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
356/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
357 * segment using its current segmentation instead.
358 */
359#define GSO_BY_FRAGS 0xFFFF
360
361typedef struct skb_frag {
362 netmem_ref netmem;
363 unsigned int len;
364 unsigned int offset;
365} skb_frag_t;
366
367/**
368 * skb_frag_size() - Returns the size of a skb fragment
369 * @frag: skb fragment
370 */
371static inline unsigned int skb_frag_size(const skb_frag_t *frag)
372{
373 return frag->len;
374}
375
376/**
377 * skb_frag_size_set() - Sets the size of a skb fragment
378 * @frag: skb fragment
379 * @size: size of fragment
380 */
381static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
382{
383 frag->len = size;
384}
385
386/**
387 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
388 * @frag: skb fragment
389 * @delta: value to add
390 */
391static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
392{
393 frag->len += delta;
394}
395
396/**
397 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
398 * @frag: skb fragment
399 * @delta: value to subtract
400 */
401static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
402{
403 frag->len -= delta;
404}
405
406/**
407 * skb_frag_must_loop - Test if %p is a high memory page
408 * @p: fragment's page
409 */
410static inline bool skb_frag_must_loop(struct page *p)
411{
412#if defined(CONFIG_HIGHMEM)
413 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
414 return true;
415#endif
416 return false;
417}
418
419/**
420 * skb_frag_foreach_page - loop over pages in a fragment
421 *
422 * @f: skb frag to operate on
423 * @f_off: offset from start of f->netmem
424 * @f_len: length from f_off to loop over
425 * @p: (temp var) current page
426 * @p_off: (temp var) offset from start of current page,
427 * non-zero only on first page.
428 * @p_len: (temp var) length in current page,
429 * < PAGE_SIZE only on first and last page.
430 * @copied: (temp var) length so far, excluding current p_len.
431 *
432 * A fragment can hold a compound page, in which case per-page
433 * operations, notably kmap_atomic, must be called for each
434 * regular page.
435 */
436#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
437 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
438 p_off = (f_off) & (PAGE_SIZE - 1), \
439 p_len = skb_frag_must_loop(p) ? \
440 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
441 copied = 0; \
442 copied < f_len; \
443 copied += p_len, p++, p_off = 0, \
444 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
445
446/**
447 * struct skb_shared_hwtstamps - hardware time stamps
448 * @hwtstamp: hardware time stamp transformed into duration
449 * since arbitrary point in time
450 * @netdev_data: address/cookie of network device driver used as
451 * reference to actual hardware time stamp
452 *
453 * Software time stamps generated by ktime_get_real() are stored in
454 * skb->tstamp.
455 *
456 * hwtstamps can only be compared against other hwtstamps from
457 * the same device.
458 *
459 * This structure is attached to packets as part of the
460 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
461 */
462struct skb_shared_hwtstamps {
463 union {
464 ktime_t hwtstamp;
465 void *netdev_data;
466 };
467};
468
469/* Definitions for tx_flags in struct skb_shared_info */
470enum {
471 /* generate hardware time stamp */
472 SKBTX_HW_TSTAMP = 1 << 0,
473
474 /* generate software time stamp when queueing packet to NIC */
475 SKBTX_SW_TSTAMP = 1 << 1,
476
477 /* device driver is going to provide hardware time stamp */
478 SKBTX_IN_PROGRESS = 1 << 2,
479
480 /* generate hardware time stamp based on cycles if supported */
481 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
482
483 /* generate wifi status information (where possible) */
484 SKBTX_WIFI_STATUS = 1 << 4,
485
486 /* determine hardware time stamp based on time or cycles */
487 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
488
489 /* generate software time stamp when entering packet scheduling */
490 SKBTX_SCHED_TSTAMP = 1 << 6,
491};
492
493#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
494 SKBTX_SCHED_TSTAMP)
495#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
496 SKBTX_HW_TSTAMP_USE_CYCLES | \
497 SKBTX_ANY_SW_TSTAMP)
498
499/* Definitions for flags in struct skb_shared_info */
500enum {
501 /* use zcopy routines */
502 SKBFL_ZEROCOPY_ENABLE = BIT(0),
503
504 /* This indicates at least one fragment might be overwritten
505 * (as in vmsplice(), sendfile() ...)
506 * If we need to compute a TX checksum, we'll need to copy
507 * all frags to avoid possible bad checksum
508 */
509 SKBFL_SHARED_FRAG = BIT(1),
510
511 /* segment contains only zerocopy data and should not be
512 * charged to the kernel memory.
513 */
514 SKBFL_PURE_ZEROCOPY = BIT(2),
515
516 SKBFL_DONT_ORPHAN = BIT(3),
517
518 /* page references are managed by the ubuf_info, so it's safe to
519 * use frags only up until ubuf_info is released
520 */
521 SKBFL_MANAGED_FRAG_REFS = BIT(4),
522};
523
524#define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
525#define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
526 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
527
528struct ubuf_info_ops {
529 void (*complete)(struct sk_buff *, struct ubuf_info *,
530 bool zerocopy_success);
531 /* has to be compatible with skb_zcopy_set() */
532 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg);
533};
534
535/*
536 * The callback notifies userspace to release buffers when skb DMA is done in
537 * lower device, the skb last reference should be 0 when calling this.
538 * The zerocopy_success argument is true if zero copy transmit occurred,
539 * false on data copy or out of memory error caused by data copy attempt.
540 * The ctx field is used to track device context.
541 * The desc field is used to track userspace buffer index.
542 */
543struct ubuf_info {
544 const struct ubuf_info_ops *ops;
545 refcount_t refcnt;
546 u8 flags;
547};
548
549struct ubuf_info_msgzc {
550 struct ubuf_info ubuf;
551
552 union {
553 struct {
554 unsigned long desc;
555 void *ctx;
556 };
557 struct {
558 u32 id;
559 u16 len;
560 u16 zerocopy:1;
561 u32 bytelen;
562 };
563 };
564
565 struct mmpin {
566 struct user_struct *user;
567 unsigned int num_pg;
568 } mmp;
569};
570
571#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
572#define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
573 ubuf)
574
575int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
576void mm_unaccount_pinned_pages(struct mmpin *mmp);
577
578/* Preserve some data across TX submission and completion.
579 *
580 * Note, this state is stored in the driver. Extending the layout
581 * might need some special care.
582 */
583struct xsk_tx_metadata_compl {
584 __u64 *tx_timestamp;
585};
586
587/* This data is invariant across clones and lives at
588 * the end of the header data, ie. at skb->end.
589 */
590struct skb_shared_info {
591 __u8 flags;
592 __u8 meta_len;
593 __u8 nr_frags;
594 __u8 tx_flags;
595 unsigned short gso_size;
596 /* Warning: this field is not always filled in (UFO)! */
597 unsigned short gso_segs;
598 struct sk_buff *frag_list;
599 union {
600 struct skb_shared_hwtstamps hwtstamps;
601 struct xsk_tx_metadata_compl xsk_meta;
602 };
603 unsigned int gso_type;
604 u32 tskey;
605
606 /*
607 * Warning : all fields before dataref are cleared in __alloc_skb()
608 */
609 atomic_t dataref;
610 unsigned int xdp_frags_size;
611
612 /* Intermediate layers must ensure that destructor_arg
613 * remains valid until skb destructor */
614 void * destructor_arg;
615
616 /* must be last field, see pskb_expand_head() */
617 skb_frag_t frags[MAX_SKB_FRAGS];
618};
619
620/**
621 * DOC: dataref and headerless skbs
622 *
623 * Transport layers send out clones of payload skbs they hold for
624 * retransmissions. To allow lower layers of the stack to prepend their headers
625 * we split &skb_shared_info.dataref into two halves.
626 * The lower 16 bits count the overall number of references.
627 * The higher 16 bits indicate how many of the references are payload-only.
628 * skb_header_cloned() checks if skb is allowed to add / write the headers.
629 *
630 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
631 * (via __skb_header_release()). Any clone created from marked skb will get
632 * &sk_buff.hdr_len populated with the available headroom.
633 * If there's the only clone in existence it's able to modify the headroom
634 * at will. The sequence of calls inside the transport layer is::
635 *
636 * <alloc skb>
637 * skb_reserve()
638 * __skb_header_release()
639 * skb_clone()
640 * // send the clone down the stack
641 *
642 * This is not a very generic construct and it depends on the transport layers
643 * doing the right thing. In practice there's usually only one payload-only skb.
644 * Having multiple payload-only skbs with different lengths of hdr_len is not
645 * possible. The payload-only skbs should never leave their owner.
646 */
647#define SKB_DATAREF_SHIFT 16
648#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
649
650
651enum {
652 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
653 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
654 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
655};
656
657enum {
658 SKB_GSO_TCPV4 = 1 << 0,
659
660 /* This indicates the skb is from an untrusted source. */
661 SKB_GSO_DODGY = 1 << 1,
662
663 /* This indicates the tcp segment has CWR set. */
664 SKB_GSO_TCP_ECN = 1 << 2,
665
666 SKB_GSO_TCP_FIXEDID = 1 << 3,
667
668 SKB_GSO_TCPV6 = 1 << 4,
669
670 SKB_GSO_FCOE = 1 << 5,
671
672 SKB_GSO_GRE = 1 << 6,
673
674 SKB_GSO_GRE_CSUM = 1 << 7,
675
676 SKB_GSO_IPXIP4 = 1 << 8,
677
678 SKB_GSO_IPXIP6 = 1 << 9,
679
680 SKB_GSO_UDP_TUNNEL = 1 << 10,
681
682 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
683
684 SKB_GSO_PARTIAL = 1 << 12,
685
686 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
687
688 SKB_GSO_SCTP = 1 << 14,
689
690 SKB_GSO_ESP = 1 << 15,
691
692 SKB_GSO_UDP = 1 << 16,
693
694 SKB_GSO_UDP_L4 = 1 << 17,
695
696 SKB_GSO_FRAGLIST = 1 << 18,
697};
698
699#if BITS_PER_LONG > 32
700#define NET_SKBUFF_DATA_USES_OFFSET 1
701#endif
702
703#ifdef NET_SKBUFF_DATA_USES_OFFSET
704typedef unsigned int sk_buff_data_t;
705#else
706typedef unsigned char *sk_buff_data_t;
707#endif
708
709enum skb_tstamp_type {
710 SKB_CLOCK_REALTIME,
711 SKB_CLOCK_MONOTONIC,
712 SKB_CLOCK_TAI,
713 __SKB_CLOCK_MAX = SKB_CLOCK_TAI,
714};
715
716/**
717 * DOC: Basic sk_buff geometry
718 *
719 * struct sk_buff itself is a metadata structure and does not hold any packet
720 * data. All the data is held in associated buffers.
721 *
722 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
723 * into two parts:
724 *
725 * - data buffer, containing headers and sometimes payload;
726 * this is the part of the skb operated on by the common helpers
727 * such as skb_put() or skb_pull();
728 * - shared info (struct skb_shared_info) which holds an array of pointers
729 * to read-only data in the (page, offset, length) format.
730 *
731 * Optionally &skb_shared_info.frag_list may point to another skb.
732 *
733 * Basic diagram may look like this::
734 *
735 * ---------------
736 * | sk_buff |
737 * ---------------
738 * ,--------------------------- + head
739 * / ,----------------- + data
740 * / / ,----------- + tail
741 * | | | , + end
742 * | | | |
743 * v v v v
744 * -----------------------------------------------
745 * | headroom | data | tailroom | skb_shared_info |
746 * -----------------------------------------------
747 * + [page frag]
748 * + [page frag]
749 * + [page frag]
750 * + [page frag] ---------
751 * + frag_list --> | sk_buff |
752 * ---------
753 *
754 */
755
756/**
757 * struct sk_buff - socket buffer
758 * @next: Next buffer in list
759 * @prev: Previous buffer in list
760 * @tstamp: Time we arrived/left
761 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
762 * for retransmit timer
763 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
764 * @list: queue head
765 * @ll_node: anchor in an llist (eg socket defer_list)
766 * @sk: Socket we are owned by
767 * @dev: Device we arrived on/are leaving by
768 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
769 * @cb: Control buffer. Free for use by every layer. Put private vars here
770 * @_skb_refdst: destination entry (with norefcount bit)
771 * @len: Length of actual data
772 * @data_len: Data length
773 * @mac_len: Length of link layer header
774 * @hdr_len: writable header length of cloned skb
775 * @csum: Checksum (must include start/offset pair)
776 * @csum_start: Offset from skb->head where checksumming should start
777 * @csum_offset: Offset from csum_start where checksum should be stored
778 * @priority: Packet queueing priority
779 * @ignore_df: allow local fragmentation
780 * @cloned: Head may be cloned (check refcnt to be sure)
781 * @ip_summed: Driver fed us an IP checksum
782 * @nohdr: Payload reference only, must not modify header
783 * @pkt_type: Packet class
784 * @fclone: skbuff clone status
785 * @ipvs_property: skbuff is owned by ipvs
786 * @inner_protocol_type: whether the inner protocol is
787 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
788 * @remcsum_offload: remote checksum offload is enabled
789 * @offload_fwd_mark: Packet was L2-forwarded in hardware
790 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
791 * @tc_skip_classify: do not classify packet. set by IFB device
792 * @tc_at_ingress: used within tc_classify to distinguish in/egress
793 * @redirected: packet was redirected by packet classifier
794 * @from_ingress: packet was redirected from the ingress path
795 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
796 * @peeked: this packet has been seen already, so stats have been
797 * done for it, don't do them again
798 * @nf_trace: netfilter packet trace flag
799 * @protocol: Packet protocol from driver
800 * @destructor: Destruct function
801 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
802 * @_sk_redir: socket redirection information for skmsg
803 * @_nfct: Associated connection, if any (with nfctinfo bits)
804 * @skb_iif: ifindex of device we arrived on
805 * @tc_index: Traffic control index
806 * @hash: the packet hash
807 * @queue_mapping: Queue mapping for multiqueue devices
808 * @head_frag: skb was allocated from page fragments,
809 * not allocated by kmalloc() or vmalloc().
810 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
811 * @pp_recycle: mark the packet for recycling instead of freeing (implies
812 * page_pool support on driver)
813 * @active_extensions: active extensions (skb_ext_id types)
814 * @ndisc_nodetype: router type (from link layer)
815 * @ooo_okay: allow the mapping of a socket to a queue to be changed
816 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
817 * ports.
818 * @sw_hash: indicates hash was computed in software stack
819 * @wifi_acked_valid: wifi_acked was set
820 * @wifi_acked: whether frame was acked on wifi or not
821 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
822 * @encapsulation: indicates the inner headers in the skbuff are valid
823 * @encap_hdr_csum: software checksum is needed
824 * @csum_valid: checksum is already valid
825 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
826 * @csum_complete_sw: checksum was completed by software
827 * @csum_level: indicates the number of consecutive checksums found in
828 * the packet minus one that have been verified as
829 * CHECKSUM_UNNECESSARY (max 3)
830 * @unreadable: indicates that at least 1 of the fragments in this skb is
831 * unreadable.
832 * @dst_pending_confirm: need to confirm neighbour
833 * @decrypted: Decrypted SKB
834 * @slow_gro: state present at GRO time, slower prepare step required
835 * @tstamp_type: When set, skb->tstamp has the
836 * delivery_time clock base of skb->tstamp.
837 * @napi_id: id of the NAPI struct this skb came from
838 * @sender_cpu: (aka @napi_id) source CPU in XPS
839 * @alloc_cpu: CPU which did the skb allocation.
840 * @secmark: security marking
841 * @mark: Generic packet mark
842 * @reserved_tailroom: (aka @mark) number of bytes of free space available
843 * at the tail of an sk_buff
844 * @vlan_all: vlan fields (proto & tci)
845 * @vlan_proto: vlan encapsulation protocol
846 * @vlan_tci: vlan tag control information
847 * @inner_protocol: Protocol (encapsulation)
848 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
849 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
850 * @inner_transport_header: Inner transport layer header (encapsulation)
851 * @inner_network_header: Network layer header (encapsulation)
852 * @inner_mac_header: Link layer header (encapsulation)
853 * @transport_header: Transport layer header
854 * @network_header: Network layer header
855 * @mac_header: Link layer header
856 * @kcov_handle: KCOV remote handle for remote coverage collection
857 * @tail: Tail pointer
858 * @end: End pointer
859 * @head: Head of buffer
860 * @data: Data head pointer
861 * @truesize: Buffer size
862 * @users: User count - see {datagram,tcp}.c
863 * @extensions: allocated extensions, valid if active_extensions is nonzero
864 */
865
866struct sk_buff {
867 union {
868 struct {
869 /* These two members must be first to match sk_buff_head. */
870 struct sk_buff *next;
871 struct sk_buff *prev;
872
873 union {
874 struct net_device *dev;
875 /* Some protocols might use this space to store information,
876 * while device pointer would be NULL.
877 * UDP receive path is one user.
878 */
879 unsigned long dev_scratch;
880 };
881 };
882 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
883 struct list_head list;
884 struct llist_node ll_node;
885 };
886
887 struct sock *sk;
888
889 union {
890 ktime_t tstamp;
891 u64 skb_mstamp_ns; /* earliest departure time */
892 };
893 /*
894 * This is the control buffer. It is free to use for every
895 * layer. Please put your private variables there. If you
896 * want to keep them across layers you have to do a skb_clone()
897 * first. This is owned by whoever has the skb queued ATM.
898 */
899 char cb[48] __aligned(8);
900
901 union {
902 struct {
903 unsigned long _skb_refdst;
904 void (*destructor)(struct sk_buff *skb);
905 };
906 struct list_head tcp_tsorted_anchor;
907#ifdef CONFIG_NET_SOCK_MSG
908 unsigned long _sk_redir;
909#endif
910 };
911
912#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
913 unsigned long _nfct;
914#endif
915 unsigned int len,
916 data_len;
917 __u16 mac_len,
918 hdr_len;
919
920 /* Following fields are _not_ copied in __copy_skb_header()
921 * Note that queue_mapping is here mostly to fill a hole.
922 */
923 __u16 queue_mapping;
924
925/* if you move cloned around you also must adapt those constants */
926#ifdef __BIG_ENDIAN_BITFIELD
927#define CLONED_MASK (1 << 7)
928#else
929#define CLONED_MASK 1
930#endif
931#define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
932
933 /* private: */
934 __u8 __cloned_offset[0];
935 /* public: */
936 __u8 cloned:1,
937 nohdr:1,
938 fclone:2,
939 peeked:1,
940 head_frag:1,
941 pfmemalloc:1,
942 pp_recycle:1; /* page_pool recycle indicator */
943#ifdef CONFIG_SKB_EXTENSIONS
944 __u8 active_extensions;
945#endif
946
947 /* Fields enclosed in headers group are copied
948 * using a single memcpy() in __copy_skb_header()
949 */
950 struct_group(headers,
951
952 /* private: */
953 __u8 __pkt_type_offset[0];
954 /* public: */
955 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
956 __u8 ignore_df:1;
957 __u8 dst_pending_confirm:1;
958 __u8 ip_summed:2;
959 __u8 ooo_okay:1;
960
961 /* private: */
962 __u8 __mono_tc_offset[0];
963 /* public: */
964 __u8 tstamp_type:2; /* See skb_tstamp_type */
965#ifdef CONFIG_NET_XGRESS
966 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
967 __u8 tc_skip_classify:1;
968#endif
969 __u8 remcsum_offload:1;
970 __u8 csum_complete_sw:1;
971 __u8 csum_level:2;
972 __u8 inner_protocol_type:1;
973
974 __u8 l4_hash:1;
975 __u8 sw_hash:1;
976#ifdef CONFIG_WIRELESS
977 __u8 wifi_acked_valid:1;
978 __u8 wifi_acked:1;
979#endif
980 __u8 no_fcs:1;
981 /* Indicates the inner headers are valid in the skbuff. */
982 __u8 encapsulation:1;
983 __u8 encap_hdr_csum:1;
984 __u8 csum_valid:1;
985#ifdef CONFIG_IPV6_NDISC_NODETYPE
986 __u8 ndisc_nodetype:2;
987#endif
988
989#if IS_ENABLED(CONFIG_IP_VS)
990 __u8 ipvs_property:1;
991#endif
992#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
993 __u8 nf_trace:1;
994#endif
995#ifdef CONFIG_NET_SWITCHDEV
996 __u8 offload_fwd_mark:1;
997 __u8 offload_l3_fwd_mark:1;
998#endif
999 __u8 redirected:1;
1000#ifdef CONFIG_NET_REDIRECT
1001 __u8 from_ingress:1;
1002#endif
1003#ifdef CONFIG_NETFILTER_SKIP_EGRESS
1004 __u8 nf_skip_egress:1;
1005#endif
1006#ifdef CONFIG_SKB_DECRYPTED
1007 __u8 decrypted:1;
1008#endif
1009 __u8 slow_gro:1;
1010#if IS_ENABLED(CONFIG_IP_SCTP)
1011 __u8 csum_not_inet:1;
1012#endif
1013 __u8 unreadable:1;
1014#if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS)
1015 __u16 tc_index; /* traffic control index */
1016#endif
1017
1018 u16 alloc_cpu;
1019
1020 union {
1021 __wsum csum;
1022 struct {
1023 __u16 csum_start;
1024 __u16 csum_offset;
1025 };
1026 };
1027 __u32 priority;
1028 int skb_iif;
1029 __u32 hash;
1030 union {
1031 u32 vlan_all;
1032 struct {
1033 __be16 vlan_proto;
1034 __u16 vlan_tci;
1035 };
1036 };
1037#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1038 union {
1039 unsigned int napi_id;
1040 unsigned int sender_cpu;
1041 };
1042#endif
1043#ifdef CONFIG_NETWORK_SECMARK
1044 __u32 secmark;
1045#endif
1046
1047 union {
1048 __u32 mark;
1049 __u32 reserved_tailroom;
1050 };
1051
1052 union {
1053 __be16 inner_protocol;
1054 __u8 inner_ipproto;
1055 };
1056
1057 __u16 inner_transport_header;
1058 __u16 inner_network_header;
1059 __u16 inner_mac_header;
1060
1061 __be16 protocol;
1062 __u16 transport_header;
1063 __u16 network_header;
1064 __u16 mac_header;
1065
1066#ifdef CONFIG_KCOV
1067 u64 kcov_handle;
1068#endif
1069
1070 ); /* end headers group */
1071
1072 /* These elements must be at the end, see alloc_skb() for details. */
1073 sk_buff_data_t tail;
1074 sk_buff_data_t end;
1075 unsigned char *head,
1076 *data;
1077 unsigned int truesize;
1078 refcount_t users;
1079
1080#ifdef CONFIG_SKB_EXTENSIONS
1081 /* only usable after checking ->active_extensions != 0 */
1082 struct skb_ext *extensions;
1083#endif
1084};
1085
1086/* if you move pkt_type around you also must adapt those constants */
1087#ifdef __BIG_ENDIAN_BITFIELD
1088#define PKT_TYPE_MAX (7 << 5)
1089#else
1090#define PKT_TYPE_MAX 7
1091#endif
1092#define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1093
1094/* if you move tc_at_ingress or tstamp_type
1095 * around, you also must adapt these constants.
1096 */
1097#ifdef __BIG_ENDIAN_BITFIELD
1098#define SKB_TSTAMP_TYPE_MASK (3 << 6)
1099#define SKB_TSTAMP_TYPE_RSHIFT (6)
1100#define TC_AT_INGRESS_MASK (1 << 5)
1101#else
1102#define SKB_TSTAMP_TYPE_MASK (3)
1103#define TC_AT_INGRESS_MASK (1 << 2)
1104#endif
1105#define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)
1106
1107#ifdef __KERNEL__
1108/*
1109 * Handling routines are only of interest to the kernel
1110 */
1111
1112#define SKB_ALLOC_FCLONE 0x01
1113#define SKB_ALLOC_RX 0x02
1114#define SKB_ALLOC_NAPI 0x04
1115
1116/**
1117 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1118 * @skb: buffer
1119 */
1120static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1121{
1122 return unlikely(skb->pfmemalloc);
1123}
1124
1125/*
1126 * skb might have a dst pointer attached, refcounted or not.
1127 * _skb_refdst low order bit is set if refcount was _not_ taken
1128 */
1129#define SKB_DST_NOREF 1UL
1130#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1131
1132/**
1133 * skb_dst - returns skb dst_entry
1134 * @skb: buffer
1135 *
1136 * Returns skb dst_entry, regardless of reference taken or not.
1137 */
1138static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1139{
1140 /* If refdst was not refcounted, check we still are in a
1141 * rcu_read_lock section
1142 */
1143 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1144 !rcu_read_lock_held() &&
1145 !rcu_read_lock_bh_held());
1146 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1147}
1148
1149/**
1150 * skb_dst_set - sets skb dst
1151 * @skb: buffer
1152 * @dst: dst entry
1153 *
1154 * Sets skb dst, assuming a reference was taken on dst and should
1155 * be released by skb_dst_drop()
1156 */
1157static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1158{
1159 skb->slow_gro |= !!dst;
1160 skb->_skb_refdst = (unsigned long)dst;
1161}
1162
1163/**
1164 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1165 * @skb: buffer
1166 * @dst: dst entry
1167 *
1168 * Sets skb dst, assuming a reference was not taken on dst.
1169 * If dst entry is cached, we do not take reference and dst_release
1170 * will be avoided by refdst_drop. If dst entry is not cached, we take
1171 * reference, so that last dst_release can destroy the dst immediately.
1172 */
1173static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1174{
1175 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1176 skb->slow_gro |= !!dst;
1177 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1178}
1179
1180/**
1181 * skb_dst_is_noref - Test if skb dst isn't refcounted
1182 * @skb: buffer
1183 */
1184static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1185{
1186 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1187}
1188
1189/* For mangling skb->pkt_type from user space side from applications
1190 * such as nft, tc, etc, we only allow a conservative subset of
1191 * possible pkt_types to be set.
1192*/
1193static inline bool skb_pkt_type_ok(u32 ptype)
1194{
1195 return ptype <= PACKET_OTHERHOST;
1196}
1197
1198/**
1199 * skb_napi_id - Returns the skb's NAPI id
1200 * @skb: buffer
1201 */
1202static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1203{
1204#ifdef CONFIG_NET_RX_BUSY_POLL
1205 return skb->napi_id;
1206#else
1207 return 0;
1208#endif
1209}
1210
1211static inline bool skb_wifi_acked_valid(const struct sk_buff *skb)
1212{
1213#ifdef CONFIG_WIRELESS
1214 return skb->wifi_acked_valid;
1215#else
1216 return 0;
1217#endif
1218}
1219
1220/**
1221 * skb_unref - decrement the skb's reference count
1222 * @skb: buffer
1223 *
1224 * Returns true if we can free the skb.
1225 */
1226static inline bool skb_unref(struct sk_buff *skb)
1227{
1228 if (unlikely(!skb))
1229 return false;
1230 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1))
1231 smp_rmb();
1232 else if (likely(!refcount_dec_and_test(&skb->users)))
1233 return false;
1234
1235 return true;
1236}
1237
1238static inline bool skb_data_unref(const struct sk_buff *skb,
1239 struct skb_shared_info *shinfo)
1240{
1241 int bias;
1242
1243 if (!skb->cloned)
1244 return true;
1245
1246 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
1247
1248 if (atomic_read(&shinfo->dataref) == bias)
1249 smp_rmb();
1250 else if (atomic_sub_return(bias, &shinfo->dataref))
1251 return false;
1252
1253 return true;
1254}
1255
1256void __fix_address sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb,
1257 enum skb_drop_reason reason);
1258
1259static inline void
1260kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
1261{
1262 sk_skb_reason_drop(NULL, skb, reason);
1263}
1264
1265/**
1266 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1267 * @skb: buffer to free
1268 */
1269static inline void kfree_skb(struct sk_buff *skb)
1270{
1271 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1272}
1273
1274void skb_release_head_state(struct sk_buff *skb);
1275void kfree_skb_list_reason(struct sk_buff *segs,
1276 enum skb_drop_reason reason);
1277void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1278void skb_tx_error(struct sk_buff *skb);
1279
1280static inline void kfree_skb_list(struct sk_buff *segs)
1281{
1282 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1283}
1284
1285#ifdef CONFIG_TRACEPOINTS
1286void consume_skb(struct sk_buff *skb);
1287#else
1288static inline void consume_skb(struct sk_buff *skb)
1289{
1290 return kfree_skb(skb);
1291}
1292#endif
1293
1294void __consume_stateless_skb(struct sk_buff *skb);
1295void __kfree_skb(struct sk_buff *skb);
1296
1297void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1298bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1299 bool *fragstolen, int *delta_truesize);
1300
1301struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1302 int node);
1303struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1304struct sk_buff *build_skb(void *data, unsigned int frag_size);
1305struct sk_buff *build_skb_around(struct sk_buff *skb,
1306 void *data, unsigned int frag_size);
1307void skb_attempt_defer_free(struct sk_buff *skb);
1308
1309struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1310struct sk_buff *slab_build_skb(void *data);
1311
1312/**
1313 * alloc_skb - allocate a network buffer
1314 * @size: size to allocate
1315 * @priority: allocation mask
1316 *
1317 * This function is a convenient wrapper around __alloc_skb().
1318 */
1319static inline struct sk_buff *alloc_skb(unsigned int size,
1320 gfp_t priority)
1321{
1322 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1323}
1324
1325struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1326 unsigned long data_len,
1327 int max_page_order,
1328 int *errcode,
1329 gfp_t gfp_mask);
1330struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1331
1332/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1333struct sk_buff_fclones {
1334 struct sk_buff skb1;
1335
1336 struct sk_buff skb2;
1337
1338 refcount_t fclone_ref;
1339};
1340
1341/**
1342 * skb_fclone_busy - check if fclone is busy
1343 * @sk: socket
1344 * @skb: buffer
1345 *
1346 * Returns true if skb is a fast clone, and its clone is not freed.
1347 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1348 * so we also check that didn't happen.
1349 */
1350static inline bool skb_fclone_busy(const struct sock *sk,
1351 const struct sk_buff *skb)
1352{
1353 const struct sk_buff_fclones *fclones;
1354
1355 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1356
1357 return skb->fclone == SKB_FCLONE_ORIG &&
1358 refcount_read(&fclones->fclone_ref) > 1 &&
1359 READ_ONCE(fclones->skb2.sk) == sk;
1360}
1361
1362/**
1363 * alloc_skb_fclone - allocate a network buffer from fclone cache
1364 * @size: size to allocate
1365 * @priority: allocation mask
1366 *
1367 * This function is a convenient wrapper around __alloc_skb().
1368 */
1369static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1370 gfp_t priority)
1371{
1372 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1373}
1374
1375struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1376void skb_headers_offset_update(struct sk_buff *skb, int off);
1377int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1378struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1379void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1380struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1381struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1382 gfp_t gfp_mask, bool fclone);
1383static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1384 gfp_t gfp_mask)
1385{
1386 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1387}
1388
1389int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1390struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1391 unsigned int headroom);
1392struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1393struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1394 int newtailroom, gfp_t priority);
1395int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1396 int offset, int len);
1397int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1398 int offset, int len);
1399int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1400int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1401
1402/**
1403 * skb_pad - zero pad the tail of an skb
1404 * @skb: buffer to pad
1405 * @pad: space to pad
1406 *
1407 * Ensure that a buffer is followed by a padding area that is zero
1408 * filled. Used by network drivers which may DMA or transfer data
1409 * beyond the buffer end onto the wire.
1410 *
1411 * May return error in out of memory cases. The skb is freed on error.
1412 */
1413static inline int skb_pad(struct sk_buff *skb, int pad)
1414{
1415 return __skb_pad(skb, pad, true);
1416}
1417#define dev_kfree_skb(a) consume_skb(a)
1418
1419int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1420 int offset, size_t size, size_t max_frags);
1421
1422struct skb_seq_state {
1423 __u32 lower_offset;
1424 __u32 upper_offset;
1425 __u32 frag_idx;
1426 __u32 stepped_offset;
1427 struct sk_buff *root_skb;
1428 struct sk_buff *cur_skb;
1429 __u8 *frag_data;
1430 __u32 frag_off;
1431};
1432
1433void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1434 unsigned int to, struct skb_seq_state *st);
1435unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1436 struct skb_seq_state *st);
1437void skb_abort_seq_read(struct skb_seq_state *st);
1438int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len);
1439
1440unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1441 unsigned int to, struct ts_config *config);
1442
1443/*
1444 * Packet hash types specify the type of hash in skb_set_hash.
1445 *
1446 * Hash types refer to the protocol layer addresses which are used to
1447 * construct a packet's hash. The hashes are used to differentiate or identify
1448 * flows of the protocol layer for the hash type. Hash types are either
1449 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1450 *
1451 * Properties of hashes:
1452 *
1453 * 1) Two packets in different flows have different hash values
1454 * 2) Two packets in the same flow should have the same hash value
1455 *
1456 * A hash at a higher layer is considered to be more specific. A driver should
1457 * set the most specific hash possible.
1458 *
1459 * A driver cannot indicate a more specific hash than the layer at which a hash
1460 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1461 *
1462 * A driver may indicate a hash level which is less specific than the
1463 * actual layer the hash was computed on. For instance, a hash computed
1464 * at L4 may be considered an L3 hash. This should only be done if the
1465 * driver can't unambiguously determine that the HW computed the hash at
1466 * the higher layer. Note that the "should" in the second property above
1467 * permits this.
1468 */
1469enum pkt_hash_types {
1470 PKT_HASH_TYPE_NONE, /* Undefined type */
1471 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1472 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1473 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1474};
1475
1476static inline void skb_clear_hash(struct sk_buff *skb)
1477{
1478 skb->hash = 0;
1479 skb->sw_hash = 0;
1480 skb->l4_hash = 0;
1481}
1482
1483static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1484{
1485 if (!skb->l4_hash)
1486 skb_clear_hash(skb);
1487}
1488
1489static inline void
1490__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1491{
1492 skb->l4_hash = is_l4;
1493 skb->sw_hash = is_sw;
1494 skb->hash = hash;
1495}
1496
1497static inline void
1498skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1499{
1500 /* Used by drivers to set hash from HW */
1501 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1502}
1503
1504static inline void
1505__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1506{
1507 __skb_set_hash(skb, hash, true, is_l4);
1508}
1509
1510u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb);
1511
1512static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb)
1513{
1514 return __skb_get_hash_symmetric_net(NULL, skb);
1515}
1516
1517void __skb_get_hash_net(const struct net *net, struct sk_buff *skb);
1518u32 skb_get_poff(const struct sk_buff *skb);
1519u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1520 const struct flow_keys_basic *keys, int hlen);
1521__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1522 const void *data, int hlen_proto);
1523
1524static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1525 int thoff, u8 ip_proto)
1526{
1527 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1528}
1529
1530void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1531 const struct flow_dissector_key *key,
1532 unsigned int key_count);
1533
1534struct bpf_flow_dissector;
1535u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1536 __be16 proto, int nhoff, int hlen, unsigned int flags);
1537
1538bool __skb_flow_dissect(const struct net *net,
1539 const struct sk_buff *skb,
1540 struct flow_dissector *flow_dissector,
1541 void *target_container, const void *data,
1542 __be16 proto, int nhoff, int hlen, unsigned int flags);
1543
1544static inline bool skb_flow_dissect(const struct sk_buff *skb,
1545 struct flow_dissector *flow_dissector,
1546 void *target_container, unsigned int flags)
1547{
1548 return __skb_flow_dissect(NULL, skb, flow_dissector,
1549 target_container, NULL, 0, 0, 0, flags);
1550}
1551
1552static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1553 struct flow_keys *flow,
1554 unsigned int flags)
1555{
1556 memset(flow, 0, sizeof(*flow));
1557 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1558 flow, NULL, 0, 0, 0, flags);
1559}
1560
1561static inline bool
1562skb_flow_dissect_flow_keys_basic(const struct net *net,
1563 const struct sk_buff *skb,
1564 struct flow_keys_basic *flow,
1565 const void *data, __be16 proto,
1566 int nhoff, int hlen, unsigned int flags)
1567{
1568 memset(flow, 0, sizeof(*flow));
1569 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1570 data, proto, nhoff, hlen, flags);
1571}
1572
1573void skb_flow_dissect_meta(const struct sk_buff *skb,
1574 struct flow_dissector *flow_dissector,
1575 void *target_container);
1576
1577/* Gets a skb connection tracking info, ctinfo map should be a
1578 * map of mapsize to translate enum ip_conntrack_info states
1579 * to user states.
1580 */
1581void
1582skb_flow_dissect_ct(const struct sk_buff *skb,
1583 struct flow_dissector *flow_dissector,
1584 void *target_container,
1585 u16 *ctinfo_map, size_t mapsize,
1586 bool post_ct, u16 zone);
1587void
1588skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1589 struct flow_dissector *flow_dissector,
1590 void *target_container);
1591
1592void skb_flow_dissect_hash(const struct sk_buff *skb,
1593 struct flow_dissector *flow_dissector,
1594 void *target_container);
1595
1596static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb)
1597{
1598 if (!skb->l4_hash && !skb->sw_hash)
1599 __skb_get_hash_net(net, skb);
1600
1601 return skb->hash;
1602}
1603
1604static inline __u32 skb_get_hash(struct sk_buff *skb)
1605{
1606 if (!skb->l4_hash && !skb->sw_hash)
1607 __skb_get_hash_net(NULL, skb);
1608
1609 return skb->hash;
1610}
1611
1612static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1613{
1614 if (!skb->l4_hash && !skb->sw_hash) {
1615 struct flow_keys keys;
1616 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1617
1618 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1619 }
1620
1621 return skb->hash;
1622}
1623
1624__u32 skb_get_hash_perturb(const struct sk_buff *skb,
1625 const siphash_key_t *perturb);
1626
1627static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1628{
1629 return skb->hash;
1630}
1631
1632static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1633{
1634 to->hash = from->hash;
1635 to->sw_hash = from->sw_hash;
1636 to->l4_hash = from->l4_hash;
1637};
1638
1639static inline int skb_cmp_decrypted(const struct sk_buff *skb1,
1640 const struct sk_buff *skb2)
1641{
1642#ifdef CONFIG_SKB_DECRYPTED
1643 return skb2->decrypted - skb1->decrypted;
1644#else
1645 return 0;
1646#endif
1647}
1648
1649static inline bool skb_is_decrypted(const struct sk_buff *skb)
1650{
1651#ifdef CONFIG_SKB_DECRYPTED
1652 return skb->decrypted;
1653#else
1654 return false;
1655#endif
1656}
1657
1658static inline void skb_copy_decrypted(struct sk_buff *to,
1659 const struct sk_buff *from)
1660{
1661#ifdef CONFIG_SKB_DECRYPTED
1662 to->decrypted = from->decrypted;
1663#endif
1664}
1665
1666#ifdef NET_SKBUFF_DATA_USES_OFFSET
1667static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1668{
1669 return skb->head + skb->end;
1670}
1671
1672static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1673{
1674 return skb->end;
1675}
1676
1677static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1678{
1679 skb->end = offset;
1680}
1681#else
1682static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1683{
1684 return skb->end;
1685}
1686
1687static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1688{
1689 return skb->end - skb->head;
1690}
1691
1692static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1693{
1694 skb->end = skb->head + offset;
1695}
1696#endif
1697
1698extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops;
1699
1700struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1701 struct ubuf_info *uarg);
1702
1703void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1704
1705int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1706 struct sk_buff *skb, struct iov_iter *from,
1707 size_t length);
1708
1709int zerocopy_fill_skb_from_iter(struct sk_buff *skb,
1710 struct iov_iter *from, size_t length);
1711
1712static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1713 struct msghdr *msg, int len)
1714{
1715 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1716}
1717
1718int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1719 struct msghdr *msg, int len,
1720 struct ubuf_info *uarg);
1721
1722/* Internal */
1723#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1724
1725static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1726{
1727 return &skb_shinfo(skb)->hwtstamps;
1728}
1729
1730static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1731{
1732 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1733
1734 return is_zcopy ? skb_uarg(skb) : NULL;
1735}
1736
1737static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1738{
1739 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1740}
1741
1742static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1743{
1744 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1745}
1746
1747static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1748 const struct sk_buff *skb2)
1749{
1750 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1751}
1752
1753static inline void net_zcopy_get(struct ubuf_info *uarg)
1754{
1755 refcount_inc(&uarg->refcnt);
1756}
1757
1758static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1759{
1760 skb_shinfo(skb)->destructor_arg = uarg;
1761 skb_shinfo(skb)->flags |= uarg->flags;
1762}
1763
1764static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1765 bool *have_ref)
1766{
1767 if (skb && uarg && !skb_zcopy(skb)) {
1768 if (unlikely(have_ref && *have_ref))
1769 *have_ref = false;
1770 else
1771 net_zcopy_get(uarg);
1772 skb_zcopy_init(skb, uarg);
1773 }
1774}
1775
1776static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1777{
1778 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1779 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1780}
1781
1782static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1783{
1784 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1785}
1786
1787static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1788{
1789 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1790}
1791
1792static inline void net_zcopy_put(struct ubuf_info *uarg)
1793{
1794 if (uarg)
1795 uarg->ops->complete(NULL, uarg, true);
1796}
1797
1798static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1799{
1800 if (uarg) {
1801 if (uarg->ops == &msg_zerocopy_ubuf_ops)
1802 msg_zerocopy_put_abort(uarg, have_uref);
1803 else if (have_uref)
1804 net_zcopy_put(uarg);
1805 }
1806}
1807
1808/* Release a reference on a zerocopy structure */
1809static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1810{
1811 struct ubuf_info *uarg = skb_zcopy(skb);
1812
1813 if (uarg) {
1814 if (!skb_zcopy_is_nouarg(skb))
1815 uarg->ops->complete(skb, uarg, zerocopy_success);
1816
1817 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1818 }
1819}
1820
1821void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1822
1823static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1824{
1825 if (unlikely(skb_zcopy_managed(skb)))
1826 __skb_zcopy_downgrade_managed(skb);
1827}
1828
1829/* Return true if frags in this skb are readable by the host. */
1830static inline bool skb_frags_readable(const struct sk_buff *skb)
1831{
1832 return !skb->unreadable;
1833}
1834
1835static inline void skb_mark_not_on_list(struct sk_buff *skb)
1836{
1837 skb->next = NULL;
1838}
1839
1840static inline void skb_poison_list(struct sk_buff *skb)
1841{
1842#ifdef CONFIG_DEBUG_NET
1843 skb->next = SKB_LIST_POISON_NEXT;
1844#endif
1845}
1846
1847/* Iterate through singly-linked GSO fragments of an skb. */
1848#define skb_list_walk_safe(first, skb, next_skb) \
1849 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1850 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1851
1852static inline void skb_list_del_init(struct sk_buff *skb)
1853{
1854 __list_del_entry(&skb->list);
1855 skb_mark_not_on_list(skb);
1856}
1857
1858/**
1859 * skb_queue_empty - check if a queue is empty
1860 * @list: queue head
1861 *
1862 * Returns true if the queue is empty, false otherwise.
1863 */
1864static inline int skb_queue_empty(const struct sk_buff_head *list)
1865{
1866 return list->next == (const struct sk_buff *) list;
1867}
1868
1869/**
1870 * skb_queue_empty_lockless - check if a queue is empty
1871 * @list: queue head
1872 *
1873 * Returns true if the queue is empty, false otherwise.
1874 * This variant can be used in lockless contexts.
1875 */
1876static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1877{
1878 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1879}
1880
1881
1882/**
1883 * skb_queue_is_last - check if skb is the last entry in the queue
1884 * @list: queue head
1885 * @skb: buffer
1886 *
1887 * Returns true if @skb is the last buffer on the list.
1888 */
1889static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1890 const struct sk_buff *skb)
1891{
1892 return skb->next == (const struct sk_buff *) list;
1893}
1894
1895/**
1896 * skb_queue_is_first - check if skb is the first entry in the queue
1897 * @list: queue head
1898 * @skb: buffer
1899 *
1900 * Returns true if @skb is the first buffer on the list.
1901 */
1902static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1903 const struct sk_buff *skb)
1904{
1905 return skb->prev == (const struct sk_buff *) list;
1906}
1907
1908/**
1909 * skb_queue_next - return the next packet in the queue
1910 * @list: queue head
1911 * @skb: current buffer
1912 *
1913 * Return the next packet in @list after @skb. It is only valid to
1914 * call this if skb_queue_is_last() evaluates to false.
1915 */
1916static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1917 const struct sk_buff *skb)
1918{
1919 /* This BUG_ON may seem severe, but if we just return then we
1920 * are going to dereference garbage.
1921 */
1922 BUG_ON(skb_queue_is_last(list, skb));
1923 return skb->next;
1924}
1925
1926/**
1927 * skb_queue_prev - return the prev packet in the queue
1928 * @list: queue head
1929 * @skb: current buffer
1930 *
1931 * Return the prev packet in @list before @skb. It is only valid to
1932 * call this if skb_queue_is_first() evaluates to false.
1933 */
1934static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1935 const struct sk_buff *skb)
1936{
1937 /* This BUG_ON may seem severe, but if we just return then we
1938 * are going to dereference garbage.
1939 */
1940 BUG_ON(skb_queue_is_first(list, skb));
1941 return skb->prev;
1942}
1943
1944/**
1945 * skb_get - reference buffer
1946 * @skb: buffer to reference
1947 *
1948 * Makes another reference to a socket buffer and returns a pointer
1949 * to the buffer.
1950 */
1951static inline struct sk_buff *skb_get(struct sk_buff *skb)
1952{
1953 refcount_inc(&skb->users);
1954 return skb;
1955}
1956
1957/*
1958 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1959 */
1960
1961/**
1962 * skb_cloned - is the buffer a clone
1963 * @skb: buffer to check
1964 *
1965 * Returns true if the buffer was generated with skb_clone() and is
1966 * one of multiple shared copies of the buffer. Cloned buffers are
1967 * shared data so must not be written to under normal circumstances.
1968 */
1969static inline int skb_cloned(const struct sk_buff *skb)
1970{
1971 return skb->cloned &&
1972 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1973}
1974
1975static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1976{
1977 might_sleep_if(gfpflags_allow_blocking(pri));
1978
1979 if (skb_cloned(skb))
1980 return pskb_expand_head(skb, 0, 0, pri);
1981
1982 return 0;
1983}
1984
1985/* This variant of skb_unclone() makes sure skb->truesize
1986 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1987 *
1988 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1989 * when various debugging features are in place.
1990 */
1991int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1992static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1993{
1994 might_sleep_if(gfpflags_allow_blocking(pri));
1995
1996 if (skb_cloned(skb))
1997 return __skb_unclone_keeptruesize(skb, pri);
1998 return 0;
1999}
2000
2001/**
2002 * skb_header_cloned - is the header a clone
2003 * @skb: buffer to check
2004 *
2005 * Returns true if modifying the header part of the buffer requires
2006 * the data to be copied.
2007 */
2008static inline int skb_header_cloned(const struct sk_buff *skb)
2009{
2010 int dataref;
2011
2012 if (!skb->cloned)
2013 return 0;
2014
2015 dataref = atomic_read(&skb_shinfo(skb)->dataref);
2016 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
2017 return dataref != 1;
2018}
2019
2020static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
2021{
2022 might_sleep_if(gfpflags_allow_blocking(pri));
2023
2024 if (skb_header_cloned(skb))
2025 return pskb_expand_head(skb, 0, 0, pri);
2026
2027 return 0;
2028}
2029
2030/**
2031 * __skb_header_release() - allow clones to use the headroom
2032 * @skb: buffer to operate on
2033 *
2034 * See "DOC: dataref and headerless skbs".
2035 */
2036static inline void __skb_header_release(struct sk_buff *skb)
2037{
2038 skb->nohdr = 1;
2039 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
2040}
2041
2042
2043/**
2044 * skb_shared - is the buffer shared
2045 * @skb: buffer to check
2046 *
2047 * Returns true if more than one person has a reference to this
2048 * buffer.
2049 */
2050static inline int skb_shared(const struct sk_buff *skb)
2051{
2052 return refcount_read(&skb->users) != 1;
2053}
2054
2055/**
2056 * skb_share_check - check if buffer is shared and if so clone it
2057 * @skb: buffer to check
2058 * @pri: priority for memory allocation
2059 *
2060 * If the buffer is shared the buffer is cloned and the old copy
2061 * drops a reference. A new clone with a single reference is returned.
2062 * If the buffer is not shared the original buffer is returned. When
2063 * being called from interrupt status or with spinlocks held pri must
2064 * be GFP_ATOMIC.
2065 *
2066 * NULL is returned on a memory allocation failure.
2067 */
2068static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2069{
2070 might_sleep_if(gfpflags_allow_blocking(pri));
2071 if (skb_shared(skb)) {
2072 struct sk_buff *nskb = skb_clone(skb, pri);
2073
2074 if (likely(nskb))
2075 consume_skb(skb);
2076 else
2077 kfree_skb(skb);
2078 skb = nskb;
2079 }
2080 return skb;
2081}
2082
2083/*
2084 * Copy shared buffers into a new sk_buff. We effectively do COW on
2085 * packets to handle cases where we have a local reader and forward
2086 * and a couple of other messy ones. The normal one is tcpdumping
2087 * a packet that's being forwarded.
2088 */
2089
2090/**
2091 * skb_unshare - make a copy of a shared buffer
2092 * @skb: buffer to check
2093 * @pri: priority for memory allocation
2094 *
2095 * If the socket buffer is a clone then this function creates a new
2096 * copy of the data, drops a reference count on the old copy and returns
2097 * the new copy with the reference count at 1. If the buffer is not a clone
2098 * the original buffer is returned. When called with a spinlock held or
2099 * from interrupt state @pri must be %GFP_ATOMIC
2100 *
2101 * %NULL is returned on a memory allocation failure.
2102 */
2103static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2104 gfp_t pri)
2105{
2106 might_sleep_if(gfpflags_allow_blocking(pri));
2107 if (skb_cloned(skb)) {
2108 struct sk_buff *nskb = skb_copy(skb, pri);
2109
2110 /* Free our shared copy */
2111 if (likely(nskb))
2112 consume_skb(skb);
2113 else
2114 kfree_skb(skb);
2115 skb = nskb;
2116 }
2117 return skb;
2118}
2119
2120/**
2121 * skb_peek - peek at the head of an &sk_buff_head
2122 * @list_: list to peek at
2123 *
2124 * Peek an &sk_buff. Unlike most other operations you _MUST_
2125 * be careful with this one. A peek leaves the buffer on the
2126 * list and someone else may run off with it. You must hold
2127 * the appropriate locks or have a private queue to do this.
2128 *
2129 * Returns %NULL for an empty list or a pointer to the head element.
2130 * The reference count is not incremented and the reference is therefore
2131 * volatile. Use with caution.
2132 */
2133static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2134{
2135 struct sk_buff *skb = list_->next;
2136
2137 if (skb == (struct sk_buff *)list_)
2138 skb = NULL;
2139 return skb;
2140}
2141
2142/**
2143 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2144 * @list_: list to peek at
2145 *
2146 * Like skb_peek(), but the caller knows that the list is not empty.
2147 */
2148static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2149{
2150 return list_->next;
2151}
2152
2153/**
2154 * skb_peek_next - peek skb following the given one from a queue
2155 * @skb: skb to start from
2156 * @list_: list to peek at
2157 *
2158 * Returns %NULL when the end of the list is met or a pointer to the
2159 * next element. The reference count is not incremented and the
2160 * reference is therefore volatile. Use with caution.
2161 */
2162static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2163 const struct sk_buff_head *list_)
2164{
2165 struct sk_buff *next = skb->next;
2166
2167 if (next == (struct sk_buff *)list_)
2168 next = NULL;
2169 return next;
2170}
2171
2172/**
2173 * skb_peek_tail - peek at the tail of an &sk_buff_head
2174 * @list_: list to peek at
2175 *
2176 * Peek an &sk_buff. Unlike most other operations you _MUST_
2177 * be careful with this one. A peek leaves the buffer on the
2178 * list and someone else may run off with it. You must hold
2179 * the appropriate locks or have a private queue to do this.
2180 *
2181 * Returns %NULL for an empty list or a pointer to the tail element.
2182 * The reference count is not incremented and the reference is therefore
2183 * volatile. Use with caution.
2184 */
2185static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2186{
2187 struct sk_buff *skb = READ_ONCE(list_->prev);
2188
2189 if (skb == (struct sk_buff *)list_)
2190 skb = NULL;
2191 return skb;
2192
2193}
2194
2195/**
2196 * skb_queue_len - get queue length
2197 * @list_: list to measure
2198 *
2199 * Return the length of an &sk_buff queue.
2200 */
2201static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2202{
2203 return list_->qlen;
2204}
2205
2206/**
2207 * skb_queue_len_lockless - get queue length
2208 * @list_: list to measure
2209 *
2210 * Return the length of an &sk_buff queue.
2211 * This variant can be used in lockless contexts.
2212 */
2213static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2214{
2215 return READ_ONCE(list_->qlen);
2216}
2217
2218/**
2219 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2220 * @list: queue to initialize
2221 *
2222 * This initializes only the list and queue length aspects of
2223 * an sk_buff_head object. This allows to initialize the list
2224 * aspects of an sk_buff_head without reinitializing things like
2225 * the spinlock. It can also be used for on-stack sk_buff_head
2226 * objects where the spinlock is known to not be used.
2227 */
2228static inline void __skb_queue_head_init(struct sk_buff_head *list)
2229{
2230 list->prev = list->next = (struct sk_buff *)list;
2231 list->qlen = 0;
2232}
2233
2234/*
2235 * This function creates a split out lock class for each invocation;
2236 * this is needed for now since a whole lot of users of the skb-queue
2237 * infrastructure in drivers have different locking usage (in hardirq)
2238 * than the networking core (in softirq only). In the long run either the
2239 * network layer or drivers should need annotation to consolidate the
2240 * main types of usage into 3 classes.
2241 */
2242static inline void skb_queue_head_init(struct sk_buff_head *list)
2243{
2244 spin_lock_init(&list->lock);
2245 __skb_queue_head_init(list);
2246}
2247
2248static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2249 struct lock_class_key *class)
2250{
2251 skb_queue_head_init(list);
2252 lockdep_set_class(&list->lock, class);
2253}
2254
2255/*
2256 * Insert an sk_buff on a list.
2257 *
2258 * The "__skb_xxxx()" functions are the non-atomic ones that
2259 * can only be called with interrupts disabled.
2260 */
2261static inline void __skb_insert(struct sk_buff *newsk,
2262 struct sk_buff *prev, struct sk_buff *next,
2263 struct sk_buff_head *list)
2264{
2265 /* See skb_queue_empty_lockless() and skb_peek_tail()
2266 * for the opposite READ_ONCE()
2267 */
2268 WRITE_ONCE(newsk->next, next);
2269 WRITE_ONCE(newsk->prev, prev);
2270 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2271 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2272 WRITE_ONCE(list->qlen, list->qlen + 1);
2273}
2274
2275static inline void __skb_queue_splice(const struct sk_buff_head *list,
2276 struct sk_buff *prev,
2277 struct sk_buff *next)
2278{
2279 struct sk_buff *first = list->next;
2280 struct sk_buff *last = list->prev;
2281
2282 WRITE_ONCE(first->prev, prev);
2283 WRITE_ONCE(prev->next, first);
2284
2285 WRITE_ONCE(last->next, next);
2286 WRITE_ONCE(next->prev, last);
2287}
2288
2289/**
2290 * skb_queue_splice - join two skb lists, this is designed for stacks
2291 * @list: the new list to add
2292 * @head: the place to add it in the first list
2293 */
2294static inline void skb_queue_splice(const struct sk_buff_head *list,
2295 struct sk_buff_head *head)
2296{
2297 if (!skb_queue_empty(list)) {
2298 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2299 head->qlen += list->qlen;
2300 }
2301}
2302
2303/**
2304 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2305 * @list: the new list to add
2306 * @head: the place to add it in the first list
2307 *
2308 * The list at @list is reinitialised
2309 */
2310static inline void skb_queue_splice_init(struct sk_buff_head *list,
2311 struct sk_buff_head *head)
2312{
2313 if (!skb_queue_empty(list)) {
2314 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2315 head->qlen += list->qlen;
2316 __skb_queue_head_init(list);
2317 }
2318}
2319
2320/**
2321 * skb_queue_splice_tail - join two skb lists, each list being a queue
2322 * @list: the new list to add
2323 * @head: the place to add it in the first list
2324 */
2325static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2326 struct sk_buff_head *head)
2327{
2328 if (!skb_queue_empty(list)) {
2329 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2330 head->qlen += list->qlen;
2331 }
2332}
2333
2334/**
2335 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2336 * @list: the new list to add
2337 * @head: the place to add it in the first list
2338 *
2339 * Each of the lists is a queue.
2340 * The list at @list is reinitialised
2341 */
2342static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2343 struct sk_buff_head *head)
2344{
2345 if (!skb_queue_empty(list)) {
2346 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2347 head->qlen += list->qlen;
2348 __skb_queue_head_init(list);
2349 }
2350}
2351
2352/**
2353 * __skb_queue_after - queue a buffer at the list head
2354 * @list: list to use
2355 * @prev: place after this buffer
2356 * @newsk: buffer to queue
2357 *
2358 * Queue a buffer int the middle of a list. This function takes no locks
2359 * and you must therefore hold required locks before calling it.
2360 *
2361 * A buffer cannot be placed on two lists at the same time.
2362 */
2363static inline void __skb_queue_after(struct sk_buff_head *list,
2364 struct sk_buff *prev,
2365 struct sk_buff *newsk)
2366{
2367 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2368}
2369
2370void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2371 struct sk_buff_head *list);
2372
2373static inline void __skb_queue_before(struct sk_buff_head *list,
2374 struct sk_buff *next,
2375 struct sk_buff *newsk)
2376{
2377 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2378}
2379
2380/**
2381 * __skb_queue_head - queue a buffer at the list head
2382 * @list: list to use
2383 * @newsk: buffer to queue
2384 *
2385 * Queue a buffer at the start of a list. This function takes no locks
2386 * and you must therefore hold required locks before calling it.
2387 *
2388 * A buffer cannot be placed on two lists at the same time.
2389 */
2390static inline void __skb_queue_head(struct sk_buff_head *list,
2391 struct sk_buff *newsk)
2392{
2393 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2394}
2395void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2396
2397/**
2398 * __skb_queue_tail - queue a buffer at the list tail
2399 * @list: list to use
2400 * @newsk: buffer to queue
2401 *
2402 * Queue a buffer at the end of a list. This function takes no locks
2403 * and you must therefore hold required locks before calling it.
2404 *
2405 * A buffer cannot be placed on two lists at the same time.
2406 */
2407static inline void __skb_queue_tail(struct sk_buff_head *list,
2408 struct sk_buff *newsk)
2409{
2410 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2411}
2412void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2413
2414/*
2415 * remove sk_buff from list. _Must_ be called atomically, and with
2416 * the list known..
2417 */
2418void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2419static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2420{
2421 struct sk_buff *next, *prev;
2422
2423 WRITE_ONCE(list->qlen, list->qlen - 1);
2424 next = skb->next;
2425 prev = skb->prev;
2426 skb->next = skb->prev = NULL;
2427 WRITE_ONCE(next->prev, prev);
2428 WRITE_ONCE(prev->next, next);
2429}
2430
2431/**
2432 * __skb_dequeue - remove from the head of the queue
2433 * @list: list to dequeue from
2434 *
2435 * Remove the head of the list. This function does not take any locks
2436 * so must be used with appropriate locks held only. The head item is
2437 * returned or %NULL if the list is empty.
2438 */
2439static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2440{
2441 struct sk_buff *skb = skb_peek(list);
2442 if (skb)
2443 __skb_unlink(skb, list);
2444 return skb;
2445}
2446struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2447
2448/**
2449 * __skb_dequeue_tail - remove from the tail of the queue
2450 * @list: list to dequeue from
2451 *
2452 * Remove the tail of the list. This function does not take any locks
2453 * so must be used with appropriate locks held only. The tail item is
2454 * returned or %NULL if the list is empty.
2455 */
2456static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2457{
2458 struct sk_buff *skb = skb_peek_tail(list);
2459 if (skb)
2460 __skb_unlink(skb, list);
2461 return skb;
2462}
2463struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2464
2465
2466static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2467{
2468 return skb->data_len;
2469}
2470
2471static inline unsigned int skb_headlen(const struct sk_buff *skb)
2472{
2473 return skb->len - skb->data_len;
2474}
2475
2476static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2477{
2478 unsigned int i, len = 0;
2479
2480 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2481 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2482 return len;
2483}
2484
2485static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2486{
2487 return skb_headlen(skb) + __skb_pagelen(skb);
2488}
2489
2490static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag,
2491 netmem_ref netmem, int off,
2492 int size)
2493{
2494 frag->netmem = netmem;
2495 frag->offset = off;
2496 skb_frag_size_set(frag, size);
2497}
2498
2499static inline void skb_frag_fill_page_desc(skb_frag_t *frag,
2500 struct page *page,
2501 int off, int size)
2502{
2503 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size);
2504}
2505
2506static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo,
2507 int i, netmem_ref netmem,
2508 int off, int size)
2509{
2510 skb_frag_t *frag = &shinfo->frags[i];
2511
2512 skb_frag_fill_netmem_desc(frag, netmem, off, size);
2513}
2514
2515static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2516 int i, struct page *page,
2517 int off, int size)
2518{
2519 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off,
2520 size);
2521}
2522
2523/**
2524 * skb_len_add - adds a number to len fields of skb
2525 * @skb: buffer to add len to
2526 * @delta: number of bytes to add
2527 */
2528static inline void skb_len_add(struct sk_buff *skb, int delta)
2529{
2530 skb->len += delta;
2531 skb->data_len += delta;
2532 skb->truesize += delta;
2533}
2534
2535/**
2536 * __skb_fill_netmem_desc - initialise a fragment in an skb
2537 * @skb: buffer containing fragment to be initialised
2538 * @i: fragment index to initialise
2539 * @netmem: the netmem to use for this fragment
2540 * @off: the offset to the data with @page
2541 * @size: the length of the data
2542 *
2543 * Initialises the @i'th fragment of @skb to point to &size bytes at
2544 * offset @off within @page.
2545 *
2546 * Does not take any additional reference on the fragment.
2547 */
2548static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i,
2549 netmem_ref netmem, int off, int size)
2550{
2551 struct page *page;
2552
2553 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size);
2554
2555 if (netmem_is_net_iov(netmem)) {
2556 skb->unreadable = true;
2557 return;
2558 }
2559
2560 page = netmem_to_page(netmem);
2561
2562 /* Propagate page pfmemalloc to the skb if we can. The problem is
2563 * that not all callers have unique ownership of the page but rely
2564 * on page_is_pfmemalloc doing the right thing(tm).
2565 */
2566 page = compound_head(page);
2567 if (page_is_pfmemalloc(page))
2568 skb->pfmemalloc = true;
2569}
2570
2571static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2572 struct page *page, int off, int size)
2573{
2574 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2575}
2576
2577static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i,
2578 netmem_ref netmem, int off, int size)
2579{
2580 __skb_fill_netmem_desc(skb, i, netmem, off, size);
2581 skb_shinfo(skb)->nr_frags = i + 1;
2582}
2583
2584/**
2585 * skb_fill_page_desc - initialise a paged fragment in an skb
2586 * @skb: buffer containing fragment to be initialised
2587 * @i: paged fragment index to initialise
2588 * @page: the page to use for this fragment
2589 * @off: the offset to the data with @page
2590 * @size: the length of the data
2591 *
2592 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2593 * @skb to point to @size bytes at offset @off within @page. In
2594 * addition updates @skb such that @i is the last fragment.
2595 *
2596 * Does not take any additional reference on the fragment.
2597 */
2598static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2599 struct page *page, int off, int size)
2600{
2601 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2602}
2603
2604/**
2605 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2606 * @skb: buffer containing fragment to be initialised
2607 * @i: paged fragment index to initialise
2608 * @page: the page to use for this fragment
2609 * @off: the offset to the data with @page
2610 * @size: the length of the data
2611 *
2612 * Variant of skb_fill_page_desc() which does not deal with
2613 * pfmemalloc, if page is not owned by us.
2614 */
2615static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2616 struct page *page, int off,
2617 int size)
2618{
2619 struct skb_shared_info *shinfo = skb_shinfo(skb);
2620
2621 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2622 shinfo->nr_frags = i + 1;
2623}
2624
2625void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
2626 int off, int size, unsigned int truesize);
2627
2628static inline void skb_add_rx_frag(struct sk_buff *skb, int i,
2629 struct page *page, int off, int size,
2630 unsigned int truesize)
2631{
2632 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size,
2633 truesize);
2634}
2635
2636void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2637 unsigned int truesize);
2638
2639#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2640
2641#ifdef NET_SKBUFF_DATA_USES_OFFSET
2642static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2643{
2644 return skb->head + skb->tail;
2645}
2646
2647static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2648{
2649 skb->tail = skb->data - skb->head;
2650}
2651
2652static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2653{
2654 skb_reset_tail_pointer(skb);
2655 skb->tail += offset;
2656}
2657
2658#else /* NET_SKBUFF_DATA_USES_OFFSET */
2659static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2660{
2661 return skb->tail;
2662}
2663
2664static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2665{
2666 skb->tail = skb->data;
2667}
2668
2669static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2670{
2671 skb->tail = skb->data + offset;
2672}
2673
2674#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2675
2676static inline void skb_assert_len(struct sk_buff *skb)
2677{
2678#ifdef CONFIG_DEBUG_NET
2679 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2680 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2681#endif /* CONFIG_DEBUG_NET */
2682}
2683
2684/*
2685 * Add data to an sk_buff
2686 */
2687void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2688void *skb_put(struct sk_buff *skb, unsigned int len);
2689static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2690{
2691 void *tmp = skb_tail_pointer(skb);
2692 SKB_LINEAR_ASSERT(skb);
2693 skb->tail += len;
2694 skb->len += len;
2695 return tmp;
2696}
2697
2698static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2699{
2700 void *tmp = __skb_put(skb, len);
2701
2702 memset(tmp, 0, len);
2703 return tmp;
2704}
2705
2706static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2707 unsigned int len)
2708{
2709 void *tmp = __skb_put(skb, len);
2710
2711 memcpy(tmp, data, len);
2712 return tmp;
2713}
2714
2715static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2716{
2717 *(u8 *)__skb_put(skb, 1) = val;
2718}
2719
2720static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2721{
2722 void *tmp = skb_put(skb, len);
2723
2724 memset(tmp, 0, len);
2725
2726 return tmp;
2727}
2728
2729static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2730 unsigned int len)
2731{
2732 void *tmp = skb_put(skb, len);
2733
2734 memcpy(tmp, data, len);
2735
2736 return tmp;
2737}
2738
2739static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2740{
2741 *(u8 *)skb_put(skb, 1) = val;
2742}
2743
2744void *skb_push(struct sk_buff *skb, unsigned int len);
2745static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2746{
2747 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2748
2749 skb->data -= len;
2750 skb->len += len;
2751 return skb->data;
2752}
2753
2754void *skb_pull(struct sk_buff *skb, unsigned int len);
2755static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2756{
2757 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2758
2759 skb->len -= len;
2760 if (unlikely(skb->len < skb->data_len)) {
2761#if defined(CONFIG_DEBUG_NET)
2762 skb->len += len;
2763 pr_err("__skb_pull(len=%u)\n", len);
2764 skb_dump(KERN_ERR, skb, false);
2765#endif
2766 BUG();
2767 }
2768 return skb->data += len;
2769}
2770
2771static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2772{
2773 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2774}
2775
2776void *skb_pull_data(struct sk_buff *skb, size_t len);
2777
2778void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2779
2780static inline enum skb_drop_reason
2781pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2782{
2783 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2784
2785 if (likely(len <= skb_headlen(skb)))
2786 return SKB_NOT_DROPPED_YET;
2787
2788 if (unlikely(len > skb->len))
2789 return SKB_DROP_REASON_PKT_TOO_SMALL;
2790
2791 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2792 return SKB_DROP_REASON_NOMEM;
2793
2794 return SKB_NOT_DROPPED_YET;
2795}
2796
2797static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2798{
2799 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2800}
2801
2802static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2803{
2804 if (!pskb_may_pull(skb, len))
2805 return NULL;
2806
2807 skb->len -= len;
2808 return skb->data += len;
2809}
2810
2811void skb_condense(struct sk_buff *skb);
2812
2813/**
2814 * skb_headroom - bytes at buffer head
2815 * @skb: buffer to check
2816 *
2817 * Return the number of bytes of free space at the head of an &sk_buff.
2818 */
2819static inline unsigned int skb_headroom(const struct sk_buff *skb)
2820{
2821 return skb->data - skb->head;
2822}
2823
2824/**
2825 * skb_tailroom - bytes at buffer end
2826 * @skb: buffer to check
2827 *
2828 * Return the number of bytes of free space at the tail of an sk_buff
2829 */
2830static inline int skb_tailroom(const struct sk_buff *skb)
2831{
2832 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2833}
2834
2835/**
2836 * skb_availroom - bytes at buffer end
2837 * @skb: buffer to check
2838 *
2839 * Return the number of bytes of free space at the tail of an sk_buff
2840 * allocated by sk_stream_alloc()
2841 */
2842static inline int skb_availroom(const struct sk_buff *skb)
2843{
2844 if (skb_is_nonlinear(skb))
2845 return 0;
2846
2847 return skb->end - skb->tail - skb->reserved_tailroom;
2848}
2849
2850/**
2851 * skb_reserve - adjust headroom
2852 * @skb: buffer to alter
2853 * @len: bytes to move
2854 *
2855 * Increase the headroom of an empty &sk_buff by reducing the tail
2856 * room. This is only allowed for an empty buffer.
2857 */
2858static inline void skb_reserve(struct sk_buff *skb, int len)
2859{
2860 skb->data += len;
2861 skb->tail += len;
2862}
2863
2864/**
2865 * skb_tailroom_reserve - adjust reserved_tailroom
2866 * @skb: buffer to alter
2867 * @mtu: maximum amount of headlen permitted
2868 * @needed_tailroom: minimum amount of reserved_tailroom
2869 *
2870 * Set reserved_tailroom so that headlen can be as large as possible but
2871 * not larger than mtu and tailroom cannot be smaller than
2872 * needed_tailroom.
2873 * The required headroom should already have been reserved before using
2874 * this function.
2875 */
2876static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2877 unsigned int needed_tailroom)
2878{
2879 SKB_LINEAR_ASSERT(skb);
2880 if (mtu < skb_tailroom(skb) - needed_tailroom)
2881 /* use at most mtu */
2882 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2883 else
2884 /* use up to all available space */
2885 skb->reserved_tailroom = needed_tailroom;
2886}
2887
2888#define ENCAP_TYPE_ETHER 0
2889#define ENCAP_TYPE_IPPROTO 1
2890
2891static inline void skb_set_inner_protocol(struct sk_buff *skb,
2892 __be16 protocol)
2893{
2894 skb->inner_protocol = protocol;
2895 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2896}
2897
2898static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2899 __u8 ipproto)
2900{
2901 skb->inner_ipproto = ipproto;
2902 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2903}
2904
2905static inline void skb_reset_inner_headers(struct sk_buff *skb)
2906{
2907 skb->inner_mac_header = skb->mac_header;
2908 skb->inner_network_header = skb->network_header;
2909 skb->inner_transport_header = skb->transport_header;
2910}
2911
2912static inline void skb_reset_mac_len(struct sk_buff *skb)
2913{
2914 skb->mac_len = skb->network_header - skb->mac_header;
2915}
2916
2917static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2918 *skb)
2919{
2920 return skb->head + skb->inner_transport_header;
2921}
2922
2923static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2924{
2925 return skb_inner_transport_header(skb) - skb->data;
2926}
2927
2928static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2929{
2930 skb->inner_transport_header = skb->data - skb->head;
2931}
2932
2933static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2934 const int offset)
2935{
2936 skb_reset_inner_transport_header(skb);
2937 skb->inner_transport_header += offset;
2938}
2939
2940static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2941{
2942 return skb->head + skb->inner_network_header;
2943}
2944
2945static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2946{
2947 skb->inner_network_header = skb->data - skb->head;
2948}
2949
2950static inline void skb_set_inner_network_header(struct sk_buff *skb,
2951 const int offset)
2952{
2953 skb_reset_inner_network_header(skb);
2954 skb->inner_network_header += offset;
2955}
2956
2957static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb)
2958{
2959 return skb->inner_network_header > 0;
2960}
2961
2962static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2963{
2964 return skb->head + skb->inner_mac_header;
2965}
2966
2967static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2968{
2969 skb->inner_mac_header = skb->data - skb->head;
2970}
2971
2972static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2973 const int offset)
2974{
2975 skb_reset_inner_mac_header(skb);
2976 skb->inner_mac_header += offset;
2977}
2978static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2979{
2980 return skb->transport_header != (typeof(skb->transport_header))~0U;
2981}
2982
2983static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2984{
2985 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2986 return skb->head + skb->transport_header;
2987}
2988
2989static inline void skb_reset_transport_header(struct sk_buff *skb)
2990{
2991 skb->transport_header = skb->data - skb->head;
2992}
2993
2994static inline void skb_set_transport_header(struct sk_buff *skb,
2995 const int offset)
2996{
2997 skb_reset_transport_header(skb);
2998 skb->transport_header += offset;
2999}
3000
3001static inline unsigned char *skb_network_header(const struct sk_buff *skb)
3002{
3003 return skb->head + skb->network_header;
3004}
3005
3006static inline void skb_reset_network_header(struct sk_buff *skb)
3007{
3008 skb->network_header = skb->data - skb->head;
3009}
3010
3011static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
3012{
3013 skb_reset_network_header(skb);
3014 skb->network_header += offset;
3015}
3016
3017static inline int skb_mac_header_was_set(const struct sk_buff *skb)
3018{
3019 return skb->mac_header != (typeof(skb->mac_header))~0U;
3020}
3021
3022static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
3023{
3024 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
3025 return skb->head + skb->mac_header;
3026}
3027
3028static inline int skb_mac_offset(const struct sk_buff *skb)
3029{
3030 return skb_mac_header(skb) - skb->data;
3031}
3032
3033static inline u32 skb_mac_header_len(const struct sk_buff *skb)
3034{
3035 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
3036 return skb->network_header - skb->mac_header;
3037}
3038
3039static inline void skb_unset_mac_header(struct sk_buff *skb)
3040{
3041 skb->mac_header = (typeof(skb->mac_header))~0U;
3042}
3043
3044static inline void skb_reset_mac_header(struct sk_buff *skb)
3045{
3046 skb->mac_header = skb->data - skb->head;
3047}
3048
3049static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
3050{
3051 skb_reset_mac_header(skb);
3052 skb->mac_header += offset;
3053}
3054
3055static inline void skb_pop_mac_header(struct sk_buff *skb)
3056{
3057 skb->mac_header = skb->network_header;
3058}
3059
3060static inline void skb_probe_transport_header(struct sk_buff *skb)
3061{
3062 struct flow_keys_basic keys;
3063
3064 if (skb_transport_header_was_set(skb))
3065 return;
3066
3067 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
3068 NULL, 0, 0, 0, 0))
3069 skb_set_transport_header(skb, keys.control.thoff);
3070}
3071
3072static inline void skb_mac_header_rebuild(struct sk_buff *skb)
3073{
3074 if (skb_mac_header_was_set(skb)) {
3075 const unsigned char *old_mac = skb_mac_header(skb);
3076
3077 skb_set_mac_header(skb, -skb->mac_len);
3078 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
3079 }
3080}
3081
3082/* Move the full mac header up to current network_header.
3083 * Leaves skb->data pointing at offset skb->mac_len into the mac_header.
3084 * Must be provided the complete mac header length.
3085 */
3086static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len)
3087{
3088 if (skb_mac_header_was_set(skb)) {
3089 const unsigned char *old_mac = skb_mac_header(skb);
3090
3091 skb_set_mac_header(skb, -full_mac_len);
3092 memmove(skb_mac_header(skb), old_mac, full_mac_len);
3093 __skb_push(skb, full_mac_len - skb->mac_len);
3094 }
3095}
3096
3097static inline int skb_checksum_start_offset(const struct sk_buff *skb)
3098{
3099 return skb->csum_start - skb_headroom(skb);
3100}
3101
3102static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
3103{
3104 return skb->head + skb->csum_start;
3105}
3106
3107static inline int skb_transport_offset(const struct sk_buff *skb)
3108{
3109 return skb_transport_header(skb) - skb->data;
3110}
3111
3112static inline u32 skb_network_header_len(const struct sk_buff *skb)
3113{
3114 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
3115 return skb->transport_header - skb->network_header;
3116}
3117
3118static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
3119{
3120 return skb->inner_transport_header - skb->inner_network_header;
3121}
3122
3123static inline int skb_network_offset(const struct sk_buff *skb)
3124{
3125 return skb_network_header(skb) - skb->data;
3126}
3127
3128static inline int skb_inner_network_offset(const struct sk_buff *skb)
3129{
3130 return skb_inner_network_header(skb) - skb->data;
3131}
3132
3133static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
3134{
3135 return pskb_may_pull(skb, skb_network_offset(skb) + len);
3136}
3137
3138/*
3139 * CPUs often take a performance hit when accessing unaligned memory
3140 * locations. The actual performance hit varies, it can be small if the
3141 * hardware handles it or large if we have to take an exception and fix it
3142 * in software.
3143 *
3144 * Since an ethernet header is 14 bytes network drivers often end up with
3145 * the IP header at an unaligned offset. The IP header can be aligned by
3146 * shifting the start of the packet by 2 bytes. Drivers should do this
3147 * with:
3148 *
3149 * skb_reserve(skb, NET_IP_ALIGN);
3150 *
3151 * The downside to this alignment of the IP header is that the DMA is now
3152 * unaligned. On some architectures the cost of an unaligned DMA is high
3153 * and this cost outweighs the gains made by aligning the IP header.
3154 *
3155 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3156 * to be overridden.
3157 */
3158#ifndef NET_IP_ALIGN
3159#define NET_IP_ALIGN 2
3160#endif
3161
3162/*
3163 * The networking layer reserves some headroom in skb data (via
3164 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3165 * the header has to grow. In the default case, if the header has to grow
3166 * 32 bytes or less we avoid the reallocation.
3167 *
3168 * Unfortunately this headroom changes the DMA alignment of the resulting
3169 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3170 * on some architectures. An architecture can override this value,
3171 * perhaps setting it to a cacheline in size (since that will maintain
3172 * cacheline alignment of the DMA). It must be a power of 2.
3173 *
3174 * Various parts of the networking layer expect at least 32 bytes of
3175 * headroom, you should not reduce this.
3176 *
3177 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3178 * to reduce average number of cache lines per packet.
3179 * get_rps_cpu() for example only access one 64 bytes aligned block :
3180 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3181 */
3182#ifndef NET_SKB_PAD
3183#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3184#endif
3185
3186int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3187
3188static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3189{
3190 if (WARN_ON(skb_is_nonlinear(skb)))
3191 return;
3192 skb->len = len;
3193 skb_set_tail_pointer(skb, len);
3194}
3195
3196static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3197{
3198 __skb_set_length(skb, len);
3199}
3200
3201void skb_trim(struct sk_buff *skb, unsigned int len);
3202
3203static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3204{
3205 if (skb->data_len)
3206 return ___pskb_trim(skb, len);
3207 __skb_trim(skb, len);
3208 return 0;
3209}
3210
3211static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3212{
3213 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3214}
3215
3216/**
3217 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3218 * @skb: buffer to alter
3219 * @len: new length
3220 *
3221 * This is identical to pskb_trim except that the caller knows that
3222 * the skb is not cloned so we should never get an error due to out-
3223 * of-memory.
3224 */
3225static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3226{
3227 int err = pskb_trim(skb, len);
3228 BUG_ON(err);
3229}
3230
3231static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3232{
3233 unsigned int diff = len - skb->len;
3234
3235 if (skb_tailroom(skb) < diff) {
3236 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3237 GFP_ATOMIC);
3238 if (ret)
3239 return ret;
3240 }
3241 __skb_set_length(skb, len);
3242 return 0;
3243}
3244
3245/**
3246 * skb_orphan - orphan a buffer
3247 * @skb: buffer to orphan
3248 *
3249 * If a buffer currently has an owner then we call the owner's
3250 * destructor function and make the @skb unowned. The buffer continues
3251 * to exist but is no longer charged to its former owner.
3252 */
3253static inline void skb_orphan(struct sk_buff *skb)
3254{
3255 if (skb->destructor) {
3256 skb->destructor(skb);
3257 skb->destructor = NULL;
3258 skb->sk = NULL;
3259 } else {
3260 BUG_ON(skb->sk);
3261 }
3262}
3263
3264/**
3265 * skb_orphan_frags - orphan the frags contained in a buffer
3266 * @skb: buffer to orphan frags from
3267 * @gfp_mask: allocation mask for replacement pages
3268 *
3269 * For each frag in the SKB which needs a destructor (i.e. has an
3270 * owner) create a copy of that frag and release the original
3271 * page by calling the destructor.
3272 */
3273static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3274{
3275 if (likely(!skb_zcopy(skb)))
3276 return 0;
3277 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3278 return 0;
3279 return skb_copy_ubufs(skb, gfp_mask);
3280}
3281
3282/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3283static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3284{
3285 if (likely(!skb_zcopy(skb)))
3286 return 0;
3287 return skb_copy_ubufs(skb, gfp_mask);
3288}
3289
3290/**
3291 * __skb_queue_purge_reason - empty a list
3292 * @list: list to empty
3293 * @reason: drop reason
3294 *
3295 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3296 * the list and one reference dropped. This function does not take the
3297 * list lock and the caller must hold the relevant locks to use it.
3298 */
3299static inline void __skb_queue_purge_reason(struct sk_buff_head *list,
3300 enum skb_drop_reason reason)
3301{
3302 struct sk_buff *skb;
3303
3304 while ((skb = __skb_dequeue(list)) != NULL)
3305 kfree_skb_reason(skb, reason);
3306}
3307
3308static inline void __skb_queue_purge(struct sk_buff_head *list)
3309{
3310 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3311}
3312
3313void skb_queue_purge_reason(struct sk_buff_head *list,
3314 enum skb_drop_reason reason);
3315
3316static inline void skb_queue_purge(struct sk_buff_head *list)
3317{
3318 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3319}
3320
3321unsigned int skb_rbtree_purge(struct rb_root *root);
3322void skb_errqueue_purge(struct sk_buff_head *list);
3323
3324void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3325
3326/**
3327 * netdev_alloc_frag - allocate a page fragment
3328 * @fragsz: fragment size
3329 *
3330 * Allocates a frag from a page for receive buffer.
3331 * Uses GFP_ATOMIC allocations.
3332 */
3333static inline void *netdev_alloc_frag(unsigned int fragsz)
3334{
3335 return __netdev_alloc_frag_align(fragsz, ~0u);
3336}
3337
3338static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3339 unsigned int align)
3340{
3341 WARN_ON_ONCE(!is_power_of_2(align));
3342 return __netdev_alloc_frag_align(fragsz, -align);
3343}
3344
3345struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3346 gfp_t gfp_mask);
3347
3348/**
3349 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3350 * @dev: network device to receive on
3351 * @length: length to allocate
3352 *
3353 * Allocate a new &sk_buff and assign it a usage count of one. The
3354 * buffer has unspecified headroom built in. Users should allocate
3355 * the headroom they think they need without accounting for the
3356 * built in space. The built in space is used for optimisations.
3357 *
3358 * %NULL is returned if there is no free memory. Although this function
3359 * allocates memory it can be called from an interrupt.
3360 */
3361static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3362 unsigned int length)
3363{
3364 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3365}
3366
3367/* legacy helper around __netdev_alloc_skb() */
3368static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3369 gfp_t gfp_mask)
3370{
3371 return __netdev_alloc_skb(NULL, length, gfp_mask);
3372}
3373
3374/* legacy helper around netdev_alloc_skb() */
3375static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3376{
3377 return netdev_alloc_skb(NULL, length);
3378}
3379
3380
3381static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3382 unsigned int length, gfp_t gfp)
3383{
3384 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3385
3386 if (NET_IP_ALIGN && skb)
3387 skb_reserve(skb, NET_IP_ALIGN);
3388 return skb;
3389}
3390
3391static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3392 unsigned int length)
3393{
3394 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3395}
3396
3397static inline void skb_free_frag(void *addr)
3398{
3399 page_frag_free(addr);
3400}
3401
3402void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3403
3404static inline void *napi_alloc_frag(unsigned int fragsz)
3405{
3406 return __napi_alloc_frag_align(fragsz, ~0u);
3407}
3408
3409static inline void *napi_alloc_frag_align(unsigned int fragsz,
3410 unsigned int align)
3411{
3412 WARN_ON_ONCE(!is_power_of_2(align));
3413 return __napi_alloc_frag_align(fragsz, -align);
3414}
3415
3416struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length);
3417void napi_consume_skb(struct sk_buff *skb, int budget);
3418
3419void napi_skb_free_stolen_head(struct sk_buff *skb);
3420void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason);
3421
3422/**
3423 * __dev_alloc_pages - allocate page for network Rx
3424 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3425 * @order: size of the allocation
3426 *
3427 * Allocate a new page.
3428 *
3429 * %NULL is returned if there is no free memory.
3430*/
3431static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask,
3432 unsigned int order)
3433{
3434 /* This piece of code contains several assumptions.
3435 * 1. This is for device Rx, therefore a cold page is preferred.
3436 * 2. The expectation is the user wants a compound page.
3437 * 3. If requesting a order 0 page it will not be compound
3438 * due to the check to see if order has a value in prep_new_page
3439 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3440 * code in gfp_to_alloc_flags that should be enforcing this.
3441 */
3442 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3443
3444 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order);
3445}
3446#define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__))
3447
3448/*
3449 * This specialized allocator has to be a macro for its allocations to be
3450 * accounted separately (to have a separate alloc_tag).
3451 */
3452#define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order)
3453
3454/**
3455 * __dev_alloc_page - allocate a page for network Rx
3456 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3457 *
3458 * Allocate a new page.
3459 *
3460 * %NULL is returned if there is no free memory.
3461 */
3462static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask)
3463{
3464 return __dev_alloc_pages_noprof(gfp_mask, 0);
3465}
3466#define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__))
3467
3468/*
3469 * This specialized allocator has to be a macro for its allocations to be
3470 * accounted separately (to have a separate alloc_tag).
3471 */
3472#define dev_alloc_page() dev_alloc_pages(0)
3473
3474/**
3475 * dev_page_is_reusable - check whether a page can be reused for network Rx
3476 * @page: the page to test
3477 *
3478 * A page shouldn't be considered for reusing/recycling if it was allocated
3479 * under memory pressure or at a distant memory node.
3480 *
3481 * Returns false if this page should be returned to page allocator, true
3482 * otherwise.
3483 */
3484static inline bool dev_page_is_reusable(const struct page *page)
3485{
3486 return likely(page_to_nid(page) == numa_mem_id() &&
3487 !page_is_pfmemalloc(page));
3488}
3489
3490/**
3491 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3492 * @page: The page that was allocated from skb_alloc_page
3493 * @skb: The skb that may need pfmemalloc set
3494 */
3495static inline void skb_propagate_pfmemalloc(const struct page *page,
3496 struct sk_buff *skb)
3497{
3498 if (page_is_pfmemalloc(page))
3499 skb->pfmemalloc = true;
3500}
3501
3502/**
3503 * skb_frag_off() - Returns the offset of a skb fragment
3504 * @frag: the paged fragment
3505 */
3506static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3507{
3508 return frag->offset;
3509}
3510
3511/**
3512 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3513 * @frag: skb fragment
3514 * @delta: value to add
3515 */
3516static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3517{
3518 frag->offset += delta;
3519}
3520
3521/**
3522 * skb_frag_off_set() - Sets the offset of a skb fragment
3523 * @frag: skb fragment
3524 * @offset: offset of fragment
3525 */
3526static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3527{
3528 frag->offset = offset;
3529}
3530
3531/**
3532 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3533 * @fragto: skb fragment where offset is set
3534 * @fragfrom: skb fragment offset is copied from
3535 */
3536static inline void skb_frag_off_copy(skb_frag_t *fragto,
3537 const skb_frag_t *fragfrom)
3538{
3539 fragto->offset = fragfrom->offset;
3540}
3541
3542/* Return: true if the skb_frag contains a net_iov. */
3543static inline bool skb_frag_is_net_iov(const skb_frag_t *frag)
3544{
3545 return netmem_is_net_iov(frag->netmem);
3546}
3547
3548/**
3549 * skb_frag_net_iov - retrieve the net_iov referred to by fragment
3550 * @frag: the fragment
3551 *
3552 * Return: the &struct net_iov associated with @frag. Returns NULL if this
3553 * frag has no associated net_iov.
3554 */
3555static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag)
3556{
3557 if (!skb_frag_is_net_iov(frag))
3558 return NULL;
3559
3560 return netmem_to_net_iov(frag->netmem);
3561}
3562
3563/**
3564 * skb_frag_page - retrieve the page referred to by a paged fragment
3565 * @frag: the paged fragment
3566 *
3567 * Return: the &struct page associated with @frag. Returns NULL if this frag
3568 * has no associated page.
3569 */
3570static inline struct page *skb_frag_page(const skb_frag_t *frag)
3571{
3572 if (skb_frag_is_net_iov(frag))
3573 return NULL;
3574
3575 return netmem_to_page(frag->netmem);
3576}
3577
3578/**
3579 * skb_frag_netmem - retrieve the netmem referred to by a fragment
3580 * @frag: the fragment
3581 *
3582 * Return: the &netmem_ref associated with @frag.
3583 */
3584static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag)
3585{
3586 return frag->netmem;
3587}
3588
3589int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
3590 unsigned int headroom);
3591int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
3592 struct bpf_prog *prog);
3593
3594/**
3595 * skb_frag_address - gets the address of the data contained in a paged fragment
3596 * @frag: the paged fragment buffer
3597 *
3598 * Returns the address of the data within @frag. The page must already
3599 * be mapped.
3600 */
3601static inline void *skb_frag_address(const skb_frag_t *frag)
3602{
3603 if (!skb_frag_page(frag))
3604 return NULL;
3605
3606 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3607}
3608
3609/**
3610 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3611 * @frag: the paged fragment buffer
3612 *
3613 * Returns the address of the data within @frag. Checks that the page
3614 * is mapped and returns %NULL otherwise.
3615 */
3616static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3617{
3618 void *ptr = page_address(skb_frag_page(frag));
3619 if (unlikely(!ptr))
3620 return NULL;
3621
3622 return ptr + skb_frag_off(frag);
3623}
3624
3625/**
3626 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3627 * @fragto: skb fragment where page is set
3628 * @fragfrom: skb fragment page is copied from
3629 */
3630static inline void skb_frag_page_copy(skb_frag_t *fragto,
3631 const skb_frag_t *fragfrom)
3632{
3633 fragto->netmem = fragfrom->netmem;
3634}
3635
3636bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3637
3638/**
3639 * skb_frag_dma_map - maps a paged fragment via the DMA API
3640 * @dev: the device to map the fragment to
3641 * @frag: the paged fragment to map
3642 * @offset: the offset within the fragment (starting at the
3643 * fragment's own offset)
3644 * @size: the number of bytes to map
3645 * @dir: the direction of the mapping (``PCI_DMA_*``)
3646 *
3647 * Maps the page associated with @frag to @device.
3648 */
3649static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3650 const skb_frag_t *frag,
3651 size_t offset, size_t size,
3652 enum dma_data_direction dir)
3653{
3654 return dma_map_page(dev, skb_frag_page(frag),
3655 skb_frag_off(frag) + offset, size, dir);
3656}
3657
3658static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3659 gfp_t gfp_mask)
3660{
3661 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3662}
3663
3664
3665static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3666 gfp_t gfp_mask)
3667{
3668 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3669}
3670
3671
3672/**
3673 * skb_clone_writable - is the header of a clone writable
3674 * @skb: buffer to check
3675 * @len: length up to which to write
3676 *
3677 * Returns true if modifying the header part of the cloned buffer
3678 * does not requires the data to be copied.
3679 */
3680static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3681{
3682 return !skb_header_cloned(skb) &&
3683 skb_headroom(skb) + len <= skb->hdr_len;
3684}
3685
3686static inline int skb_try_make_writable(struct sk_buff *skb,
3687 unsigned int write_len)
3688{
3689 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3690 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3691}
3692
3693static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3694 int cloned)
3695{
3696 int delta = 0;
3697
3698 if (headroom > skb_headroom(skb))
3699 delta = headroom - skb_headroom(skb);
3700
3701 if (delta || cloned)
3702 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3703 GFP_ATOMIC);
3704 return 0;
3705}
3706
3707/**
3708 * skb_cow - copy header of skb when it is required
3709 * @skb: buffer to cow
3710 * @headroom: needed headroom
3711 *
3712 * If the skb passed lacks sufficient headroom or its data part
3713 * is shared, data is reallocated. If reallocation fails, an error
3714 * is returned and original skb is not changed.
3715 *
3716 * The result is skb with writable area skb->head...skb->tail
3717 * and at least @headroom of space at head.
3718 */
3719static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3720{
3721 return __skb_cow(skb, headroom, skb_cloned(skb));
3722}
3723
3724/**
3725 * skb_cow_head - skb_cow but only making the head writable
3726 * @skb: buffer to cow
3727 * @headroom: needed headroom
3728 *
3729 * This function is identical to skb_cow except that we replace the
3730 * skb_cloned check by skb_header_cloned. It should be used when
3731 * you only need to push on some header and do not need to modify
3732 * the data.
3733 */
3734static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3735{
3736 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3737}
3738
3739/**
3740 * skb_padto - pad an skbuff up to a minimal size
3741 * @skb: buffer to pad
3742 * @len: minimal length
3743 *
3744 * Pads up a buffer to ensure the trailing bytes exist and are
3745 * blanked. If the buffer already contains sufficient data it
3746 * is untouched. Otherwise it is extended. Returns zero on
3747 * success. The skb is freed on error.
3748 */
3749static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3750{
3751 unsigned int size = skb->len;
3752 if (likely(size >= len))
3753 return 0;
3754 return skb_pad(skb, len - size);
3755}
3756
3757/**
3758 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3759 * @skb: buffer to pad
3760 * @len: minimal length
3761 * @free_on_error: free buffer on error
3762 *
3763 * Pads up a buffer to ensure the trailing bytes exist and are
3764 * blanked. If the buffer already contains sufficient data it
3765 * is untouched. Otherwise it is extended. Returns zero on
3766 * success. The skb is freed on error if @free_on_error is true.
3767 */
3768static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3769 unsigned int len,
3770 bool free_on_error)
3771{
3772 unsigned int size = skb->len;
3773
3774 if (unlikely(size < len)) {
3775 len -= size;
3776 if (__skb_pad(skb, len, free_on_error))
3777 return -ENOMEM;
3778 __skb_put(skb, len);
3779 }
3780 return 0;
3781}
3782
3783/**
3784 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3785 * @skb: buffer to pad
3786 * @len: minimal length
3787 *
3788 * Pads up a buffer to ensure the trailing bytes exist and are
3789 * blanked. If the buffer already contains sufficient data it
3790 * is untouched. Otherwise it is extended. Returns zero on
3791 * success. The skb is freed on error.
3792 */
3793static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3794{
3795 return __skb_put_padto(skb, len, true);
3796}
3797
3798bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i)
3799 __must_check;
3800
3801static inline int skb_add_data(struct sk_buff *skb,
3802 struct iov_iter *from, int copy)
3803{
3804 const int off = skb->len;
3805
3806 if (skb->ip_summed == CHECKSUM_NONE) {
3807 __wsum csum = 0;
3808 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3809 &csum, from)) {
3810 skb->csum = csum_block_add(skb->csum, csum, off);
3811 return 0;
3812 }
3813 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3814 return 0;
3815
3816 __skb_trim(skb, off);
3817 return -EFAULT;
3818}
3819
3820static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3821 const struct page *page, int off)
3822{
3823 if (skb_zcopy(skb))
3824 return false;
3825 if (i) {
3826 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3827
3828 return page == skb_frag_page(frag) &&
3829 off == skb_frag_off(frag) + skb_frag_size(frag);
3830 }
3831 return false;
3832}
3833
3834static inline int __skb_linearize(struct sk_buff *skb)
3835{
3836 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3837}
3838
3839/**
3840 * skb_linearize - convert paged skb to linear one
3841 * @skb: buffer to linarize
3842 *
3843 * If there is no free memory -ENOMEM is returned, otherwise zero
3844 * is returned and the old skb data released.
3845 */
3846static inline int skb_linearize(struct sk_buff *skb)
3847{
3848 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3849}
3850
3851/**
3852 * skb_has_shared_frag - can any frag be overwritten
3853 * @skb: buffer to test
3854 *
3855 * Return true if the skb has at least one frag that might be modified
3856 * by an external entity (as in vmsplice()/sendfile())
3857 */
3858static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3859{
3860 return skb_is_nonlinear(skb) &&
3861 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3862}
3863
3864/**
3865 * skb_linearize_cow - make sure skb is linear and writable
3866 * @skb: buffer to process
3867 *
3868 * If there is no free memory -ENOMEM is returned, otherwise zero
3869 * is returned and the old skb data released.
3870 */
3871static inline int skb_linearize_cow(struct sk_buff *skb)
3872{
3873 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3874 __skb_linearize(skb) : 0;
3875}
3876
3877static __always_inline void
3878__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3879 unsigned int off)
3880{
3881 if (skb->ip_summed == CHECKSUM_COMPLETE)
3882 skb->csum = csum_block_sub(skb->csum,
3883 csum_partial(start, len, 0), off);
3884 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3885 skb_checksum_start_offset(skb) < 0)
3886 skb->ip_summed = CHECKSUM_NONE;
3887}
3888
3889/**
3890 * skb_postpull_rcsum - update checksum for received skb after pull
3891 * @skb: buffer to update
3892 * @start: start of data before pull
3893 * @len: length of data pulled
3894 *
3895 * After doing a pull on a received packet, you need to call this to
3896 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3897 * CHECKSUM_NONE so that it can be recomputed from scratch.
3898 */
3899static inline void skb_postpull_rcsum(struct sk_buff *skb,
3900 const void *start, unsigned int len)
3901{
3902 if (skb->ip_summed == CHECKSUM_COMPLETE)
3903 skb->csum = wsum_negate(csum_partial(start, len,
3904 wsum_negate(skb->csum)));
3905 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3906 skb_checksum_start_offset(skb) < 0)
3907 skb->ip_summed = CHECKSUM_NONE;
3908}
3909
3910static __always_inline void
3911__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3912 unsigned int off)
3913{
3914 if (skb->ip_summed == CHECKSUM_COMPLETE)
3915 skb->csum = csum_block_add(skb->csum,
3916 csum_partial(start, len, 0), off);
3917}
3918
3919/**
3920 * skb_postpush_rcsum - update checksum for received skb after push
3921 * @skb: buffer to update
3922 * @start: start of data after push
3923 * @len: length of data pushed
3924 *
3925 * After doing a push on a received packet, you need to call this to
3926 * update the CHECKSUM_COMPLETE checksum.
3927 */
3928static inline void skb_postpush_rcsum(struct sk_buff *skb,
3929 const void *start, unsigned int len)
3930{
3931 __skb_postpush_rcsum(skb, start, len, 0);
3932}
3933
3934void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3935
3936/**
3937 * skb_push_rcsum - push skb and update receive checksum
3938 * @skb: buffer to update
3939 * @len: length of data pulled
3940 *
3941 * This function performs an skb_push on the packet and updates
3942 * the CHECKSUM_COMPLETE checksum. It should be used on
3943 * receive path processing instead of skb_push unless you know
3944 * that the checksum difference is zero (e.g., a valid IP header)
3945 * or you are setting ip_summed to CHECKSUM_NONE.
3946 */
3947static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3948{
3949 skb_push(skb, len);
3950 skb_postpush_rcsum(skb, skb->data, len);
3951 return skb->data;
3952}
3953
3954int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3955/**
3956 * pskb_trim_rcsum - trim received skb and update checksum
3957 * @skb: buffer to trim
3958 * @len: new length
3959 *
3960 * This is exactly the same as pskb_trim except that it ensures the
3961 * checksum of received packets are still valid after the operation.
3962 * It can change skb pointers.
3963 */
3964
3965static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3966{
3967 if (likely(len >= skb->len))
3968 return 0;
3969 return pskb_trim_rcsum_slow(skb, len);
3970}
3971
3972static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3973{
3974 if (skb->ip_summed == CHECKSUM_COMPLETE)
3975 skb->ip_summed = CHECKSUM_NONE;
3976 __skb_trim(skb, len);
3977 return 0;
3978}
3979
3980static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3981{
3982 if (skb->ip_summed == CHECKSUM_COMPLETE)
3983 skb->ip_summed = CHECKSUM_NONE;
3984 return __skb_grow(skb, len);
3985}
3986
3987#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3988#define skb_rb_first(root) rb_to_skb(rb_first(root))
3989#define skb_rb_last(root) rb_to_skb(rb_last(root))
3990#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3991#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3992
3993#define skb_queue_walk(queue, skb) \
3994 for (skb = (queue)->next; \
3995 skb != (struct sk_buff *)(queue); \
3996 skb = skb->next)
3997
3998#define skb_queue_walk_safe(queue, skb, tmp) \
3999 for (skb = (queue)->next, tmp = skb->next; \
4000 skb != (struct sk_buff *)(queue); \
4001 skb = tmp, tmp = skb->next)
4002
4003#define skb_queue_walk_from(queue, skb) \
4004 for (; skb != (struct sk_buff *)(queue); \
4005 skb = skb->next)
4006
4007#define skb_rbtree_walk(skb, root) \
4008 for (skb = skb_rb_first(root); skb != NULL; \
4009 skb = skb_rb_next(skb))
4010
4011#define skb_rbtree_walk_from(skb) \
4012 for (; skb != NULL; \
4013 skb = skb_rb_next(skb))
4014
4015#define skb_rbtree_walk_from_safe(skb, tmp) \
4016 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
4017 skb = tmp)
4018
4019#define skb_queue_walk_from_safe(queue, skb, tmp) \
4020 for (tmp = skb->next; \
4021 skb != (struct sk_buff *)(queue); \
4022 skb = tmp, tmp = skb->next)
4023
4024#define skb_queue_reverse_walk(queue, skb) \
4025 for (skb = (queue)->prev; \
4026 skb != (struct sk_buff *)(queue); \
4027 skb = skb->prev)
4028
4029#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
4030 for (skb = (queue)->prev, tmp = skb->prev; \
4031 skb != (struct sk_buff *)(queue); \
4032 skb = tmp, tmp = skb->prev)
4033
4034#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
4035 for (tmp = skb->prev; \
4036 skb != (struct sk_buff *)(queue); \
4037 skb = tmp, tmp = skb->prev)
4038
4039static inline bool skb_has_frag_list(const struct sk_buff *skb)
4040{
4041 return skb_shinfo(skb)->frag_list != NULL;
4042}
4043
4044static inline void skb_frag_list_init(struct sk_buff *skb)
4045{
4046 skb_shinfo(skb)->frag_list = NULL;
4047}
4048
4049#define skb_walk_frags(skb, iter) \
4050 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
4051
4052
4053int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
4054 int *err, long *timeo_p,
4055 const struct sk_buff *skb);
4056struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
4057 struct sk_buff_head *queue,
4058 unsigned int flags,
4059 int *off, int *err,
4060 struct sk_buff **last);
4061struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
4062 struct sk_buff_head *queue,
4063 unsigned int flags, int *off, int *err,
4064 struct sk_buff **last);
4065struct sk_buff *__skb_recv_datagram(struct sock *sk,
4066 struct sk_buff_head *sk_queue,
4067 unsigned int flags, int *off, int *err);
4068struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
4069__poll_t datagram_poll(struct file *file, struct socket *sock,
4070 struct poll_table_struct *wait);
4071int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
4072 struct iov_iter *to, int size);
4073static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
4074 struct msghdr *msg, int size)
4075{
4076 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
4077}
4078int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
4079 struct msghdr *msg);
4080int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
4081 struct iov_iter *to, int len,
4082 struct ahash_request *hash);
4083int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
4084 struct iov_iter *from, int len);
4085int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
4086void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
4087int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
4088int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
4089int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
4090__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
4091 int len);
4092int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
4093 struct pipe_inode_info *pipe, unsigned int len,
4094 unsigned int flags);
4095int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
4096 int len);
4097int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
4098void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
4099unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
4100int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
4101 int len, int hlen);
4102void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
4103int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
4104void skb_scrub_packet(struct sk_buff *skb, bool xnet);
4105struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
4106struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
4107 unsigned int offset);
4108struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
4109int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
4110int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev);
4111int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
4112int skb_vlan_pop(struct sk_buff *skb);
4113int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4114int skb_eth_pop(struct sk_buff *skb);
4115int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4116 const unsigned char *src);
4117int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4118 int mac_len, bool ethernet);
4119int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4120 bool ethernet);
4121int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4122int skb_mpls_dec_ttl(struct sk_buff *skb);
4123struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4124 gfp_t gfp);
4125
4126static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4127{
4128 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4129}
4130
4131static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4132{
4133 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4134}
4135
4136struct skb_checksum_ops {
4137 __wsum (*update)(const void *mem, int len, __wsum wsum);
4138 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4139};
4140
4141extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4142
4143__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4144 __wsum csum, const struct skb_checksum_ops *ops);
4145__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4146 __wsum csum);
4147
4148static inline void * __must_check
4149__skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4150 const void *data, int hlen, void *buffer)
4151{
4152 if (likely(hlen - offset >= len))
4153 return (void *)data + offset;
4154
4155 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4156 return NULL;
4157
4158 return buffer;
4159}
4160
4161static inline void * __must_check
4162skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4163{
4164 return __skb_header_pointer(skb, offset, len, skb->data,
4165 skb_headlen(skb), buffer);
4166}
4167
4168static inline void * __must_check
4169skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len)
4170{
4171 if (likely(skb_headlen(skb) - offset >= len))
4172 return skb->data + offset;
4173 return NULL;
4174}
4175
4176/**
4177 * skb_needs_linearize - check if we need to linearize a given skb
4178 * depending on the given device features.
4179 * @skb: socket buffer to check
4180 * @features: net device features
4181 *
4182 * Returns true if either:
4183 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4184 * 2. skb is fragmented and the device does not support SG.
4185 */
4186static inline bool skb_needs_linearize(struct sk_buff *skb,
4187 netdev_features_t features)
4188{
4189 return skb_is_nonlinear(skb) &&
4190 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4191 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4192}
4193
4194static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4195 void *to,
4196 const unsigned int len)
4197{
4198 memcpy(to, skb->data, len);
4199}
4200
4201static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4202 const int offset, void *to,
4203 const unsigned int len)
4204{
4205 memcpy(to, skb->data + offset, len);
4206}
4207
4208static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4209 const void *from,
4210 const unsigned int len)
4211{
4212 memcpy(skb->data, from, len);
4213}
4214
4215static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4216 const int offset,
4217 const void *from,
4218 const unsigned int len)
4219{
4220 memcpy(skb->data + offset, from, len);
4221}
4222
4223void skb_init(void);
4224
4225static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4226{
4227 return skb->tstamp;
4228}
4229
4230/**
4231 * skb_get_timestamp - get timestamp from a skb
4232 * @skb: skb to get stamp from
4233 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4234 *
4235 * Timestamps are stored in the skb as offsets to a base timestamp.
4236 * This function converts the offset back to a struct timeval and stores
4237 * it in stamp.
4238 */
4239static inline void skb_get_timestamp(const struct sk_buff *skb,
4240 struct __kernel_old_timeval *stamp)
4241{
4242 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4243}
4244
4245static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4246 struct __kernel_sock_timeval *stamp)
4247{
4248 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4249
4250 stamp->tv_sec = ts.tv_sec;
4251 stamp->tv_usec = ts.tv_nsec / 1000;
4252}
4253
4254static inline void skb_get_timestampns(const struct sk_buff *skb,
4255 struct __kernel_old_timespec *stamp)
4256{
4257 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4258
4259 stamp->tv_sec = ts.tv_sec;
4260 stamp->tv_nsec = ts.tv_nsec;
4261}
4262
4263static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4264 struct __kernel_timespec *stamp)
4265{
4266 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4267
4268 stamp->tv_sec = ts.tv_sec;
4269 stamp->tv_nsec = ts.tv_nsec;
4270}
4271
4272static inline void __net_timestamp(struct sk_buff *skb)
4273{
4274 skb->tstamp = ktime_get_real();
4275 skb->tstamp_type = SKB_CLOCK_REALTIME;
4276}
4277
4278static inline ktime_t net_timedelta(ktime_t t)
4279{
4280 return ktime_sub(ktime_get_real(), t);
4281}
4282
4283static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4284 u8 tstamp_type)
4285{
4286 skb->tstamp = kt;
4287
4288 if (kt)
4289 skb->tstamp_type = tstamp_type;
4290 else
4291 skb->tstamp_type = SKB_CLOCK_REALTIME;
4292}
4293
4294static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb,
4295 ktime_t kt, clockid_t clockid)
4296{
4297 u8 tstamp_type = SKB_CLOCK_REALTIME;
4298
4299 switch (clockid) {
4300 case CLOCK_REALTIME:
4301 break;
4302 case CLOCK_MONOTONIC:
4303 tstamp_type = SKB_CLOCK_MONOTONIC;
4304 break;
4305 case CLOCK_TAI:
4306 tstamp_type = SKB_CLOCK_TAI;
4307 break;
4308 default:
4309 WARN_ON_ONCE(1);
4310 kt = 0;
4311 }
4312
4313 skb_set_delivery_time(skb, kt, tstamp_type);
4314}
4315
4316DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4317
4318/* It is used in the ingress path to clear the delivery_time.
4319 * If needed, set the skb->tstamp to the (rcv) timestamp.
4320 */
4321static inline void skb_clear_delivery_time(struct sk_buff *skb)
4322{
4323 if (skb->tstamp_type) {
4324 skb->tstamp_type = SKB_CLOCK_REALTIME;
4325 if (static_branch_unlikely(&netstamp_needed_key))
4326 skb->tstamp = ktime_get_real();
4327 else
4328 skb->tstamp = 0;
4329 }
4330}
4331
4332static inline void skb_clear_tstamp(struct sk_buff *skb)
4333{
4334 if (skb->tstamp_type)
4335 return;
4336
4337 skb->tstamp = 0;
4338}
4339
4340static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4341{
4342 if (skb->tstamp_type)
4343 return 0;
4344
4345 return skb->tstamp;
4346}
4347
4348static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4349{
4350 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp)
4351 return skb->tstamp;
4352
4353 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4354 return ktime_get_real();
4355
4356 return 0;
4357}
4358
4359static inline u8 skb_metadata_len(const struct sk_buff *skb)
4360{
4361 return skb_shinfo(skb)->meta_len;
4362}
4363
4364static inline void *skb_metadata_end(const struct sk_buff *skb)
4365{
4366 return skb_mac_header(skb);
4367}
4368
4369static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4370 const struct sk_buff *skb_b,
4371 u8 meta_len)
4372{
4373 const void *a = skb_metadata_end(skb_a);
4374 const void *b = skb_metadata_end(skb_b);
4375 u64 diffs = 0;
4376
4377 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
4378 BITS_PER_LONG != 64)
4379 goto slow;
4380
4381 /* Using more efficient variant than plain call to memcmp(). */
4382 switch (meta_len) {
4383#define __it(x, op) (x -= sizeof(u##op))
4384#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4385 case 32: diffs |= __it_diff(a, b, 64);
4386 fallthrough;
4387 case 24: diffs |= __it_diff(a, b, 64);
4388 fallthrough;
4389 case 16: diffs |= __it_diff(a, b, 64);
4390 fallthrough;
4391 case 8: diffs |= __it_diff(a, b, 64);
4392 break;
4393 case 28: diffs |= __it_diff(a, b, 64);
4394 fallthrough;
4395 case 20: diffs |= __it_diff(a, b, 64);
4396 fallthrough;
4397 case 12: diffs |= __it_diff(a, b, 64);
4398 fallthrough;
4399 case 4: diffs |= __it_diff(a, b, 32);
4400 break;
4401 default:
4402slow:
4403 return memcmp(a - meta_len, b - meta_len, meta_len);
4404 }
4405 return diffs;
4406}
4407
4408static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4409 const struct sk_buff *skb_b)
4410{
4411 u8 len_a = skb_metadata_len(skb_a);
4412 u8 len_b = skb_metadata_len(skb_b);
4413
4414 if (!(len_a | len_b))
4415 return false;
4416
4417 return len_a != len_b ?
4418 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4419}
4420
4421static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4422{
4423 skb_shinfo(skb)->meta_len = meta_len;
4424}
4425
4426static inline void skb_metadata_clear(struct sk_buff *skb)
4427{
4428 skb_metadata_set(skb, 0);
4429}
4430
4431struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4432
4433#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4434
4435void skb_clone_tx_timestamp(struct sk_buff *skb);
4436bool skb_defer_rx_timestamp(struct sk_buff *skb);
4437
4438#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4439
4440static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4441{
4442}
4443
4444static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4445{
4446 return false;
4447}
4448
4449#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4450
4451/**
4452 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4453 *
4454 * PHY drivers may accept clones of transmitted packets for
4455 * timestamping via their phy_driver.txtstamp method. These drivers
4456 * must call this function to return the skb back to the stack with a
4457 * timestamp.
4458 *
4459 * @skb: clone of the original outgoing packet
4460 * @hwtstamps: hardware time stamps
4461 *
4462 */
4463void skb_complete_tx_timestamp(struct sk_buff *skb,
4464 struct skb_shared_hwtstamps *hwtstamps);
4465
4466void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4467 struct skb_shared_hwtstamps *hwtstamps,
4468 struct sock *sk, int tstype);
4469
4470/**
4471 * skb_tstamp_tx - queue clone of skb with send time stamps
4472 * @orig_skb: the original outgoing packet
4473 * @hwtstamps: hardware time stamps, may be NULL if not available
4474 *
4475 * If the skb has a socket associated, then this function clones the
4476 * skb (thus sharing the actual data and optional structures), stores
4477 * the optional hardware time stamping information (if non NULL) or
4478 * generates a software time stamp (otherwise), then queues the clone
4479 * to the error queue of the socket. Errors are silently ignored.
4480 */
4481void skb_tstamp_tx(struct sk_buff *orig_skb,
4482 struct skb_shared_hwtstamps *hwtstamps);
4483
4484/**
4485 * skb_tx_timestamp() - Driver hook for transmit timestamping
4486 *
4487 * Ethernet MAC Drivers should call this function in their hard_xmit()
4488 * function immediately before giving the sk_buff to the MAC hardware.
4489 *
4490 * Specifically, one should make absolutely sure that this function is
4491 * called before TX completion of this packet can trigger. Otherwise
4492 * the packet could potentially already be freed.
4493 *
4494 * @skb: A socket buffer.
4495 */
4496static inline void skb_tx_timestamp(struct sk_buff *skb)
4497{
4498 skb_clone_tx_timestamp(skb);
4499 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4500 skb_tstamp_tx(skb, NULL);
4501}
4502
4503/**
4504 * skb_complete_wifi_ack - deliver skb with wifi status
4505 *
4506 * @skb: the original outgoing packet
4507 * @acked: ack status
4508 *
4509 */
4510void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4511
4512__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4513__sum16 __skb_checksum_complete(struct sk_buff *skb);
4514
4515static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4516{
4517 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4518 skb->csum_valid ||
4519 (skb->ip_summed == CHECKSUM_PARTIAL &&
4520 skb_checksum_start_offset(skb) >= 0));
4521}
4522
4523/**
4524 * skb_checksum_complete - Calculate checksum of an entire packet
4525 * @skb: packet to process
4526 *
4527 * This function calculates the checksum over the entire packet plus
4528 * the value of skb->csum. The latter can be used to supply the
4529 * checksum of a pseudo header as used by TCP/UDP. It returns the
4530 * checksum.
4531 *
4532 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4533 * this function can be used to verify that checksum on received
4534 * packets. In that case the function should return zero if the
4535 * checksum is correct. In particular, this function will return zero
4536 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4537 * hardware has already verified the correctness of the checksum.
4538 */
4539static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4540{
4541 return skb_csum_unnecessary(skb) ?
4542 0 : __skb_checksum_complete(skb);
4543}
4544
4545static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4546{
4547 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4548 if (skb->csum_level == 0)
4549 skb->ip_summed = CHECKSUM_NONE;
4550 else
4551 skb->csum_level--;
4552 }
4553}
4554
4555static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4556{
4557 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4558 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4559 skb->csum_level++;
4560 } else if (skb->ip_summed == CHECKSUM_NONE) {
4561 skb->ip_summed = CHECKSUM_UNNECESSARY;
4562 skb->csum_level = 0;
4563 }
4564}
4565
4566static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4567{
4568 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4569 skb->ip_summed = CHECKSUM_NONE;
4570 skb->csum_level = 0;
4571 }
4572}
4573
4574/* Check if we need to perform checksum complete validation.
4575 *
4576 * Returns true if checksum complete is needed, false otherwise
4577 * (either checksum is unnecessary or zero checksum is allowed).
4578 */
4579static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4580 bool zero_okay,
4581 __sum16 check)
4582{
4583 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4584 skb->csum_valid = 1;
4585 __skb_decr_checksum_unnecessary(skb);
4586 return false;
4587 }
4588
4589 return true;
4590}
4591
4592/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4593 * in checksum_init.
4594 */
4595#define CHECKSUM_BREAK 76
4596
4597/* Unset checksum-complete
4598 *
4599 * Unset checksum complete can be done when packet is being modified
4600 * (uncompressed for instance) and checksum-complete value is
4601 * invalidated.
4602 */
4603static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4604{
4605 if (skb->ip_summed == CHECKSUM_COMPLETE)
4606 skb->ip_summed = CHECKSUM_NONE;
4607}
4608
4609/* Validate (init) checksum based on checksum complete.
4610 *
4611 * Return values:
4612 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4613 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4614 * checksum is stored in skb->csum for use in __skb_checksum_complete
4615 * non-zero: value of invalid checksum
4616 *
4617 */
4618static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4619 bool complete,
4620 __wsum psum)
4621{
4622 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4623 if (!csum_fold(csum_add(psum, skb->csum))) {
4624 skb->csum_valid = 1;
4625 return 0;
4626 }
4627 }
4628
4629 skb->csum = psum;
4630
4631 if (complete || skb->len <= CHECKSUM_BREAK) {
4632 __sum16 csum;
4633
4634 csum = __skb_checksum_complete(skb);
4635 skb->csum_valid = !csum;
4636 return csum;
4637 }
4638
4639 return 0;
4640}
4641
4642static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4643{
4644 return 0;
4645}
4646
4647/* Perform checksum validate (init). Note that this is a macro since we only
4648 * want to calculate the pseudo header which is an input function if necessary.
4649 * First we try to validate without any computation (checksum unnecessary) and
4650 * then calculate based on checksum complete calling the function to compute
4651 * pseudo header.
4652 *
4653 * Return values:
4654 * 0: checksum is validated or try to in skb_checksum_complete
4655 * non-zero: value of invalid checksum
4656 */
4657#define __skb_checksum_validate(skb, proto, complete, \
4658 zero_okay, check, compute_pseudo) \
4659({ \
4660 __sum16 __ret = 0; \
4661 skb->csum_valid = 0; \
4662 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4663 __ret = __skb_checksum_validate_complete(skb, \
4664 complete, compute_pseudo(skb, proto)); \
4665 __ret; \
4666})
4667
4668#define skb_checksum_init(skb, proto, compute_pseudo) \
4669 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4670
4671#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4672 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4673
4674#define skb_checksum_validate(skb, proto, compute_pseudo) \
4675 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4676
4677#define skb_checksum_validate_zero_check(skb, proto, check, \
4678 compute_pseudo) \
4679 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4680
4681#define skb_checksum_simple_validate(skb) \
4682 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4683
4684static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4685{
4686 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4687}
4688
4689static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4690{
4691 skb->csum = ~pseudo;
4692 skb->ip_summed = CHECKSUM_COMPLETE;
4693}
4694
4695#define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4696do { \
4697 if (__skb_checksum_convert_check(skb)) \
4698 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4699} while (0)
4700
4701static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4702 u16 start, u16 offset)
4703{
4704 skb->ip_summed = CHECKSUM_PARTIAL;
4705 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4706 skb->csum_offset = offset - start;
4707}
4708
4709/* Update skbuf and packet to reflect the remote checksum offload operation.
4710 * When called, ptr indicates the starting point for skb->csum when
4711 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4712 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4713 */
4714static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4715 int start, int offset, bool nopartial)
4716{
4717 __wsum delta;
4718
4719 if (!nopartial) {
4720 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4721 return;
4722 }
4723
4724 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4725 __skb_checksum_complete(skb);
4726 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4727 }
4728
4729 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4730
4731 /* Adjust skb->csum since we changed the packet */
4732 skb->csum = csum_add(skb->csum, delta);
4733}
4734
4735static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4736{
4737#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4738 return (void *)(skb->_nfct & NFCT_PTRMASK);
4739#else
4740 return NULL;
4741#endif
4742}
4743
4744static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4745{
4746#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4747 return skb->_nfct;
4748#else
4749 return 0UL;
4750#endif
4751}
4752
4753static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4754{
4755#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4756 skb->slow_gro |= !!nfct;
4757 skb->_nfct = nfct;
4758#endif
4759}
4760
4761#ifdef CONFIG_SKB_EXTENSIONS
4762enum skb_ext_id {
4763#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4764 SKB_EXT_BRIDGE_NF,
4765#endif
4766#ifdef CONFIG_XFRM
4767 SKB_EXT_SEC_PATH,
4768#endif
4769#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4770 TC_SKB_EXT,
4771#endif
4772#if IS_ENABLED(CONFIG_MPTCP)
4773 SKB_EXT_MPTCP,
4774#endif
4775#if IS_ENABLED(CONFIG_MCTP_FLOWS)
4776 SKB_EXT_MCTP,
4777#endif
4778 SKB_EXT_NUM, /* must be last */
4779};
4780
4781/**
4782 * struct skb_ext - sk_buff extensions
4783 * @refcnt: 1 on allocation, deallocated on 0
4784 * @offset: offset to add to @data to obtain extension address
4785 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4786 * @data: start of extension data, variable sized
4787 *
4788 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4789 * to use 'u8' types while allowing up to 2kb worth of extension data.
4790 */
4791struct skb_ext {
4792 refcount_t refcnt;
4793 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4794 u8 chunks; /* same */
4795 char data[] __aligned(8);
4796};
4797
4798struct skb_ext *__skb_ext_alloc(gfp_t flags);
4799void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4800 struct skb_ext *ext);
4801void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4802void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4803void __skb_ext_put(struct skb_ext *ext);
4804
4805static inline void skb_ext_put(struct sk_buff *skb)
4806{
4807 if (skb->active_extensions)
4808 __skb_ext_put(skb->extensions);
4809}
4810
4811static inline void __skb_ext_copy(struct sk_buff *dst,
4812 const struct sk_buff *src)
4813{
4814 dst->active_extensions = src->active_extensions;
4815
4816 if (src->active_extensions) {
4817 struct skb_ext *ext = src->extensions;
4818
4819 refcount_inc(&ext->refcnt);
4820 dst->extensions = ext;
4821 }
4822}
4823
4824static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4825{
4826 skb_ext_put(dst);
4827 __skb_ext_copy(dst, src);
4828}
4829
4830static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4831{
4832 return !!ext->offset[i];
4833}
4834
4835static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4836{
4837 return skb->active_extensions & (1 << id);
4838}
4839
4840static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4841{
4842 if (skb_ext_exist(skb, id))
4843 __skb_ext_del(skb, id);
4844}
4845
4846static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4847{
4848 if (skb_ext_exist(skb, id)) {
4849 struct skb_ext *ext = skb->extensions;
4850
4851 return (void *)ext + (ext->offset[id] << 3);
4852 }
4853
4854 return NULL;
4855}
4856
4857static inline void skb_ext_reset(struct sk_buff *skb)
4858{
4859 if (unlikely(skb->active_extensions)) {
4860 __skb_ext_put(skb->extensions);
4861 skb->active_extensions = 0;
4862 }
4863}
4864
4865static inline bool skb_has_extensions(struct sk_buff *skb)
4866{
4867 return unlikely(skb->active_extensions);
4868}
4869#else
4870static inline void skb_ext_put(struct sk_buff *skb) {}
4871static inline void skb_ext_reset(struct sk_buff *skb) {}
4872static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4873static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4874static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4875static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4876#endif /* CONFIG_SKB_EXTENSIONS */
4877
4878static inline void nf_reset_ct(struct sk_buff *skb)
4879{
4880#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4881 nf_conntrack_put(skb_nfct(skb));
4882 skb->_nfct = 0;
4883#endif
4884}
4885
4886static inline void nf_reset_trace(struct sk_buff *skb)
4887{
4888#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4889 skb->nf_trace = 0;
4890#endif
4891}
4892
4893static inline void ipvs_reset(struct sk_buff *skb)
4894{
4895#if IS_ENABLED(CONFIG_IP_VS)
4896 skb->ipvs_property = 0;
4897#endif
4898}
4899
4900/* Note: This doesn't put any conntrack info in dst. */
4901static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4902 bool copy)
4903{
4904#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4905 dst->_nfct = src->_nfct;
4906 nf_conntrack_get(skb_nfct(src));
4907#endif
4908#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4909 if (copy)
4910 dst->nf_trace = src->nf_trace;
4911#endif
4912}
4913
4914static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4915{
4916#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4917 nf_conntrack_put(skb_nfct(dst));
4918#endif
4919 dst->slow_gro = src->slow_gro;
4920 __nf_copy(dst, src, true);
4921}
4922
4923#ifdef CONFIG_NETWORK_SECMARK
4924static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4925{
4926 to->secmark = from->secmark;
4927}
4928
4929static inline void skb_init_secmark(struct sk_buff *skb)
4930{
4931 skb->secmark = 0;
4932}
4933#else
4934static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4935{ }
4936
4937static inline void skb_init_secmark(struct sk_buff *skb)
4938{ }
4939#endif
4940
4941static inline int secpath_exists(const struct sk_buff *skb)
4942{
4943#ifdef CONFIG_XFRM
4944 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4945#else
4946 return 0;
4947#endif
4948}
4949
4950static inline bool skb_irq_freeable(const struct sk_buff *skb)
4951{
4952 return !skb->destructor &&
4953 !secpath_exists(skb) &&
4954 !skb_nfct(skb) &&
4955 !skb->_skb_refdst &&
4956 !skb_has_frag_list(skb);
4957}
4958
4959static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4960{
4961 skb->queue_mapping = queue_mapping;
4962}
4963
4964static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4965{
4966 return skb->queue_mapping;
4967}
4968
4969static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4970{
4971 to->queue_mapping = from->queue_mapping;
4972}
4973
4974static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4975{
4976 skb->queue_mapping = rx_queue + 1;
4977}
4978
4979static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4980{
4981 return skb->queue_mapping - 1;
4982}
4983
4984static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4985{
4986 return skb->queue_mapping != 0;
4987}
4988
4989static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4990{
4991 skb->dst_pending_confirm = val;
4992}
4993
4994static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4995{
4996 return skb->dst_pending_confirm != 0;
4997}
4998
4999static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
5000{
5001#ifdef CONFIG_XFRM
5002 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
5003#else
5004 return NULL;
5005#endif
5006}
5007
5008static inline bool skb_is_gso(const struct sk_buff *skb)
5009{
5010 return skb_shinfo(skb)->gso_size;
5011}
5012
5013/* Note: Should be called only if skb_is_gso(skb) is true */
5014static inline bool skb_is_gso_v6(const struct sk_buff *skb)
5015{
5016 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
5017}
5018
5019/* Note: Should be called only if skb_is_gso(skb) is true */
5020static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
5021{
5022 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
5023}
5024
5025/* Note: Should be called only if skb_is_gso(skb) is true */
5026static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
5027{
5028 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
5029}
5030
5031static inline void skb_gso_reset(struct sk_buff *skb)
5032{
5033 skb_shinfo(skb)->gso_size = 0;
5034 skb_shinfo(skb)->gso_segs = 0;
5035 skb_shinfo(skb)->gso_type = 0;
5036}
5037
5038static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
5039 u16 increment)
5040{
5041 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5042 return;
5043 shinfo->gso_size += increment;
5044}
5045
5046static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
5047 u16 decrement)
5048{
5049 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5050 return;
5051 shinfo->gso_size -= decrement;
5052}
5053
5054void __skb_warn_lro_forwarding(const struct sk_buff *skb);
5055
5056static inline bool skb_warn_if_lro(const struct sk_buff *skb)
5057{
5058 /* LRO sets gso_size but not gso_type, whereas if GSO is really
5059 * wanted then gso_type will be set. */
5060 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5061
5062 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
5063 unlikely(shinfo->gso_type == 0)) {
5064 __skb_warn_lro_forwarding(skb);
5065 return true;
5066 }
5067 return false;
5068}
5069
5070static inline void skb_forward_csum(struct sk_buff *skb)
5071{
5072 /* Unfortunately we don't support this one. Any brave souls? */
5073 if (skb->ip_summed == CHECKSUM_COMPLETE)
5074 skb->ip_summed = CHECKSUM_NONE;
5075}
5076
5077/**
5078 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
5079 * @skb: skb to check
5080 *
5081 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
5082 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
5083 * use this helper, to document places where we make this assertion.
5084 */
5085static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5086{
5087 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5088}
5089
5090bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5091
5092int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5093struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5094 unsigned int transport_len,
5095 __sum16(*skb_chkf)(struct sk_buff *skb));
5096
5097/**
5098 * skb_head_is_locked - Determine if the skb->head is locked down
5099 * @skb: skb to check
5100 *
5101 * The head on skbs build around a head frag can be removed if they are
5102 * not cloned. This function returns true if the skb head is locked down
5103 * due to either being allocated via kmalloc, or by being a clone with
5104 * multiple references to the head.
5105 */
5106static inline bool skb_head_is_locked(const struct sk_buff *skb)
5107{
5108 return !skb->head_frag || skb_cloned(skb);
5109}
5110
5111/* Local Checksum Offload.
5112 * Compute outer checksum based on the assumption that the
5113 * inner checksum will be offloaded later.
5114 * See Documentation/networking/checksum-offloads.rst for
5115 * explanation of how this works.
5116 * Fill in outer checksum adjustment (e.g. with sum of outer
5117 * pseudo-header) before calling.
5118 * Also ensure that inner checksum is in linear data area.
5119 */
5120static inline __wsum lco_csum(struct sk_buff *skb)
5121{
5122 unsigned char *csum_start = skb_checksum_start(skb);
5123 unsigned char *l4_hdr = skb_transport_header(skb);
5124 __wsum partial;
5125
5126 /* Start with complement of inner checksum adjustment */
5127 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5128 skb->csum_offset));
5129
5130 /* Add in checksum of our headers (incl. outer checksum
5131 * adjustment filled in by caller) and return result.
5132 */
5133 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5134}
5135
5136static inline bool skb_is_redirected(const struct sk_buff *skb)
5137{
5138 return skb->redirected;
5139}
5140
5141static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5142{
5143 skb->redirected = 1;
5144#ifdef CONFIG_NET_REDIRECT
5145 skb->from_ingress = from_ingress;
5146 if (skb->from_ingress)
5147 skb_clear_tstamp(skb);
5148#endif
5149}
5150
5151static inline void skb_reset_redirect(struct sk_buff *skb)
5152{
5153 skb->redirected = 0;
5154}
5155
5156static inline void skb_set_redirected_noclear(struct sk_buff *skb,
5157 bool from_ingress)
5158{
5159 skb->redirected = 1;
5160#ifdef CONFIG_NET_REDIRECT
5161 skb->from_ingress = from_ingress;
5162#endif
5163}
5164
5165static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5166{
5167#if IS_ENABLED(CONFIG_IP_SCTP)
5168 return skb->csum_not_inet;
5169#else
5170 return 0;
5171#endif
5172}
5173
5174static inline void skb_reset_csum_not_inet(struct sk_buff *skb)
5175{
5176 skb->ip_summed = CHECKSUM_NONE;
5177#if IS_ENABLED(CONFIG_IP_SCTP)
5178 skb->csum_not_inet = 0;
5179#endif
5180}
5181
5182static inline void skb_set_kcov_handle(struct sk_buff *skb,
5183 const u64 kcov_handle)
5184{
5185#ifdef CONFIG_KCOV
5186 skb->kcov_handle = kcov_handle;
5187#endif
5188}
5189
5190static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5191{
5192#ifdef CONFIG_KCOV
5193 return skb->kcov_handle;
5194#else
5195 return 0;
5196#endif
5197}
5198
5199static inline void skb_mark_for_recycle(struct sk_buff *skb)
5200{
5201#ifdef CONFIG_PAGE_POOL
5202 skb->pp_recycle = 1;
5203#endif
5204}
5205
5206ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
5207 ssize_t maxsize, gfp_t gfp);
5208
5209#endif /* __KERNEL__ */
5210#endif /* _LINUX_SKBUFF_H */