blob: b04b5bd43f541456a0717429cf361ab12e8390c5 [file] [log] [blame]
David Brazdil0f672f62019-12-10 10:32:29 +00001/* SPDX-License-Identifier: GPL-2.0-or-later */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002/*
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>
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00008 */
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>
David Brazdil0f672f62019-12-10 10:32:29 +000017#include <linux/bvec.h>
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000018#include <linux/cache.h>
19#include <linux/rbtree.h>
20#include <linux/socket.h>
21#include <linux/refcount.h>
22
23#include <linux/atomic.h>
24#include <asm/types.h>
25#include <linux/spinlock.h>
26#include <linux/net.h>
27#include <linux/textsearch.h>
28#include <net/checksum.h>
29#include <linux/rcupdate.h>
30#include <linux/hrtimer.h>
31#include <linux/dma-mapping.h>
32#include <linux/netdev_features.h>
33#include <linux/sched.h>
34#include <linux/sched/clock.h>
35#include <net/flow_dissector.h>
36#include <linux/splice.h>
37#include <linux/in6.h>
38#include <linux/if_packet.h>
39#include <net/flow.h>
David Brazdil0f672f62019-12-10 10:32:29 +000040#if IS_ENABLED(CONFIG_NF_CONNTRACK)
41#include <linux/netfilter/nf_conntrack_common.h>
42#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000043
44/* The interface for checksum offload between the stack and networking drivers
45 * is as follows...
46 *
47 * A. IP checksum related features
48 *
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
52 * to its device.
53 *
54 * The checksum related features are:
55 *
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
61 *
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
70 *
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
80 *
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
86 *
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
89 *
90 * CHECKSUM_NONE:
91 *
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
95 *
96 * CHECKSUM_UNNECESSARY:
97 *
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
104 *
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 * FCOE: indicates the CRC in FC frame has been validated.
113 *
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
123 *
124 * CHECKSUM_COMPLETE:
125 *
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
129 *
130 * Notes:
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
134 *
135 * CHECKSUM_PARTIAL:
136 *
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
145 * be verified.
146 *
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
149 *
150 * CHECKSUM_PARTIAL:
151 *
152 * The driver is required to checksum the packet as seen by hard_start_xmit()
153 * from skb->csum_start up to the end, and to record/write the checksum at
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
159 * correctly.
160 *
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
165 *
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
173 *
174 * CHECKSUM_NONE:
175 *
176 * The skb was already checksummed by the protocol, or a checksum is not
177 * required.
178 *
179 * CHECKSUM_UNNECESSARY:
180 *
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
182 * output.
183 *
184 * CHECKSUM_COMPLETE:
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
187 *
188 * D. Non-IP checksum (CRC) offloads
189 *
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
199 *
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
207 *
208 * E. Checksumming on output with GSO.
209 *
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
217 */
218
219/* Don't change this without changing skb_csum_unnecessary! */
220#define CHECKSUM_NONE 0
221#define CHECKSUM_UNNECESSARY 1
222#define CHECKSUM_COMPLETE 2
223#define CHECKSUM_PARTIAL 3
224
225/* Maximum value in skb->csum_level */
226#define SKB_MAX_CSUM_LEVEL 3
227
228#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229#define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231#define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
235
236/* return minimum truesize of one skb containing X bytes of data */
237#define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
240
241struct net_device;
242struct scatterlist;
243struct pipe_inode_info;
244struct iov_iter;
245struct napi_struct;
David Brazdil0f672f62019-12-10 10:32:29 +0000246struct bpf_prog;
247union bpf_attr;
248struct skb_ext;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000249
250#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
251struct nf_bridge_info {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000252 enum {
253 BRNF_PROTO_UNCHANGED,
254 BRNF_PROTO_8021Q,
255 BRNF_PROTO_PPPOE
256 } orig_proto:8;
257 u8 pkt_otherhost:1;
258 u8 in_prerouting:1;
259 u8 bridged_dnat:1;
260 __u16 frag_max_size;
261 struct net_device *physindev;
262
263 /* always valid & non-NULL from FORWARD on, for physdev match */
264 struct net_device *physoutdev;
265 union {
266 /* prerouting: detect dnat in orig/reply direction */
267 __be32 ipv4_daddr;
268 struct in6_addr ipv6_daddr;
269
270 /* after prerouting + nat detected: store original source
271 * mac since neigh resolution overwrites it, only used while
272 * skb is out in neigh layer.
273 */
274 char neigh_header[8];
275 };
276};
277#endif
278
David Brazdil0f672f62019-12-10 10:32:29 +0000279#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
280/* Chain in tc_skb_ext will be used to share the tc chain with
281 * ovs recirc_id. It will be set to the current chain by tc
282 * and read by ovs to recirc_id.
283 */
284struct tc_skb_ext {
285 __u32 chain;
286};
287#endif
288
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000289struct sk_buff_head {
290 /* These two members must be first. */
291 struct sk_buff *next;
292 struct sk_buff *prev;
293
294 __u32 qlen;
295 spinlock_t lock;
296};
297
298struct sk_buff;
299
300/* To allow 64K frame to be packed as single skb without frag_list we
301 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
302 * buffers which do not start on a page boundary.
303 *
304 * Since GRO uses frags we allocate at least 16 regardless of page
305 * size.
306 */
307#if (65536/PAGE_SIZE + 1) < 16
308#define MAX_SKB_FRAGS 16UL
309#else
310#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
311#endif
312extern int sysctl_max_skb_frags;
313
314/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
315 * segment using its current segmentation instead.
316 */
317#define GSO_BY_FRAGS 0xFFFF
318
David Brazdil0f672f62019-12-10 10:32:29 +0000319typedef struct bio_vec skb_frag_t;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000320
David Brazdil0f672f62019-12-10 10:32:29 +0000321/**
322 * skb_frag_size() - Returns the size of a skb fragment
323 * @frag: skb fragment
324 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000325static inline unsigned int skb_frag_size(const skb_frag_t *frag)
326{
David Brazdil0f672f62019-12-10 10:32:29 +0000327 return frag->bv_len;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000328}
329
David Brazdil0f672f62019-12-10 10:32:29 +0000330/**
331 * skb_frag_size_set() - Sets the size of a skb fragment
332 * @frag: skb fragment
333 * @size: size of fragment
334 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000335static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
336{
David Brazdil0f672f62019-12-10 10:32:29 +0000337 frag->bv_len = size;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000338}
339
David Brazdil0f672f62019-12-10 10:32:29 +0000340/**
341 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
342 * @frag: skb fragment
343 * @delta: value to add
344 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000345static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
346{
David Brazdil0f672f62019-12-10 10:32:29 +0000347 frag->bv_len += delta;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000348}
349
David Brazdil0f672f62019-12-10 10:32:29 +0000350/**
351 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
352 * @frag: skb fragment
353 * @delta: value to subtract
354 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000355static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
356{
David Brazdil0f672f62019-12-10 10:32:29 +0000357 frag->bv_len -= delta;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000358}
359
David Brazdil0f672f62019-12-10 10:32:29 +0000360/**
361 * skb_frag_must_loop - Test if %p is a high memory page
362 * @p: fragment's page
363 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000364static inline bool skb_frag_must_loop(struct page *p)
365{
366#if defined(CONFIG_HIGHMEM)
367 if (PageHighMem(p))
368 return true;
369#endif
370 return false;
371}
372
373/**
374 * skb_frag_foreach_page - loop over pages in a fragment
375 *
376 * @f: skb frag to operate on
David Brazdil0f672f62019-12-10 10:32:29 +0000377 * @f_off: offset from start of f->bv_page
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000378 * @f_len: length from f_off to loop over
379 * @p: (temp var) current page
380 * @p_off: (temp var) offset from start of current page,
381 * non-zero only on first page.
382 * @p_len: (temp var) length in current page,
383 * < PAGE_SIZE only on first and last page.
384 * @copied: (temp var) length so far, excluding current p_len.
385 *
386 * A fragment can hold a compound page, in which case per-page
387 * operations, notably kmap_atomic, must be called for each
388 * regular page.
389 */
390#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
391 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
392 p_off = (f_off) & (PAGE_SIZE - 1), \
393 p_len = skb_frag_must_loop(p) ? \
394 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
395 copied = 0; \
396 copied < f_len; \
397 copied += p_len, p++, p_off = 0, \
398 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
399
400#define HAVE_HW_TIME_STAMP
401
402/**
403 * struct skb_shared_hwtstamps - hardware time stamps
404 * @hwtstamp: hardware time stamp transformed into duration
405 * since arbitrary point in time
406 *
407 * Software time stamps generated by ktime_get_real() are stored in
408 * skb->tstamp.
409 *
410 * hwtstamps can only be compared against other hwtstamps from
411 * the same device.
412 *
413 * This structure is attached to packets as part of the
414 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
415 */
416struct skb_shared_hwtstamps {
417 ktime_t hwtstamp;
418};
419
420/* Definitions for tx_flags in struct skb_shared_info */
421enum {
422 /* generate hardware time stamp */
423 SKBTX_HW_TSTAMP = 1 << 0,
424
425 /* generate software time stamp when queueing packet to NIC */
426 SKBTX_SW_TSTAMP = 1 << 1,
427
428 /* device driver is going to provide hardware time stamp */
429 SKBTX_IN_PROGRESS = 1 << 2,
430
431 /* device driver supports TX zero-copy buffers */
432 SKBTX_DEV_ZEROCOPY = 1 << 3,
433
434 /* generate wifi status information (where possible) */
435 SKBTX_WIFI_STATUS = 1 << 4,
436
437 /* This indicates at least one fragment might be overwritten
438 * (as in vmsplice(), sendfile() ...)
439 * If we need to compute a TX checksum, we'll need to copy
440 * all frags to avoid possible bad checksum
441 */
442 SKBTX_SHARED_FRAG = 1 << 5,
443
444 /* generate software time stamp when entering packet scheduling */
445 SKBTX_SCHED_TSTAMP = 1 << 6,
446};
447
448#define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
449#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
450 SKBTX_SCHED_TSTAMP)
451#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
452
453/*
454 * The callback notifies userspace to release buffers when skb DMA is done in
455 * lower device, the skb last reference should be 0 when calling this.
456 * The zerocopy_success argument is true if zero copy transmit occurred,
457 * false on data copy or out of memory error caused by data copy attempt.
458 * The ctx field is used to track device context.
459 * The desc field is used to track userspace buffer index.
460 */
461struct ubuf_info {
462 void (*callback)(struct ubuf_info *, bool zerocopy_success);
463 union {
464 struct {
465 unsigned long desc;
466 void *ctx;
467 };
468 struct {
469 u32 id;
470 u16 len;
471 u16 zerocopy:1;
472 u32 bytelen;
473 };
474 };
475 refcount_t refcnt;
476
477 struct mmpin {
478 struct user_struct *user;
479 unsigned int num_pg;
480 } mmp;
481};
482
483#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
484
485int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
486void mm_unaccount_pinned_pages(struct mmpin *mmp);
487
488struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
489struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
490 struct ubuf_info *uarg);
491
492static inline void sock_zerocopy_get(struct ubuf_info *uarg)
493{
494 refcount_inc(&uarg->refcnt);
495}
496
497void sock_zerocopy_put(struct ubuf_info *uarg);
David Brazdil0f672f62019-12-10 10:32:29 +0000498void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000499
500void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
501
David Brazdil0f672f62019-12-10 10:32:29 +0000502int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000503int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
504 struct msghdr *msg, int len,
505 struct ubuf_info *uarg);
506
507/* This data is invariant across clones and lives at
508 * the end of the header data, ie. at skb->end.
509 */
510struct skb_shared_info {
511 __u8 __unused;
512 __u8 meta_len;
513 __u8 nr_frags;
514 __u8 tx_flags;
515 unsigned short gso_size;
516 /* Warning: this field is not always filled in (UFO)! */
517 unsigned short gso_segs;
518 struct sk_buff *frag_list;
519 struct skb_shared_hwtstamps hwtstamps;
520 unsigned int gso_type;
521 u32 tskey;
522
523 /*
524 * Warning : all fields before dataref are cleared in __alloc_skb()
525 */
526 atomic_t dataref;
527
528 /* Intermediate layers must ensure that destructor_arg
529 * remains valid until skb destructor */
530 void * destructor_arg;
531
532 /* must be last field, see pskb_expand_head() */
533 skb_frag_t frags[MAX_SKB_FRAGS];
534};
535
536/* We divide dataref into two halves. The higher 16 bits hold references
537 * to the payload part of skb->data. The lower 16 bits hold references to
538 * the entire skb->data. A clone of a headerless skb holds the length of
539 * the header in skb->hdr_len.
540 *
541 * All users must obey the rule that the skb->data reference count must be
542 * greater than or equal to the payload reference count.
543 *
544 * Holding a reference to the payload part means that the user does not
545 * care about modifications to the header part of skb->data.
546 */
547#define SKB_DATAREF_SHIFT 16
548#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
549
550
551enum {
552 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
553 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
554 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
555};
556
557enum {
558 SKB_GSO_TCPV4 = 1 << 0,
559
560 /* This indicates the skb is from an untrusted source. */
561 SKB_GSO_DODGY = 1 << 1,
562
563 /* This indicates the tcp segment has CWR set. */
564 SKB_GSO_TCP_ECN = 1 << 2,
565
566 SKB_GSO_TCP_FIXEDID = 1 << 3,
567
568 SKB_GSO_TCPV6 = 1 << 4,
569
570 SKB_GSO_FCOE = 1 << 5,
571
572 SKB_GSO_GRE = 1 << 6,
573
574 SKB_GSO_GRE_CSUM = 1 << 7,
575
576 SKB_GSO_IPXIP4 = 1 << 8,
577
578 SKB_GSO_IPXIP6 = 1 << 9,
579
580 SKB_GSO_UDP_TUNNEL = 1 << 10,
581
582 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
583
584 SKB_GSO_PARTIAL = 1 << 12,
585
586 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
587
588 SKB_GSO_SCTP = 1 << 14,
589
590 SKB_GSO_ESP = 1 << 15,
591
592 SKB_GSO_UDP = 1 << 16,
593
594 SKB_GSO_UDP_L4 = 1 << 17,
595};
596
597#if BITS_PER_LONG > 32
598#define NET_SKBUFF_DATA_USES_OFFSET 1
599#endif
600
601#ifdef NET_SKBUFF_DATA_USES_OFFSET
602typedef unsigned int sk_buff_data_t;
603#else
604typedef unsigned char *sk_buff_data_t;
605#endif
606
David Brazdil0f672f62019-12-10 10:32:29 +0000607/**
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000608 * struct sk_buff - socket buffer
609 * @next: Next buffer in list
610 * @prev: Previous buffer in list
611 * @tstamp: Time we arrived/left
612 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
613 * @sk: Socket we are owned by
614 * @dev: Device we arrived on/are leaving by
615 * @cb: Control buffer. Free for use by every layer. Put private vars here
616 * @_skb_refdst: destination entry (with norefcount bit)
617 * @sp: the security path, used for xfrm
618 * @len: Length of actual data
619 * @data_len: Data length
620 * @mac_len: Length of link layer header
621 * @hdr_len: writable header length of cloned skb
622 * @csum: Checksum (must include start/offset pair)
623 * @csum_start: Offset from skb->head where checksumming should start
624 * @csum_offset: Offset from csum_start where checksum should be stored
625 * @priority: Packet queueing priority
626 * @ignore_df: allow local fragmentation
627 * @cloned: Head may be cloned (check refcnt to be sure)
628 * @ip_summed: Driver fed us an IP checksum
629 * @nohdr: Payload reference only, must not modify header
630 * @pkt_type: Packet class
631 * @fclone: skbuff clone status
632 * @ipvs_property: skbuff is owned by ipvs
David Brazdil0f672f62019-12-10 10:32:29 +0000633 * @offload_fwd_mark: Packet was L2-forwarded in hardware
634 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000635 * @tc_skip_classify: do not classify packet. set by IFB device
636 * @tc_at_ingress: used within tc_classify to distinguish in/egress
Olivier Deprez0e641232021-09-23 10:07:05 +0200637 * @redirected: packet was redirected by packet classifier
638 * @from_ingress: packet was redirected from the ingress path
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000639 * @peeked: this packet has been seen already, so stats have been
640 * done for it, don't do them again
641 * @nf_trace: netfilter packet trace flag
642 * @protocol: Packet protocol from driver
643 * @destructor: Destruct function
644 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
645 * @_nfct: Associated connection, if any (with nfctinfo bits)
646 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
647 * @skb_iif: ifindex of device we arrived on
648 * @tc_index: Traffic control index
649 * @hash: the packet hash
650 * @queue_mapping: Queue mapping for multiqueue devices
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000651 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
David Brazdil0f672f62019-12-10 10:32:29 +0000652 * @active_extensions: active extensions (skb_ext_id types)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000653 * @ndisc_nodetype: router type (from link layer)
654 * @ooo_okay: allow the mapping of a socket to a queue to be changed
655 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
656 * ports.
657 * @sw_hash: indicates hash was computed in software stack
658 * @wifi_acked_valid: wifi_acked was set
659 * @wifi_acked: whether frame was acked on wifi or not
660 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
661 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
662 * @dst_pending_confirm: need to confirm neighbour
663 * @decrypted: Decrypted SKB
David Brazdil0f672f62019-12-10 10:32:29 +0000664 * @napi_id: id of the NAPI struct this skb came from
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000665 * @secmark: security marking
666 * @mark: Generic packet mark
667 * @vlan_proto: vlan encapsulation protocol
668 * @vlan_tci: vlan tag control information
669 * @inner_protocol: Protocol (encapsulation)
670 * @inner_transport_header: Inner transport layer header (encapsulation)
671 * @inner_network_header: Network layer header (encapsulation)
672 * @inner_mac_header: Link layer header (encapsulation)
673 * @transport_header: Transport layer header
674 * @network_header: Network layer header
675 * @mac_header: Link layer header
676 * @tail: Tail pointer
677 * @end: End pointer
678 * @head: Head of buffer
679 * @data: Data head pointer
680 * @truesize: Buffer size
681 * @users: User count - see {datagram,tcp}.c
David Brazdil0f672f62019-12-10 10:32:29 +0000682 * @extensions: allocated extensions, valid if active_extensions is nonzero
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000683 */
684
685struct sk_buff {
686 union {
687 struct {
688 /* These two members must be first. */
689 struct sk_buff *next;
690 struct sk_buff *prev;
691
692 union {
693 struct net_device *dev;
694 /* Some protocols might use this space to store information,
695 * while device pointer would be NULL.
696 * UDP receive path is one user.
697 */
698 unsigned long dev_scratch;
699 };
700 };
701 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
702 struct list_head list;
703 };
704
705 union {
706 struct sock *sk;
707 int ip_defrag_offset;
708 };
709
710 union {
711 ktime_t tstamp;
David Brazdil0f672f62019-12-10 10:32:29 +0000712 u64 skb_mstamp_ns; /* earliest departure time */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000713 };
714 /*
715 * This is the control buffer. It is free to use for every
716 * layer. Please put your private variables there. If you
717 * want to keep them across layers you have to do a skb_clone()
718 * first. This is owned by whoever has the skb queued ATM.
719 */
720 char cb[48] __aligned(8);
721
722 union {
723 struct {
724 unsigned long _skb_refdst;
725 void (*destructor)(struct sk_buff *skb);
726 };
727 struct list_head tcp_tsorted_anchor;
728 };
729
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000730#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
731 unsigned long _nfct;
732#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000733 unsigned int len,
734 data_len;
735 __u16 mac_len,
736 hdr_len;
737
738 /* Following fields are _not_ copied in __copy_skb_header()
739 * Note that queue_mapping is here mostly to fill a hole.
740 */
741 __u16 queue_mapping;
742
743/* if you move cloned around you also must adapt those constants */
744#ifdef __BIG_ENDIAN_BITFIELD
745#define CLONED_MASK (1 << 7)
746#else
747#define CLONED_MASK 1
748#endif
749#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
750
751 __u8 __cloned_offset[0];
752 __u8 cloned:1,
753 nohdr:1,
754 fclone:2,
755 peeked:1,
756 head_frag:1,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000757 pfmemalloc:1;
David Brazdil0f672f62019-12-10 10:32:29 +0000758#ifdef CONFIG_SKB_EXTENSIONS
759 __u8 active_extensions;
760#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000761 /* fields enclosed in headers_start/headers_end are copied
762 * using a single memcpy() in __copy_skb_header()
763 */
764 /* private: */
765 __u32 headers_start[0];
766 /* public: */
767
768/* if you move pkt_type around you also must adapt those constants */
769#ifdef __BIG_ENDIAN_BITFIELD
770#define PKT_TYPE_MAX (7 << 5)
771#else
772#define PKT_TYPE_MAX 7
773#endif
774#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
775
776 __u8 __pkt_type_offset[0];
777 __u8 pkt_type:3;
778 __u8 ignore_df:1;
779 __u8 nf_trace:1;
780 __u8 ip_summed:2;
781 __u8 ooo_okay:1;
782
783 __u8 l4_hash:1;
784 __u8 sw_hash:1;
785 __u8 wifi_acked_valid:1;
786 __u8 wifi_acked:1;
787 __u8 no_fcs:1;
788 /* Indicates the inner headers are valid in the skbuff. */
789 __u8 encapsulation:1;
790 __u8 encap_hdr_csum:1;
791 __u8 csum_valid:1;
792
David Brazdil0f672f62019-12-10 10:32:29 +0000793#ifdef __BIG_ENDIAN_BITFIELD
794#define PKT_VLAN_PRESENT_BIT 7
795#else
796#define PKT_VLAN_PRESENT_BIT 0
797#endif
798#define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
799 __u8 __pkt_vlan_present_offset[0];
800 __u8 vlan_present:1;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000801 __u8 csum_complete_sw:1;
802 __u8 csum_level:2;
803 __u8 csum_not_inet:1;
804 __u8 dst_pending_confirm:1;
805#ifdef CONFIG_IPV6_NDISC_NODETYPE
806 __u8 ndisc_nodetype:2;
807#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000808
David Brazdil0f672f62019-12-10 10:32:29 +0000809 __u8 ipvs_property:1;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000810 __u8 inner_protocol_type:1;
811 __u8 remcsum_offload:1;
812#ifdef CONFIG_NET_SWITCHDEV
813 __u8 offload_fwd_mark:1;
David Brazdil0f672f62019-12-10 10:32:29 +0000814 __u8 offload_l3_fwd_mark:1;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000815#endif
816#ifdef CONFIG_NET_CLS_ACT
817 __u8 tc_skip_classify:1;
818 __u8 tc_at_ingress:1;
Olivier Deprez0e641232021-09-23 10:07:05 +0200819#endif
820#ifdef CONFIG_NET_REDIRECT
821 __u8 redirected:1;
822 __u8 from_ingress:1;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000823#endif
824#ifdef CONFIG_TLS_DEVICE
825 __u8 decrypted:1;
826#endif
827
828#ifdef CONFIG_NET_SCHED
829 __u16 tc_index; /* traffic control index */
830#endif
831
832 union {
833 __wsum csum;
834 struct {
835 __u16 csum_start;
836 __u16 csum_offset;
837 };
838 };
839 __u32 priority;
840 int skb_iif;
841 __u32 hash;
842 __be16 vlan_proto;
843 __u16 vlan_tci;
844#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
845 union {
846 unsigned int napi_id;
847 unsigned int sender_cpu;
848 };
849#endif
850#ifdef CONFIG_NETWORK_SECMARK
851 __u32 secmark;
852#endif
853
854 union {
855 __u32 mark;
856 __u32 reserved_tailroom;
857 };
858
859 union {
860 __be16 inner_protocol;
861 __u8 inner_ipproto;
862 };
863
864 __u16 inner_transport_header;
865 __u16 inner_network_header;
866 __u16 inner_mac_header;
867
868 __be16 protocol;
869 __u16 transport_header;
870 __u16 network_header;
871 __u16 mac_header;
872
873 /* private: */
874 __u32 headers_end[0];
875 /* public: */
876
877 /* These elements must be at the end, see alloc_skb() for details. */
878 sk_buff_data_t tail;
879 sk_buff_data_t end;
880 unsigned char *head,
881 *data;
882 unsigned int truesize;
883 refcount_t users;
David Brazdil0f672f62019-12-10 10:32:29 +0000884
885#ifdef CONFIG_SKB_EXTENSIONS
886 /* only useable after checking ->active_extensions != 0 */
887 struct skb_ext *extensions;
888#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000889};
890
891#ifdef __KERNEL__
892/*
893 * Handling routines are only of interest to the kernel
894 */
895
896#define SKB_ALLOC_FCLONE 0x01
897#define SKB_ALLOC_RX 0x02
898#define SKB_ALLOC_NAPI 0x04
899
David Brazdil0f672f62019-12-10 10:32:29 +0000900/**
901 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
902 * @skb: buffer
903 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000904static inline bool skb_pfmemalloc(const struct sk_buff *skb)
905{
906 return unlikely(skb->pfmemalloc);
907}
908
909/*
910 * skb might have a dst pointer attached, refcounted or not.
911 * _skb_refdst low order bit is set if refcount was _not_ taken
912 */
913#define SKB_DST_NOREF 1UL
914#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
915
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000916/**
917 * skb_dst - returns skb dst_entry
918 * @skb: buffer
919 *
920 * Returns skb dst_entry, regardless of reference taken or not.
921 */
922static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
923{
David Brazdil0f672f62019-12-10 10:32:29 +0000924 /* If refdst was not refcounted, check we still are in a
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000925 * rcu_read_lock section
926 */
927 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
928 !rcu_read_lock_held() &&
929 !rcu_read_lock_bh_held());
930 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
931}
932
933/**
934 * skb_dst_set - sets skb dst
935 * @skb: buffer
936 * @dst: dst entry
937 *
938 * Sets skb dst, assuming a reference was taken on dst and should
939 * be released by skb_dst_drop()
940 */
941static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
942{
943 skb->_skb_refdst = (unsigned long)dst;
944}
945
946/**
947 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
948 * @skb: buffer
949 * @dst: dst entry
950 *
951 * Sets skb dst, assuming a reference was not taken on dst.
952 * If dst entry is cached, we do not take reference and dst_release
953 * will be avoided by refdst_drop. If dst entry is not cached, we take
954 * reference, so that last dst_release can destroy the dst immediately.
955 */
956static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
957{
958 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
959 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
960}
961
962/**
963 * skb_dst_is_noref - Test if skb dst isn't refcounted
964 * @skb: buffer
965 */
966static inline bool skb_dst_is_noref(const struct sk_buff *skb)
967{
968 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
969}
970
David Brazdil0f672f62019-12-10 10:32:29 +0000971/**
972 * skb_rtable - Returns the skb &rtable
973 * @skb: buffer
974 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000975static inline struct rtable *skb_rtable(const struct sk_buff *skb)
976{
977 return (struct rtable *)skb_dst(skb);
978}
979
980/* For mangling skb->pkt_type from user space side from applications
981 * such as nft, tc, etc, we only allow a conservative subset of
982 * possible pkt_types to be set.
983*/
984static inline bool skb_pkt_type_ok(u32 ptype)
985{
986 return ptype <= PACKET_OTHERHOST;
987}
988
David Brazdil0f672f62019-12-10 10:32:29 +0000989/**
990 * skb_napi_id - Returns the skb's NAPI id
991 * @skb: buffer
992 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000993static inline unsigned int skb_napi_id(const struct sk_buff *skb)
994{
995#ifdef CONFIG_NET_RX_BUSY_POLL
996 return skb->napi_id;
997#else
998 return 0;
999#endif
1000}
1001
David Brazdil0f672f62019-12-10 10:32:29 +00001002/**
1003 * skb_unref - decrement the skb's reference count
1004 * @skb: buffer
1005 *
1006 * Returns true if we can free the skb.
1007 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001008static inline bool skb_unref(struct sk_buff *skb)
1009{
1010 if (unlikely(!skb))
1011 return false;
1012 if (likely(refcount_read(&skb->users) == 1))
1013 smp_rmb();
1014 else if (likely(!refcount_dec_and_test(&skb->users)))
1015 return false;
1016
1017 return true;
1018}
1019
1020void skb_release_head_state(struct sk_buff *skb);
1021void kfree_skb(struct sk_buff *skb);
1022void kfree_skb_list(struct sk_buff *segs);
David Brazdil0f672f62019-12-10 10:32:29 +00001023void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001024void skb_tx_error(struct sk_buff *skb);
1025void consume_skb(struct sk_buff *skb);
1026void __consume_stateless_skb(struct sk_buff *skb);
1027void __kfree_skb(struct sk_buff *skb);
1028extern struct kmem_cache *skbuff_head_cache;
1029
1030void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1031bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1032 bool *fragstolen, int *delta_truesize);
1033
1034struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1035 int node);
1036struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1037struct sk_buff *build_skb(void *data, unsigned int frag_size);
David Brazdil0f672f62019-12-10 10:32:29 +00001038struct sk_buff *build_skb_around(struct sk_buff *skb,
1039 void *data, unsigned int frag_size);
1040
1041/**
1042 * alloc_skb - allocate a network buffer
1043 * @size: size to allocate
1044 * @priority: allocation mask
1045 *
1046 * This function is a convenient wrapper around __alloc_skb().
1047 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001048static inline struct sk_buff *alloc_skb(unsigned int size,
1049 gfp_t priority)
1050{
1051 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1052}
1053
1054struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1055 unsigned long data_len,
1056 int max_page_order,
1057 int *errcode,
1058 gfp_t gfp_mask);
David Brazdil0f672f62019-12-10 10:32:29 +00001059struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001060
1061/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1062struct sk_buff_fclones {
1063 struct sk_buff skb1;
1064
1065 struct sk_buff skb2;
1066
1067 refcount_t fclone_ref;
1068};
1069
1070/**
1071 * skb_fclone_busy - check if fclone is busy
1072 * @sk: socket
1073 * @skb: buffer
1074 *
1075 * Returns true if skb is a fast clone, and its clone is not freed.
1076 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1077 * so we also check that this didnt happen.
1078 */
1079static inline bool skb_fclone_busy(const struct sock *sk,
1080 const struct sk_buff *skb)
1081{
1082 const struct sk_buff_fclones *fclones;
1083
1084 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1085
1086 return skb->fclone == SKB_FCLONE_ORIG &&
1087 refcount_read(&fclones->fclone_ref) > 1 &&
1088 fclones->skb2.sk == sk;
1089}
1090
David Brazdil0f672f62019-12-10 10:32:29 +00001091/**
1092 * alloc_skb_fclone - allocate a network buffer from fclone cache
1093 * @size: size to allocate
1094 * @priority: allocation mask
1095 *
1096 * This function is a convenient wrapper around __alloc_skb().
1097 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001098static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1099 gfp_t priority)
1100{
1101 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1102}
1103
1104struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1105void skb_headers_offset_update(struct sk_buff *skb, int off);
1106int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1107struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1108void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1109struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1110struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1111 gfp_t gfp_mask, bool fclone);
1112static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1113 gfp_t gfp_mask)
1114{
1115 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1116}
1117
1118int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1119struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1120 unsigned int headroom);
1121struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1122 int newtailroom, gfp_t priority);
1123int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1124 int offset, int len);
1125int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1126 int offset, int len);
1127int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1128int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1129
1130/**
1131 * skb_pad - zero pad the tail of an skb
1132 * @skb: buffer to pad
1133 * @pad: space to pad
1134 *
1135 * Ensure that a buffer is followed by a padding area that is zero
1136 * filled. Used by network drivers which may DMA or transfer data
1137 * beyond the buffer end onto the wire.
1138 *
1139 * May return error in out of memory cases. The skb is freed on error.
1140 */
1141static inline int skb_pad(struct sk_buff *skb, int pad)
1142{
1143 return __skb_pad(skb, pad, true);
1144}
1145#define dev_kfree_skb(a) consume_skb(a)
1146
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001147int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1148 int offset, size_t size);
1149
1150struct skb_seq_state {
1151 __u32 lower_offset;
1152 __u32 upper_offset;
1153 __u32 frag_idx;
1154 __u32 stepped_offset;
1155 struct sk_buff *root_skb;
1156 struct sk_buff *cur_skb;
1157 __u8 *frag_data;
1158};
1159
1160void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1161 unsigned int to, struct skb_seq_state *st);
1162unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1163 struct skb_seq_state *st);
1164void skb_abort_seq_read(struct skb_seq_state *st);
1165
1166unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1167 unsigned int to, struct ts_config *config);
1168
1169/*
1170 * Packet hash types specify the type of hash in skb_set_hash.
1171 *
1172 * Hash types refer to the protocol layer addresses which are used to
1173 * construct a packet's hash. The hashes are used to differentiate or identify
1174 * flows of the protocol layer for the hash type. Hash types are either
1175 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1176 *
1177 * Properties of hashes:
1178 *
1179 * 1) Two packets in different flows have different hash values
1180 * 2) Two packets in the same flow should have the same hash value
1181 *
1182 * A hash at a higher layer is considered to be more specific. A driver should
1183 * set the most specific hash possible.
1184 *
1185 * A driver cannot indicate a more specific hash than the layer at which a hash
1186 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1187 *
1188 * A driver may indicate a hash level which is less specific than the
1189 * actual layer the hash was computed on. For instance, a hash computed
1190 * at L4 may be considered an L3 hash. This should only be done if the
1191 * driver can't unambiguously determine that the HW computed the hash at
1192 * the higher layer. Note that the "should" in the second property above
1193 * permits this.
1194 */
1195enum pkt_hash_types {
1196 PKT_HASH_TYPE_NONE, /* Undefined type */
1197 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1198 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1199 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1200};
1201
1202static inline void skb_clear_hash(struct sk_buff *skb)
1203{
1204 skb->hash = 0;
1205 skb->sw_hash = 0;
1206 skb->l4_hash = 0;
1207}
1208
1209static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1210{
1211 if (!skb->l4_hash)
1212 skb_clear_hash(skb);
1213}
1214
1215static inline void
1216__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1217{
1218 skb->l4_hash = is_l4;
1219 skb->sw_hash = is_sw;
1220 skb->hash = hash;
1221}
1222
1223static inline void
1224skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1225{
1226 /* Used by drivers to set hash from HW */
1227 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1228}
1229
1230static inline void
1231__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1232{
1233 __skb_set_hash(skb, hash, true, is_l4);
1234}
1235
1236void __skb_get_hash(struct sk_buff *skb);
1237u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1238u32 skb_get_poff(const struct sk_buff *skb);
1239u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1240 const struct flow_keys_basic *keys, int hlen);
1241__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1242 void *data, int hlen_proto);
1243
1244static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1245 int thoff, u8 ip_proto)
1246{
1247 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1248}
1249
1250void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1251 const struct flow_dissector_key *key,
1252 unsigned int key_count);
1253
David Brazdil0f672f62019-12-10 10:32:29 +00001254#ifdef CONFIG_NET
1255int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1256 union bpf_attr __user *uattr);
1257int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1258 struct bpf_prog *prog);
1259
1260int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1261#else
1262static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1263 union bpf_attr __user *uattr)
1264{
1265 return -EOPNOTSUPP;
1266}
1267
1268static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1269 struct bpf_prog *prog)
1270{
1271 return -EOPNOTSUPP;
1272}
1273
1274static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1275{
1276 return -EOPNOTSUPP;
1277}
1278#endif
1279
1280struct bpf_flow_dissector;
1281bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1282 __be16 proto, int nhoff, int hlen, unsigned int flags);
1283
1284bool __skb_flow_dissect(const struct net *net,
1285 const struct sk_buff *skb,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001286 struct flow_dissector *flow_dissector,
1287 void *target_container,
1288 void *data, __be16 proto, int nhoff, int hlen,
1289 unsigned int flags);
1290
1291static inline bool skb_flow_dissect(const struct sk_buff *skb,
1292 struct flow_dissector *flow_dissector,
1293 void *target_container, unsigned int flags)
1294{
David Brazdil0f672f62019-12-10 10:32:29 +00001295 return __skb_flow_dissect(NULL, skb, flow_dissector,
1296 target_container, NULL, 0, 0, 0, flags);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001297}
1298
1299static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1300 struct flow_keys *flow,
1301 unsigned int flags)
1302{
1303 memset(flow, 0, sizeof(*flow));
David Brazdil0f672f62019-12-10 10:32:29 +00001304 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1305 flow, NULL, 0, 0, 0, flags);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001306}
1307
1308static inline bool
David Brazdil0f672f62019-12-10 10:32:29 +00001309skb_flow_dissect_flow_keys_basic(const struct net *net,
1310 const struct sk_buff *skb,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001311 struct flow_keys_basic *flow, void *data,
1312 __be16 proto, int nhoff, int hlen,
1313 unsigned int flags)
1314{
1315 memset(flow, 0, sizeof(*flow));
David Brazdil0f672f62019-12-10 10:32:29 +00001316 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001317 data, proto, nhoff, hlen, flags);
1318}
1319
David Brazdil0f672f62019-12-10 10:32:29 +00001320void skb_flow_dissect_meta(const struct sk_buff *skb,
1321 struct flow_dissector *flow_dissector,
1322 void *target_container);
1323
1324/* Gets a skb connection tracking info, ctinfo map should be a
1325 * a map of mapsize to translate enum ip_conntrack_info states
1326 * to user states.
1327 */
1328void
1329skb_flow_dissect_ct(const struct sk_buff *skb,
1330 struct flow_dissector *flow_dissector,
1331 void *target_container,
1332 u16 *ctinfo_map,
1333 size_t mapsize);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001334void
1335skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1336 struct flow_dissector *flow_dissector,
1337 void *target_container);
1338
1339static inline __u32 skb_get_hash(struct sk_buff *skb)
1340{
1341 if (!skb->l4_hash && !skb->sw_hash)
1342 __skb_get_hash(skb);
1343
1344 return skb->hash;
1345}
1346
1347static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1348{
1349 if (!skb->l4_hash && !skb->sw_hash) {
1350 struct flow_keys keys;
1351 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1352
1353 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1354 }
1355
1356 return skb->hash;
1357}
1358
David Brazdil0f672f62019-12-10 10:32:29 +00001359__u32 skb_get_hash_perturb(const struct sk_buff *skb,
1360 const siphash_key_t *perturb);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001361
1362static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1363{
1364 return skb->hash;
1365}
1366
1367static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1368{
1369 to->hash = from->hash;
1370 to->sw_hash = from->sw_hash;
1371 to->l4_hash = from->l4_hash;
1372};
1373
David Brazdil0f672f62019-12-10 10:32:29 +00001374static inline void skb_copy_decrypted(struct sk_buff *to,
1375 const struct sk_buff *from)
1376{
1377#ifdef CONFIG_TLS_DEVICE
1378 to->decrypted = from->decrypted;
1379#endif
1380}
1381
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001382#ifdef NET_SKBUFF_DATA_USES_OFFSET
1383static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1384{
1385 return skb->head + skb->end;
1386}
1387
1388static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1389{
1390 return skb->end;
1391}
1392#else
1393static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1394{
1395 return skb->end;
1396}
1397
1398static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1399{
1400 return skb->end - skb->head;
1401}
1402#endif
1403
1404/* Internal */
1405#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1406
1407static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1408{
1409 return &skb_shinfo(skb)->hwtstamps;
1410}
1411
1412static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1413{
1414 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1415
1416 return is_zcopy ? skb_uarg(skb) : NULL;
1417}
1418
David Brazdil0f672f62019-12-10 10:32:29 +00001419static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1420 bool *have_ref)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001421{
1422 if (skb && uarg && !skb_zcopy(skb)) {
David Brazdil0f672f62019-12-10 10:32:29 +00001423 if (unlikely(have_ref && *have_ref))
1424 *have_ref = false;
1425 else
1426 sock_zerocopy_get(uarg);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001427 skb_shinfo(skb)->destructor_arg = uarg;
1428 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1429 }
1430}
1431
1432static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1433{
1434 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1435 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1436}
1437
1438static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1439{
1440 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1441}
1442
1443static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1444{
1445 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1446}
1447
1448/* Release a reference on a zerocopy structure */
1449static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1450{
1451 struct ubuf_info *uarg = skb_zcopy(skb);
1452
1453 if (uarg) {
David Brazdil0f672f62019-12-10 10:32:29 +00001454 if (skb_zcopy_is_nouarg(skb)) {
1455 /* no notification callback */
1456 } else if (uarg->callback == sock_zerocopy_callback) {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001457 uarg->zerocopy = uarg->zerocopy && zerocopy;
1458 sock_zerocopy_put(uarg);
David Brazdil0f672f62019-12-10 10:32:29 +00001459 } else {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001460 uarg->callback(uarg, zerocopy);
1461 }
1462
1463 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1464 }
1465}
1466
1467/* Abort a zerocopy operation and revert zckey on error in send syscall */
1468static inline void skb_zcopy_abort(struct sk_buff *skb)
1469{
1470 struct ubuf_info *uarg = skb_zcopy(skb);
1471
1472 if (uarg) {
David Brazdil0f672f62019-12-10 10:32:29 +00001473 sock_zerocopy_put_abort(uarg, false);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001474 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1475 }
1476}
1477
1478static inline void skb_mark_not_on_list(struct sk_buff *skb)
1479{
1480 skb->next = NULL;
1481}
1482
Olivier Deprez0e641232021-09-23 10:07:05 +02001483/* Iterate through singly-linked GSO fragments of an skb. */
1484#define skb_list_walk_safe(first, skb, next_skb) \
1485 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1486 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1487
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001488static inline void skb_list_del_init(struct sk_buff *skb)
1489{
1490 __list_del_entry(&skb->list);
1491 skb_mark_not_on_list(skb);
1492}
1493
1494/**
1495 * skb_queue_empty - check if a queue is empty
1496 * @list: queue head
1497 *
1498 * Returns true if the queue is empty, false otherwise.
1499 */
1500static inline int skb_queue_empty(const struct sk_buff_head *list)
1501{
1502 return list->next == (const struct sk_buff *) list;
1503}
1504
1505/**
David Brazdil0f672f62019-12-10 10:32:29 +00001506 * skb_queue_empty_lockless - check if a queue is empty
1507 * @list: queue head
1508 *
1509 * Returns true if the queue is empty, false otherwise.
1510 * This variant can be used in lockless contexts.
1511 */
1512static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1513{
1514 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1515}
1516
1517
1518/**
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001519 * skb_queue_is_last - check if skb is the last entry in the queue
1520 * @list: queue head
1521 * @skb: buffer
1522 *
1523 * Returns true if @skb is the last buffer on the list.
1524 */
1525static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1526 const struct sk_buff *skb)
1527{
1528 return skb->next == (const struct sk_buff *) list;
1529}
1530
1531/**
1532 * skb_queue_is_first - check if skb is the first entry in the queue
1533 * @list: queue head
1534 * @skb: buffer
1535 *
1536 * Returns true if @skb is the first buffer on the list.
1537 */
1538static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1539 const struct sk_buff *skb)
1540{
1541 return skb->prev == (const struct sk_buff *) list;
1542}
1543
1544/**
1545 * skb_queue_next - return the next packet in the queue
1546 * @list: queue head
1547 * @skb: current buffer
1548 *
1549 * Return the next packet in @list after @skb. It is only valid to
1550 * call this if skb_queue_is_last() evaluates to false.
1551 */
1552static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1553 const struct sk_buff *skb)
1554{
1555 /* This BUG_ON may seem severe, but if we just return then we
1556 * are going to dereference garbage.
1557 */
1558 BUG_ON(skb_queue_is_last(list, skb));
1559 return skb->next;
1560}
1561
1562/**
1563 * skb_queue_prev - return the prev packet in the queue
1564 * @list: queue head
1565 * @skb: current buffer
1566 *
1567 * Return the prev packet in @list before @skb. It is only valid to
1568 * call this if skb_queue_is_first() evaluates to false.
1569 */
1570static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1571 const struct sk_buff *skb)
1572{
1573 /* This BUG_ON may seem severe, but if we just return then we
1574 * are going to dereference garbage.
1575 */
1576 BUG_ON(skb_queue_is_first(list, skb));
1577 return skb->prev;
1578}
1579
1580/**
1581 * skb_get - reference buffer
1582 * @skb: buffer to reference
1583 *
1584 * Makes another reference to a socket buffer and returns a pointer
1585 * to the buffer.
1586 */
1587static inline struct sk_buff *skb_get(struct sk_buff *skb)
1588{
1589 refcount_inc(&skb->users);
1590 return skb;
1591}
1592
1593/*
1594 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1595 */
1596
1597/**
1598 * skb_cloned - is the buffer a clone
1599 * @skb: buffer to check
1600 *
1601 * Returns true if the buffer was generated with skb_clone() and is
1602 * one of multiple shared copies of the buffer. Cloned buffers are
1603 * shared data so must not be written to under normal circumstances.
1604 */
1605static inline int skb_cloned(const struct sk_buff *skb)
1606{
1607 return skb->cloned &&
1608 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1609}
1610
1611static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1612{
1613 might_sleep_if(gfpflags_allow_blocking(pri));
1614
1615 if (skb_cloned(skb))
1616 return pskb_expand_head(skb, 0, 0, pri);
1617
1618 return 0;
1619}
1620
1621/**
1622 * skb_header_cloned - is the header a clone
1623 * @skb: buffer to check
1624 *
1625 * Returns true if modifying the header part of the buffer requires
1626 * the data to be copied.
1627 */
1628static inline int skb_header_cloned(const struct sk_buff *skb)
1629{
1630 int dataref;
1631
1632 if (!skb->cloned)
1633 return 0;
1634
1635 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1636 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1637 return dataref != 1;
1638}
1639
1640static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1641{
1642 might_sleep_if(gfpflags_allow_blocking(pri));
1643
1644 if (skb_header_cloned(skb))
1645 return pskb_expand_head(skb, 0, 0, pri);
1646
1647 return 0;
1648}
1649
1650/**
1651 * __skb_header_release - release reference to header
1652 * @skb: buffer to operate on
1653 */
1654static inline void __skb_header_release(struct sk_buff *skb)
1655{
1656 skb->nohdr = 1;
1657 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1658}
1659
1660
1661/**
1662 * skb_shared - is the buffer shared
1663 * @skb: buffer to check
1664 *
1665 * Returns true if more than one person has a reference to this
1666 * buffer.
1667 */
1668static inline int skb_shared(const struct sk_buff *skb)
1669{
1670 return refcount_read(&skb->users) != 1;
1671}
1672
1673/**
1674 * skb_share_check - check if buffer is shared and if so clone it
1675 * @skb: buffer to check
1676 * @pri: priority for memory allocation
1677 *
1678 * If the buffer is shared the buffer is cloned and the old copy
1679 * drops a reference. A new clone with a single reference is returned.
1680 * If the buffer is not shared the original buffer is returned. When
1681 * being called from interrupt status or with spinlocks held pri must
1682 * be GFP_ATOMIC.
1683 *
1684 * NULL is returned on a memory allocation failure.
1685 */
1686static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1687{
1688 might_sleep_if(gfpflags_allow_blocking(pri));
1689 if (skb_shared(skb)) {
1690 struct sk_buff *nskb = skb_clone(skb, pri);
1691
1692 if (likely(nskb))
1693 consume_skb(skb);
1694 else
1695 kfree_skb(skb);
1696 skb = nskb;
1697 }
1698 return skb;
1699}
1700
1701/*
1702 * Copy shared buffers into a new sk_buff. We effectively do COW on
1703 * packets to handle cases where we have a local reader and forward
1704 * and a couple of other messy ones. The normal one is tcpdumping
1705 * a packet thats being forwarded.
1706 */
1707
1708/**
1709 * skb_unshare - make a copy of a shared buffer
1710 * @skb: buffer to check
1711 * @pri: priority for memory allocation
1712 *
1713 * If the socket buffer is a clone then this function creates a new
1714 * copy of the data, drops a reference count on the old copy and returns
1715 * the new copy with the reference count at 1. If the buffer is not a clone
1716 * the original buffer is returned. When called with a spinlock held or
1717 * from interrupt state @pri must be %GFP_ATOMIC
1718 *
1719 * %NULL is returned on a memory allocation failure.
1720 */
1721static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1722 gfp_t pri)
1723{
1724 might_sleep_if(gfpflags_allow_blocking(pri));
1725 if (skb_cloned(skb)) {
1726 struct sk_buff *nskb = skb_copy(skb, pri);
1727
1728 /* Free our shared copy */
1729 if (likely(nskb))
1730 consume_skb(skb);
1731 else
1732 kfree_skb(skb);
1733 skb = nskb;
1734 }
1735 return skb;
1736}
1737
1738/**
1739 * skb_peek - peek at the head of an &sk_buff_head
1740 * @list_: list to peek at
1741 *
1742 * Peek an &sk_buff. Unlike most other operations you _MUST_
1743 * be careful with this one. A peek leaves the buffer on the
1744 * list and someone else may run off with it. You must hold
1745 * the appropriate locks or have a private queue to do this.
1746 *
1747 * Returns %NULL for an empty list or a pointer to the head element.
1748 * The reference count is not incremented and the reference is therefore
1749 * volatile. Use with caution.
1750 */
1751static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1752{
1753 struct sk_buff *skb = list_->next;
1754
1755 if (skb == (struct sk_buff *)list_)
1756 skb = NULL;
1757 return skb;
1758}
1759
1760/**
David Brazdil0f672f62019-12-10 10:32:29 +00001761 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1762 * @list_: list to peek at
1763 *
1764 * Like skb_peek(), but the caller knows that the list is not empty.
1765 */
1766static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1767{
1768 return list_->next;
1769}
1770
1771/**
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001772 * skb_peek_next - peek skb following the given one from a queue
1773 * @skb: skb to start from
1774 * @list_: list to peek at
1775 *
1776 * Returns %NULL when the end of the list is met or a pointer to the
1777 * next element. The reference count is not incremented and the
1778 * reference is therefore volatile. Use with caution.
1779 */
1780static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1781 const struct sk_buff_head *list_)
1782{
1783 struct sk_buff *next = skb->next;
1784
1785 if (next == (struct sk_buff *)list_)
1786 next = NULL;
1787 return next;
1788}
1789
1790/**
1791 * skb_peek_tail - peek at the tail of an &sk_buff_head
1792 * @list_: list to peek at
1793 *
1794 * Peek an &sk_buff. Unlike most other operations you _MUST_
1795 * be careful with this one. A peek leaves the buffer on the
1796 * list and someone else may run off with it. You must hold
1797 * the appropriate locks or have a private queue to do this.
1798 *
1799 * Returns %NULL for an empty list or a pointer to the tail element.
1800 * The reference count is not incremented and the reference is therefore
1801 * volatile. Use with caution.
1802 */
1803static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1804{
Olivier Deprez0e641232021-09-23 10:07:05 +02001805 struct sk_buff *skb = READ_ONCE(list_->prev);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001806
1807 if (skb == (struct sk_buff *)list_)
1808 skb = NULL;
1809 return skb;
1810
1811}
1812
1813/**
1814 * skb_queue_len - get queue length
1815 * @list_: list to measure
1816 *
1817 * Return the length of an &sk_buff queue.
1818 */
1819static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1820{
1821 return list_->qlen;
1822}
1823
1824/**
Olivier Deprez0e641232021-09-23 10:07:05 +02001825 * skb_queue_len_lockless - get queue length
1826 * @list_: list to measure
1827 *
1828 * Return the length of an &sk_buff queue.
1829 * This variant can be used in lockless contexts.
1830 */
1831static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
1832{
1833 return READ_ONCE(list_->qlen);
1834}
1835
1836/**
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001837 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1838 * @list: queue to initialize
1839 *
1840 * This initializes only the list and queue length aspects of
1841 * an sk_buff_head object. This allows to initialize the list
1842 * aspects of an sk_buff_head without reinitializing things like
1843 * the spinlock. It can also be used for on-stack sk_buff_head
1844 * objects where the spinlock is known to not be used.
1845 */
1846static inline void __skb_queue_head_init(struct sk_buff_head *list)
1847{
1848 list->prev = list->next = (struct sk_buff *)list;
1849 list->qlen = 0;
1850}
1851
1852/*
1853 * This function creates a split out lock class for each invocation;
1854 * this is needed for now since a whole lot of users of the skb-queue
1855 * infrastructure in drivers have different locking usage (in hardirq)
1856 * than the networking core (in softirq only). In the long run either the
1857 * network layer or drivers should need annotation to consolidate the
1858 * main types of usage into 3 classes.
1859 */
1860static inline void skb_queue_head_init(struct sk_buff_head *list)
1861{
1862 spin_lock_init(&list->lock);
1863 __skb_queue_head_init(list);
1864}
1865
1866static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1867 struct lock_class_key *class)
1868{
1869 skb_queue_head_init(list);
1870 lockdep_set_class(&list->lock, class);
1871}
1872
1873/*
1874 * Insert an sk_buff on a list.
1875 *
1876 * The "__skb_xxxx()" functions are the non-atomic ones that
1877 * can only be called with interrupts disabled.
1878 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001879static inline void __skb_insert(struct sk_buff *newsk,
1880 struct sk_buff *prev, struct sk_buff *next,
1881 struct sk_buff_head *list)
1882{
Olivier Deprez0e641232021-09-23 10:07:05 +02001883 /* See skb_queue_empty_lockless() and skb_peek_tail()
1884 * for the opposite READ_ONCE()
1885 */
David Brazdil0f672f62019-12-10 10:32:29 +00001886 WRITE_ONCE(newsk->next, next);
1887 WRITE_ONCE(newsk->prev, prev);
1888 WRITE_ONCE(next->prev, newsk);
1889 WRITE_ONCE(prev->next, newsk);
Olivier Deprez0e641232021-09-23 10:07:05 +02001890 WRITE_ONCE(list->qlen, list->qlen + 1);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001891}
1892
1893static inline void __skb_queue_splice(const struct sk_buff_head *list,
1894 struct sk_buff *prev,
1895 struct sk_buff *next)
1896{
1897 struct sk_buff *first = list->next;
1898 struct sk_buff *last = list->prev;
1899
David Brazdil0f672f62019-12-10 10:32:29 +00001900 WRITE_ONCE(first->prev, prev);
1901 WRITE_ONCE(prev->next, first);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001902
David Brazdil0f672f62019-12-10 10:32:29 +00001903 WRITE_ONCE(last->next, next);
1904 WRITE_ONCE(next->prev, last);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001905}
1906
1907/**
1908 * skb_queue_splice - join two skb lists, this is designed for stacks
1909 * @list: the new list to add
1910 * @head: the place to add it in the first list
1911 */
1912static inline void skb_queue_splice(const struct sk_buff_head *list,
1913 struct sk_buff_head *head)
1914{
1915 if (!skb_queue_empty(list)) {
1916 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1917 head->qlen += list->qlen;
1918 }
1919}
1920
1921/**
1922 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1923 * @list: the new list to add
1924 * @head: the place to add it in the first list
1925 *
1926 * The list at @list is reinitialised
1927 */
1928static inline void skb_queue_splice_init(struct sk_buff_head *list,
1929 struct sk_buff_head *head)
1930{
1931 if (!skb_queue_empty(list)) {
1932 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1933 head->qlen += list->qlen;
1934 __skb_queue_head_init(list);
1935 }
1936}
1937
1938/**
1939 * skb_queue_splice_tail - join two skb lists, each list being a queue
1940 * @list: the new list to add
1941 * @head: the place to add it in the first list
1942 */
1943static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1944 struct sk_buff_head *head)
1945{
1946 if (!skb_queue_empty(list)) {
1947 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1948 head->qlen += list->qlen;
1949 }
1950}
1951
1952/**
1953 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1954 * @list: the new list to add
1955 * @head: the place to add it in the first list
1956 *
1957 * Each of the lists is a queue.
1958 * The list at @list is reinitialised
1959 */
1960static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1961 struct sk_buff_head *head)
1962{
1963 if (!skb_queue_empty(list)) {
1964 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1965 head->qlen += list->qlen;
1966 __skb_queue_head_init(list);
1967 }
1968}
1969
1970/**
1971 * __skb_queue_after - queue a buffer at the list head
1972 * @list: list to use
1973 * @prev: place after this buffer
1974 * @newsk: buffer to queue
1975 *
1976 * Queue a buffer int the middle of a list. This function takes no locks
1977 * and you must therefore hold required locks before calling it.
1978 *
1979 * A buffer cannot be placed on two lists at the same time.
1980 */
1981static inline void __skb_queue_after(struct sk_buff_head *list,
1982 struct sk_buff *prev,
1983 struct sk_buff *newsk)
1984{
1985 __skb_insert(newsk, prev, prev->next, list);
1986}
1987
1988void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1989 struct sk_buff_head *list);
1990
1991static inline void __skb_queue_before(struct sk_buff_head *list,
1992 struct sk_buff *next,
1993 struct sk_buff *newsk)
1994{
1995 __skb_insert(newsk, next->prev, next, list);
1996}
1997
1998/**
1999 * __skb_queue_head - queue a buffer at the list head
2000 * @list: list to use
2001 * @newsk: buffer to queue
2002 *
2003 * Queue a buffer at the start of a list. This function takes no locks
2004 * and you must therefore hold required locks before calling it.
2005 *
2006 * A buffer cannot be placed on two lists at the same time.
2007 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002008static inline void __skb_queue_head(struct sk_buff_head *list,
2009 struct sk_buff *newsk)
2010{
2011 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2012}
David Brazdil0f672f62019-12-10 10:32:29 +00002013void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002014
2015/**
2016 * __skb_queue_tail - queue a buffer at the list tail
2017 * @list: list to use
2018 * @newsk: buffer to queue
2019 *
2020 * Queue a buffer at the end of a list. This function takes no locks
2021 * and you must therefore hold required locks before calling it.
2022 *
2023 * A buffer cannot be placed on two lists at the same time.
2024 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002025static inline void __skb_queue_tail(struct sk_buff_head *list,
2026 struct sk_buff *newsk)
2027{
2028 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2029}
David Brazdil0f672f62019-12-10 10:32:29 +00002030void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002031
2032/*
2033 * remove sk_buff from list. _Must_ be called atomically, and with
2034 * the list known..
2035 */
2036void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2037static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2038{
2039 struct sk_buff *next, *prev;
2040
Olivier Deprez0e641232021-09-23 10:07:05 +02002041 WRITE_ONCE(list->qlen, list->qlen - 1);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002042 next = skb->next;
2043 prev = skb->prev;
2044 skb->next = skb->prev = NULL;
David Brazdil0f672f62019-12-10 10:32:29 +00002045 WRITE_ONCE(next->prev, prev);
2046 WRITE_ONCE(prev->next, next);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002047}
2048
2049/**
2050 * __skb_dequeue - remove from the head of the queue
2051 * @list: list to dequeue from
2052 *
2053 * Remove the head of the list. This function does not take any locks
2054 * so must be used with appropriate locks held only. The head item is
2055 * returned or %NULL if the list is empty.
2056 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002057static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2058{
2059 struct sk_buff *skb = skb_peek(list);
2060 if (skb)
2061 __skb_unlink(skb, list);
2062 return skb;
2063}
David Brazdil0f672f62019-12-10 10:32:29 +00002064struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002065
2066/**
2067 * __skb_dequeue_tail - remove from the tail of the queue
2068 * @list: list to dequeue from
2069 *
2070 * Remove the tail of the list. This function does not take any locks
2071 * so must be used with appropriate locks held only. The tail item is
2072 * returned or %NULL if the list is empty.
2073 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002074static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2075{
2076 struct sk_buff *skb = skb_peek_tail(list);
2077 if (skb)
2078 __skb_unlink(skb, list);
2079 return skb;
2080}
David Brazdil0f672f62019-12-10 10:32:29 +00002081struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002082
2083
2084static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2085{
2086 return skb->data_len;
2087}
2088
2089static inline unsigned int skb_headlen(const struct sk_buff *skb)
2090{
2091 return skb->len - skb->data_len;
2092}
2093
2094static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2095{
2096 unsigned int i, len = 0;
2097
2098 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2099 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2100 return len;
2101}
2102
2103static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2104{
2105 return skb_headlen(skb) + __skb_pagelen(skb);
2106}
2107
2108/**
2109 * __skb_fill_page_desc - initialise a paged fragment in an skb
2110 * @skb: buffer containing fragment to be initialised
2111 * @i: paged fragment index to initialise
2112 * @page: the page to use for this fragment
2113 * @off: the offset to the data with @page
2114 * @size: the length of the data
2115 *
2116 * Initialises the @i'th fragment of @skb to point to &size bytes at
2117 * offset @off within @page.
2118 *
2119 * Does not take any additional reference on the fragment.
2120 */
2121static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2122 struct page *page, int off, int size)
2123{
2124 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2125
2126 /*
2127 * Propagate page pfmemalloc to the skb if we can. The problem is
2128 * that not all callers have unique ownership of the page but rely
2129 * on page_is_pfmemalloc doing the right thing(tm).
2130 */
David Brazdil0f672f62019-12-10 10:32:29 +00002131 frag->bv_page = page;
2132 frag->bv_offset = off;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002133 skb_frag_size_set(frag, size);
2134
2135 page = compound_head(page);
2136 if (page_is_pfmemalloc(page))
2137 skb->pfmemalloc = true;
2138}
2139
2140/**
2141 * skb_fill_page_desc - initialise a paged fragment in an skb
2142 * @skb: buffer containing fragment to be initialised
2143 * @i: paged fragment index to initialise
2144 * @page: the page to use for this fragment
2145 * @off: the offset to the data with @page
2146 * @size: the length of the data
2147 *
2148 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2149 * @skb to point to @size bytes at offset @off within @page. In
2150 * addition updates @skb such that @i is the last fragment.
2151 *
2152 * Does not take any additional reference on the fragment.
2153 */
2154static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2155 struct page *page, int off, int size)
2156{
2157 __skb_fill_page_desc(skb, i, page, off, size);
2158 skb_shinfo(skb)->nr_frags = i + 1;
2159}
2160
2161void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2162 int size, unsigned int truesize);
2163
2164void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2165 unsigned int truesize);
2166
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002167#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2168
2169#ifdef NET_SKBUFF_DATA_USES_OFFSET
2170static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2171{
2172 return skb->head + skb->tail;
2173}
2174
2175static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2176{
2177 skb->tail = skb->data - skb->head;
2178}
2179
2180static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2181{
2182 skb_reset_tail_pointer(skb);
2183 skb->tail += offset;
2184}
2185
2186#else /* NET_SKBUFF_DATA_USES_OFFSET */
2187static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2188{
2189 return skb->tail;
2190}
2191
2192static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2193{
2194 skb->tail = skb->data;
2195}
2196
2197static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2198{
2199 skb->tail = skb->data + offset;
2200}
2201
2202#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2203
2204/*
2205 * Add data to an sk_buff
2206 */
2207void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2208void *skb_put(struct sk_buff *skb, unsigned int len);
2209static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2210{
2211 void *tmp = skb_tail_pointer(skb);
2212 SKB_LINEAR_ASSERT(skb);
2213 skb->tail += len;
2214 skb->len += len;
2215 return tmp;
2216}
2217
2218static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2219{
2220 void *tmp = __skb_put(skb, len);
2221
2222 memset(tmp, 0, len);
2223 return tmp;
2224}
2225
2226static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2227 unsigned int len)
2228{
2229 void *tmp = __skb_put(skb, len);
2230
2231 memcpy(tmp, data, len);
2232 return tmp;
2233}
2234
2235static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2236{
2237 *(u8 *)__skb_put(skb, 1) = val;
2238}
2239
2240static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2241{
2242 void *tmp = skb_put(skb, len);
2243
2244 memset(tmp, 0, len);
2245
2246 return tmp;
2247}
2248
2249static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2250 unsigned int len)
2251{
2252 void *tmp = skb_put(skb, len);
2253
2254 memcpy(tmp, data, len);
2255
2256 return tmp;
2257}
2258
2259static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2260{
2261 *(u8 *)skb_put(skb, 1) = val;
2262}
2263
2264void *skb_push(struct sk_buff *skb, unsigned int len);
2265static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2266{
2267 skb->data -= len;
2268 skb->len += len;
2269 return skb->data;
2270}
2271
2272void *skb_pull(struct sk_buff *skb, unsigned int len);
2273static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2274{
2275 skb->len -= len;
2276 BUG_ON(skb->len < skb->data_len);
2277 return skb->data += len;
2278}
2279
2280static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2281{
2282 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2283}
2284
2285void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2286
2287static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2288{
2289 if (len > skb_headlen(skb) &&
2290 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2291 return NULL;
2292 skb->len -= len;
2293 return skb->data += len;
2294}
2295
2296static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2297{
2298 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2299}
2300
2301static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2302{
2303 if (likely(len <= skb_headlen(skb)))
2304 return 1;
2305 if (unlikely(len > skb->len))
2306 return 0;
2307 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2308}
2309
2310void skb_condense(struct sk_buff *skb);
2311
2312/**
2313 * skb_headroom - bytes at buffer head
2314 * @skb: buffer to check
2315 *
2316 * Return the number of bytes of free space at the head of an &sk_buff.
2317 */
2318static inline unsigned int skb_headroom(const struct sk_buff *skb)
2319{
2320 return skb->data - skb->head;
2321}
2322
2323/**
2324 * skb_tailroom - bytes at buffer end
2325 * @skb: buffer to check
2326 *
2327 * Return the number of bytes of free space at the tail of an sk_buff
2328 */
2329static inline int skb_tailroom(const struct sk_buff *skb)
2330{
2331 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2332}
2333
2334/**
2335 * skb_availroom - bytes at buffer end
2336 * @skb: buffer to check
2337 *
2338 * Return the number of bytes of free space at the tail of an sk_buff
2339 * allocated by sk_stream_alloc()
2340 */
2341static inline int skb_availroom(const struct sk_buff *skb)
2342{
2343 if (skb_is_nonlinear(skb))
2344 return 0;
2345
2346 return skb->end - skb->tail - skb->reserved_tailroom;
2347}
2348
2349/**
2350 * skb_reserve - adjust headroom
2351 * @skb: buffer to alter
2352 * @len: bytes to move
2353 *
2354 * Increase the headroom of an empty &sk_buff by reducing the tail
2355 * room. This is only allowed for an empty buffer.
2356 */
2357static inline void skb_reserve(struct sk_buff *skb, int len)
2358{
2359 skb->data += len;
2360 skb->tail += len;
2361}
2362
2363/**
2364 * skb_tailroom_reserve - adjust reserved_tailroom
2365 * @skb: buffer to alter
2366 * @mtu: maximum amount of headlen permitted
2367 * @needed_tailroom: minimum amount of reserved_tailroom
2368 *
2369 * Set reserved_tailroom so that headlen can be as large as possible but
2370 * not larger than mtu and tailroom cannot be smaller than
2371 * needed_tailroom.
2372 * The required headroom should already have been reserved before using
2373 * this function.
2374 */
2375static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2376 unsigned int needed_tailroom)
2377{
2378 SKB_LINEAR_ASSERT(skb);
2379 if (mtu < skb_tailroom(skb) - needed_tailroom)
2380 /* use at most mtu */
2381 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2382 else
2383 /* use up to all available space */
2384 skb->reserved_tailroom = needed_tailroom;
2385}
2386
2387#define ENCAP_TYPE_ETHER 0
2388#define ENCAP_TYPE_IPPROTO 1
2389
2390static inline void skb_set_inner_protocol(struct sk_buff *skb,
2391 __be16 protocol)
2392{
2393 skb->inner_protocol = protocol;
2394 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2395}
2396
2397static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2398 __u8 ipproto)
2399{
2400 skb->inner_ipproto = ipproto;
2401 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2402}
2403
2404static inline void skb_reset_inner_headers(struct sk_buff *skb)
2405{
2406 skb->inner_mac_header = skb->mac_header;
2407 skb->inner_network_header = skb->network_header;
2408 skb->inner_transport_header = skb->transport_header;
2409}
2410
2411static inline void skb_reset_mac_len(struct sk_buff *skb)
2412{
2413 skb->mac_len = skb->network_header - skb->mac_header;
2414}
2415
2416static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2417 *skb)
2418{
2419 return skb->head + skb->inner_transport_header;
2420}
2421
2422static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2423{
2424 return skb_inner_transport_header(skb) - skb->data;
2425}
2426
2427static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2428{
2429 skb->inner_transport_header = skb->data - skb->head;
2430}
2431
2432static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2433 const int offset)
2434{
2435 skb_reset_inner_transport_header(skb);
2436 skb->inner_transport_header += offset;
2437}
2438
2439static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2440{
2441 return skb->head + skb->inner_network_header;
2442}
2443
2444static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2445{
2446 skb->inner_network_header = skb->data - skb->head;
2447}
2448
2449static inline void skb_set_inner_network_header(struct sk_buff *skb,
2450 const int offset)
2451{
2452 skb_reset_inner_network_header(skb);
2453 skb->inner_network_header += offset;
2454}
2455
2456static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2457{
2458 return skb->head + skb->inner_mac_header;
2459}
2460
2461static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2462{
2463 skb->inner_mac_header = skb->data - skb->head;
2464}
2465
2466static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2467 const int offset)
2468{
2469 skb_reset_inner_mac_header(skb);
2470 skb->inner_mac_header += offset;
2471}
2472static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2473{
2474 return skb->transport_header != (typeof(skb->transport_header))~0U;
2475}
2476
2477static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2478{
2479 return skb->head + skb->transport_header;
2480}
2481
2482static inline void skb_reset_transport_header(struct sk_buff *skb)
2483{
2484 skb->transport_header = skb->data - skb->head;
2485}
2486
2487static inline void skb_set_transport_header(struct sk_buff *skb,
2488 const int offset)
2489{
2490 skb_reset_transport_header(skb);
2491 skb->transport_header += offset;
2492}
2493
2494static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2495{
2496 return skb->head + skb->network_header;
2497}
2498
2499static inline void skb_reset_network_header(struct sk_buff *skb)
2500{
2501 skb->network_header = skb->data - skb->head;
2502}
2503
2504static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2505{
2506 skb_reset_network_header(skb);
2507 skb->network_header += offset;
2508}
2509
2510static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2511{
2512 return skb->head + skb->mac_header;
2513}
2514
2515static inline int skb_mac_offset(const struct sk_buff *skb)
2516{
2517 return skb_mac_header(skb) - skb->data;
2518}
2519
2520static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2521{
2522 return skb->network_header - skb->mac_header;
2523}
2524
2525static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2526{
2527 return skb->mac_header != (typeof(skb->mac_header))~0U;
2528}
2529
2530static inline void skb_reset_mac_header(struct sk_buff *skb)
2531{
2532 skb->mac_header = skb->data - skb->head;
2533}
2534
2535static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2536{
2537 skb_reset_mac_header(skb);
2538 skb->mac_header += offset;
2539}
2540
2541static inline void skb_pop_mac_header(struct sk_buff *skb)
2542{
2543 skb->mac_header = skb->network_header;
2544}
2545
David Brazdil0f672f62019-12-10 10:32:29 +00002546static inline void skb_probe_transport_header(struct sk_buff *skb)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002547{
2548 struct flow_keys_basic keys;
2549
2550 if (skb_transport_header_was_set(skb))
2551 return;
2552
David Brazdil0f672f62019-12-10 10:32:29 +00002553 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2554 NULL, 0, 0, 0, 0))
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002555 skb_set_transport_header(skb, keys.control.thoff);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002556}
2557
2558static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2559{
2560 if (skb_mac_header_was_set(skb)) {
2561 const unsigned char *old_mac = skb_mac_header(skb);
2562
2563 skb_set_mac_header(skb, -skb->mac_len);
2564 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2565 }
2566}
2567
2568static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2569{
2570 return skb->csum_start - skb_headroom(skb);
2571}
2572
2573static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2574{
2575 return skb->head + skb->csum_start;
2576}
2577
2578static inline int skb_transport_offset(const struct sk_buff *skb)
2579{
2580 return skb_transport_header(skb) - skb->data;
2581}
2582
2583static inline u32 skb_network_header_len(const struct sk_buff *skb)
2584{
2585 return skb->transport_header - skb->network_header;
2586}
2587
2588static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2589{
2590 return skb->inner_transport_header - skb->inner_network_header;
2591}
2592
2593static inline int skb_network_offset(const struct sk_buff *skb)
2594{
2595 return skb_network_header(skb) - skb->data;
2596}
2597
2598static inline int skb_inner_network_offset(const struct sk_buff *skb)
2599{
2600 return skb_inner_network_header(skb) - skb->data;
2601}
2602
2603static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2604{
2605 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2606}
2607
2608/*
2609 * CPUs often take a performance hit when accessing unaligned memory
2610 * locations. The actual performance hit varies, it can be small if the
2611 * hardware handles it or large if we have to take an exception and fix it
2612 * in software.
2613 *
2614 * Since an ethernet header is 14 bytes network drivers often end up with
2615 * the IP header at an unaligned offset. The IP header can be aligned by
2616 * shifting the start of the packet by 2 bytes. Drivers should do this
2617 * with:
2618 *
2619 * skb_reserve(skb, NET_IP_ALIGN);
2620 *
2621 * The downside to this alignment of the IP header is that the DMA is now
2622 * unaligned. On some architectures the cost of an unaligned DMA is high
2623 * and this cost outweighs the gains made by aligning the IP header.
2624 *
2625 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2626 * to be overridden.
2627 */
2628#ifndef NET_IP_ALIGN
2629#define NET_IP_ALIGN 2
2630#endif
2631
2632/*
2633 * The networking layer reserves some headroom in skb data (via
2634 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2635 * the header has to grow. In the default case, if the header has to grow
2636 * 32 bytes or less we avoid the reallocation.
2637 *
2638 * Unfortunately this headroom changes the DMA alignment of the resulting
2639 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2640 * on some architectures. An architecture can override this value,
2641 * perhaps setting it to a cacheline in size (since that will maintain
2642 * cacheline alignment of the DMA). It must be a power of 2.
2643 *
2644 * Various parts of the networking layer expect at least 32 bytes of
2645 * headroom, you should not reduce this.
2646 *
2647 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2648 * to reduce average number of cache lines per packet.
2649 * get_rps_cpus() for example only access one 64 bytes aligned block :
2650 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2651 */
2652#ifndef NET_SKB_PAD
2653#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2654#endif
2655
2656int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2657
2658static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2659{
David Brazdil0f672f62019-12-10 10:32:29 +00002660 if (WARN_ON(skb_is_nonlinear(skb)))
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002661 return;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002662 skb->len = len;
2663 skb_set_tail_pointer(skb, len);
2664}
2665
2666static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2667{
2668 __skb_set_length(skb, len);
2669}
2670
2671void skb_trim(struct sk_buff *skb, unsigned int len);
2672
2673static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2674{
2675 if (skb->data_len)
2676 return ___pskb_trim(skb, len);
2677 __skb_trim(skb, len);
2678 return 0;
2679}
2680
2681static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2682{
2683 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2684}
2685
2686/**
2687 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2688 * @skb: buffer to alter
2689 * @len: new length
2690 *
2691 * This is identical to pskb_trim except that the caller knows that
2692 * the skb is not cloned so we should never get an error due to out-
2693 * of-memory.
2694 */
2695static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2696{
2697 int err = pskb_trim(skb, len);
2698 BUG_ON(err);
2699}
2700
2701static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2702{
2703 unsigned int diff = len - skb->len;
2704
2705 if (skb_tailroom(skb) < diff) {
2706 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2707 GFP_ATOMIC);
2708 if (ret)
2709 return ret;
2710 }
2711 __skb_set_length(skb, len);
2712 return 0;
2713}
2714
2715/**
2716 * skb_orphan - orphan a buffer
2717 * @skb: buffer to orphan
2718 *
2719 * If a buffer currently has an owner then we call the owner's
2720 * destructor function and make the @skb unowned. The buffer continues
2721 * to exist but is no longer charged to its former owner.
2722 */
2723static inline void skb_orphan(struct sk_buff *skb)
2724{
2725 if (skb->destructor) {
2726 skb->destructor(skb);
2727 skb->destructor = NULL;
2728 skb->sk = NULL;
2729 } else {
2730 BUG_ON(skb->sk);
2731 }
2732}
2733
2734/**
2735 * skb_orphan_frags - orphan the frags contained in a buffer
2736 * @skb: buffer to orphan frags from
2737 * @gfp_mask: allocation mask for replacement pages
2738 *
2739 * For each frag in the SKB which needs a destructor (i.e. has an
2740 * owner) create a copy of that frag and release the original
2741 * page by calling the destructor.
2742 */
2743static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2744{
2745 if (likely(!skb_zcopy(skb)))
2746 return 0;
David Brazdil0f672f62019-12-10 10:32:29 +00002747 if (!skb_zcopy_is_nouarg(skb) &&
2748 skb_uarg(skb)->callback == sock_zerocopy_callback)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002749 return 0;
2750 return skb_copy_ubufs(skb, gfp_mask);
2751}
2752
2753/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2754static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2755{
2756 if (likely(!skb_zcopy(skb)))
2757 return 0;
2758 return skb_copy_ubufs(skb, gfp_mask);
2759}
2760
2761/**
2762 * __skb_queue_purge - empty a list
2763 * @list: list to empty
2764 *
2765 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2766 * the list and one reference dropped. This function does not take the
2767 * list lock and the caller must hold the relevant locks to use it.
2768 */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002769static inline void __skb_queue_purge(struct sk_buff_head *list)
2770{
2771 struct sk_buff *skb;
2772 while ((skb = __skb_dequeue(list)) != NULL)
2773 kfree_skb(skb);
2774}
David Brazdil0f672f62019-12-10 10:32:29 +00002775void skb_queue_purge(struct sk_buff_head *list);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002776
2777unsigned int skb_rbtree_purge(struct rb_root *root);
2778
2779void *netdev_alloc_frag(unsigned int fragsz);
2780
2781struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2782 gfp_t gfp_mask);
2783
2784/**
2785 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2786 * @dev: network device to receive on
2787 * @length: length to allocate
2788 *
2789 * Allocate a new &sk_buff and assign it a usage count of one. The
2790 * buffer has unspecified headroom built in. Users should allocate
2791 * the headroom they think they need without accounting for the
2792 * built in space. The built in space is used for optimisations.
2793 *
2794 * %NULL is returned if there is no free memory. Although this function
2795 * allocates memory it can be called from an interrupt.
2796 */
2797static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2798 unsigned int length)
2799{
2800 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2801}
2802
2803/* legacy helper around __netdev_alloc_skb() */
2804static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2805 gfp_t gfp_mask)
2806{
2807 return __netdev_alloc_skb(NULL, length, gfp_mask);
2808}
2809
2810/* legacy helper around netdev_alloc_skb() */
2811static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2812{
2813 return netdev_alloc_skb(NULL, length);
2814}
2815
2816
2817static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2818 unsigned int length, gfp_t gfp)
2819{
2820 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2821
2822 if (NET_IP_ALIGN && skb)
2823 skb_reserve(skb, NET_IP_ALIGN);
2824 return skb;
2825}
2826
2827static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2828 unsigned int length)
2829{
2830 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2831}
2832
2833static inline void skb_free_frag(void *addr)
2834{
2835 page_frag_free(addr);
2836}
2837
2838void *napi_alloc_frag(unsigned int fragsz);
2839struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2840 unsigned int length, gfp_t gfp_mask);
2841static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2842 unsigned int length)
2843{
2844 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2845}
2846void napi_consume_skb(struct sk_buff *skb, int budget);
2847
2848void __kfree_skb_flush(void);
2849void __kfree_skb_defer(struct sk_buff *skb);
2850
2851/**
2852 * __dev_alloc_pages - allocate page for network Rx
2853 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2854 * @order: size of the allocation
2855 *
2856 * Allocate a new page.
2857 *
2858 * %NULL is returned if there is no free memory.
2859*/
2860static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2861 unsigned int order)
2862{
2863 /* This piece of code contains several assumptions.
2864 * 1. This is for device Rx, therefor a cold page is preferred.
2865 * 2. The expectation is the user wants a compound page.
2866 * 3. If requesting a order 0 page it will not be compound
2867 * due to the check to see if order has a value in prep_new_page
2868 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2869 * code in gfp_to_alloc_flags that should be enforcing this.
2870 */
2871 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2872
2873 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2874}
2875
2876static inline struct page *dev_alloc_pages(unsigned int order)
2877{
2878 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2879}
2880
2881/**
2882 * __dev_alloc_page - allocate a page for network Rx
2883 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2884 *
2885 * Allocate a new page.
2886 *
2887 * %NULL is returned if there is no free memory.
2888 */
2889static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2890{
2891 return __dev_alloc_pages(gfp_mask, 0);
2892}
2893
2894static inline struct page *dev_alloc_page(void)
2895{
2896 return dev_alloc_pages(0);
2897}
2898
2899/**
2900 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2901 * @page: The page that was allocated from skb_alloc_page
2902 * @skb: The skb that may need pfmemalloc set
2903 */
2904static inline void skb_propagate_pfmemalloc(struct page *page,
2905 struct sk_buff *skb)
2906{
2907 if (page_is_pfmemalloc(page))
2908 skb->pfmemalloc = true;
2909}
2910
2911/**
David Brazdil0f672f62019-12-10 10:32:29 +00002912 * skb_frag_off() - Returns the offset of a skb fragment
2913 * @frag: the paged fragment
2914 */
2915static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2916{
2917 return frag->bv_offset;
2918}
2919
2920/**
2921 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
2922 * @frag: skb fragment
2923 * @delta: value to add
2924 */
2925static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
2926{
2927 frag->bv_offset += delta;
2928}
2929
2930/**
2931 * skb_frag_off_set() - Sets the offset of a skb fragment
2932 * @frag: skb fragment
2933 * @offset: offset of fragment
2934 */
2935static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
2936{
2937 frag->bv_offset = offset;
2938}
2939
2940/**
2941 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
2942 * @fragto: skb fragment where offset is set
2943 * @fragfrom: skb fragment offset is copied from
2944 */
2945static inline void skb_frag_off_copy(skb_frag_t *fragto,
2946 const skb_frag_t *fragfrom)
2947{
2948 fragto->bv_offset = fragfrom->bv_offset;
2949}
2950
2951/**
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002952 * skb_frag_page - retrieve the page referred to by a paged fragment
2953 * @frag: the paged fragment
2954 *
2955 * Returns the &struct page associated with @frag.
2956 */
2957static inline struct page *skb_frag_page(const skb_frag_t *frag)
2958{
David Brazdil0f672f62019-12-10 10:32:29 +00002959 return frag->bv_page;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002960}
2961
2962/**
2963 * __skb_frag_ref - take an addition reference on a paged fragment.
2964 * @frag: the paged fragment
2965 *
2966 * Takes an additional reference on the paged fragment @frag.
2967 */
2968static inline void __skb_frag_ref(skb_frag_t *frag)
2969{
2970 get_page(skb_frag_page(frag));
2971}
2972
2973/**
2974 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2975 * @skb: the buffer
2976 * @f: the fragment offset.
2977 *
2978 * Takes an additional reference on the @f'th paged fragment of @skb.
2979 */
2980static inline void skb_frag_ref(struct sk_buff *skb, int f)
2981{
2982 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2983}
2984
2985/**
2986 * __skb_frag_unref - release a reference on a paged fragment.
2987 * @frag: the paged fragment
2988 *
2989 * Releases a reference on the paged fragment @frag.
2990 */
2991static inline void __skb_frag_unref(skb_frag_t *frag)
2992{
2993 put_page(skb_frag_page(frag));
2994}
2995
2996/**
2997 * skb_frag_unref - release a reference on a paged fragment of an skb.
2998 * @skb: the buffer
2999 * @f: the fragment offset
3000 *
3001 * Releases a reference on the @f'th paged fragment of @skb.
3002 */
3003static inline void skb_frag_unref(struct sk_buff *skb, int f)
3004{
3005 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
3006}
3007
3008/**
3009 * skb_frag_address - gets the address of the data contained in a paged fragment
3010 * @frag: the paged fragment buffer
3011 *
3012 * Returns the address of the data within @frag. The page must already
3013 * be mapped.
3014 */
3015static inline void *skb_frag_address(const skb_frag_t *frag)
3016{
David Brazdil0f672f62019-12-10 10:32:29 +00003017 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003018}
3019
3020/**
3021 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3022 * @frag: the paged fragment buffer
3023 *
3024 * Returns the address of the data within @frag. Checks that the page
3025 * is mapped and returns %NULL otherwise.
3026 */
3027static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3028{
3029 void *ptr = page_address(skb_frag_page(frag));
3030 if (unlikely(!ptr))
3031 return NULL;
3032
David Brazdil0f672f62019-12-10 10:32:29 +00003033 return ptr + skb_frag_off(frag);
3034}
3035
3036/**
3037 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3038 * @fragto: skb fragment where page is set
3039 * @fragfrom: skb fragment page is copied from
3040 */
3041static inline void skb_frag_page_copy(skb_frag_t *fragto,
3042 const skb_frag_t *fragfrom)
3043{
3044 fragto->bv_page = fragfrom->bv_page;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003045}
3046
3047/**
3048 * __skb_frag_set_page - sets the page contained in a paged fragment
3049 * @frag: the paged fragment
3050 * @page: the page to set
3051 *
3052 * Sets the fragment @frag to contain @page.
3053 */
3054static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3055{
David Brazdil0f672f62019-12-10 10:32:29 +00003056 frag->bv_page = page;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003057}
3058
3059/**
3060 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3061 * @skb: the buffer
3062 * @f: the fragment offset
3063 * @page: the page to set
3064 *
3065 * Sets the @f'th fragment of @skb to contain @page.
3066 */
3067static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3068 struct page *page)
3069{
3070 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3071}
3072
3073bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3074
3075/**
3076 * skb_frag_dma_map - maps a paged fragment via the DMA API
3077 * @dev: the device to map the fragment to
3078 * @frag: the paged fragment to map
3079 * @offset: the offset within the fragment (starting at the
3080 * fragment's own offset)
3081 * @size: the number of bytes to map
3082 * @dir: the direction of the mapping (``PCI_DMA_*``)
3083 *
3084 * Maps the page associated with @frag to @device.
3085 */
3086static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3087 const skb_frag_t *frag,
3088 size_t offset, size_t size,
3089 enum dma_data_direction dir)
3090{
3091 return dma_map_page(dev, skb_frag_page(frag),
David Brazdil0f672f62019-12-10 10:32:29 +00003092 skb_frag_off(frag) + offset, size, dir);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003093}
3094
3095static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3096 gfp_t gfp_mask)
3097{
3098 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3099}
3100
3101
3102static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3103 gfp_t gfp_mask)
3104{
3105 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3106}
3107
3108
3109/**
3110 * skb_clone_writable - is the header of a clone writable
3111 * @skb: buffer to check
3112 * @len: length up to which to write
3113 *
3114 * Returns true if modifying the header part of the cloned buffer
3115 * does not requires the data to be copied.
3116 */
3117static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3118{
3119 return !skb_header_cloned(skb) &&
3120 skb_headroom(skb) + len <= skb->hdr_len;
3121}
3122
3123static inline int skb_try_make_writable(struct sk_buff *skb,
3124 unsigned int write_len)
3125{
3126 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3127 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3128}
3129
3130static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3131 int cloned)
3132{
3133 int delta = 0;
3134
3135 if (headroom > skb_headroom(skb))
3136 delta = headroom - skb_headroom(skb);
3137
3138 if (delta || cloned)
3139 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3140 GFP_ATOMIC);
3141 return 0;
3142}
3143
3144/**
3145 * skb_cow - copy header of skb when it is required
3146 * @skb: buffer to cow
3147 * @headroom: needed headroom
3148 *
3149 * If the skb passed lacks sufficient headroom or its data part
3150 * is shared, data is reallocated. If reallocation fails, an error
3151 * is returned and original skb is not changed.
3152 *
3153 * The result is skb with writable area skb->head...skb->tail
3154 * and at least @headroom of space at head.
3155 */
3156static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3157{
3158 return __skb_cow(skb, headroom, skb_cloned(skb));
3159}
3160
3161/**
3162 * skb_cow_head - skb_cow but only making the head writable
3163 * @skb: buffer to cow
3164 * @headroom: needed headroom
3165 *
3166 * This function is identical to skb_cow except that we replace the
3167 * skb_cloned check by skb_header_cloned. It should be used when
3168 * you only need to push on some header and do not need to modify
3169 * the data.
3170 */
3171static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3172{
3173 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3174}
3175
3176/**
3177 * skb_padto - pad an skbuff up to a minimal size
3178 * @skb: buffer to pad
3179 * @len: minimal length
3180 *
3181 * Pads up a buffer to ensure the trailing bytes exist and are
3182 * blanked. If the buffer already contains sufficient data it
3183 * is untouched. Otherwise it is extended. Returns zero on
3184 * success. The skb is freed on error.
3185 */
3186static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3187{
3188 unsigned int size = skb->len;
3189 if (likely(size >= len))
3190 return 0;
3191 return skb_pad(skb, len - size);
3192}
3193
3194/**
David Brazdil0f672f62019-12-10 10:32:29 +00003195 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003196 * @skb: buffer to pad
3197 * @len: minimal length
3198 * @free_on_error: free buffer on error
3199 *
3200 * Pads up a buffer to ensure the trailing bytes exist and are
3201 * blanked. If the buffer already contains sufficient data it
3202 * is untouched. Otherwise it is extended. Returns zero on
3203 * success. The skb is freed on error if @free_on_error is true.
3204 */
Olivier Deprez0e641232021-09-23 10:07:05 +02003205static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3206 unsigned int len,
3207 bool free_on_error)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003208{
3209 unsigned int size = skb->len;
3210
3211 if (unlikely(size < len)) {
3212 len -= size;
3213 if (__skb_pad(skb, len, free_on_error))
3214 return -ENOMEM;
3215 __skb_put(skb, len);
3216 }
3217 return 0;
3218}
3219
3220/**
3221 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3222 * @skb: buffer to pad
3223 * @len: minimal length
3224 *
3225 * Pads up a buffer to ensure the trailing bytes exist and are
3226 * blanked. If the buffer already contains sufficient data it
3227 * is untouched. Otherwise it is extended. Returns zero on
3228 * success. The skb is freed on error.
3229 */
Olivier Deprez0e641232021-09-23 10:07:05 +02003230static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003231{
3232 return __skb_put_padto(skb, len, true);
3233}
3234
3235static inline int skb_add_data(struct sk_buff *skb,
3236 struct iov_iter *from, int copy)
3237{
3238 const int off = skb->len;
3239
3240 if (skb->ip_summed == CHECKSUM_NONE) {
3241 __wsum csum = 0;
3242 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3243 &csum, from)) {
3244 skb->csum = csum_block_add(skb->csum, csum, off);
3245 return 0;
3246 }
3247 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3248 return 0;
3249
3250 __skb_trim(skb, off);
3251 return -EFAULT;
3252}
3253
3254static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3255 const struct page *page, int off)
3256{
3257 if (skb_zcopy(skb))
3258 return false;
3259 if (i) {
David Brazdil0f672f62019-12-10 10:32:29 +00003260 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003261
3262 return page == skb_frag_page(frag) &&
David Brazdil0f672f62019-12-10 10:32:29 +00003263 off == skb_frag_off(frag) + skb_frag_size(frag);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003264 }
3265 return false;
3266}
3267
3268static inline int __skb_linearize(struct sk_buff *skb)
3269{
3270 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3271}
3272
3273/**
3274 * skb_linearize - convert paged skb to linear one
3275 * @skb: buffer to linarize
3276 *
3277 * If there is no free memory -ENOMEM is returned, otherwise zero
3278 * is returned and the old skb data released.
3279 */
3280static inline int skb_linearize(struct sk_buff *skb)
3281{
3282 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3283}
3284
3285/**
3286 * skb_has_shared_frag - can any frag be overwritten
3287 * @skb: buffer to test
3288 *
3289 * Return true if the skb has at least one frag that might be modified
3290 * by an external entity (as in vmsplice()/sendfile())
3291 */
3292static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3293{
3294 return skb_is_nonlinear(skb) &&
3295 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3296}
3297
3298/**
3299 * skb_linearize_cow - make sure skb is linear and writable
3300 * @skb: buffer to process
3301 *
3302 * If there is no free memory -ENOMEM is returned, otherwise zero
3303 * is returned and the old skb data released.
3304 */
3305static inline int skb_linearize_cow(struct sk_buff *skb)
3306{
3307 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3308 __skb_linearize(skb) : 0;
3309}
3310
3311static __always_inline void
3312__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3313 unsigned int off)
3314{
3315 if (skb->ip_summed == CHECKSUM_COMPLETE)
3316 skb->csum = csum_block_sub(skb->csum,
3317 csum_partial(start, len, 0), off);
3318 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3319 skb_checksum_start_offset(skb) < 0)
3320 skb->ip_summed = CHECKSUM_NONE;
3321}
3322
3323/**
3324 * skb_postpull_rcsum - update checksum for received skb after pull
3325 * @skb: buffer to update
3326 * @start: start of data before pull
3327 * @len: length of data pulled
3328 *
3329 * After doing a pull on a received packet, you need to call this to
3330 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3331 * CHECKSUM_NONE so that it can be recomputed from scratch.
3332 */
3333static inline void skb_postpull_rcsum(struct sk_buff *skb,
3334 const void *start, unsigned int len)
3335{
3336 __skb_postpull_rcsum(skb, start, len, 0);
3337}
3338
3339static __always_inline void
3340__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3341 unsigned int off)
3342{
3343 if (skb->ip_summed == CHECKSUM_COMPLETE)
3344 skb->csum = csum_block_add(skb->csum,
3345 csum_partial(start, len, 0), off);
3346}
3347
3348/**
3349 * skb_postpush_rcsum - update checksum for received skb after push
3350 * @skb: buffer to update
3351 * @start: start of data after push
3352 * @len: length of data pushed
3353 *
3354 * After doing a push on a received packet, you need to call this to
3355 * update the CHECKSUM_COMPLETE checksum.
3356 */
3357static inline void skb_postpush_rcsum(struct sk_buff *skb,
3358 const void *start, unsigned int len)
3359{
3360 __skb_postpush_rcsum(skb, start, len, 0);
3361}
3362
3363void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3364
3365/**
3366 * skb_push_rcsum - push skb and update receive checksum
3367 * @skb: buffer to update
3368 * @len: length of data pulled
3369 *
3370 * This function performs an skb_push on the packet and updates
3371 * the CHECKSUM_COMPLETE checksum. It should be used on
3372 * receive path processing instead of skb_push unless you know
3373 * that the checksum difference is zero (e.g., a valid IP header)
3374 * or you are setting ip_summed to CHECKSUM_NONE.
3375 */
3376static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3377{
3378 skb_push(skb, len);
3379 skb_postpush_rcsum(skb, skb->data, len);
3380 return skb->data;
3381}
3382
3383int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3384/**
3385 * pskb_trim_rcsum - trim received skb and update checksum
3386 * @skb: buffer to trim
3387 * @len: new length
3388 *
3389 * This is exactly the same as pskb_trim except that it ensures the
3390 * checksum of received packets are still valid after the operation.
David Brazdil0f672f62019-12-10 10:32:29 +00003391 * It can change skb pointers.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003392 */
3393
3394static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3395{
3396 if (likely(len >= skb->len))
3397 return 0;
3398 return pskb_trim_rcsum_slow(skb, len);
3399}
3400
3401static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3402{
3403 if (skb->ip_summed == CHECKSUM_COMPLETE)
3404 skb->ip_summed = CHECKSUM_NONE;
3405 __skb_trim(skb, len);
3406 return 0;
3407}
3408
3409static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3410{
3411 if (skb->ip_summed == CHECKSUM_COMPLETE)
3412 skb->ip_summed = CHECKSUM_NONE;
3413 return __skb_grow(skb, len);
3414}
3415
3416#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3417#define skb_rb_first(root) rb_to_skb(rb_first(root))
3418#define skb_rb_last(root) rb_to_skb(rb_last(root))
3419#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3420#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3421
3422#define skb_queue_walk(queue, skb) \
3423 for (skb = (queue)->next; \
3424 skb != (struct sk_buff *)(queue); \
3425 skb = skb->next)
3426
3427#define skb_queue_walk_safe(queue, skb, tmp) \
3428 for (skb = (queue)->next, tmp = skb->next; \
3429 skb != (struct sk_buff *)(queue); \
3430 skb = tmp, tmp = skb->next)
3431
3432#define skb_queue_walk_from(queue, skb) \
3433 for (; skb != (struct sk_buff *)(queue); \
3434 skb = skb->next)
3435
3436#define skb_rbtree_walk(skb, root) \
3437 for (skb = skb_rb_first(root); skb != NULL; \
3438 skb = skb_rb_next(skb))
3439
3440#define skb_rbtree_walk_from(skb) \
3441 for (; skb != NULL; \
3442 skb = skb_rb_next(skb))
3443
3444#define skb_rbtree_walk_from_safe(skb, tmp) \
3445 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3446 skb = tmp)
3447
3448#define skb_queue_walk_from_safe(queue, skb, tmp) \
3449 for (tmp = skb->next; \
3450 skb != (struct sk_buff *)(queue); \
3451 skb = tmp, tmp = skb->next)
3452
3453#define skb_queue_reverse_walk(queue, skb) \
3454 for (skb = (queue)->prev; \
3455 skb != (struct sk_buff *)(queue); \
3456 skb = skb->prev)
3457
3458#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3459 for (skb = (queue)->prev, tmp = skb->prev; \
3460 skb != (struct sk_buff *)(queue); \
3461 skb = tmp, tmp = skb->prev)
3462
3463#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3464 for (tmp = skb->prev; \
3465 skb != (struct sk_buff *)(queue); \
3466 skb = tmp, tmp = skb->prev)
3467
3468static inline bool skb_has_frag_list(const struct sk_buff *skb)
3469{
3470 return skb_shinfo(skb)->frag_list != NULL;
3471}
3472
3473static inline void skb_frag_list_init(struct sk_buff *skb)
3474{
3475 skb_shinfo(skb)->frag_list = NULL;
3476}
3477
3478#define skb_walk_frags(skb, iter) \
3479 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3480
3481
3482int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3483 const struct sk_buff *skb);
3484struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3485 struct sk_buff_head *queue,
3486 unsigned int flags,
3487 void (*destructor)(struct sock *sk,
3488 struct sk_buff *skb),
David Brazdil0f672f62019-12-10 10:32:29 +00003489 int *off, int *err,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003490 struct sk_buff **last);
3491struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3492 void (*destructor)(struct sock *sk,
3493 struct sk_buff *skb),
David Brazdil0f672f62019-12-10 10:32:29 +00003494 int *off, int *err,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003495 struct sk_buff **last);
3496struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3497 void (*destructor)(struct sock *sk,
3498 struct sk_buff *skb),
David Brazdil0f672f62019-12-10 10:32:29 +00003499 int *off, int *err);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003500struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3501 int *err);
3502__poll_t datagram_poll(struct file *file, struct socket *sock,
3503 struct poll_table_struct *wait);
3504int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3505 struct iov_iter *to, int size);
3506static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3507 struct msghdr *msg, int size)
3508{
3509 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3510}
3511int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3512 struct msghdr *msg);
David Brazdil0f672f62019-12-10 10:32:29 +00003513int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3514 struct iov_iter *to, int len,
3515 struct ahash_request *hash);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003516int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3517 struct iov_iter *from, int len);
3518int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3519void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3520void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3521static inline void skb_free_datagram_locked(struct sock *sk,
3522 struct sk_buff *skb)
3523{
3524 __skb_free_datagram_locked(sk, skb, 0);
3525}
3526int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3527int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3528int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3529__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3530 int len, __wsum csum);
3531int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3532 struct pipe_inode_info *pipe, unsigned int len,
3533 unsigned int flags);
3534int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3535 int len);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003536void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3537unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3538int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3539 int len, int hlen);
3540void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3541int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3542void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3543bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3544bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3545struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3546struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3547int skb_ensure_writable(struct sk_buff *skb, int write_len);
3548int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3549int skb_vlan_pop(struct sk_buff *skb);
3550int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
David Brazdil0f672f62019-12-10 10:32:29 +00003551int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
Olivier Deprez0e641232021-09-23 10:07:05 +02003552 int mac_len, bool ethernet);
3553int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3554 bool ethernet);
David Brazdil0f672f62019-12-10 10:32:29 +00003555int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3556int skb_mpls_dec_ttl(struct sk_buff *skb);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003557struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3558 gfp_t gfp);
3559
3560static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3561{
3562 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3563}
3564
3565static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3566{
3567 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3568}
3569
3570struct skb_checksum_ops {
3571 __wsum (*update)(const void *mem, int len, __wsum wsum);
3572 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3573};
3574
3575extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3576
3577__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3578 __wsum csum, const struct skb_checksum_ops *ops);
3579__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3580 __wsum csum);
3581
3582static inline void * __must_check
3583__skb_header_pointer(const struct sk_buff *skb, int offset,
3584 int len, void *data, int hlen, void *buffer)
3585{
3586 if (hlen - offset >= len)
3587 return data + offset;
3588
3589 if (!skb ||
3590 skb_copy_bits(skb, offset, buffer, len) < 0)
3591 return NULL;
3592
3593 return buffer;
3594}
3595
3596static inline void * __must_check
3597skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3598{
3599 return __skb_header_pointer(skb, offset, len, skb->data,
3600 skb_headlen(skb), buffer);
3601}
3602
3603/**
3604 * skb_needs_linearize - check if we need to linearize a given skb
3605 * depending on the given device features.
3606 * @skb: socket buffer to check
3607 * @features: net device features
3608 *
3609 * Returns true if either:
3610 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3611 * 2. skb is fragmented and the device does not support SG.
3612 */
3613static inline bool skb_needs_linearize(struct sk_buff *skb,
3614 netdev_features_t features)
3615{
3616 return skb_is_nonlinear(skb) &&
3617 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3618 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3619}
3620
3621static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3622 void *to,
3623 const unsigned int len)
3624{
3625 memcpy(to, skb->data, len);
3626}
3627
3628static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3629 const int offset, void *to,
3630 const unsigned int len)
3631{
3632 memcpy(to, skb->data + offset, len);
3633}
3634
3635static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3636 const void *from,
3637 const unsigned int len)
3638{
3639 memcpy(skb->data, from, len);
3640}
3641
3642static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3643 const int offset,
3644 const void *from,
3645 const unsigned int len)
3646{
3647 memcpy(skb->data + offset, from, len);
3648}
3649
3650void skb_init(void);
3651
3652static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3653{
3654 return skb->tstamp;
3655}
3656
3657/**
3658 * skb_get_timestamp - get timestamp from a skb
3659 * @skb: skb to get stamp from
David Brazdil0f672f62019-12-10 10:32:29 +00003660 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003661 *
3662 * Timestamps are stored in the skb as offsets to a base timestamp.
3663 * This function converts the offset back to a struct timeval and stores
3664 * it in stamp.
3665 */
3666static inline void skb_get_timestamp(const struct sk_buff *skb,
David Brazdil0f672f62019-12-10 10:32:29 +00003667 struct __kernel_old_timeval *stamp)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003668{
David Brazdil0f672f62019-12-10 10:32:29 +00003669 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3670}
3671
3672static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3673 struct __kernel_sock_timeval *stamp)
3674{
3675 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3676
3677 stamp->tv_sec = ts.tv_sec;
3678 stamp->tv_usec = ts.tv_nsec / 1000;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003679}
3680
3681static inline void skb_get_timestampns(const struct sk_buff *skb,
3682 struct timespec *stamp)
3683{
3684 *stamp = ktime_to_timespec(skb->tstamp);
3685}
3686
David Brazdil0f672f62019-12-10 10:32:29 +00003687static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3688 struct __kernel_timespec *stamp)
3689{
3690 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3691
3692 stamp->tv_sec = ts.tv_sec;
3693 stamp->tv_nsec = ts.tv_nsec;
3694}
3695
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003696static inline void __net_timestamp(struct sk_buff *skb)
3697{
3698 skb->tstamp = ktime_get_real();
3699}
3700
3701static inline ktime_t net_timedelta(ktime_t t)
3702{
3703 return ktime_sub(ktime_get_real(), t);
3704}
3705
3706static inline ktime_t net_invalid_timestamp(void)
3707{
3708 return 0;
3709}
3710
3711static inline u8 skb_metadata_len(const struct sk_buff *skb)
3712{
3713 return skb_shinfo(skb)->meta_len;
3714}
3715
3716static inline void *skb_metadata_end(const struct sk_buff *skb)
3717{
3718 return skb_mac_header(skb);
3719}
3720
3721static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3722 const struct sk_buff *skb_b,
3723 u8 meta_len)
3724{
3725 const void *a = skb_metadata_end(skb_a);
3726 const void *b = skb_metadata_end(skb_b);
3727 /* Using more efficient varaiant than plain call to memcmp(). */
3728#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3729 u64 diffs = 0;
3730
3731 switch (meta_len) {
3732#define __it(x, op) (x -= sizeof(u##op))
3733#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3734 case 32: diffs |= __it_diff(a, b, 64);
David Brazdil0f672f62019-12-10 10:32:29 +00003735 /* fall through */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003736 case 24: diffs |= __it_diff(a, b, 64);
David Brazdil0f672f62019-12-10 10:32:29 +00003737 /* fall through */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003738 case 16: diffs |= __it_diff(a, b, 64);
David Brazdil0f672f62019-12-10 10:32:29 +00003739 /* fall through */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003740 case 8: diffs |= __it_diff(a, b, 64);
3741 break;
3742 case 28: diffs |= __it_diff(a, b, 64);
David Brazdil0f672f62019-12-10 10:32:29 +00003743 /* fall through */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003744 case 20: diffs |= __it_diff(a, b, 64);
David Brazdil0f672f62019-12-10 10:32:29 +00003745 /* fall through */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003746 case 12: diffs |= __it_diff(a, b, 64);
David Brazdil0f672f62019-12-10 10:32:29 +00003747 /* fall through */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00003748 case 4: diffs |= __it_diff(a, b, 32);
3749 break;
3750 }
3751 return diffs;
3752#else
3753 return memcmp(a - meta_len, b - meta_len, meta_len);
3754#endif
3755}
3756
3757static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3758 const struct sk_buff *skb_b)
3759{
3760 u8 len_a = skb_metadata_len(skb_a);
3761 u8 len_b = skb_metadata_len(skb_b);
3762
3763 if (!(len_a | len_b))
3764 return false;
3765
3766 return len_a != len_b ?
3767 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3768}
3769
3770static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3771{
3772 skb_shinfo(skb)->meta_len = meta_len;
3773}
3774
3775static inline void skb_metadata_clear(struct sk_buff *skb)
3776{
3777 skb_metadata_set(skb, 0);
3778}
3779
3780struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3781
3782#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3783
3784void skb_clone_tx_timestamp(struct sk_buff *skb);
3785bool skb_defer_rx_timestamp(struct sk_buff *skb);
3786
3787#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3788
3789static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3790{
3791}
3792
3793static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3794{
3795 return false;
3796}
3797
3798#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3799
3800/**
3801 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3802 *
3803 * PHY drivers may accept clones of transmitted packets for
3804 * timestamping via their phy_driver.txtstamp method. These drivers
3805 * must call this function to return the skb back to the stack with a
3806 * timestamp.
3807 *
3808 * @skb: clone of the the original outgoing packet
3809 * @hwtstamps: hardware time stamps
3810 *
3811 */
3812void skb_complete_tx_timestamp(struct sk_buff *skb,
3813 struct skb_shared_hwtstamps *hwtstamps);
3814
3815void __skb_tstamp_tx(struct sk_buff *orig_skb,
3816 struct skb_shared_hwtstamps *hwtstamps,
3817 struct sock *sk, int tstype);
3818
3819/**
3820 * skb_tstamp_tx - queue clone of skb with send time stamps
3821 * @orig_skb: the original outgoing packet
3822 * @hwtstamps: hardware time stamps, may be NULL if not available
3823 *
3824 * If the skb has a socket associated, then this function clones the
3825 * skb (thus sharing the actual data and optional structures), stores
3826 * the optional hardware time stamping information (if non NULL) or
3827 * generates a software time stamp (otherwise), then queues the clone
3828 * to the error queue of the socket. Errors are silently ignored.
3829 */
3830void skb_tstamp_tx(struct sk_buff *orig_skb,
3831 struct skb_shared_hwtstamps *hwtstamps);
3832
3833/**
3834 * skb_tx_timestamp() - Driver hook for transmit timestamping
3835 *
3836 * Ethernet MAC Drivers should call this function in their hard_xmit()
3837 * function immediately before giving the sk_buff to the MAC hardware.
3838 *
3839 * Specifically, one should make absolutely sure that this function is
3840 * called before TX completion of this packet can trigger. Otherwise
3841 * the packet could potentially already be freed.
3842 *
3843 * @skb: A socket buffer.
3844 */
3845static inline void skb_tx_timestamp(struct sk_buff *skb)
3846{
3847 skb_clone_tx_timestamp(skb);
3848 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3849 skb_tstamp_tx(skb, NULL);
3850}
3851
3852/**
3853 * skb_complete_wifi_ack - deliver skb with wifi status
3854 *
3855 * @skb: the original outgoing packet
3856 * @acked: ack status
3857 *
3858 */
3859void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3860
3861__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3862__sum16 __skb_checksum_complete(struct sk_buff *skb);
3863
3864static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3865{
3866 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3867 skb->csum_valid ||
3868 (skb->ip_summed == CHECKSUM_PARTIAL &&
3869 skb_checksum_start_offset(skb) >= 0));
3870}
3871
3872/**
3873 * skb_checksum_complete - Calculate checksum of an entire packet
3874 * @skb: packet to process
3875 *
3876 * This function calculates the checksum over the entire packet plus
3877 * the value of skb->csum. The latter can be used to supply the
3878 * checksum of a pseudo header as used by TCP/UDP. It returns the
3879 * checksum.
3880 *
3881 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3882 * this function can be used to verify that checksum on received
3883 * packets. In that case the function should return zero if the
3884 * checksum is correct. In particular, this function will return zero
3885 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3886 * hardware has already verified the correctness of the checksum.
3887 */
3888static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3889{
3890 return skb_csum_unnecessary(skb) ?
3891 0 : __skb_checksum_complete(skb);
3892}
3893
3894static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3895{
3896 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3897 if (skb->csum_level == 0)
3898 skb->ip_summed = CHECKSUM_NONE;
3899 else
3900 skb->csum_level--;
3901 }
3902}
3903
3904static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3905{
3906 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3907 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3908 skb->csum_level++;
3909 } else if (skb->ip_summed == CHECKSUM_NONE) {
3910 skb->ip_summed = CHECKSUM_UNNECESSARY;
3911 skb->csum_level = 0;
3912 }
3913}
3914
3915/* Check if we need to perform checksum complete validation.
3916 *
3917 * Returns true if checksum complete is needed, false otherwise
3918 * (either checksum is unnecessary or zero checksum is allowed).
3919 */
3920static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3921 bool zero_okay,
3922 __sum16 check)
3923{
3924 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3925 skb->csum_valid = 1;
3926 __skb_decr_checksum_unnecessary(skb);
3927 return false;
3928 }
3929
3930 return true;
3931}
3932
3933/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3934 * in checksum_init.
3935 */
3936#define CHECKSUM_BREAK 76
3937
3938/* Unset checksum-complete
3939 *
3940 * Unset checksum complete can be done when packet is being modified
3941 * (uncompressed for instance) and checksum-complete value is
3942 * invalidated.
3943 */
3944static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3945{
3946 if (skb->ip_summed == CHECKSUM_COMPLETE)
3947 skb->ip_summed = CHECKSUM_NONE;
3948}
3949
3950/* Validate (init) checksum based on checksum complete.
3951 *
3952 * Return values:
3953 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3954 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3955 * checksum is stored in skb->csum for use in __skb_checksum_complete
3956 * non-zero: value of invalid checksum
3957 *
3958 */
3959static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3960 bool complete,
3961 __wsum psum)
3962{
3963 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3964 if (!csum_fold(csum_add(psum, skb->csum))) {
3965 skb->csum_valid = 1;
3966 return 0;
3967 }
3968 }
3969
3970 skb->csum = psum;
3971
3972 if (complete || skb->len <= CHECKSUM_BREAK) {
3973 __sum16 csum;
3974
3975 csum = __skb_checksum_complete(skb);
3976 skb->csum_valid = !csum;
3977 return csum;
3978 }
3979
3980 return 0;
3981}
3982
3983static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3984{
3985 return 0;
3986}
3987
3988/* Perform checksum validate (init). Note that this is a macro since we only
3989 * want to calculate the pseudo header which is an input function if necessary.
3990 * First we try to validate without any computation (checksum unnecessary) and
3991 * then calculate based on checksum complete calling the function to compute
3992 * pseudo header.
3993 *
3994 * Return values:
3995 * 0: checksum is validated or try to in skb_checksum_complete
3996 * non-zero: value of invalid checksum
3997 */
3998#define __skb_checksum_validate(skb, proto, complete, \
3999 zero_okay, check, compute_pseudo) \
4000({ \
4001 __sum16 __ret = 0; \
4002 skb->csum_valid = 0; \
4003 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4004 __ret = __skb_checksum_validate_complete(skb, \
4005 complete, compute_pseudo(skb, proto)); \
4006 __ret; \
4007})
4008
4009#define skb_checksum_init(skb, proto, compute_pseudo) \
4010 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4011
4012#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4013 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4014
4015#define skb_checksum_validate(skb, proto, compute_pseudo) \
4016 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4017
4018#define skb_checksum_validate_zero_check(skb, proto, check, \
4019 compute_pseudo) \
4020 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4021
4022#define skb_checksum_simple_validate(skb) \
4023 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4024
4025static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4026{
4027 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4028}
4029
David Brazdil0f672f62019-12-10 10:32:29 +00004030static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004031{
4032 skb->csum = ~pseudo;
4033 skb->ip_summed = CHECKSUM_COMPLETE;
4034}
4035
David Brazdil0f672f62019-12-10 10:32:29 +00004036#define skb_checksum_try_convert(skb, proto, compute_pseudo) \
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004037do { \
4038 if (__skb_checksum_convert_check(skb)) \
David Brazdil0f672f62019-12-10 10:32:29 +00004039 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004040} while (0)
4041
4042static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4043 u16 start, u16 offset)
4044{
4045 skb->ip_summed = CHECKSUM_PARTIAL;
4046 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4047 skb->csum_offset = offset - start;
4048}
4049
4050/* Update skbuf and packet to reflect the remote checksum offload operation.
4051 * When called, ptr indicates the starting point for skb->csum when
4052 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4053 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4054 */
4055static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4056 int start, int offset, bool nopartial)
4057{
4058 __wsum delta;
4059
4060 if (!nopartial) {
4061 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4062 return;
4063 }
4064
4065 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4066 __skb_checksum_complete(skb);
4067 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4068 }
4069
4070 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4071
4072 /* Adjust skb->csum since we changed the packet */
4073 skb->csum = csum_add(skb->csum, delta);
4074}
4075
4076static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4077{
4078#if IS_ENABLED(CONFIG_NF_CONNTRACK)
David Brazdil0f672f62019-12-10 10:32:29 +00004079 return (void *)(skb->_nfct & NFCT_PTRMASK);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004080#else
4081 return NULL;
4082#endif
4083}
4084
David Brazdil0f672f62019-12-10 10:32:29 +00004085static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004086{
David Brazdil0f672f62019-12-10 10:32:29 +00004087#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4088 return skb->_nfct;
4089#else
4090 return 0UL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004091#endif
David Brazdil0f672f62019-12-10 10:32:29 +00004092}
4093
4094static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4095{
4096#if IS_ENABLED(CONFIG_NF_CONNTRACK)
4097 skb->_nfct = nfct;
4098#endif
4099}
4100
4101#ifdef CONFIG_SKB_EXTENSIONS
4102enum skb_ext_id {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004103#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
David Brazdil0f672f62019-12-10 10:32:29 +00004104 SKB_EXT_BRIDGE_NF,
4105#endif
4106#ifdef CONFIG_XFRM
4107 SKB_EXT_SEC_PATH,
4108#endif
4109#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4110 TC_SKB_EXT,
4111#endif
4112 SKB_EXT_NUM, /* must be last */
4113};
4114
4115/**
4116 * struct skb_ext - sk_buff extensions
4117 * @refcnt: 1 on allocation, deallocated on 0
4118 * @offset: offset to add to @data to obtain extension address
4119 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4120 * @data: start of extension data, variable sized
4121 *
4122 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4123 * to use 'u8' types while allowing up to 2kb worth of extension data.
4124 */
4125struct skb_ext {
4126 refcount_t refcnt;
4127 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4128 u8 chunks; /* same */
4129 char data[0] __aligned(8);
4130};
4131
4132void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4133void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4134void __skb_ext_put(struct skb_ext *ext);
4135
4136static inline void skb_ext_put(struct sk_buff *skb)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004137{
David Brazdil0f672f62019-12-10 10:32:29 +00004138 if (skb->active_extensions)
4139 __skb_ext_put(skb->extensions);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004140}
David Brazdil0f672f62019-12-10 10:32:29 +00004141
4142static inline void __skb_ext_copy(struct sk_buff *dst,
4143 const struct sk_buff *src)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004144{
David Brazdil0f672f62019-12-10 10:32:29 +00004145 dst->active_extensions = src->active_extensions;
4146
4147 if (src->active_extensions) {
4148 struct skb_ext *ext = src->extensions;
4149
4150 refcount_inc(&ext->refcnt);
4151 dst->extensions = ext;
4152 }
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004153}
David Brazdil0f672f62019-12-10 10:32:29 +00004154
4155static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4156{
4157 skb_ext_put(dst);
4158 __skb_ext_copy(dst, src);
4159}
4160
4161static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4162{
4163 return !!ext->offset[i];
4164}
4165
4166static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4167{
4168 return skb->active_extensions & (1 << id);
4169}
4170
4171static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4172{
4173 if (skb_ext_exist(skb, id))
4174 __skb_ext_del(skb, id);
4175}
4176
4177static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4178{
4179 if (skb_ext_exist(skb, id)) {
4180 struct skb_ext *ext = skb->extensions;
4181
4182 return (void *)ext + (ext->offset[id] << 3);
4183 }
4184
4185 return NULL;
4186}
4187
4188static inline void skb_ext_reset(struct sk_buff *skb)
4189{
4190 if (unlikely(skb->active_extensions)) {
4191 __skb_ext_put(skb->extensions);
4192 skb->active_extensions = 0;
4193 }
4194}
4195
4196static inline bool skb_has_extensions(struct sk_buff *skb)
4197{
4198 return unlikely(skb->active_extensions);
4199}
4200#else
4201static inline void skb_ext_put(struct sk_buff *skb) {}
4202static inline void skb_ext_reset(struct sk_buff *skb) {}
4203static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4204static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4205static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4206static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4207#endif /* CONFIG_SKB_EXTENSIONS */
4208
4209static inline void nf_reset_ct(struct sk_buff *skb)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004210{
4211#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4212 nf_conntrack_put(skb_nfct(skb));
4213 skb->_nfct = 0;
4214#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004215}
4216
4217static inline void nf_reset_trace(struct sk_buff *skb)
4218{
4219#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4220 skb->nf_trace = 0;
4221#endif
4222}
4223
4224static inline void ipvs_reset(struct sk_buff *skb)
4225{
4226#if IS_ENABLED(CONFIG_IP_VS)
4227 skb->ipvs_property = 0;
4228#endif
4229}
4230
David Brazdil0f672f62019-12-10 10:32:29 +00004231/* Note: This doesn't put any conntrack info in dst. */
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004232static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4233 bool copy)
4234{
4235#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4236 dst->_nfct = src->_nfct;
4237 nf_conntrack_get(skb_nfct(src));
4238#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004239#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4240 if (copy)
4241 dst->nf_trace = src->nf_trace;
4242#endif
4243}
4244
4245static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4246{
4247#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4248 nf_conntrack_put(skb_nfct(dst));
4249#endif
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004250 __nf_copy(dst, src, true);
4251}
4252
4253#ifdef CONFIG_NETWORK_SECMARK
4254static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4255{
4256 to->secmark = from->secmark;
4257}
4258
4259static inline void skb_init_secmark(struct sk_buff *skb)
4260{
4261 skb->secmark = 0;
4262}
4263#else
4264static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4265{ }
4266
4267static inline void skb_init_secmark(struct sk_buff *skb)
4268{ }
4269#endif
4270
David Brazdil0f672f62019-12-10 10:32:29 +00004271static inline int secpath_exists(const struct sk_buff *skb)
4272{
4273#ifdef CONFIG_XFRM
4274 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4275#else
4276 return 0;
4277#endif
4278}
4279
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004280static inline bool skb_irq_freeable(const struct sk_buff *skb)
4281{
4282 return !skb->destructor &&
David Brazdil0f672f62019-12-10 10:32:29 +00004283 !secpath_exists(skb) &&
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004284 !skb_nfct(skb) &&
4285 !skb->_skb_refdst &&
4286 !skb_has_frag_list(skb);
4287}
4288
4289static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4290{
4291 skb->queue_mapping = queue_mapping;
4292}
4293
4294static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4295{
4296 return skb->queue_mapping;
4297}
4298
4299static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4300{
4301 to->queue_mapping = from->queue_mapping;
4302}
4303
4304static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4305{
4306 skb->queue_mapping = rx_queue + 1;
4307}
4308
4309static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4310{
4311 return skb->queue_mapping - 1;
4312}
4313
4314static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4315{
4316 return skb->queue_mapping != 0;
4317}
4318
4319static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4320{
4321 skb->dst_pending_confirm = val;
4322}
4323
4324static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4325{
4326 return skb->dst_pending_confirm != 0;
4327}
4328
David Brazdil0f672f62019-12-10 10:32:29 +00004329static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004330{
4331#ifdef CONFIG_XFRM
David Brazdil0f672f62019-12-10 10:32:29 +00004332 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004333#else
4334 return NULL;
4335#endif
4336}
4337
4338/* Keeps track of mac header offset relative to skb->head.
4339 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4340 * For non-tunnel skb it points to skb_mac_header() and for
4341 * tunnel skb it points to outer mac header.
4342 * Keeps track of level of encapsulation of network headers.
4343 */
4344struct skb_gso_cb {
4345 union {
4346 int mac_offset;
4347 int data_offset;
4348 };
4349 int encap_level;
4350 __wsum csum;
4351 __u16 csum_start;
4352};
4353#define SKB_SGO_CB_OFFSET 32
4354#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4355
4356static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4357{
4358 return (skb_mac_header(inner_skb) - inner_skb->head) -
4359 SKB_GSO_CB(inner_skb)->mac_offset;
4360}
4361
4362static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4363{
4364 int new_headroom, headroom;
4365 int ret;
4366
4367 headroom = skb_headroom(skb);
4368 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4369 if (ret)
4370 return ret;
4371
4372 new_headroom = skb_headroom(skb);
4373 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4374 return 0;
4375}
4376
4377static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4378{
4379 /* Do not update partial checksums if remote checksum is enabled. */
4380 if (skb->remcsum_offload)
4381 return;
4382
4383 SKB_GSO_CB(skb)->csum = res;
4384 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4385}
4386
4387/* Compute the checksum for a gso segment. First compute the checksum value
4388 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4389 * then add in skb->csum (checksum from csum_start to end of packet).
4390 * skb->csum and csum_start are then updated to reflect the checksum of the
4391 * resultant packet starting from the transport header-- the resultant checksum
4392 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4393 * header.
4394 */
4395static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4396{
4397 unsigned char *csum_start = skb_transport_header(skb);
4398 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4399 __wsum partial = SKB_GSO_CB(skb)->csum;
4400
4401 SKB_GSO_CB(skb)->csum = res;
4402 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4403
4404 return csum_fold(csum_partial(csum_start, plen, partial));
4405}
4406
4407static inline bool skb_is_gso(const struct sk_buff *skb)
4408{
4409 return skb_shinfo(skb)->gso_size;
4410}
4411
4412/* Note: Should be called only if skb_is_gso(skb) is true */
4413static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4414{
4415 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4416}
4417
4418/* Note: Should be called only if skb_is_gso(skb) is true */
4419static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4420{
4421 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4422}
4423
David Brazdil0f672f62019-12-10 10:32:29 +00004424/* Note: Should be called only if skb_is_gso(skb) is true */
4425static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4426{
4427 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4428}
4429
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004430static inline void skb_gso_reset(struct sk_buff *skb)
4431{
4432 skb_shinfo(skb)->gso_size = 0;
4433 skb_shinfo(skb)->gso_segs = 0;
4434 skb_shinfo(skb)->gso_type = 0;
4435}
4436
4437static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4438 u16 increment)
4439{
4440 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4441 return;
4442 shinfo->gso_size += increment;
4443}
4444
4445static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4446 u16 decrement)
4447{
4448 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4449 return;
4450 shinfo->gso_size -= decrement;
4451}
4452
4453void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4454
4455static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4456{
4457 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4458 * wanted then gso_type will be set. */
4459 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4460
4461 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4462 unlikely(shinfo->gso_type == 0)) {
4463 __skb_warn_lro_forwarding(skb);
4464 return true;
4465 }
4466 return false;
4467}
4468
4469static inline void skb_forward_csum(struct sk_buff *skb)
4470{
4471 /* Unfortunately we don't support this one. Any brave souls? */
4472 if (skb->ip_summed == CHECKSUM_COMPLETE)
4473 skb->ip_summed = CHECKSUM_NONE;
4474}
4475
4476/**
4477 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4478 * @skb: skb to check
4479 *
4480 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4481 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4482 * use this helper, to document places where we make this assertion.
4483 */
4484static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4485{
4486#ifdef DEBUG
4487 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4488#endif
4489}
4490
4491bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4492
4493int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4494struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4495 unsigned int transport_len,
4496 __sum16(*skb_chkf)(struct sk_buff *skb));
4497
4498/**
4499 * skb_head_is_locked - Determine if the skb->head is locked down
4500 * @skb: skb to check
4501 *
4502 * The head on skbs build around a head frag can be removed if they are
4503 * not cloned. This function returns true if the skb head is locked down
4504 * due to either being allocated via kmalloc, or by being a clone with
4505 * multiple references to the head.
4506 */
4507static inline bool skb_head_is_locked(const struct sk_buff *skb)
4508{
4509 return !skb->head_frag || skb_cloned(skb);
4510}
4511
4512/* Local Checksum Offload.
4513 * Compute outer checksum based on the assumption that the
4514 * inner checksum will be offloaded later.
David Brazdil0f672f62019-12-10 10:32:29 +00004515 * See Documentation/networking/checksum-offloads.rst for
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004516 * explanation of how this works.
4517 * Fill in outer checksum adjustment (e.g. with sum of outer
4518 * pseudo-header) before calling.
4519 * Also ensure that inner checksum is in linear data area.
4520 */
4521static inline __wsum lco_csum(struct sk_buff *skb)
4522{
4523 unsigned char *csum_start = skb_checksum_start(skb);
4524 unsigned char *l4_hdr = skb_transport_header(skb);
4525 __wsum partial;
4526
4527 /* Start with complement of inner checksum adjustment */
4528 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4529 skb->csum_offset));
4530
4531 /* Add in checksum of our headers (incl. outer checksum
4532 * adjustment filled in by caller) and return result.
4533 */
4534 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4535}
4536
Olivier Deprez0e641232021-09-23 10:07:05 +02004537static inline bool skb_is_redirected(const struct sk_buff *skb)
4538{
4539#ifdef CONFIG_NET_REDIRECT
4540 return skb->redirected;
4541#else
4542 return false;
4543#endif
4544}
4545
4546static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
4547{
4548#ifdef CONFIG_NET_REDIRECT
4549 skb->redirected = 1;
4550 skb->from_ingress = from_ingress;
4551 if (skb->from_ingress)
4552 skb->tstamp = 0;
4553#endif
4554}
4555
4556static inline void skb_reset_redirect(struct sk_buff *skb)
4557{
4558#ifdef CONFIG_NET_REDIRECT
4559 skb->redirected = 0;
4560#endif
4561}
4562
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00004563#endif /* __KERNEL__ */
4564#endif /* _LINUX_SKBUFF_H */