONFIG_HIGHMEM)
415 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
416 return true;
417#endif
418 return false;
419}
420
421/**
422 * skb_frag_foreach_page - loop over pages in a fragment
423 *
424 * @f: skb frag to operate on
425 * @f_off: offset from start of f->netmem
426 * @f_len: length from f_off to loop over
427 * @p: (temp var) current page
428 * @p_off: (temp var) offset from start of current page,
429 * non-zero only on first page.
430 * @p_len: (temp var) length in current page,
431 * < PAGE_SIZE only on first and last page.
432 * @copied: (temp var) length so far, excluding current p_len.
433 *
434 * A fragment can hold a compound page, in which case per-page
435 * operations, notably kmap_atomic, must be called for each
436 * regular page.
437 */
438#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
439 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
440 p_off = (f_off) & (PAGE_SIZE - 1), \
441 p_len = skb_frag_must_loop(p) ? \
442 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
443 copied = 0; \
444 copied < f_len; \
445 copied += p_len, p++, p_off = 0, \
446 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
447
448/**
449 * struct skb_shared_hwtstamps - hardware time stamps
450 * @hwtstamp: hardware time stamp transformed into duration
451 * since arbitrary point in time
452 * @netdev_data: address/cookie of network device driver used as
453 * reference to actual hardware time stamp
454 *
455 * Software time stamps generated by ktime_get_real() are stored in
456 * skb->tstamp.
457 *
458 * hwtstamps can only be compared against other hwtstamps from
459 * the same device.
460 *
461 * This structure is attached to packets as part of the
462 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
463 */
464struct skb_shared_hwtstamps {
465 union {
466 ktime_t hwtstamp;
467 void *netdev_data;
468 };
469};
470
471/* Definitions for tx_flags in struct skb_shared_info */
472enum {
473 /* generate hardware time stamp */
474 SKBTX_HW_TSTAMP = 1 << 0,
475
476 /* generate software time stamp when queueing packet to NIC */
477 SKBTX_SW_TSTAMP = 1 << 1,
478
479 /* device driver is going to provide hardware time stamp */
480 SKBTX_IN_PROGRESS = 1 << 2,
481
482 /* generate hardware time stamp based on cycles if supported */
483 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
484
485 /* generate wifi status information (where possible) */
486 SKBTX_WIFI_STATUS = 1 << 4,
487
488 /* determine hardware time stamp based on time or cycles */
489 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
490
491 /* generate software time stamp when entering packet scheduling */
492 SKBTX_SCHED_TSTAMP = 1 << 6,
493};
494
495#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
496 SKBTX_SCHED_TSTAMP)
497#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
498 SKBTX_HW_TSTAMP_USE_CYCLES | \
499 SKBTX_ANY_SW_TSTAMP)
500
501/* Definitions for flags in struct skb_shared_info */
502enum {
503 /* use zcopy routines */
504 SKBFL_ZEROCOPY_ENABLE = BIT(0),
505
506 /* This indicates at least one fragment might be overwritten
507 * (as in vmsplice(), sendfile() ...)
508 * If we need to compute a TX checksum, we'll need to copy
509 * all frags to avoid possible bad checksum
510 */
511 SKBFL_SHARED_FRAG = BIT(1),
512
513 /* segment contains only zerocopy data and should not be
514 * charged to the kernel memory.
515 */
516 SKBFL_PURE_ZEROCOPY = BIT(2),
517
518 SKBFL_DONT_ORPHAN = BIT(3),
519
520 /* page references are managed by the ubuf_info, so it's safe to
521 * use frags only up until ubuf_info is released
522 */
523 SKBFL_MANAGED_FRAG_REFS = BIT(4),
524};
525
526#define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
527#define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
528 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
529
530/*
531 * The callback notifies userspace to release buffers when skb DMA is done in
532 * lower device, the skb last reference should be 0 when calling this.
533 * The zerocopy_success argument is true if zero copy transmit occurred,
534 * false on data copy or out of memory error caused by data copy attempt.
535 * The ctx field is used to track device context.
536 * The desc field is used to track userspace buffer index.
537 */
538struct ubuf_info {
539 void (*callback)(struct sk_buff *, struct ubuf_info *,
540 bool zerocopy_success);
541 refcount_t refcnt;
542 u8 flags;
543};
544
545struct ubuf_info_msgzc {
546 struct ubuf_info ubuf;
547
548 union {
549 struct {
550 unsigned long desc;
551 void *ctx;
552 };
553 struct {
554 u32 id;
555 u16 len;
556 u16 zerocopy:1;
557 u32 bytelen;
558 };
559 };
560
561 struct mmpin {
562 struct user_struct *user;
563 unsigned int num_pg;
564 } mmp;
565};
566
567#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
568#define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
569 ubuf)
570
571int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
572void mm_unaccount_pinned_pages(struct mmpin *mmp);
573
574/* Preserve some data across TX submission and completion.
575 *
576 * Note, this state is stored in the driver. Extending the layout
577 * might need some special care.
578 */
579struct xsk_tx_metadata_compl {
580 __u64 *tx_timestamp;
581};
582
583/* This data is invariant across clones and lives at
584 * the end of the header data, ie. at skb->end.
585 */
586struct skb_shared_info {
587 __u8 flags;
588 __u8 meta_len;
589 __u8 nr_frags;
590 __u8 tx_flags;
591 unsigned short gso_size;
592 /* Warning: this field is not always filled in (UFO)! */
593 unsigned short gso_segs;
594 struct sk_buff *frag_list;
595 union {
596 struct skb_shared_hwtstamps hwtstamps;
597 struct xsk_tx_metadata_compl xsk_meta;
598 };
599 unsigned int gso_type;
600 u32 tskey;
601
602 /*
603 * Warning : all fields before dataref are cleared in __alloc_skb()
604 */
605 atomic_t dataref;
606 unsigned int xdp_frags_size;
607
608 /* Intermediate layers must ensure that destructor_arg
609 * remains valid until skb destructor */
610 void * destructor_arg;
611
612 /* must be last field, see pskb_expand_head() */
613 skb_frag_t frags[MAX_SKB_FRAGS];
614};
615
616/**
617 * DOC: dataref and headerless skbs
618 *
619 * Transport layers send out clones of payload skbs they hold for
620 * retransmissions. To allow lower layers of the stack to prepend their headers
621 * we split &skb_shared_info.dataref into two halves.
622 * The lower 16 bits count the overall number of references.
623 * The higher 16 bits indicate how many of the references are payload-only.
624 * skb_header_cloned() checks if skb is allowed to add / write the headers.
625 *
626 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
627 * (via __skb_header_release()). Any clone created from marked skb will get
628 * &sk_buff.hdr_len populated with the available headroom.
629 * If there's the only clone in existence it's able to modify the headroom
630 * at will. The sequence of calls inside the transport layer is::
631 *
632 * <alloc skb>
633 * skb_reserve()
634 * __skb_header_release()
635 * skb_clone()
636 * // send the clone down the stack
637 *
638 * This is not a very generic construct and it depends on the transport layers
639 * doing the right thing. In practice there's usually only one payload-only skb.
640 * Having multiple payload-only skbs with different lengths of hdr_len is not
641 * possible. The payload-only skbs should never leave their owner.
642 */
643#define SKB_DATAREF_SHIFT 16
644#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
645
646
647enum {
648 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
649 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
650 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
651};
652
653enum {
654 SKB_GSO_TCPV4 = 1 << 0,
655
656 /* This indicates the skb is from an untrusted source. */
657 SKB_GSO_DODGY = 1 << 1,
658
659 /* This indicates the tcp segment has CWR set. */
660 SKB_GSO_TCP_ECN = 1 << 2,
661
662 SKB_GSO_TCP_FIXEDID = 1 << 3,
663
664 SKB_GSO_TCPV6 = 1 << 4,
665
666 SKB_GSO_FCOE = 1 << 5,
667
668 SKB_GSO_GRE = 1 << 6,
669
670 SKB_GSO_GRE_CSUM = 1 << 7,
671
672 SKB_GSO_IPXIP4 = 1 << 8,
673
674 SKB_GSO_IPXIP6 = 1 << 9,
675
676 SKB_GSO_UDP_TUNNEL = 1 << 10,
677
678 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
679
680 SKB_GSO_PARTIAL = 1 << 12,
681
682 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
683
684 SKB_GSO_SCTP = 1 << 14,
685
686 SKB_GSO_ESP = 1 << 15,
687
688 SKB_GSO_UDP = 1 << 16,
689
690 SKB_GSO_UDP_L4 = 1 << 17,
691
692 SKB_GSO_FRAGLIST = 1 << 18,
693};
694
695#if BITS_PER_LONG > 32
696#define NET_SKBUFF_DATA_USES_OFFSET 1
697#endif
698
699#ifdef NET_SKBUFF_DATA_USES_OFFSET
700typedef unsigned int sk_buff_data_t;
701#else
702typedef unsigned char *sk_buff_data_t;
703#endif
704
705/**
706 * DOC: Basic sk_buff geometry
707 *
708 * struct sk_buff itself is a metadata structure and does not hold any packet
709 * data. All the data is held in associated buffers.
710 *
711 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
712 * into two parts:
713 *
714 * - data buffer, containing headers and sometimes payload;
715 * this is the part of the skb operated on by the common helpers
716 * such as skb_put() or skb_pull();
717 * - shared info (struct skb_shared_info) which holds an array of pointers
718 * to read-only data in the (page, offset, length) format.
719 *
720 * Optionally &skb_shared_info.frag_list may point to another skb.
721 *
722 * Basic diagram may look like this::
723 *
724 * ---------------
725 * | sk_buff |
726 * ---------------
727 * ,--------------------------- + head
728 * / ,----------------- + data
729 * / / ,----------- + tail
730 * | | | , + end
731 * | | | |
732 * v v v v
733 * -----------------------------------------------
734 * | headroom | data | tailroom | skb_shared_info |
735 * -----------------------------------------------
736 * + [page frag]
737 * + [page frag]
738 * + [page frag]
739 * + [page frag] ---------
740 * + frag_list --> | sk_buff |
741 * ---------
742 *
743 */
744
745/**
746 * struct sk_buff - socket buffer
747 * @next: Next buffer in list
748 * @prev: Previous buffer in list
749 * @tstamp: Time we arrived/left
750 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
751 * for retransmit timer
752 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
753 * @list: queue head
754 * @ll_node: anchor in an llist (eg socket defer_list)
755 * @sk: Socket we are owned by
756 * @dev: Device we arrived on/are leaving by
757 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
758 * @cb: Control buffer. Free for use by every layer. Put private vars here
759 * @_skb_refdst: destination entry (with norefcount bit)
760 * @len: Length of actual data
761 * @data_len: Data length
762 * @mac_len: Length of link layer header
763 * @hdr_len: writable header length of cloned skb
764 * @csum: Checksum (must include start/offset pair)
765 * @csum_start: Offset from skb->head where checksumming should start
766 * @csum_offset: Offset from csum_start where checksum should be stored
767 * @priority: Packet queueing priority
768 * @ignore_df: allow local fragmentation
769 * @cloned: Head may be cloned (check refcnt to be sure)
770 * @ip_summed: Driver fed us an IP checksum
771 * @nohdr: Payload reference only, must not modify header
772 * @pkt_type: Packet class
773 * @fclone: skbuff clone status
774 * @ipvs_property: skbuff is owned by ipvs
775 * @inner_protocol_type: whether the inner protocol is
776 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
777 * @remcsum_offload: remote checksum offload is enabled
778 * @offload_fwd_mark: Packet was L2-forwarded in hardware
779 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
780 * @tc_skip_classify: do not classify packet. set by IFB device
781 * @tc_at_ingress: used within tc_classify to distinguish in/egress
782 * @redirected: packet was redirected by packet classifier
783 * @from_ingress: packet was redirected from the ingress path
784 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
785 * @peeked: this packet has been seen already, so stats have been
786 * done for it, don't do them again
787 * @nf_trace: netfilter packet trace flag
788 * @protocol: Packet protocol from driver
789 * @destructor: Destruct function
790 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
791 * @_sk_redir: socket redirection information for skmsg
792 * @_nfct: Associated connection, if any (with nfctinfo bits)
793 * @skb_iif: ifindex of device we arrived on
794 * @tc_index: Traffic control index
795 * @hash: the packet hash
796 * @queue_mapping: Queue mapping for multiqueue devices
797 * @head_frag: skb was allocated from page fragments,
798 * not allocated by kmalloc() or vmalloc().
799 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
800 * @pp_recycle: mark the packet for recycling instead of freeing (implies
801 * page_pool support on driver)
802 * @active_extensions: active extensions (skb_ext_id types)
803 * @ndisc_nodetype: router type (from link layer)
804 * @ooo_okay: allow the mapping of a socket to a queue to be changed
805 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
806 * ports.
807 * @sw_hash: indicates hash was computed in software stack
808 * @wifi_acked_valid: wifi_acked was set
809 * @wifi_acked: whether frame was acked on wifi or not
810 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
811 * @encapsulation: indicates the inner headers in the skbuff are valid
812 * @encap_hdr_csum: software checksum is needed
813 * @csum_valid: checksum is already valid
814 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
815 * @csum_complete_sw: checksum was completed by software
816 * @csum_level: indicates the number of consecutive checksums found in
817 * the packet minus one that have been verified as
818 * CHECKSUM_UNNECESSARY (max 3)
819 * @dst_pending_confirm: need to confirm neighbour
820 * @decrypted: Decrypted SKB
821 * @slow_gro: state present at GRO time, slower prepare step required
822 * @mono_delivery_time: When set, skb->tstamp has the
823 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
824 * skb->tstamp has the (rcv) timestamp at ingress and
825 * delivery_time at egress.
826 * @napi_id: id of the NAPI struct this skb came from
827 * @sender_cpu: (aka @napi_id) source CPU in XPS
828 * @alloc_cpu: CPU which did the skb allocation.
829 * @secmark: security marking
830 * @mark: Generic packet mark
831 * @reserved_tailroom: (aka @mark) number of bytes of free space available
832 * at the tail of an sk_buff
833 * @vlan_all: vlan fields (proto & tci)
834 * @vlan_proto: vlan encapsulation protocol
835 * @vlan_tci: vlan tag control information
836 * @inner_protocol: Protocol (encapsulation)
837 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
838 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
839 * @inner_transport_header: Inner transport layer header (encapsulation)
840 * @inner_network_header: Network layer header (encapsulation)
841 * @inner_mac_header: Link layer header (encapsulation)
842 * @transport_header: Transport layer header
843 * @network_header: Network layer header
844 * @mac_header: Link layer header
845 * @kcov_handle: KCOV remote handle for remote coverage collection
846 * @tail: Tail pointer
847 * @end: End pointer
848 * @head: Head of buffer
849 * @data: Data head pointer
850 * @truesize: Buffer size
851 * @users: User count - see {datagram,tcp}.c
852 * @extensions: allocated extensions, valid if active_extensions is nonzero
853 */
854
855struct sk_buff {
856 union {
857 struct {
858 /* These two members must be first to match sk_buff_head. */
859 struct sk_buff *next;
860 struct sk_buff *prev;
861
862 union {
863 struct net_device *dev;
864 /* Some protocols might use this space to store information,
865 * while device pointer would be NULL.
866 * UDP receive path is one user.
867 */
868 unsigned long dev_scratch;
869 };
870 };
871 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
872 struct list_head list;
873 struct llist_node ll_node;
874 };
875
876 struct sock *sk;
877
878 union {
879 ktime_t tstamp;
880 u64 skb_mstamp_ns; /* earliest departure time */
881 };
882 /*
883 * This is the control buffer. It is free to use for every
884 * layer. Please put your private variables there. If you
885 * want to keep them across layers you have to do a skb_clone()
886 * first. This is owned by whoever has the skb queued ATM.
887 */
888 char cb[48] __aligned(8);
889
890 union {
891 struct {
892 unsigned long _skb_refdst;
893 void (*destructor)(struct sk_buff *skb);
894 };
895 struct list_head tcp_tsorted_anchor;
896#ifdef CONFIG_NET_SOCK_MSG
897 unsigned long _sk_redir;
898#endif
899 };
900
901#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
902 unsigned long _nfct;
903#endif
904 unsigned int len,
905 data_len;
906 __u16 mac_len,
907 hdr_len;
908
909 /* Following fields are _not_ copied in __copy_skb_header()
910 * Note that queue_mapping is here mostly to fill a hole.
911 */
912 __u16 queue_mapping;
913
914/* if you move cloned around you also must adapt those constants */
915#ifdef __BIG_ENDIAN_BITFIELD
916#define CLONED_MASK (1 << 7)
917#else
918#define CLONED_MASK 1
919#endif
920#define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
921
922 /* private: */
923 __u8 __cloned_offset[0];
924 /* public: */
925 __u8 cloned:1,
926 nohdr:1,
927 fclone:2,
928 peeked:1,
929 head_frag:1,
930 pfmemalloc:1,
931 pp_recycle:1; /* page_pool recycle indicator */
932#ifdef CONFIG_SKB_EXTENSIONS
933 __u8 active_extensions;
934#endif
935
936 /* Fields enclosed in headers group are copied
937 * using a single memcpy() in __copy_skb_header()
938 */
939 struct_group(headers,
940
941 /* private: */
942 __u8 __pkt_type_offset[0];
943 /* public: */
944 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
945 __u8 ignore_df:1;
946 __u8 dst_pending_confirm:1;
947 __u8 ip_summed:2;
948 __u8 ooo_okay:1;
949
950 /* private: */
951 __u8 __mono_tc_offset[0];
952 /* public: */
953 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
954#ifdef CONFIG_NET_XGRESS
955 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
956 __u8 tc_skip_classify:1;
957#endif
958 __u8 remcsum_offload:1;
959 __u8 csum_complete_sw:1;
960 __u8 csum_level:2;
961 __u8 inner_protocol_type:1;
962
963 __u8 l4_hash:1;
964 __u8 sw_hash:1;
965#ifdef CONFIG_WIRELESS
966 __u8 wifi_acked_valid:1;
967 __u8 wifi_acked:1;
968#endif
969 __u8 no_fcs:1;
970 /* Indicates the inner headers are valid in the skbuff. */
971 __u8 encapsulation:1;
972 __u8 encap_hdr_csum:1;
973 __u8 csum_valid:1;
974#ifdef CONFIG_IPV6_NDISC_NODETYPE
975 __u8 ndisc_nodetype:2;
976#endif
977
978#if IS_ENABLED(CONFIG_IP_VS)
979 __u8 ipvs_property:1;
980#endif
981#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
982 __u8 nf_trace:1;
983#endif
984#ifdef CONFIG_NET_SWITCHDEV
985 __u8 offload_fwd_mark:1;
986 __u8 offload_l3_fwd_mark:1;
987#endif
988 __u8 redirected:1;
989#ifdef CONFIG_NET_REDIRECT
990 __u8 from_ingress:1;
991#endif
992#ifdef CONFIG_NETFILTER_SKIP_EGRESS
993 __u8 nf_skip_egress:1;
994#endif
995#ifdef CONFIG_TLS_DEVICE
996 __u8 decrypted:1;
997#endif
998 __u8 slow_gro:1;
999#if IS_ENABLED(CONFIG_IP_SCTP)
1000 __u8 csum_not_inet:1;
1001#endif
1002
1003#if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS)
1004 __u16 tc_index; /* traffic control index */
1005#endif
1006
1007 u16 alloc_cpu;
1008
1009 union {
1010 __wsum csum;
1011 struct {
1012 __u16 csum_start;
1013 __u16 csum_offset;
1014 };
1015 };
1016 __u32 priority;
1017 int skb_iif;
1018 __u32 hash;
1019 union {
1020 u32 vlan_all;
1021 struct {
1022 __be16 vlan_proto;
1023 __u16 vlan_tci;
1024 };
1025 };
1026#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1027 union {
1028 unsigned int napi_id;
1029 unsigned int sender_cpu;
1030 };
1031#endif
1032#ifdef CONFIG_NETWORK_SECMARK
1033 __u32 secmark;
1034#endif
1035
1036 union {
1037 __u32 mark;
1038 __u32 reserved_tailroom;
1039 };
1040
1041 union {
1042 __be16 inner_protocol;
1043 __u8 inner_ipproto;
1044 };
1045
1046 __u16 inner_transport_header;
1047 __u16 inner_network_header;
1048 __u16 inner_mac_header;
1049
1050 __be16 protocol;
1051 __u16 transport_header;
1052 __u16 network_header;
1053 __u16 mac_header;
1054
1055#ifdef CONFIG_KCOV
1056 u64 kcov_handle;
1057#endif
1058
1059 ); /* end headers group */
1060
1061 /* These elements must be at the end, see alloc_skb() for details. */
1062 sk_buff_data_t tail;
1063 sk_buff_data_t end;
1064 unsigned char *head,
1065 *data;
1066 unsigned int truesize;
1067 refcount_t users;
1068
1069#ifdef CONFIG_SKB_EXTENSIONS
1070 /* only usable after checking ->active_extensions != 0 */
1071 struct skb_ext *extensions;
1072#endif
1073};
1074
1075/* if you move pkt_type around you also must adapt those constants */
1076#ifdef __BIG_ENDIAN_BITFIELD
1077#define PKT_TYPE_MAX (7 << 5)
1078#else
1079#define PKT_TYPE_MAX 7
1080#endif
1081#define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1082
1083/* if you move tc_at_ingress or mono_delivery_time
1084 * around, you also must adapt these constants.
1085 */
1086#ifdef __BIG_ENDIAN_BITFIELD
1087#define SKB_MONO_DELIVERY_TIME_MASK (1 << 7)
1088#define TC_AT_INGRESS_MASK (1 << 6)
1089#else
1090#define SKB_MONO_DELIVERY_TIME_MASK (1 << 0)
1091#define TC_AT_INGRESS_MASK (1 << 1)
1092#endif
1093#define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)
1094
1095#ifdef __KERNEL__
1096/*
1097 * Handling routines are only of interest to the kernel
1098 */
1099
1100#define SKB_ALLOC_FCLONE 0x01
1101#define SKB_ALLOC_RX 0x02
1102#define SKB_ALLOC_NAPI 0x04
1103
1104/**
1105 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1106 * @skb: buffer
1107 */
1108static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1109{
1110 return unlikely(skb->pfmemalloc);
1111}
1112
1113/*
1114 * skb might have a dst pointer attached, refcounted or not.
1115 * _skb_refdst low order bit is set if refcount was _not_ taken
1116 */
1117#define SKB_DST_NOREF 1UL
1118#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1119
1120/**
1121 * skb_dst - returns skb dst_entry
1122 * @skb: buffer
1123 *
1124 * Returns skb dst_entry, regardless of reference taken or not.
1125 */
1126static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1127{
1128 /* If refdst was not refcounted, check we still are in a
1129 * rcu_read_lock section
1130 */
1131 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1132 !rcu_read_lock_held() &&
1133 !rcu_read_lock_bh_held());
1134 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1135}
1136
1137/**
1138 * skb_dst_set - sets skb dst
1139 * @skb: buffer
1140 * @dst: dst entry
1141 *
1142 * Sets skb dst, assuming a reference was taken on dst and should
1143 * be released by skb_dst_drop()
1144 */
1145static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1146{
1147 skb->slow_gro |= !!dst;
1148 skb->_skb_refdst = (unsigned long)dst;
1149}
1150
1151/**
1152 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1153 * @skb: buffer
1154 * @dst: dst entry
1155 *
1156 * Sets skb dst, assuming a reference was not taken on dst.
1157 * If dst entry is cached, we do not take reference and dst_release
1158 * will be avoided by refdst_drop. If dst entry is not cached, we take
1159 * reference, so that last dst_release can destroy the dst immediately.
1160 */
1161static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1162{
1163 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1164 skb->slow_gro |= !!dst;
1165 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1166}
1167
1168/**
1169 * skb_dst_is_noref - Test if skb dst isn't refcounted
1170 * @skb: buffer
1171 */
1172static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1173{
1174 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1175}
1176
1177/**
1178 * skb_rtable - Returns the skb &rtable
1179 * @skb: buffer
1180 */
1181static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1182{
1183 return (struct rtable *)skb_dst(skb);
1184}
1185
1186/* For mangling skb->pkt_type from user space side from applications
1187 * such as nft, tc, etc, we only allow a conservative subset of
1188 * possible pkt_types to be set.
1189*/
1190static inline bool skb_pkt_type_ok(u32 ptype)
1191{
1192 return ptype <= PACKET_OTHERHOST;
1193}
1194
1195/**
1196 * skb_napi_id - Returns the skb's NAPI id
1197 * @skb: buffer
1198 */
1199static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1200{
1201#ifdef CONFIG_NET_RX_BUSY_POLL
1202 return skb->napi_id;
1203#else
1204 return 0;
1205#endif
1206}
1207
1208static inline bool skb_wifi_acked_valid(const struct sk_buff *skb)
1209{
1210#ifdef CONFIG_WIRELESS
1211 return skb->wifi_acked_valid;
1212#else
1213 return 0;
1214#endif
1215}
1216
1217/**
1218 * skb_unref - decrement the skb's reference count
1219 * @skb: buffer
1220 *
1221 * Returns true if we can free the skb.
1222 */
1223static inline bool skb_unref(struct sk_buff *skb)
1224{
1225 if (unlikely(!skb))
1226 return false;
1227 if (likely(refcount_read(&skb->users) == 1))
1228 smp_rmb();
1229 else if (likely(!refcount_dec_and_test(&skb->users)))
1230 return false;
1231
1232 return true;
1233}
1234
1235static inline bool skb_data_unref(const struct sk_buff *skb,
1236 struct skb_shared_info *shinfo)
1237{
1238 int bias;
1239
1240 if (!skb->cloned)
1241 return true;
1242
1243 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
1244
1245 if (atomic_read(&shinfo->dataref) == bias)
1246 smp_rmb();
1247 else if (atomic_sub_return(bias, &shinfo->dataref))
1248 return false;
1249
1250 return true;
1251}
1252
1253void __fix_address
1254kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1255
1256/**
1257 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1258 * @skb: buffer to free
1259 */
1260static inline void kfree_skb(struct sk_buff *skb)
1261{
1262 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1263}
1264
1265void skb_release_head_state(struct sk_buff *skb);
1266void kfree_skb_list_reason(struct sk_buff *segs,
1267 enum skb_drop_reason reason);
1268void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1269void skb_tx_error(struct sk_buff *skb);
1270
1271static inline void kfree_skb_list(struct sk_buff *segs)
1272{
1273 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1274}
1275
1276#ifdef CONFIG_TRACEPOINTS
1277void consume_skb(struct sk_buff *skb);
1278#else
1279static inline void consume_skb(struct sk_buff *skb)
1280{
1281 return kfree_skb(skb);
1282}
1283#endif
1284
1285void __consume_stateless_skb(struct sk_buff *skb);
1286void __kfree_skb(struct sk_buff *skb);
1287
1288void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1289bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1290 bool *fragstolen, int *delta_truesize);
1291
1292struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1293 int node);
1294struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1295struct sk_buff *build_skb(void *data, unsigned int frag_size);
1296struct sk_buff *build_skb_around(struct sk_buff *skb,
1297 void *data, unsigned int frag_size);
1298void skb_attempt_defer_free(struct sk_buff *skb);
1299
1300struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1301struct sk_buff *slab_build_skb(void *data);
1302
1303/**
1304 * alloc_skb - allocate a network buffer
1305 * @size: size to allocate
1306 * @priority: allocation mask
1307 *
1308 * This function is a convenient wrapper around __alloc_skb().
1309 */
1310static inline struct sk_buff *alloc_skb(unsigned int size,
1311 gfp_t priority)
1312{
1313 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1314}
1315
1316struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1317 unsigned long data_len,
1318 int max_page_order,
1319 int *errcode,
1320 gfp_t gfp_mask);
1321struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1322
1323/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1324struct sk_buff_fclones {
1325 struct sk_buff skb1;
1326
1327 struct sk_buff skb2;
1328
1329 refcount_t fclone_ref;
1330};
1331
1332/**
1333 * skb_fclone_busy - check if fclone is busy
1334 * @sk: socket
1335 * @skb: buffer
1336 *
1337 * Returns true if skb is a fast clone, and its clone is not freed.
1338 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1339 * so we also check that didn't happen.
1340 */
1341static inline bool skb_fclone_busy(const struct sock *sk,
1342 const struct sk_buff *skb)
1343{
1344 const struct sk_buff_fclones *fclones;
1345
1346 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1347
1348 return skb->fclone == SKB_FCLONE_ORIG &&
1349 refcount_read(&fclones->fclone_ref) > 1 &&
1350 READ_ONCE(fclones->skb2.sk) == sk;
1351}
1352
1353/**
1354 * alloc_skb_fclone - allocate a network buffer from fclone cache
1355 * @size: size to allocate
1356 * @priority: allocation mask
1357 *
1358 * This function is a convenient wrapper around __alloc_skb().
1359 */
1360static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1361 gfp_t priority)
1362{
1363 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1364}
1365
1366struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1367void skb_headers_offset_update(struct sk_buff *skb, int off);
1368int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1369struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1370void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1371struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1372struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1373 gfp_t gfp_mask, bool fclone);
1374static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1375 gfp_t gfp_mask)
1376{
1377 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1378}
1379
1380int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1381struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1382 unsigned int headroom);
1383struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1384struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1385 int newtailroom, gfp_t priority);
1386int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1387 int offset, int len);
1388int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1389 int offset, int len);
1390int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1391int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1392
1393/**
1394 * skb_pad - zero pad the tail of an skb
1395 * @skb: buffer to pad
1396 * @pad: space to pad
1397 *
1398 * Ensure that a buffer is followed by a padding area that is zero
1399 * filled. Used by network drivers which may DMA or transfer data
1400 * beyond the buffer end onto the wire.
1401 *
1402 * May return error in out of memory cases. The skb is freed on error.
1403 */
1404static inline int skb_pad(struct sk_buff *skb, int pad)
1405{
1406 return __skb_pad(skb, pad, true);
1407}
1408#define dev_kfree_skb(a) consume_skb(a)
1409
1410int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1411 int offset, size_t size, size_t max_frags);
1412
1413struct skb_seq_state {
1414 __u32 lower_offset;
1415 __u32 upper_offset;
1416 __u32 frag_idx;
1417 __u32 stepped_offset;
1418 struct sk_buff *root_skb;
1419 struct sk_buff *cur_skb;
1420 __u8 *frag_data;
1421 __u32 frag_off;
1422};
1423
1424void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1425 unsigned int to, struct skb_seq_state *st);
1426unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1427 struct skb_seq_state *st);
1428void skb_abort_seq_read(struct skb_seq_state *st);
1429
1430unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1431 unsigned int to, struct ts_config *config);
1432
1433/*
1434 * Packet hash types specify the type of hash in skb_set_hash.
1435 *
1436 * Hash types refer to the protocol layer addresses which are used to
1437 * construct a packet's hash. The hashes are used to differentiate or identify
1438 * flows of the protocol layer for the hash type. Hash types are either
1439 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1440 *
1441 * Properties of hashes:
1442 *
1443 * 1) Two packets in different flows have different hash values
1444 * 2) Two packets in the same flow should have the same hash value
1445 *
1446 * A hash at a higher layer is considered to be more specific. A driver should
1447 * set the most specific hash possible.
1448 *
1449 * A driver cannot indicate a more specific hash than the layer at which a hash
1450 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1451 *
1452 * A driver may indicate a hash level which is less specific than the
1453 * actual layer the hash was computed on. For instance, a hash computed
1454 * at L4 may be considered an L3 hash. This should only be done if the
1455 * driver can't unambiguously determine that the HW computed the hash at
1456 * the higher layer. Note that the "should" in the second property above
1457 * permits this.
1458 */
1459enum pkt_hash_types {
1460 PKT_HASH_TYPE_NONE, /* Undefined type */
1461 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1462 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1463 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1464};
1465
1466static inline void skb_clear_hash(struct sk_buff *skb)
1467{
1468 skb->hash = 0;
1469 skb->sw_hash = 0;
1470 skb->l4_hash = 0;
1471}
1472
1473static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1474{
1475 if (!skb->l4_hash)
1476 skb_clear_hash(skb);
1477}
1478
1479static inline void
1480__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1481{
1482 skb->l4_hash = is_l4;
1483 skb->sw_hash = is_sw;
1484 skb->hash = hash;
1485}
1486
1487static inline void
1488skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1489{
1490 /* Used by drivers to set hash from HW */
1491 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1492}
1493
1494static inline void
1495__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1496{
1497 __skb_set_hash(skb, hash, true, is_l4);
1498}
1499
1500void __skb_get_hash(struct sk_buff *skb);
1501u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1502u32 skb_get_poff(const struct sk_buff *skb);
1503u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1504 const struct flow_keys_basic *keys, int hlen);
1505__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1506 const void *data, int hlen_proto);
1507
1508static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1509 int thoff, u8 ip_proto)
1510{
1511 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1512}
1513
1514void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1515 const struct flow_dissector_key *key,
1516 unsigned int key_count);
1517
1518struct bpf_flow_dissector;
1519u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1520 __be16 proto, int nhoff, int hlen, unsigned int flags);
1521
1522bool __skb_flow_dissect(const struct net *net,
1523 const struct sk_buff *skb,
1524 struct flow_dissector *flow_dissector,
1525 void *target_container, const void *data,
1526 __be16 proto, int nhoff, int hlen, unsigned int flags);
1527
1528static inline bool skb_flow_dissect(const struct sk_buff *skb,
1529 struct flow_dissector *flow_dissector,
1530 void *target_container, unsigned int flags)
1531{
1532 return __skb_flow_dissect(NULL, skb, flow_dissector,
1533 target_container, NULL, 0, 0, 0, flags);
1534}
1535
1536static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1537 struct flow_keys *flow,
1538 unsigned int flags)
1539{
1540 memset(flow, 0, sizeof(*flow));
1541 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1542 flow, NULL, 0, 0, 0, flags);
1543}
1544
1545static inline bool
1546skb_flow_dissect_flow_keys_basic(const struct net *net,
1547 const struct sk_buff *skb,
1548 struct flow_keys_basic *flow,
1549 const void *data, __be16 proto,
1550 int nhoff, int hlen, unsigned int flags)
1551{
1552 memset(flow, 0, sizeof(*flow));
1553 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1554 data, proto, nhoff, hlen, flags);
1555}
1556
1557void skb_flow_dissect_meta(const struct sk_buff *skb,
1558 struct flow_dissector *flow_dissector,
1559 void *target_container);
1560
1561/* Gets a skb connection tracking info, ctinfo map should be a
1562 * map of mapsize to translate enum ip_conntrack_info states
1563 * to user states.
1564 */
1565void
1566skb_flow_dissect_ct(const struct sk_buff *skb,
1567 struct flow_dissector *flow_dissector,
1568 void *target_container,
1569 u16 *ctinfo_map, size_t mapsize,
1570 bool post_ct, u16 zone);
1571void
1572skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1573 struct flow_dissector *flow_dissector,
1574 void *target_container);
1575
1576void skb_flow_dissect_hash(const struct sk_buff *skb,
1577 struct flow_dissector *flow_dissector,
1578 void *target_container);
1579
1580static inline __u32 skb_get_hash(struct sk_buff *skb)
1581{
1582 if (!skb->l4_hash && !skb->sw_hash)
1583 __skb_get_hash(skb);
1584
1585 return skb->hash;
1586}
1587
1588static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1589{
1590 if (!skb->l4_hash && !skb->sw_hash) {
1591 struct flow_keys keys;
1592 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1593
1594 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1595 }
1596
1597 return skb->hash;
1598}
1599
1600__u32 skb_get_hash_perturb(const struct sk_buff *skb,
1601 const siphash_key_t *perturb);
1602
1603static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1604{
1605 return skb->hash;
1606}
1607
1608static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1609{
1610 to->hash = from->hash;
1611 to->sw_hash = from->sw_hash;
1612 to->l4_hash = from->l4_hash;
1613};
1614
1615static inline int skb_cmp_decrypted(const struct sk_buff *skb1,
1616 const struct sk_buff *skb2)
1617{
1618#ifdef CONFIG_TLS_DEVICE
1619 return skb2->decrypted - skb1->decrypted;
1620#else
1621 return 0;
1622#endif
1623}
1624
1625static inline void skb_copy_decrypted(struct sk_buff *to,
1626 const struct sk_buff *from)
1627{
1628#ifdef CONFIG_TLS_DEVICE
1629 to->decrypted = from->decrypted;
1630#endif
1631}
1632
1633#ifdef NET_SKBUFF_DATA_USES_OFFSET
1634static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1635{
1636 return skb->head + skb->end;
1637}
1638
1639static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1640{
1641 return skb->end;
1642}
1643
1644static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1645{
1646 skb->end = offset;
1647}
1648#else
1649static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1650{
1651 return skb->end;
1652}
1653
1654static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1655{
1656 return skb->end - skb->head;
1657}
1658
1659static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1660{
1661 skb->end = skb->head + offset;
1662}
1663#endif
1664
1665struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1666 struct ubuf_info *uarg);
1667
1668void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1669
1670void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1671 bool success);
1672
1673int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1674 struct sk_buff *skb, struct iov_iter *from,
1675 size_t length);
1676
1677static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1678 struct msghdr *msg, int len)
1679{
1680 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1681}
1682
1683int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1684 struct msghdr *msg, int len,
1685 struct ubuf_info *uarg);
1686
1687/* Internal */
1688#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1689
1690static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1691{
1692 return &skb_shinfo(skb)->hwtstamps;
1693}
1694
1695static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1696{
1697 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1698
1699 return is_zcopy ? skb_uarg(skb) : NULL;
1700}
1701
1702static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1703{
1704 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1705}
1706
1707static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1708{
1709 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1710}
1711
1712static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1713 const struct sk_buff *skb2)
1714{
1715 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1716}
1717
1718static inline void net_zcopy_get(struct ubuf_info *uarg)
1719{
1720 refcount_inc(&uarg->refcnt);
1721}
1722
1723static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1724{
1725 skb_shinfo(skb)->destructor_arg = uarg;
1726 skb_shinfo(skb)->flags |= uarg->flags;
1727}
1728
1729static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1730 bool *have_ref)
1731{
1732 if (skb && uarg && !skb_zcopy(skb)) {
1733 if (unlikely(have_ref && *have_ref))
1734 *have_ref = false;
1735 else
1736 net_zcopy_get(uarg);
1737 skb_zcopy_init(skb, uarg);
1738 }
1739}
1740
1741static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1742{
1743 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1744 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1745}
1746
1747static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1748{
1749 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1750}
1751
1752static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1753{
1754 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1755}
1756
1757static inline void net_zcopy_put(struct ubuf_info *uarg)
1758{
1759 if (uarg)
1760 uarg->callback(NULL, uarg, true);
1761}
1762
1763static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1764{
1765 if (uarg) {
1766 if (uarg->callback == msg_zerocopy_callback)
1767 msg_zerocopy_put_abort(uarg, have_uref);
1768 else if (have_uref)
1769 net_zcopy_put(uarg);
1770 }
1771}
1772
1773/* Release a reference on a zerocopy structure */
1774static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1775{
1776 struct ubuf_info *uarg = skb_zcopy(skb);
1777
1778 if (uarg) {
1779 if (!skb_zcopy_is_nouarg(skb))
1780 uarg->callback(skb, uarg, zerocopy_success);
1781
1782 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1783 }
1784}
1785
1786void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1787
1788static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1789{
1790 if (unlikely(skb_zcopy_managed(skb)))
1791 __skb_zcopy_downgrade_managed(skb);
1792}
1793
1794static inline void skb_mark_not_on_list(struct sk_buff *skb)
1795{
1796 skb->next = NULL;
1797}
1798
1799static inline void skb_poison_list(struct sk_buff *skb)
1800{
1801#ifdef CONFIG_DEBUG_NET
1802 skb->next = SKB_LIST_POISON_NEXT;
1803#endif
1804}
1805
1806/* Iterate through singly-linked GSO fragments of an skb. */
1807#define skb_list_walk_safe(first, skb, next_skb) \
1808 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1809 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1810
1811static inline void skb_list_del_init(struct sk_buff *skb)
1812{
1813 __list_del_entry(&skb->list);
1814 skb_mark_not_on_list(skb);
1815}
1816
1817/**
1818 * skb_queue_empty - check if a queue is empty
1819 * @list: queue head
1820 *
1821 * Returns true if the queue is empty, false otherwise.
1822 */
1823static inline int skb_queue_empty(const struct sk_buff_head *list)
1824{
1825 return list->next == (const struct sk_buff *) list;
1826}
1827
1828/**
1829 * skb_queue_empty_lockless - check if a queue is empty
1830 * @list: queue head
1831 *
1832 * Returns true if the queue is empty, false otherwise.
1833 * This variant can be used in lockless contexts.
1834 */
1835static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1836{
1837 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1838}
1839
1840
1841/**
1842 * skb_queue_is_last - check if skb is the last entry in the queue
1843 * @list: queue head
1844 * @skb: buffer
1845 *
1846 * Returns true if @skb is the last buffer on the list.
1847 */
1848static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1849 const struct sk_buff *skb)
1850{
1851 return skb->next == (const struct sk_buff *) list;
1852}
1853
1854/**
1855 * skb_queue_is_first - check if skb is the first entry in the queue
1856 * @list: queue head
1857 * @skb: buffer
1858 *
1859 * Returns true if @skb is the first buffer on the list.
1860 */
1861static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1862 const struct sk_buff *skb)
1863{
1864 return skb->prev == (const struct sk_buff *) list;
1865}
1866
1867/**
1868 * skb_queue_next - return the next packet in the queue
1869 * @list: queue head
1870 * @skb: current buffer
1871 *
1872 * Return the next packet in @list after @skb. It is only valid to
1873 * call this if skb_queue_is_last() evaluates to false.
1874 */
1875static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1876 const struct sk_buff *skb)
1877{
1878 /* This BUG_ON may seem severe, but if we just return then we
1879 * are going to dereference garbage.
1880 */
1881 BUG_ON(skb_queue_is_last(list, skb));
1882 return skb->next;
1883}
1884
1885/**
1886 * skb_queue_prev - return the prev packet in the queue
1887 * @list: queue head
1888 * @skb: current buffer
1889 *
1890 * Return the prev packet in @list before @skb. It is only valid to
1891 * call this if skb_queue_is_first() evaluates to false.
1892 */
1893static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1894 const struct sk_buff *skb)
1895{
1896 /* This BUG_ON may seem severe, but if we just return then we
1897 * are going to dereference garbage.
1898 */
1899 BUG_ON(skb_queue_is_first(list, skb));
1900 return skb->prev;
1901}
1902
1903/**
1904 * skb_get - reference buffer
1905 * @skb: buffer to reference
1906 *
1907 * Makes another reference to a socket buffer and returns a pointer
1908 * to the buffer.
1909 */
1910static inline struct sk_buff *skb_get(struct sk_buff *skb)
1911{
1912 refcount_inc(&skb->users);
1913 return skb;
1914}
1915
1916/*
1917 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1918 */
1919
1920/**
1921 * skb_cloned - is the buffer a clone
1922 * @skb: buffer to check
1923 *
1924 * Returns true if the buffer was generated with skb_clone() and is
1925 * one of multiple shared copies of the buffer. Cloned buffers are
1926 * shared data so must not be written to under normal circumstances.
1927 */
1928static inline int skb_cloned(const struct sk_buff *skb)
1929{
1930 return skb->cloned &&
1931 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1932}
1933
1934static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1935{
1936 might_sleep_if(gfpflags_allow_blocking(pri));
1937
1938 if (skb_cloned(skb))
1939 return pskb_expand_head(skb, 0, 0, pri);
1940
1941 return 0;
1942}
1943
1944/* This variant of skb_unclone() makes sure skb->truesize
1945 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1946 *
1947 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1948 * when various debugging features are in place.
1949 */
1950int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1951static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1952{
1953 might_sleep_if(gfpflags_allow_blocking(pri));
1954
1955 if (skb_cloned(skb))
1956 return __skb_unclone_keeptruesize(skb, pri);
1957 return 0;
1958}
1959
1960/**
1961 * skb_header_cloned - is the header a clone
1962 * @skb: buffer to check
1963 *
1964 * Returns true if modifying the header part of the buffer requires
1965 * the data to be copied.
1966 */
1967static inline int skb_header_cloned(const struct sk_buff *skb)
1968{
1969 int dataref;
1970
1971 if (!skb->cloned)
1972 return 0;
1973
1974 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1975 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1976 return dataref != 1;
1977}
1978
1979static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1980{
1981 might_sleep_if(gfpflags_allow_blocking(pri));
1982
1983 if (skb_header_cloned(skb))
1984 return pskb_expand_head(skb, 0, 0, pri);
1985
1986 return 0;
1987}
1988
1989/**
1990 * __skb_header_release() - allow clones to use the headroom
1991 * @skb: buffer to operate on
1992 *
1993 * See "DOC: dataref and headerless skbs".
1994 */
1995static inline void __skb_header_release(struct sk_buff *skb)
1996{
1997 skb->nohdr = 1;
1998 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1999}
2000
2001
2002/**
2003 * skb_shared - is the buffer shared
2004 * @skb: buffer to check
2005 *
2006 * Returns true if more than one person has a reference to this
2007 * buffer.
2008 */
2009static inline int skb_shared(const struct sk_buff *skb)
2010{
2011 return refcount_read(&skb->users) != 1;
2012}
2013
2014/**
2015 * skb_share_check - check if buffer is shared and if so clone it
2016 * @skb: buffer to check
2017 * @pri: priority for memory allocation
2018 *
2019 * If the buffer is shared the buffer is cloned and the old copy
2020 * drops a reference. A new clone with a single reference is returned.
2021 * If the buffer is not shared the original buffer is returned. When
2022 * being called from interrupt status or with spinlocks held pri must
2023 * be GFP_ATOMIC.
2024 *
2025 * NULL is returned on a memory allocation failure.
2026 */
2027static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2028{
2029 might_sleep_if(gfpflags_allow_blocking(pri));
2030 if (skb_shared(skb)) {
2031 struct sk_buff *nskb = skb_clone(skb, pri);
2032
2033 if (likely(nskb))
2034 consume_skb(skb);
2035 else
2036 kfree_skb(skb);
2037 skb = nskb;
2038 }
2039 return skb;
2040}
2041
2042/*
2043 * Copy shared buffers into a new sk_buff. We effectively do COW on
2044 * packets to handle cases where we have a local reader and forward
2045 * and a couple of other messy ones. The normal one is tcpdumping
2046 * a packet that's being forwarded.
2047 */
2048
2049/**
2050 * skb_unshare - make a copy of a shared buffer
2051 * @skb: buffer to check
2052 * @pri: priority for memory allocation
2053 *
2054 * If the socket buffer is a clone then this function creates a new
2055 * copy of the data, drops a reference count on the old copy and returns
2056 * the new copy with the reference count at 1. If the buffer is not a clone
2057 * the original buffer is returned. When called with a spinlock held or
2058 * from interrupt state @pri must be %GFP_ATOMIC
2059 *
2060 * %NULL is returned on a memory allocation failure.
2061 */
2062static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2063 gfp_t pri)
2064{
2065 might_sleep_if(gfpflags_allow_blocking(pri));
2066 if (skb_cloned(skb)) {
2067 struct sk_buff *nskb = skb_copy(skb, pri);
2068
2069 /* Free our shared copy */
2070 if (likely(nskb))
2071 consume_skb(skb);
2072 else
2073 kfree_skb(skb);
2074 skb = nskb;
2075 }
2076 return skb;
2077}
2078
2079/**
2080 * skb_peek - peek at the head of an &sk_buff_head
2081 * @list_: list to peek at
2082 *
2083 * Peek an &sk_buff. Unlike most other operations you _MUST_
2084 * be careful with this one. A peek leaves the buffer on the
2085 * list and someone else may run off with it. You must hold
2086 * the appropriate locks or have a private queue to do this.
2087 *
2088 * Returns %NULL for an empty list or a pointer to the head element.
2089 * The reference count is not incremented and the reference is therefore
2090 * volatile. Use with caution.
2091 */
2092static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2093{
2094 struct sk_buff *skb = list_->next;
2095
2096 if (skb == (struct sk_buff *)list_)
2097 skb = NULL;
2098 return skb;
2099}
2100
2101/**
2102 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2103 * @list_: list to peek at
2104 *
2105 * Like skb_peek(), but the caller knows that the list is not empty.
2106 */
2107static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2108{
2109 return list_->next;
2110}
2111
2112/**
2113 * skb_peek_next - peek skb following the given one from a queue
2114 * @skb: skb to start from
2115 * @list_: list to peek at
2116 *
2117 * Returns %NULL when the end of the list is met or a pointer to the
2118 * next element. The reference count is not incremented and the
2119 * reference is therefore volatile. Use with caution.
2120 */
2121static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2122 const struct sk_buff_head *list_)
2123{
2124 struct sk_buff *next = skb->next;
2125
2126 if (next == (struct sk_buff *)list_)
2127 next = NULL;
2128 return next;
2129}
2130
2131/**
2132 * skb_peek_tail - peek at the tail of an &sk_buff_head
2133 * @list_: list to peek at
2134 *
2135 * Peek an &sk_buff. Unlike most other operations you _MUST_
2136 * be careful with this one. A peek leaves the buffer on the
2137 * list and someone else may run off with it. You must hold
2138 * the appropriate locks or have a private queue to do this.
2139 *
2140 * Returns %NULL for an empty list or a pointer to the tail element.
2141 * The reference count is not incremented and the reference is therefore
2142 * volatile. Use with caution.
2143 */
2144static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2145{
2146 struct sk_buff *skb = READ_ONCE(list_->prev);
2147
2148 if (skb == (struct sk_buff *)list_)
2149 skb = NULL;
2150 return skb;
2151
2152}
2153
2154/**
2155 * skb_queue_len - get queue length
2156 * @list_: list to measure
2157 *
2158 * Return the length of an &sk_buff queue.
2159 */
2160static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2161{
2162 return list_->qlen;
2163}
2164
2165/**
2166 * skb_queue_len_lockless - get queue length
2167 * @list_: list to measure
2168 *
2169 * Return the length of an &sk_buff queue.
2170 * This variant can be used in lockless contexts.
2171 */
2172static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2173{
2174 return READ_ONCE(list_->qlen);
2175}
2176
2177/**
2178 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2179 * @list: queue to initialize
2180 *
2181 * This initializes only the list and queue length aspects of
2182 * an sk_buff_head object. This allows to initialize the list
2183 * aspects of an sk_buff_head without reinitializing things like
2184 * the spinlock. It can also be used for on-stack sk_buff_head
2185 * objects where the spinlock is known to not be used.
2186 */
2187static inline void __skb_queue_head_init(struct sk_buff_head *list)
2188{
2189 list->prev = list->next = (struct sk_buff *)list;
2190 list->qlen = 0;
2191}
2192
2193/*
2194 * This function creates a split out lock class for each invocation;
2195 * this is needed for now since a whole lot of users of the skb-queue
2196 * infrastructure in drivers have different locking usage (in hardirq)
2197 * than the networking core (in softirq only). In the long run either the
2198 * network layer or drivers should need annotation to consolidate the
2199 * main types of usage into 3 classes.
2200 */
2201static inline void skb_queue_head_init(struct sk_buff_head *list)
2202{
2203 spin_lock_init(&list->lock);
2204 __skb_queue_head_init(list);
2205}
2206
2207static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2208 struct lock_class_key *class)
2209{
2210 skb_queue_head_init(list);
2211 lockdep_set_class(&list->lock, class);
2212}
2213
2214/*
2215 * Insert an sk_buff on a list.
2216 *
2217 * The "__skb_xxxx()" functions are the non-atomic ones that
2218 * can only be called with interrupts disabled.
2219 */
2220static inline void __skb_insert(struct sk_buff *newsk,
2221 struct sk_buff *prev, struct sk_buff *next,
2222 struct sk_buff_head *list)
2223{
2224 /* See skb_queue_empty_lockless() and skb_peek_tail()
2225 * for the opposite READ_ONCE()
2226 */
2227 WRITE_ONCE(newsk->next, next);
2228 WRITE_ONCE(newsk->prev, prev);
2229 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2230 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2231 WRITE_ONCE(list->qlen, list->qlen + 1);
2232}
2233
2234static inline void __skb_queue_splice(const struct sk_buff_head *list,
2235 struct sk_buff *prev,
2236 struct sk_buff *next)
2237{
2238 struct sk_buff *first = list->next;
2239 struct sk_buff *last = list->prev;
2240
2241 WRITE_ONCE(first->prev, prev);
2242 WRITE_ONCE(prev->next, first);
2243
2244 WRITE_ONCE(last->next, next);
2245 WRITE_ONCE(next->prev, last);
2246}
2247
2248/**
2249 * skb_queue_splice - join two skb lists, this is designed for stacks
2250 * @list: the new list to add
2251 * @head: the place to add it in the first list
2252 */
2253static inline void skb_queue_splice(const struct sk_buff_head *list,
2254 struct sk_buff_head *head)
2255{
2256 if (!skb_queue_empty(list)) {
2257 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2258 head->qlen += list->qlen;
2259 }
2260}
2261
2262/**
2263 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2264 * @list: the new list to add
2265 * @head: the place to add it in the first list
2266 *
2267 * The list at @list is reinitialised
2268 */
2269static inline void skb_queue_splice_init(struct sk_buff_head *list,
2270 struct sk_buff_head *head)
2271{
2272 if (!skb_queue_empty(list)) {
2273 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2274 head->qlen += list->qlen;
2275 __skb_queue_head_init(list);
2276 }
2277}
2278
2279/**
2280 * skb_queue_splice_tail - join two skb lists, each list being a queue
2281 * @list: the new list to add
2282 * @head: the place to add it in the first list
2283 */
2284static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2285 struct sk_buff_head *head)
2286{
2287 if (!skb_queue_empty(list)) {
2288 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2289 head->qlen += list->qlen;
2290 }
2291}
2292
2293/**
2294 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2295 * @list: the new list to add
2296 * @head: the place to add it in the first list
2297 *
2298 * Each of the lists is a queue.
2299 * The list at @list is reinitialised
2300 */
2301static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2302 struct sk_buff_head *head)
2303{
2304 if (!skb_queue_empty(list)) {
2305 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2306 head->qlen += list->qlen;
2307 __skb_queue_head_init(list);
2308 }
2309}
2310
2311/**
2312 * __skb_queue_after - queue a buffer at the list head
2313 * @list: list to use
2314 * @prev: place after this buffer
2315 * @newsk: buffer to queue
2316 *
2317 * Queue a buffer int the middle of a list. This function takes no locks
2318 * and you must therefore hold required locks before calling it.
2319 *
2320 * A buffer cannot be placed on two lists at the same time.
2321 */
2322static inline void __skb_queue_after(struct sk_buff_head *list,
2323 struct sk_buff *prev,
2324 struct sk_buff *newsk)
2325{
2326 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2327}
2328
2329void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2330 struct sk_buff_head *list);
2331
2332static inline void __skb_queue_before(struct sk_buff_head *list,
2333 struct sk_buff *next,
2334 struct sk_buff *newsk)
2335{
2336 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2337}
2338
2339/**
2340 * __skb_queue_head - queue a buffer at the list head
2341 * @list: list to use
2342 * @newsk: buffer to queue
2343 *
2344 * Queue a buffer at the start of a list. This function takes no locks
2345 * and you must therefore hold required locks before calling it.
2346 *
2347 * A buffer cannot be placed on two lists at the same time.
2348 */
2349static inline void __skb_queue_head(struct sk_buff_head *list,
2350 struct sk_buff *newsk)
2351{
2352 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2353}
2354void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2355
2356/**
2357 * __skb_queue_tail - queue a buffer at the list tail
2358 * @list: list to use
2359 * @newsk: buffer to queue
2360 *
2361 * Queue a buffer at the end of a list. This function takes no locks
2362 * and you must therefore hold required locks before calling it.
2363 *
2364 * A buffer cannot be placed on two lists at the same time.
2365 */
2366static inline void __skb_queue_tail(struct sk_buff_head *list,
2367 struct sk_buff *newsk)
2368{
2369 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2370}
2371void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2372
2373/*
2374 * remove sk_buff from list. _Must_ be called atomically, and with
2375 * the list known..
2376 */
2377void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2378static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2379{
2380 struct sk_buff *next, *prev;
2381
2382 WRITE_ONCE(list->qlen, list->qlen - 1);
2383 next = skb->next;
2384 prev = skb->prev;
2385 skb->next = skb->prev = NULL;
2386 WRITE_ONCE(next->prev, prev);
2387 WRITE_ONCE(prev->next, next);
2388}
2389
2390/**
2391 * __skb_dequeue - remove from the head of the queue
2392 * @list: list to dequeue from
2393 *
2394 * Remove the head of the list. This function does not take any locks
2395 * so must be used with appropriate locks held only. The head item is
2396 * returned or %NULL if the list is empty.
2397 */
2398static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2399{
2400 struct sk_buff *skb = skb_peek(list);
2401 if (skb)
2402 __skb_unlink(skb, list);
2403 return skb;
2404}
2405struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2406
2407/**
2408 * __skb_dequeue_tail - remove from the tail of the queue
2409 * @list: list to dequeue from
2410 *
2411 * Remove the tail of the list. This function does not take any locks
2412 * so must be used with appropriate locks held only. The tail item is
2413 * returned or %NULL if the list is empty.
2414 */
2415static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2416{
2417 struct sk_buff *skb = skb_peek_tail(list);
2418 if (skb)
2419 __skb_unlink(skb, list);
2420 return skb;
2421}
2422struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2423
2424
2425static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2426{
2427 return skb->data_len;
2428}
2429
2430static inline unsigned int skb_headlen(const struct sk_buff *skb)
2431{
2432 return skb->len - skb->data_len;
2433}
2434
2435static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2436{
2437 unsigned int i, len = 0;
2438
2439 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2440 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2441 return len;
2442}
2443
2444static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2445{
2446 return skb_headlen(skb) + __skb_pagelen(skb);
2447}
2448
2449static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag,
2450 netmem_ref netmem, int off,
2451 int size)
2452{
2453 frag->netmem = netmem;
2454 frag->offset = off;
2455 skb_frag_size_set(frag, size);
2456}
2457
2458static inline void skb_frag_fill_page_desc(skb_frag_t *frag,
2459 struct page *page,
2460 int off, int size)
2461{
2462 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size);
2463}
2464
2465static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo,
2466 int i, netmem_ref netmem,
2467 int off, int size)
2468{
2469 skb_frag_t *frag = &shinfo->frags[i];
2470
2471 skb_frag_fill_netmem_desc(frag, netmem, off, size);
2472}
2473
2474static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2475 int i, struct page *page,
2476 int off, int size)
2477{
2478 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off,
2479 size);
2480}
2481
2482/**
2483 * skb_len_add - adds a number to len fields of skb
2484 * @skb: buffer to add len to
2485 * @delta: number of bytes to add
2486 */
2487static inline void skb_len_add(struct sk_buff *skb, int delta)
2488{
2489 skb->len += delta;
2490 skb->data_len += delta;
2491 skb->truesize += delta;
2492}
2493
2494/**
2495 * __skb_fill_netmem_desc - initialise a fragment in an skb
2496 * @skb: buffer containing fragment to be initialised
2497 * @i: fragment index to initialise
2498 * @netmem: the netmem to use for this fragment
2499 * @off: the offset to the data with @page
2500 * @size: the length of the data
2501 *
2502 * Initialises the @i'th fragment of @skb to point to &size bytes at
2503 * offset @off within @page.
2504 *
2505 * Does not take any additional reference on the fragment.
2506 */
2507static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i,
2508 netmem_ref netmem, int off, int size)
2509{
2510 struct page *page = netmem_to_page(netmem);
2511
2512 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size);
2513
2514 /* Propagate page pfmemalloc to the skb if we can. The problem is
2515 * that not all callers have unique ownership of the page but rely
2516 * on page_is_pfmemalloc doing the right thing(tm).
2517 */
2518 page = compound_head(page);
2519 if (page_is_pfmemalloc(page))
2520 skb->pfmemalloc = true;
2521}
2522
2523static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2524 struct page *page, int off, int size)
2525{
2526 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2527}
2528
2529static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i,
2530 netmem_ref netmem, int off, int size)
2531{
2532 __skb_fill_netmem_desc(skb, i, netmem, off, size);
2533 skb_shinfo(skb)->nr_frags = i + 1;
2534}
2535
2536/**
2537 * skb_fill_page_desc - initialise a paged fragment in an skb
2538 * @skb: buffer containing fragment to be initialised
2539 * @i: paged fragment index to initialise
2540 * @page: the page to use for this fragment
2541 * @off: the offset to the data with @page
2542 * @size: the length of the data
2543 *
2544 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2545 * @skb to point to @size bytes at offset @off within @page. In
2546 * addition updates @skb such that @i is the last fragment.
2547 *
2548 * Does not take any additional reference on the fragment.
2549 */
2550static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2551 struct page *page, int off, int size)
2552{
2553 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2554}
2555
2556/**
2557 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2558 * @skb: buffer containing fragment to be initialised
2559 * @i: paged fragment index to initialise
2560 * @page: the page to use for this fragment
2561 * @off: the offset to the data with @page
2562 * @size: the length of the data
2563 *
2564 * Variant of skb_fill_page_desc() which does not deal with
2565 * pfmemalloc, if page is not owned by us.
2566 */
2567static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2568 struct page *page, int off,
2569 int size)
2570{
2571 struct skb_shared_info *shinfo = skb_shinfo(skb);
2572
2573 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2574 shinfo->nr_frags = i + 1;
2575}
2576
2577void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
2578 int off, int size, unsigned int truesize);
2579
2580static inline void skb_add_rx_frag(struct sk_buff *skb, int i,
2581 struct page *page, int off, int size,
2582 unsigned int truesize)
2583{
2584 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size,
2585 truesize);
2586}
2587
2588void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2589 unsigned int truesize);
2590
2591#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2592
2593#ifdef NET_SKBUFF_DATA_USES_OFFSET
2594static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2595{
2596 return skb->head + skb->tail;
2597}
2598
2599static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2600{
2601 skb->tail = skb->data - skb->head;
2602}
2603
2604static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2605{
2606 skb_reset_tail_pointer(skb);
2607 skb->tail += offset;
2608}
2609
2610#else /* NET_SKBUFF_DATA_USES_OFFSET */
2611static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2612{
2613 return skb->tail;
2614}
2615
2616static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2617{
2618 skb->tail = skb->data;
2619}
2620
2621static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2622{
2623 skb->tail = skb->data + offset;
2624}
2625
2626#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2627
2628static inline void skb_assert_len(struct sk_buff *skb)
2629{
2630#ifdef CONFIG_DEBUG_NET
2631 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2632 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2633#endif /* CONFIG_DEBUG_NET */
2634}
2635
2636/*
2637 * Add data to an sk_buff
2638 */
2639void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2640void *skb_put(struct sk_buff *skb, unsigned int len);
2641static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2642{
2643 void *tmp = skb_tail_pointer(skb);
2644 SKB_LINEAR_ASSERT(skb);
2645 skb->tail += len;
2646 skb->len += len;
2647 return tmp;
2648}
2649
2650static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2651{
2652 void *tmp = __skb_put(skb, len);
2653
2654 memset(tmp, 0, len);
2655 return tmp;
2656}
2657
2658static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2659 unsigned int len)
2660{
2661 void *tmp = __skb_put(skb, len);
2662
2663 memcpy(tmp, data, len);
2664 return tmp;
2665}
2666
2667static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2668{
2669 *(u8 *)__skb_put(skb, 1) = val;
2670}
2671
2672static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2673{
2674 void *tmp = skb_put(skb, len);
2675
2676 memset(tmp, 0, len);
2677
2678 return tmp;
2679}
2680
2681static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2682 unsigned int len)
2683{
2684 void *tmp = skb_put(skb, len);
2685
2686 memcpy(tmp, data, len);
2687
2688 return tmp;
2689}
2690
2691static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2692{
2693 *(u8 *)skb_put(skb, 1) = val;
2694}
2695
2696void *skb_push(struct sk_buff *skb, unsigned int len);
2697static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2698{
2699 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2700
2701 skb->data -= len;
2702 skb->len += len;
2703 return skb->data;
2704}
2705
2706void *skb_pull(struct sk_buff *skb, unsigned int len);
2707static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2708{
2709 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2710
2711 skb->len -= len;
2712 if (unlikely(skb->len < skb->data_len)) {
2713#if defined(CONFIG_DEBUG_NET)
2714 skb->len += len;
2715 pr_err("__skb_pull(len=%u)\n", len);
2716 skb_dump(KERN_ERR, skb, false);
2717#endif
2718 BUG();
2719 }
2720 return skb->data += len;
2721}
2722
2723static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2724{
2725 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2726}
2727
2728void *skb_pull_data(struct sk_buff *skb, size_t len);
2729
2730void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2731
2732static inline enum skb_drop_reason
2733pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2734{
2735 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2736
2737 if (likely(len <= skb_headlen(skb)))
2738 return SKB_NOT_DROPPED_YET;
2739
2740 if (unlikely(len > skb->len))
2741 return SKB_DROP_REASON_PKT_TOO_SMALL;
2742
2743 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2744 return SKB_DROP_REASON_NOMEM;
2745
2746 return SKB_NOT_DROPPED_YET;
2747}
2748
2749static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2750{
2751 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2752}
2753
2754static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2755{
2756 if (!pskb_may_pull(skb, len))
2757 return NULL;
2758
2759 skb->len -= len;
2760 return skb->data += len;
2761}
2762
2763void skb_condense(struct sk_buff *skb);
2764
2765/**
2766 * skb_headroom - bytes at buffer head
2767 * @skb: buffer to check
2768 *
2769 * Return the number of bytes of free space at the head of an &sk_buff.
2770 */
2771static inline unsigned int skb_headroom(const struct sk_buff *skb)
2772{
2773 return skb->data - skb->head;
2774}
2775
2776/**
2777 * skb_tailroom - bytes at buffer end
2778 * @skb: buffer to check
2779 *
2780 * Return the number of bytes of free space at the tail of an sk_buff
2781 */
2782static inline int skb_tailroom(const struct sk_buff *skb)
2783{
2784 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2785}
2786
2787/**
2788 * skb_availroom - bytes at buffer end
2789 * @skb: buffer to check
2790 *
2791 * Return the number of bytes of free space at the tail of an sk_buff
2792 * allocated by sk_stream_alloc()
2793 */
2794static inline int skb_availroom(const struct sk_buff *skb)
2795{
2796 if (skb_is_nonlinear(skb))
2797 return 0;
2798
2799 return skb->end - skb->tail - skb->reserved_tailroom;
2800}
2801
2802/**
2803 * skb_reserve - adjust headroom
2804 * @skb: buffer to alter
2805 * @len: bytes to move
2806 *
2807 * Increase the headroom of an empty &sk_buff by reducing the tail
2808 * room. This is only allowed for an empty buffer.
2809 */
2810static inline void skb_reserve(struct sk_buff *skb, int len)
2811{
2812 skb->data += len;
2813 skb->tail += len;
2814}
2815
2816/**
2817 * skb_tailroom_reserve - adjust reserved_tailroom
2818 * @skb: buffer to alter
2819 * @mtu: maximum amount of headlen permitted
2820 * @needed_tailroom: minimum amount of reserved_tailroom
2821 *
2822 * Set reserved_tailroom so that headlen can be as large as possible but
2823 * not larger than mtu and tailroom cannot be smaller than
2824 * needed_tailroom.
2825 * The required headroom should already have been reserved before using
2826 * this function.
2827 */
2828static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2829 unsigned int needed_tailroom)
2830{
2831 SKB_LINEAR_ASSERT(skb);
2832 if (mtu < skb_tailroom(skb) - needed_tailroom)
2833 /* use at most mtu */
2834 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2835 else
2836 /* use up to all available space */
2837 skb->reserved_tailroom = needed_tailroom;
2838}
2839
2840#define ENCAP_TYPE_ETHER 0
2841#define ENCAP_TYPE_IPPROTO 1
2842
2843static inline void skb_set_inner_protocol(struct sk_buff *skb,
2844 __be16 protocol)
2845{
2846 skb->inner_protocol = protocol;
2847 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2848}
2849
2850static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2851 __u8 ipproto)
2852{
2853 skb->inner_ipproto = ipproto;
2854 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2855}
2856
2857static inline void skb_reset_inner_headers(struct sk_buff *skb)
2858{
2859 skb->inner_mac_header = skb->mac_header;
2860 skb->inner_network_header = skb->network_header;
2861 skb->inner_transport_header = skb->transport_header;
2862}
2863
2864static inline void skb_reset_mac_len(struct sk_buff *skb)
2865{
2866 skb->mac_len = skb->network_header - skb->mac_header;
2867}
2868
2869static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2870 *skb)
2871{
2872 return skb->head + skb->inner_transport_header;
2873}
2874
2875static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2876{
2877 return skb_inner_transport_header(skb) - skb->data;
2878}
2879
2880static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2881{
2882 skb->inner_transport_header = skb->data - skb->head;
2883}
2884
2885static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2886 const int offset)
2887{
2888 skb_reset_inner_transport_header(skb);
2889 skb->inner_transport_header += offset;
2890}
2891
2892static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2893{
2894 return skb->head + skb->inner_network_header;
2895}
2896
2897static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2898{
2899 skb->inner_network_header = skb->data - skb->head;
2900}
2901
2902static inline void skb_set_inner_network_header(struct sk_buff *skb,
2903 const int offset)
2904{
2905 skb_reset_inner_network_header(skb);
2906 skb->inner_network_header += offset;
2907}
2908
2909static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb)
2910{
2911 return skb->inner_network_header > 0;
2912}
2913
2914static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2915{
2916 return skb->head + skb->inner_mac_header;
2917}
2918
2919static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2920{
2921 skb->inner_mac_header = skb->data - skb->head;
2922}
2923
2924static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2925 const int offset)
2926{
2927 skb_reset_inner_mac_header(skb);
2928 skb->inner_mac_header += offset;
2929}
2930static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2931{
2932 return skb->transport_header != (typeof(skb->transport_header))~0U;
2933}
2934
2935static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2936{
2937 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2938 return skb->head + skb->transport_header;
2939}
2940
2941static inline void skb_reset_transport_header(struct sk_buff *skb)
2942{
2943 skb->transport_header = skb->data - skb->head;
2944}
2945
2946static inline void skb_set_transport_header(struct sk_buff *skb,
2947 const int offset)
2948{
2949 skb_reset_transport_header(skb);
2950 skb->transport_header += offset;
2951}
2952
2953static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2954{
2955 return skb->head + skb->network_header;
2956}
2957
2958static inline void skb_reset_network_header(struct sk_buff *skb)
2959{
2960 skb->network_header = skb->data - skb->head;
2961}
2962
2963static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2964{
2965 skb_reset_network_header(skb);
2966 skb->network_header += offset;
2967}
2968
2969static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2970{
2971 return skb->mac_header != (typeof(skb->mac_header))~0U;
2972}
2973
2974static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2975{
2976 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2977 return skb->head + skb->mac_header;
2978}
2979
2980static inline int skb_mac_offset(const struct sk_buff *skb)
2981{
2982 return skb_mac_header(skb) - skb->data;
2983}
2984
2985static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2986{
2987 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2988 return skb->network_header - skb->mac_header;
2989}
2990
2991static inline void skb_unset_mac_header(struct sk_buff *skb)
2992{
2993 skb->mac_header = (typeof(skb->mac_header))~0U;
2994}
2995
2996static inline void skb_reset_mac_header(struct sk_buff *skb)
2997{
2998 skb->mac_header = skb->data - skb->head;
2999}
3000
3001static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
3002{
3003 skb_reset_mac_header(skb);
3004 skb->mac_header += offset;
3005}
3006
3007static inline void skb_pop_mac_header(struct sk_buff *skb)
3008{
3009 skb->mac_header = skb->network_header;
3010}
3011
3012static inline void skb_probe_transport_header(struct sk_buff *skb)
3013{
3014 struct flow_keys_basic keys;
3015
3016 if (skb_transport_header_was_set(skb))
3017 return;
3018
3019 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
3020 NULL, 0, 0, 0, 0))
3021 skb_set_transport_header(skb, keys.control.thoff);
3022}
3023
3024static inline void skb_mac_header_rebuild(struct sk_buff *skb)
3025{
3026 if (skb_mac_header_was_set(skb)) {
3027 const unsigned char *old_mac = skb_mac_header(skb);
3028
3029 skb_set_mac_header(skb, -skb->mac_len);
3030 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
3031 }
3032}
3033
3034/* Move the full mac header up to current network_header.
3035 * Leaves skb->data pointing at offset skb->mac_len into the mac_header.
3036 * Must be provided the complete mac header length.
3037 */
3038static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len)
3039{
3040 if (skb_mac_header_was_set(skb)) {
3041 const unsigned char *old_mac = skb_mac_header(skb);
3042
3043 skb_set_mac_header(skb, -full_mac_len);
3044 memmove(skb_mac_header(skb), old_mac, full_mac_len);
3045 __skb_push(skb, full_mac_len - skb->mac_len);
3046 }
3047}
3048
3049static inline int skb_checksum_start_offset(const struct sk_buff *skb)
3050{
3051 return skb->csum_start - skb_headroom(skb);
3052}
3053
3054static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
3055{
3056 return skb->head + skb->csum_start;
3057}
3058
3059static inline int skb_transport_offset(const struct sk_buff *skb)
3060{
3061 return skb_transport_header(skb) - skb->data;
3062}
3063
3064static inline u32 skb_network_header_len(const struct sk_buff *skb)
3065{
3066 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
3067 return skb->transport_header - skb->network_header;
3068}
3069
3070static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
3071{
3072 return skb->inner_transport_header - skb->inner_network_header;
3073}
3074
3075static inline int skb_network_offset(const struct sk_buff *skb)
3076{
3077 return skb_network_header(skb) - skb->data;
3078}
3079
3080static inline int skb_inner_network_offset(const struct sk_buff *skb)
3081{
3082 return skb_inner_network_header(skb) - skb->data;
3083}
3084
3085static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
3086{
3087 return pskb_may_pull(skb, skb_network_offset(skb) + len);
3088}
3089
3090/*
3091 * CPUs often take a performance hit when accessing unaligned memory
3092 * locations. The actual performance hit varies, it can be small if the
3093 * hardware handles it or large if we have to take an exception and fix it
3094 * in software.
3095 *
3096 * Since an ethernet header is 14 bytes network drivers often end up with
3097 * the IP header at an unaligned offset. The IP header can be aligned by
3098 * shifting the start of the packet by 2 bytes. Drivers should do this
3099 * with:
3100 *
3101 * skb_reserve(skb, NET_IP_ALIGN);
3102 *
3103 * The downside to this alignment of the IP header is that the DMA is now
3104 * unaligned. On some architectures the cost of an unaligned DMA is high
3105 * and this cost outweighs the gains made by aligning the IP header.
3106 *
3107 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3108 * to be overridden.
3109 */
3110#ifndef NET_IP_ALIGN
3111#define NET_IP_ALIGN 2
3112#endif
3113
3114/*
3115 * The networking layer reserves some headroom in skb data (via
3116 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3117 * the header has to grow. In the default case, if the header has to grow
3118 * 32 bytes or less we avoid the reallocation.
3119 *
3120 * Unfortunately this headroom changes the DMA alignment of the resulting
3121 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3122 * on some architectures. An architecture can override this value,
3123 * perhaps setting it to a cacheline in size (since that will maintain
3124 * cacheline alignment of the DMA). It must be a power of 2.
3125 *
3126 * Various parts of the networking layer expect at least 32 bytes of
3127 * headroom, you should not reduce this.
3128 *
3129 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3130 * to reduce average number of cache lines per packet.
3131 * get_rps_cpu() for example only access one 64 bytes aligned block :
3132 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3133 */
3134#ifndef NET_SKB_PAD
3135#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3136#endif
3137
3138int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3139
3140static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3141{
3142 if (WARN_ON(skb_is_nonlinear(skb)))
3143 return;
3144 skb->len = len;
3145 skb_set_tail_pointer(skb, len);
3146}
3147
3148static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3149{
3150 __skb_set_length(skb, len);
3151}
3152
3153void skb_trim(struct sk_buff *skb, unsigned int len);
3154
3155static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3156{
3157 if (skb->data_len)
3158 return ___pskb_trim(skb, len);
3159 __skb_trim(skb, len);
3160 return 0;
3161}
3162
3163static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3164{
3165 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3166}
3167
3168/**
3169 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3170 * @skb: buffer to alter
3171 * @len: new length
3172 *
3173 * This is identical to pskb_trim except that the caller knows that
3174 * the skb is not cloned so we should never get an error due to out-
3175 * of-memory.
3176 */
3177static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3178{
3179 int err = pskb_trim(skb, len);
3180 BUG_ON(err);
3181}
3182
3183static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3184{
3185 unsigned int diff = len - skb->len;
3186
3187 if (skb_tailroom(skb) < diff) {
3188 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3189 GFP_ATOMIC);
3190 if (ret)
3191 return ret;
3192 }
3193 __skb_set_length(skb, len);
3194 return 0;
3195}
3196
3197/**
3198 * skb_orphan - orphan a buffer
3199 * @skb: buffer to orphan
3200 *
3201 * If a buffer currently has an owner then we call the owner's
3202 * destructor function and make the @skb unowned. The buffer continues
3203 * to exist but is no longer charged to its former owner.
3204 */
3205static inline void skb_orphan(struct sk_buff *skb)
3206{
3207 if (skb->destructor) {
3208 skb->destructor(skb);
3209 skb->destructor = NULL;
3210 skb->sk = NULL;
3211 } else {
3212 BUG_ON(skb->sk);
3213 }
3214}
3215
3216/**
3217 * skb_orphan_frags - orphan the frags contained in a buffer
3218 * @skb: buffer to orphan frags from
3219 * @gfp_mask: allocation mask for replacement pages
3220 *
3221 * For each frag in the SKB which needs a destructor (i.e. has an
3222 * owner) create a copy of that frag and release the original
3223 * page by calling the destructor.
3224 */
3225static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3226{
3227 if (likely(!skb_zcopy(skb)))
3228 return 0;
3229 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3230 return 0;
3231 return skb_copy_ubufs(skb, gfp_mask);
3232}
3233
3234/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3235static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3236{
3237 if (likely(!skb_zcopy(skb)))
3238 return 0;
3239 return skb_copy_ubufs(skb, gfp_mask);
3240}
3241
3242/**
3243 * __skb_queue_purge_reason - empty a list
3244 * @list: list to empty
3245 * @reason: drop reason
3246 *
3247 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3248 * the list and one reference dropped. This function does not take the
3249 * list lock and the caller must hold the relevant locks to use it.
3250 */
3251static inline void __skb_queue_purge_reason(struct sk_buff_head *list,
3252 enum skb_drop_reason reason)
3253{
3254 struct sk_buff *skb;
3255
3256 while ((skb = __skb_dequeue(list)) != NULL)
3257 kfree_skb_reason(skb, reason);
3258}
3259
3260static inline void __skb_queue_purge(struct sk_buff_head *list)
3261{
3262 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3263}
3264
3265void skb_queue_purge_reason(struct sk_buff_head *list,
3266 enum skb_drop_reason reason);
3267
3268static inline void skb_queue_purge(struct sk_buff_head *list)
3269{
3270 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3271}
3272
3273unsigned int skb_rbtree_purge(struct rb_root *root);
3274void skb_errqueue_purge(struct sk_buff_head *list);
3275
3276void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3277
3278/**
3279 * netdev_alloc_frag - allocate a page fragment
3280 * @fragsz: fragment size
3281 *
3282 * Allocates a frag from a page for receive buffer.
3283 * Uses GFP_ATOMIC allocations.
3284 */
3285static inline void *netdev_alloc_frag(unsigned int fragsz)
3286{
3287 return __netdev_alloc_frag_align(fragsz, ~0u);
3288}
3289
3290static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3291 unsigned int align)
3292{
3293 WARN_ON_ONCE(!is_power_of_2(align));
3294 return __netdev_alloc_frag_align(fragsz, -align);
3295}
3296
3297struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3298 gfp_t gfp_mask);
3299
3300/**
3301 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3302 * @dev: network device to receive on
3303 * @length: length to allocate
3304 *
3305 * Allocate a new &sk_buff and assign it a usage count of one. The
3306 * buffer has unspecified headroom built in. Users should allocate
3307 * the headroom they think they need without accounting for the
3308 * built in space. The built in space is used for optimisations.
3309 *
3310 * %NULL is returned if there is no free memory. Although this function
3311 * allocates memory it can be called from an interrupt.
3312 */
3313static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3314 unsigned int length)
3315{
3316 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3317}
3318
3319/* legacy helper around __netdev_alloc_skb() */
3320static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3321 gfp_t gfp_mask)
3322{
3323 return __netdev_alloc_skb(NULL, length, gfp_mask);
3324}
3325
3326/* legacy helper around netdev_alloc_skb() */
3327static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3328{
3329 return netdev_alloc_skb(NULL, length);
3330}
3331
3332
3333static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3334 unsigned int length, gfp_t gfp)
3335{
3336 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3337
3338 if (NET_IP_ALIGN && skb)
3339 skb_reserve(skb, NET_IP_ALIGN);
3340 return skb;
3341}
3342
3343static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3344 unsigned int length)
3345{
3346 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3347}
3348
3349static inline void skb_free_frag(void *addr)
3350{
3351 page_frag_free(addr);
3352}
3353
3354void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3355
3356static inline void *napi_alloc_frag(unsigned int fragsz)
3357{
3358 return __napi_alloc_frag_align(fragsz, ~0u);
3359}
3360
3361static inline void *napi_alloc_frag_align(unsigned int fragsz,
3362 unsigned int align)
3363{
3364 WARN_ON_ONCE(!is_power_of_2(align));
3365 return __napi_alloc_frag_align(fragsz, -align);
3366}
3367
3368struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3369 unsigned int length, gfp_t gfp_mask);
3370static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3371 unsigned int length)
3372{
3373 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3374}
3375void napi_consume_skb(struct sk_buff *skb, int budget);
3376
3377void napi_skb_free_stolen_head(struct sk_buff *skb);
3378void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason);
3379
3380/**
3381 * __dev_alloc_pages - allocate page for network Rx
3382 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3383 * @order: size of the allocation
3384 *
3385 * Allocate a new page.
3386 *
3387 * %NULL is returned if there is no free memory.
3388*/
3389static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3390 unsigned int order)
3391{
3392 /* This piece of code contains several assumptions.
3393 * 1. This is for device Rx, therefore a cold page is preferred.
3394 * 2. The expectation is the user wants a compound page.
3395 * 3. If requesting a order 0 page it will not be compound
3396 * due to the check to see if order has a value in prep_new_page
3397 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3398 * code in gfp_to_alloc_flags that should be enforcing this.
3399 */
3400 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3401
3402 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3403}
3404
3405static inline struct page *dev_alloc_pages(unsigned int order)
3406{
3407 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3408}
3409
3410/**
3411 * __dev_alloc_page - allocate a page for network Rx
3412 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3413 *
3414 * Allocate a new page.
3415 *
3416 * %NULL is returned if there is no free memory.
3417 */
3418static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3419{
3420 return __dev_alloc_pages(gfp_mask, 0);
3421}
3422
3423static inline struct page *dev_alloc_page(void)
3424{
3425 return dev_alloc_pages(0);
3426}
3427
3428/**
3429 * dev_page_is_reusable - check whether a page can be reused for network Rx
3430 * @page: the page to test
3431 *
3432 * A page shouldn't be considered for reusing/recycling if it was allocated
3433 * under memory pressure or at a distant memory node.
3434 *
3435 * Returns false if this page should be returned to page allocator, true
3436 * otherwise.
3437 */
3438static inline bool dev_page_is_reusable(const struct page *page)
3439{
3440 return likely(page_to_nid(page) == numa_mem_id() &&
3441 !page_is_pfmemalloc(page));
3442}
3443
3444/**
3445 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3446 * @page: The page that was allocated from skb_alloc_page
3447 * @skb: The skb that may need pfmemalloc set
3448 */
3449static inline void skb_propagate_pfmemalloc(const struct page *page,
3450 struct sk_buff *skb)
3451{
3452 if (page_is_pfmemalloc(page))
3453 skb->pfmemalloc = true;
3454}
3455
3456/**
3457 * skb_frag_off() - Returns the offset of a skb fragment
3458 * @frag: the paged fragment
3459 */
3460static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3461{
3462 return frag->offset;
3463}
3464
3465/**
3466 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3467 * @frag: skb fragment
3468 * @delta: value to add
3469 */
3470static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3471{
3472 frag->offset += delta;
3473}
3474
3475/**
3476 * skb_frag_off_set() - Sets the offset of a skb fragment
3477 * @frag: skb fragment
3478 * @offset: offset of fragment
3479 */
3480static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3481{
3482 frag->offset = offset;
3483}
3484
3485/**
3486 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3487 * @fragto: skb fragment where offset is set
3488 * @fragfrom: skb fragment offset is copied from
3489 */
3490static inline void skb_frag_off_copy(skb_frag_t *fragto,
3491 const skb_frag_t *fragfrom)
3492{
3493 fragto->offset = fragfrom->offset;
3494}
3495
3496/**
3497 * skb_frag_page - retrieve the page referred to by a paged fragment
3498 * @frag: the paged fragment
3499 *
3500 * Returns the &struct page associated with @frag.
3501 */
3502static inline struct page *skb_frag_page(const skb_frag_t *frag)
3503{
3504 return netmem_to_page(frag->netmem);
3505}
3506
3507/**
3508 * __skb_frag_ref - take an addition reference on a paged fragment.
3509 * @frag: the paged fragment
3510 *
3511 * Takes an additional reference on the paged fragment @frag.
3512 */
3513static inline void __skb_frag_ref(skb_frag_t *frag)
3514{
3515 get_page(skb_frag_page(frag));
3516}
3517
3518/**
3519 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3520 * @skb: the buffer
3521 * @f: the fragment offset.
3522 *
3523 * Takes an additional reference on the @f'th paged fragment of @skb.
3524 */
3525static inline void skb_frag_ref(struct sk_buff *skb, int f)
3526{
3527 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3528}
3529
3530int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
3531 unsigned int headroom);
3532int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
3533 struct bpf_prog *prog);
3534bool napi_pp_put_page(struct page *page, bool napi_safe);
3535
3536static inline void
3537skb_page_unref(const struct sk_buff *skb, struct page *page, bool napi_safe)
3538{
3539#ifdef CONFIG_PAGE_POOL
3540 if (skb->pp_recycle && napi_pp_put_page(page, napi_safe))
3541 return;
3542#endif
3543 put_page(page);
3544}
3545
3546static inline void
3547napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe)
3548{
3549 struct page *page = skb_frag_page(frag);
3550
3551#ifdef CONFIG_PAGE_POOL
3552 if (recycle && napi_pp_put_page(page, napi_safe))
3553 return;
3554#endif
3555 put_page(page);
3556}
3557
3558/**
3559 * __skb_frag_unref - release a reference on a paged fragment.
3560 * @frag: the paged fragment
3561 * @recycle: recycle the page if allocated via page_pool
3562 *
3563 * Releases a reference on the paged fragment @frag
3564 * or recycles the page via the page_pool API.
3565 */
3566static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3567{
3568 napi_frag_unref(frag, recycle, false);
3569}
3570
3571/**
3572 * skb_frag_unref - release a reference on a paged fragment of an skb.
3573 * @skb: the buffer
3574 * @f: the fragment offset
3575 *
3576 * Releases a reference on the @f'th paged fragment of @skb.
3577 */
3578static inline void skb_frag_unref(struct sk_buff *skb, int f)
3579{
3580 struct skb_shared_info *shinfo = skb_shinfo(skb);
3581
3582 if (!skb_zcopy_managed(skb))
3583 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3584}
3585
3586/**
3587 * skb_frag_address - gets the address of the data contained in a paged fragment
3588 * @frag: the paged fragment buffer
3589 *
3590 * Returns the address of the data within @frag. The page must already
3591 * be mapped.
3592 */
3593static inline void *skb_frag_address(const skb_frag_t *frag)
3594{
3595 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3596}
3597
3598/**
3599 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3600 * @frag: the paged fragment buffer
3601 *
3602 * Returns the address of the data within @frag. Checks that the page
3603 * is mapped and returns %NULL otherwise.
3604 */
3605static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3606{
3607 void *ptr = page_address(skb_frag_page(frag));
3608 if (unlikely(!ptr))
3609 return NULL;
3610
3611 return ptr + skb_frag_off(frag);
3612}
3613
3614/**
3615 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3616 * @fragto: skb fragment where page is set
3617 * @fragfrom: skb fragment page is copied from
3618 */
3619static inline void skb_frag_page_copy(skb_frag_t *fragto,
3620 const skb_frag_t *fragfrom)
3621{
3622 fragto->netmem = fragfrom->netmem;
3623}
3624
3625bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3626
3627/**
3628 * skb_frag_dma_map - maps a paged fragment via the DMA API
3629 * @dev: the device to map the fragment to
3630 * @frag: the paged fragment to map
3631 * @offset: the offset within the fragment (starting at the
3632 * fragment's own offset)
3633 * @size: the number of bytes to map
3634 * @dir: the direction of the mapping (``PCI_DMA_*``)
3635 *
3636 * Maps the page associated with @frag to @device.
3637 */
3638static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3639 const skb_frag_t *frag,
3640 size_t offset, size_t size,
3641 enum dma_data_direction dir)
3642{
3643 return dma_map_page(dev, skb_frag_page(frag),
3644 skb_frag_off(frag) + offset, size, dir);
3645}
3646
3647static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3648 gfp_t gfp_mask)
3649{
3650 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3651}
3652
3653
3654static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3655 gfp_t gfp_mask)
3656{
3657 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3658}
3659
3660
3661/**
3662 * skb_clone_writable - is the header of a clone writable
3663 * @skb: buffer to check
3664 * @len: length up to which to write
3665 *
3666 * Returns true if modifying the header part of the cloned buffer
3667 * does not requires the data to be copied.
3668 */
3669static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3670{
3671 return !skb_header_cloned(skb) &&
3672 skb_headroom(skb) + len <= skb->hdr_len;
3673}
3674
3675static inline int skb_try_make_writable(struct sk_buff *skb,
3676 unsigned int write_len)
3677{
3678 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3679 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3680}
3681
3682static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3683 int cloned)
3684{
3685 int delta = 0;
3686
3687 if (headroom > skb_headroom(skb))
3688 delta = headroom - skb_headroom(skb);
3689
3690 if (delta || cloned)
3691 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3692 GFP_ATOMIC);
3693 return 0;
3694}
3695
3696/**
3697 * skb_cow - copy header of skb when it is required
3698 * @skb: buffer to cow
3699 * @headroom: needed headroom
3700 *
3701 * If the skb passed lacks sufficient headroom or its data part
3702 * is shared, data is reallocated. If reallocation fails, an error
3703 * is returned and original skb is not changed.
3704 *
3705 * The result is skb with writable area skb->head...skb->tail
3706 * and at least @headroom of space at head.
3707 */
3708static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3709{
3710 return __skb_cow(skb, headroom, skb_cloned(skb));
3711}
3712
3713/**
3714 * skb_cow_head - skb_cow but only making the head writable
3715 * @skb: buffer to cow
3716 * @headroom: needed headroom
3717 *
3718 * This function is identical to skb_cow except that we replace the
3719 * skb_cloned check by skb_header_cloned. It should be used when
3720 * you only need to push on some header and do not need to modify
3721 * the data.
3722 */
3723static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3724{
3725 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3726}
3727
3728/**
3729 * skb_padto - pad an skbuff up to a minimal size
3730 * @skb: buffer to pad
3731 * @len: minimal length
3732 *
3733 * Pads up a buffer to ensure the trailing bytes exist and are
3734 * blanked. If the buffer already contains sufficient data it
3735 * is untouched. Otherwise it is extended. Returns zero on
3736 * success. The skb is freed on error.
3737 */
3738static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3739{
3740 unsigned int size = skb->len;
3741 if (likely(size >= len))
3742 return 0;
3743 return skb_pad(skb, len - size);
3744}
3745
3746/**
3747 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3748 * @skb: buffer to pad
3749 * @len: minimal length
3750 * @free_on_error: free buffer on error
3751 *
3752 * Pads up a buffer to ensure the trailing bytes exist and are
3753 * blanked. If the buffer already contains sufficient data it
3754 * is untouched. Otherwise it is extended. Returns zero on
3755 * success. The skb is freed on error if @free_on_error is true.
3756 */
3757static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3758 unsigned int len,
3759 bool free_on_error)
3760{
3761 unsigned int size = skb->len;
3762
3763 if (unlikely(size < len)) {
3764 len -= size;
3765 if (__skb_pad(skb, len, free_on_error))
3766 return -ENOMEM;
3767 __skb_put(skb, len);
3768 }
3769 return 0;
3770}
3771
3772/**
3773 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3774 * @skb: buffer to pad
3775 * @len: minimal length
3776 *
3777 * Pads up a buffer to ensure the trailing bytes exist and are
3778 * blanked. If the buffer already contains sufficient data it
3779 * is untouched. Otherwise it is extended. Returns zero on
3780 * success. The skb is freed on error.
3781 */
3782static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3783{
3784 return __skb_put_padto(skb, len, true);
3785}
3786
3787bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i)
3788 __must_check;
3789
3790static inline int skb_add_data(struct sk_buff *skb,
3791 struct iov_iter *from, int copy)
3792{
3793 const int off = skb->len;
3794
3795 if (skb->ip_summed == CHECKSUM_NONE) {
3796 __wsum csum = 0;
3797 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3798 &csum, from)) {
3799 skb->csum = csum_block_add(skb->csum, csum, off);
3800 return 0;
3801 }
3802 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3803 return 0;
3804
3805 __skb_trim(skb, off);
3806 return -EFAULT;
3807}
3808
3809static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3810 const struct page *page, int off)
3811{
3812 if (skb_zcopy(skb))
3813 return false;
3814 if (i) {
3815 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3816
3817 return page == skb_frag_page(frag) &&
3818 off == skb_frag_off(frag) + skb_frag_size(frag);
3819 }
3820 return false;
3821}
3822
3823static inline int __skb_linearize(struct sk_buff *skb)
3824{
3825 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3826}
3827
3828/**
3829 * skb_linearize - convert paged skb to linear one
3830 * @skb: buffer to linarize
3831 *
3832 * If there is no free memory -ENOMEM is returned, otherwise zero
3833 * is returned and the old skb data released.
3834 */
3835static inline int skb_linearize(struct sk_buff *skb)
3836{
3837 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3838}
3839
3840/**
3841 * skb_has_shared_frag - can any frag be overwritten
3842 * @skb: buffer to test
3843 *
3844 * Return true if the skb has at least one frag that might be modified
3845 * by an external entity (as in vmsplice()/sendfile())
3846 */
3847static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3848{
3849 return skb_is_nonlinear(skb) &&
3850 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3851}
3852
3853/**
3854 * skb_linearize_cow - make sure skb is linear and writable
3855 * @skb: buffer to process
3856 *
3857 * If there is no free memory -ENOMEM is returned, otherwise zero
3858 * is returned and the old skb data released.
3859 */
3860static inline int skb_linearize_cow(struct sk_buff *skb)
3861{
3862 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3863 __skb_linearize(skb) : 0;
3864}
3865
3866static __always_inline void
3867__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3868 unsigned int off)
3869{
3870 if (skb->ip_summed == CHECKSUM_COMPLETE)
3871 skb->csum = csum_block_sub(skb->csum,
3872 csum_partial(start, len, 0), off);
3873 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3874 skb_checksum_start_offset(skb) < 0)
3875 skb->ip_summed = CHECKSUM_NONE;
3876}
3877
3878/**
3879 * skb_postpull_rcsum - update checksum for received skb after pull
3880 * @skb: buffer to update
3881 * @start: start of data before pull
3882 * @len: length of data pulled
3883 *
3884 * After doing a pull on a received packet, you need to call this to
3885 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3886 * CHECKSUM_NONE so that it can be recomputed from scratch.
3887 */
3888static inline void skb_postpull_rcsum(struct sk_buff *skb,
3889 const void *start, unsigned int len)
3890{
3891 if (skb->ip_summed == CHECKSUM_COMPLETE)
3892 skb->csum = wsum_negate(csum_partial(start, len,
3893 wsum_negate(skb->csum)));
3894 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3895 skb_checksum_start_offset(skb) < 0)
3896 skb->ip_summed = CHECKSUM_NONE;
3897}
3898
3899static __always_inline void
3900__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3901 unsigned int off)
3902{
3903 if (skb->ip_summed == CHECKSUM_COMPLETE)
3904 skb->csum = csum_block_add(skb->csum,
3905 csum_partial(start, len, 0), off);
3906}
3907
3908/**
3909 * skb_postpush_rcsum - update checksum for received skb after push
3910 * @skb: buffer to update
3911 * @start: start of data after push
3912 * @len: length of data pushed
3913 *
3914 * After doing a push on a received packet, you need to call this to
3915 * update the CHECKSUM_COMPLETE checksum.
3916 */
3917static inline void skb_postpush_rcsum(struct sk_buff *skb,
3918 const void *start, unsigned int len)
3919{
3920 __skb_postpush_rcsum(skb, start, len, 0);
3921}
3922
3923void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3924
3925/**
3926 * skb_push_rcsum - push skb and update receive checksum
3927 * @skb: buffer to update
3928 * @len: length of data pulled
3929 *
3930 * This function performs an skb_push on the packet and updates
3931 * the CHECKSUM_COMPLETE checksum. It should be used on
3932 * receive path processing instead of skb_push unless you know
3933 * that the checksum difference is zero (e.g., a valid IP header)
3934 * or you are setting ip_summed to CHECKSUM_NONE.
3935 */
3936static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3937{
3938 skb_push(skb, len);
3939 skb_postpush_rcsum(skb, skb->data, len);
3940 return skb->data;
3941}
3942
3943int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3944/**
3945 * pskb_trim_rcsum - trim received skb and update checksum
3946 * @skb: buffer to trim
3947 * @len: new length
3948 *
3949 * This is exactly the same as pskb_trim except that it ensures the
3950 * checksum of received packets are still valid after the operation.
3951 * It can change skb pointers.
3952 */
3953
3954static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3955{
3956 if (likely(len >= skb->len))
3957 return 0;
3958 return pskb_trim_rcsum_slow(skb, len);
3959}
3960
3961static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3962{
3963 if (skb->ip_summed == CHECKSUM_COMPLETE)
3964 skb->ip_summed = CHECKSUM_NONE;
3965 __skb_trim(skb, len);
3966 return 0;
3967}
3968
3969static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3970{
3971 if (skb->ip_summed == CHECKSUM_COMPLETE)
3972 skb->ip_summed = CHECKSUM_NONE;
3973 return __skb_grow(skb, len);
3974}
3975
3976#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3977#define skb_rb_first(root) rb_to_skb(rb_first(root))
3978#define skb_rb_last(root) rb_to_skb(rb_last(root))
3979#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3980#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3981
3982#define skb_queue_walk(queue, skb) \
3983 for (skb = (queue)->next; \
3984 skb != (struct sk_buff *)(queue); \
3985 skb = skb->next)
3986
3987#define skb_queue_walk_safe(queue, skb, tmp) \
3988 for (skb = (queue)->next, tmp = skb->next; \
3989 skb != (struct sk_buff *)(queue); \
3990 skb = tmp, tmp = skb->next)
3991
3992#define skb_queue_walk_from(queue, skb) \
3993 for (; skb != (struct sk_buff *)(queue); \
3994 skb = skb->next)
3995
3996#define skb_rbtree_walk(skb, root) \
3997 for (skb = skb_rb_first(root); skb != NULL; \
3998 skb = skb_rb_next(skb))
3999
4000#define skb_rbtree_walk_from(skb) \
4001 for (; skb != NULL; \
4002 skb = skb_rb_next(skb))
4003
4004#define skb_rbtree_walk_from_safe(skb, tmp) \
4005 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
4006 skb = tmp)
4007
4008#define skb_queue_walk_from_safe(queue, skb, tmp) \
4009 for (tmp = skb->next; \
4010 skb != (struct sk_buff *)(queue); \
4011 skb = tmp, tmp = skb->next)
4012
4013#define skb_queue_reverse_walk(queue, skb) \
4014 for (skb = (queue)->prev; \
4015 skb != (struct sk_buff *)(queue); \
4016 skb = skb->prev)
4017
4018#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
4019 for (skb = (queue)->prev, tmp = skb->prev; \
4020 skb != (struct sk_buff *)(queue); \
4021 skb = tmp, tmp = skb->prev)
4022
4023#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
4024 for (tmp = skb->prev; \
4025 skb != (struct sk_buff *)(queue); \
4026 skb = tmp, tmp = skb->prev)
4027
4028static inline bool skb_has_frag_list(const struct sk_buff *skb)
4029{
4030 return skb_shinfo(skb)->frag_list != NULL;
4031}
4032
4033static inline void skb_frag_list_init(struct sk_buff *skb)
4034{
4035 skb_shinfo(skb)->frag_list = NULL;
4036}
4037
4038#define skb_walk_frags(skb, iter) \
4039 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
4040
4041
4042int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
4043 int *err, long *timeo_p,
4044 const struct sk_buff *skb);
4045struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
4046 struct sk_buff_head *queue,
4047 unsigned int flags,
4048 int *off, int *err,
4049 struct sk_buff **last);
4050struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
4051 struct sk_buff_head *queue,
4052 unsigned int flags, int *off, int *err,
4053 struct sk_buff **last);
4054struct sk_buff *__skb_recv_datagram(struct sock *sk,
4055 struct sk_buff_head *sk_queue,
4056 unsigned int flags, int *off, int *err);
4057struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
4058__poll_t datagram_poll(struct file *file, struct socket *sock,
4059 struct poll_table_struct *wait);
4060int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
4061 struct iov_iter *to, int size);
4062static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
4063 struct msghdr *msg, int size)
4064{
4065 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
4066}
4067int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
4068 struct msghdr *msg);
4069int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
4070 struct iov_iter *to, int len,
4071 struct ahash_request *hash);
4072int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
4073 struct iov_iter *from, int len);
4074int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
4075void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
4076void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
4077static inline void skb_free_datagram_locked(struct sock *sk,
4078 struct sk_buff *skb)
4079{
4080 __skb_free_datagram_locked(sk, skb, 0);
4081}
4082int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
4083int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
4084int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
4085__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
4086 int len);
4087int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
4088 struct pipe_inode_info *pipe, unsigned int len,
4089 unsigned int flags);
4090int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
4091 int len);
4092int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
4093void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
4094unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
4095int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
4096 int len, int hlen);
4097void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
4098int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
4099void skb_scrub_packet(struct sk_buff *skb, bool xnet);
4100struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
4101struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
4102 unsigned int offset);
4103struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
4104int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
4105int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev);
4106int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
4107int skb_vlan_pop(struct sk_buff *skb);
4108int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4109int skb_eth_pop(struct sk_buff *skb);
4110int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4111 const unsigned char *src);
4112int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4113 int mac_len, bool ethernet);
4114int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4115 bool ethernet);
4116int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4117int skb_mpls_dec_ttl(struct sk_buff *skb);
4118struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4119 gfp_t gfp);
4120
4121static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4122{
4123 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4124}
4125
4126static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4127{
4128 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4129}
4130
4131struct skb_checksum_ops {
4132 __wsum (*update)(const void *mem, int len, __wsum wsum);
4133 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4134};
4135
4136extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4137
4138__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4139 __wsum csum, const struct skb_checksum_ops *ops);
4140__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4141 __wsum csum);
4142
4143static inline void * __must_check
4144__skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4145 const void *data, int hlen, void *buffer)
4146{
4147 if (likely(hlen - offset >= len))
4148 return (void *)data + offset;
4149
4150 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4151 return NULL;
4152
4153 return buffer;
4154}
4155
4156static inline void * __must_check
4157skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4158{
4159 return __skb_header_pointer(skb, offset, len, skb->data,
4160 skb_headlen(skb), buffer);
4161}
4162
4163static inline void * __must_check
4164skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len)
4165{
4166 if (likely(skb_headlen(skb) - offset >= len))
4167 return skb->data + offset;
4168 return NULL;
4169}
4170
4171/**
4172 * skb_needs_linearize - check if we need to linearize a given skb
4173 * depending on the given device features.
4174 * @skb: socket buffer to check
4175 * @features: net device features
4176 *
4177 * Returns true if either:
4178 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4179 * 2. skb is fragmented and the device does not support SG.
4180 */
4181static inline bool skb_needs_linearize(struct sk_buff *skb,
4182 netdev_features_t features)
4183{
4184 return skb_is_nonlinear(skb) &&
4185 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4186 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4187}
4188
4189static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4190 void *to,
4191 const unsigned int len)
4192{
4193 memcpy(to, skb->data, len);
4194}
4195
4196static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4197 const int offset, void *to,
4198 const unsigned int len)
4199{
4200 memcpy(to, skb->data + offset, len);
4201}
4202
4203static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4204 const void *from,
4205 const unsigned int len)
4206{
4207 memcpy(skb->data, from, len);
4208}
4209
4210static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4211 const int offset,
4212 const void *from,
4213 const unsigned int len)
4214{
4215 memcpy(skb->data + offset, from, len);
4216}
4217
4218void skb_init(void);
4219
4220static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4221{
4222 return skb->tstamp;
4223}
4224
4225/**
4226 * skb_get_timestamp - get timestamp from a skb
4227 * @skb: skb to get stamp from
4228 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4229 *
4230 * Timestamps are stored in the skb as offsets to a base timestamp.
4231 * This function converts the offset back to a struct timeval and stores
4232 * it in stamp.
4233 */
4234static inline void skb_get_timestamp(const struct sk_buff *skb,
4235 struct __kernel_old_timeval *stamp)
4236{
4237 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4238}
4239
4240static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4241 struct __kernel_sock_timeval *stamp)
4242{
4243 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4244
4245 stamp->tv_sec = ts.tv_sec;
4246 stamp->tv_usec = ts.tv_nsec / 1000;
4247}
4248
4249static inline void skb_get_timestampns(const struct sk_buff *skb,
4250 struct __kernel_old_timespec *stamp)
4251{
4252 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4253
4254 stamp->tv_sec = ts.tv_sec;
4255 stamp->tv_nsec = ts.tv_nsec;
4256}
4257
4258static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4259 struct __kernel_timespec *stamp)
4260{
4261 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4262
4263 stamp->tv_sec = ts.tv_sec;
4264 stamp->tv_nsec = ts.tv_nsec;
4265}
4266
4267static inline void __net_timestamp(struct sk_buff *skb)
4268{
4269 skb->tstamp = ktime_get_real();
4270 skb->mono_delivery_time = 0;
4271}
4272
4273static inline ktime_t net_timedelta(ktime_t t)
4274{
4275 return ktime_sub(ktime_get_real(), t);
4276}
4277
4278static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4279 bool mono)
4280{
4281 skb->tstamp = kt;
4282 skb->mono_delivery_time = kt && mono;
4283}
4284
4285DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4286
4287/* It is used in the ingress path to clear the delivery_time.
4288 * If needed, set the skb->tstamp to the (rcv) timestamp.
4289 */
4290static inline void skb_clear_delivery_time(struct sk_buff *skb)
4291{
4292 if (skb->mono_delivery_time) {
4293 skb->mono_delivery_time = 0;
4294 if (static_branch_unlikely(&netstamp_needed_key))
4295 skb->tstamp = ktime_get_real();
4296 else
4297 skb->tstamp = 0;
4298 }
4299}
4300
4301static inline void skb_clear_tstamp(struct sk_buff *skb)
4302{
4303 if (skb->mono_delivery_time)
4304 return;
4305
4306 skb->tstamp = 0;
4307}
4308
4309static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4310{
4311 if (skb->mono_delivery_time)
4312 return 0;
4313
4314 return skb->tstamp;
4315}
4316
4317static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4318{
4319 if (!skb->mono_delivery_time && skb->tstamp)
4320 return skb->tstamp;
4321
4322 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4323 return ktime_get_real();
4324
4325 return 0;
4326}
4327
4328static inline u8 skb_metadata_len(const struct sk_buff *skb)
4329{
4330 return skb_shinfo(skb)->meta_len;
4331}
4332
4333static inline void *skb_metadata_end(const struct sk_buff *skb)
4334{
4335 return skb_mac_header(skb);
4336}
4337
4338static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4339 const struct sk_buff *skb_b,
4340 u8 meta_len)
4341{
4342 const void *a = skb_metadata_end(skb_a);
4343 const void *b = skb_metadata_end(skb_b);
4344 u64 diffs = 0;
4345
4346 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
4347 BITS_PER_LONG != 64)
4348 goto slow;
4349
4350 /* Using more efficient variant than plain call to memcmp(). */
4351 switch (meta_len) {
4352#define __it(x, op) (x -= sizeof(u##op))
4353#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4354 case 32: diffs |= __it_diff(a, b, 64);
4355 fallthrough;
4356 case 24: diffs |= __it_diff(a, b, 64);
4357 fallthrough;
4358 case 16: diffs |= __it_diff(a, b, 64);
4359 fallthrough;
4360 case 8: diffs |= __it_diff(a, b, 64);
4361 break;
4362 case 28: diffs |= __it_diff(a, b, 64);
4363 fallthrough;
4364 case 20: diffs |= __it_diff(a, b, 64);
4365 fallthrough;
4366 case 12: diffs |= __it_diff(a, b, 64);
4367 fallthrough;
4368 case 4: diffs |= __it_diff(a, b, 32);
4369 break;
4370 default:
4371slow:
4372 return memcmp(a - meta_len, b - meta_len, meta_len);
4373 }
4374 return diffs;
4375}
4376
4377static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4378 const struct sk_buff *skb_b)
4379{
4380 u8 len_a = skb_metadata_len(skb_a);
4381 u8 len_b = skb_metadata_len(skb_b);
4382
4383 if (!(len_a | len_b))
4384 return false;
4385
4386 return len_a != len_b ?
4387 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4388}
4389
4390static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4391{
4392 skb_shinfo(skb)->meta_len = meta_len;
4393}
4394
4395static inline void skb_metadata_clear(struct sk_buff *skb)
4396{
4397 skb_metadata_set(skb, 0);
4398}
4399
4400struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4401
4402#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4403
4404void skb_clone_tx_timestamp(struct sk_buff *skb);
4405bool skb_defer_rx_timestamp(struct sk_buff *skb);
4406
4407#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4408
4409static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4410{
4411}
4412
4413static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4414{
4415 return false;
4416}
4417
4418#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4419
4420/**
4421 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4422 *
4423 * PHY drivers may accept clones of transmitted packets for
4424 * timestamping via their phy_driver.txtstamp method. These drivers
4425 * must call this function to return the skb back to the stack with a
4426 * timestamp.
4427 *
4428 * @skb: clone of the original outgoing packet
4429 * @hwtstamps: hardware time stamps
4430 *
4431 */
4432void skb_complete_tx_timestamp(struct sk_buff *skb,
4433 struct skb_shared_hwtstamps *hwtstamps);
4434
4435void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4436 struct skb_shared_hwtstamps *hwtstamps,
4437 struct sock *sk, int tstype);
4438
4439/**
4440 * skb_tstamp_tx - queue clone of skb with send time stamps
4441 * @orig_skb: the original outgoing packet
4442 * @hwtstamps: hardware time stamps, may be NULL if not available
4443 *
4444 * If the skb has a socket associated, then this function clones the
4445 * skb (thus sharing the actual data and optional structures), stores
4446 * the optional hardware time stamping information (if non NULL) or
4447 * generates a software time stamp (otherwise), then queues the clone
4448 * to the error queue of the socket. Errors are silently ignored.
4449 */
4450void skb_tstamp_tx(struct sk_buff *orig_skb,
4451 struct skb_shared_hwtstamps *hwtstamps);
4452
4453/**
4454 * skb_tx_timestamp() - Driver hook for transmit timestamping
4455 *
4456 * Ethernet MAC Drivers should call this function in their hard_xmit()
4457 * function immediately before giving the sk_buff to the MAC hardware.
4458 *
4459 * Specifically, one should make absolutely sure that this function is
4460 * called before TX completion of this packet can trigger. Otherwise
4461 * the packet could potentially already be freed.
4462 *
4463 * @skb: A socket buffer.
4464 */
4465static inline void skb_tx_timestamp(struct sk_buff *skb)
4466{
4467 skb_clone_tx_timestamp(skb);
4468 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4469 skb_tstamp_tx(skb, NULL);
4470}
4471
4472/**
4473 * skb_complete_wifi_ack - deliver skb with wifi status
4474 *
4475 * @skb: the original outgoing packet
4476 * @acked: ack status
4477 *
4478 */
4479void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4480
4481__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4482__sum16 __skb_checksum_complete(struct sk_buff *skb);
4483
4484static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4485{
4486 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4487 skb->csum_valid ||
4488 (skb->ip_summed == CHECKSUM_PARTIAL &&
4489 skb_checksum_start_offset(skb) >= 0));
4490}
4491
4492/**
4493 * skb_checksum_complete - Calculate checksum of an entire packet
4494 * @skb: packet to process
4495 *
4496 * This function calculates the checksum over the entire packet plus
4497 * the value of skb->csum. The latter can be used to supply the
4498 * checksum of a pseudo header as used by TCP/UDP. It returns the
4499 * checksum.
4500 *
4501 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4502 * this function can be used to verify that checksum on received
4503 * packets. In that case the function should return zero if the
4504 * checksum is correct. In particular, this function will return zero
4505 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4506 * hardware has already verified the correctness of the checksum.
4507 */
4508static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4509{
4510 return skb_csum_unnecessary(skb) ?
4511 0 : __skb_checksum_complete(skb);
4512}
4513
4514static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4515{
4516 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4517 if (skb->csum_level == 0)
4518 skb->ip_summed = CHECKSUM_NONE;
4519 else
4520 skb->csum_level--;
4521 }
4522}
4523
4524static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4525{
4526 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4527 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4528 skb->csum_level++;
4529 } else if (skb->ip_summed == CHECKSUM_NONE) {
4530 skb->ip_summed = CHECKSUM_UNNECESSARY;
4531 skb->csum_level = 0;
4532 }
4533}
4534
4535static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4536{
4537 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4538 skb->ip_summed = CHECKSUM_NONE;
4539 skb->csum_level = 0;
4540 }
4541}
4542
4543/* Check if we need to perform checksum complete validation.
4544 *
4545 * Returns true if checksum complete is needed, false otherwise
4546 * (either checksum is unnecessary or zero checksum is allowed).
4547 */
4548static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4549 bool zero_okay,
4550 __sum16 check)
4551{
4552 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4553 skb->csum_valid = 1;
4554 __skb_decr_checksum_unnecessary(skb);
4555 return false;
4556 }
4557
4558 return true;
4559}
4560
4561/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4562 * in checksum_init.
4563 */
4564#define CHECKSUM_BREAK 76
4565
4566/* Unset checksum-complete
4567 *
4568 * Unset checksum complete can be done when packet is being modified
4569 * (uncompressed for instance) and checksum-complete value is
4570 * invalidated.
4571 */
4572static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4573{
4574 if (skb->ip_summed == CHECKSUM_COMPLETE)
4575 skb->ip_summed = CHECKSUM_NONE;
4576}
4577
4578/* Validate (init) checksum based on checksum complete.
4579 *
4580 * Return values:
4581 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4582 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4583 * checksum is stored in skb->csum for use in __skb_checksum_complete
4584 * non-zero: value of invalid checksum
4585 *
4586 */
4587static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4588 bool complete,
4589 __wsum psum)
4590{
4591 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4592 if (!csum_fold(csum_add(psum, skb->csum))) {
4593 skb->csum_valid = 1;
4594 return 0;
4595 }
4596 }
4597
4598 skb->csum = psum;
4599
4600 if (complete || skb->len <= CHECKSUM_BREAK) {
4601 __sum16 csum;
4602
4603 csum = __skb_checksum_complete(skb);
4604 skb->csum_valid = !csum;
4605 return csum;
4606 }
4607
4608 return 0;
4609}
4610
4611static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4612{
4613 return 0;
4614}
4615
4616/* Perform checksum validate (init). Note that this is a macro since we only
4617 * want to calculate the pseudo header which is an input function if necessary.
4618 * First we try to validate without any computation (checksum unnecessary) and
4619 * then calculate based on checksum complete calling the function to compute
4620 * pseudo header.
4621 *
4622 * Return values:
4623 * 0: checksum is validated or try to in skb_checksum_complete
4624 * non-zero: value of invalid checksum
4625 */
4626#define __skb_checksum_validate(skb, proto, complete, \
4627 zero_okay, check, compute_pseudo) \
4628({ \
4629 __sum16 __ret = 0; \
4630 skb->csum_valid = 0; \
4631 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4632 __ret = __skb_checksum_validate_complete(skb, \
4633 complete, compute_pseudo(skb, proto)); \
4634 __ret; \
4635})
4636
4637#define skb_checksum_init(skb, proto, compute_pseudo) \
4638 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4639
4640#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4641 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4642
4643#define skb_checksum_validate(skb, proto, compute_pseudo) \
4644 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4645
4646#define skb_checksum_validate_zero_check(skb, proto, check, \
4647 compute_pseudo) \
4648 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4649
4650#define skb_checksum_simple_validate(skb) \
4651 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4652
4653static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4654{
4655 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4656}
4657
4658static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4659{
4660 skb->csum = ~pseudo;
4661 skb->ip_summed = CHECKSUM_COMPLETE;
4662}
4663
4664#define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4665do { \
4666 if (__skb_checksum_convert_check(skb)) \
4667 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4668} while (0)
4669
4670static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4671 u16 start, u16 offset)
4672{
4673 skb->ip_summed = CHECKSUM_PARTIAL;
4674 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4675 skb->csum_offset = offset - start;
4676}
4677
4678/* Update skbuf and packet to reflect the remote checksum offload operation.
4679 * When called, ptr indicates the starting point for skb->csum when
4680 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4681 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4682 */
4683static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4684 int start, int offset, bool nopartial)
4685{
4686 __wsum delta;
4687
4688 if (!nopartial) {
4689 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4690 return;
4691 }
4692
4693 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4694 __skb_checksum_complete(skb);
4695 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4696 }
4697
4698 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4699
4700 /* Adjust skb->csum since we changed the packet */
4701 skb->csum = csum_add(skb->csum, delta);
4702}
4703
4704static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4705{
4706#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4707 return (void *)(skb->_nfct & NFCT_PTRMASK);
4708#else
4709 return NULL;
4710#endif
4711}
4712
4713static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4714{
4715#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4716 return skb->_nfct;
4717#else
4718 return 0UL;
4719#endif
4720}
4721
4722static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4723{
4724#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4725 skb->slow_gro |= !!nfct;
4726 skb->_nfct = nfct;
4727#endif
4728}
4729
4730#ifdef CONFIG_SKB_EXTENSIONS
4731enum skb_ext_id {
4732#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4733 SKB_EXT_BRIDGE_NF,
4734#endif
4735#ifdef CONFIG_XFRM
4736 SKB_EXT_SEC_PATH,
4737#endif
4738#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4739 TC_SKB_EXT,
4740#endif
4741#if IS_ENABLED(CONFIG_MPTCP)
4742 SKB_EXT_MPTCP,
4743#endif
4744#if IS_ENABLED(CONFIG_MCTP_FLOWS)
4745 SKB_EXT_MCTP,
4746#endif
4747 SKB_EXT_NUM, /* must be last */
4748};
4749
4750/**
4751 * struct skb_ext - sk_buff extensions
4752 * @refcnt: 1 on allocation, deallocated on 0
4753 * @offset: offset to add to @data to obtain extension address
4754 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4755 * @data: start of extension data, variable sized
4756 *
4757 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4758 * to use 'u8' types while allowing up to 2kb worth of extension data.
4759 */
4760struct skb_ext {
4761 refcount_t refcnt;
4762 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4763 u8 chunks; /* same */
4764 char data[] __aligned(8);
4765};
4766
4767struct skb_ext *__skb_ext_alloc(gfp_t flags);
4768void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4769 struct skb_ext *ext);
4770void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4771void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4772void __skb_ext_put(struct skb_ext *ext);
4773
4774static inline void skb_ext_put(struct sk_buff *skb)
4775{
4776 if (skb->active_extensions)
4777 __skb_ext_put(skb->extensions);
4778}
4779
4780static inline void __skb_ext_copy(struct sk_buff *dst,
4781 const struct sk_buff *src)
4782{
4783 dst->active_extensions = src->active_extensions;
4784
4785 if (src->active_extensions) {
4786 struct skb_ext *ext = src->extensions;
4787
4788 refcount_inc(&ext->refcnt);
4789 dst->extensions = ext;
4790 }
4791}
4792
4793static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4794{
4795 skb_ext_put(dst);
4796 __skb_ext_copy(dst, src);
4797}
4798
4799static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4800{
4801 return !!ext->offset[i];
4802}
4803
4804static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4805{
4806 return skb->active_extensions & (1 << id);
4807}
4808
4809static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4810{
4811 if (skb_ext_exist(skb, id))
4812 __skb_ext_del(skb, id);
4813}
4814
4815static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4816{
4817 if (skb_ext_exist(skb, id)) {
4818 struct skb_ext *ext = skb->extensions;
4819
4820 return (void *)ext + (ext->offset[id] << 3);
4821 }
4822
4823 return NULL;
4824}
4825
4826static inline void skb_ext_reset(struct sk_buff *skb)
4827{
4828 if (unlikely(skb->active_extensions)) {
4829 __skb_ext_put(skb->extensions);
4830 skb->active_extensions = 0;
4831 }
4832}
4833
4834static inline bool skb_has_extensions(struct sk_buff *skb)
4835{
4836 return unlikely(skb->active_extensions);
4837}
4838#else
4839static inline void skb_ext_put(struct sk_buff *skb) {}
4840static inline void skb_ext_reset(struct sk_buff *skb) {}
4841static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4842static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4843static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4844static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4845#endif /* CONFIG_SKB_EXTENSIONS */
4846
4847static inline void nf_reset_ct(struct sk_buff *skb)
4848{
4849#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4850 nf_conntrack_put(skb_nfct(skb));
4851 skb->_nfct = 0;
4852#endif
4853}
4854
4855static inline void nf_reset_trace(struct sk_buff *skb)
4856{
4857#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4858 skb->nf_trace = 0;
4859#endif
4860}
4861
4862static inline void ipvs_reset(struct sk_buff *skb)
4863{
4864#if IS_ENABLED(CONFIG_IP_VS)
4865 skb->ipvs_property = 0;
4866#endif
4867}
4868
4869/* Note: This doesn't put any conntrack info in dst. */
4870static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4871 bool copy)
4872{
4873#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4874 dst->_nfct = src->_nfct;
4875 nf_conntrack_get(skb_nfct(src));
4876#endif
4877#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4878 if (copy)
4879 dst->nf_trace = src->nf_trace;
4880#endif
4881}
4882
4883static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4884{
4885#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4886 nf_conntrack_put(skb_nfct(dst));
4887#endif
4888 dst->slow_gro = src->slow_gro;
4889 __nf_copy(dst, src, true);
4890}
4891
4892#ifdef CONFIG_NETWORK_SECMARK
4893static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4894{
4895 to->secmark = from->secmark;
4896}
4897
4898static inline void skb_init_secmark(struct sk_buff *skb)
4899{
4900 skb->secmark = 0;
4901}
4902#else
4903static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4904{ }
4905
4906static inline void skb_init_secmark(struct sk_buff *skb)
4907{ }
4908#endif
4909
4910static inline int secpath_exists(const struct sk_buff *skb)
4911{
4912#ifdef CONFIG_XFRM
4913 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4914#else
4915 return 0;
4916#endif
4917}
4918
4919static inline bool skb_irq_freeable(const struct sk_buff *skb)
4920{
4921 return !skb->destructor &&
4922 !secpath_exists(skb) &&
4923 !skb_nfct(skb) &&
4924 !skb->_skb_refdst &&
4925 !skb_has_frag_list(skb);
4926}
4927
4928static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4929{
4930 skb->queue_mapping = queue_mapping;
4931}
4932
4933static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4934{
4935 return skb->queue_mapping;
4936}
4937
4938static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4939{
4940 to->queue_mapping = from->queue_mapping;
4941}
4942
4943static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4944{
4945 skb->queue_mapping = rx_queue + 1;
4946}
4947
4948static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4949{
4950 return skb->queue_mapping - 1;
4951}
4952
4953static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4954{
4955 return skb->queue_mapping != 0;
4956}
4957
4958static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4959{
4960 skb->dst_pending_confirm = val;
4961}
4962
4963static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4964{
4965 return skb->dst_pending_confirm != 0;
4966}
4967
4968static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4969{
4970#ifdef CONFIG_XFRM
4971 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4972#else
4973 return NULL;
4974#endif
4975}
4976
4977static inline bool skb_is_gso(const struct sk_buff *skb)
4978{
4979 return skb_shinfo(skb)->gso_size;
4980}
4981
4982/* Note: Should be called only if skb_is_gso(skb) is true */
4983static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4984{
4985 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4986}
4987
4988/* Note: Should be called only if skb_is_gso(skb) is true */
4989static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4990{
4991 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4992}
4993
4994/* Note: Should be called only if skb_is_gso(skb) is true */
4995static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4996{
4997 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4998}
4999
5000static inline void skb_gso_reset(struct sk_buff *skb)
5001{
5002 skb_shinfo(skb)->gso_size = 0;
5003 skb_shinfo(skb)->gso_segs = 0;
5004 skb_shinfo(skb)->gso_type = 0;
5005}
5006
5007static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
5008 u16 increment)
5009{
5010 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5011 return;
5012 shinfo->gso_size += increment;
5013}
5014
5015static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
5016 u16 decrement)
5017{
5018 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5019 return;
5020 shinfo->gso_size -= decrement;
5021}
5022
5023void __skb_warn_lro_forwarding(const struct sk_buff *skb);
5024
5025static inline bool skb_warn_if_lro(const struct sk_buff *skb)
5026{
5027 /* LRO sets gso_size but not gso_type, whereas if GSO is really
5028 * wanted then gso_type will be set. */
5029 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5030
5031 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
5032 unlikely(shinfo->gso_type == 0)) {
5033 __skb_warn_lro_forwarding(skb);
5034 return true;
5035 }
5036 return false;
5037}
5038
5039static inline void skb_forward_csum(struct sk_buff *skb)
5040{
5041 /* Unfortunately we don't support this one. Any brave souls? */
5042 if (skb->ip_summed == CHECKSUM_COMPLETE)
5043 skb->ip_summed = CHECKSUM_NONE;
5044}
5045
5046/**
5047 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
5048 * @skb: skb to check
5049 *
5050 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
5051 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
5052 * use this helper, to document places where we make this assertion.
5053 */
5054static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5055{
5056 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5057}
5058
5059bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5060
5061int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5062struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5063 unsigned int transport_len,
5064 __sum16(*skb_chkf)(struct sk_buff *skb));
5065
5066/**
5067 * skb_head_is_locked - Determine if the skb->head is locked down
5068 * @skb: skb to check
5069 *
5070 * The head on skbs build around a head frag can be removed if they are
5071 * not cloned. This function returns true if the skb head is locked down
5072 * due to either being allocated via kmalloc, or by being a clone with
5073 * multiple references to the head.
5074 */
5075static inline bool skb_head_is_locked(const struct sk_buff *skb)
5076{
5077 return !skb->head_frag || skb_cloned(skb);
5078}
5079
5080/* Local Checksum Offload.
5081 * Compute outer checksum based on the assumption that the
5082 * inner checksum will be offloaded later.
5083 * See Documentation/networking/checksum-offloads.rst for
5084 * explanation of how this works.
5085 * Fill in outer checksum adjustment (e.g. with sum of outer
5086 * pseudo-header) before calling.
5087 * Also ensure that inner checksum is in linear data area.
5088 */
5089static inline __wsum lco_csum(struct sk_buff *skb)
5090{
5091 unsigned char *csum_start = skb_checksum_start(skb);
5092 unsigned char *l4_hdr = skb_transport_header(skb);
5093 __wsum partial;
5094
5095 /* Start with complement of inner checksum adjustment */
5096 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5097 skb->csum_offset));
5098
5099 /* Add in checksum of our headers (incl. outer checksum
5100 * adjustment filled in by caller) and return result.
5101 */
5102 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5103}
5104
5105static inline bool skb_is_redirected(const struct sk_buff *skb)
5106{
5107 return skb->redirected;
5108}
5109
5110static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5111{
5112 skb->redirected = 1;
5113#ifdef CONFIG_NET_REDIRECT
5114 skb->from_ingress = from_ingress;
5115 if (skb->from_ingress)
5116 skb_clear_tstamp(skb);
5117#endif
5118}
5119
5120static inline void skb_reset_redirect(struct sk_buff *skb)
5121{
5122 skb->redirected = 0;
5123}
5124
5125static inline void skb_set_redirected_noclear(struct sk_buff *skb,
5126 bool from_ingress)
5127{
5128 skb->redirected = 1;
5129#ifdef CONFIG_NET_REDIRECT
5130 skb->from_ingress = from_ingress;
5131#endif
5132}
5133
5134static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5135{
5136#if IS_ENABLED(CONFIG_IP_SCTP)
5137 return skb->csum_not_inet;
5138#else
5139 return 0;
5140#endif
5141}
5142
5143static inline void skb_reset_csum_not_inet(struct sk_buff *skb)
5144{
5145 skb->ip_summed = CHECKSUM_NONE;
5146#if IS_ENABLED(CONFIG_IP_SCTP)
5147 skb->csum_not_inet = 0;
5148#endif
5149}
5150
5151static inline void skb_set_kcov_handle(struct sk_buff *skb,
5152 const u64 kcov_handle)
5153{
5154#ifdef CONFIG_KCOV
5155 skb->kcov_handle = kcov_handle;
5156#endif
5157}
5158
5159static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5160{
5161#ifdef CONFIG_KCOV
5162 return skb->kcov_handle;
5163#else
5164 return 0;
5165#endif
5166}
5167
5168static inline void skb_mark_for_recycle(struct sk_buff *skb)
5169{
5170#ifdef CONFIG_PAGE_POOL
5171 skb->pp_recycle = 1;
5172#endif
5173}
5174
5175ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
5176 ssize_t maxsize, gfp_t gfp);
5177
5178#endif /* __KERNEL__ */
5179#endif /* _LINUX_SKBUFF_H */